drivers/crypto/ccp/psp-dev.c | 3 + drivers/crypto/ccp/tee-dev.c | 119 ++++++++++++++++++---------- drivers/crypto/ccp/tee-dev.h | 11 +-- drivers/tee/amdtee/amdtee_private.h | 18 ++--- drivers/tee/amdtee/call.c | 74 ++++++++--------- drivers/tee/amdtee/core.c | 72 ++++++++++------- drivers/tee/amdtee/shm_pool.c | 21 ++--- include/linux/psp-tee.h | 47 +++++++++++ 8 files changed, 221 insertions(+), 144 deletions(-)
From: Jeshwanth Kumar N K <JESHWANTHKUMAR.NK@amd.com> At present, the shared memory for TEE ring buffer, command buffer and data buffer is allocated using get_free_pages(). The driver shares the physical address of these buffers with PSP so that it can be mapped by the Trusted OS. In this patch series we have replaced get_free_pages() with dma_alloc_coherent() to allocate shared memory to cleanup the existing allocation method. Rijo Thomas (3): crypto: ccp - Add function to allocate and free memory using DMA APIs crypto: ccp - Use psp_tee_alloc_buffer() and psp_tee_free_buffer() tee: amdtee: Use psp_tee_alloc_buffer() and psp_tee_free_buffer() drivers/crypto/ccp/psp-dev.c | 3 + drivers/crypto/ccp/tee-dev.c | 119 ++++++++++++++++++---------- drivers/crypto/ccp/tee-dev.h | 11 +-- drivers/tee/amdtee/amdtee_private.h | 18 ++--- drivers/tee/amdtee/call.c | 74 ++++++++--------- drivers/tee/amdtee/core.c | 72 ++++++++++------- drivers/tee/amdtee/shm_pool.c | 21 ++--- include/linux/psp-tee.h | 47 +++++++++++ 8 files changed, 221 insertions(+), 144 deletions(-) -- 2.25.1
Hi Jeshwank, On Wed, 25 Oct 2023 at 12:27, jeshwank <JESHWANTHKUMAR.NK@amd.com> wrote: > > From: Jeshwanth Kumar N K <JESHWANTHKUMAR.NK@amd.com> > > At present, the shared memory for TEE ring buffer, command buffer and > data buffer is allocated using get_free_pages(). The driver shares the > physical address of these buffers with PSP so that it can be mapped by > the Trusted OS. > > In this patch series we have replaced get_free_pages() with > dma_alloc_coherent() to allocate shared memory to cleanup the existing > allocation method. Thanks for putting this together but I can't find the reasoning behind this change neither in this commit message and nor in the patch descriptions. Care to explain why? -Sumit > > Rijo Thomas (3): > crypto: ccp - Add function to allocate and free memory using DMA APIs > crypto: ccp - Use psp_tee_alloc_buffer() and psp_tee_free_buffer() > tee: amdtee: Use psp_tee_alloc_buffer() and psp_tee_free_buffer() > > drivers/crypto/ccp/psp-dev.c | 3 + > drivers/crypto/ccp/tee-dev.c | 119 ++++++++++++++++++---------- > drivers/crypto/ccp/tee-dev.h | 11 +-- > drivers/tee/amdtee/amdtee_private.h | 18 ++--- > drivers/tee/amdtee/call.c | 74 ++++++++--------- > drivers/tee/amdtee/core.c | 72 ++++++++++------- > drivers/tee/amdtee/shm_pool.c | 21 ++--- > include/linux/psp-tee.h | 47 +++++++++++ > 8 files changed, 221 insertions(+), 144 deletions(-) > > -- > 2.25.1 >
On 25-Oct-23 7:01 PM, Sumit Garg wrote: > Hi Jeshwank, > > On Wed, 25 Oct 2023 at 12:27, jeshwank <JESHWANTHKUMAR.NK@amd.com> wrote: >> From: Jeshwanth Kumar N K <JESHWANTHKUMAR.NK@amd.com> >> >> At present, the shared memory for TEE ring buffer, command buffer and >> data buffer is allocated using get_free_pages(). The driver shares the >> physical address of these buffers with PSP so that it can be mapped by >> the Trusted OS. >> >> In this patch series we have replaced get_free_pages() with >> dma_alloc_coherent() to allocate shared memory to cleanup the existing >> allocation method. > Thanks for putting this together but I can't find the reasoning behind > this change neither in this commit message and nor in the patch > descriptions. Care to explain why? > > -Sumit Hi Sumit, We see that there is an advantage in using dma_alloc_coherent() over get_free_pages(). The dma-ops associated with PSP PCIe device can be overridden. This capability will be helpful when we enable virtualization support. We plan to post a virtualization related patch in future. Regards, Jeshwanth > >> Rijo Thomas (3): >> crypto: ccp - Add function to allocate and free memory using DMA APIs >> crypto: ccp - Use psp_tee_alloc_buffer() and psp_tee_free_buffer() >> tee: amdtee: Use psp_tee_alloc_buffer() and psp_tee_free_buffer() >> >> drivers/crypto/ccp/psp-dev.c | 3 + >> drivers/crypto/ccp/tee-dev.c | 119 ++++++++++++++++++---------- >> drivers/crypto/ccp/tee-dev.h | 11 +-- >> drivers/tee/amdtee/amdtee_private.h | 18 ++--- >> drivers/tee/amdtee/call.c | 74 ++++++++--------- >> drivers/tee/amdtee/core.c | 72 ++++++++++------- >> drivers/tee/amdtee/shm_pool.c | 21 ++--- >> include/linux/psp-tee.h | 47 +++++++++++ >> 8 files changed, 221 insertions(+), 144 deletions(-) >> >> -- >> 2.25.1 >>
On 10/26/23 05:30, NK, JESHWANTHKUMAR wrote: > > On 25-Oct-23 7:01 PM, Sumit Garg wrote: >> Hi Jeshwank, >> >> On Wed, 25 Oct 2023 at 12:27, jeshwank <JESHWANTHKUMAR.NK@amd.com> wrote: >>> From: Jeshwanth Kumar N K <JESHWANTHKUMAR.NK@amd.com> >>> >>> At present, the shared memory for TEE ring buffer, command buffer and >>> data buffer is allocated using get_free_pages(). The driver shares the >>> physical address of these buffers with PSP so that it can be mapped by >>> the Trusted OS. >>> >>> In this patch series we have replaced get_free_pages() with >>> dma_alloc_coherent() to allocate shared memory to cleanup the existing >>> allocation method. >> Thanks for putting this together but I can't find the reasoning behind >> this change neither in this commit message and nor in the patch >> descriptions. Care to explain why? >> >> -Sumit > Hi Sumit, > > We see that there is an advantage in using dma_alloc_coherent() over > get_free_pages(). The dma-ops associated with PSP PCIe device can be > overridden. This capability will be helpful when we enable virtualization > support. We plan to post a virtualization related patch in future. To be specific, you are referring to Xen virtualization support, correct? Because I don't see how this works in a Qemu/KVM environment where you would get a GPA and not an SPA. If that is the case, you should clearly specify that. Also, this looks like it should be introduced with the virtualization support that you submit in the future and not before. Thanks, Tom > > Regards, > > Jeshwanth > >> >>> Rijo Thomas (3): >>> crypto: ccp - Add function to allocate and free memory using DMA APIs >>> crypto: ccp - Use psp_tee_alloc_buffer() and psp_tee_free_buffer() >>> tee: amdtee: Use psp_tee_alloc_buffer() and psp_tee_free_buffer() >>> >>> drivers/crypto/ccp/psp-dev.c | 3 + >>> drivers/crypto/ccp/tee-dev.c | 119 ++++++++++++++++++---------- >>> drivers/crypto/ccp/tee-dev.h | 11 +-- >>> drivers/tee/amdtee/amdtee_private.h | 18 ++--- >>> drivers/tee/amdtee/call.c | 74 ++++++++--------- >>> drivers/tee/amdtee/core.c | 72 ++++++++++------- >>> drivers/tee/amdtee/shm_pool.c | 21 ++--- >>> include/linux/psp-tee.h | 47 +++++++++++ >>> 8 files changed, 221 insertions(+), 144 deletions(-) >>> >>> -- >>> 2.25.1 >>>
Hi Tom, On 26-Oct-23 8:23 PM, Tom Lendacky wrote: > On 10/26/23 05:30, NK, JESHWANTHKUMAR wrote: >> >> On 25-Oct-23 7:01 PM, Sumit Garg wrote: >>> Hi Jeshwank, >>> >>> On Wed, 25 Oct 2023 at 12:27, jeshwank <JESHWANTHKUMAR.NK@amd.com> >>> wrote: >>>> From: Jeshwanth Kumar N K <JESHWANTHKUMAR.NK@amd.com> >>>> >>>> At present, the shared memory for TEE ring buffer, command buffer and >>>> data buffer is allocated using get_free_pages(). The driver shares the >>>> physical address of these buffers with PSP so that it can be mapped by >>>> the Trusted OS. >>>> >>>> In this patch series we have replaced get_free_pages() with >>>> dma_alloc_coherent() to allocate shared memory to cleanup the existing >>>> allocation method. >>> Thanks for putting this together but I can't find the reasoning behind >>> this change neither in this commit message and nor in the patch >>> descriptions. Care to explain why? >>> >>> -Sumit >> Hi Sumit, >> >> We see that there is an advantage in using dma_alloc_coherent() over >> get_free_pages(). The dma-ops associated with PSP PCIe device can be >> overridden. This capability will be helpful when we enable >> virtualization support. We plan to post a virtualization related >> patch in future. > > To be specific, you are referring to Xen virtualization support, > correct? Because I don't see how this works in a Qemu/KVM environment > where you would get a GPA and not an SPA. The patch is not specific to Xen. We have verified it in Qemu/KVM and Xen PV mode. Support for Xen PVH mode will be added as a separate patch. > > If that is the case, you should clearly specify that. Also, this looks > like it should be introduced with the virtualization support that you > submit in the future and not before. I will update the commit message in the next version of the patch series to include these details. > Thanks, > Tom > >> >> Regards, >> >> Jeshwanth >> >>> >>>> Rijo Thomas (3): >>>> crypto: ccp - Add function to allocate and free memory using DMA >>>> APIs >>>> crypto: ccp - Use psp_tee_alloc_buffer() and psp_tee_free_buffer() >>>> tee: amdtee: Use psp_tee_alloc_buffer() and psp_tee_free_buffer() >>>> >>>> drivers/crypto/ccp/psp-dev.c | 3 + >>>> drivers/crypto/ccp/tee-dev.c | 119 >>>> ++++++++++++++++++---------- >>>> drivers/crypto/ccp/tee-dev.h | 11 +-- >>>> drivers/tee/amdtee/amdtee_private.h | 18 ++--- >>>> drivers/tee/amdtee/call.c | 74 ++++++++--------- >>>> drivers/tee/amdtee/core.c | 72 ++++++++++------- >>>> drivers/tee/amdtee/shm_pool.c | 21 ++--- >>>> include/linux/psp-tee.h | 47 +++++++++++ >>>> 8 files changed, 221 insertions(+), 144 deletions(-) >>>> >>>> -- >>>> 2.25.1 >>>>
© 2016 - 2026 Red Hat, Inc.