accel/stubs/Makefile.objs | 1 + accel/stubs/nvmm-stub.c | 43 ++ configure | 37 ++ cpus.c | 58 ++ include/sysemu/hw_accel.h | 14 + include/sysemu/nvmm.h | 35 ++ qemu-options.hx | 16 +- target/i386/Makefile.objs | 1 + target/i386/helper.c | 2 +- target/i386/nvmm-all.c | 1221 +++++++++++++++++++++++++++++++++++++ 10 files changed, 1419 insertions(+), 9 deletions(-) create mode 100644 accel/stubs/nvmm-stub.c create mode 100644 include/sysemu/nvmm.h create mode 100644 target/i386/nvmm-all.c
Hello QEMU Community! Over the past year the NetBSD team has been working hard on a new user-mode API for our hypervisor that will be released as part of the upcoming NetBSD 9.0. This new API adds user-mode capabilities to create and manage virtual machines, configure memory mappings for guest machines, and create and control execution of virtual processors. With this new API we are now able to bring our hypervisor to the QEMU community! The following patches implement the NetBSD Virtual Machine Monitor accelerator (NVMM) for QEMU on NetBSD 9.0 and newer hosts. When compiling QEMU for x86_64 passing the --enable-nvmm flag will compile the accelerator for use. At runtime using the '-accel nvmm' should see a significant performance improvement over emulation, much like when using 'hax' on NetBSD. The documentation for this new API is visible at https://man.netbsd.org under the libnvmm(3) and nvmm(4) pages. NVMM was designed and implemented by Maxime Villard. Thank you for your feedback. Refrences: https://m00nbsd.net/4e0798b7f2620c965d0dd9d6a7a2f296.html Test plan: 1. Download a NetBSD 9.0 pre-release snapshot: http://nycdn.netbsd.org/pub/NetBSD-daily/netbsd-9/latest/images/NetBSD-9.0_RC1-amd64.iso 2. Install it natively on a not too old x86_64 hardware (Intel or AMD). There is no support for nested virtualization in NVMM. 3. Setup the system. export PKG_PATH=http://www.ki.nu/pkgsrc/packages/current/NetBSD-9.0_RC1/All pkg_add git gmake python37 glib2 bison pkgconf pixman Install mozilla-rootcerts and follow post-install instructions. pkg_add mozilla-rootcerts More information: https://wiki.qemu.org/Hosts/BSD#NetBSD 4. Build qemu mkdir build cd build ../configure --python=python3.7 gmake gmake check 5. Test qemu -accel nvmm ... History: v2 -> v3: - Register nvmm in targetos NetBSD check - Stop including hw/boards.h - Rephrase old code comments (remove XXX) v1 -> v2: - Included the testing plan as requested by Philippe Mathieu-Daude - Formatting nit fix in qemu-options.hx - Document NVMM in the accel section of qemu-options.hx Maxime Villard (4): Add the NVMM vcpu API Add the NetBSD Virtual Machine Monitor accelerator. Introduce the NVMM impl Add the NVMM acceleration enlightenments accel/stubs/Makefile.objs | 1 + accel/stubs/nvmm-stub.c | 43 ++ configure | 37 ++ cpus.c | 58 ++ include/sysemu/hw_accel.h | 14 + include/sysemu/nvmm.h | 35 ++ qemu-options.hx | 16 +- target/i386/Makefile.objs | 1 + target/i386/helper.c | 2 +- target/i386/nvmm-all.c | 1221 +++++++++++++++++++++++++++++++++++++ 10 files changed, 1419 insertions(+), 9 deletions(-) create mode 100644 accel/stubs/nvmm-stub.c create mode 100644 include/sysemu/nvmm.h create mode 100644 target/i386/nvmm-all.c -- 2.25.0
Hello QEMU Community! Over the past year the NetBSD team has been working hard on a new user-mode API for our hypervisor that will be released as part of the upcoming NetBSD 9.0. This new API adds user-mode capabilities to create and manage virtual machines, configure memory mappings for guest machines, and create and control execution of virtual processors. With this new API we are now able to bring our hypervisor to the QEMU community! The following patches implement the NetBSD Virtual Machine Monitor accelerator (NVMM) for QEMU on NetBSD 9.0 and newer hosts. When compiling QEMU for x86_64 passing the --enable-nvmm flag will compile the accelerator for use. At runtime using the '-accel nvmm' should see a significant performance improvement over emulation, much like when using 'hax' on NetBSD. The documentation for this new API is visible at https://man.netbsd.org under the libnvmm(3) and nvmm(4) pages. NVMM was designed and implemented by Maxime Villard. Thank you for your feedback. Refrences: https://m00nbsd.net/4e0798b7f2620c965d0dd9d6a7a2f296.html Test plan: 1. Download a NetBSD 9.0 pre-release snapshot: http://nycdn.netbsd.org/pub/NetBSD-daily/netbsd-9/latest/images/NetBSD-9.0_RC1-amd64.iso 2. Install it natively on a not too old x86_64 hardware (Intel or AMD). There is no support for nested virtualization in NVMM. 3. Setup the system. export PKG_PATH=http://www.ki.nu/pkgsrc/packages/current/NetBSD-9.0_RC1/All pkg_add git gmake python37 glib2 bison pkgconf pixman Install mozilla-rootcerts and follow post-install instructions. pkg_add mozilla-rootcerts More information: https://wiki.qemu.org/Hosts/BSD#NetBSD 4. Build qemu mkdir build cd build ../configure --python=python3.7 gmake gmake check 5. Test qemu -accel nvmm ... History: v3 -> v4: - Correct build warning by adding a missing include - Do not set R8-R16 registers unless TARGET_X86_64 v2 -> v3: - Register nvmm in targetos NetBSD check - Stop including hw/boards.h - Rephrase old code comments (remove XXX) v1 -> v2: - Included the testing plan as requested by Philippe Mathieu-Daude - Formatting nit fix in qemu-options.hx - Document NVMM in the accel section of qemu-options.hx Maxime Villard (4): Add the NVMM vcpu API Add the NetBSD Virtual Machine Monitor accelerator. Introduce the NVMM impl Add the NVMM acceleration enlightenments accel/stubs/Makefile.objs | 1 + accel/stubs/nvmm-stub.c | 43 ++ configure | 37 ++ cpus.c | 58 ++ include/sysemu/hw_accel.h | 14 + include/sysemu/nvmm.h | 35 ++ qemu-options.hx | 16 +- target/i386/Makefile.objs | 1 + target/i386/helper.c | 2 +- target/i386/nvmm-all.c | 1226 +++++++++++++++++++++++++++++++++++++ 10 files changed, 1424 insertions(+), 9 deletions(-) create mode 100644 accel/stubs/nvmm-stub.c create mode 100644 include/sysemu/nvmm.h create mode 100644 target/i386/nvmm-all.c -- 2.25.0
Ping? On 06.02.2020 22:32, Kamil Rytarowski wrote: > Hello QEMU Community! > > Over the past year the NetBSD team has been working hard on a new user-mode API > for our hypervisor that will be released as part of the upcoming NetBSD 9.0. > This new API adds user-mode capabilities to create and manage virtual machines, > configure memory mappings for guest machines, and create and control execution > of virtual processors. > > With this new API we are now able to bring our hypervisor to the QEMU > community! The following patches implement the NetBSD Virtual Machine Monitor > accelerator (NVMM) for QEMU on NetBSD 9.0 and newer hosts. > > When compiling QEMU for x86_64 passing the --enable-nvmm flag will compile the > accelerator for use. At runtime using the '-accel nvmm' should see a > significant performance improvement over emulation, much like when using 'hax' > on NetBSD. > > The documentation for this new API is visible at https://man.netbsd.org under > the libnvmm(3) and nvmm(4) pages. > > NVMM was designed and implemented by Maxime Villard. > > Thank you for your feedback. > > Refrences: > https://m00nbsd.net/4e0798b7f2620c965d0dd9d6a7a2f296.html > > Test plan: > > 1. Download a NetBSD 9.0 pre-release snapshot: > http://nycdn.netbsd.org/pub/NetBSD-daily/netbsd-9/latest/images/NetBSD-9.0_RC1-amd64.iso > > 2. Install it natively on a not too old x86_64 hardware (Intel or AMD). > > There is no support for nested virtualization in NVMM. > > 3. Setup the system. > > export PKG_PATH=http://www.ki.nu/pkgsrc/packages/current/NetBSD-9.0_RC1/All > pkg_add git gmake python37 glib2 bison pkgconf pixman > > Install mozilla-rootcerts and follow post-install instructions. > > pkg_add mozilla-rootcerts > > More information: https://wiki.qemu.org/Hosts/BSD#NetBSD > > 4. Build qemu > > mkdir build > cd build > ../configure --python=python3.7 > gmake > gmake check > > 5. Test > > qemu -accel nvmm ... > > > History: > v3 -> v4: > - Correct build warning by adding a missing include > - Do not set R8-R16 registers unless TARGET_X86_64 > v2 -> v3: > - Register nvmm in targetos NetBSD check > - Stop including hw/boards.h > - Rephrase old code comments (remove XXX) > v1 -> v2: > - Included the testing plan as requested by Philippe Mathieu-Daude > - Formatting nit fix in qemu-options.hx > - Document NVMM in the accel section of qemu-options.hx > > Maxime Villard (4): > Add the NVMM vcpu API > Add the NetBSD Virtual Machine Monitor accelerator. > Introduce the NVMM impl > Add the NVMM acceleration enlightenments > > accel/stubs/Makefile.objs | 1 + > accel/stubs/nvmm-stub.c | 43 ++ > configure | 37 ++ > cpus.c | 58 ++ > include/sysemu/hw_accel.h | 14 + > include/sysemu/nvmm.h | 35 ++ > qemu-options.hx | 16 +- > target/i386/Makefile.objs | 1 + > target/i386/helper.c | 2 +- > target/i386/nvmm-all.c | 1226 +++++++++++++++++++++++++++++++++++++ > 10 files changed, 1424 insertions(+), 9 deletions(-) > create mode 100644 accel/stubs/nvmm-stub.c > create mode 100644 include/sysemu/nvmm.h > create mode 100644 target/i386/nvmm-all.c > > -- > 2.25.0 > >
Ping? On 17.02.2020 10:07, Kamil Rytarowski wrote: > Ping? > > On 06.02.2020 22:32, Kamil Rytarowski wrote: >> Hello QEMU Community! >> >> Over the past year the NetBSD team has been working hard on a new user-mode API >> for our hypervisor that will be released as part of the upcoming NetBSD 9.0. >> This new API adds user-mode capabilities to create and manage virtual machines, >> configure memory mappings for guest machines, and create and control execution >> of virtual processors. >> >> With this new API we are now able to bring our hypervisor to the QEMU >> community! The following patches implement the NetBSD Virtual Machine Monitor >> accelerator (NVMM) for QEMU on NetBSD 9.0 and newer hosts. >> >> When compiling QEMU for x86_64 passing the --enable-nvmm flag will compile the >> accelerator for use. At runtime using the '-accel nvmm' should see a >> significant performance improvement over emulation, much like when using 'hax' >> on NetBSD. >> >> The documentation for this new API is visible at https://man.netbsd.org under >> the libnvmm(3) and nvmm(4) pages. >> >> NVMM was designed and implemented by Maxime Villard. >> >> Thank you for your feedback. >> >> Refrences: >> https://m00nbsd.net/4e0798b7f2620c965d0dd9d6a7a2f296.html >> >> Test plan: >> >> 1. Download a NetBSD 9.0 pre-release snapshot: >> http://nycdn.netbsd.org/pub/NetBSD-daily/netbsd-9/latest/images/NetBSD-9.0_RC1-amd64.iso >> >> 2. Install it natively on a not too old x86_64 hardware (Intel or AMD). >> >> There is no support for nested virtualization in NVMM. >> >> 3. Setup the system. >> >> export PKG_PATH=http://www.ki.nu/pkgsrc/packages/current/NetBSD-9.0_RC1/All >> pkg_add git gmake python37 glib2 bison pkgconf pixman >> >> Install mozilla-rootcerts and follow post-install instructions. >> >> pkg_add mozilla-rootcerts >> >> More information: https://wiki.qemu.org/Hosts/BSD#NetBSD >> >> 4. Build qemu >> >> mkdir build >> cd build >> ../configure --python=python3.7 >> gmake >> gmake check >> >> 5. Test >> >> qemu -accel nvmm ... >> >> >> History: >> v3 -> v4: >> - Correct build warning by adding a missing include >> - Do not set R8-R16 registers unless TARGET_X86_64 >> v2 -> v3: >> - Register nvmm in targetos NetBSD check >> - Stop including hw/boards.h >> - Rephrase old code comments (remove XXX) >> v1 -> v2: >> - Included the testing plan as requested by Philippe Mathieu-Daude >> - Formatting nit fix in qemu-options.hx >> - Document NVMM in the accel section of qemu-options.hx >> >> Maxime Villard (4): >> Add the NVMM vcpu API >> Add the NetBSD Virtual Machine Monitor accelerator. >> Introduce the NVMM impl >> Add the NVMM acceleration enlightenments >> >> accel/stubs/Makefile.objs | 1 + >> accel/stubs/nvmm-stub.c | 43 ++ >> configure | 37 ++ >> cpus.c | 58 ++ >> include/sysemu/hw_accel.h | 14 + >> include/sysemu/nvmm.h | 35 ++ >> qemu-options.hx | 16 +- >> target/i386/Makefile.objs | 1 + >> target/i386/helper.c | 2 +- >> target/i386/nvmm-all.c | 1226 +++++++++++++++++++++++++++++++++++++ >> 10 files changed, 1424 insertions(+), 9 deletions(-) >> create mode 100644 accel/stubs/nvmm-stub.c >> create mode 100644 include/sysemu/nvmm.h >> create mode 100644 target/i386/nvmm-all.c >> >> -- >> 2.25.0 >> >> >
Ping? On 24.02.2020 16:17, Kamil Rytarowski wrote: > Ping? > > On 17.02.2020 10:07, Kamil Rytarowski wrote: >> Ping? >> >> On 06.02.2020 22:32, Kamil Rytarowski wrote: >>> Hello QEMU Community! >>> >>> Over the past year the NetBSD team has been working hard on a new user-mode API >>> for our hypervisor that will be released as part of the upcoming NetBSD 9.0. >>> This new API adds user-mode capabilities to create and manage virtual machines, >>> configure memory mappings for guest machines, and create and control execution >>> of virtual processors. >>> >>> With this new API we are now able to bring our hypervisor to the QEMU >>> community! The following patches implement the NetBSD Virtual Machine Monitor >>> accelerator (NVMM) for QEMU on NetBSD 9.0 and newer hosts. >>> >>> When compiling QEMU for x86_64 passing the --enable-nvmm flag will compile the >>> accelerator for use. At runtime using the '-accel nvmm' should see a >>> significant performance improvement over emulation, much like when using 'hax' >>> on NetBSD. >>> >>> The documentation for this new API is visible at https://man.netbsd.org under >>> the libnvmm(3) and nvmm(4) pages. >>> >>> NVMM was designed and implemented by Maxime Villard. >>> >>> Thank you for your feedback. >>> >>> Refrences: >>> https://m00nbsd.net/4e0798b7f2620c965d0dd9d6a7a2f296.html >>> >>> Test plan: >>> >>> 1. Download a NetBSD 9.0 pre-release snapshot: >>> http://nycdn.netbsd.org/pub/NetBSD-daily/netbsd-9/latest/images/NetBSD-9.0_RC1-amd64.iso >>> >>> 2. Install it natively on a not too old x86_64 hardware (Intel or AMD). >>> >>> There is no support for nested virtualization in NVMM. >>> >>> 3. Setup the system. >>> >>> export PKG_PATH=http://www.ki.nu/pkgsrc/packages/current/NetBSD-9.0_RC1/All >>> pkg_add git gmake python37 glib2 bison pkgconf pixman >>> >>> Install mozilla-rootcerts and follow post-install instructions. >>> >>> pkg_add mozilla-rootcerts >>> >>> More information: https://wiki.qemu.org/Hosts/BSD#NetBSD >>> >>> 4. Build qemu >>> >>> mkdir build >>> cd build >>> ../configure --python=python3.7 >>> gmake >>> gmake check >>> >>> 5. Test >>> >>> qemu -accel nvmm ... >>> >>> >>> History: >>> v3 -> v4: >>> - Correct build warning by adding a missing include >>> - Do not set R8-R16 registers unless TARGET_X86_64 >>> v2 -> v3: >>> - Register nvmm in targetos NetBSD check >>> - Stop including hw/boards.h >>> - Rephrase old code comments (remove XXX) >>> v1 -> v2: >>> - Included the testing plan as requested by Philippe Mathieu-Daude >>> - Formatting nit fix in qemu-options.hx >>> - Document NVMM in the accel section of qemu-options.hx >>> >>> Maxime Villard (4): >>> Add the NVMM vcpu API >>> Add the NetBSD Virtual Machine Monitor accelerator. >>> Introduce the NVMM impl >>> Add the NVMM acceleration enlightenments >>> >>> accel/stubs/Makefile.objs | 1 + >>> accel/stubs/nvmm-stub.c | 43 ++ >>> configure | 37 ++ >>> cpus.c | 58 ++ >>> include/sysemu/hw_accel.h | 14 + >>> include/sysemu/nvmm.h | 35 ++ >>> qemu-options.hx | 16 +- >>> target/i386/Makefile.objs | 1 + >>> target/i386/helper.c | 2 +- >>> target/i386/nvmm-all.c | 1226 +++++++++++++++++++++++++++++++++++++ >>> 10 files changed, 1424 insertions(+), 9 deletions(-) >>> create mode 100644 accel/stubs/nvmm-stub.c >>> create mode 100644 include/sysemu/nvmm.h >>> create mode 100644 target/i386/nvmm-all.c >>> >>> -- >>> 2.25.0 >>> >>> >> >
Hi Kamil, Maxime, I haven't managed to reserve time to review this, sorry for that. I hope others can chime in before I do. Would any of you be willing to be included as maintainer of the new code on MAINTAINERS? On Mon, Mar 02, 2020 at 06:02:18PM +0100, Kamil Rytarowski wrote: > Ping? > > On 24.02.2020 16:17, Kamil Rytarowski wrote: > > Ping? > > > > On 17.02.2020 10:07, Kamil Rytarowski wrote: > >> Ping? > >> > >> On 06.02.2020 22:32, Kamil Rytarowski wrote: > >>> Hello QEMU Community! > >>> > >>> Over the past year the NetBSD team has been working hard on a new user-mode API > >>> for our hypervisor that will be released as part of the upcoming NetBSD 9.0. > >>> This new API adds user-mode capabilities to create and manage virtual machines, > >>> configure memory mappings for guest machines, and create and control execution > >>> of virtual processors. > >>> > >>> With this new API we are now able to bring our hypervisor to the QEMU > >>> community! The following patches implement the NetBSD Virtual Machine Monitor > >>> accelerator (NVMM) for QEMU on NetBSD 9.0 and newer hosts. > >>> > >>> When compiling QEMU for x86_64 passing the --enable-nvmm flag will compile the > >>> accelerator for use. At runtime using the '-accel nvmm' should see a > >>> significant performance improvement over emulation, much like when using 'hax' > >>> on NetBSD. > >>> > >>> The documentation for this new API is visible at https://man.netbsd.org under > >>> the libnvmm(3) and nvmm(4) pages. > >>> > >>> NVMM was designed and implemented by Maxime Villard. > >>> > >>> Thank you for your feedback. > >>> > >>> Refrences: > >>> https://m00nbsd.net/4e0798b7f2620c965d0dd9d6a7a2f296.html > >>> > >>> Test plan: > >>> > >>> 1. Download a NetBSD 9.0 pre-release snapshot: > >>> http://nycdn.netbsd.org/pub/NetBSD-daily/netbsd-9/latest/images/NetBSD-9.0_RC1-amd64.iso > >>> > >>> 2. Install it natively on a not too old x86_64 hardware (Intel or AMD). > >>> > >>> There is no support for nested virtualization in NVMM. > >>> > >>> 3. Setup the system. > >>> > >>> export PKG_PATH=http://www.ki.nu/pkgsrc/packages/current/NetBSD-9.0_RC1/All > >>> pkg_add git gmake python37 glib2 bison pkgconf pixman > >>> > >>> Install mozilla-rootcerts and follow post-install instructions. > >>> > >>> pkg_add mozilla-rootcerts > >>> > >>> More information: https://wiki.qemu.org/Hosts/BSD#NetBSD > >>> > >>> 4. Build qemu > >>> > >>> mkdir build > >>> cd build > >>> ../configure --python=python3.7 > >>> gmake > >>> gmake check > >>> > >>> 5. Test > >>> > >>> qemu -accel nvmm ... > >>> > >>> > >>> History: > >>> v3 -> v4: > >>> - Correct build warning by adding a missing include > >>> - Do not set R8-R16 registers unless TARGET_X86_64 > >>> v2 -> v3: > >>> - Register nvmm in targetos NetBSD check > >>> - Stop including hw/boards.h > >>> - Rephrase old code comments (remove XXX) > >>> v1 -> v2: > >>> - Included the testing plan as requested by Philippe Mathieu-Daude > >>> - Formatting nit fix in qemu-options.hx > >>> - Document NVMM in the accel section of qemu-options.hx > >>> > >>> Maxime Villard (4): > >>> Add the NVMM vcpu API > >>> Add the NetBSD Virtual Machine Monitor accelerator. > >>> Introduce the NVMM impl > >>> Add the NVMM acceleration enlightenments > >>> > >>> accel/stubs/Makefile.objs | 1 + > >>> accel/stubs/nvmm-stub.c | 43 ++ > >>> configure | 37 ++ > >>> cpus.c | 58 ++ > >>> include/sysemu/hw_accel.h | 14 + > >>> include/sysemu/nvmm.h | 35 ++ > >>> qemu-options.hx | 16 +- > >>> target/i386/Makefile.objs | 1 + > >>> target/i386/helper.c | 2 +- > >>> target/i386/nvmm-all.c | 1226 +++++++++++++++++++++++++++++++++++++ > >>> 10 files changed, 1424 insertions(+), 9 deletions(-) > >>> create mode 100644 accel/stubs/nvmm-stub.c > >>> create mode 100644 include/sysemu/nvmm.h > >>> create mode 100644 target/i386/nvmm-all.c > >>> > >>> -- > >>> 2.25.0 > >>> > >>> > >> > > > -- Eduardo
On 02.03.2020 18:10, Eduardo Habkost wrote: > Hi Kamil, Maxime, > > I haven't managed to reserve time to review this, sorry for that. > I hope others can chime in before I do. > > Would any of you be willing to be included as maintainer of the > new code on MAINTAINERS? > I'm already mentioned as the NetBSD maintainer and NVMM is NetBSD-only (at least today). > > On Mon, Mar 02, 2020 at 06:02:18PM +0100, Kamil Rytarowski wrote: >> Ping? >> >> On 24.02.2020 16:17, Kamil Rytarowski wrote: >>> Ping? >>> >>> On 17.02.2020 10:07, Kamil Rytarowski wrote: >>>> Ping? >>>> >>>> On 06.02.2020 22:32, Kamil Rytarowski wrote: >>>>> Hello QEMU Community! >>>>> >>>>> Over the past year the NetBSD team has been working hard on a new user-mode API >>>>> for our hypervisor that will be released as part of the upcoming NetBSD 9.0. >>>>> This new API adds user-mode capabilities to create and manage virtual machines, >>>>> configure memory mappings for guest machines, and create and control execution >>>>> of virtual processors. >>>>> >>>>> With this new API we are now able to bring our hypervisor to the QEMU >>>>> community! The following patches implement the NetBSD Virtual Machine Monitor >>>>> accelerator (NVMM) for QEMU on NetBSD 9.0 and newer hosts. >>>>> >>>>> When compiling QEMU for x86_64 passing the --enable-nvmm flag will compile the >>>>> accelerator for use. At runtime using the '-accel nvmm' should see a >>>>> significant performance improvement over emulation, much like when using 'hax' >>>>> on NetBSD. >>>>> >>>>> The documentation for this new API is visible at https://man.netbsd.org under >>>>> the libnvmm(3) and nvmm(4) pages. >>>>> >>>>> NVMM was designed and implemented by Maxime Villard. >>>>> >>>>> Thank you for your feedback. >>>>> >>>>> Refrences: >>>>> https://m00nbsd.net/4e0798b7f2620c965d0dd9d6a7a2f296.html >>>>> >>>>> Test plan: >>>>> >>>>> 1. Download a NetBSD 9.0 pre-release snapshot: >>>>> http://nycdn.netbsd.org/pub/NetBSD-daily/netbsd-9/latest/images/NetBSD-9.0_RC1-amd64.iso >>>>> >>>>> 2. Install it natively on a not too old x86_64 hardware (Intel or AMD). >>>>> >>>>> There is no support for nested virtualization in NVMM. >>>>> >>>>> 3. Setup the system. >>>>> >>>>> export PKG_PATH=http://www.ki.nu/pkgsrc/packages/current/NetBSD-9.0_RC1/All >>>>> pkg_add git gmake python37 glib2 bison pkgconf pixman >>>>> >>>>> Install mozilla-rootcerts and follow post-install instructions. >>>>> >>>>> pkg_add mozilla-rootcerts >>>>> >>>>> More information: https://wiki.qemu.org/Hosts/BSD#NetBSD >>>>> >>>>> 4. Build qemu >>>>> >>>>> mkdir build >>>>> cd build >>>>> ../configure --python=python3.7 >>>>> gmake >>>>> gmake check >>>>> >>>>> 5. Test >>>>> >>>>> qemu -accel nvmm ... >>>>> >>>>> >>>>> History: >>>>> v3 -> v4: >>>>> - Correct build warning by adding a missing include >>>>> - Do not set R8-R16 registers unless TARGET_X86_64 >>>>> v2 -> v3: >>>>> - Register nvmm in targetos NetBSD check >>>>> - Stop including hw/boards.h >>>>> - Rephrase old code comments (remove XXX) >>>>> v1 -> v2: >>>>> - Included the testing plan as requested by Philippe Mathieu-Daude >>>>> - Formatting nit fix in qemu-options.hx >>>>> - Document NVMM in the accel section of qemu-options.hx >>>>> >>>>> Maxime Villard (4): >>>>> Add the NVMM vcpu API >>>>> Add the NetBSD Virtual Machine Monitor accelerator. >>>>> Introduce the NVMM impl >>>>> Add the NVMM acceleration enlightenments >>>>> >>>>> accel/stubs/Makefile.objs | 1 + >>>>> accel/stubs/nvmm-stub.c | 43 ++ >>>>> configure | 37 ++ >>>>> cpus.c | 58 ++ >>>>> include/sysemu/hw_accel.h | 14 + >>>>> include/sysemu/nvmm.h | 35 ++ >>>>> qemu-options.hx | 16 +- >>>>> target/i386/Makefile.objs | 1 + >>>>> target/i386/helper.c | 2 +- >>>>> target/i386/nvmm-all.c | 1226 +++++++++++++++++++++++++++++++++++++ >>>>> 10 files changed, 1424 insertions(+), 9 deletions(-) >>>>> create mode 100644 accel/stubs/nvmm-stub.c >>>>> create mode 100644 include/sysemu/nvmm.h >>>>> create mode 100644 target/i386/nvmm-all.c >>>>> >>>>> -- >>>>> 2.25.0 >>>>> >>>>> >>>> >>> >> >
On Mon, Mar 02, 2020 at 06:10:50PM +0100, Kamil Rytarowski wrote: > On 02.03.2020 18:10, Eduardo Habkost wrote: > > Hi Kamil, Maxime, > > > > I haven't managed to reserve time to review this, sorry for that. > > I hope others can chime in before I do. > > > > Would any of you be willing to be included as maintainer of the > > new code on MAINTAINERS? > > > > I'm already mentioned as the NetBSD maintainer and NVMM is NetBSD-only > (at least today). I don't see the new files (accel/stubs/nvmm-stub.c, include/sysemu/nvmm.h, target/i386/nvmm-all.c) being added to MAINTAINERS. Can you add it as a follow up patch? -- Eduardo
From: Maxime Villard <max@m00nbsd.net>
Adds support for the NetBSD Virtual Machine Monitor (NVMM) stubs and
introduces the nvmm.h sysemu API for managing the vcpu scheduling and
management.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
accel/stubs/Makefile.objs | 1 +
accel/stubs/nvmm-stub.c | 43 +++++++++++++++++++++++++++++++++++++++
include/sysemu/nvmm.h | 35 +++++++++++++++++++++++++++++++
3 files changed, 79 insertions(+)
create mode 100644 accel/stubs/nvmm-stub.c
create mode 100644 include/sysemu/nvmm.h
diff --git a/accel/stubs/Makefile.objs b/accel/stubs/Makefile.objs
index 3894caf95d..09f2d3e1dd 100644
--- a/accel/stubs/Makefile.objs
+++ b/accel/stubs/Makefile.objs
@@ -1,5 +1,6 @@
obj-$(call lnot,$(CONFIG_HAX)) += hax-stub.o
obj-$(call lnot,$(CONFIG_HVF)) += hvf-stub.o
obj-$(call lnot,$(CONFIG_WHPX)) += whpx-stub.o
+obj-$(call lnot,$(CONFIG_NVMM)) += nvmm-stub.o
obj-$(call lnot,$(CONFIG_KVM)) += kvm-stub.o
obj-$(call lnot,$(CONFIG_TCG)) += tcg-stub.o
diff --git a/accel/stubs/nvmm-stub.c b/accel/stubs/nvmm-stub.c
new file mode 100644
index 0000000000..c2208b84a3
--- /dev/null
+++ b/accel/stubs/nvmm-stub.c
@@ -0,0 +1,43 @@
+/*
+ * Copyright (c) 2018-2019 Maxime Villard, All rights reserved.
+ *
+ * NetBSD Virtual Machine Monitor (NVMM) accelerator stub.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu-common.h"
+#include "cpu.h"
+#include "sysemu/nvmm.h"
+
+int nvmm_init_vcpu(CPUState *cpu)
+{
+ return -1;
+}
+
+int nvmm_vcpu_exec(CPUState *cpu)
+{
+ return -1;
+}
+
+void nvmm_destroy_vcpu(CPUState *cpu)
+{
+}
+
+void nvmm_cpu_synchronize_state(CPUState *cpu)
+{
+}
+
+void nvmm_cpu_synchronize_post_reset(CPUState *cpu)
+{
+}
+
+void nvmm_cpu_synchronize_post_init(CPUState *cpu)
+{
+}
+
+void nvmm_cpu_synchronize_pre_loadvm(CPUState *cpu)
+{
+}
diff --git a/include/sysemu/nvmm.h b/include/sysemu/nvmm.h
new file mode 100644
index 0000000000..10496f3980
--- /dev/null
+++ b/include/sysemu/nvmm.h
@@ -0,0 +1,35 @@
+/*
+ * Copyright (c) 2018-2019 Maxime Villard, All rights reserved.
+ *
+ * NetBSD Virtual Machine Monitor (NVMM) accelerator support.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#ifndef QEMU_NVMM_H
+#define QEMU_NVMM_H
+
+#include "config-host.h"
+#include "qemu-common.h"
+
+int nvmm_init_vcpu(CPUState *);
+int nvmm_vcpu_exec(CPUState *);
+void nvmm_destroy_vcpu(CPUState *);
+
+void nvmm_cpu_synchronize_state(CPUState *);
+void nvmm_cpu_synchronize_post_reset(CPUState *);
+void nvmm_cpu_synchronize_post_init(CPUState *);
+void nvmm_cpu_synchronize_pre_loadvm(CPUState *);
+
+#ifdef CONFIG_NVMM
+
+int nvmm_enabled(void);
+
+#else /* CONFIG_NVMM */
+
+#define nvmm_enabled() (0)
+
+#endif /* CONFIG_NVMM */
+
+#endif /* CONFIG_NVMM */
--
2.25.0
From: Maxime Villard <max@m00nbsd.net>
Adds support for the NetBSD Virtual Machine Monitor (NVMM) stubs and
introduces the nvmm.h sysemu API for managing the vcpu scheduling and
management.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
accel/stubs/Makefile.objs | 1 +
accel/stubs/nvmm-stub.c | 43 +++++++++++++++++++++++++++++++++++++++
include/sysemu/nvmm.h | 35 +++++++++++++++++++++++++++++++
3 files changed, 79 insertions(+)
create mode 100644 accel/stubs/nvmm-stub.c
create mode 100644 include/sysemu/nvmm.h
diff --git a/accel/stubs/Makefile.objs b/accel/stubs/Makefile.objs
index bbd14e71fb..38660a0b9b 100644
--- a/accel/stubs/Makefile.objs
+++ b/accel/stubs/Makefile.objs
@@ -1,6 +1,7 @@
obj-$(call lnot,$(CONFIG_HAX)) += hax-stub.o
obj-$(call lnot,$(CONFIG_HVF)) += hvf-stub.o
obj-$(call lnot,$(CONFIG_WHPX)) += whpx-stub.o
+obj-$(call lnot,$(CONFIG_NVMM)) += nvmm-stub.o
obj-$(call lnot,$(CONFIG_KVM)) += kvm-stub.o
obj-$(call lnot,$(CONFIG_TCG)) += tcg-stub.o
obj-$(call lnot,$(CONFIG_XEN)) += xen-stub.o
diff --git a/accel/stubs/nvmm-stub.c b/accel/stubs/nvmm-stub.c
new file mode 100644
index 0000000000..c2208b84a3
--- /dev/null
+++ b/accel/stubs/nvmm-stub.c
@@ -0,0 +1,43 @@
+/*
+ * Copyright (c) 2018-2019 Maxime Villard, All rights reserved.
+ *
+ * NetBSD Virtual Machine Monitor (NVMM) accelerator stub.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu-common.h"
+#include "cpu.h"
+#include "sysemu/nvmm.h"
+
+int nvmm_init_vcpu(CPUState *cpu)
+{
+ return -1;
+}
+
+int nvmm_vcpu_exec(CPUState *cpu)
+{
+ return -1;
+}
+
+void nvmm_destroy_vcpu(CPUState *cpu)
+{
+}
+
+void nvmm_cpu_synchronize_state(CPUState *cpu)
+{
+}
+
+void nvmm_cpu_synchronize_post_reset(CPUState *cpu)
+{
+}
+
+void nvmm_cpu_synchronize_post_init(CPUState *cpu)
+{
+}
+
+void nvmm_cpu_synchronize_pre_loadvm(CPUState *cpu)
+{
+}
diff --git a/include/sysemu/nvmm.h b/include/sysemu/nvmm.h
new file mode 100644
index 0000000000..10496f3980
--- /dev/null
+++ b/include/sysemu/nvmm.h
@@ -0,0 +1,35 @@
+/*
+ * Copyright (c) 2018-2019 Maxime Villard, All rights reserved.
+ *
+ * NetBSD Virtual Machine Monitor (NVMM) accelerator support.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#ifndef QEMU_NVMM_H
+#define QEMU_NVMM_H
+
+#include "config-host.h"
+#include "qemu-common.h"
+
+int nvmm_init_vcpu(CPUState *);
+int nvmm_vcpu_exec(CPUState *);
+void nvmm_destroy_vcpu(CPUState *);
+
+void nvmm_cpu_synchronize_state(CPUState *);
+void nvmm_cpu_synchronize_post_reset(CPUState *);
+void nvmm_cpu_synchronize_post_init(CPUState *);
+void nvmm_cpu_synchronize_pre_loadvm(CPUState *);
+
+#ifdef CONFIG_NVMM
+
+int nvmm_enabled(void);
+
+#else /* CONFIG_NVMM */
+
+#define nvmm_enabled() (0)
+
+#endif /* CONFIG_NVMM */
+
+#endif /* CONFIG_NVMM */
--
2.24.1
From: Maxime Villard <max@m00nbsd.net>
Introduces the configure support for the new NetBSD Virtual Machine Monitor that
allows for hypervisor acceleration from usermode components on the NetBSD
platform.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
configure | 37 +++++++++++++++++++++++++++++++++++++
qemu-options.hx | 10 +++++-----
2 files changed, 42 insertions(+), 5 deletions(-)
diff --git a/configure b/configure
index 2acc4d1465..fb9ffba2bf 100755
--- a/configure
+++ b/configure
@@ -246,6 +246,17 @@ supported_whpx_target() {
return 1
}
+supported_nvmm_target() {
+ test "$nvmm" = "yes" || return 1
+ glob "$1" "*-softmmu" || return 1
+ case "${1%-softmmu}" in
+ i386|x86_64)
+ return 0
+ ;;
+ esac
+ return 1
+}
+
supported_target() {
case "$1" in
*-softmmu)
@@ -273,6 +284,7 @@ supported_target() {
supported_hax_target "$1" && return 0
supported_hvf_target "$1" && return 0
supported_whpx_target "$1" && return 0
+ supported_nvmm_target "$1" && return 0
print_error "TCG disabled, but hardware accelerator not available for '$target'"
return 1
}
@@ -395,6 +407,7 @@ kvm="no"
hax="no"
hvf="no"
whpx="no"
+nvmm="no"
rdma=""
pvrdma=""
gprof="no"
@@ -847,6 +860,7 @@ DragonFly)
NetBSD)
bsd="yes"
hax="yes"
+ nvmm="yes"
make="${MAKE-gmake}"
audio_drv_list="oss try-sdl"
audio_possible_drivers="oss sdl"
@@ -1233,6 +1247,10 @@ for opt do
;;
--enable-whpx) whpx="yes"
;;
+ --disable-nvmm) nvmm="no"
+ ;;
+ --enable-nvmm) nvmm="yes"
+ ;;
--disable-tcg-interpreter) tcg_interpreter="no"
;;
--enable-tcg-interpreter) tcg_interpreter="yes"
@@ -1879,6 +1897,7 @@ disabled with --disable-FEATURE, default is enabled if available:
hax HAX acceleration support
hvf Hypervisor.framework acceleration support
whpx Windows Hypervisor Platform acceleration support
+ nvmm NetBSD Virtual Machine Monitor acceleration support
rdma Enable RDMA-based migration
pvrdma Enable PVRDMA support
vde support for vde network
@@ -2965,6 +2984,20 @@ if test "$whpx" != "no" ; then
fi
fi
+##########################################
+# NetBSD Virtual Machine Monitor (NVMM) accelerator check
+if test "$nvmm" != "no" ; then
+ if check_include "nvmm.h" ; then
+ nvmm="yes"
+ LIBS="-lnvmm $LIBS"
+ else
+ if test "$nvmm" = "yes"; then
+ feature_not_found "NVMM" "NVMM is not available"
+ fi
+ nvmm="no"
+ fi
+fi
+
##########################################
# Sparse probe
if test "$sparse" != "no" ; then
@@ -6934,6 +6967,7 @@ echo "KVM support $kvm"
echo "HAX support $hax"
echo "HVF support $hvf"
echo "WHPX support $whpx"
+echo "NVMM support $nvmm"
echo "TCG support $tcg"
if test "$tcg" = "yes" ; then
echo "TCG debug enabled $debug_tcg"
@@ -8332,6 +8366,9 @@ fi
if test "$target_aligned_only" = "yes" ; then
echo "TARGET_ALIGNED_ONLY=y" >> $config_target_mak
fi
+if supported_nvmm_target $target; then
+ echo "CONFIG_NVMM=y" >> $config_target_mak
+fi
if test "$target_bigendian" = "yes" ; then
echo "TARGET_WORDS_BIGENDIAN=y" >> $config_target_mak
fi
diff --git a/qemu-options.hx b/qemu-options.hx
index 708583b4ce..697accaa7e 100644
--- a/qemu-options.hx
+++ b/qemu-options.hx
@@ -26,7 +26,7 @@ DEF("machine", HAS_ARG, QEMU_OPTION_machine, \
"-machine [type=]name[,prop[=value][,...]]\n"
" selects emulated machine ('-machine help' for list)\n"
" property accel=accel1[:accel2[:...]] selects accelerator\n"
- " supported accelerators are kvm, xen, hax, hvf, whpx or tcg (default: tcg)\n"
+ " supported accelerators are kvm, xen, hax, hvf, nvmm, whpx or tcg (default: tcg)\n"
" vmport=on|off|auto controls emulation of vmport (default: auto)\n"
" dump-guest-core=on|off include guest memory in a core dump (default=on)\n"
" mem-merge=on|off controls memory merge support (default: on)\n"
@@ -58,7 +58,7 @@ SRST
``accel=accels1[:accels2[:...]]``
This is used to enable an accelerator. Depending on the target
- architecture, kvm, xen, hax, hvf, whpx or tcg can be available.
+ architecture, kvm, xen, hax, hvf, nvmm, whpx or tcg can be available.
By default, tcg is used. If there is more than one accelerator
specified, the next one is used if the previous one fails to
initialize.
@@ -119,7 +119,7 @@ ERST
DEF("accel", HAS_ARG, QEMU_OPTION_accel,
"-accel [accel=]accelerator[,prop[=value][,...]]\n"
- " select accelerator (kvm, xen, hax, hvf, whpx or tcg; use 'help' for a list)\n"
+ " select accelerator (kvm, xen, hax, hvf, nvmm, whpx or tcg; use 'help' for a list)\n"
" igd-passthru=on|off (enable Xen integrated Intel graphics passthrough, default=off)\n"
" kernel-irqchip=on|off|split controls accelerated irqchip support (default=on)\n"
" kvm-shadow-mem=size of KVM shadow MMU in bytes\n"
@@ -128,8 +128,8 @@ DEF("accel", HAS_ARG, QEMU_OPTION_accel,
SRST
``-accel name[,prop=value[,...]]``
This is used to enable an accelerator. Depending on the target
- architecture, kvm, xen, hax, hvf, whpx or tcg can be available. By
- default, tcg is used. If there is more than one accelerator
+ architecture, kvm, xen, hax, hvf, nvmm whpx or tcg can be available.
+ By default, tcg is used. If there is more than one accelerator
specified, the next one is used if the previous one fails to
initialize.
--
2.24.1
From: Maxime Villard <max@m00nbsd.net>
Implements the NetBSD Virtual Machine Monitor (NVMM) target. Which
acts as a hypervisor accelerator for QEMU on the NetBSD platform. This enables
QEMU much greater speed over the emulated x86_64 path's that are taken on
NetBSD today.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
target/i386/Makefile.objs | 1 +
target/i386/nvmm-all.c | 1226 +++++++++++++++++++++++++++++++++++++
2 files changed, 1227 insertions(+)
create mode 100644 target/i386/nvmm-all.c
diff --git a/target/i386/Makefile.objs b/target/i386/Makefile.objs
index 0b93143e27..ff0df68404 100644
--- a/target/i386/Makefile.objs
+++ b/target/i386/Makefile.objs
@@ -18,6 +18,7 @@ obj-$(CONFIG_HAX) += hax-all.o hax-mem.o hax-posix.o
endif
obj-$(CONFIG_HVF) += hvf/
obj-$(CONFIG_WHPX) += whpx-all.o
+obj-$(CONFIG_NVMM) += nvmm-all.o
endif
obj-$(CONFIG_SEV) += sev.o
obj-$(call lnot,$(CONFIG_SEV)) += sev-stub.o
diff --git a/target/i386/nvmm-all.c b/target/i386/nvmm-all.c
new file mode 100644
index 0000000000..408f7305b9
--- /dev/null
+++ b/target/i386/nvmm-all.c
@@ -0,0 +1,1226 @@
+/*
+ * Copyright (c) 2018-2019 Maxime Villard, All rights reserved.
+ *
+ * NetBSD Virtual Machine Monitor (NVMM) accelerator for QEMU.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "cpu.h"
+#include "exec/address-spaces.h"
+#include "exec/ioport.h"
+#include "qemu-common.h"
+#include "strings.h"
+#include "sysemu/accel.h"
+#include "sysemu/nvmm.h"
+#include "sysemu/runstate.h"
+#include "sysemu/sysemu.h"
+#include "sysemu/cpus.h"
+#include "qemu/main-loop.h"
+#include "qemu/error-report.h"
+#include "qemu/queue.h"
+#include "qapi/error.h"
+#include "migration/blocker.h"
+
+#include <nvmm.h>
+
+struct qemu_vcpu {
+ struct nvmm_vcpu vcpu;
+ uint8_t tpr;
+ bool stop;
+
+ /* Window-exiting for INTs/NMIs. */
+ bool int_window_exit;
+ bool nmi_window_exit;
+
+ /* The guest is in an interrupt shadow (POP SS, etc). */
+ bool int_shadow;
+};
+
+struct qemu_machine {
+ struct nvmm_capability cap;
+ struct nvmm_machine mach;
+};
+
+/* -------------------------------------------------------------------------- */
+
+static bool nvmm_allowed;
+static struct qemu_machine qemu_mach;
+
+static struct qemu_vcpu *
+get_qemu_vcpu(CPUState *cpu)
+{
+ return (struct qemu_vcpu *)cpu->hax_vcpu;
+}
+
+static struct nvmm_machine *
+get_nvmm_mach(void)
+{
+ return &qemu_mach.mach;
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_set_segment(struct nvmm_x64_state_seg *nseg, const SegmentCache *qseg)
+{
+ uint32_t attrib = qseg->flags;
+
+ nseg->selector = qseg->selector;
+ nseg->limit = qseg->limit;
+ nseg->base = qseg->base;
+ nseg->attrib.type = __SHIFTOUT(attrib, DESC_TYPE_MASK);
+ nseg->attrib.s = __SHIFTOUT(attrib, DESC_S_MASK);
+ nseg->attrib.dpl = __SHIFTOUT(attrib, DESC_DPL_MASK);
+ nseg->attrib.p = __SHIFTOUT(attrib, DESC_P_MASK);
+ nseg->attrib.avl = __SHIFTOUT(attrib, DESC_AVL_MASK);
+ nseg->attrib.l = __SHIFTOUT(attrib, DESC_L_MASK);
+ nseg->attrib.def = __SHIFTOUT(attrib, DESC_B_MASK);
+ nseg->attrib.g = __SHIFTOUT(attrib, DESC_G_MASK);
+}
+
+static void
+nvmm_set_registers(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t bitmap;
+ size_t i;
+ int ret;
+
+ assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
+
+ /* GPRs. */
+ state->gprs[NVMM_X64_GPR_RAX] = env->regs[R_EAX];
+ state->gprs[NVMM_X64_GPR_RCX] = env->regs[R_ECX];
+ state->gprs[NVMM_X64_GPR_RDX] = env->regs[R_EDX];
+ state->gprs[NVMM_X64_GPR_RBX] = env->regs[R_EBX];
+ state->gprs[NVMM_X64_GPR_RSP] = env->regs[R_ESP];
+ state->gprs[NVMM_X64_GPR_RBP] = env->regs[R_EBP];
+ state->gprs[NVMM_X64_GPR_RSI] = env->regs[R_ESI];
+ state->gprs[NVMM_X64_GPR_RDI] = env->regs[R_EDI];
+#ifdef TARGET_X86_64
+ state->gprs[NVMM_X64_GPR_R8] = env->regs[R_R8];
+ state->gprs[NVMM_X64_GPR_R9] = env->regs[R_R9];
+ state->gprs[NVMM_X64_GPR_R10] = env->regs[R_R10];
+ state->gprs[NVMM_X64_GPR_R11] = env->regs[R_R11];
+ state->gprs[NVMM_X64_GPR_R12] = env->regs[R_R12];
+ state->gprs[NVMM_X64_GPR_R13] = env->regs[R_R13];
+ state->gprs[NVMM_X64_GPR_R14] = env->regs[R_R14];
+ state->gprs[NVMM_X64_GPR_R15] = env->regs[R_R15];
+#endif
+
+ /* RIP and RFLAGS. */
+ state->gprs[NVMM_X64_GPR_RIP] = env->eip;
+ state->gprs[NVMM_X64_GPR_RFLAGS] = env->eflags;
+
+ /* Segments. */
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_CS], &env->segs[R_CS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_DS], &env->segs[R_DS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_ES], &env->segs[R_ES]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_FS], &env->segs[R_FS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_GS], &env->segs[R_GS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_SS], &env->segs[R_SS]);
+
+ /* Special segments. */
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_GDT], &env->gdt);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_LDT], &env->ldt);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_TR], &env->tr);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_IDT], &env->idt);
+
+ /* Control registers. */
+ state->crs[NVMM_X64_CR_CR0] = env->cr[0];
+ state->crs[NVMM_X64_CR_CR2] = env->cr[2];
+ state->crs[NVMM_X64_CR_CR3] = env->cr[3];
+ state->crs[NVMM_X64_CR_CR4] = env->cr[4];
+ state->crs[NVMM_X64_CR_CR8] = qcpu->tpr;
+ state->crs[NVMM_X64_CR_XCR0] = env->xcr0;
+
+ /* Debug registers. */
+ state->drs[NVMM_X64_DR_DR0] = env->dr[0];
+ state->drs[NVMM_X64_DR_DR1] = env->dr[1];
+ state->drs[NVMM_X64_DR_DR2] = env->dr[2];
+ state->drs[NVMM_X64_DR_DR3] = env->dr[3];
+ state->drs[NVMM_X64_DR_DR6] = env->dr[6];
+ state->drs[NVMM_X64_DR_DR7] = env->dr[7];
+
+ /* FPU. */
+ state->fpu.fx_cw = env->fpuc;
+ state->fpu.fx_sw = (env->fpus & ~0x3800) | ((env->fpstt & 0x7) << 11);
+ state->fpu.fx_tw = 0;
+ for (i = 0; i < 8; i++) {
+ state->fpu.fx_tw |= (!env->fptags[i]) << i;
+ }
+ state->fpu.fx_opcode = env->fpop;
+ state->fpu.fx_ip.fa_64 = env->fpip;
+ state->fpu.fx_dp.fa_64 = env->fpdp;
+ state->fpu.fx_mxcsr = env->mxcsr;
+ state->fpu.fx_mxcsr_mask = 0x0000FFFF;
+ assert(sizeof(state->fpu.fx_87_ac) == sizeof(env->fpregs));
+ memcpy(state->fpu.fx_87_ac, env->fpregs, sizeof(env->fpregs));
+ for (i = 0; i < 16; i++) {
+ memcpy(&state->fpu.fx_xmm[i].xmm_bytes[0],
+ &env->xmm_regs[i].ZMM_Q(0), 8);
+ memcpy(&state->fpu.fx_xmm[i].xmm_bytes[8],
+ &env->xmm_regs[i].ZMM_Q(1), 8);
+ }
+
+ /* MSRs. */
+ state->msrs[NVMM_X64_MSR_EFER] = env->efer;
+ state->msrs[NVMM_X64_MSR_STAR] = env->star;
+#ifdef TARGET_X86_64
+ state->msrs[NVMM_X64_MSR_LSTAR] = env->lstar;
+ state->msrs[NVMM_X64_MSR_CSTAR] = env->cstar;
+ state->msrs[NVMM_X64_MSR_SFMASK] = env->fmask;
+ state->msrs[NVMM_X64_MSR_KERNELGSBASE] = env->kernelgsbase;
+#endif
+ state->msrs[NVMM_X64_MSR_SYSENTER_CS] = env->sysenter_cs;
+ state->msrs[NVMM_X64_MSR_SYSENTER_ESP] = env->sysenter_esp;
+ state->msrs[NVMM_X64_MSR_SYSENTER_EIP] = env->sysenter_eip;
+ state->msrs[NVMM_X64_MSR_PAT] = env->pat;
+ state->msrs[NVMM_X64_MSR_TSC] = env->tsc;
+
+ bitmap =
+ NVMM_X64_STATE_SEGS |
+ NVMM_X64_STATE_GPRS |
+ NVMM_X64_STATE_CRS |
+ NVMM_X64_STATE_DRS |
+ NVMM_X64_STATE_MSRS |
+ NVMM_X64_STATE_FPU;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, bitmap);
+ if (ret == -1) {
+ error_report("NVMM: Failed to set virtual processor context,"
+ " error=%d", errno);
+ }
+}
+
+static void
+nvmm_get_segment(SegmentCache *qseg, const struct nvmm_x64_state_seg *nseg)
+{
+ qseg->selector = nseg->selector;
+ qseg->limit = nseg->limit;
+ qseg->base = nseg->base;
+
+ qseg->flags =
+ __SHIFTIN((uint32_t)nseg->attrib.type, DESC_TYPE_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.s, DESC_S_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.dpl, DESC_DPL_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.p, DESC_P_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.avl, DESC_AVL_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.l, DESC_L_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.def, DESC_B_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.g, DESC_G_MASK);
+}
+
+static void
+nvmm_get_registers(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t bitmap, tpr;
+ size_t i;
+ int ret;
+
+ assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
+
+ bitmap =
+ NVMM_X64_STATE_SEGS |
+ NVMM_X64_STATE_GPRS |
+ NVMM_X64_STATE_CRS |
+ NVMM_X64_STATE_DRS |
+ NVMM_X64_STATE_MSRS |
+ NVMM_X64_STATE_FPU;
+
+ ret = nvmm_vcpu_getstate(mach, vcpu, bitmap);
+ if (ret == -1) {
+ error_report("NVMM: Failed to get virtual processor context,"
+ " error=%d", errno);
+ }
+
+ /* GPRs. */
+ env->regs[R_EAX] = state->gprs[NVMM_X64_GPR_RAX];
+ env->regs[R_ECX] = state->gprs[NVMM_X64_GPR_RCX];
+ env->regs[R_EDX] = state->gprs[NVMM_X64_GPR_RDX];
+ env->regs[R_EBX] = state->gprs[NVMM_X64_GPR_RBX];
+ env->regs[R_ESP] = state->gprs[NVMM_X64_GPR_RSP];
+ env->regs[R_EBP] = state->gprs[NVMM_X64_GPR_RBP];
+ env->regs[R_ESI] = state->gprs[NVMM_X64_GPR_RSI];
+ env->regs[R_EDI] = state->gprs[NVMM_X64_GPR_RDI];
+#ifdef TARGET_X86_64
+ env->regs[R_R8] = state->gprs[NVMM_X64_GPR_R8];
+ env->regs[R_R9] = state->gprs[NVMM_X64_GPR_R9];
+ env->regs[R_R10] = state->gprs[NVMM_X64_GPR_R10];
+ env->regs[R_R11] = state->gprs[NVMM_X64_GPR_R11];
+ env->regs[R_R12] = state->gprs[NVMM_X64_GPR_R12];
+ env->regs[R_R13] = state->gprs[NVMM_X64_GPR_R13];
+ env->regs[R_R14] = state->gprs[NVMM_X64_GPR_R14];
+ env->regs[R_R15] = state->gprs[NVMM_X64_GPR_R15];
+#endif
+
+ /* RIP and RFLAGS. */
+ env->eip = state->gprs[NVMM_X64_GPR_RIP];
+ env->eflags = state->gprs[NVMM_X64_GPR_RFLAGS];
+
+ /* Segments. */
+ nvmm_get_segment(&env->segs[R_ES], &state->segs[NVMM_X64_SEG_ES]);
+ nvmm_get_segment(&env->segs[R_CS], &state->segs[NVMM_X64_SEG_CS]);
+ nvmm_get_segment(&env->segs[R_SS], &state->segs[NVMM_X64_SEG_SS]);
+ nvmm_get_segment(&env->segs[R_DS], &state->segs[NVMM_X64_SEG_DS]);
+ nvmm_get_segment(&env->segs[R_FS], &state->segs[NVMM_X64_SEG_FS]);
+ nvmm_get_segment(&env->segs[R_GS], &state->segs[NVMM_X64_SEG_GS]);
+
+ /* Special segments. */
+ nvmm_get_segment(&env->gdt, &state->segs[NVMM_X64_SEG_GDT]);
+ nvmm_get_segment(&env->ldt, &state->segs[NVMM_X64_SEG_LDT]);
+ nvmm_get_segment(&env->tr, &state->segs[NVMM_X64_SEG_TR]);
+ nvmm_get_segment(&env->idt, &state->segs[NVMM_X64_SEG_IDT]);
+
+ /* Control registers. */
+ env->cr[0] = state->crs[NVMM_X64_CR_CR0];
+ env->cr[2] = state->crs[NVMM_X64_CR_CR2];
+ env->cr[3] = state->crs[NVMM_X64_CR_CR3];
+ env->cr[4] = state->crs[NVMM_X64_CR_CR4];
+ tpr = state->crs[NVMM_X64_CR_CR8];
+ if (tpr != qcpu->tpr) {
+ qcpu->tpr = tpr;
+ cpu_set_apic_tpr(x86_cpu->apic_state, tpr);
+ }
+ env->xcr0 = state->crs[NVMM_X64_CR_XCR0];
+
+ /* Debug registers. */
+ env->dr[0] = state->drs[NVMM_X64_DR_DR0];
+ env->dr[1] = state->drs[NVMM_X64_DR_DR1];
+ env->dr[2] = state->drs[NVMM_X64_DR_DR2];
+ env->dr[3] = state->drs[NVMM_X64_DR_DR3];
+ env->dr[6] = state->drs[NVMM_X64_DR_DR6];
+ env->dr[7] = state->drs[NVMM_X64_DR_DR7];
+
+ /* FPU. */
+ env->fpuc = state->fpu.fx_cw;
+ env->fpstt = (state->fpu.fx_sw >> 11) & 0x7;
+ env->fpus = state->fpu.fx_sw & ~0x3800;
+ for (i = 0; i < 8; i++) {
+ env->fptags[i] = !((state->fpu.fx_tw >> i) & 1);
+ }
+ env->fpop = state->fpu.fx_opcode;
+ env->fpip = state->fpu.fx_ip.fa_64;
+ env->fpdp = state->fpu.fx_dp.fa_64;
+ env->mxcsr = state->fpu.fx_mxcsr;
+ assert(sizeof(state->fpu.fx_87_ac) == sizeof(env->fpregs));
+ memcpy(env->fpregs, state->fpu.fx_87_ac, sizeof(env->fpregs));
+ for (i = 0; i < 16; i++) {
+ memcpy(&env->xmm_regs[i].ZMM_Q(0),
+ &state->fpu.fx_xmm[i].xmm_bytes[0], 8);
+ memcpy(&env->xmm_regs[i].ZMM_Q(1),
+ &state->fpu.fx_xmm[i].xmm_bytes[8], 8);
+ }
+
+ /* MSRs. */
+ env->efer = state->msrs[NVMM_X64_MSR_EFER];
+ env->star = state->msrs[NVMM_X64_MSR_STAR];
+#ifdef TARGET_X86_64
+ env->lstar = state->msrs[NVMM_X64_MSR_LSTAR];
+ env->cstar = state->msrs[NVMM_X64_MSR_CSTAR];
+ env->fmask = state->msrs[NVMM_X64_MSR_SFMASK];
+ env->kernelgsbase = state->msrs[NVMM_X64_MSR_KERNELGSBASE];
+#endif
+ env->sysenter_cs = state->msrs[NVMM_X64_MSR_SYSENTER_CS];
+ env->sysenter_esp = state->msrs[NVMM_X64_MSR_SYSENTER_ESP];
+ env->sysenter_eip = state->msrs[NVMM_X64_MSR_SYSENTER_EIP];
+ env->pat = state->msrs[NVMM_X64_MSR_PAT];
+ env->tsc = state->msrs[NVMM_X64_MSR_TSC];
+
+ x86_update_hflags(env);
+}
+
+static bool
+nvmm_can_take_int(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ struct nvmm_machine *mach = get_nvmm_mach();
+
+ if (qcpu->int_window_exit) {
+ return false;
+ }
+
+ if (qcpu->int_shadow || !(env->eflags & IF_MASK)) {
+ struct nvmm_x64_state *state = vcpu->state;
+
+ /* Exit on interrupt window. */
+ nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_INTR);
+ state->intr.int_window_exiting = 1;
+ nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_INTR);
+
+ return false;
+ }
+
+ return true;
+}
+
+static bool
+nvmm_can_take_nmi(CPUState *cpu)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+
+ /*
+ * Contrary to INTs, NMIs always schedule an exit when they are
+ * completed. Therefore, if window-exiting is enabled, it means
+ * NMIs are blocked.
+ */
+ if (qcpu->nmi_window_exit) {
+ return false;
+ }
+
+ return true;
+}
+
+/*
+ * Called before the VCPU is run. We inject events generated by the I/O
+ * thread, and synchronize the guest TPR.
+ */
+static void
+nvmm_vcpu_pre_run(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ struct nvmm_vcpu_event *event = vcpu->event;
+ bool has_event = false;
+ bool sync_tpr = false;
+ uint8_t tpr;
+ int ret;
+
+ qemu_mutex_lock_iothread();
+
+ tpr = cpu_get_apic_tpr(x86_cpu->apic_state);
+ if (tpr != qcpu->tpr) {
+ qcpu->tpr = tpr;
+ sync_tpr = true;
+ }
+
+ /*
+ * Force the VCPU out of its inner loop to process any INIT requests
+ * or commit pending TPR access.
+ */
+ if (cpu->interrupt_request & (CPU_INTERRUPT_INIT | CPU_INTERRUPT_TPR)) {
+ cpu->exit_request = 1;
+ }
+
+ if (!has_event && (cpu->interrupt_request & CPU_INTERRUPT_NMI)) {
+ if (nvmm_can_take_nmi(cpu)) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_NMI;
+ event->type = NVMM_VCPU_EVENT_INTR;
+ event->vector = 2;
+ has_event = true;
+ }
+ }
+
+ if (!has_event && (cpu->interrupt_request & CPU_INTERRUPT_HARD)) {
+ if (nvmm_can_take_int(cpu)) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_HARD;
+ event->type = NVMM_VCPU_EVENT_INTR;
+ event->vector = cpu_get_pic_interrupt(env);
+ has_event = true;
+ }
+ }
+
+ /* Don't want SMIs. */
+ if (cpu->interrupt_request & CPU_INTERRUPT_SMI) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_SMI;
+ }
+
+ if (sync_tpr) {
+ ret = nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_CRS);
+ if (ret == -1) {
+ error_report("NVMM: Failed to get CPU state,"
+ " error=%d", errno);
+ }
+
+ state->crs[NVMM_X64_CR_CR8] = qcpu->tpr;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_CRS);
+ if (ret == -1) {
+ error_report("NVMM: Failed to set CPU state,"
+ " error=%d", errno);
+ }
+ }
+
+ if (has_event) {
+ ret = nvmm_vcpu_inject(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: Failed to inject event,"
+ " error=%d", errno);
+ }
+ }
+
+ qemu_mutex_unlock_iothread();
+}
+
+/*
+ * Called after the VCPU ran. We synchronize the host view of the TPR and
+ * RFLAGS.
+ */
+static void
+nvmm_vcpu_post_run(CPUState *cpu, struct nvmm_vcpu_exit *exit)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ uint64_t tpr;
+
+ env->eflags = exit->exitstate.rflags;
+ qcpu->int_shadow = exit->exitstate.int_shadow;
+ qcpu->int_window_exit = exit->exitstate.int_window_exiting;
+ qcpu->nmi_window_exit = exit->exitstate.nmi_window_exiting;
+
+ tpr = exit->exitstate.cr8;
+ if (qcpu->tpr != tpr) {
+ qcpu->tpr = tpr;
+ qemu_mutex_lock_iothread();
+ cpu_set_apic_tpr(x86_cpu->apic_state, qcpu->tpr);
+ qemu_mutex_unlock_iothread();
+ }
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_io_callback(struct nvmm_io *io)
+{
+ MemTxAttrs attrs = { 0 };
+ int ret;
+
+ ret = address_space_rw(&address_space_io, io->port, attrs, io->data,
+ io->size, !io->in);
+ if (ret != MEMTX_OK) {
+ error_report("NVMM: I/O Transaction Failed "
+ "[%s, port=%u, size=%zu]", (io->in ? "in" : "out"),
+ io->port, io->size);
+ }
+
+ /* Needed, otherwise infinite loop. */
+ current_cpu->vcpu_dirty = false;
+}
+
+static void
+nvmm_mem_callback(struct nvmm_mem *mem)
+{
+ cpu_physical_memory_rw(mem->gpa, mem->data, mem->size, mem->write);
+
+ /* XXX Needed, otherwise infinite loop. */
+ current_cpu->vcpu_dirty = false;
+}
+
+static struct nvmm_assist_callbacks nvmm_callbacks = {
+ .io = nvmm_io_callback,
+ .mem = nvmm_mem_callback
+};
+
+/* -------------------------------------------------------------------------- */
+
+static int
+nvmm_handle_mem(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
+{
+ int ret;
+
+ ret = nvmm_assist_mem(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: Mem Assist Failed [gpa=%p]",
+ (void *)vcpu->exit->u.mem.gpa);
+ }
+
+ return ret;
+}
+
+static int
+nvmm_handle_io(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
+{
+ int ret;
+
+ ret = nvmm_assist_io(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: I/O Assist Failed [port=%d]",
+ (int)vcpu->exit->u.io.port);
+ }
+
+ return ret;
+}
+
+static int
+nvmm_handle_rdmsr(struct nvmm_machine *mach, CPUState *cpu,
+ struct nvmm_vcpu_exit *exit)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t val;
+ int ret;
+
+ switch (exit->u.rdmsr.msr) {
+ case MSR_IA32_APICBASE:
+ val = cpu_get_apic_base(x86_cpu->apic_state);
+ break;
+ case MSR_MTRRcap:
+ case MSR_MTRRdefType:
+ case MSR_MCG_CAP:
+ case MSR_MCG_STATUS:
+ val = 0;
+ break;
+ default: /* More MSRs to add? */
+ val = 0;
+ error_report("NVMM: Unexpected RDMSR 0x%x, ignored",
+ exit->u.rdmsr.msr);
+ break;
+ }
+
+ ret = nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ state->gprs[NVMM_X64_GPR_RAX] = (val & 0xFFFFFFFF);
+ state->gprs[NVMM_X64_GPR_RDX] = (val >> 32);
+ state->gprs[NVMM_X64_GPR_RIP] = exit->u.rdmsr.npc;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ return 0;
+}
+
+static int
+nvmm_handle_wrmsr(struct nvmm_machine *mach, CPUState *cpu,
+ struct nvmm_vcpu_exit *exit)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t val;
+ int ret;
+
+ val = exit->u.wrmsr.val;
+
+ switch (exit->u.wrmsr.msr) {
+ case MSR_IA32_APICBASE:
+ cpu_set_apic_base(x86_cpu->apic_state, val);
+ break;
+ case MSR_MTRRdefType:
+ case MSR_MCG_STATUS:
+ break;
+ default: /* More MSRs to add? */
+ error_report("NVMM: Unexpected WRMSR 0x%x [val=0x%lx], ignored",
+ exit->u.wrmsr.msr, val);
+ break;
+ }
+
+ ret = nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ state->gprs[NVMM_X64_GPR_RIP] = exit->u.wrmsr.npc;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ return 0;
+}
+
+static int
+nvmm_handle_halted(struct nvmm_machine *mach, CPUState *cpu,
+ struct nvmm_vcpu_exit *exit)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ int ret = 0;
+
+ qemu_mutex_lock_iothread();
+
+ if (!((cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
+ (env->eflags & IF_MASK)) &&
+ !(cpu->interrupt_request & CPU_INTERRUPT_NMI)) {
+ cpu->exception_index = EXCP_HLT;
+ cpu->halted = true;
+ ret = 1;
+ }
+
+ qemu_mutex_unlock_iothread();
+
+ return ret;
+}
+
+static int
+nvmm_inject_ud(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
+{
+ struct nvmm_vcpu_event *event = vcpu->event;
+
+ event->type = NVMM_VCPU_EVENT_EXCP;
+ event->vector = 6;
+ event->u.excp.error = 0;
+
+ return nvmm_vcpu_inject(mach, vcpu);
+}
+
+static int
+nvmm_vcpu_loop(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_vcpu_exit *exit = vcpu->exit;
+ int ret;
+
+ /*
+ * Some asynchronous events must be handled outside of the inner
+ * VCPU loop. They are handled here.
+ */
+ if (cpu->interrupt_request & CPU_INTERRUPT_INIT) {
+ nvmm_cpu_synchronize_state(cpu);
+ do_cpu_init(x86_cpu);
+ /* set int/nmi windows back to the reset state */
+ }
+ if (cpu->interrupt_request & CPU_INTERRUPT_POLL) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_POLL;
+ apic_poll_irq(x86_cpu->apic_state);
+ }
+ if (((cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
+ (env->eflags & IF_MASK)) ||
+ (cpu->interrupt_request & CPU_INTERRUPT_NMI)) {
+ cpu->halted = false;
+ }
+ if (cpu->interrupt_request & CPU_INTERRUPT_SIPI) {
+ nvmm_cpu_synchronize_state(cpu);
+ do_cpu_sipi(x86_cpu);
+ }
+ if (cpu->interrupt_request & CPU_INTERRUPT_TPR) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_TPR;
+ nvmm_cpu_synchronize_state(cpu);
+ apic_handle_tpr_access_report(x86_cpu->apic_state, env->eip,
+ env->tpr_access_type);
+ }
+
+ if (cpu->halted) {
+ cpu->exception_index = EXCP_HLT;
+ atomic_set(&cpu->exit_request, false);
+ return 0;
+ }
+
+ qemu_mutex_unlock_iothread();
+ cpu_exec_start(cpu);
+
+ /*
+ * Inner VCPU loop.
+ */
+ do {
+ if (cpu->vcpu_dirty) {
+ nvmm_set_registers(cpu);
+ cpu->vcpu_dirty = false;
+ }
+
+ if (qcpu->stop) {
+ cpu->exception_index = EXCP_INTERRUPT;
+ qcpu->stop = false;
+ ret = 1;
+ break;
+ }
+
+ nvmm_vcpu_pre_run(cpu);
+
+ if (atomic_read(&cpu->exit_request)) {
+ qemu_cpu_kick_self();
+ }
+
+ ret = nvmm_vcpu_run(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: Failed to exec a virtual processor,"
+ " error=%d", errno);
+ break;
+ }
+
+ nvmm_vcpu_post_run(cpu, exit);
+
+ switch (exit->reason) {
+ case NVMM_VCPU_EXIT_NONE:
+ break;
+ case NVMM_VCPU_EXIT_MEMORY:
+ ret = nvmm_handle_mem(mach, vcpu);
+ break;
+ case NVMM_VCPU_EXIT_IO:
+ ret = nvmm_handle_io(mach, vcpu);
+ break;
+ case NVMM_VCPU_EXIT_INT_READY:
+ case NVMM_VCPU_EXIT_NMI_READY:
+ case NVMM_VCPU_EXIT_TPR_CHANGED:
+ break;
+ case NVMM_VCPU_EXIT_HALTED:
+ ret = nvmm_handle_halted(mach, cpu, exit);
+ break;
+ case NVMM_VCPU_EXIT_SHUTDOWN:
+ qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
+ cpu->exception_index = EXCP_INTERRUPT;
+ ret = 1;
+ break;
+ case NVMM_VCPU_EXIT_RDMSR:
+ ret = nvmm_handle_rdmsr(mach, cpu, exit);
+ break;
+ case NVMM_VCPU_EXIT_WRMSR:
+ ret = nvmm_handle_wrmsr(mach, cpu, exit);
+ break;
+ case NVMM_VCPU_EXIT_MONITOR:
+ case NVMM_VCPU_EXIT_MWAIT:
+ ret = nvmm_inject_ud(mach, vcpu);
+ break;
+ default:
+ error_report("NVMM: Unexpected VM exit code 0x%lx [hw=0x%lx]",
+ exit->reason, exit->u.inv.hwcode);
+ nvmm_get_registers(cpu);
+ qemu_mutex_lock_iothread();
+ qemu_system_guest_panicked(cpu_get_crash_info(cpu));
+ qemu_mutex_unlock_iothread();
+ ret = -1;
+ break;
+ }
+ } while (ret == 0);
+
+ cpu_exec_end(cpu);
+ qemu_mutex_lock_iothread();
+ current_cpu = cpu;
+
+ atomic_set(&cpu->exit_request, false);
+
+ return ret < 0;
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+do_nvmm_cpu_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
+{
+ nvmm_get_registers(cpu);
+ cpu->vcpu_dirty = true;
+}
+
+static void
+do_nvmm_cpu_synchronize_post_reset(CPUState *cpu, run_on_cpu_data arg)
+{
+ nvmm_set_registers(cpu);
+ cpu->vcpu_dirty = false;
+}
+
+static void
+do_nvmm_cpu_synchronize_post_init(CPUState *cpu, run_on_cpu_data arg)
+{
+ nvmm_set_registers(cpu);
+ cpu->vcpu_dirty = false;
+}
+
+static void
+do_nvmm_cpu_synchronize_pre_loadvm(CPUState *cpu, run_on_cpu_data arg)
+{
+ cpu->vcpu_dirty = true;
+}
+
+void nvmm_cpu_synchronize_state(CPUState *cpu)
+{
+ if (!cpu->vcpu_dirty) {
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_state, RUN_ON_CPU_NULL);
+ }
+}
+
+void nvmm_cpu_synchronize_post_reset(CPUState *cpu)
+{
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_post_reset, RUN_ON_CPU_NULL);
+}
+
+void nvmm_cpu_synchronize_post_init(CPUState *cpu)
+{
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_post_init, RUN_ON_CPU_NULL);
+}
+
+void nvmm_cpu_synchronize_pre_loadvm(CPUState *cpu)
+{
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_pre_loadvm, RUN_ON_CPU_NULL);
+}
+
+/* -------------------------------------------------------------------------- */
+
+static Error *nvmm_migration_blocker;
+
+static void
+nvmm_ipi_signal(int sigcpu)
+{
+ struct qemu_vcpu *qcpu;
+
+ if (current_cpu) {
+ qcpu = get_qemu_vcpu(current_cpu);
+ qcpu->stop = true;
+ }
+}
+
+static void
+nvmm_init_cpu_signals(void)
+{
+ struct sigaction sigact;
+ sigset_t set;
+
+ /* Install the IPI handler. */
+ memset(&sigact, 0, sizeof(sigact));
+ sigact.sa_handler = nvmm_ipi_signal;
+ sigaction(SIG_IPI, &sigact, NULL);
+
+ /* Allow IPIs on the current thread. */
+ sigprocmask(SIG_BLOCK, NULL, &set);
+ sigdelset(&set, SIG_IPI);
+ pthread_sigmask(SIG_SETMASK, &set, NULL);
+}
+
+int
+nvmm_init_vcpu(CPUState *cpu)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct nvmm_vcpu_conf_cpuid cpuid;
+ struct nvmm_vcpu_conf_tpr tpr;
+ Error *local_error = NULL;
+ struct qemu_vcpu *qcpu;
+ int ret, err;
+
+ nvmm_init_cpu_signals();
+
+ if (nvmm_migration_blocker == NULL) {
+ error_setg(&nvmm_migration_blocker,
+ "NVMM: Migration not supported");
+
+ (void)migrate_add_blocker(nvmm_migration_blocker, &local_error);
+ if (local_error) {
+ error_report_err(local_error);
+ migrate_del_blocker(nvmm_migration_blocker);
+ error_free(nvmm_migration_blocker);
+ return -EINVAL;
+ }
+ }
+
+ qcpu = g_malloc0(sizeof(*qcpu));
+ if (qcpu == NULL) {
+ error_report("NVMM: Failed to allocate VCPU context.");
+ return -ENOMEM;
+ }
+
+ ret = nvmm_vcpu_create(mach, cpu->cpu_index, &qcpu->vcpu);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to create a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+
+ memset(&cpuid, 0, sizeof(cpuid));
+ cpuid.mask = 1;
+ cpuid.leaf = 0x00000001;
+ cpuid.u.mask.set.edx = CPUID_MCE | CPUID_MCA | CPUID_MTRR;
+ ret = nvmm_vcpu_configure(mach, &qcpu->vcpu, NVMM_VCPU_CONF_CPUID,
+ &cpuid);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to configure a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+
+ ret = nvmm_vcpu_configure(mach, &qcpu->vcpu, NVMM_VCPU_CONF_CALLBACKS,
+ &nvmm_callbacks);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to configure a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+
+ if (qemu_mach.cap.arch.vcpu_conf_support & NVMM_CAP_ARCH_VCPU_CONF_TPR) {
+ memset(&tpr, 0, sizeof(tpr));
+ tpr.exit_changed = 1;
+ ret = nvmm_vcpu_configure(mach, &qcpu->vcpu, NVMM_VCPU_CONF_TPR, &tpr);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to configure a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+ }
+
+ cpu->vcpu_dirty = true;
+ cpu->hax_vcpu = (struct hax_vcpu_state *)qcpu;
+
+ return 0;
+}
+
+int
+nvmm_vcpu_exec(CPUState *cpu)
+{
+ int ret, fatal;
+
+ while (1) {
+ if (cpu->exception_index >= EXCP_INTERRUPT) {
+ ret = cpu->exception_index;
+ cpu->exception_index = -1;
+ break;
+ }
+
+ fatal = nvmm_vcpu_loop(cpu);
+
+ if (fatal) {
+ error_report("NVMM: Failed to execute a VCPU.");
+ abort();
+ }
+ }
+
+ return ret;
+}
+
+void
+nvmm_destroy_vcpu(CPUState *cpu)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+
+ nvmm_vcpu_destroy(mach, &qcpu->vcpu);
+ g_free(cpu->hax_vcpu);
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_update_mapping(hwaddr start_pa, ram_addr_t size, uintptr_t hva,
+ bool add, bool rom, const char *name)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ int ret, prot;
+
+ if (add) {
+ prot = PROT_READ | PROT_EXEC;
+ if (!rom) {
+ prot |= PROT_WRITE;
+ }
+ ret = nvmm_gpa_map(mach, hva, start_pa, size, prot);
+ } else {
+ ret = nvmm_gpa_unmap(mach, hva, start_pa, size);
+ }
+
+ if (ret == -1) {
+ error_report("NVMM: Failed to %s GPA range '%s' PA:%p, "
+ "Size:%p bytes, HostVA:%p, error=%d",
+ (add ? "map" : "unmap"), name, (void *)(uintptr_t)start_pa,
+ (void *)size, (void *)hva, errno);
+ }
+}
+
+static void
+nvmm_process_section(MemoryRegionSection *section, int add)
+{
+ MemoryRegion *mr = section->mr;
+ hwaddr start_pa = section->offset_within_address_space;
+ ram_addr_t size = int128_get64(section->size);
+ unsigned int delta;
+ uintptr_t hva;
+
+ if (!memory_region_is_ram(mr)) {
+ return;
+ }
+
+ /* Adjust start_pa and size so that they are page-aligned. */
+ delta = qemu_real_host_page_size - (start_pa & ~qemu_real_host_page_mask);
+ delta &= ~qemu_real_host_page_mask;
+ if (delta > size) {
+ return;
+ }
+ start_pa += delta;
+ size -= delta;
+ size &= qemu_real_host_page_mask;
+ if (!size || (start_pa & ~qemu_real_host_page_mask)) {
+ return;
+ }
+
+ hva = (uintptr_t)memory_region_get_ram_ptr(mr) +
+ section->offset_within_region + delta;
+
+ nvmm_update_mapping(start_pa, size, hva, add,
+ memory_region_is_rom(mr), mr->name);
+}
+
+static void
+nvmm_region_add(MemoryListener *listener, MemoryRegionSection *section)
+{
+ memory_region_ref(section->mr);
+ nvmm_process_section(section, 1);
+}
+
+static void
+nvmm_region_del(MemoryListener *listener, MemoryRegionSection *section)
+{
+ nvmm_process_section(section, 0);
+ memory_region_unref(section->mr);
+}
+
+static void
+nvmm_transaction_begin(MemoryListener *listener)
+{
+ /* nothing */
+}
+
+static void
+nvmm_transaction_commit(MemoryListener *listener)
+{
+ /* nothing */
+}
+
+static void
+nvmm_log_sync(MemoryListener *listener, MemoryRegionSection *section)
+{
+ MemoryRegion *mr = section->mr;
+
+ if (!memory_region_is_ram(mr)) {
+ return;
+ }
+
+ memory_region_set_dirty(mr, 0, int128_get64(section->size));
+}
+
+static MemoryListener nvmm_memory_listener = {
+ .begin = nvmm_transaction_begin,
+ .commit = nvmm_transaction_commit,
+ .region_add = nvmm_region_add,
+ .region_del = nvmm_region_del,
+ .log_sync = nvmm_log_sync,
+ .priority = 10,
+};
+
+static void
+nvmm_ram_block_added(RAMBlockNotifier *n, void *host, size_t size)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ uintptr_t hva = (uintptr_t)host;
+ int ret;
+
+ ret = nvmm_hva_map(mach, hva, size);
+
+ if (ret == -1) {
+ error_report("NVMM: Failed to map HVA, HostVA:%p "
+ "Size:%p bytes, error=%d",
+ (void *)hva, (void *)size, errno);
+ }
+}
+
+static struct RAMBlockNotifier nvmm_ram_notifier = {
+ .ram_block_added = nvmm_ram_block_added
+};
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_handle_interrupt(CPUState *cpu, int mask)
+{
+ cpu->interrupt_request |= mask;
+
+ if (!qemu_cpu_is_self(cpu)) {
+ qemu_cpu_kick(cpu);
+ }
+}
+
+/* -------------------------------------------------------------------------- */
+
+static int
+nvmm_accel_init(MachineState *ms)
+{
+ int ret, err;
+
+ ret = nvmm_init();
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Initialization failed, error=%d", errno);
+ return -err;
+ }
+
+ ret = nvmm_capability(&qemu_mach.cap);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Unable to fetch capability, error=%d", errno);
+ return -err;
+ }
+ if (qemu_mach.cap.version != NVMM_KERN_VERSION) {
+ error_report("NVMM: Unsupported version %u", qemu_mach.cap.version);
+ return -EPROGMISMATCH;
+ }
+ if (qemu_mach.cap.state_size != sizeof(struct nvmm_x64_state)) {
+ error_report("NVMM: Wrong state size %u", qemu_mach.cap.state_size);
+ return -EPROGMISMATCH;
+ }
+
+ ret = nvmm_machine_create(&qemu_mach.mach);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Machine creation failed, error=%d", errno);
+ return -err;
+ }
+
+ memory_listener_register(&nvmm_memory_listener, &address_space_memory);
+ ram_block_notifier_add(&nvmm_ram_notifier);
+
+ cpu_interrupt_handler = nvmm_handle_interrupt;
+
+ printf("NetBSD Virtual Machine Monitor accelerator is operational\n");
+ return 0;
+}
+
+int
+nvmm_enabled(void)
+{
+ return nvmm_allowed;
+}
+
+static void
+nvmm_accel_class_init(ObjectClass *oc, void *data)
+{
+ AccelClass *ac = ACCEL_CLASS(oc);
+ ac->name = "NVMM";
+ ac->init_machine = nvmm_accel_init;
+ ac->allowed = &nvmm_allowed;
+}
+
+static const TypeInfo nvmm_accel_type = {
+ .name = ACCEL_CLASS_NAME("nvmm"),
+ .parent = TYPE_ACCEL,
+ .class_init = nvmm_accel_class_init,
+};
+
+static void
+nvmm_type_init(void)
+{
+ type_register_static(&nvmm_accel_type);
+}
+
+type_init(nvmm_type_init);
--
2.24.1
From: Maxime Villard <max@m00nbsd.net>
Implements the NVMM accelerator cpu enlightenments to actually use the nvmm-all
accelerator on NetBSD platforms.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
include/sysemu/hw_accel.h | 14 ++++++++++
softmmu/cpus.c | 58 +++++++++++++++++++++++++++++++++++++++
target/i386/helper.c | 2 +-
3 files changed, 73 insertions(+), 1 deletion(-)
diff --git a/include/sysemu/hw_accel.h b/include/sysemu/hw_accel.h
index e128f8b06b..9e19f5794c 100644
--- a/include/sysemu/hw_accel.h
+++ b/include/sysemu/hw_accel.h
@@ -16,6 +16,7 @@
#include "sysemu/kvm.h"
#include "sysemu/hvf.h"
#include "sysemu/whpx.h"
+#include "sysemu/nvmm.h"
static inline void cpu_synchronize_state(CPUState *cpu)
{
@@ -31,6 +32,9 @@ static inline void cpu_synchronize_state(CPUState *cpu)
if (whpx_enabled()) {
whpx_cpu_synchronize_state(cpu);
}
+ if (nvmm_enabled()) {
+ nvmm_cpu_synchronize_state(cpu);
+ }
}
static inline void cpu_synchronize_post_reset(CPUState *cpu)
@@ -47,6 +51,10 @@ static inline void cpu_synchronize_post_reset(CPUState *cpu)
if (whpx_enabled()) {
whpx_cpu_synchronize_post_reset(cpu);
}
+ if (nvmm_enabled()) {
+ nvmm_cpu_synchronize_post_reset(cpu);
+ }
+
}
static inline void cpu_synchronize_post_init(CPUState *cpu)
@@ -63,6 +71,9 @@ static inline void cpu_synchronize_post_init(CPUState *cpu)
if (whpx_enabled()) {
whpx_cpu_synchronize_post_init(cpu);
}
+ if (nvmm_enabled()) {
+ nvmm_cpu_synchronize_post_init(cpu);
+ }
}
static inline void cpu_synchronize_pre_loadvm(CPUState *cpu)
@@ -79,6 +90,9 @@ static inline void cpu_synchronize_pre_loadvm(CPUState *cpu)
if (whpx_enabled()) {
whpx_cpu_synchronize_pre_loadvm(cpu);
}
+ if (nvmm_enabled()) {
+ nvmm_cpu_synchronize_pre_loadvm(cpu);
+ }
}
#endif /* QEMU_HW_ACCEL_H */
diff --git a/softmmu/cpus.c b/softmmu/cpus.c
index a802e899ab..3b44b92830 100644
--- a/softmmu/cpus.c
+++ b/softmmu/cpus.c
@@ -43,6 +43,7 @@
#include "sysemu/hax.h"
#include "sysemu/hvf.h"
#include "sysemu/whpx.h"
+#include "sysemu/nvmm.h"
#include "exec/exec-all.h"
#include "qemu/thread.h"
@@ -1621,6 +1622,48 @@ static void *qemu_whpx_cpu_thread_fn(void *arg)
return NULL;
}
+static void *qemu_nvmm_cpu_thread_fn(void *arg)
+{
+ CPUState *cpu = arg;
+ int r;
+
+ assert(nvmm_enabled());
+
+ rcu_register_thread();
+
+ qemu_mutex_lock_iothread();
+ qemu_thread_get_self(cpu->thread);
+ cpu->thread_id = qemu_get_thread_id();
+ current_cpu = cpu;
+
+ r = nvmm_init_vcpu(cpu);
+ if (r < 0) {
+ fprintf(stderr, "nvmm_init_vcpu failed: %s\n", strerror(-r));
+ exit(1);
+ }
+
+ /* signal CPU creation */
+ cpu->created = true;
+ qemu_cond_signal(&qemu_cpu_cond);
+
+ do {
+ if (cpu_can_run(cpu)) {
+ r = nvmm_vcpu_exec(cpu);
+ if (r == EXCP_DEBUG) {
+ cpu_handle_guest_debug(cpu);
+ }
+ }
+ qemu_wait_io_event(cpu);
+ } while (!cpu->unplug || cpu_can_run(cpu));
+
+ nvmm_destroy_vcpu(cpu);
+ cpu->created = false;
+ qemu_cond_signal(&qemu_cpu_cond);
+ qemu_mutex_unlock_iothread();
+ rcu_unregister_thread();
+ return NULL;
+}
+
#ifdef _WIN32
static void CALLBACK dummy_apc_func(ULONG_PTR unused)
{
@@ -1998,6 +2041,19 @@ static void qemu_whpx_start_vcpu(CPUState *cpu)
#endif
}
+static void qemu_nvmm_start_vcpu(CPUState *cpu)
+{
+ char thread_name[VCPU_THREAD_NAME_SIZE];
+
+ cpu->thread = g_malloc0(sizeof(QemuThread));
+ cpu->halt_cond = g_malloc0(sizeof(QemuCond));
+ qemu_cond_init(cpu->halt_cond);
+ snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/NVMM",
+ cpu->cpu_index);
+ qemu_thread_create(cpu->thread, thread_name, qemu_nvmm_cpu_thread_fn,
+ cpu, QEMU_THREAD_JOINABLE);
+}
+
static void qemu_dummy_start_vcpu(CPUState *cpu)
{
char thread_name[VCPU_THREAD_NAME_SIZE];
@@ -2038,6 +2094,8 @@ void qemu_init_vcpu(CPUState *cpu)
qemu_tcg_init_vcpu(cpu);
} else if (whpx_enabled()) {
qemu_whpx_start_vcpu(cpu);
+ } else if (nvmm_enabled()) {
+ qemu_nvmm_start_vcpu(cpu);
} else {
qemu_dummy_start_vcpu(cpu);
}
diff --git a/target/i386/helper.c b/target/i386/helper.c
index 70be53e2c3..c2f1aef65c 100644
--- a/target/i386/helper.c
+++ b/target/i386/helper.c
@@ -983,7 +983,7 @@ void cpu_report_tpr_access(CPUX86State *env, TPRAccess access)
X86CPU *cpu = env_archcpu(env);
CPUState *cs = env_cpu(env);
- if (kvm_enabled() || whpx_enabled()) {
+ if (kvm_enabled() || whpx_enabled() || nvmm_enabled()) {
env->tpr_access_type = access;
cpu_interrupt(cs, CPU_INTERRUPT_TPR);
--
2.24.1
From: Maxime Villard <max@m00nbsd.net>
Adds support for the NetBSD Virtual Machine Monitor (NVMM) stubs and
introduces the nvmm.h sysemu API for managing the vcpu scheduling and
management.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
accel/stubs/Makefile.objs | 1 +
accel/stubs/nvmm-stub.c | 43 +++++++++++++++++++++++++++++++++++++++
include/sysemu/nvmm.h | 35 +++++++++++++++++++++++++++++++
3 files changed, 79 insertions(+)
create mode 100644 accel/stubs/nvmm-stub.c
create mode 100644 include/sysemu/nvmm.h
diff --git a/accel/stubs/Makefile.objs b/accel/stubs/Makefile.objs
index bbd14e71fb..38660a0b9b 100644
--- a/accel/stubs/Makefile.objs
+++ b/accel/stubs/Makefile.objs
@@ -1,6 +1,7 @@
obj-$(call lnot,$(CONFIG_HAX)) += hax-stub.o
obj-$(call lnot,$(CONFIG_HVF)) += hvf-stub.o
obj-$(call lnot,$(CONFIG_WHPX)) += whpx-stub.o
+obj-$(call lnot,$(CONFIG_NVMM)) += nvmm-stub.o
obj-$(call lnot,$(CONFIG_KVM)) += kvm-stub.o
obj-$(call lnot,$(CONFIG_TCG)) += tcg-stub.o
obj-$(call lnot,$(CONFIG_XEN)) += xen-stub.o
diff --git a/accel/stubs/nvmm-stub.c b/accel/stubs/nvmm-stub.c
new file mode 100644
index 0000000000..c2208b84a3
--- /dev/null
+++ b/accel/stubs/nvmm-stub.c
@@ -0,0 +1,43 @@
+/*
+ * Copyright (c) 2018-2019 Maxime Villard, All rights reserved.
+ *
+ * NetBSD Virtual Machine Monitor (NVMM) accelerator stub.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu-common.h"
+#include "cpu.h"
+#include "sysemu/nvmm.h"
+
+int nvmm_init_vcpu(CPUState *cpu)
+{
+ return -1;
+}
+
+int nvmm_vcpu_exec(CPUState *cpu)
+{
+ return -1;
+}
+
+void nvmm_destroy_vcpu(CPUState *cpu)
+{
+}
+
+void nvmm_cpu_synchronize_state(CPUState *cpu)
+{
+}
+
+void nvmm_cpu_synchronize_post_reset(CPUState *cpu)
+{
+}
+
+void nvmm_cpu_synchronize_post_init(CPUState *cpu)
+{
+}
+
+void nvmm_cpu_synchronize_pre_loadvm(CPUState *cpu)
+{
+}
diff --git a/include/sysemu/nvmm.h b/include/sysemu/nvmm.h
new file mode 100644
index 0000000000..10496f3980
--- /dev/null
+++ b/include/sysemu/nvmm.h
@@ -0,0 +1,35 @@
+/*
+ * Copyright (c) 2018-2019 Maxime Villard, All rights reserved.
+ *
+ * NetBSD Virtual Machine Monitor (NVMM) accelerator support.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#ifndef QEMU_NVMM_H
+#define QEMU_NVMM_H
+
+#include "config-host.h"
+#include "qemu-common.h"
+
+int nvmm_init_vcpu(CPUState *);
+int nvmm_vcpu_exec(CPUState *);
+void nvmm_destroy_vcpu(CPUState *);
+
+void nvmm_cpu_synchronize_state(CPUState *);
+void nvmm_cpu_synchronize_post_reset(CPUState *);
+void nvmm_cpu_synchronize_post_init(CPUState *);
+void nvmm_cpu_synchronize_pre_loadvm(CPUState *);
+
+#ifdef CONFIG_NVMM
+
+int nvmm_enabled(void);
+
+#else /* CONFIG_NVMM */
+
+#define nvmm_enabled() (0)
+
+#endif /* CONFIG_NVMM */
+
+#endif /* CONFIG_NVMM */
--
2.28.0
Ping?
On 11.08.2020 15:01, Kamil Rytarowski wrote:
> From: Maxime Villard <max@m00nbsd.net>
>
> Adds support for the NetBSD Virtual Machine Monitor (NVMM) stubs and
> introduces the nvmm.h sysemu API for managing the vcpu scheduling and
> management.
>
> Signed-off-by: Maxime Villard <max@m00nbsd.net>
> Signed-off-by: Kamil Rytarowski <n54@gmx.com>
> Reviewed-by: Sergio Lopez <slp@redhat.com>
> Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
> Tested-by: Jared McNeill <jmcneill@invisible.ca>
> ---
> accel/stubs/Makefile.objs | 1 +
> accel/stubs/nvmm-stub.c | 43 +++++++++++++++++++++++++++++++++++++++
> include/sysemu/nvmm.h | 35 +++++++++++++++++++++++++++++++
> 3 files changed, 79 insertions(+)
> create mode 100644 accel/stubs/nvmm-stub.c
> create mode 100644 include/sysemu/nvmm.h
>
> diff --git a/accel/stubs/Makefile.objs b/accel/stubs/Makefile.objs
> index bbd14e71fb..38660a0b9b 100644
> --- a/accel/stubs/Makefile.objs
> +++ b/accel/stubs/Makefile.objs
> @@ -1,6 +1,7 @@
> obj-$(call lnot,$(CONFIG_HAX)) += hax-stub.o
> obj-$(call lnot,$(CONFIG_HVF)) += hvf-stub.o
> obj-$(call lnot,$(CONFIG_WHPX)) += whpx-stub.o
> +obj-$(call lnot,$(CONFIG_NVMM)) += nvmm-stub.o
> obj-$(call lnot,$(CONFIG_KVM)) += kvm-stub.o
> obj-$(call lnot,$(CONFIG_TCG)) += tcg-stub.o
> obj-$(call lnot,$(CONFIG_XEN)) += xen-stub.o
> diff --git a/accel/stubs/nvmm-stub.c b/accel/stubs/nvmm-stub.c
> new file mode 100644
> index 0000000000..c2208b84a3
> --- /dev/null
> +++ b/accel/stubs/nvmm-stub.c
> @@ -0,0 +1,43 @@
> +/*
> + * Copyright (c) 2018-2019 Maxime Villard, All rights reserved.
> + *
> + * NetBSD Virtual Machine Monitor (NVMM) accelerator stub.
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> + * See the COPYING file in the top-level directory.
> + */
> +
> +#include "qemu/osdep.h"
> +#include "qemu-common.h"
> +#include "cpu.h"
> +#include "sysemu/nvmm.h"
> +
> +int nvmm_init_vcpu(CPUState *cpu)
> +{
> + return -1;
> +}
> +
> +int nvmm_vcpu_exec(CPUState *cpu)
> +{
> + return -1;
> +}
> +
> +void nvmm_destroy_vcpu(CPUState *cpu)
> +{
> +}
> +
> +void nvmm_cpu_synchronize_state(CPUState *cpu)
> +{
> +}
> +
> +void nvmm_cpu_synchronize_post_reset(CPUState *cpu)
> +{
> +}
> +
> +void nvmm_cpu_synchronize_post_init(CPUState *cpu)
> +{
> +}
> +
> +void nvmm_cpu_synchronize_pre_loadvm(CPUState *cpu)
> +{
> +}
> diff --git a/include/sysemu/nvmm.h b/include/sysemu/nvmm.h
> new file mode 100644
> index 0000000000..10496f3980
> --- /dev/null
> +++ b/include/sysemu/nvmm.h
> @@ -0,0 +1,35 @@
> +/*
> + * Copyright (c) 2018-2019 Maxime Villard, All rights reserved.
> + *
> + * NetBSD Virtual Machine Monitor (NVMM) accelerator support.
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> + * See the COPYING file in the top-level directory.
> + */
> +
> +#ifndef QEMU_NVMM_H
> +#define QEMU_NVMM_H
> +
> +#include "config-host.h"
> +#include "qemu-common.h"
> +
> +int nvmm_init_vcpu(CPUState *);
> +int nvmm_vcpu_exec(CPUState *);
> +void nvmm_destroy_vcpu(CPUState *);
> +
> +void nvmm_cpu_synchronize_state(CPUState *);
> +void nvmm_cpu_synchronize_post_reset(CPUState *);
> +void nvmm_cpu_synchronize_post_init(CPUState *);
> +void nvmm_cpu_synchronize_pre_loadvm(CPUState *);
> +
> +#ifdef CONFIG_NVMM
> +
> +int nvmm_enabled(void);
> +
> +#else /* CONFIG_NVMM */
> +
> +#define nvmm_enabled() (0)
> +
> +#endif /* CONFIG_NVMM */
> +
> +#endif /* CONFIG_NVMM */
> --
> 2.28.0
>
From: Maxime Villard <max@m00nbsd.net>
Introduces the configure support for the new NetBSD Virtual Machine Monitor that
allows for hypervisor acceleration from usermode components on the NetBSD
platform.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
configure | 37 +++++++++++++++++++++++++++++++++++++
qemu-options.hx | 10 +++++-----
2 files changed, 42 insertions(+), 5 deletions(-)
diff --git a/configure b/configure
index 2acc4d1465..fb9ffba2bf 100755
--- a/configure
+++ b/configure
@@ -246,6 +246,17 @@ supported_whpx_target() {
return 1
}
+supported_nvmm_target() {
+ test "$nvmm" = "yes" || return 1
+ glob "$1" "*-softmmu" || return 1
+ case "${1%-softmmu}" in
+ i386|x86_64)
+ return 0
+ ;;
+ esac
+ return 1
+}
+
supported_target() {
case "$1" in
*-softmmu)
@@ -273,6 +284,7 @@ supported_target() {
supported_hax_target "$1" && return 0
supported_hvf_target "$1" && return 0
supported_whpx_target "$1" && return 0
+ supported_nvmm_target "$1" && return 0
print_error "TCG disabled, but hardware accelerator not available for '$target'"
return 1
}
@@ -395,6 +407,7 @@ kvm="no"
hax="no"
hvf="no"
whpx="no"
+nvmm="no"
rdma=""
pvrdma=""
gprof="no"
@@ -847,6 +860,7 @@ DragonFly)
NetBSD)
bsd="yes"
hax="yes"
+ nvmm="yes"
make="${MAKE-gmake}"
audio_drv_list="oss try-sdl"
audio_possible_drivers="oss sdl"
@@ -1233,6 +1247,10 @@ for opt do
;;
--enable-whpx) whpx="yes"
;;
+ --disable-nvmm) nvmm="no"
+ ;;
+ --enable-nvmm) nvmm="yes"
+ ;;
--disable-tcg-interpreter) tcg_interpreter="no"
;;
--enable-tcg-interpreter) tcg_interpreter="yes"
@@ -1879,6 +1897,7 @@ disabled with --disable-FEATURE, default is enabled if available:
hax HAX acceleration support
hvf Hypervisor.framework acceleration support
whpx Windows Hypervisor Platform acceleration support
+ nvmm NetBSD Virtual Machine Monitor acceleration support
rdma Enable RDMA-based migration
pvrdma Enable PVRDMA support
vde support for vde network
@@ -2965,6 +2984,20 @@ if test "$whpx" != "no" ; then
fi
fi
+##########################################
+# NetBSD Virtual Machine Monitor (NVMM) accelerator check
+if test "$nvmm" != "no" ; then
+ if check_include "nvmm.h" ; then
+ nvmm="yes"
+ LIBS="-lnvmm $LIBS"
+ else
+ if test "$nvmm" = "yes"; then
+ feature_not_found "NVMM" "NVMM is not available"
+ fi
+ nvmm="no"
+ fi
+fi
+
##########################################
# Sparse probe
if test "$sparse" != "no" ; then
@@ -6934,6 +6967,7 @@ echo "KVM support $kvm"
echo "HAX support $hax"
echo "HVF support $hvf"
echo "WHPX support $whpx"
+echo "NVMM support $nvmm"
echo "TCG support $tcg"
if test "$tcg" = "yes" ; then
echo "TCG debug enabled $debug_tcg"
@@ -8332,6 +8366,9 @@ fi
if test "$target_aligned_only" = "yes" ; then
echo "TARGET_ALIGNED_ONLY=y" >> $config_target_mak
fi
+if supported_nvmm_target $target; then
+ echo "CONFIG_NVMM=y" >> $config_target_mak
+fi
if test "$target_bigendian" = "yes" ; then
echo "TARGET_WORDS_BIGENDIAN=y" >> $config_target_mak
fi
diff --git a/qemu-options.hx b/qemu-options.hx
index 708583b4ce..697accaa7e 100644
--- a/qemu-options.hx
+++ b/qemu-options.hx
@@ -26,7 +26,7 @@ DEF("machine", HAS_ARG, QEMU_OPTION_machine, \
"-machine [type=]name[,prop[=value][,...]]\n"
" selects emulated machine ('-machine help' for list)\n"
" property accel=accel1[:accel2[:...]] selects accelerator\n"
- " supported accelerators are kvm, xen, hax, hvf, whpx or tcg (default: tcg)\n"
+ " supported accelerators are kvm, xen, hax, hvf, nvmm, whpx or tcg (default: tcg)\n"
" vmport=on|off|auto controls emulation of vmport (default: auto)\n"
" dump-guest-core=on|off include guest memory in a core dump (default=on)\n"
" mem-merge=on|off controls memory merge support (default: on)\n"
@@ -58,7 +58,7 @@ SRST
``accel=accels1[:accels2[:...]]``
This is used to enable an accelerator. Depending on the target
- architecture, kvm, xen, hax, hvf, whpx or tcg can be available.
+ architecture, kvm, xen, hax, hvf, nvmm, whpx or tcg can be available.
By default, tcg is used. If there is more than one accelerator
specified, the next one is used if the previous one fails to
initialize.
@@ -119,7 +119,7 @@ ERST
DEF("accel", HAS_ARG, QEMU_OPTION_accel,
"-accel [accel=]accelerator[,prop[=value][,...]]\n"
- " select accelerator (kvm, xen, hax, hvf, whpx or tcg; use 'help' for a list)\n"
+ " select accelerator (kvm, xen, hax, hvf, nvmm, whpx or tcg; use 'help' for a list)\n"
" igd-passthru=on|off (enable Xen integrated Intel graphics passthrough, default=off)\n"
" kernel-irqchip=on|off|split controls accelerated irqchip support (default=on)\n"
" kvm-shadow-mem=size of KVM shadow MMU in bytes\n"
@@ -128,8 +128,8 @@ DEF("accel", HAS_ARG, QEMU_OPTION_accel,
SRST
``-accel name[,prop=value[,...]]``
This is used to enable an accelerator. Depending on the target
- architecture, kvm, xen, hax, hvf, whpx or tcg can be available. By
- default, tcg is used. If there is more than one accelerator
+ architecture, kvm, xen, hax, hvf, nvmm whpx or tcg can be available.
+ By default, tcg is used. If there is more than one accelerator
specified, the next one is used if the previous one fails to
initialize.
--
2.28.0
From: Maxime Villard <max@m00nbsd.net>
Implements the NetBSD Virtual Machine Monitor (NVMM) target. Which
acts as a hypervisor accelerator for QEMU on the NetBSD platform. This enables
QEMU much greater speed over the emulated x86_64 path's that are taken on
NetBSD today.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
target/i386/Makefile.objs | 1 +
target/i386/nvmm-all.c | 1226 +++++++++++++++++++++++++++++++++++++
2 files changed, 1227 insertions(+)
create mode 100644 target/i386/nvmm-all.c
diff --git a/target/i386/Makefile.objs b/target/i386/Makefile.objs
index 0b93143e27..ff0df68404 100644
--- a/target/i386/Makefile.objs
+++ b/target/i386/Makefile.objs
@@ -18,6 +18,7 @@ obj-$(CONFIG_HAX) += hax-all.o hax-mem.o hax-posix.o
endif
obj-$(CONFIG_HVF) += hvf/
obj-$(CONFIG_WHPX) += whpx-all.o
+obj-$(CONFIG_NVMM) += nvmm-all.o
endif
obj-$(CONFIG_SEV) += sev.o
obj-$(call lnot,$(CONFIG_SEV)) += sev-stub.o
diff --git a/target/i386/nvmm-all.c b/target/i386/nvmm-all.c
new file mode 100644
index 0000000000..408f7305b9
--- /dev/null
+++ b/target/i386/nvmm-all.c
@@ -0,0 +1,1226 @@
+/*
+ * Copyright (c) 2018-2019 Maxime Villard, All rights reserved.
+ *
+ * NetBSD Virtual Machine Monitor (NVMM) accelerator for QEMU.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "cpu.h"
+#include "exec/address-spaces.h"
+#include "exec/ioport.h"
+#include "qemu-common.h"
+#include "strings.h"
+#include "sysemu/accel.h"
+#include "sysemu/nvmm.h"
+#include "sysemu/runstate.h"
+#include "sysemu/sysemu.h"
+#include "sysemu/cpus.h"
+#include "qemu/main-loop.h"
+#include "qemu/error-report.h"
+#include "qemu/queue.h"
+#include "qapi/error.h"
+#include "migration/blocker.h"
+
+#include <nvmm.h>
+
+struct qemu_vcpu {
+ struct nvmm_vcpu vcpu;
+ uint8_t tpr;
+ bool stop;
+
+ /* Window-exiting for INTs/NMIs. */
+ bool int_window_exit;
+ bool nmi_window_exit;
+
+ /* The guest is in an interrupt shadow (POP SS, etc). */
+ bool int_shadow;
+};
+
+struct qemu_machine {
+ struct nvmm_capability cap;
+ struct nvmm_machine mach;
+};
+
+/* -------------------------------------------------------------------------- */
+
+static bool nvmm_allowed;
+static struct qemu_machine qemu_mach;
+
+static struct qemu_vcpu *
+get_qemu_vcpu(CPUState *cpu)
+{
+ return (struct qemu_vcpu *)cpu->hax_vcpu;
+}
+
+static struct nvmm_machine *
+get_nvmm_mach(void)
+{
+ return &qemu_mach.mach;
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_set_segment(struct nvmm_x64_state_seg *nseg, const SegmentCache *qseg)
+{
+ uint32_t attrib = qseg->flags;
+
+ nseg->selector = qseg->selector;
+ nseg->limit = qseg->limit;
+ nseg->base = qseg->base;
+ nseg->attrib.type = __SHIFTOUT(attrib, DESC_TYPE_MASK);
+ nseg->attrib.s = __SHIFTOUT(attrib, DESC_S_MASK);
+ nseg->attrib.dpl = __SHIFTOUT(attrib, DESC_DPL_MASK);
+ nseg->attrib.p = __SHIFTOUT(attrib, DESC_P_MASK);
+ nseg->attrib.avl = __SHIFTOUT(attrib, DESC_AVL_MASK);
+ nseg->attrib.l = __SHIFTOUT(attrib, DESC_L_MASK);
+ nseg->attrib.def = __SHIFTOUT(attrib, DESC_B_MASK);
+ nseg->attrib.g = __SHIFTOUT(attrib, DESC_G_MASK);
+}
+
+static void
+nvmm_set_registers(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t bitmap;
+ size_t i;
+ int ret;
+
+ assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
+
+ /* GPRs. */
+ state->gprs[NVMM_X64_GPR_RAX] = env->regs[R_EAX];
+ state->gprs[NVMM_X64_GPR_RCX] = env->regs[R_ECX];
+ state->gprs[NVMM_X64_GPR_RDX] = env->regs[R_EDX];
+ state->gprs[NVMM_X64_GPR_RBX] = env->regs[R_EBX];
+ state->gprs[NVMM_X64_GPR_RSP] = env->regs[R_ESP];
+ state->gprs[NVMM_X64_GPR_RBP] = env->regs[R_EBP];
+ state->gprs[NVMM_X64_GPR_RSI] = env->regs[R_ESI];
+ state->gprs[NVMM_X64_GPR_RDI] = env->regs[R_EDI];
+#ifdef TARGET_X86_64
+ state->gprs[NVMM_X64_GPR_R8] = env->regs[R_R8];
+ state->gprs[NVMM_X64_GPR_R9] = env->regs[R_R9];
+ state->gprs[NVMM_X64_GPR_R10] = env->regs[R_R10];
+ state->gprs[NVMM_X64_GPR_R11] = env->regs[R_R11];
+ state->gprs[NVMM_X64_GPR_R12] = env->regs[R_R12];
+ state->gprs[NVMM_X64_GPR_R13] = env->regs[R_R13];
+ state->gprs[NVMM_X64_GPR_R14] = env->regs[R_R14];
+ state->gprs[NVMM_X64_GPR_R15] = env->regs[R_R15];
+#endif
+
+ /* RIP and RFLAGS. */
+ state->gprs[NVMM_X64_GPR_RIP] = env->eip;
+ state->gprs[NVMM_X64_GPR_RFLAGS] = env->eflags;
+
+ /* Segments. */
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_CS], &env->segs[R_CS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_DS], &env->segs[R_DS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_ES], &env->segs[R_ES]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_FS], &env->segs[R_FS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_GS], &env->segs[R_GS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_SS], &env->segs[R_SS]);
+
+ /* Special segments. */
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_GDT], &env->gdt);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_LDT], &env->ldt);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_TR], &env->tr);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_IDT], &env->idt);
+
+ /* Control registers. */
+ state->crs[NVMM_X64_CR_CR0] = env->cr[0];
+ state->crs[NVMM_X64_CR_CR2] = env->cr[2];
+ state->crs[NVMM_X64_CR_CR3] = env->cr[3];
+ state->crs[NVMM_X64_CR_CR4] = env->cr[4];
+ state->crs[NVMM_X64_CR_CR8] = qcpu->tpr;
+ state->crs[NVMM_X64_CR_XCR0] = env->xcr0;
+
+ /* Debug registers. */
+ state->drs[NVMM_X64_DR_DR0] = env->dr[0];
+ state->drs[NVMM_X64_DR_DR1] = env->dr[1];
+ state->drs[NVMM_X64_DR_DR2] = env->dr[2];
+ state->drs[NVMM_X64_DR_DR3] = env->dr[3];
+ state->drs[NVMM_X64_DR_DR6] = env->dr[6];
+ state->drs[NVMM_X64_DR_DR7] = env->dr[7];
+
+ /* FPU. */
+ state->fpu.fx_cw = env->fpuc;
+ state->fpu.fx_sw = (env->fpus & ~0x3800) | ((env->fpstt & 0x7) << 11);
+ state->fpu.fx_tw = 0;
+ for (i = 0; i < 8; i++) {
+ state->fpu.fx_tw |= (!env->fptags[i]) << i;
+ }
+ state->fpu.fx_opcode = env->fpop;
+ state->fpu.fx_ip.fa_64 = env->fpip;
+ state->fpu.fx_dp.fa_64 = env->fpdp;
+ state->fpu.fx_mxcsr = env->mxcsr;
+ state->fpu.fx_mxcsr_mask = 0x0000FFFF;
+ assert(sizeof(state->fpu.fx_87_ac) == sizeof(env->fpregs));
+ memcpy(state->fpu.fx_87_ac, env->fpregs, sizeof(env->fpregs));
+ for (i = 0; i < 16; i++) {
+ memcpy(&state->fpu.fx_xmm[i].xmm_bytes[0],
+ &env->xmm_regs[i].ZMM_Q(0), 8);
+ memcpy(&state->fpu.fx_xmm[i].xmm_bytes[8],
+ &env->xmm_regs[i].ZMM_Q(1), 8);
+ }
+
+ /* MSRs. */
+ state->msrs[NVMM_X64_MSR_EFER] = env->efer;
+ state->msrs[NVMM_X64_MSR_STAR] = env->star;
+#ifdef TARGET_X86_64
+ state->msrs[NVMM_X64_MSR_LSTAR] = env->lstar;
+ state->msrs[NVMM_X64_MSR_CSTAR] = env->cstar;
+ state->msrs[NVMM_X64_MSR_SFMASK] = env->fmask;
+ state->msrs[NVMM_X64_MSR_KERNELGSBASE] = env->kernelgsbase;
+#endif
+ state->msrs[NVMM_X64_MSR_SYSENTER_CS] = env->sysenter_cs;
+ state->msrs[NVMM_X64_MSR_SYSENTER_ESP] = env->sysenter_esp;
+ state->msrs[NVMM_X64_MSR_SYSENTER_EIP] = env->sysenter_eip;
+ state->msrs[NVMM_X64_MSR_PAT] = env->pat;
+ state->msrs[NVMM_X64_MSR_TSC] = env->tsc;
+
+ bitmap =
+ NVMM_X64_STATE_SEGS |
+ NVMM_X64_STATE_GPRS |
+ NVMM_X64_STATE_CRS |
+ NVMM_X64_STATE_DRS |
+ NVMM_X64_STATE_MSRS |
+ NVMM_X64_STATE_FPU;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, bitmap);
+ if (ret == -1) {
+ error_report("NVMM: Failed to set virtual processor context,"
+ " error=%d", errno);
+ }
+}
+
+static void
+nvmm_get_segment(SegmentCache *qseg, const struct nvmm_x64_state_seg *nseg)
+{
+ qseg->selector = nseg->selector;
+ qseg->limit = nseg->limit;
+ qseg->base = nseg->base;
+
+ qseg->flags =
+ __SHIFTIN((uint32_t)nseg->attrib.type, DESC_TYPE_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.s, DESC_S_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.dpl, DESC_DPL_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.p, DESC_P_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.avl, DESC_AVL_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.l, DESC_L_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.def, DESC_B_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.g, DESC_G_MASK);
+}
+
+static void
+nvmm_get_registers(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t bitmap, tpr;
+ size_t i;
+ int ret;
+
+ assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
+
+ bitmap =
+ NVMM_X64_STATE_SEGS |
+ NVMM_X64_STATE_GPRS |
+ NVMM_X64_STATE_CRS |
+ NVMM_X64_STATE_DRS |
+ NVMM_X64_STATE_MSRS |
+ NVMM_X64_STATE_FPU;
+
+ ret = nvmm_vcpu_getstate(mach, vcpu, bitmap);
+ if (ret == -1) {
+ error_report("NVMM: Failed to get virtual processor context,"
+ " error=%d", errno);
+ }
+
+ /* GPRs. */
+ env->regs[R_EAX] = state->gprs[NVMM_X64_GPR_RAX];
+ env->regs[R_ECX] = state->gprs[NVMM_X64_GPR_RCX];
+ env->regs[R_EDX] = state->gprs[NVMM_X64_GPR_RDX];
+ env->regs[R_EBX] = state->gprs[NVMM_X64_GPR_RBX];
+ env->regs[R_ESP] = state->gprs[NVMM_X64_GPR_RSP];
+ env->regs[R_EBP] = state->gprs[NVMM_X64_GPR_RBP];
+ env->regs[R_ESI] = state->gprs[NVMM_X64_GPR_RSI];
+ env->regs[R_EDI] = state->gprs[NVMM_X64_GPR_RDI];
+#ifdef TARGET_X86_64
+ env->regs[R_R8] = state->gprs[NVMM_X64_GPR_R8];
+ env->regs[R_R9] = state->gprs[NVMM_X64_GPR_R9];
+ env->regs[R_R10] = state->gprs[NVMM_X64_GPR_R10];
+ env->regs[R_R11] = state->gprs[NVMM_X64_GPR_R11];
+ env->regs[R_R12] = state->gprs[NVMM_X64_GPR_R12];
+ env->regs[R_R13] = state->gprs[NVMM_X64_GPR_R13];
+ env->regs[R_R14] = state->gprs[NVMM_X64_GPR_R14];
+ env->regs[R_R15] = state->gprs[NVMM_X64_GPR_R15];
+#endif
+
+ /* RIP and RFLAGS. */
+ env->eip = state->gprs[NVMM_X64_GPR_RIP];
+ env->eflags = state->gprs[NVMM_X64_GPR_RFLAGS];
+
+ /* Segments. */
+ nvmm_get_segment(&env->segs[R_ES], &state->segs[NVMM_X64_SEG_ES]);
+ nvmm_get_segment(&env->segs[R_CS], &state->segs[NVMM_X64_SEG_CS]);
+ nvmm_get_segment(&env->segs[R_SS], &state->segs[NVMM_X64_SEG_SS]);
+ nvmm_get_segment(&env->segs[R_DS], &state->segs[NVMM_X64_SEG_DS]);
+ nvmm_get_segment(&env->segs[R_FS], &state->segs[NVMM_X64_SEG_FS]);
+ nvmm_get_segment(&env->segs[R_GS], &state->segs[NVMM_X64_SEG_GS]);
+
+ /* Special segments. */
+ nvmm_get_segment(&env->gdt, &state->segs[NVMM_X64_SEG_GDT]);
+ nvmm_get_segment(&env->ldt, &state->segs[NVMM_X64_SEG_LDT]);
+ nvmm_get_segment(&env->tr, &state->segs[NVMM_X64_SEG_TR]);
+ nvmm_get_segment(&env->idt, &state->segs[NVMM_X64_SEG_IDT]);
+
+ /* Control registers. */
+ env->cr[0] = state->crs[NVMM_X64_CR_CR0];
+ env->cr[2] = state->crs[NVMM_X64_CR_CR2];
+ env->cr[3] = state->crs[NVMM_X64_CR_CR3];
+ env->cr[4] = state->crs[NVMM_X64_CR_CR4];
+ tpr = state->crs[NVMM_X64_CR_CR8];
+ if (tpr != qcpu->tpr) {
+ qcpu->tpr = tpr;
+ cpu_set_apic_tpr(x86_cpu->apic_state, tpr);
+ }
+ env->xcr0 = state->crs[NVMM_X64_CR_XCR0];
+
+ /* Debug registers. */
+ env->dr[0] = state->drs[NVMM_X64_DR_DR0];
+ env->dr[1] = state->drs[NVMM_X64_DR_DR1];
+ env->dr[2] = state->drs[NVMM_X64_DR_DR2];
+ env->dr[3] = state->drs[NVMM_X64_DR_DR3];
+ env->dr[6] = state->drs[NVMM_X64_DR_DR6];
+ env->dr[7] = state->drs[NVMM_X64_DR_DR7];
+
+ /* FPU. */
+ env->fpuc = state->fpu.fx_cw;
+ env->fpstt = (state->fpu.fx_sw >> 11) & 0x7;
+ env->fpus = state->fpu.fx_sw & ~0x3800;
+ for (i = 0; i < 8; i++) {
+ env->fptags[i] = !((state->fpu.fx_tw >> i) & 1);
+ }
+ env->fpop = state->fpu.fx_opcode;
+ env->fpip = state->fpu.fx_ip.fa_64;
+ env->fpdp = state->fpu.fx_dp.fa_64;
+ env->mxcsr = state->fpu.fx_mxcsr;
+ assert(sizeof(state->fpu.fx_87_ac) == sizeof(env->fpregs));
+ memcpy(env->fpregs, state->fpu.fx_87_ac, sizeof(env->fpregs));
+ for (i = 0; i < 16; i++) {
+ memcpy(&env->xmm_regs[i].ZMM_Q(0),
+ &state->fpu.fx_xmm[i].xmm_bytes[0], 8);
+ memcpy(&env->xmm_regs[i].ZMM_Q(1),
+ &state->fpu.fx_xmm[i].xmm_bytes[8], 8);
+ }
+
+ /* MSRs. */
+ env->efer = state->msrs[NVMM_X64_MSR_EFER];
+ env->star = state->msrs[NVMM_X64_MSR_STAR];
+#ifdef TARGET_X86_64
+ env->lstar = state->msrs[NVMM_X64_MSR_LSTAR];
+ env->cstar = state->msrs[NVMM_X64_MSR_CSTAR];
+ env->fmask = state->msrs[NVMM_X64_MSR_SFMASK];
+ env->kernelgsbase = state->msrs[NVMM_X64_MSR_KERNELGSBASE];
+#endif
+ env->sysenter_cs = state->msrs[NVMM_X64_MSR_SYSENTER_CS];
+ env->sysenter_esp = state->msrs[NVMM_X64_MSR_SYSENTER_ESP];
+ env->sysenter_eip = state->msrs[NVMM_X64_MSR_SYSENTER_EIP];
+ env->pat = state->msrs[NVMM_X64_MSR_PAT];
+ env->tsc = state->msrs[NVMM_X64_MSR_TSC];
+
+ x86_update_hflags(env);
+}
+
+static bool
+nvmm_can_take_int(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ struct nvmm_machine *mach = get_nvmm_mach();
+
+ if (qcpu->int_window_exit) {
+ return false;
+ }
+
+ if (qcpu->int_shadow || !(env->eflags & IF_MASK)) {
+ struct nvmm_x64_state *state = vcpu->state;
+
+ /* Exit on interrupt window. */
+ nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_INTR);
+ state->intr.int_window_exiting = 1;
+ nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_INTR);
+
+ return false;
+ }
+
+ return true;
+}
+
+static bool
+nvmm_can_take_nmi(CPUState *cpu)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+
+ /*
+ * Contrary to INTs, NMIs always schedule an exit when they are
+ * completed. Therefore, if window-exiting is enabled, it means
+ * NMIs are blocked.
+ */
+ if (qcpu->nmi_window_exit) {
+ return false;
+ }
+
+ return true;
+}
+
+/*
+ * Called before the VCPU is run. We inject events generated by the I/O
+ * thread, and synchronize the guest TPR.
+ */
+static void
+nvmm_vcpu_pre_run(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ struct nvmm_vcpu_event *event = vcpu->event;
+ bool has_event = false;
+ bool sync_tpr = false;
+ uint8_t tpr;
+ int ret;
+
+ qemu_mutex_lock_iothread();
+
+ tpr = cpu_get_apic_tpr(x86_cpu->apic_state);
+ if (tpr != qcpu->tpr) {
+ qcpu->tpr = tpr;
+ sync_tpr = true;
+ }
+
+ /*
+ * Force the VCPU out of its inner loop to process any INIT requests
+ * or commit pending TPR access.
+ */
+ if (cpu->interrupt_request & (CPU_INTERRUPT_INIT | CPU_INTERRUPT_TPR)) {
+ cpu->exit_request = 1;
+ }
+
+ if (!has_event && (cpu->interrupt_request & CPU_INTERRUPT_NMI)) {
+ if (nvmm_can_take_nmi(cpu)) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_NMI;
+ event->type = NVMM_VCPU_EVENT_INTR;
+ event->vector = 2;
+ has_event = true;
+ }
+ }
+
+ if (!has_event && (cpu->interrupt_request & CPU_INTERRUPT_HARD)) {
+ if (nvmm_can_take_int(cpu)) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_HARD;
+ event->type = NVMM_VCPU_EVENT_INTR;
+ event->vector = cpu_get_pic_interrupt(env);
+ has_event = true;
+ }
+ }
+
+ /* Don't want SMIs. */
+ if (cpu->interrupt_request & CPU_INTERRUPT_SMI) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_SMI;
+ }
+
+ if (sync_tpr) {
+ ret = nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_CRS);
+ if (ret == -1) {
+ error_report("NVMM: Failed to get CPU state,"
+ " error=%d", errno);
+ }
+
+ state->crs[NVMM_X64_CR_CR8] = qcpu->tpr;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_CRS);
+ if (ret == -1) {
+ error_report("NVMM: Failed to set CPU state,"
+ " error=%d", errno);
+ }
+ }
+
+ if (has_event) {
+ ret = nvmm_vcpu_inject(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: Failed to inject event,"
+ " error=%d", errno);
+ }
+ }
+
+ qemu_mutex_unlock_iothread();
+}
+
+/*
+ * Called after the VCPU ran. We synchronize the host view of the TPR and
+ * RFLAGS.
+ */
+static void
+nvmm_vcpu_post_run(CPUState *cpu, struct nvmm_vcpu_exit *exit)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ uint64_t tpr;
+
+ env->eflags = exit->exitstate.rflags;
+ qcpu->int_shadow = exit->exitstate.int_shadow;
+ qcpu->int_window_exit = exit->exitstate.int_window_exiting;
+ qcpu->nmi_window_exit = exit->exitstate.nmi_window_exiting;
+
+ tpr = exit->exitstate.cr8;
+ if (qcpu->tpr != tpr) {
+ qcpu->tpr = tpr;
+ qemu_mutex_lock_iothread();
+ cpu_set_apic_tpr(x86_cpu->apic_state, qcpu->tpr);
+ qemu_mutex_unlock_iothread();
+ }
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_io_callback(struct nvmm_io *io)
+{
+ MemTxAttrs attrs = { 0 };
+ int ret;
+
+ ret = address_space_rw(&address_space_io, io->port, attrs, io->data,
+ io->size, !io->in);
+ if (ret != MEMTX_OK) {
+ error_report("NVMM: I/O Transaction Failed "
+ "[%s, port=%u, size=%zu]", (io->in ? "in" : "out"),
+ io->port, io->size);
+ }
+
+ /* Needed, otherwise infinite loop. */
+ current_cpu->vcpu_dirty = false;
+}
+
+static void
+nvmm_mem_callback(struct nvmm_mem *mem)
+{
+ cpu_physical_memory_rw(mem->gpa, mem->data, mem->size, mem->write);
+
+ /* XXX Needed, otherwise infinite loop. */
+ current_cpu->vcpu_dirty = false;
+}
+
+static struct nvmm_assist_callbacks nvmm_callbacks = {
+ .io = nvmm_io_callback,
+ .mem = nvmm_mem_callback
+};
+
+/* -------------------------------------------------------------------------- */
+
+static int
+nvmm_handle_mem(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
+{
+ int ret;
+
+ ret = nvmm_assist_mem(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: Mem Assist Failed [gpa=%p]",
+ (void *)vcpu->exit->u.mem.gpa);
+ }
+
+ return ret;
+}
+
+static int
+nvmm_handle_io(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
+{
+ int ret;
+
+ ret = nvmm_assist_io(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: I/O Assist Failed [port=%d]",
+ (int)vcpu->exit->u.io.port);
+ }
+
+ return ret;
+}
+
+static int
+nvmm_handle_rdmsr(struct nvmm_machine *mach, CPUState *cpu,
+ struct nvmm_vcpu_exit *exit)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t val;
+ int ret;
+
+ switch (exit->u.rdmsr.msr) {
+ case MSR_IA32_APICBASE:
+ val = cpu_get_apic_base(x86_cpu->apic_state);
+ break;
+ case MSR_MTRRcap:
+ case MSR_MTRRdefType:
+ case MSR_MCG_CAP:
+ case MSR_MCG_STATUS:
+ val = 0;
+ break;
+ default: /* More MSRs to add? */
+ val = 0;
+ error_report("NVMM: Unexpected RDMSR 0x%x, ignored",
+ exit->u.rdmsr.msr);
+ break;
+ }
+
+ ret = nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ state->gprs[NVMM_X64_GPR_RAX] = (val & 0xFFFFFFFF);
+ state->gprs[NVMM_X64_GPR_RDX] = (val >> 32);
+ state->gprs[NVMM_X64_GPR_RIP] = exit->u.rdmsr.npc;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ return 0;
+}
+
+static int
+nvmm_handle_wrmsr(struct nvmm_machine *mach, CPUState *cpu,
+ struct nvmm_vcpu_exit *exit)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t val;
+ int ret;
+
+ val = exit->u.wrmsr.val;
+
+ switch (exit->u.wrmsr.msr) {
+ case MSR_IA32_APICBASE:
+ cpu_set_apic_base(x86_cpu->apic_state, val);
+ break;
+ case MSR_MTRRdefType:
+ case MSR_MCG_STATUS:
+ break;
+ default: /* More MSRs to add? */
+ error_report("NVMM: Unexpected WRMSR 0x%x [val=0x%lx], ignored",
+ exit->u.wrmsr.msr, val);
+ break;
+ }
+
+ ret = nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ state->gprs[NVMM_X64_GPR_RIP] = exit->u.wrmsr.npc;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ return 0;
+}
+
+static int
+nvmm_handle_halted(struct nvmm_machine *mach, CPUState *cpu,
+ struct nvmm_vcpu_exit *exit)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ int ret = 0;
+
+ qemu_mutex_lock_iothread();
+
+ if (!((cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
+ (env->eflags & IF_MASK)) &&
+ !(cpu->interrupt_request & CPU_INTERRUPT_NMI)) {
+ cpu->exception_index = EXCP_HLT;
+ cpu->halted = true;
+ ret = 1;
+ }
+
+ qemu_mutex_unlock_iothread();
+
+ return ret;
+}
+
+static int
+nvmm_inject_ud(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
+{
+ struct nvmm_vcpu_event *event = vcpu->event;
+
+ event->type = NVMM_VCPU_EVENT_EXCP;
+ event->vector = 6;
+ event->u.excp.error = 0;
+
+ return nvmm_vcpu_inject(mach, vcpu);
+}
+
+static int
+nvmm_vcpu_loop(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_vcpu_exit *exit = vcpu->exit;
+ int ret;
+
+ /*
+ * Some asynchronous events must be handled outside of the inner
+ * VCPU loop. They are handled here.
+ */
+ if (cpu->interrupt_request & CPU_INTERRUPT_INIT) {
+ nvmm_cpu_synchronize_state(cpu);
+ do_cpu_init(x86_cpu);
+ /* set int/nmi windows back to the reset state */
+ }
+ if (cpu->interrupt_request & CPU_INTERRUPT_POLL) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_POLL;
+ apic_poll_irq(x86_cpu->apic_state);
+ }
+ if (((cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
+ (env->eflags & IF_MASK)) ||
+ (cpu->interrupt_request & CPU_INTERRUPT_NMI)) {
+ cpu->halted = false;
+ }
+ if (cpu->interrupt_request & CPU_INTERRUPT_SIPI) {
+ nvmm_cpu_synchronize_state(cpu);
+ do_cpu_sipi(x86_cpu);
+ }
+ if (cpu->interrupt_request & CPU_INTERRUPT_TPR) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_TPR;
+ nvmm_cpu_synchronize_state(cpu);
+ apic_handle_tpr_access_report(x86_cpu->apic_state, env->eip,
+ env->tpr_access_type);
+ }
+
+ if (cpu->halted) {
+ cpu->exception_index = EXCP_HLT;
+ atomic_set(&cpu->exit_request, false);
+ return 0;
+ }
+
+ qemu_mutex_unlock_iothread();
+ cpu_exec_start(cpu);
+
+ /*
+ * Inner VCPU loop.
+ */
+ do {
+ if (cpu->vcpu_dirty) {
+ nvmm_set_registers(cpu);
+ cpu->vcpu_dirty = false;
+ }
+
+ if (qcpu->stop) {
+ cpu->exception_index = EXCP_INTERRUPT;
+ qcpu->stop = false;
+ ret = 1;
+ break;
+ }
+
+ nvmm_vcpu_pre_run(cpu);
+
+ if (atomic_read(&cpu->exit_request)) {
+ qemu_cpu_kick_self();
+ }
+
+ ret = nvmm_vcpu_run(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: Failed to exec a virtual processor,"
+ " error=%d", errno);
+ break;
+ }
+
+ nvmm_vcpu_post_run(cpu, exit);
+
+ switch (exit->reason) {
+ case NVMM_VCPU_EXIT_NONE:
+ break;
+ case NVMM_VCPU_EXIT_MEMORY:
+ ret = nvmm_handle_mem(mach, vcpu);
+ break;
+ case NVMM_VCPU_EXIT_IO:
+ ret = nvmm_handle_io(mach, vcpu);
+ break;
+ case NVMM_VCPU_EXIT_INT_READY:
+ case NVMM_VCPU_EXIT_NMI_READY:
+ case NVMM_VCPU_EXIT_TPR_CHANGED:
+ break;
+ case NVMM_VCPU_EXIT_HALTED:
+ ret = nvmm_handle_halted(mach, cpu, exit);
+ break;
+ case NVMM_VCPU_EXIT_SHUTDOWN:
+ qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
+ cpu->exception_index = EXCP_INTERRUPT;
+ ret = 1;
+ break;
+ case NVMM_VCPU_EXIT_RDMSR:
+ ret = nvmm_handle_rdmsr(mach, cpu, exit);
+ break;
+ case NVMM_VCPU_EXIT_WRMSR:
+ ret = nvmm_handle_wrmsr(mach, cpu, exit);
+ break;
+ case NVMM_VCPU_EXIT_MONITOR:
+ case NVMM_VCPU_EXIT_MWAIT:
+ ret = nvmm_inject_ud(mach, vcpu);
+ break;
+ default:
+ error_report("NVMM: Unexpected VM exit code 0x%lx [hw=0x%lx]",
+ exit->reason, exit->u.inv.hwcode);
+ nvmm_get_registers(cpu);
+ qemu_mutex_lock_iothread();
+ qemu_system_guest_panicked(cpu_get_crash_info(cpu));
+ qemu_mutex_unlock_iothread();
+ ret = -1;
+ break;
+ }
+ } while (ret == 0);
+
+ cpu_exec_end(cpu);
+ qemu_mutex_lock_iothread();
+ current_cpu = cpu;
+
+ atomic_set(&cpu->exit_request, false);
+
+ return ret < 0;
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+do_nvmm_cpu_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
+{
+ nvmm_get_registers(cpu);
+ cpu->vcpu_dirty = true;
+}
+
+static void
+do_nvmm_cpu_synchronize_post_reset(CPUState *cpu, run_on_cpu_data arg)
+{
+ nvmm_set_registers(cpu);
+ cpu->vcpu_dirty = false;
+}
+
+static void
+do_nvmm_cpu_synchronize_post_init(CPUState *cpu, run_on_cpu_data arg)
+{
+ nvmm_set_registers(cpu);
+ cpu->vcpu_dirty = false;
+}
+
+static void
+do_nvmm_cpu_synchronize_pre_loadvm(CPUState *cpu, run_on_cpu_data arg)
+{
+ cpu->vcpu_dirty = true;
+}
+
+void nvmm_cpu_synchronize_state(CPUState *cpu)
+{
+ if (!cpu->vcpu_dirty) {
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_state, RUN_ON_CPU_NULL);
+ }
+}
+
+void nvmm_cpu_synchronize_post_reset(CPUState *cpu)
+{
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_post_reset, RUN_ON_CPU_NULL);
+}
+
+void nvmm_cpu_synchronize_post_init(CPUState *cpu)
+{
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_post_init, RUN_ON_CPU_NULL);
+}
+
+void nvmm_cpu_synchronize_pre_loadvm(CPUState *cpu)
+{
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_pre_loadvm, RUN_ON_CPU_NULL);
+}
+
+/* -------------------------------------------------------------------------- */
+
+static Error *nvmm_migration_blocker;
+
+static void
+nvmm_ipi_signal(int sigcpu)
+{
+ struct qemu_vcpu *qcpu;
+
+ if (current_cpu) {
+ qcpu = get_qemu_vcpu(current_cpu);
+ qcpu->stop = true;
+ }
+}
+
+static void
+nvmm_init_cpu_signals(void)
+{
+ struct sigaction sigact;
+ sigset_t set;
+
+ /* Install the IPI handler. */
+ memset(&sigact, 0, sizeof(sigact));
+ sigact.sa_handler = nvmm_ipi_signal;
+ sigaction(SIG_IPI, &sigact, NULL);
+
+ /* Allow IPIs on the current thread. */
+ sigprocmask(SIG_BLOCK, NULL, &set);
+ sigdelset(&set, SIG_IPI);
+ pthread_sigmask(SIG_SETMASK, &set, NULL);
+}
+
+int
+nvmm_init_vcpu(CPUState *cpu)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct nvmm_vcpu_conf_cpuid cpuid;
+ struct nvmm_vcpu_conf_tpr tpr;
+ Error *local_error = NULL;
+ struct qemu_vcpu *qcpu;
+ int ret, err;
+
+ nvmm_init_cpu_signals();
+
+ if (nvmm_migration_blocker == NULL) {
+ error_setg(&nvmm_migration_blocker,
+ "NVMM: Migration not supported");
+
+ (void)migrate_add_blocker(nvmm_migration_blocker, &local_error);
+ if (local_error) {
+ error_report_err(local_error);
+ migrate_del_blocker(nvmm_migration_blocker);
+ error_free(nvmm_migration_blocker);
+ return -EINVAL;
+ }
+ }
+
+ qcpu = g_malloc0(sizeof(*qcpu));
+ if (qcpu == NULL) {
+ error_report("NVMM: Failed to allocate VCPU context.");
+ return -ENOMEM;
+ }
+
+ ret = nvmm_vcpu_create(mach, cpu->cpu_index, &qcpu->vcpu);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to create a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+
+ memset(&cpuid, 0, sizeof(cpuid));
+ cpuid.mask = 1;
+ cpuid.leaf = 0x00000001;
+ cpuid.u.mask.set.edx = CPUID_MCE | CPUID_MCA | CPUID_MTRR;
+ ret = nvmm_vcpu_configure(mach, &qcpu->vcpu, NVMM_VCPU_CONF_CPUID,
+ &cpuid);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to configure a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+
+ ret = nvmm_vcpu_configure(mach, &qcpu->vcpu, NVMM_VCPU_CONF_CALLBACKS,
+ &nvmm_callbacks);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to configure a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+
+ if (qemu_mach.cap.arch.vcpu_conf_support & NVMM_CAP_ARCH_VCPU_CONF_TPR) {
+ memset(&tpr, 0, sizeof(tpr));
+ tpr.exit_changed = 1;
+ ret = nvmm_vcpu_configure(mach, &qcpu->vcpu, NVMM_VCPU_CONF_TPR, &tpr);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to configure a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+ }
+
+ cpu->vcpu_dirty = true;
+ cpu->hax_vcpu = (struct hax_vcpu_state *)qcpu;
+
+ return 0;
+}
+
+int
+nvmm_vcpu_exec(CPUState *cpu)
+{
+ int ret, fatal;
+
+ while (1) {
+ if (cpu->exception_index >= EXCP_INTERRUPT) {
+ ret = cpu->exception_index;
+ cpu->exception_index = -1;
+ break;
+ }
+
+ fatal = nvmm_vcpu_loop(cpu);
+
+ if (fatal) {
+ error_report("NVMM: Failed to execute a VCPU.");
+ abort();
+ }
+ }
+
+ return ret;
+}
+
+void
+nvmm_destroy_vcpu(CPUState *cpu)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+
+ nvmm_vcpu_destroy(mach, &qcpu->vcpu);
+ g_free(cpu->hax_vcpu);
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_update_mapping(hwaddr start_pa, ram_addr_t size, uintptr_t hva,
+ bool add, bool rom, const char *name)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ int ret, prot;
+
+ if (add) {
+ prot = PROT_READ | PROT_EXEC;
+ if (!rom) {
+ prot |= PROT_WRITE;
+ }
+ ret = nvmm_gpa_map(mach, hva, start_pa, size, prot);
+ } else {
+ ret = nvmm_gpa_unmap(mach, hva, start_pa, size);
+ }
+
+ if (ret == -1) {
+ error_report("NVMM: Failed to %s GPA range '%s' PA:%p, "
+ "Size:%p bytes, HostVA:%p, error=%d",
+ (add ? "map" : "unmap"), name, (void *)(uintptr_t)start_pa,
+ (void *)size, (void *)hva, errno);
+ }
+}
+
+static void
+nvmm_process_section(MemoryRegionSection *section, int add)
+{
+ MemoryRegion *mr = section->mr;
+ hwaddr start_pa = section->offset_within_address_space;
+ ram_addr_t size = int128_get64(section->size);
+ unsigned int delta;
+ uintptr_t hva;
+
+ if (!memory_region_is_ram(mr)) {
+ return;
+ }
+
+ /* Adjust start_pa and size so that they are page-aligned. */
+ delta = qemu_real_host_page_size - (start_pa & ~qemu_real_host_page_mask);
+ delta &= ~qemu_real_host_page_mask;
+ if (delta > size) {
+ return;
+ }
+ start_pa += delta;
+ size -= delta;
+ size &= qemu_real_host_page_mask;
+ if (!size || (start_pa & ~qemu_real_host_page_mask)) {
+ return;
+ }
+
+ hva = (uintptr_t)memory_region_get_ram_ptr(mr) +
+ section->offset_within_region + delta;
+
+ nvmm_update_mapping(start_pa, size, hva, add,
+ memory_region_is_rom(mr), mr->name);
+}
+
+static void
+nvmm_region_add(MemoryListener *listener, MemoryRegionSection *section)
+{
+ memory_region_ref(section->mr);
+ nvmm_process_section(section, 1);
+}
+
+static void
+nvmm_region_del(MemoryListener *listener, MemoryRegionSection *section)
+{
+ nvmm_process_section(section, 0);
+ memory_region_unref(section->mr);
+}
+
+static void
+nvmm_transaction_begin(MemoryListener *listener)
+{
+ /* nothing */
+}
+
+static void
+nvmm_transaction_commit(MemoryListener *listener)
+{
+ /* nothing */
+}
+
+static void
+nvmm_log_sync(MemoryListener *listener, MemoryRegionSection *section)
+{
+ MemoryRegion *mr = section->mr;
+
+ if (!memory_region_is_ram(mr)) {
+ return;
+ }
+
+ memory_region_set_dirty(mr, 0, int128_get64(section->size));
+}
+
+static MemoryListener nvmm_memory_listener = {
+ .begin = nvmm_transaction_begin,
+ .commit = nvmm_transaction_commit,
+ .region_add = nvmm_region_add,
+ .region_del = nvmm_region_del,
+ .log_sync = nvmm_log_sync,
+ .priority = 10,
+};
+
+static void
+nvmm_ram_block_added(RAMBlockNotifier *n, void *host, size_t size)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ uintptr_t hva = (uintptr_t)host;
+ int ret;
+
+ ret = nvmm_hva_map(mach, hva, size);
+
+ if (ret == -1) {
+ error_report("NVMM: Failed to map HVA, HostVA:%p "
+ "Size:%p bytes, error=%d",
+ (void *)hva, (void *)size, errno);
+ }
+}
+
+static struct RAMBlockNotifier nvmm_ram_notifier = {
+ .ram_block_added = nvmm_ram_block_added
+};
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_handle_interrupt(CPUState *cpu, int mask)
+{
+ cpu->interrupt_request |= mask;
+
+ if (!qemu_cpu_is_self(cpu)) {
+ qemu_cpu_kick(cpu);
+ }
+}
+
+/* -------------------------------------------------------------------------- */
+
+static int
+nvmm_accel_init(MachineState *ms)
+{
+ int ret, err;
+
+ ret = nvmm_init();
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Initialization failed, error=%d", errno);
+ return -err;
+ }
+
+ ret = nvmm_capability(&qemu_mach.cap);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Unable to fetch capability, error=%d", errno);
+ return -err;
+ }
+ if (qemu_mach.cap.version != NVMM_KERN_VERSION) {
+ error_report("NVMM: Unsupported version %u", qemu_mach.cap.version);
+ return -EPROGMISMATCH;
+ }
+ if (qemu_mach.cap.state_size != sizeof(struct nvmm_x64_state)) {
+ error_report("NVMM: Wrong state size %u", qemu_mach.cap.state_size);
+ return -EPROGMISMATCH;
+ }
+
+ ret = nvmm_machine_create(&qemu_mach.mach);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Machine creation failed, error=%d", errno);
+ return -err;
+ }
+
+ memory_listener_register(&nvmm_memory_listener, &address_space_memory);
+ ram_block_notifier_add(&nvmm_ram_notifier);
+
+ cpu_interrupt_handler = nvmm_handle_interrupt;
+
+ printf("NetBSD Virtual Machine Monitor accelerator is operational\n");
+ return 0;
+}
+
+int
+nvmm_enabled(void)
+{
+ return nvmm_allowed;
+}
+
+static void
+nvmm_accel_class_init(ObjectClass *oc, void *data)
+{
+ AccelClass *ac = ACCEL_CLASS(oc);
+ ac->name = "NVMM";
+ ac->init_machine = nvmm_accel_init;
+ ac->allowed = &nvmm_allowed;
+}
+
+static const TypeInfo nvmm_accel_type = {
+ .name = ACCEL_CLASS_NAME("nvmm"),
+ .parent = TYPE_ACCEL,
+ .class_init = nvmm_accel_class_init,
+};
+
+static void
+nvmm_type_init(void)
+{
+ type_register_static(&nvmm_accel_type);
+}
+
+type_init(nvmm_type_init);
--
2.28.0
From: Maxime Villard <max@m00nbsd.net>
Implements the NVMM accelerator cpu enlightenments to actually use the nvmm-all
accelerator on NetBSD platforms.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
include/sysemu/hw_accel.h | 14 ++++++++++
softmmu/cpus.c | 58 +++++++++++++++++++++++++++++++++++++++
target/i386/helper.c | 2 +-
3 files changed, 73 insertions(+), 1 deletion(-)
diff --git a/include/sysemu/hw_accel.h b/include/sysemu/hw_accel.h
index e128f8b06b..9e19f5794c 100644
--- a/include/sysemu/hw_accel.h
+++ b/include/sysemu/hw_accel.h
@@ -16,6 +16,7 @@
#include "sysemu/kvm.h"
#include "sysemu/hvf.h"
#include "sysemu/whpx.h"
+#include "sysemu/nvmm.h"
static inline void cpu_synchronize_state(CPUState *cpu)
{
@@ -31,6 +32,9 @@ static inline void cpu_synchronize_state(CPUState *cpu)
if (whpx_enabled()) {
whpx_cpu_synchronize_state(cpu);
}
+ if (nvmm_enabled()) {
+ nvmm_cpu_synchronize_state(cpu);
+ }
}
static inline void cpu_synchronize_post_reset(CPUState *cpu)
@@ -47,6 +51,10 @@ static inline void cpu_synchronize_post_reset(CPUState *cpu)
if (whpx_enabled()) {
whpx_cpu_synchronize_post_reset(cpu);
}
+ if (nvmm_enabled()) {
+ nvmm_cpu_synchronize_post_reset(cpu);
+ }
+
}
static inline void cpu_synchronize_post_init(CPUState *cpu)
@@ -63,6 +71,9 @@ static inline void cpu_synchronize_post_init(CPUState *cpu)
if (whpx_enabled()) {
whpx_cpu_synchronize_post_init(cpu);
}
+ if (nvmm_enabled()) {
+ nvmm_cpu_synchronize_post_init(cpu);
+ }
}
static inline void cpu_synchronize_pre_loadvm(CPUState *cpu)
@@ -79,6 +90,9 @@ static inline void cpu_synchronize_pre_loadvm(CPUState *cpu)
if (whpx_enabled()) {
whpx_cpu_synchronize_pre_loadvm(cpu);
}
+ if (nvmm_enabled()) {
+ nvmm_cpu_synchronize_pre_loadvm(cpu);
+ }
}
#endif /* QEMU_HW_ACCEL_H */
diff --git a/softmmu/cpus.c b/softmmu/cpus.c
index a802e899ab..3b44b92830 100644
--- a/softmmu/cpus.c
+++ b/softmmu/cpus.c
@@ -43,6 +43,7 @@
#include "sysemu/hax.h"
#include "sysemu/hvf.h"
#include "sysemu/whpx.h"
+#include "sysemu/nvmm.h"
#include "exec/exec-all.h"
#include "qemu/thread.h"
@@ -1621,6 +1622,48 @@ static void *qemu_whpx_cpu_thread_fn(void *arg)
return NULL;
}
+static void *qemu_nvmm_cpu_thread_fn(void *arg)
+{
+ CPUState *cpu = arg;
+ int r;
+
+ assert(nvmm_enabled());
+
+ rcu_register_thread();
+
+ qemu_mutex_lock_iothread();
+ qemu_thread_get_self(cpu->thread);
+ cpu->thread_id = qemu_get_thread_id();
+ current_cpu = cpu;
+
+ r = nvmm_init_vcpu(cpu);
+ if (r < 0) {
+ fprintf(stderr, "nvmm_init_vcpu failed: %s\n", strerror(-r));
+ exit(1);
+ }
+
+ /* signal CPU creation */
+ cpu->created = true;
+ qemu_cond_signal(&qemu_cpu_cond);
+
+ do {
+ if (cpu_can_run(cpu)) {
+ r = nvmm_vcpu_exec(cpu);
+ if (r == EXCP_DEBUG) {
+ cpu_handle_guest_debug(cpu);
+ }
+ }
+ qemu_wait_io_event(cpu);
+ } while (!cpu->unplug || cpu_can_run(cpu));
+
+ nvmm_destroy_vcpu(cpu);
+ cpu->created = false;
+ qemu_cond_signal(&qemu_cpu_cond);
+ qemu_mutex_unlock_iothread();
+ rcu_unregister_thread();
+ return NULL;
+}
+
#ifdef _WIN32
static void CALLBACK dummy_apc_func(ULONG_PTR unused)
{
@@ -1998,6 +2041,19 @@ static void qemu_whpx_start_vcpu(CPUState *cpu)
#endif
}
+static void qemu_nvmm_start_vcpu(CPUState *cpu)
+{
+ char thread_name[VCPU_THREAD_NAME_SIZE];
+
+ cpu->thread = g_malloc0(sizeof(QemuThread));
+ cpu->halt_cond = g_malloc0(sizeof(QemuCond));
+ qemu_cond_init(cpu->halt_cond);
+ snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/NVMM",
+ cpu->cpu_index);
+ qemu_thread_create(cpu->thread, thread_name, qemu_nvmm_cpu_thread_fn,
+ cpu, QEMU_THREAD_JOINABLE);
+}
+
static void qemu_dummy_start_vcpu(CPUState *cpu)
{
char thread_name[VCPU_THREAD_NAME_SIZE];
@@ -2038,6 +2094,8 @@ void qemu_init_vcpu(CPUState *cpu)
qemu_tcg_init_vcpu(cpu);
} else if (whpx_enabled()) {
qemu_whpx_start_vcpu(cpu);
+ } else if (nvmm_enabled()) {
+ qemu_nvmm_start_vcpu(cpu);
} else {
qemu_dummy_start_vcpu(cpu);
}
diff --git a/target/i386/helper.c b/target/i386/helper.c
index 70be53e2c3..c2f1aef65c 100644
--- a/target/i386/helper.c
+++ b/target/i386/helper.c
@@ -983,7 +983,7 @@ void cpu_report_tpr_access(CPUX86State *env, TPRAccess access)
X86CPU *cpu = env_archcpu(env);
CPUState *cs = env_cpu(env);
- if (kvm_enabled() || whpx_enabled()) {
+ if (kvm_enabled() || whpx_enabled() || nvmm_enabled()) {
env->tpr_access_type = access;
cpu_interrupt(cs, CPU_INTERRUPT_TPR);
--
2.28.0
From: Maxime Villard <max@m00nbsd.net>
Introduces the configure support for the new NetBSD Virtual Machine Monitor that
allows for hypervisor acceleration from usermode components on the NetBSD
platform.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
configure | 37 +++++++++++++++++++++++++++++++++++++
qemu-options.hx | 16 ++++++++--------
2 files changed, 45 insertions(+), 8 deletions(-)
diff --git a/configure b/configure
index 115dc38085..d4a837cf9d 100755
--- a/configure
+++ b/configure
@@ -241,6 +241,17 @@ supported_whpx_target() {
return 1
}
+supported_nvmm_target() {
+ test "$nvmm" = "yes" || return 1
+ glob "$1" "*-softmmu" || return 1
+ case "${1%-softmmu}" in
+ i386|x86_64)
+ return 0
+ ;;
+ esac
+ return 1
+}
+
supported_target() {
case "$1" in
*-softmmu)
@@ -268,6 +279,7 @@ supported_target() {
supported_hax_target "$1" && return 0
supported_hvf_target "$1" && return 0
supported_whpx_target "$1" && return 0
+ supported_nvmm_target "$1" && return 0
print_error "TCG disabled, but hardware accelerator not available for '$target'"
return 1
}
@@ -388,6 +400,7 @@ kvm="no"
hax="no"
hvf="no"
whpx="no"
+nvmm="no"
rdma=""
pvrdma=""
gprof="no"
@@ -823,6 +836,7 @@ DragonFly)
NetBSD)
bsd="yes"
hax="yes"
+ nvmm="yes"
make="${MAKE-gmake}"
audio_drv_list="oss try-sdl"
audio_possible_drivers="oss sdl"
@@ -1169,6 +1183,10 @@ for opt do
;;
--enable-whpx) whpx="yes"
;;
+ --disable-nvmm) nvmm="no"
+ ;;
+ --enable-nvmm) nvmm="yes"
+ ;;
--disable-tcg-interpreter) tcg_interpreter="no"
;;
--enable-tcg-interpreter) tcg_interpreter="yes"
@@ -1773,6 +1791,7 @@ disabled with --disable-FEATURE, default is enabled if available:
hax HAX acceleration support
hvf Hypervisor.framework acceleration support
whpx Windows Hypervisor Platform acceleration support
+ nvmm NetBSD Virtual Machine Monitor acceleration support
rdma Enable RDMA-based migration
pvrdma Enable PVRDMA support
vde support for vde network
@@ -2764,6 +2783,20 @@ if test "$whpx" != "no" ; then
fi
fi
+##########################################
+# NetBSD Virtual Machine Monitor (NVMM) accelerator check
+if test "$nvmm" != "no" ; then
+ if check_include "nvmm.h" ; then
+ nvmm="yes"
+ LIBS="-lnvmm $LIBS"
+ else
+ if test "$nvmm" = "yes"; then
+ feature_not_found "NVMM" "NVMM is not available"
+ fi
+ nvmm="no"
+ fi
+fi
+
##########################################
# Sparse probe
if test "$sparse" != "no" ; then
@@ -6543,6 +6576,7 @@ echo "KVM support $kvm"
echo "HAX support $hax"
echo "HVF support $hvf"
echo "WHPX support $whpx"
+echo "NVMM support $nvmm"
echo "TCG support $tcg"
if test "$tcg" = "yes" ; then
echo "TCG debug enabled $debug_tcg"
@@ -7828,6 +7862,9 @@ fi
if test "$target_aligned_only" = "yes" ; then
echo "TARGET_ALIGNED_ONLY=y" >> $config_target_mak
fi
+if supported_nvmm_target $target; then
+ echo "CONFIG_NVMM=y" >> $config_target_mak
+fi
if test "$target_bigendian" = "yes" ; then
echo "TARGET_WORDS_BIGENDIAN=y" >> $config_target_mak
fi
diff --git a/qemu-options.hx b/qemu-options.hx
index 224a8e8712..10c046c916 100644
--- a/qemu-options.hx
+++ b/qemu-options.hx
@@ -31,7 +31,7 @@ DEF("machine", HAS_ARG, QEMU_OPTION_machine, \
"-machine [type=]name[,prop[=value][,...]]\n"
" selects emulated machine ('-machine help' for list)\n"
" property accel=accel1[:accel2[:...]] selects accelerator\n"
- " supported accelerators are kvm, xen, hax, hvf, whpx or tcg (default: tcg)\n"
+ " supported accelerators are kvm, xen, hax, hvf, nvmm, whpx or tcg (default: tcg)\n"
" vmport=on|off|auto controls emulation of vmport (default: auto)\n"
" dump-guest-core=on|off include guest memory in a core dump (default=on)\n"
" mem-merge=on|off controls memory merge support (default: on)\n"
@@ -64,9 +64,9 @@ Supported machine properties are:
@table @option
@item accel=@var{accels1}[:@var{accels2}[:...]]
This is used to enable an accelerator. Depending on the target architecture,
-kvm, xen, hax, hvf, whpx or tcg can be available. By default, tcg is used. If there is
-more than one accelerator specified, the next one is used if the previous one
-fails to initialize.
+kvm, xen, hax, hvf, nvmm, whpx or tcg can be available. By default, tcg is used.
+If there is more than one accelerator specified, the next one is used if the
+previous one fails to initialize.
@item vmport=on|off|auto
Enables emulation of VMWare IO port, for vmmouse etc. auto says to select the
value based on accel. For accel=xen the default is off otherwise the default
@@ -114,7 +114,7 @@ ETEXI
DEF("accel", HAS_ARG, QEMU_OPTION_accel,
"-accel [accel=]accelerator[,prop[=value][,...]]\n"
- " select accelerator (kvm, xen, hax, hvf, whpx or tcg; use 'help' for a list)\n"
+ " select accelerator (kvm, xen, hax, hvf, nvmm, whpx or tcg; use 'help' for a list)\n"
" igd-passthru=on|off (enable Xen integrated Intel graphics passthrough, default=off)\n"
" kernel-irqchip=on|off|split controls accelerated irqchip support (default=on)\n"
" kvm-shadow-mem=size of KVM shadow MMU in bytes\n"
@@ -124,9 +124,9 @@ STEXI
@item -accel @var{name}[,prop=@var{value}[,...]]
@findex -accel
This is used to enable an accelerator. Depending on the target architecture,
-kvm, xen, hax, hvf, whpx or tcg can be available. By default, tcg is used. If there is
-more than one accelerator specified, the next one is used if the previous one
-fails to initialize.
+kvm, xen, hax, hvf, nvmm, whpx or tcg can be available. By default, tcg is used.
+If there is more than one accelerator specified, the next one is used if the
+previous one fails to initialize.
@table @option
@item igd-passthru=on|off
When Xen is in use, this option controls whether Intel integrated graphics
--
2.25.0
From: Maxime Villard <max@m00nbsd.net>
Implements the NetBSD Virtual Machine Monitor (NVMM) target. Which
acts as a hypervisor accelerator for QEMU on the NetBSD platform. This enables
QEMU much greater speed over the emulated x86_64 path's that are taken on
NetBSD today.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
target/i386/Makefile.objs | 1 +
target/i386/nvmm-all.c | 1226 +++++++++++++++++++++++++++++++++++++
2 files changed, 1227 insertions(+)
create mode 100644 target/i386/nvmm-all.c
diff --git a/target/i386/Makefile.objs b/target/i386/Makefile.objs
index 48e0c28434..bdcdb32e93 100644
--- a/target/i386/Makefile.objs
+++ b/target/i386/Makefile.objs
@@ -17,6 +17,7 @@ obj-$(CONFIG_HAX) += hax-all.o hax-mem.o hax-posix.o
endif
obj-$(CONFIG_HVF) += hvf/
obj-$(CONFIG_WHPX) += whpx-all.o
+obj-$(CONFIG_NVMM) += nvmm-all.o
endif
obj-$(CONFIG_SEV) += sev.o
obj-$(call lnot,$(CONFIG_SEV)) += sev-stub.o
diff --git a/target/i386/nvmm-all.c b/target/i386/nvmm-all.c
new file mode 100644
index 0000000000..a21908f46a
--- /dev/null
+++ b/target/i386/nvmm-all.c
@@ -0,0 +1,1226 @@
+/*
+ * Copyright (c) 2018-2019 Maxime Villard, All rights reserved.
+ *
+ * NetBSD Virtual Machine Monitor (NVMM) accelerator for QEMU.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "cpu.h"
+#include "exec/address-spaces.h"
+#include "exec/ioport.h"
+#include "qemu-common.h"
+#include "strings.h"
+#include "sysemu/accel.h"
+#include "sysemu/nvmm.h"
+#include "sysemu/runstate.h"
+#include "sysemu/sysemu.h"
+#include "sysemu/cpus.h"
+#include "qemu/main-loop.h"
+#include "qemu/error-report.h"
+#include "qemu/queue.h"
+#include "qapi/error.h"
+#include "migration/blocker.h"
+
+#include <nvmm.h>
+
+struct qemu_vcpu {
+ struct nvmm_vcpu vcpu;
+ uint8_t tpr;
+ bool stop;
+
+ /* Window-exiting for INTs/NMIs. */
+ bool int_window_exit;
+ bool nmi_window_exit;
+
+ /* The guest is in an interrupt shadow (POP SS, etc). */
+ bool int_shadow;
+};
+
+struct qemu_machine {
+ struct nvmm_capability cap;
+ struct nvmm_machine mach;
+};
+
+/* -------------------------------------------------------------------------- */
+
+static bool nvmm_allowed;
+static struct qemu_machine qemu_mach;
+
+static struct qemu_vcpu *
+get_qemu_vcpu(CPUState *cpu)
+{
+ return (struct qemu_vcpu *)cpu->hax_vcpu;
+}
+
+static struct nvmm_machine *
+get_nvmm_mach(void)
+{
+ return &qemu_mach.mach;
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_set_segment(struct nvmm_x64_state_seg *nseg, const SegmentCache *qseg)
+{
+ uint32_t attrib = qseg->flags;
+
+ nseg->selector = qseg->selector;
+ nseg->limit = qseg->limit;
+ nseg->base = qseg->base;
+ nseg->attrib.type = __SHIFTOUT(attrib, DESC_TYPE_MASK);
+ nseg->attrib.s = __SHIFTOUT(attrib, DESC_S_MASK);
+ nseg->attrib.dpl = __SHIFTOUT(attrib, DESC_DPL_MASK);
+ nseg->attrib.p = __SHIFTOUT(attrib, DESC_P_MASK);
+ nseg->attrib.avl = __SHIFTOUT(attrib, DESC_AVL_MASK);
+ nseg->attrib.l = __SHIFTOUT(attrib, DESC_L_MASK);
+ nseg->attrib.def = __SHIFTOUT(attrib, DESC_B_MASK);
+ nseg->attrib.g = __SHIFTOUT(attrib, DESC_G_MASK);
+}
+
+static void
+nvmm_set_registers(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t bitmap;
+ size_t i;
+ int ret;
+
+ assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
+
+ /* GPRs. */
+ state->gprs[NVMM_X64_GPR_RAX] = env->regs[R_EAX];
+ state->gprs[NVMM_X64_GPR_RCX] = env->regs[R_ECX];
+ state->gprs[NVMM_X64_GPR_RDX] = env->regs[R_EDX];
+ state->gprs[NVMM_X64_GPR_RBX] = env->regs[R_EBX];
+ state->gprs[NVMM_X64_GPR_RSP] = env->regs[R_ESP];
+ state->gprs[NVMM_X64_GPR_RBP] = env->regs[R_EBP];
+ state->gprs[NVMM_X64_GPR_RSI] = env->regs[R_ESI];
+ state->gprs[NVMM_X64_GPR_RDI] = env->regs[R_EDI];
+#ifdef TARGET_X86_64
+ state->gprs[NVMM_X64_GPR_R8] = env->regs[R_R8];
+ state->gprs[NVMM_X64_GPR_R9] = env->regs[R_R9];
+ state->gprs[NVMM_X64_GPR_R10] = env->regs[R_R10];
+ state->gprs[NVMM_X64_GPR_R11] = env->regs[R_R11];
+ state->gprs[NVMM_X64_GPR_R12] = env->regs[R_R12];
+ state->gprs[NVMM_X64_GPR_R13] = env->regs[R_R13];
+ state->gprs[NVMM_X64_GPR_R14] = env->regs[R_R14];
+ state->gprs[NVMM_X64_GPR_R15] = env->regs[R_R15];
+#endif
+
+ /* RIP and RFLAGS. */
+ state->gprs[NVMM_X64_GPR_RIP] = env->eip;
+ state->gprs[NVMM_X64_GPR_RFLAGS] = env->eflags;
+
+ /* Segments. */
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_CS], &env->segs[R_CS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_DS], &env->segs[R_DS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_ES], &env->segs[R_ES]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_FS], &env->segs[R_FS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_GS], &env->segs[R_GS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_SS], &env->segs[R_SS]);
+
+ /* Special segments. */
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_GDT], &env->gdt);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_LDT], &env->ldt);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_TR], &env->tr);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_IDT], &env->idt);
+
+ /* Control registers. */
+ state->crs[NVMM_X64_CR_CR0] = env->cr[0];
+ state->crs[NVMM_X64_CR_CR2] = env->cr[2];
+ state->crs[NVMM_X64_CR_CR3] = env->cr[3];
+ state->crs[NVMM_X64_CR_CR4] = env->cr[4];
+ state->crs[NVMM_X64_CR_CR8] = qcpu->tpr;
+ state->crs[NVMM_X64_CR_XCR0] = env->xcr0;
+
+ /* Debug registers. */
+ state->drs[NVMM_X64_DR_DR0] = env->dr[0];
+ state->drs[NVMM_X64_DR_DR1] = env->dr[1];
+ state->drs[NVMM_X64_DR_DR2] = env->dr[2];
+ state->drs[NVMM_X64_DR_DR3] = env->dr[3];
+ state->drs[NVMM_X64_DR_DR6] = env->dr[6];
+ state->drs[NVMM_X64_DR_DR7] = env->dr[7];
+
+ /* FPU. */
+ state->fpu.fx_cw = env->fpuc;
+ state->fpu.fx_sw = (env->fpus & ~0x3800) | ((env->fpstt & 0x7) << 11);
+ state->fpu.fx_tw = 0;
+ for (i = 0; i < 8; i++) {
+ state->fpu.fx_tw |= (!env->fptags[i]) << i;
+ }
+ state->fpu.fx_opcode = env->fpop;
+ state->fpu.fx_ip.fa_64 = env->fpip;
+ state->fpu.fx_dp.fa_64 = env->fpdp;
+ state->fpu.fx_mxcsr = env->mxcsr;
+ state->fpu.fx_mxcsr_mask = 0x0000FFFF;
+ assert(sizeof(state->fpu.fx_87_ac) == sizeof(env->fpregs));
+ memcpy(state->fpu.fx_87_ac, env->fpregs, sizeof(env->fpregs));
+ for (i = 0; i < 16; i++) {
+ memcpy(&state->fpu.fx_xmm[i].xmm_bytes[0],
+ &env->xmm_regs[i].ZMM_Q(0), 8);
+ memcpy(&state->fpu.fx_xmm[i].xmm_bytes[8],
+ &env->xmm_regs[i].ZMM_Q(1), 8);
+ }
+
+ /* MSRs. */
+ state->msrs[NVMM_X64_MSR_EFER] = env->efer;
+ state->msrs[NVMM_X64_MSR_STAR] = env->star;
+#ifdef TARGET_X86_64
+ state->msrs[NVMM_X64_MSR_LSTAR] = env->lstar;
+ state->msrs[NVMM_X64_MSR_CSTAR] = env->cstar;
+ state->msrs[NVMM_X64_MSR_SFMASK] = env->fmask;
+ state->msrs[NVMM_X64_MSR_KERNELGSBASE] = env->kernelgsbase;
+#endif
+ state->msrs[NVMM_X64_MSR_SYSENTER_CS] = env->sysenter_cs;
+ state->msrs[NVMM_X64_MSR_SYSENTER_ESP] = env->sysenter_esp;
+ state->msrs[NVMM_X64_MSR_SYSENTER_EIP] = env->sysenter_eip;
+ state->msrs[NVMM_X64_MSR_PAT] = env->pat;
+ state->msrs[NVMM_X64_MSR_TSC] = env->tsc;
+
+ bitmap =
+ NVMM_X64_STATE_SEGS |
+ NVMM_X64_STATE_GPRS |
+ NVMM_X64_STATE_CRS |
+ NVMM_X64_STATE_DRS |
+ NVMM_X64_STATE_MSRS |
+ NVMM_X64_STATE_FPU;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, bitmap);
+ if (ret == -1) {
+ error_report("NVMM: Failed to set virtual processor context,"
+ " error=%d", errno);
+ }
+}
+
+static void
+nvmm_get_segment(SegmentCache *qseg, const struct nvmm_x64_state_seg *nseg)
+{
+ qseg->selector = nseg->selector;
+ qseg->limit = nseg->limit;
+ qseg->base = nseg->base;
+
+ qseg->flags =
+ __SHIFTIN((uint32_t)nseg->attrib.type, DESC_TYPE_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.s, DESC_S_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.dpl, DESC_DPL_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.p, DESC_P_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.avl, DESC_AVL_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.l, DESC_L_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.def, DESC_B_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.g, DESC_G_MASK);
+}
+
+static void
+nvmm_get_registers(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t bitmap, tpr;
+ size_t i;
+ int ret;
+
+ assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
+
+ bitmap =
+ NVMM_X64_STATE_SEGS |
+ NVMM_X64_STATE_GPRS |
+ NVMM_X64_STATE_CRS |
+ NVMM_X64_STATE_DRS |
+ NVMM_X64_STATE_MSRS |
+ NVMM_X64_STATE_FPU;
+
+ ret = nvmm_vcpu_getstate(mach, vcpu, bitmap);
+ if (ret == -1) {
+ error_report("NVMM: Failed to get virtual processor context,"
+ " error=%d", errno);
+ }
+
+ /* GPRs. */
+ env->regs[R_EAX] = state->gprs[NVMM_X64_GPR_RAX];
+ env->regs[R_ECX] = state->gprs[NVMM_X64_GPR_RCX];
+ env->regs[R_EDX] = state->gprs[NVMM_X64_GPR_RDX];
+ env->regs[R_EBX] = state->gprs[NVMM_X64_GPR_RBX];
+ env->regs[R_ESP] = state->gprs[NVMM_X64_GPR_RSP];
+ env->regs[R_EBP] = state->gprs[NVMM_X64_GPR_RBP];
+ env->regs[R_ESI] = state->gprs[NVMM_X64_GPR_RSI];
+ env->regs[R_EDI] = state->gprs[NVMM_X64_GPR_RDI];
+#ifdef TARGET_X86_64
+ env->regs[R_R8] = state->gprs[NVMM_X64_GPR_R8];
+ env->regs[R_R9] = state->gprs[NVMM_X64_GPR_R9];
+ env->regs[R_R10] = state->gprs[NVMM_X64_GPR_R10];
+ env->regs[R_R11] = state->gprs[NVMM_X64_GPR_R11];
+ env->regs[R_R12] = state->gprs[NVMM_X64_GPR_R12];
+ env->regs[R_R13] = state->gprs[NVMM_X64_GPR_R13];
+ env->regs[R_R14] = state->gprs[NVMM_X64_GPR_R14];
+ env->regs[R_R15] = state->gprs[NVMM_X64_GPR_R15];
+#endif
+
+ /* RIP and RFLAGS. */
+ env->eip = state->gprs[NVMM_X64_GPR_RIP];
+ env->eflags = state->gprs[NVMM_X64_GPR_RFLAGS];
+
+ /* Segments. */
+ nvmm_get_segment(&env->segs[R_ES], &state->segs[NVMM_X64_SEG_ES]);
+ nvmm_get_segment(&env->segs[R_CS], &state->segs[NVMM_X64_SEG_CS]);
+ nvmm_get_segment(&env->segs[R_SS], &state->segs[NVMM_X64_SEG_SS]);
+ nvmm_get_segment(&env->segs[R_DS], &state->segs[NVMM_X64_SEG_DS]);
+ nvmm_get_segment(&env->segs[R_FS], &state->segs[NVMM_X64_SEG_FS]);
+ nvmm_get_segment(&env->segs[R_GS], &state->segs[NVMM_X64_SEG_GS]);
+
+ /* Special segments. */
+ nvmm_get_segment(&env->gdt, &state->segs[NVMM_X64_SEG_GDT]);
+ nvmm_get_segment(&env->ldt, &state->segs[NVMM_X64_SEG_LDT]);
+ nvmm_get_segment(&env->tr, &state->segs[NVMM_X64_SEG_TR]);
+ nvmm_get_segment(&env->idt, &state->segs[NVMM_X64_SEG_IDT]);
+
+ /* Control registers. */
+ env->cr[0] = state->crs[NVMM_X64_CR_CR0];
+ env->cr[2] = state->crs[NVMM_X64_CR_CR2];
+ env->cr[3] = state->crs[NVMM_X64_CR_CR3];
+ env->cr[4] = state->crs[NVMM_X64_CR_CR4];
+ tpr = state->crs[NVMM_X64_CR_CR8];
+ if (tpr != qcpu->tpr) {
+ qcpu->tpr = tpr;
+ cpu_set_apic_tpr(x86_cpu->apic_state, tpr);
+ }
+ env->xcr0 = state->crs[NVMM_X64_CR_XCR0];
+
+ /* Debug registers. */
+ env->dr[0] = state->drs[NVMM_X64_DR_DR0];
+ env->dr[1] = state->drs[NVMM_X64_DR_DR1];
+ env->dr[2] = state->drs[NVMM_X64_DR_DR2];
+ env->dr[3] = state->drs[NVMM_X64_DR_DR3];
+ env->dr[6] = state->drs[NVMM_X64_DR_DR6];
+ env->dr[7] = state->drs[NVMM_X64_DR_DR7];
+
+ /* FPU. */
+ env->fpuc = state->fpu.fx_cw;
+ env->fpstt = (state->fpu.fx_sw >> 11) & 0x7;
+ env->fpus = state->fpu.fx_sw & ~0x3800;
+ for (i = 0; i < 8; i++) {
+ env->fptags[i] = !((state->fpu.fx_tw >> i) & 1);
+ }
+ env->fpop = state->fpu.fx_opcode;
+ env->fpip = state->fpu.fx_ip.fa_64;
+ env->fpdp = state->fpu.fx_dp.fa_64;
+ env->mxcsr = state->fpu.fx_mxcsr;
+ assert(sizeof(state->fpu.fx_87_ac) == sizeof(env->fpregs));
+ memcpy(env->fpregs, state->fpu.fx_87_ac, sizeof(env->fpregs));
+ for (i = 0; i < 16; i++) {
+ memcpy(&env->xmm_regs[i].ZMM_Q(0),
+ &state->fpu.fx_xmm[i].xmm_bytes[0], 8);
+ memcpy(&env->xmm_regs[i].ZMM_Q(1),
+ &state->fpu.fx_xmm[i].xmm_bytes[8], 8);
+ }
+
+ /* MSRs. */
+ env->efer = state->msrs[NVMM_X64_MSR_EFER];
+ env->star = state->msrs[NVMM_X64_MSR_STAR];
+#ifdef TARGET_X86_64
+ env->lstar = state->msrs[NVMM_X64_MSR_LSTAR];
+ env->cstar = state->msrs[NVMM_X64_MSR_CSTAR];
+ env->fmask = state->msrs[NVMM_X64_MSR_SFMASK];
+ env->kernelgsbase = state->msrs[NVMM_X64_MSR_KERNELGSBASE];
+#endif
+ env->sysenter_cs = state->msrs[NVMM_X64_MSR_SYSENTER_CS];
+ env->sysenter_esp = state->msrs[NVMM_X64_MSR_SYSENTER_ESP];
+ env->sysenter_eip = state->msrs[NVMM_X64_MSR_SYSENTER_EIP];
+ env->pat = state->msrs[NVMM_X64_MSR_PAT];
+ env->tsc = state->msrs[NVMM_X64_MSR_TSC];
+
+ x86_update_hflags(env);
+}
+
+static bool
+nvmm_can_take_int(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ struct nvmm_machine *mach = get_nvmm_mach();
+
+ if (qcpu->int_window_exit) {
+ return false;
+ }
+
+ if (qcpu->int_shadow || !(env->eflags & IF_MASK)) {
+ struct nvmm_x64_state *state = vcpu->state;
+
+ /* Exit on interrupt window. */
+ nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_INTR);
+ state->intr.int_window_exiting = 1;
+ nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_INTR);
+
+ return false;
+ }
+
+ return true;
+}
+
+static bool
+nvmm_can_take_nmi(CPUState *cpu)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+
+ /*
+ * Contrary to INTs, NMIs always schedule an exit when they are
+ * completed. Therefore, if window-exiting is enabled, it means
+ * NMIs are blocked.
+ */
+ if (qcpu->nmi_window_exit) {
+ return false;
+ }
+
+ return true;
+}
+
+/*
+ * Called before the VCPU is run. We inject events generated by the I/O
+ * thread, and synchronize the guest TPR.
+ */
+static void
+nvmm_vcpu_pre_run(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ struct nvmm_vcpu_event *event = vcpu->event;
+ bool has_event = false;
+ bool sync_tpr = false;
+ uint8_t tpr;
+ int ret;
+
+ qemu_mutex_lock_iothread();
+
+ tpr = cpu_get_apic_tpr(x86_cpu->apic_state);
+ if (tpr != qcpu->tpr) {
+ qcpu->tpr = tpr;
+ sync_tpr = true;
+ }
+
+ /*
+ * Force the VCPU out of its inner loop to process any INIT requests
+ * or commit pending TPR access.
+ */
+ if (cpu->interrupt_request & (CPU_INTERRUPT_INIT | CPU_INTERRUPT_TPR)) {
+ cpu->exit_request = 1;
+ }
+
+ if (!has_event && (cpu->interrupt_request & CPU_INTERRUPT_NMI)) {
+ if (nvmm_can_take_nmi(cpu)) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_NMI;
+ event->type = NVMM_VCPU_EVENT_INTR;
+ event->vector = 2;
+ has_event = true;
+ }
+ }
+
+ if (!has_event && (cpu->interrupt_request & CPU_INTERRUPT_HARD)) {
+ if (nvmm_can_take_int(cpu)) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_HARD;
+ event->type = NVMM_VCPU_EVENT_INTR;
+ event->vector = cpu_get_pic_interrupt(env);
+ has_event = true;
+ }
+ }
+
+ /* Don't want SMIs. */
+ if (cpu->interrupt_request & CPU_INTERRUPT_SMI) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_SMI;
+ }
+
+ if (sync_tpr) {
+ ret = nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_CRS);
+ if (ret == -1) {
+ error_report("NVMM: Failed to get CPU state,"
+ " error=%d", errno);
+ }
+
+ state->crs[NVMM_X64_CR_CR8] = qcpu->tpr;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_CRS);
+ if (ret == -1) {
+ error_report("NVMM: Failed to set CPU state,"
+ " error=%d", errno);
+ }
+ }
+
+ if (has_event) {
+ ret = nvmm_vcpu_inject(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: Failed to inject event,"
+ " error=%d", errno);
+ }
+ }
+
+ qemu_mutex_unlock_iothread();
+}
+
+/*
+ * Called after the VCPU ran. We synchronize the host view of the TPR and
+ * RFLAGS.
+ */
+static void
+nvmm_vcpu_post_run(CPUState *cpu, struct nvmm_vcpu_exit *exit)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ uint64_t tpr;
+
+ env->eflags = exit->exitstate.rflags;
+ qcpu->int_shadow = exit->exitstate.int_shadow;
+ qcpu->int_window_exit = exit->exitstate.int_window_exiting;
+ qcpu->nmi_window_exit = exit->exitstate.nmi_window_exiting;
+
+ tpr = exit->exitstate.cr8;
+ if (qcpu->tpr != tpr) {
+ qcpu->tpr = tpr;
+ qemu_mutex_lock_iothread();
+ cpu_set_apic_tpr(x86_cpu->apic_state, qcpu->tpr);
+ qemu_mutex_unlock_iothread();
+ }
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_io_callback(struct nvmm_io *io)
+{
+ MemTxAttrs attrs = { 0 };
+ int ret;
+
+ ret = address_space_rw(&address_space_io, io->port, attrs, io->data,
+ io->size, !io->in);
+ if (ret != MEMTX_OK) {
+ error_report("NVMM: I/O Transaction Failed "
+ "[%s, port=%u, size=%zu]", (io->in ? "in" : "out"),
+ io->port, io->size);
+ }
+
+ /* Needed, otherwise infinite loop. */
+ current_cpu->vcpu_dirty = false;
+}
+
+static void
+nvmm_mem_callback(struct nvmm_mem *mem)
+{
+ cpu_physical_memory_rw(mem->gpa, mem->data, mem->size, mem->write);
+
+ /* XXX Needed, otherwise infinite loop. */
+ current_cpu->vcpu_dirty = false;
+}
+
+static struct nvmm_assist_callbacks nvmm_callbacks = {
+ .io = nvmm_io_callback,
+ .mem = nvmm_mem_callback
+};
+
+/* -------------------------------------------------------------------------- */
+
+static int
+nvmm_handle_mem(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
+{
+ int ret;
+
+ ret = nvmm_assist_mem(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: Mem Assist Failed [gpa=%p]",
+ (void *)vcpu->exit->u.mem.gpa);
+ }
+
+ return ret;
+}
+
+static int
+nvmm_handle_io(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
+{
+ int ret;
+
+ ret = nvmm_assist_io(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: I/O Assist Failed [port=%d]",
+ (int)vcpu->exit->u.io.port);
+ }
+
+ return ret;
+}
+
+static int
+nvmm_handle_rdmsr(struct nvmm_machine *mach, CPUState *cpu,
+ struct nvmm_vcpu_exit *exit)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t val;
+ int ret;
+
+ switch (exit->u.rdmsr.msr) {
+ case MSR_IA32_APICBASE:
+ val = cpu_get_apic_base(x86_cpu->apic_state);
+ break;
+ case MSR_MTRRcap:
+ case MSR_MTRRdefType:
+ case MSR_MCG_CAP:
+ case MSR_MCG_STATUS:
+ val = 0;
+ break;
+ default: /* More MSRs to add? */
+ val = 0;
+ error_report("NVMM: Unexpected RDMSR 0x%x, ignored",
+ exit->u.rdmsr.msr);
+ break;
+ }
+
+ ret = nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ state->gprs[NVMM_X64_GPR_RAX] = (val & 0xFFFFFFFF);
+ state->gprs[NVMM_X64_GPR_RDX] = (val >> 32);
+ state->gprs[NVMM_X64_GPR_RIP] = exit->u.rdmsr.npc;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ return 0;
+}
+
+static int
+nvmm_handle_wrmsr(struct nvmm_machine *mach, CPUState *cpu,
+ struct nvmm_vcpu_exit *exit)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t val;
+ int ret;
+
+ val = exit->u.wrmsr.val;
+
+ switch (exit->u.wrmsr.msr) {
+ case MSR_IA32_APICBASE:
+ cpu_set_apic_base(x86_cpu->apic_state, val);
+ break;
+ case MSR_MTRRdefType:
+ case MSR_MCG_STATUS:
+ break;
+ default: /* More MSRs to add? */
+ error_report("NVMM: Unexpected WRMSR 0x%x [val=0x%lx], ignored",
+ exit->u.wrmsr.msr, val);
+ break;
+ }
+
+ ret = nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ state->gprs[NVMM_X64_GPR_RIP] = exit->u.wrmsr.npc;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ return 0;
+}
+
+static int
+nvmm_handle_halted(struct nvmm_machine *mach, CPUState *cpu,
+ struct nvmm_vcpu_exit *exit)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ int ret = 0;
+
+ qemu_mutex_lock_iothread();
+
+ if (!((cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
+ (env->eflags & IF_MASK)) &&
+ !(cpu->interrupt_request & CPU_INTERRUPT_NMI)) {
+ cpu->exception_index = EXCP_HLT;
+ cpu->halted = true;
+ ret = 1;
+ }
+
+ qemu_mutex_unlock_iothread();
+
+ return ret;
+}
+
+static int
+nvmm_inject_ud(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
+{
+ struct nvmm_vcpu_event *event = vcpu->event;
+
+ event->type = NVMM_VCPU_EVENT_EXCP;
+ event->vector = 6;
+ event->u.excp.error = 0;
+
+ return nvmm_vcpu_inject(mach, vcpu);
+}
+
+static int
+nvmm_vcpu_loop(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_vcpu_exit *exit = vcpu->exit;
+ int ret;
+
+ /*
+ * Some asynchronous events must be handled outside of the inner
+ * VCPU loop. They are handled here.
+ */
+ if (cpu->interrupt_request & CPU_INTERRUPT_INIT) {
+ nvmm_cpu_synchronize_state(cpu);
+ do_cpu_init(x86_cpu);
+ /* set int/nmi windows back to the reset state */
+ }
+ if (cpu->interrupt_request & CPU_INTERRUPT_POLL) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_POLL;
+ apic_poll_irq(x86_cpu->apic_state);
+ }
+ if (((cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
+ (env->eflags & IF_MASK)) ||
+ (cpu->interrupt_request & CPU_INTERRUPT_NMI)) {
+ cpu->halted = false;
+ }
+ if (cpu->interrupt_request & CPU_INTERRUPT_SIPI) {
+ nvmm_cpu_synchronize_state(cpu);
+ do_cpu_sipi(x86_cpu);
+ }
+ if (cpu->interrupt_request & CPU_INTERRUPT_TPR) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_TPR;
+ nvmm_cpu_synchronize_state(cpu);
+ apic_handle_tpr_access_report(x86_cpu->apic_state, env->eip,
+ env->tpr_access_type);
+ }
+
+ if (cpu->halted) {
+ cpu->exception_index = EXCP_HLT;
+ atomic_set(&cpu->exit_request, false);
+ return 0;
+ }
+
+ qemu_mutex_unlock_iothread();
+ cpu_exec_start(cpu);
+
+ /*
+ * Inner VCPU loop.
+ */
+ do {
+ if (cpu->vcpu_dirty) {
+ nvmm_set_registers(cpu);
+ cpu->vcpu_dirty = false;
+ }
+
+ if (qcpu->stop) {
+ cpu->exception_index = EXCP_INTERRUPT;
+ qcpu->stop = false;
+ ret = 1;
+ break;
+ }
+
+ nvmm_vcpu_pre_run(cpu);
+
+ if (atomic_read(&cpu->exit_request)) {
+ qemu_cpu_kick_self();
+ }
+
+ ret = nvmm_vcpu_run(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: Failed to exec a virtual processor,"
+ " error=%d", errno);
+ break;
+ }
+
+ nvmm_vcpu_post_run(cpu, exit);
+
+ switch (exit->reason) {
+ case NVMM_VCPU_EXIT_NONE:
+ break;
+ case NVMM_VCPU_EXIT_MEMORY:
+ ret = nvmm_handle_mem(mach, vcpu);
+ break;
+ case NVMM_VCPU_EXIT_IO:
+ ret = nvmm_handle_io(mach, vcpu);
+ break;
+ case NVMM_VCPU_EXIT_INT_READY:
+ case NVMM_VCPU_EXIT_NMI_READY:
+ case NVMM_VCPU_EXIT_TPR_CHANGED:
+ break;
+ case NVMM_VCPU_EXIT_HALTED:
+ ret = nvmm_handle_halted(mach, cpu, exit);
+ break;
+ case NVMM_VCPU_EXIT_SHUTDOWN:
+ qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
+ cpu->exception_index = EXCP_INTERRUPT;
+ ret = 1;
+ break;
+ case NVMM_VCPU_EXIT_RDMSR:
+ ret = nvmm_handle_rdmsr(mach, cpu, exit);
+ break;
+ case NVMM_VCPU_EXIT_WRMSR:
+ ret = nvmm_handle_wrmsr(mach, cpu, exit);
+ break;
+ case NVMM_VCPU_EXIT_MONITOR:
+ case NVMM_VCPU_EXIT_MWAIT:
+ ret = nvmm_inject_ud(mach, vcpu);
+ break;
+ default:
+ error_report("NVMM: Unexpected VM exit code 0x%lx [hw=0x%lx]",
+ exit->reason, exit->u.inv.hwcode);
+ nvmm_get_registers(cpu);
+ qemu_mutex_lock_iothread();
+ qemu_system_guest_panicked(cpu_get_crash_info(cpu));
+ qemu_mutex_unlock_iothread();
+ ret = -1;
+ break;
+ }
+ } while (ret == 0);
+
+ cpu_exec_end(cpu);
+ qemu_mutex_lock_iothread();
+ current_cpu = cpu;
+
+ atomic_set(&cpu->exit_request, false);
+
+ return ret < 0;
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+do_nvmm_cpu_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
+{
+ nvmm_get_registers(cpu);
+ cpu->vcpu_dirty = true;
+}
+
+static void
+do_nvmm_cpu_synchronize_post_reset(CPUState *cpu, run_on_cpu_data arg)
+{
+ nvmm_set_registers(cpu);
+ cpu->vcpu_dirty = false;
+}
+
+static void
+do_nvmm_cpu_synchronize_post_init(CPUState *cpu, run_on_cpu_data arg)
+{
+ nvmm_set_registers(cpu);
+ cpu->vcpu_dirty = false;
+}
+
+static void
+do_nvmm_cpu_synchronize_pre_loadvm(CPUState *cpu, run_on_cpu_data arg)
+{
+ cpu->vcpu_dirty = true;
+}
+
+void nvmm_cpu_synchronize_state(CPUState *cpu)
+{
+ if (!cpu->vcpu_dirty) {
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_state, RUN_ON_CPU_NULL);
+ }
+}
+
+void nvmm_cpu_synchronize_post_reset(CPUState *cpu)
+{
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_post_reset, RUN_ON_CPU_NULL);
+}
+
+void nvmm_cpu_synchronize_post_init(CPUState *cpu)
+{
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_post_init, RUN_ON_CPU_NULL);
+}
+
+void nvmm_cpu_synchronize_pre_loadvm(CPUState *cpu)
+{
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_pre_loadvm, RUN_ON_CPU_NULL);
+}
+
+/* -------------------------------------------------------------------------- */
+
+static Error *nvmm_migration_blocker;
+
+static void
+nvmm_ipi_signal(int sigcpu)
+{
+ struct qemu_vcpu *qcpu;
+
+ if (current_cpu) {
+ qcpu = get_qemu_vcpu(current_cpu);
+ qcpu->stop = true;
+ }
+}
+
+static void
+nvmm_init_cpu_signals(void)
+{
+ struct sigaction sigact;
+ sigset_t set;
+
+ /* Install the IPI handler. */
+ memset(&sigact, 0, sizeof(sigact));
+ sigact.sa_handler = nvmm_ipi_signal;
+ sigaction(SIG_IPI, &sigact, NULL);
+
+ /* Allow IPIs on the current thread. */
+ sigprocmask(SIG_BLOCK, NULL, &set);
+ sigdelset(&set, SIG_IPI);
+ pthread_sigmask(SIG_SETMASK, &set, NULL);
+}
+
+int
+nvmm_init_vcpu(CPUState *cpu)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct nvmm_vcpu_conf_cpuid cpuid;
+ struct nvmm_vcpu_conf_tpr tpr;
+ Error *local_error = NULL;
+ struct qemu_vcpu *qcpu;
+ int ret, err;
+
+ nvmm_init_cpu_signals();
+
+ if (nvmm_migration_blocker == NULL) {
+ error_setg(&nvmm_migration_blocker,
+ "NVMM: Migration not supported");
+
+ (void)migrate_add_blocker(nvmm_migration_blocker, &local_error);
+ if (local_error) {
+ error_report_err(local_error);
+ migrate_del_blocker(nvmm_migration_blocker);
+ error_free(nvmm_migration_blocker);
+ return -EINVAL;
+ }
+ }
+
+ qcpu = g_malloc0(sizeof(*qcpu));
+ if (qcpu == NULL) {
+ error_report("NVMM: Failed to allocate VCPU context.");
+ return -ENOMEM;
+ }
+
+ ret = nvmm_vcpu_create(mach, cpu->cpu_index, &qcpu->vcpu);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to create a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+
+ memset(&cpuid, 0, sizeof(cpuid));
+ cpuid.mask = 1;
+ cpuid.leaf = 0x00000001;
+ cpuid.u.mask.set.edx = CPUID_MCE | CPUID_MCA | CPUID_MTRR;
+ ret = nvmm_vcpu_configure(mach, &qcpu->vcpu, NVMM_VCPU_CONF_CPUID,
+ &cpuid);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to configure a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+
+ ret = nvmm_vcpu_configure(mach, &qcpu->vcpu, NVMM_VCPU_CONF_CALLBACKS,
+ &nvmm_callbacks);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to configure a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+
+ if (qemu_mach.cap.arch.vcpu_conf_support & NVMM_CAP_ARCH_VCPU_CONF_TPR) {
+ memset(&tpr, 0, sizeof(tpr));
+ tpr.exit_changed = 1;
+ ret = nvmm_vcpu_configure(mach, &qcpu->vcpu, NVMM_VCPU_CONF_TPR, &tpr);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to configure a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+ }
+
+ cpu->vcpu_dirty = true;
+ cpu->hax_vcpu = (struct hax_vcpu_state *)qcpu;
+
+ return 0;
+}
+
+int
+nvmm_vcpu_exec(CPUState *cpu)
+{
+ int ret, fatal;
+
+ while (1) {
+ if (cpu->exception_index >= EXCP_INTERRUPT) {
+ ret = cpu->exception_index;
+ cpu->exception_index = -1;
+ break;
+ }
+
+ fatal = nvmm_vcpu_loop(cpu);
+
+ if (fatal) {
+ error_report("NVMM: Failed to execute a VCPU.");
+ abort();
+ }
+ }
+
+ return ret;
+}
+
+void
+nvmm_destroy_vcpu(CPUState *cpu)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+
+ nvmm_vcpu_destroy(mach, &qcpu->vcpu);
+ g_free(cpu->hax_vcpu);
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_update_mapping(hwaddr start_pa, ram_addr_t size, uintptr_t hva,
+ bool add, bool rom, const char *name)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ int ret, prot;
+
+ if (add) {
+ prot = PROT_READ | PROT_EXEC;
+ if (!rom) {
+ prot |= PROT_WRITE;
+ }
+ ret = nvmm_gpa_map(mach, hva, start_pa, size, prot);
+ } else {
+ ret = nvmm_gpa_unmap(mach, hva, start_pa, size);
+ }
+
+ if (ret == -1) {
+ error_report("NVMM: Failed to %s GPA range '%s' PA:%p, "
+ "Size:%p bytes, HostVA:%p, error=%d",
+ (add ? "map" : "unmap"), name, (void *)(uintptr_t)start_pa,
+ (void *)size, (void *)hva, errno);
+ }
+}
+
+static void
+nvmm_process_section(MemoryRegionSection *section, int add)
+{
+ MemoryRegion *mr = section->mr;
+ hwaddr start_pa = section->offset_within_address_space;
+ ram_addr_t size = int128_get64(section->size);
+ unsigned int delta;
+ uintptr_t hva;
+
+ if (!memory_region_is_ram(mr)) {
+ return;
+ }
+
+ /* Adjust start_pa and size so that they are page-aligned. */
+ delta = qemu_real_host_page_size - (start_pa & ~qemu_real_host_page_mask);
+ delta &= ~qemu_real_host_page_mask;
+ if (delta > size) {
+ return;
+ }
+ start_pa += delta;
+ size -= delta;
+ size &= qemu_real_host_page_mask;
+ if (!size || (start_pa & ~qemu_real_host_page_mask)) {
+ return;
+ }
+
+ hva = (uintptr_t)memory_region_get_ram_ptr(mr) +
+ section->offset_within_region + delta;
+
+ nvmm_update_mapping(start_pa, size, hva, add,
+ memory_region_is_rom(mr), mr->name);
+}
+
+static void
+nvmm_region_add(MemoryListener *listener, MemoryRegionSection *section)
+{
+ memory_region_ref(section->mr);
+ nvmm_process_section(section, 1);
+}
+
+static void
+nvmm_region_del(MemoryListener *listener, MemoryRegionSection *section)
+{
+ nvmm_process_section(section, 0);
+ memory_region_unref(section->mr);
+}
+
+static void
+nvmm_transaction_begin(MemoryListener *listener)
+{
+ /* nothing */
+}
+
+static void
+nvmm_transaction_commit(MemoryListener *listener)
+{
+ /* nothing */
+}
+
+static void
+nvmm_log_sync(MemoryListener *listener, MemoryRegionSection *section)
+{
+ MemoryRegion *mr = section->mr;
+
+ if (!memory_region_is_ram(mr)) {
+ return;
+ }
+
+ memory_region_set_dirty(mr, 0, int128_get64(section->size));
+}
+
+static MemoryListener nvmm_memory_listener = {
+ .begin = nvmm_transaction_begin,
+ .commit = nvmm_transaction_commit,
+ .region_add = nvmm_region_add,
+ .region_del = nvmm_region_del,
+ .log_sync = nvmm_log_sync,
+ .priority = 10,
+};
+
+static void
+nvmm_ram_block_added(RAMBlockNotifier *n, void *host, size_t size)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ uintptr_t hva = (uintptr_t)host;
+ int ret;
+
+ ret = nvmm_hva_map(mach, hva, size);
+
+ if (ret == -1) {
+ error_report("NVMM: Failed to map HVA, HostVA:%p "
+ "Size:%p bytes, error=%d",
+ (void *)hva, (void *)size, errno);
+ }
+}
+
+static struct RAMBlockNotifier nvmm_ram_notifier = {
+ .ram_block_added = nvmm_ram_block_added
+};
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_handle_interrupt(CPUState *cpu, int mask)
+{
+ cpu->interrupt_request |= mask;
+
+ if (!qemu_cpu_is_self(cpu)) {
+ qemu_cpu_kick(cpu);
+ }
+}
+
+/* -------------------------------------------------------------------------- */
+
+static int
+nvmm_accel_init(MachineState *ms)
+{
+ int ret, err;
+
+ ret = nvmm_init();
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Initialization failed, error=%d", errno);
+ return -err;
+ }
+
+ ret = nvmm_capability(&qemu_mach.cap);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Unable to fetch capability, error=%d", errno);
+ return -err;
+ }
+ if (qemu_mach.cap.version != 1) {
+ error_report("NVMM: Unsupported version %u", qemu_mach.cap.version);
+ return -EPROGMISMATCH;
+ }
+ if (qemu_mach.cap.state_size != sizeof(struct nvmm_x64_state)) {
+ error_report("NVMM: Wrong state size %u", qemu_mach.cap.state_size);
+ return -EPROGMISMATCH;
+ }
+
+ ret = nvmm_machine_create(&qemu_mach.mach);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Machine creation failed, error=%d", errno);
+ return -err;
+ }
+
+ memory_listener_register(&nvmm_memory_listener, &address_space_memory);
+ ram_block_notifier_add(&nvmm_ram_notifier);
+
+ cpu_interrupt_handler = nvmm_handle_interrupt;
+
+ printf("NetBSD Virtual Machine Monitor accelerator is operational\n");
+ return 0;
+}
+
+int
+nvmm_enabled(void)
+{
+ return nvmm_allowed;
+}
+
+static void
+nvmm_accel_class_init(ObjectClass *oc, void *data)
+{
+ AccelClass *ac = ACCEL_CLASS(oc);
+ ac->name = "NVMM";
+ ac->init_machine = nvmm_accel_init;
+ ac->allowed = &nvmm_allowed;
+}
+
+static const TypeInfo nvmm_accel_type = {
+ .name = ACCEL_CLASS_NAME("nvmm"),
+ .parent = TYPE_ACCEL,
+ .class_init = nvmm_accel_class_init,
+};
+
+static void
+nvmm_type_init(void)
+{
+ type_register_static(&nvmm_accel_type);
+}
+
+type_init(nvmm_type_init);
--
2.25.0
On 06/02/20 22:32, Kamil Rytarowski wrote:
> +get_qemu_vcpu(CPUState *cpu)
> +{
> + return (struct qemu_vcpu *)cpu->hax_vcpu;
> +}
Please make hax_vcpu a void * and rename it to "accel_data".
> + nseg->attrib.g = __SHIFTOUT(attrib, DESC_G_MASK);
> + __SHIFTIN((uint32_t)nseg->attrib.g, DESC_G_MASK);
What are __SHIFTOUT and __SHIFTIN?
>
> + if (qcpu->int_window_exit) {
Should it assert the condition in the "if" below?
> + return false;
> + }
> +
> + if (qcpu->int_shadow || !(env->eflags & IF_MASK)) {
> + struct nvmm_x64_state *state = vcpu->state;
> +
> + /* Exit on interrupt window. */
> + nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_INTR);
> + state->intr.int_window_exiting = 1;
> + nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_INTR);
... and should this set qcpu->int_window_exit?
> +
> + return false;
> + }
Have you tried running kvm-unit-tests?
> +
> + /* Needed, otherwise infinite loop. */
> + current_cpu->vcpu_dirty = false;
Can you explain this?
> + break;
> + default: /* More MSRs to add? */
> + val = 0;
I would add at least MSR_IA32_TSC.
>
> +
> + if (qcpu->stop) {
> + cpu->exception_index = EXCP_INTERRUPT;
> + qcpu->stop = false;
> + ret = 1;
> + break;
> + }
> +
> + nvmm_vcpu_pre_run(cpu);
> +
> + if (atomic_read(&cpu->exit_request)) {
> + qemu_cpu_kick_self();
> + }
> +
This is racy without something like KVM's immediate_exit mechanism.
This should be fixed in NVMM.
Paolo
Le 02/03/2020 à 19:13, Paolo Bonzini a écrit :
> On 06/02/20 22:32, Kamil Rytarowski wrote:
>> +get_qemu_vcpu(CPUState *cpu)
>> +{
>> + return (struct qemu_vcpu *)cpu->hax_vcpu;
>> +}
>
> Please make hax_vcpu a void * and rename it to "accel_data".
NVMM reproduces the existing logic in the other accelerators. I agree
that it should be "accel_data" with void *, but that should be done
in all accelerators in a separate commit, unrelated to NVMM.
>> + nseg->attrib.g = __SHIFTOUT(attrib, DESC_G_MASK);
>
>> + __SHIFTIN((uint32_t)nseg->attrib.g, DESC_G_MASK);
>
> What are __SHIFTOUT and __SHIFTIN?
They are macros in NetBSD.
>> + if (qcpu->int_window_exit) {
>
> Should it assert the condition in the "if" below?
No, because if int_window_exit is set, then state->intr.int_window_exiting
is set too, so there is no point doing get+set.
>> + return false;
>> + }
>> +
>> + if (qcpu->int_shadow || !(env->eflags & IF_MASK)) {
>> + struct nvmm_x64_state *state = vcpu->state;
>> +
>> + /* Exit on interrupt window. */
>> + nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_INTR);
>> + state->intr.int_window_exiting = 1;
>> + nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_INTR);
>
> ... and should this set qcpu->int_window_exit?
Mmh, maybe. Not a big problem though, because at worst it just means we
set int_window_exiting to one while it was already one. I'll think about
that for a future commit. (I'm not immediately able to test.)
>> +
>> + return false;
>> + }
>
> Have you tried running kvm-unit-tests?
I didn't know kvm-unit-tests (until now). I developed my own tests and
ran them in qemu-nvmm. But good to know, I'll try these tests.
>> + /* Needed, otherwise infinite loop. */
>> + current_cpu->vcpu_dirty = false;
>
> Can you explain this?
If vcpu_dirty remains true, we land here in the next iteration of the
loop:
if (cpu->vcpu_dirty) {
nvmm_set_registers(cpu);
cpu->vcpu_dirty = false;
}
And the (now updated) register values are lost. The guest stays on the
same instruction.
>> + break;
>> + default: /* More MSRs to add? */
>> + val = 0;
>
> I would add at least MSR_IA32_TSC.
MSR_IA32_TSC is handled by the kernel, that's why it isn't there. The
values do get synced:
state->msrs[NVMM_X64_MSR_TSC] = env->tsc;
...
env->tsc = state->msrs[NVMM_X64_MSR_TSC];
>> + if (qcpu->stop) {
>> + cpu->exception_index = EXCP_INTERRUPT;
>> + qcpu->stop = false;
>> + ret = 1;
>> + break;
>> + }
>> +
>> + nvmm_vcpu_pre_run(cpu);
>> +
>> + if (atomic_read(&cpu->exit_request)) {
>> + qemu_cpu_kick_self();
>> + }
>> +
>
> This is racy without something like KVM's immediate_exit mechanism.
> This should be fixed in NVMM.
I don't immediately see how this is racy. It reproduces the existing
logic found in whpx-all.c, and if there is a real problem it can be
fixed in a future commit along with WHPX.
Maxime
Il lun 2 mar 2020, 20:28 Maxime Villard <max@m00nbsd.net> ha scritto:
>
> >> + nvmm_vcpu_pre_run(cpu);
> >> +
> >> + if (atomic_read(&cpu->exit_request)) {
> >> + qemu_cpu_kick_self();
> >> + }
> >> +
> >
> > This is racy without something like KVM's immediate_exit mechanism.
> > This should be fixed in NVMM.
>
> I don't immediately see how this is racy.
You can get an IPI signal immediately after reading cpu->exit_request.
It reproduces the existing
> logic found in whpx-all.c, and if there is a real problem it can be
> fixed in a future commit along with WHPX.
>
It's buggy there too and it has to be fixed in the hypervisor so it can't
be done at the same time I'm both. KVM does it right by having a flag
("immediate_exit") that is set by the signal handler and checked by the
hypervisor.
An earlier version of KVM instead atomically unblocked the signal while
executing the guest, and then ate it with a sigwaitinfo after exiting back
to userspace.
You don't have to fix it immediately, but adding a FIXME would be a good
idea.
Paolo
> Maxime
>
>
Le 02/03/2020 à 20:35, Paolo Bonzini a écrit :
>
>
> Il lun 2 mar 2020, 20:28 Maxime Villard <max@m00nbsd.net <mailto:max@m00nbsd.net>> ha scritto:
>
>
> >> + nvmm_vcpu_pre_run(cpu);
> >> +
> >> + if (atomic_read(&cpu->exit_request)) {
> >> + qemu_cpu_kick_self();
> >> + }
> >> +
> >
> > This is racy without something like KVM's immediate_exit mechanism.
> > This should be fixed in NVMM.
>
> I don't immediately see how this is racy.
>
>
> You can get an IPI signal immediately after reading cpu->exit_request.
>
> It reproduces the existing
> logic found in whpx-all.c, and if there is a real problem it can be
> fixed in a future commit along with WHPX.
>
>
> It's buggy there too and it has to be fixed in the hypervisor so it can't be done at the same time I'm both. KVM does it right by having a flag ("immediate_exit") that is set by the signal handler and checked by the hypervisor.
>
> An earlier version of KVM instead atomically unblocked the signal while executing the guest, and then ate it with a sigwaitinfo after exiting back to userspace.
>
> You don't have to fix it immediately, but adding a FIXME would be a good idea.
>
> Paolo
Kamil, please add /* FIXME: possible race here */ before the atomic_read().
Thanks
On 10.03.2020 07:45, Maxime Villard wrote:
> Le 02/03/2020 à 20:35, Paolo Bonzini a écrit :
>>
>>
>> Il lun 2 mar 2020, 20:28 Maxime Villard <max@m00nbsd.net <mailto:max@m00nbsd.net>> ha scritto:
>>
>>
>> >> + nvmm_vcpu_pre_run(cpu);
>> >> +
>> >> + if (atomic_read(&cpu->exit_request)) {
>> >> + qemu_cpu_kick_self();
>> >> + }
>> >> +
>> >
>> > This is racy without something like KVM's immediate_exit mechanism.
>> > This should be fixed in NVMM.
>>
>> I don't immediately see how this is racy.
>>
>>
>> You can get an IPI signal immediately after reading cpu->exit_request.
>>
>> It reproduces the existing
>> logic found in whpx-all.c, and if there is a real problem it can be
>> fixed in a future commit along with WHPX.
>>
>>
>> It's buggy there too and it has to be fixed in the hypervisor so it can't be done at the same time I'm both. KVM does it right by having a flag ("immediate_exit") that is set by the signal handler and checked by the hypervisor.
>>
>> An earlier version of KVM instead atomically unblocked the signal while executing the guest, and then ate it with a sigwaitinfo after exiting back to userspace.
>>
>> You don't have to fix it immediately, but adding a FIXME would be a good idea.
>>
>> Paolo
>
> Kamil, please add /* FIXME: possible race here */ before the atomic_read().
>
> Thanks
>
I will do it and submit a new patchset revision.
On 10/03/20 07:45, Maxime Villard wrote:
> > It reproduces the existing logic found in whpx-all.c, and if there is
> >
>
> It's buggy there too and it has to be fixed in the hypervisor so it
> can't be done at the same time I'm both. KVM does it right by having
> a flag ("immediate_exit") that is set by the signal handler and
> checked by the hypervisor.
For what it's worth, WHPX's whpx_kick_vcpu invokes a "cancel entry" API
that is probably similar to what KVM does, so there's nothing to do there.
Paolo
Le 10/03/2020 à 11:58, Paolo Bonzini a écrit :
> On 10/03/20 07:45, Maxime Villard wrote:
>>> It reproduces the existing logic found in whpx-all.c, and if there is
>>>
>>
>> It's buggy there too and it has to be fixed in the hypervisor so it
>> can't be done at the same time I'm both. KVM does it right by having
>> a flag ("immediate_exit") that is set by the signal handler and
>> checked by the hypervisor.
>
> For what it's worth, WHPX's whpx_kick_vcpu invokes a "cancel entry" API
> that is probably similar to what KVM does, so there's nothing to do there.
>
> Paolo
Having had some time to look at the actual code today, I'm wondering why
WHPX uses whpx_vcpu_kick(), and why I even use qemu_cpu_kick_self() in
NVMM.
whpx_vcpu_kick() cancels the RunVirtualProcessor on the given VCPU. But
that VCPU is not executing anyway, since we're the calling thread, and if
we are in userland then we cannot be in the guest. It seems to me the code
should just break the loop directly instead of doing a self-kick, given
that it is known that the VCPU is not executing.
Maybe, whpx_vcpu_kick() causes a WHvRunVpExitReasonCanceled in the
WHvRunVirtualProcessor() call that follows, which in turn causes "ret=1"
to leave the loop. That is, maybe the next WHvRunVirtualProcessor() acks
the cancellation and leaves without doing anything, even if the
cancellation was received when this function wasn't executing. So there is
no bad effect, given that we still end up leaving the loop, which is the
desired functional behavior.
Looking at NVMM now, it seems to me there is the same thing. We do a
self-kick but we're the calling thread and know the VCPU isn't executing.
As a result of the self-kick the IPI handler sets
qcpu->stop = true;
And in the next iteration of the loop, we break because this bool is set.
Again, we took a difficult path, but this is the desired final behavior.
In both WHPX and NVMM, it seems that we are doing unnecessary self-kicks,
and could just break the loop right away.
What am I getting wrong? The thread logic in Qemu is not really easy to
grasp..
Thanks,
Maxime
On 10/03/20 20:14, Maxime Villard wrote: > Maybe, whpx_vcpu_kick() causes a WHvRunVpExitReasonCanceled in the > WHvRunVirtualProcessor() call that follows, which in turn causes "ret=1" > to leave the loop. That is, maybe the next WHvRunVirtualProcessor() acks > the cancellation and leaves without doing anything, even if the > cancellation was received when this function wasn't executing. So there is > no bad effect, given that we still end up leaving the loop, which is the > desired functional behavior. Yes, that's exactly the effect, and it solves the race in the same way as KVM's run->immediate_exit flag. > Looking at NVMM now, it seems to me there is the same thing. We do a > self-kick but we're the calling thread and know the VCPU isn't executing. > As a result of the self-kick the IPI handler sets > qcpu->stop = true; > And in the next iteration of the loop, we break because this bool is set The problem is that qcpu->stop is checked _before_ entering the hypervisor and not after, so there is a small race window. Paolo
Le 11/03/2020 à 19:03, Paolo Bonzini a écrit : > On 10/03/20 20:14, Maxime Villard wrote: >> Maybe, whpx_vcpu_kick() causes a WHvRunVpExitReasonCanceled in the >> WHvRunVirtualProcessor() call that follows, which in turn causes "ret=1" >> to leave the loop. That is, maybe the next WHvRunVirtualProcessor() acks >> the cancellation and leaves without doing anything, even if the >> cancellation was received when this function wasn't executing. So there is >> no bad effect, given that we still end up leaving the loop, which is the >> desired functional behavior. > > Yes, that's exactly the effect, and it solves the race in the same way > as KVM's run->immediate_exit flag. > >> Looking at NVMM now, it seems to me there is the same thing. We do a >> self-kick but we're the calling thread and know the VCPU isn't executing. >> As a result of the self-kick the IPI handler sets >> qcpu->stop = true; >> And in the next iteration of the loop, we break because this bool is set > > The problem is that qcpu->stop is checked _before_ entering the > hypervisor and not after, so there is a small race window. Ok. I don't understand what's supposed to be the race here. If we get an IPI between the check and the call to nvmm_vcpu_run() then we'll just do one run and stop in the next iteration, because the IPI will have set qcpu->stop. Is this extra iteration undesired?
On 11/03/20 21:14, Maxime Villard wrote: >> The problem is that qcpu->stop is checked _before_ entering the >> hypervisor and not after, so there is a small race window. > Ok. I don't understand what's supposed to be the race here. If we get an > IPI between the check and the call to nvmm_vcpu_run() then we'll just do > one run and stop in the next iteration, because the IPI will have set > qcpu->stop. Is this extra iteration undesired? Yes, you don't know how long that run would take. I don't know about NVMM but for KVM it may even never leave if the guest is in HLT state. Paolo
Le 11/03/2020 à 21:42, Paolo Bonzini a écrit :
> On 11/03/20 21:14, Maxime Villard wrote:
>>> The problem is that qcpu->stop is checked _before_ entering the
>>> hypervisor and not after, so there is a small race window.
>> Ok. I don't understand what's supposed to be the race here. If we get an
>> IPI between the check and the call to nvmm_vcpu_run() then we'll just do
>> one run and stop in the next iteration, because the IPI will have set
>> qcpu->stop. Is this extra iteration undesired?
>
> Yes, you don't know how long that run would take. I don't know about
> NVMM but for KVM it may even never leave if the guest is in HLT state.
Ok, I see, thanks.
In NVMM the runs are short, the syscalls are fast, and pending signals
cause returns to userland. Therefore, in practice, it's not a big problem,
because (1) the window is small and (2) if we have a miss it's not going
to take long to come back to Qemu.
I see a quick kernel change I can make to reduce 95% of the window
already in the current state. The remaining 5% will need a new
nvmm_vcpu_kick() function.
For now this issue is unimportant and no Qemu change is required.
Kamil, please also drop the XXX in
/* XXX Needed, otherwise infinite loop. */
It's not a bug.
Thanks,
Maxime
On 11.03.2020 22:21, Maxime Villard wrote: > Le 11/03/2020 à 21:42, Paolo Bonzini a écrit : >> On 11/03/20 21:14, Maxime Villard wrote: >>>> The problem is that qcpu->stop is checked _before_ entering the >>>> hypervisor and not after, so there is a small race window. >>> Ok. I don't understand what's supposed to be the race here. If we get an >>> IPI between the check and the call to nvmm_vcpu_run() then we'll just do >>> one run and stop in the next iteration, because the IPI will have set >>> qcpu->stop. Is this extra iteration undesired? >> >> Yes, you don't know how long that run would take. I don't know about >> NVMM but for KVM it may even never leave if the guest is in HLT state. > > Ok, I see, thanks. > > In NVMM the runs are short, the syscalls are fast, and pending signals > cause returns to userland. Therefore, in practice, it's not a big problem, > because (1) the window is small and (2) if we have a miss it's not going > to take long to come back to Qemu. > > I see a quick kernel change I can make to reduce 95% of the window > already in the current state. The remaining 5% will need a new > nvmm_vcpu_kick() function. > > For now this issue is unimportant and no Qemu change is required. > > Kamil, please also drop the XXX in > /* XXX Needed, otherwise infinite loop. */ > It's not a bug. > OK. I will do it. > Thanks, > Maxime >
On 11/03/20 22:21, Maxime Villard wrote: >> Yes, you don't know how long that run would take. I don't know about >> NVMM but for KVM it may even never leave if the guest is in HLT state. > Ok, I see, thanks. > > In NVMM the runs are short How do you ensure that a guest with interrupts off exits promptly? > , the syscalls are fast, and pending signals > cause returns to userland. Therefore, in practice, it's not a big problem, > because (1) the window is small and (2) if we have a miss it's not going > to take long to come back to Qemu. > > I see a quick kernel change I can make to reduce 95% of the window > already in the current state. The remaining 5% will need a new > nvmm_vcpu_kick() function. You can also do what KVM did until a few years ago: swap the signal mask atomically when you enter the hypervisor (e.g. unmasking SIGUSR1---this has to be done in the kernel) and when you leave it. Then in QEMU you keep SIGUSR1 masked and "eat" it with sigwaitinfo. > For now this issue is unimportant and no Qemu change is required. If you say so. Paolo
Le 11/03/2020 à 22:44, Paolo Bonzini a écrit : > On 11/03/20 22:21, Maxime Villard wrote: >>> Yes, you don't know how long that run would take. I don't know about >>> NVMM but for KVM it may even never leave if the guest is in HLT state. >> Ok, I see, thanks. >> >> In NVMM the runs are short > > How do you ensure that a guest with interrupts off exits promptly? In NVMM there are several conditions unrelated to the guest state which cause returns to userland. These are reschedulings, signals and softints. They happen "regularly". As the man page states: "this gives a chance for emulator software to halt the VM in its tracks". There was a specific reason this design was chosen, but it's true that a nvmm_vcpu_kick() is more precise and warranted here. >> , the syscalls are fast, and pending signals >> cause returns to userland. Therefore, in practice, it's not a big problem, >> because (1) the window is small and (2) if we have a miss it's not going >> to take long to come back to Qemu. >> >> I see a quick kernel change I can make to reduce 95% of the window >> already in the current state. The remaining 5% will need a new >> nvmm_vcpu_kick() function. > > You can also do what KVM did until a few years ago: swap the signal mask > atomically when you enter the hypervisor (e.g. unmasking SIGUSR1---this > has to be done in the kernel) and when you leave it. Then in QEMU you > keep SIGUSR1 masked and "eat" it with sigwaitinfo. > >> For now this issue is unimportant and no Qemu change is required. > > If you say so. At first I thought the race was an actual locking problem. In fact it's just a delay which on NVMM happens to be small, so yeah, not a very important issue, it will be addressed in a future patch set soon. Thanks, Maxime
On 10.03.2020 07:45, Maxime Villard wrote:
> Le 02/03/2020 à 20:35, Paolo Bonzini a écrit :
>>
>>
>> Il lun 2 mar 2020, 20:28 Maxime Villard <max@m00nbsd.net <mailto:max@m00nbsd.net>> ha scritto:
>>
>>
>> >> + nvmm_vcpu_pre_run(cpu);
>> >> +
>> >> + if (atomic_read(&cpu->exit_request)) {
>> >> + qemu_cpu_kick_self();
>> >> + }
>> >> +
>> >
>> > This is racy without something like KVM's immediate_exit mechanism.
>> > This should be fixed in NVMM.
>>
>> I don't immediately see how this is racy.
>>
>>
>> You can get an IPI signal immediately after reading cpu->exit_request.
>>
>> It reproduces the existing
>> logic found in whpx-all.c, and if there is a real problem it can be
>> fixed in a future commit along with WHPX.
>>
>>
>> It's buggy there too and it has to be fixed in the hypervisor so it can't be done at the same time I'm both. KVM does it right by having a flag ("immediate_exit") that is set by the signal handler and checked by the hypervisor.
>>
>> An earlier version of KVM instead atomically unblocked the signal while executing the guest, and then ate it with a sigwaitinfo after exiting back to userspace.
>>
>> You don't have to fix it immediately, but adding a FIXME would be a good idea.
>>
>> Paolo
>
> Kamil, please add /* FIXME: possible race here */ before the atomic_read().
>
> Thanks
>
So, is this still considered as a possible race? Were there any other
changes to be introduced?
Can we see this patchset merged?
From: Maxime Villard <max@m00nbsd.net>
Implements the NetBSD Virtual Machine Monitor (NVMM) target. Which
acts as a hypervisor accelerator for QEMU on the NetBSD platform. This enables
QEMU much greater speed over the emulated x86_64 path's that are taken on
NetBSD today.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
target/i386/Makefile.objs | 1 +
target/i386/nvmm-all.c | 1226 +++++++++++++++++++++++++++++++++++++
2 files changed, 1227 insertions(+)
create mode 100644 target/i386/nvmm-all.c
diff --git a/target/i386/Makefile.objs b/target/i386/Makefile.objs
index 48e0c28434..bdcdb32e93 100644
--- a/target/i386/Makefile.objs
+++ b/target/i386/Makefile.objs
@@ -17,6 +17,7 @@ obj-$(CONFIG_HAX) += hax-all.o hax-mem.o hax-posix.o
endif
obj-$(CONFIG_HVF) += hvf/
obj-$(CONFIG_WHPX) += whpx-all.o
+obj-$(CONFIG_NVMM) += nvmm-all.o
endif
obj-$(CONFIG_SEV) += sev.o
obj-$(call lnot,$(CONFIG_SEV)) += sev-stub.o
diff --git a/target/i386/nvmm-all.c b/target/i386/nvmm-all.c
new file mode 100644
index 0000000000..b3f1c11984
--- /dev/null
+++ b/target/i386/nvmm-all.c
@@ -0,0 +1,1226 @@
+/*
+ * Copyright (c) 2018-2019 Maxime Villard, All rights reserved.
+ *
+ * NetBSD Virtual Machine Monitor (NVMM) accelerator for QEMU.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "cpu.h"
+#include "exec/address-spaces.h"
+#include "exec/ioport.h"
+#include "qemu-common.h"
+#include "strings.h"
+#include "sysemu/accel.h"
+#include "sysemu/nvmm.h"
+#include "sysemu/runstate.h"
+#include "sysemu/sysemu.h"
+#include "sysemu/cpus.h"
+#include "qemu/main-loop.h"
+#include "qemu/error-report.h"
+#include "qemu/queue.h"
+#include "qapi/error.h"
+#include "migration/blocker.h"
+
+#include <nvmm.h>
+
+struct qemu_vcpu {
+ struct nvmm_vcpu vcpu;
+ uint8_t tpr;
+ bool stop;
+
+ /* Window-exiting for INTs/NMIs. */
+ bool int_window_exit;
+ bool nmi_window_exit;
+
+ /* The guest is in an interrupt shadow (POP SS, etc). */
+ bool int_shadow;
+};
+
+struct qemu_machine {
+ struct nvmm_capability cap;
+ struct nvmm_machine mach;
+};
+
+/* -------------------------------------------------------------------------- */
+
+static bool nvmm_allowed;
+static struct qemu_machine qemu_mach;
+
+static struct qemu_vcpu *
+get_qemu_vcpu(CPUState *cpu)
+{
+ return (struct qemu_vcpu *)cpu->hax_vcpu;
+}
+
+static struct nvmm_machine *
+get_nvmm_mach(void)
+{
+ return &qemu_mach.mach;
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_set_segment(struct nvmm_x64_state_seg *nseg, const SegmentCache *qseg)
+{
+ uint32_t attrib = qseg->flags;
+
+ nseg->selector = qseg->selector;
+ nseg->limit = qseg->limit;
+ nseg->base = qseg->base;
+ nseg->attrib.type = __SHIFTOUT(attrib, DESC_TYPE_MASK);
+ nseg->attrib.s = __SHIFTOUT(attrib, DESC_S_MASK);
+ nseg->attrib.dpl = __SHIFTOUT(attrib, DESC_DPL_MASK);
+ nseg->attrib.p = __SHIFTOUT(attrib, DESC_P_MASK);
+ nseg->attrib.avl = __SHIFTOUT(attrib, DESC_AVL_MASK);
+ nseg->attrib.l = __SHIFTOUT(attrib, DESC_L_MASK);
+ nseg->attrib.def = __SHIFTOUT(attrib, DESC_B_MASK);
+ nseg->attrib.g = __SHIFTOUT(attrib, DESC_G_MASK);
+}
+
+static void
+nvmm_set_registers(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t bitmap;
+ size_t i;
+ int ret;
+
+ assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
+
+ /* GPRs. */
+ state->gprs[NVMM_X64_GPR_RAX] = env->regs[R_EAX];
+ state->gprs[NVMM_X64_GPR_RCX] = env->regs[R_ECX];
+ state->gprs[NVMM_X64_GPR_RDX] = env->regs[R_EDX];
+ state->gprs[NVMM_X64_GPR_RBX] = env->regs[R_EBX];
+ state->gprs[NVMM_X64_GPR_RSP] = env->regs[R_ESP];
+ state->gprs[NVMM_X64_GPR_RBP] = env->regs[R_EBP];
+ state->gprs[NVMM_X64_GPR_RSI] = env->regs[R_ESI];
+ state->gprs[NVMM_X64_GPR_RDI] = env->regs[R_EDI];
+#ifdef TARGET_X86_64
+ state->gprs[NVMM_X64_GPR_R8] = env->regs[R_R8];
+ state->gprs[NVMM_X64_GPR_R9] = env->regs[R_R9];
+ state->gprs[NVMM_X64_GPR_R10] = env->regs[R_R10];
+ state->gprs[NVMM_X64_GPR_R11] = env->regs[R_R11];
+ state->gprs[NVMM_X64_GPR_R12] = env->regs[R_R12];
+ state->gprs[NVMM_X64_GPR_R13] = env->regs[R_R13];
+ state->gprs[NVMM_X64_GPR_R14] = env->regs[R_R14];
+ state->gprs[NVMM_X64_GPR_R15] = env->regs[R_R15];
+#endif
+
+ /* RIP and RFLAGS. */
+ state->gprs[NVMM_X64_GPR_RIP] = env->eip;
+ state->gprs[NVMM_X64_GPR_RFLAGS] = env->eflags;
+
+ /* Segments. */
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_CS], &env->segs[R_CS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_DS], &env->segs[R_DS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_ES], &env->segs[R_ES]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_FS], &env->segs[R_FS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_GS], &env->segs[R_GS]);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_SS], &env->segs[R_SS]);
+
+ /* Special segments. */
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_GDT], &env->gdt);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_LDT], &env->ldt);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_TR], &env->tr);
+ nvmm_set_segment(&state->segs[NVMM_X64_SEG_IDT], &env->idt);
+
+ /* Control registers. */
+ state->crs[NVMM_X64_CR_CR0] = env->cr[0];
+ state->crs[NVMM_X64_CR_CR2] = env->cr[2];
+ state->crs[NVMM_X64_CR_CR3] = env->cr[3];
+ state->crs[NVMM_X64_CR_CR4] = env->cr[4];
+ state->crs[NVMM_X64_CR_CR8] = qcpu->tpr;
+ state->crs[NVMM_X64_CR_XCR0] = env->xcr0;
+
+ /* Debug registers. */
+ state->drs[NVMM_X64_DR_DR0] = env->dr[0];
+ state->drs[NVMM_X64_DR_DR1] = env->dr[1];
+ state->drs[NVMM_X64_DR_DR2] = env->dr[2];
+ state->drs[NVMM_X64_DR_DR3] = env->dr[3];
+ state->drs[NVMM_X64_DR_DR6] = env->dr[6];
+ state->drs[NVMM_X64_DR_DR7] = env->dr[7];
+
+ /* FPU. */
+ state->fpu.fx_cw = env->fpuc;
+ state->fpu.fx_sw = (env->fpus & ~0x3800) | ((env->fpstt & 0x7) << 11);
+ state->fpu.fx_tw = 0;
+ for (i = 0; i < 8; i++) {
+ state->fpu.fx_tw |= (!env->fptags[i]) << i;
+ }
+ state->fpu.fx_opcode = env->fpop;
+ state->fpu.fx_ip.fa_64 = env->fpip;
+ state->fpu.fx_dp.fa_64 = env->fpdp;
+ state->fpu.fx_mxcsr = env->mxcsr;
+ state->fpu.fx_mxcsr_mask = 0x0000FFFF;
+ assert(sizeof(state->fpu.fx_87_ac) == sizeof(env->fpregs));
+ memcpy(state->fpu.fx_87_ac, env->fpregs, sizeof(env->fpregs));
+ for (i = 0; i < CPU_NB_REGS; i++) {
+ memcpy(&state->fpu.fx_xmm[i].xmm_bytes[0],
+ &env->xmm_regs[i].ZMM_Q(0), 8);
+ memcpy(&state->fpu.fx_xmm[i].xmm_bytes[8],
+ &env->xmm_regs[i].ZMM_Q(1), 8);
+ }
+
+ /* MSRs. */
+ state->msrs[NVMM_X64_MSR_EFER] = env->efer;
+ state->msrs[NVMM_X64_MSR_STAR] = env->star;
+#ifdef TARGET_X86_64
+ state->msrs[NVMM_X64_MSR_LSTAR] = env->lstar;
+ state->msrs[NVMM_X64_MSR_CSTAR] = env->cstar;
+ state->msrs[NVMM_X64_MSR_SFMASK] = env->fmask;
+ state->msrs[NVMM_X64_MSR_KERNELGSBASE] = env->kernelgsbase;
+#endif
+ state->msrs[NVMM_X64_MSR_SYSENTER_CS] = env->sysenter_cs;
+ state->msrs[NVMM_X64_MSR_SYSENTER_ESP] = env->sysenter_esp;
+ state->msrs[NVMM_X64_MSR_SYSENTER_EIP] = env->sysenter_eip;
+ state->msrs[NVMM_X64_MSR_PAT] = env->pat;
+ state->msrs[NVMM_X64_MSR_TSC] = env->tsc;
+
+ bitmap =
+ NVMM_X64_STATE_SEGS |
+ NVMM_X64_STATE_GPRS |
+ NVMM_X64_STATE_CRS |
+ NVMM_X64_STATE_DRS |
+ NVMM_X64_STATE_MSRS |
+ NVMM_X64_STATE_FPU;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, bitmap);
+ if (ret == -1) {
+ error_report("NVMM: Failed to set virtual processor context,"
+ " error=%d", errno);
+ }
+}
+
+static void
+nvmm_get_segment(SegmentCache *qseg, const struct nvmm_x64_state_seg *nseg)
+{
+ qseg->selector = nseg->selector;
+ qseg->limit = nseg->limit;
+ qseg->base = nseg->base;
+
+ qseg->flags =
+ __SHIFTIN((uint32_t)nseg->attrib.type, DESC_TYPE_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.s, DESC_S_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.dpl, DESC_DPL_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.p, DESC_P_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.avl, DESC_AVL_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.l, DESC_L_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.def, DESC_B_MASK) |
+ __SHIFTIN((uint32_t)nseg->attrib.g, DESC_G_MASK);
+}
+
+static void
+nvmm_get_registers(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t bitmap, tpr;
+ size_t i;
+ int ret;
+
+ assert(cpu_is_stopped(cpu) || qemu_cpu_is_self(cpu));
+
+ bitmap =
+ NVMM_X64_STATE_SEGS |
+ NVMM_X64_STATE_GPRS |
+ NVMM_X64_STATE_CRS |
+ NVMM_X64_STATE_DRS |
+ NVMM_X64_STATE_MSRS |
+ NVMM_X64_STATE_FPU;
+
+ ret = nvmm_vcpu_getstate(mach, vcpu, bitmap);
+ if (ret == -1) {
+ error_report("NVMM: Failed to get virtual processor context,"
+ " error=%d", errno);
+ }
+
+ /* GPRs. */
+ env->regs[R_EAX] = state->gprs[NVMM_X64_GPR_RAX];
+ env->regs[R_ECX] = state->gprs[NVMM_X64_GPR_RCX];
+ env->regs[R_EDX] = state->gprs[NVMM_X64_GPR_RDX];
+ env->regs[R_EBX] = state->gprs[NVMM_X64_GPR_RBX];
+ env->regs[R_ESP] = state->gprs[NVMM_X64_GPR_RSP];
+ env->regs[R_EBP] = state->gprs[NVMM_X64_GPR_RBP];
+ env->regs[R_ESI] = state->gprs[NVMM_X64_GPR_RSI];
+ env->regs[R_EDI] = state->gprs[NVMM_X64_GPR_RDI];
+#ifdef TARGET_X86_64
+ env->regs[R_R8] = state->gprs[NVMM_X64_GPR_R8];
+ env->regs[R_R9] = state->gprs[NVMM_X64_GPR_R9];
+ env->regs[R_R10] = state->gprs[NVMM_X64_GPR_R10];
+ env->regs[R_R11] = state->gprs[NVMM_X64_GPR_R11];
+ env->regs[R_R12] = state->gprs[NVMM_X64_GPR_R12];
+ env->regs[R_R13] = state->gprs[NVMM_X64_GPR_R13];
+ env->regs[R_R14] = state->gprs[NVMM_X64_GPR_R14];
+ env->regs[R_R15] = state->gprs[NVMM_X64_GPR_R15];
+#endif
+
+ /* RIP and RFLAGS. */
+ env->eip = state->gprs[NVMM_X64_GPR_RIP];
+ env->eflags = state->gprs[NVMM_X64_GPR_RFLAGS];
+
+ /* Segments. */
+ nvmm_get_segment(&env->segs[R_ES], &state->segs[NVMM_X64_SEG_ES]);
+ nvmm_get_segment(&env->segs[R_CS], &state->segs[NVMM_X64_SEG_CS]);
+ nvmm_get_segment(&env->segs[R_SS], &state->segs[NVMM_X64_SEG_SS]);
+ nvmm_get_segment(&env->segs[R_DS], &state->segs[NVMM_X64_SEG_DS]);
+ nvmm_get_segment(&env->segs[R_FS], &state->segs[NVMM_X64_SEG_FS]);
+ nvmm_get_segment(&env->segs[R_GS], &state->segs[NVMM_X64_SEG_GS]);
+
+ /* Special segments. */
+ nvmm_get_segment(&env->gdt, &state->segs[NVMM_X64_SEG_GDT]);
+ nvmm_get_segment(&env->ldt, &state->segs[NVMM_X64_SEG_LDT]);
+ nvmm_get_segment(&env->tr, &state->segs[NVMM_X64_SEG_TR]);
+ nvmm_get_segment(&env->idt, &state->segs[NVMM_X64_SEG_IDT]);
+
+ /* Control registers. */
+ env->cr[0] = state->crs[NVMM_X64_CR_CR0];
+ env->cr[2] = state->crs[NVMM_X64_CR_CR2];
+ env->cr[3] = state->crs[NVMM_X64_CR_CR3];
+ env->cr[4] = state->crs[NVMM_X64_CR_CR4];
+ tpr = state->crs[NVMM_X64_CR_CR8];
+ if (tpr != qcpu->tpr) {
+ qcpu->tpr = tpr;
+ cpu_set_apic_tpr(x86_cpu->apic_state, tpr);
+ }
+ env->xcr0 = state->crs[NVMM_X64_CR_XCR0];
+
+ /* Debug registers. */
+ env->dr[0] = state->drs[NVMM_X64_DR_DR0];
+ env->dr[1] = state->drs[NVMM_X64_DR_DR1];
+ env->dr[2] = state->drs[NVMM_X64_DR_DR2];
+ env->dr[3] = state->drs[NVMM_X64_DR_DR3];
+ env->dr[6] = state->drs[NVMM_X64_DR_DR6];
+ env->dr[7] = state->drs[NVMM_X64_DR_DR7];
+
+ /* FPU. */
+ env->fpuc = state->fpu.fx_cw;
+ env->fpstt = (state->fpu.fx_sw >> 11) & 0x7;
+ env->fpus = state->fpu.fx_sw & ~0x3800;
+ for (i = 0; i < 8; i++) {
+ env->fptags[i] = !((state->fpu.fx_tw >> i) & 1);
+ }
+ env->fpop = state->fpu.fx_opcode;
+ env->fpip = state->fpu.fx_ip.fa_64;
+ env->fpdp = state->fpu.fx_dp.fa_64;
+ env->mxcsr = state->fpu.fx_mxcsr;
+ assert(sizeof(state->fpu.fx_87_ac) == sizeof(env->fpregs));
+ memcpy(env->fpregs, state->fpu.fx_87_ac, sizeof(env->fpregs));
+ for (i = 0; i < CPU_NB_REGS; i++) {
+ memcpy(&env->xmm_regs[i].ZMM_Q(0),
+ &state->fpu.fx_xmm[i].xmm_bytes[0], 8);
+ memcpy(&env->xmm_regs[i].ZMM_Q(1),
+ &state->fpu.fx_xmm[i].xmm_bytes[8], 8);
+ }
+
+ /* MSRs. */
+ env->efer = state->msrs[NVMM_X64_MSR_EFER];
+ env->star = state->msrs[NVMM_X64_MSR_STAR];
+#ifdef TARGET_X86_64
+ env->lstar = state->msrs[NVMM_X64_MSR_LSTAR];
+ env->cstar = state->msrs[NVMM_X64_MSR_CSTAR];
+ env->fmask = state->msrs[NVMM_X64_MSR_SFMASK];
+ env->kernelgsbase = state->msrs[NVMM_X64_MSR_KERNELGSBASE];
+#endif
+ env->sysenter_cs = state->msrs[NVMM_X64_MSR_SYSENTER_CS];
+ env->sysenter_esp = state->msrs[NVMM_X64_MSR_SYSENTER_ESP];
+ env->sysenter_eip = state->msrs[NVMM_X64_MSR_SYSENTER_EIP];
+ env->pat = state->msrs[NVMM_X64_MSR_PAT];
+ env->tsc = state->msrs[NVMM_X64_MSR_TSC];
+
+ x86_update_hflags(env);
+}
+
+static bool
+nvmm_can_take_int(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ struct nvmm_machine *mach = get_nvmm_mach();
+
+ if (qcpu->int_window_exit) {
+ return false;
+ }
+
+ if (qcpu->int_shadow || !(env->eflags & IF_MASK)) {
+ struct nvmm_x64_state *state = vcpu->state;
+
+ /* Exit on interrupt window. */
+ nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_INTR);
+ state->intr.int_window_exiting = 1;
+ nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_INTR);
+
+ return false;
+ }
+
+ return true;
+}
+
+static bool
+nvmm_can_take_nmi(CPUState *cpu)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+
+ /*
+ * Contrary to INTs, NMIs always schedule an exit when they are
+ * completed. Therefore, if window-exiting is enabled, it means
+ * NMIs are blocked.
+ */
+ if (qcpu->nmi_window_exit) {
+ return false;
+ }
+
+ return true;
+}
+
+/*
+ * Called before the VCPU is run. We inject events generated by the I/O
+ * thread, and synchronize the guest TPR.
+ */
+static void
+nvmm_vcpu_pre_run(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ struct nvmm_vcpu_event *event = vcpu->event;
+ bool has_event = false;
+ bool sync_tpr = false;
+ uint8_t tpr;
+ int ret;
+
+ qemu_mutex_lock_iothread();
+
+ tpr = cpu_get_apic_tpr(x86_cpu->apic_state);
+ if (tpr != qcpu->tpr) {
+ qcpu->tpr = tpr;
+ sync_tpr = true;
+ }
+
+ /*
+ * Force the VCPU out of its inner loop to process any INIT requests
+ * or commit pending TPR access.
+ */
+ if (cpu->interrupt_request & (CPU_INTERRUPT_INIT | CPU_INTERRUPT_TPR)) {
+ cpu->exit_request = 1;
+ }
+
+ if (!has_event && (cpu->interrupt_request & CPU_INTERRUPT_NMI)) {
+ if (nvmm_can_take_nmi(cpu)) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_NMI;
+ event->type = NVMM_VCPU_EVENT_INTR;
+ event->vector = 2;
+ has_event = true;
+ }
+ }
+
+ if (!has_event && (cpu->interrupt_request & CPU_INTERRUPT_HARD)) {
+ if (nvmm_can_take_int(cpu)) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_HARD;
+ event->type = NVMM_VCPU_EVENT_INTR;
+ event->vector = cpu_get_pic_interrupt(env);
+ has_event = true;
+ }
+ }
+
+ /* Don't want SMIs. */
+ if (cpu->interrupt_request & CPU_INTERRUPT_SMI) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_SMI;
+ }
+
+ if (sync_tpr) {
+ ret = nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_CRS);
+ if (ret == -1) {
+ error_report("NVMM: Failed to get CPU state,"
+ " error=%d", errno);
+ }
+
+ state->crs[NVMM_X64_CR_CR8] = qcpu->tpr;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_CRS);
+ if (ret == -1) {
+ error_report("NVMM: Failed to set CPU state,"
+ " error=%d", errno);
+ }
+ }
+
+ if (has_event) {
+ ret = nvmm_vcpu_inject(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: Failed to inject event,"
+ " error=%d", errno);
+ }
+ }
+
+ qemu_mutex_unlock_iothread();
+}
+
+/*
+ * Called after the VCPU ran. We synchronize the host view of the TPR and
+ * RFLAGS.
+ */
+static void
+nvmm_vcpu_post_run(CPUState *cpu, struct nvmm_vcpu_exit *exit)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ uint64_t tpr;
+
+ env->eflags = exit->exitstate.rflags;
+ qcpu->int_shadow = exit->exitstate.int_shadow;
+ qcpu->int_window_exit = exit->exitstate.int_window_exiting;
+ qcpu->nmi_window_exit = exit->exitstate.nmi_window_exiting;
+
+ tpr = exit->exitstate.cr8;
+ if (qcpu->tpr != tpr) {
+ qcpu->tpr = tpr;
+ qemu_mutex_lock_iothread();
+ cpu_set_apic_tpr(x86_cpu->apic_state, qcpu->tpr);
+ qemu_mutex_unlock_iothread();
+ }
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_io_callback(struct nvmm_io *io)
+{
+ MemTxAttrs attrs = { 0 };
+ int ret;
+
+ ret = address_space_rw(&address_space_io, io->port, attrs, io->data,
+ io->size, !io->in);
+ if (ret != MEMTX_OK) {
+ error_report("NVMM: I/O Transaction Failed "
+ "[%s, port=%u, size=%zu]", (io->in ? "in" : "out"),
+ io->port, io->size);
+ }
+
+ /* Needed, otherwise infinite loop. */
+ current_cpu->vcpu_dirty = false;
+}
+
+static void
+nvmm_mem_callback(struct nvmm_mem *mem)
+{
+ cpu_physical_memory_rw(mem->gpa, mem->data, mem->size, mem->write);
+
+ /* XXX Needed, otherwise infinite loop. */
+ current_cpu->vcpu_dirty = false;
+}
+
+static struct nvmm_assist_callbacks nvmm_callbacks = {
+ .io = nvmm_io_callback,
+ .mem = nvmm_mem_callback
+};
+
+/* -------------------------------------------------------------------------- */
+
+static int
+nvmm_handle_mem(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
+{
+ int ret;
+
+ ret = nvmm_assist_mem(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: Mem Assist Failed [gpa=%p]",
+ (void *)vcpu->exit->u.mem.gpa);
+ }
+
+ return ret;
+}
+
+static int
+nvmm_handle_io(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
+{
+ int ret;
+
+ ret = nvmm_assist_io(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: I/O Assist Failed [port=%d]",
+ (int)vcpu->exit->u.io.port);
+ }
+
+ return ret;
+}
+
+static int
+nvmm_handle_rdmsr(struct nvmm_machine *mach, CPUState *cpu,
+ struct nvmm_vcpu_exit *exit)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t val;
+ int ret;
+
+ switch (exit->u.rdmsr.msr) {
+ case MSR_IA32_APICBASE:
+ val = cpu_get_apic_base(x86_cpu->apic_state);
+ break;
+ case MSR_MTRRcap:
+ case MSR_MTRRdefType:
+ case MSR_MCG_CAP:
+ case MSR_MCG_STATUS:
+ val = 0;
+ break;
+ default: /* More MSRs to add? */
+ val = 0;
+ error_report("NVMM: Unexpected RDMSR 0x%x, ignored",
+ exit->u.rdmsr.msr);
+ break;
+ }
+
+ ret = nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ state->gprs[NVMM_X64_GPR_RAX] = (val & 0xFFFFFFFF);
+ state->gprs[NVMM_X64_GPR_RDX] = (val >> 32);
+ state->gprs[NVMM_X64_GPR_RIP] = exit->u.rdmsr.npc;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ return 0;
+}
+
+static int
+nvmm_handle_wrmsr(struct nvmm_machine *mach, CPUState *cpu,
+ struct nvmm_vcpu_exit *exit)
+{
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_x64_state *state = vcpu->state;
+ uint64_t val;
+ int ret;
+
+ val = exit->u.wrmsr.val;
+
+ switch (exit->u.wrmsr.msr) {
+ case MSR_IA32_APICBASE:
+ cpu_set_apic_base(x86_cpu->apic_state, val);
+ break;
+ case MSR_MTRRdefType:
+ case MSR_MCG_STATUS:
+ break;
+ default: /* More MSRs to add? */
+ error_report("NVMM: Unexpected WRMSR 0x%x [val=0x%lx], ignored",
+ exit->u.wrmsr.msr, val);
+ break;
+ }
+
+ ret = nvmm_vcpu_getstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ state->gprs[NVMM_X64_GPR_RIP] = exit->u.wrmsr.npc;
+
+ ret = nvmm_vcpu_setstate(mach, vcpu, NVMM_X64_STATE_GPRS);
+ if (ret == -1) {
+ return -1;
+ }
+
+ return 0;
+}
+
+static int
+nvmm_handle_halted(struct nvmm_machine *mach, CPUState *cpu,
+ struct nvmm_vcpu_exit *exit)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ int ret = 0;
+
+ qemu_mutex_lock_iothread();
+
+ if (!((cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
+ (env->eflags & IF_MASK)) &&
+ !(cpu->interrupt_request & CPU_INTERRUPT_NMI)) {
+ cpu->exception_index = EXCP_HLT;
+ cpu->halted = true;
+ ret = 1;
+ }
+
+ qemu_mutex_unlock_iothread();
+
+ return ret;
+}
+
+static int
+nvmm_inject_ud(struct nvmm_machine *mach, struct nvmm_vcpu *vcpu)
+{
+ struct nvmm_vcpu_event *event = vcpu->event;
+
+ event->type = NVMM_VCPU_EVENT_EXCP;
+ event->vector = 6;
+ event->u.excp.error = 0;
+
+ return nvmm_vcpu_inject(mach, vcpu);
+}
+
+static int
+nvmm_vcpu_loop(CPUState *cpu)
+{
+ struct CPUX86State *env = (CPUArchState *)cpu->env_ptr;
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+ struct nvmm_vcpu *vcpu = &qcpu->vcpu;
+ X86CPU *x86_cpu = X86_CPU(cpu);
+ struct nvmm_vcpu_exit *exit = vcpu->exit;
+ int ret;
+
+ /*
+ * Some asynchronous events must be handled outside of the inner
+ * VCPU loop. They are handled here.
+ */
+ if (cpu->interrupt_request & CPU_INTERRUPT_INIT) {
+ nvmm_cpu_synchronize_state(cpu);
+ do_cpu_init(x86_cpu);
+ /* set int/nmi windows back to the reset state */
+ }
+ if (cpu->interrupt_request & CPU_INTERRUPT_POLL) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_POLL;
+ apic_poll_irq(x86_cpu->apic_state);
+ }
+ if (((cpu->interrupt_request & CPU_INTERRUPT_HARD) &&
+ (env->eflags & IF_MASK)) ||
+ (cpu->interrupt_request & CPU_INTERRUPT_NMI)) {
+ cpu->halted = false;
+ }
+ if (cpu->interrupt_request & CPU_INTERRUPT_SIPI) {
+ nvmm_cpu_synchronize_state(cpu);
+ do_cpu_sipi(x86_cpu);
+ }
+ if (cpu->interrupt_request & CPU_INTERRUPT_TPR) {
+ cpu->interrupt_request &= ~CPU_INTERRUPT_TPR;
+ nvmm_cpu_synchronize_state(cpu);
+ apic_handle_tpr_access_report(x86_cpu->apic_state, env->eip,
+ env->tpr_access_type);
+ }
+
+ if (cpu->halted) {
+ cpu->exception_index = EXCP_HLT;
+ atomic_set(&cpu->exit_request, false);
+ return 0;
+ }
+
+ qemu_mutex_unlock_iothread();
+ cpu_exec_start(cpu);
+
+ /*
+ * Inner VCPU loop.
+ */
+ do {
+ if (cpu->vcpu_dirty) {
+ nvmm_set_registers(cpu);
+ cpu->vcpu_dirty = false;
+ }
+
+ if (qcpu->stop) {
+ cpu->exception_index = EXCP_INTERRUPT;
+ qcpu->stop = false;
+ ret = 1;
+ break;
+ }
+
+ nvmm_vcpu_pre_run(cpu);
+
+ if (atomic_read(&cpu->exit_request)) {
+ qemu_cpu_kick_self();
+ }
+
+ ret = nvmm_vcpu_run(mach, vcpu);
+ if (ret == -1) {
+ error_report("NVMM: Failed to exec a virtual processor,"
+ " error=%d", errno);
+ break;
+ }
+
+ nvmm_vcpu_post_run(cpu, exit);
+
+ switch (exit->reason) {
+ case NVMM_VCPU_EXIT_NONE:
+ break;
+ case NVMM_VCPU_EXIT_MEMORY:
+ ret = nvmm_handle_mem(mach, vcpu);
+ break;
+ case NVMM_VCPU_EXIT_IO:
+ ret = nvmm_handle_io(mach, vcpu);
+ break;
+ case NVMM_VCPU_EXIT_INT_READY:
+ case NVMM_VCPU_EXIT_NMI_READY:
+ case NVMM_VCPU_EXIT_TPR_CHANGED:
+ break;
+ case NVMM_VCPU_EXIT_HALTED:
+ ret = nvmm_handle_halted(mach, cpu, exit);
+ break;
+ case NVMM_VCPU_EXIT_SHUTDOWN:
+ qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
+ cpu->exception_index = EXCP_INTERRUPT;
+ ret = 1;
+ break;
+ case NVMM_VCPU_EXIT_RDMSR:
+ ret = nvmm_handle_rdmsr(mach, cpu, exit);
+ break;
+ case NVMM_VCPU_EXIT_WRMSR:
+ ret = nvmm_handle_wrmsr(mach, cpu, exit);
+ break;
+ case NVMM_VCPU_EXIT_MONITOR:
+ case NVMM_VCPU_EXIT_MWAIT:
+ ret = nvmm_inject_ud(mach, vcpu);
+ break;
+ default:
+ error_report("NVMM: Unexpected VM exit code 0x%lx [hw=0x%lx]",
+ exit->reason, exit->u.inv.hwcode);
+ nvmm_get_registers(cpu);
+ qemu_mutex_lock_iothread();
+ qemu_system_guest_panicked(cpu_get_crash_info(cpu));
+ qemu_mutex_unlock_iothread();
+ ret = -1;
+ break;
+ }
+ } while (ret == 0);
+
+ cpu_exec_end(cpu);
+ qemu_mutex_lock_iothread();
+ current_cpu = cpu;
+
+ atomic_set(&cpu->exit_request, false);
+
+ return ret < 0;
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+do_nvmm_cpu_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
+{
+ nvmm_get_registers(cpu);
+ cpu->vcpu_dirty = true;
+}
+
+static void
+do_nvmm_cpu_synchronize_post_reset(CPUState *cpu, run_on_cpu_data arg)
+{
+ nvmm_set_registers(cpu);
+ cpu->vcpu_dirty = false;
+}
+
+static void
+do_nvmm_cpu_synchronize_post_init(CPUState *cpu, run_on_cpu_data arg)
+{
+ nvmm_set_registers(cpu);
+ cpu->vcpu_dirty = false;
+}
+
+static void
+do_nvmm_cpu_synchronize_pre_loadvm(CPUState *cpu, run_on_cpu_data arg)
+{
+ cpu->vcpu_dirty = true;
+}
+
+void nvmm_cpu_synchronize_state(CPUState *cpu)
+{
+ if (!cpu->vcpu_dirty) {
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_state, RUN_ON_CPU_NULL);
+ }
+}
+
+void nvmm_cpu_synchronize_post_reset(CPUState *cpu)
+{
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_post_reset, RUN_ON_CPU_NULL);
+}
+
+void nvmm_cpu_synchronize_post_init(CPUState *cpu)
+{
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_post_init, RUN_ON_CPU_NULL);
+}
+
+void nvmm_cpu_synchronize_pre_loadvm(CPUState *cpu)
+{
+ run_on_cpu(cpu, do_nvmm_cpu_synchronize_pre_loadvm, RUN_ON_CPU_NULL);
+}
+
+/* -------------------------------------------------------------------------- */
+
+static Error *nvmm_migration_blocker;
+
+static void
+nvmm_ipi_signal(int sigcpu)
+{
+ struct qemu_vcpu *qcpu;
+
+ if (current_cpu) {
+ qcpu = get_qemu_vcpu(current_cpu);
+ qcpu->stop = true;
+ }
+}
+
+static void
+nvmm_init_cpu_signals(void)
+{
+ struct sigaction sigact;
+ sigset_t set;
+
+ /* Install the IPI handler. */
+ memset(&sigact, 0, sizeof(sigact));
+ sigact.sa_handler = nvmm_ipi_signal;
+ sigaction(SIG_IPI, &sigact, NULL);
+
+ /* Allow IPIs on the current thread. */
+ sigprocmask(SIG_BLOCK, NULL, &set);
+ sigdelset(&set, SIG_IPI);
+ pthread_sigmask(SIG_SETMASK, &set, NULL);
+}
+
+int
+nvmm_init_vcpu(CPUState *cpu)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct nvmm_vcpu_conf_cpuid cpuid;
+ struct nvmm_vcpu_conf_tpr tpr;
+ Error *local_error = NULL;
+ struct qemu_vcpu *qcpu;
+ int ret, err;
+
+ nvmm_init_cpu_signals();
+
+ if (nvmm_migration_blocker == NULL) {
+ error_setg(&nvmm_migration_blocker,
+ "NVMM: Migration not supported");
+
+ (void)migrate_add_blocker(nvmm_migration_blocker, &local_error);
+ if (local_error) {
+ error_report_err(local_error);
+ migrate_del_blocker(nvmm_migration_blocker);
+ error_free(nvmm_migration_blocker);
+ return -EINVAL;
+ }
+ }
+
+ qcpu = g_malloc0(sizeof(*qcpu));
+ if (qcpu == NULL) {
+ error_report("NVMM: Failed to allocate VCPU context.");
+ return -ENOMEM;
+ }
+
+ ret = nvmm_vcpu_create(mach, cpu->cpu_index, &qcpu->vcpu);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to create a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+
+ memset(&cpuid, 0, sizeof(cpuid));
+ cpuid.mask = 1;
+ cpuid.leaf = 0x00000001;
+ cpuid.u.mask.set.edx = CPUID_MCE | CPUID_MCA | CPUID_MTRR;
+ ret = nvmm_vcpu_configure(mach, &qcpu->vcpu, NVMM_VCPU_CONF_CPUID,
+ &cpuid);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to configure a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+
+ ret = nvmm_vcpu_configure(mach, &qcpu->vcpu, NVMM_VCPU_CONF_CALLBACKS,
+ &nvmm_callbacks);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to configure a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+
+ if (qemu_mach.cap.arch.vcpu_conf_support & NVMM_CAP_ARCH_VCPU_CONF_TPR) {
+ memset(&tpr, 0, sizeof(tpr));
+ tpr.exit_changed = 1;
+ ret = nvmm_vcpu_configure(mach, &qcpu->vcpu, NVMM_VCPU_CONF_TPR, &tpr);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Failed to configure a virtual processor,"
+ " error=%d", err);
+ g_free(qcpu);
+ return -err;
+ }
+ }
+
+ cpu->vcpu_dirty = true;
+ cpu->hax_vcpu = (struct hax_vcpu_state *)qcpu;
+
+ return 0;
+}
+
+int
+nvmm_vcpu_exec(CPUState *cpu)
+{
+ int ret, fatal;
+
+ while (1) {
+ if (cpu->exception_index >= EXCP_INTERRUPT) {
+ ret = cpu->exception_index;
+ cpu->exception_index = -1;
+ break;
+ }
+
+ fatal = nvmm_vcpu_loop(cpu);
+
+ if (fatal) {
+ error_report("NVMM: Failed to execute a VCPU.");
+ abort();
+ }
+ }
+
+ return ret;
+}
+
+void
+nvmm_destroy_vcpu(CPUState *cpu)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ struct qemu_vcpu *qcpu = get_qemu_vcpu(cpu);
+
+ nvmm_vcpu_destroy(mach, &qcpu->vcpu);
+ g_free(cpu->hax_vcpu);
+}
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_update_mapping(hwaddr start_pa, ram_addr_t size, uintptr_t hva,
+ bool add, bool rom, const char *name)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ int ret, prot;
+
+ if (add) {
+ prot = PROT_READ | PROT_EXEC;
+ if (!rom) {
+ prot |= PROT_WRITE;
+ }
+ ret = nvmm_gpa_map(mach, hva, start_pa, size, prot);
+ } else {
+ ret = nvmm_gpa_unmap(mach, hva, start_pa, size);
+ }
+
+ if (ret == -1) {
+ error_report("NVMM: Failed to %s GPA range '%s' PA:%p, "
+ "Size:%p bytes, HostVA:%p, error=%d",
+ (add ? "map" : "unmap"), name, (void *)(uintptr_t)start_pa,
+ (void *)size, (void *)hva, errno);
+ }
+}
+
+static void
+nvmm_process_section(MemoryRegionSection *section, int add)
+{
+ MemoryRegion *mr = section->mr;
+ hwaddr start_pa = section->offset_within_address_space;
+ ram_addr_t size = int128_get64(section->size);
+ unsigned int delta;
+ uintptr_t hva;
+
+ if (!memory_region_is_ram(mr)) {
+ return;
+ }
+
+ /* Adjust start_pa and size so that they are page-aligned. */
+ delta = qemu_real_host_page_size - (start_pa & ~qemu_real_host_page_mask);
+ delta &= ~qemu_real_host_page_mask;
+ if (delta > size) {
+ return;
+ }
+ start_pa += delta;
+ size -= delta;
+ size &= qemu_real_host_page_mask;
+ if (!size || (start_pa & ~qemu_real_host_page_mask)) {
+ return;
+ }
+
+ hva = (uintptr_t)memory_region_get_ram_ptr(mr) +
+ section->offset_within_region + delta;
+
+ nvmm_update_mapping(start_pa, size, hva, add,
+ memory_region_is_rom(mr), mr->name);
+}
+
+static void
+nvmm_region_add(MemoryListener *listener, MemoryRegionSection *section)
+{
+ memory_region_ref(section->mr);
+ nvmm_process_section(section, 1);
+}
+
+static void
+nvmm_region_del(MemoryListener *listener, MemoryRegionSection *section)
+{
+ nvmm_process_section(section, 0);
+ memory_region_unref(section->mr);
+}
+
+static void
+nvmm_transaction_begin(MemoryListener *listener)
+{
+ /* nothing */
+}
+
+static void
+nvmm_transaction_commit(MemoryListener *listener)
+{
+ /* nothing */
+}
+
+static void
+nvmm_log_sync(MemoryListener *listener, MemoryRegionSection *section)
+{
+ MemoryRegion *mr = section->mr;
+
+ if (!memory_region_is_ram(mr)) {
+ return;
+ }
+
+ memory_region_set_dirty(mr, 0, int128_get64(section->size));
+}
+
+static MemoryListener nvmm_memory_listener = {
+ .begin = nvmm_transaction_begin,
+ .commit = nvmm_transaction_commit,
+ .region_add = nvmm_region_add,
+ .region_del = nvmm_region_del,
+ .log_sync = nvmm_log_sync,
+ .priority = 10,
+};
+
+static void
+nvmm_ram_block_added(RAMBlockNotifier *n, void *host, size_t size)
+{
+ struct nvmm_machine *mach = get_nvmm_mach();
+ uintptr_t hva = (uintptr_t)host;
+ int ret;
+
+ ret = nvmm_hva_map(mach, hva, size);
+
+ if (ret == -1) {
+ error_report("NVMM: Failed to map HVA, HostVA:%p "
+ "Size:%p bytes, error=%d",
+ (void *)hva, (void *)size, errno);
+ }
+}
+
+static struct RAMBlockNotifier nvmm_ram_notifier = {
+ .ram_block_added = nvmm_ram_block_added
+};
+
+/* -------------------------------------------------------------------------- */
+
+static void
+nvmm_handle_interrupt(CPUState *cpu, int mask)
+{
+ cpu->interrupt_request |= mask;
+
+ if (!qemu_cpu_is_self(cpu)) {
+ qemu_cpu_kick(cpu);
+ }
+}
+
+/* -------------------------------------------------------------------------- */
+
+static int
+nvmm_accel_init(MachineState *ms)
+{
+ int ret, err;
+
+ ret = nvmm_init();
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Initialization failed, error=%d", errno);
+ return -err;
+ }
+
+ ret = nvmm_capability(&qemu_mach.cap);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Unable to fetch capability, error=%d", errno);
+ return -err;
+ }
+ if (qemu_mach.cap.version != 1) {
+ error_report("NVMM: Unsupported version %u", qemu_mach.cap.version);
+ return -EPROGMISMATCH;
+ }
+ if (qemu_mach.cap.state_size != sizeof(struct nvmm_x64_state)) {
+ error_report("NVMM: Wrong state size %u", qemu_mach.cap.state_size);
+ return -EPROGMISMATCH;
+ }
+
+ ret = nvmm_machine_create(&qemu_mach.mach);
+ if (ret == -1) {
+ err = errno;
+ error_report("NVMM: Machine creation failed, error=%d", errno);
+ return -err;
+ }
+
+ memory_listener_register(&nvmm_memory_listener, &address_space_memory);
+ ram_block_notifier_add(&nvmm_ram_notifier);
+
+ cpu_interrupt_handler = nvmm_handle_interrupt;
+
+ printf("NetBSD Virtual Machine Monitor accelerator is operational\n");
+ return 0;
+}
+
+int
+nvmm_enabled(void)
+{
+ return nvmm_allowed;
+}
+
+static void
+nvmm_accel_class_init(ObjectClass *oc, void *data)
+{
+ AccelClass *ac = ACCEL_CLASS(oc);
+ ac->name = "NVMM";
+ ac->init_machine = nvmm_accel_init;
+ ac->allowed = &nvmm_allowed;
+}
+
+static const TypeInfo nvmm_accel_type = {
+ .name = ACCEL_CLASS_NAME("nvmm"),
+ .parent = TYPE_ACCEL,
+ .class_init = nvmm_accel_class_init,
+};
+
+static void
+nvmm_type_init(void)
+{
+ type_register_static(&nvmm_accel_type);
+}
+
+type_init(nvmm_type_init);
--
2.25.0
From: Maxime Villard <max@m00nbsd.net>
Implements the NVMM accelerator cpu enlightenments to actually use the nvmm-all
accelerator on NetBSD platforms.
Signed-off-by: Maxime Villard <max@m00nbsd.net>
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Jared McNeill <jmcneill@invisible.ca>
---
cpus.c | 58 +++++++++++++++++++++++++++++++++++++++
include/sysemu/hw_accel.h | 14 ++++++++++
target/i386/helper.c | 2 +-
3 files changed, 73 insertions(+), 1 deletion(-)
diff --git a/cpus.c b/cpus.c
index b4f8b84b61..f833da4a60 100644
--- a/cpus.c
+++ b/cpus.c
@@ -42,6 +42,7 @@
#include "sysemu/hax.h"
#include "sysemu/hvf.h"
#include "sysemu/whpx.h"
+#include "sysemu/nvmm.h"
#include "exec/exec-all.h"
#include "qemu/thread.h"
@@ -1670,6 +1671,48 @@ static void *qemu_whpx_cpu_thread_fn(void *arg)
return NULL;
}
+static void *qemu_nvmm_cpu_thread_fn(void *arg)
+{
+ CPUState *cpu = arg;
+ int r;
+
+ assert(nvmm_enabled());
+
+ rcu_register_thread();
+
+ qemu_mutex_lock_iothread();
+ qemu_thread_get_self(cpu->thread);
+ cpu->thread_id = qemu_get_thread_id();
+ current_cpu = cpu;
+
+ r = nvmm_init_vcpu(cpu);
+ if (r < 0) {
+ fprintf(stderr, "nvmm_init_vcpu failed: %s\n", strerror(-r));
+ exit(1);
+ }
+
+ /* signal CPU creation */
+ cpu->created = true;
+ qemu_cond_signal(&qemu_cpu_cond);
+
+ do {
+ if (cpu_can_run(cpu)) {
+ r = nvmm_vcpu_exec(cpu);
+ if (r == EXCP_DEBUG) {
+ cpu_handle_guest_debug(cpu);
+ }
+ }
+ qemu_wait_io_event(cpu);
+ } while (!cpu->unplug || cpu_can_run(cpu));
+
+ nvmm_destroy_vcpu(cpu);
+ cpu->created = false;
+ qemu_cond_signal(&qemu_cpu_cond);
+ qemu_mutex_unlock_iothread();
+ rcu_unregister_thread();
+ return NULL;
+}
+
#ifdef _WIN32
static void CALLBACK dummy_apc_func(ULONG_PTR unused)
{
@@ -2038,6 +2081,19 @@ static void qemu_whpx_start_vcpu(CPUState *cpu)
#endif
}
+static void qemu_nvmm_start_vcpu(CPUState *cpu)
+{
+ char thread_name[VCPU_THREAD_NAME_SIZE];
+
+ cpu->thread = g_malloc0(sizeof(QemuThread));
+ cpu->halt_cond = g_malloc0(sizeof(QemuCond));
+ qemu_cond_init(cpu->halt_cond);
+ snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/NVMM",
+ cpu->cpu_index);
+ qemu_thread_create(cpu->thread, thread_name, qemu_nvmm_cpu_thread_fn,
+ cpu, QEMU_THREAD_JOINABLE);
+}
+
static void qemu_dummy_start_vcpu(CPUState *cpu)
{
char thread_name[VCPU_THREAD_NAME_SIZE];
@@ -2078,6 +2134,8 @@ void qemu_init_vcpu(CPUState *cpu)
qemu_tcg_init_vcpu(cpu);
} else if (whpx_enabled()) {
qemu_whpx_start_vcpu(cpu);
+ } else if (nvmm_enabled()) {
+ qemu_nvmm_start_vcpu(cpu);
} else {
qemu_dummy_start_vcpu(cpu);
}
diff --git a/include/sysemu/hw_accel.h b/include/sysemu/hw_accel.h
index 0ec2372477..dbfa7a02f9 100644
--- a/include/sysemu/hw_accel.h
+++ b/include/sysemu/hw_accel.h
@@ -15,6 +15,7 @@
#include "sysemu/hax.h"
#include "sysemu/kvm.h"
#include "sysemu/whpx.h"
+#include "sysemu/nvmm.h"
static inline void cpu_synchronize_state(CPUState *cpu)
{
@@ -27,6 +28,9 @@ static inline void cpu_synchronize_state(CPUState *cpu)
if (whpx_enabled()) {
whpx_cpu_synchronize_state(cpu);
}
+ if (nvmm_enabled()) {
+ nvmm_cpu_synchronize_state(cpu);
+ }
}
static inline void cpu_synchronize_post_reset(CPUState *cpu)
@@ -40,6 +44,10 @@ static inline void cpu_synchronize_post_reset(CPUState *cpu)
if (whpx_enabled()) {
whpx_cpu_synchronize_post_reset(cpu);
}
+ if (nvmm_enabled()) {
+ nvmm_cpu_synchronize_post_reset(cpu);
+ }
+
}
static inline void cpu_synchronize_post_init(CPUState *cpu)
@@ -53,6 +61,9 @@ static inline void cpu_synchronize_post_init(CPUState *cpu)
if (whpx_enabled()) {
whpx_cpu_synchronize_post_init(cpu);
}
+ if (nvmm_enabled()) {
+ nvmm_cpu_synchronize_post_init(cpu);
+ }
}
static inline void cpu_synchronize_pre_loadvm(CPUState *cpu)
@@ -66,6 +77,9 @@ static inline void cpu_synchronize_pre_loadvm(CPUState *cpu)
if (whpx_enabled()) {
whpx_cpu_synchronize_pre_loadvm(cpu);
}
+ if (nvmm_enabled()) {
+ nvmm_cpu_synchronize_pre_loadvm(cpu);
+ }
}
#endif /* QEMU_HW_ACCEL_H */
diff --git a/target/i386/helper.c b/target/i386/helper.c
index c3a6e4fabe..2e79d61329 100644
--- a/target/i386/helper.c
+++ b/target/i386/helper.c
@@ -981,7 +981,7 @@ void cpu_report_tpr_access(CPUX86State *env, TPRAccess access)
X86CPU *cpu = env_archcpu(env);
CPUState *cs = env_cpu(env);
- if (kvm_enabled() || whpx_enabled()) {
+ if (kvm_enabled() || whpx_enabled() || nvmm_enabled()) {
env->tpr_access_type = access;
cpu_interrupt(cs, CPU_INTERRUPT_TPR);
--
2.25.0
Patchew URL: https://patchew.org/QEMU/20200206115731.13552-1-n54@gmx.com/
Hi,
This series seems to have some coding style problems. See output below for
more information:
Subject: [PATCH v3 0/4] Implements the NetBSD Virtual Machine Monitor accelerator
Message-id: 20200206115731.13552-1-n54@gmx.com
Type: series
=== TEST SCRIPT BEGIN ===
#!/bin/bash
git rev-parse base > /dev/null || exit 0
git config --local diff.renamelimit 0
git config --local diff.renames True
git config --local diff.algorithm histogram
./scripts/checkpatch.pl --mailback base..
=== TEST SCRIPT END ===
Updating 3c8cf5a9c21ff8782164d1def7f44bd888713384
fatal: git fetch_pack: expected ACK/NAK, got 'ERR upload-pack: not our ref 1c298dad3d820f7a2161054ff581cf2fa65ee1b4'
fatal: The remote end hung up unexpectedly
error: Could not fetch 3c8cf5a9c21ff8782164d1def7f44bd888713384
Traceback (most recent call last):
File "patchew-tester/src/patchew-cli", line 521, in test_one
git_clone_repo(clone, r["repo"], r["head"], logf, True)
File "patchew-tester/src/patchew-cli", line 48, in git_clone_repo
stdout=logf, stderr=logf)
File "/opt/rh/rh-python36/root/usr/lib64/python3.6/subprocess.py", line 291, in check_call
raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '['git', 'remote', 'add', '-f', '--mirror=fetch', '3c8cf5a9c21ff8782164d1def7f44bd888713384', 'https://github.com/patchew-project/qemu']' returned non-zero exit status 1.
The full log is available at
http://patchew.org/logs/20200206115731.13552-1-n54@gmx.com/testing.checkpatch/?type=message.
---
Email generated automatically by Patchew [https://patchew.org/].
Please send your feedback to patchew-devel@redhat.com
Am I supposed to do something with this or is this an issue in a script? On 06.02.2020 14:13, no-reply@patchew.org wrote: > Patchew URL: https://patchew.org/QEMU/20200206115731.13552-1-n54@gmx.com/ > > > > Hi, > > This series seems to have some coding style problems. See output below for > more information: > > Subject: [PATCH v3 0/4] Implements the NetBSD Virtual Machine Monitor accelerator > Message-id: 20200206115731.13552-1-n54@gmx.com > Type: series > > === TEST SCRIPT BEGIN === > #!/bin/bash > git rev-parse base > /dev/null || exit 0 > git config --local diff.renamelimit 0 > git config --local diff.renames True > git config --local diff.algorithm histogram > ./scripts/checkpatch.pl --mailback base.. > === TEST SCRIPT END === > > Updating 3c8cf5a9c21ff8782164d1def7f44bd888713384 > fatal: git fetch_pack: expected ACK/NAK, got 'ERR upload-pack: not our ref 1c298dad3d820f7a2161054ff581cf2fa65ee1b4' > fatal: The remote end hung up unexpectedly > error: Could not fetch 3c8cf5a9c21ff8782164d1def7f44bd888713384 > Traceback (most recent call last): > File "patchew-tester/src/patchew-cli", line 521, in test_one > git_clone_repo(clone, r["repo"], r["head"], logf, True) > File "patchew-tester/src/patchew-cli", line 48, in git_clone_repo > stdout=logf, stderr=logf) > File "/opt/rh/rh-python36/root/usr/lib64/python3.6/subprocess.py", line 291, in check_call > raise CalledProcessError(retcode, cmd) > subprocess.CalledProcessError: Command '['git', 'remote', 'add', '-f', '--mirror=fetch', '3c8cf5a9c21ff8782164d1def7f44bd888713384', 'https://github.com/patchew-project/qemu']' returned non-zero exit status 1. > > > > The full log is available at > http://patchew.org/logs/20200206115731.13552-1-n54@gmx.com/testing.checkpatch/?type=message. > --- > Email generated automatically by Patchew [https://patchew.org/]. > Please send your feedback to patchew-devel@redhat.com >
On 2/6/20 2:21 PM, Kamil Rytarowski wrote: > Am I supposed to do something with this or is this an issue in a script? I think either storage full or network failure. > On 06.02.2020 14:13, no-reply@patchew.org wrote: >> Patchew URL: https://patchew.org/QEMU/20200206115731.13552-1-n54@gmx.com/ >> >> >> >> Hi, >> >> This series seems to have some coding style problems. See output below for >> more information: >> >> Subject: [PATCH v3 0/4] Implements the NetBSD Virtual Machine Monitor accelerator >> Message-id: 20200206115731.13552-1-n54@gmx.com >> Type: series >> >> === TEST SCRIPT BEGIN === >> #!/bin/bash >> git rev-parse base > /dev/null || exit 0 >> git config --local diff.renamelimit 0 >> git config --local diff.renames True >> git config --local diff.algorithm histogram >> ./scripts/checkpatch.pl --mailback base.. >> === TEST SCRIPT END === >> >> Updating 3c8cf5a9c21ff8782164d1def7f44bd888713384 >> fatal: git fetch_pack: expected ACK/NAK, got 'ERR upload-pack: not our ref 1c298dad3d820f7a2161054ff581cf2fa65ee1b4' >> fatal: The remote end hung up unexpectedly >> error: Could not fetch 3c8cf5a9c21ff8782164d1def7f44bd888713384 >> Traceback (most recent call last): >> File "patchew-tester/src/patchew-cli", line 521, in test_one >> git_clone_repo(clone, r["repo"], r["head"], logf, True) >> File "patchew-tester/src/patchew-cli", line 48, in git_clone_repo >> stdout=logf, stderr=logf) >> File "/opt/rh/rh-python36/root/usr/lib64/python3.6/subprocess.py", line 291, in check_call >> raise CalledProcessError(retcode, cmd) >> subprocess.CalledProcessError: Command '['git', 'remote', 'add', '-f', '--mirror=fetch', '3c8cf5a9c21ff8782164d1def7f44bd888713384', 'https://github.com/patchew-project/qemu']' returned non-zero exit status 1. >> >> >> >> The full log is available at >> http://patchew.org/logs/20200206115731.13552-1-n54@gmx.com/testing.checkpatch/?type=message. _______________________________________________ Patchew-devel mailing list Patchew-devel@redhat.com https://www.redhat.com/mailman/listinfo/patchew-devel
© 2016 - 2024 Red Hat, Inc.