Documentation/dev-tools/autofdo.rst | 168 ++++++++++++++++++++++++++ Documentation/dev-tools/index.rst | 2 + Documentation/dev-tools/propeller.rst | 162 +++++++++++++++++++++++++ MAINTAINERS | 14 +++ Makefile | 2 + arch/Kconfig | 39 ++++++ arch/sparc/kernel/vmlinux.lds.S | 5 + arch/x86/Kconfig | 2 + arch/x86/kernel/vmlinux.lds.S | 4 + include/asm-generic/vmlinux.lds.h | 49 ++++++-- scripts/Makefile.autofdo | 24 ++++ scripts/Makefile.lib | 20 +++ scripts/Makefile.propeller | 28 +++++ tools/objtool/check.c | 2 + tools/objtool/elf.c | 15 ++- 15 files changed, 520 insertions(+), 16 deletions(-) create mode 100644 Documentation/dev-tools/autofdo.rst create mode 100644 Documentation/dev-tools/propeller.rst create mode 100644 scripts/Makefile.autofdo create mode 100644 scripts/Makefile.propeller
Hi,
This patch series is to integrate AutoFDO and Propeller support into
the Linux kernel. AutoFDO is a profile-guided optimization technique
that leverages hardware sampling to enhance binary performance.
Unlike Instrumentation-based FDO (iFDO), AutoFDO offers a user-friendly
and straightforward application process. While iFDO generally yields
superior profile quality and performance, our findings reveal that
AutoFDO achieves remarkable effectiveness, bringing performance close
to iFDO for benchmark applications.
Propeller is a profile-guided, post-link optimizer that improves
the performance of large-scale applications compiled with LLVM. It
operates by relinking the binary based on an additional round of runtime
profiles, enabling precise optimizations that are not possible at
compile time. Similar to AutoFDO, Propeller too utilizes hardware
sampling to collect profiles and apply post-link optimizations to improve
the benchmark’s performance over and above AutoFDO.
Our empirical data demonstrates significant performance improvements
with AutoFDO and Propeller, up to 10% on microbenchmarks and up to 5%
on large warehouse-scale benchmarks. This makes a strong case for their
inclusion as supported features in the upstream kernel.
Background
A significant fraction of fleet processing cycles (excluding idle time)
from data center workloads are attributable to the kernel. Ware-house
scale workloads maximize performance by optimizing the production kernel
using iFDO (a.k.a instrumented PGO, Profile Guided Optimization).
iFDO can significantly enhance application performance but its use
within the kernel has raised concerns. AutoFDO is a variant of FDO that
uses the hardware’s Performance Monitoring Unit (PMU) to collect
profiling data. While AutoFDO typically yields smaller performance
gains than iFDO, it presents unique benefits for optimizing kernels.
AutoFDO eliminates the need for instrumented kernels, allowing a single
optimized kernel to serve both execution and profile collection. It also
minimizes slowdown during profile collection, potentially yielding
higher-fidelity profiling, especially for time-sensitive code, compared
to iFDO. Additionally, AutoFDO profiles can be obtained from production
environments via the hardware’s PMU whereas iFDO profiles require
carefully curated load tests that are representative of real-world
traffic.
AutoFDO facilitates profile collection across diverse targets.
Preliminary studies indicate significant variation in kernel hot spots
within Google’s infrastructure, suggesting potential performance gains
through target-specific kernel customization.
Furthermore, other advanced compiler optimization techniques, including
ThinLTO and Propeller can be stacked on top of AutoFDO, similar to iFDO.
ThinLTO achieves better runtime performance through whole-program
analysis and cross module optimizations. The main difference between
traditional LTO and ThinLTO is that the latter is scalable in time and
memory.
This patch series adds AutoFDO and Propeller support to the kernel. The
actual solution comes in six parts:
[P 1] Add the build support for using AutoFDO in Clang
Add the basic support for AutoFDO build and provide the
instructions for using AutoFDO.
[P 2] Fix objtool for bogus warnings when -ffunction-sections is enabled
[P 3] Adjust symbol ordering in text output sections
[P 4] Add markers for text_unlikely and text_hot sections
[P 5] Enable –ffunction-sections for the AutoFDO build
[P 6] Enable Machine Function Split (MFS) optimization for AutoFDO
[P 7] Add Propeller configuration to the kernel build
Patch 1 provides basic AutoFDO build support. Patches 2 to 6 further
enhance the performance of AutoFDO builds and are functionally dependent
on Patch 1. Patch 7 enables support for Propeller and is dependent on
patch 2 to patch 4.
Caveats
AutoFDO is compatible with both GCC and Clang, but the patches in this
series are exclusively applicable to LLVM 17 or newer for AutoFDO and
LLVM 19 or newer for Propeller. For profile conversion, two different
tools could be used, llvm_profgen or create_llvm_prof. llvm_profgen
needs to be the LLVM 19 or newer, or just the LLVM trunk. Alternatively,
create_llvm_prof v0.30.1 or newer can be used instead of llvm-profgen.
Additionally, the build is only supported on x86 platforms equipped
with PMU capabilities, such as LBR on Intel machines. More
specifically:
* Intel platforms: works on every platform that supports LBR;
we have tested on Skylake.
* AMD platforms: tested on AMD Zen3 with the BRS feature. The kernel
needs to be configured with “CONFIG_PERF_EVENTS_AMD_BRS=y", To
check, use
$ cat /proc/cpuinfo | grep “ brs”
For the AMD Zen4, AMD LBRV2 is supported, but we suspect a bug with
AMD LBRv2 implementation in Genoa which blocks the usage.
For ARM, we plan to send patches for SPE-based Propeller when
AutoFDO for Arm is ready.
Experiments and Results
Experiments were conducted to compare the performance of AutoFDO-optimized
kernel images (version 6.9.x) against default builds.. The evaluation
encompassed both open source microbenchmarks and real-world production
services from Google and Meta. The selected microbenchmarks included Neper,
a network subsystem benchmark, and UnixBench which is a comprehensive suite
for assessing various kernel operations.
For Neper, AutoFDO optimization resulted in a 6.1% increase in throughput
and a 10.6% reduction in latency. UnixBench saw a 2.2% improvement in its
index score under low system load and a 2.6% improvement under high system
load.
For further details on the improvements observed in Google and Meta's
production services, please refer to the LLVM discourse post:
https://discourse.llvm.org/t/optimizing-the-linux-kernel-with-autofdo-including-thinlto-and-propeller/79108
Thanks,
Rong Xu and Han Shen
---
Change-Logs in V2:
Rebased to commit e32cde8d2bd7 ("Merge tag 'sched_ext-for-6.12-rc1-fixes-1'
of git://git.kernel.org/pub/scm/linux/kernel/git/tj/sched_ext")
1. [P 0]: moved the Propeller description to the top (Peter Zijlstra)
2. [P 1]: (1) Makefile: fixed file order (Masahiro Yamada)
(2) scripts/Makefile.lib: used is-kernel-object to exclude
files (Masahiro Yamada)
(3) scripts/Makefile.autofdo: improved the code (Masahiro Yamada)
(4) scripts/Makefile.autofdo: handled when DEBUG_INFO disabled
(Nick Desaulniers)
3. [P 2]: tools/objtool/elf.c: updated the comments (Peter Zijlstra)
4. [P 3]: include/asm-generic/vmlinux.lds.h:
(1) explicit set cold text function aligned (Peter Zijlstra and
Peter Anvin)
(2) set hot-text page aligned
5. [P 6]: (1) include/asm-generic/vmlinux.lds.h: made Propeller not
depending on AutoFDO
(2) Makefile: fixed file order (Masahiro Yamada)
(3) scripts/Makefile.lib: used is-kernel-object to exclude
files (Masahiro Yamada). This removed the change in
arch/x86/platform/efi/Makefile,
drivers/firmware/efi/libstub/Makefile, and
arch/x86/boot/compressed/Makefile.
And this also addressed the comment from Arnd Bergmann
regarding arch/x86/purgatory/Makefile
(4) scripts/Makefile.propeller: improved the code
(Masahiro Yamada)
Change-Logs in V3:
Rebased to commit eb952c47d154 ("Merge tag 'for-6.12-rc2-tag' of
git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux")
Integrated the following changes suggested by Mike Rapoport.
1. [P 1]: autofdo.rst: removed code-block directives and used "::"
2. [P 6]: propeller.rst: removed code-block directives and use "::"
Change-Logs in V4:
1. [P 1]: autofdo.rst: fixed a typo for create_llvm_prof command.
Change-Logs in V5:
Added "Tested-by: Yonghong Song <yonghong.song@linux.dev>" to all patches.
Integrated the following changes suggested by Masahiro Yamada.
1. [P 0]: (1) moved ARM related remark from patch 6 to here
2. [P 1]: (1) autofdo.rst: improved the documentation
(2) scripts/Makefile.autofdo: improved comments and used ifdef
instead of ifeq
3. [P 3]: Make the layout change unconditionally
4. [P 4]: Split the patch into two: this patch only added the markers, and
the AutoFDO change went to new P_5
5. [P 7]: (1) propeller.rst: improved the documentation
(2) scripts/Makefile.propeller: improved comments and used ifdef
instead of ifeq
(3) arch/Kconfig: made Propeller build independent of AutoFDO
build
(4) moved ARM related remarks to the cover letter
Change-Logs in V6:
Added "Tested-by: Yabin Cui <yabinc@google.com>" to AutoFDO patches.
1. [P 3]: (1) changed patch title
(2) fixed the build error in sparc64
Change-Logs in V7:
Rebased to commit 11066801dd4b7 (Merge tag 'linux_kselftest-fixes-6.12-rc6'
of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest)
Added "Tested-by: Nathan Chancellor <nathan@kernel.org>" to [P 1] to [P 7]
Added "Reviewed-by: Kees Cook <kees@kernel.org>" to [P 1] to [P 7]
Added "Acked-by: Josh Poimboeuf <jpoimboe@kernel.org>" to [P 2]
Integrated the following changes suggested by Masahiro Yamada.
1. [P 1]: autofdo.rst: fixed format
2. [P 3]: commit message: described the rationale behind the new layout
Rong Xu (7):
Add AutoFDO support for Clang build
objtool: Fix unreachable instruction warnings for weak functions
Adjust symbol ordering in text output section
Add markers for text_unlikely and text_hot sections
AutoFDO: Enable -ffunction-sections for the AutoFDO build
AutoFDO: Enable machine function split optimization for AutoFDO
Add Propeller configuration for kernel build
Documentation/dev-tools/autofdo.rst | 168 ++++++++++++++++++++++++++
Documentation/dev-tools/index.rst | 2 +
Documentation/dev-tools/propeller.rst | 162 +++++++++++++++++++++++++
MAINTAINERS | 14 +++
Makefile | 2 +
arch/Kconfig | 39 ++++++
arch/sparc/kernel/vmlinux.lds.S | 5 +
arch/x86/Kconfig | 2 +
arch/x86/kernel/vmlinux.lds.S | 4 +
include/asm-generic/vmlinux.lds.h | 49 ++++++--
scripts/Makefile.autofdo | 24 ++++
scripts/Makefile.lib | 20 +++
scripts/Makefile.propeller | 28 +++++
tools/objtool/check.c | 2 +
tools/objtool/elf.c | 15 ++-
15 files changed, 520 insertions(+), 16 deletions(-)
create mode 100644 Documentation/dev-tools/autofdo.rst
create mode 100644 Documentation/dev-tools/propeller.rst
create mode 100644 scripts/Makefile.autofdo
create mode 100644 scripts/Makefile.propeller
base-commit: 11066801dd4b7c4d75fce65c812723a80c1481ae
--
2.47.0.163.g1226f6d8fa-goog
On Sun, Nov 3, 2024 at 2:51 AM Rong Xu <xur@google.com> wrote: > > Hi, > > This patch series is to integrate AutoFDO and Propeller support into > the Linux kernel. AutoFDO is a profile-guided optimization technique > that leverages hardware sampling to enhance binary performance. > Unlike Instrumentation-based FDO (iFDO), AutoFDO offers a user-friendly > and straightforward application process. While iFDO generally yields > superior profile quality and performance, our findings reveal that > AutoFDO achieves remarkable effectiveness, bringing performance close > to iFDO for benchmark applications. > > Propeller is a profile-guided, post-link optimizer that improves > the performance of large-scale applications compiled with LLVM. It > operates by relinking the binary based on an additional round of runtime > profiles, enabling precise optimizations that are not possible at > compile time. Similar to AutoFDO, Propeller too utilizes hardware > sampling to collect profiles and apply post-link optimizations to improve > the benchmark’s performance over and above AutoFDO. > > Our empirical data demonstrates significant performance improvements > with AutoFDO and Propeller, up to 10% on microbenchmarks and up to 5% > on large warehouse-scale benchmarks. This makes a strong case for their > inclusion as supported features in the upstream kernel. > > Background > > A significant fraction of fleet processing cycles (excluding idle time) > from data center workloads are attributable to the kernel. Ware-house > scale workloads maximize performance by optimizing the production kernel > using iFDO (a.k.a instrumented PGO, Profile Guided Optimization). > > iFDO can significantly enhance application performance but its use > within the kernel has raised concerns. AutoFDO is a variant of FDO that > uses the hardware’s Performance Monitoring Unit (PMU) to collect > profiling data. While AutoFDO typically yields smaller performance > gains than iFDO, it presents unique benefits for optimizing kernels. > > AutoFDO eliminates the need for instrumented kernels, allowing a single > optimized kernel to serve both execution and profile collection. It also > minimizes slowdown during profile collection, potentially yielding > higher-fidelity profiling, especially for time-sensitive code, compared > to iFDO. Additionally, AutoFDO profiles can be obtained from production > environments via the hardware’s PMU whereas iFDO profiles require > carefully curated load tests that are representative of real-world > traffic. > > AutoFDO facilitates profile collection across diverse targets. > Preliminary studies indicate significant variation in kernel hot spots > within Google’s infrastructure, suggesting potential performance gains > through target-specific kernel customization. > > Furthermore, other advanced compiler optimization techniques, including > ThinLTO and Propeller can be stacked on top of AutoFDO, similar to iFDO. > ThinLTO achieves better runtime performance through whole-program > analysis and cross module optimizations. The main difference between > traditional LTO and ThinLTO is that the latter is scalable in time and > memory. > > This patch series adds AutoFDO and Propeller support to the kernel. The > actual solution comes in six parts: > > [P 1] Add the build support for using AutoFDO in Clang > > Add the basic support for AutoFDO build and provide the > instructions for using AutoFDO. > > [P 2] Fix objtool for bogus warnings when -ffunction-sections is enabled > > [P 3] Adjust symbol ordering in text output sections > > [P 4] Add markers for text_unlikely and text_hot sections > > [P 5] Enable –ffunction-sections for the AutoFDO build > > [P 6] Enable Machine Function Split (MFS) optimization for AutoFDO > > [P 7] Add Propeller configuration to the kernel build > > Patch 1 provides basic AutoFDO build support. Patches 2 to 6 further > enhance the performance of AutoFDO builds and are functionally dependent > on Patch 1. Patch 7 enables support for Propeller and is dependent on > patch 2 to patch 4. > > Caveats > > AutoFDO is compatible with both GCC and Clang, but the patches in this > series are exclusively applicable to LLVM 17 or newer for AutoFDO and > LLVM 19 or newer for Propeller. For profile conversion, two different > tools could be used, llvm_profgen or create_llvm_prof. llvm_profgen > needs to be the LLVM 19 or newer, or just the LLVM trunk. Alternatively, > create_llvm_prof v0.30.1 or newer can be used instead of llvm-profgen. > > Additionally, the build is only supported on x86 platforms equipped > with PMU capabilities, such as LBR on Intel machines. More > specifically: > * Intel platforms: works on every platform that supports LBR; > we have tested on Skylake. > * AMD platforms: tested on AMD Zen3 with the BRS feature. The kernel > needs to be configured with “CONFIG_PERF_EVENTS_AMD_BRS=y", To > check, use > $ cat /proc/cpuinfo | grep “ brs” > For the AMD Zen4, AMD LBRV2 is supported, but we suspect a bug with > AMD LBRv2 implementation in Genoa which blocks the usage. > > For ARM, we plan to send patches for SPE-based Propeller when > AutoFDO for Arm is ready. > > Experiments and Results > > Experiments were conducted to compare the performance of AutoFDO-optimized > kernel images (version 6.9.x) against default builds.. The evaluation > encompassed both open source microbenchmarks and real-world production > services from Google and Meta. The selected microbenchmarks included Neper, > a network subsystem benchmark, and UnixBench which is a comprehensive suite > for assessing various kernel operations. > > For Neper, AutoFDO optimization resulted in a 6.1% increase in throughput > and a 10.6% reduction in latency. UnixBench saw a 2.2% improvement in its > index score under low system load and a 2.6% improvement under high system > load. > > For further details on the improvements observed in Google and Meta's > production services, please refer to the LLVM discourse post: > https://discourse.llvm.org/t/optimizing-the-linux-kernel-with-autofdo-including-thinlto-and-propeller/79108 > > Thanks, > > Rong Xu and Han Shen I applied this series to linux-kbuild. As I mentioned before, I do not like #ifdef because it hides (not fixes) issues only for default cases. -- Best Regards Masahiro Yamada
On Wed, Nov 6, 2024 at 8:09 AM Masahiro Yamada <masahiroy@kernel.org> wrote: > > On Sun, Nov 3, 2024 at 2:51 AM Rong Xu <xur@google.com> wrote: > > > > Hi, > > > > This patch series is to integrate AutoFDO and Propeller support into > > the Linux kernel. AutoFDO is a profile-guided optimization technique > > that leverages hardware sampling to enhance binary performance. > > Unlike Instrumentation-based FDO (iFDO), AutoFDO offers a user-friendly > > and straightforward application process. While iFDO generally yields > > superior profile quality and performance, our findings reveal that > > AutoFDO achieves remarkable effectiveness, bringing performance close > > to iFDO for benchmark applications. > > > > Propeller is a profile-guided, post-link optimizer that improves > > the performance of large-scale applications compiled with LLVM. It > > operates by relinking the binary based on an additional round of runtime > > profiles, enabling precise optimizations that are not possible at > > compile time. Similar to AutoFDO, Propeller too utilizes hardware > > sampling to collect profiles and apply post-link optimizations to improve > > the benchmark’s performance over and above AutoFDO. > > > > Our empirical data demonstrates significant performance improvements > > with AutoFDO and Propeller, up to 10% on microbenchmarks and up to 5% > > on large warehouse-scale benchmarks. This makes a strong case for their > > inclusion as supported features in the upstream kernel. > > > > Background > > > > A significant fraction of fleet processing cycles (excluding idle time) > > from data center workloads are attributable to the kernel. Ware-house > > scale workloads maximize performance by optimizing the production kernel > > using iFDO (a.k.a instrumented PGO, Profile Guided Optimization). > > > > iFDO can significantly enhance application performance but its use > > within the kernel has raised concerns. AutoFDO is a variant of FDO that > > uses the hardware’s Performance Monitoring Unit (PMU) to collect > > profiling data. While AutoFDO typically yields smaller performance > > gains than iFDO, it presents unique benefits for optimizing kernels. > > > > AutoFDO eliminates the need for instrumented kernels, allowing a single > > optimized kernel to serve both execution and profile collection. It also > > minimizes slowdown during profile collection, potentially yielding > > higher-fidelity profiling, especially for time-sensitive code, compared > > to iFDO. Additionally, AutoFDO profiles can be obtained from production > > environments via the hardware’s PMU whereas iFDO profiles require > > carefully curated load tests that are representative of real-world > > traffic. > > > > AutoFDO facilitates profile collection across diverse targets. > > Preliminary studies indicate significant variation in kernel hot spots > > within Google’s infrastructure, suggesting potential performance gains > > through target-specific kernel customization. > > > > Furthermore, other advanced compiler optimization techniques, including > > ThinLTO and Propeller can be stacked on top of AutoFDO, similar to iFDO. > > ThinLTO achieves better runtime performance through whole-program > > analysis and cross module optimizations. The main difference between > > traditional LTO and ThinLTO is that the latter is scalable in time and > > memory. > > > > This patch series adds AutoFDO and Propeller support to the kernel. The > > actual solution comes in six parts: > > > > [P 1] Add the build support for using AutoFDO in Clang > > > > Add the basic support for AutoFDO build and provide the > > instructions for using AutoFDO. > > > > [P 2] Fix objtool for bogus warnings when -ffunction-sections is enabled > > > > [P 3] Adjust symbol ordering in text output sections > > > > [P 4] Add markers for text_unlikely and text_hot sections > > > > [P 5] Enable –ffunction-sections for the AutoFDO build > > > > [P 6] Enable Machine Function Split (MFS) optimization for AutoFDO > > > > [P 7] Add Propeller configuration to the kernel build > > > > Patch 1 provides basic AutoFDO build support. Patches 2 to 6 further > > enhance the performance of AutoFDO builds and are functionally dependent > > on Patch 1. Patch 7 enables support for Propeller and is dependent on > > patch 2 to patch 4. > > > > Caveats > > > > AutoFDO is compatible with both GCC and Clang, but the patches in this > > series are exclusively applicable to LLVM 17 or newer for AutoFDO and > > LLVM 19 or newer for Propeller. For profile conversion, two different > > tools could be used, llvm_profgen or create_llvm_prof. llvm_profgen > > needs to be the LLVM 19 or newer, or just the LLVM trunk. Alternatively, > > create_llvm_prof v0.30.1 or newer can be used instead of llvm-profgen. > > > > Additionally, the build is only supported on x86 platforms equipped > > with PMU capabilities, such as LBR on Intel machines. More > > specifically: > > * Intel platforms: works on every platform that supports LBR; > > we have tested on Skylake. > > * AMD platforms: tested on AMD Zen3 with the BRS feature. The kernel > > needs to be configured with “CONFIG_PERF_EVENTS_AMD_BRS=y", To > > check, use > > $ cat /proc/cpuinfo | grep “ brs” > > For the AMD Zen4, AMD LBRV2 is supported, but we suspect a bug with > > AMD LBRv2 implementation in Genoa which blocks the usage. > > > > For ARM, we plan to send patches for SPE-based Propeller when > > AutoFDO for Arm is ready. > > > > Experiments and Results > > > > Experiments were conducted to compare the performance of AutoFDO-optimized > > kernel images (version 6.9.x) against default builds.. The evaluation > > encompassed both open source microbenchmarks and real-world production > > services from Google and Meta. The selected microbenchmarks included Neper, > > a network subsystem benchmark, and UnixBench which is a comprehensive suite > > for assessing various kernel operations. > > > > For Neper, AutoFDO optimization resulted in a 6.1% increase in throughput > > and a 10.6% reduction in latency. UnixBench saw a 2.2% improvement in its > > index score under low system load and a 2.6% improvement under high system > > load. > > > > For further details on the improvements observed in Google and Meta's > > production services, please refer to the LLVM discourse post: > > https://discourse.llvm.org/t/optimizing-the-linux-kernel-with-autofdo-including-thinlto-and-propeller/79108 > > > > Thanks, > > > > Rong Xu and Han Shen > > > I applied this series to linux-kbuild. > Thanks for taking the patch! > As I mentioned before, I do not like #ifdef because > it hides (not fixes) issues only for default cases. We followed the suggestion and removed most of the #if (or #ifdef) in the linker script. I just checked: there are two #ifdef remaining: (1) in the propeller patch for .llvm_bb_addr_map (2) in linker script patch for arch/sparc/kernel/vmlinux.lds.S. I think it's likely safe to remove the checks for head_64.o in non-SPARC64 builds and .llvm_bb_addr_map symbols in non-propeller builds. SPARC64 builds should always produce head_64.o, and non-SPARC64 builds shouldn't. Propeller builds always generate .llvm_bb_addr_map symbols, and the linker will omit the section if it's empty in non-propeller builds. Keeping the checks is harmless and might slightly reduce linker workload for matching. But If you'd prefer to remove them, I'm happy to provide a patch. Best regards, -Rong > > -- > Best Regards > Masahiro Yamada
On Thu, Nov 7, 2024 at 4:00 AM Rong Xu <xur@google.com> wrote: > > On Wed, Nov 6, 2024 at 8:09 AM Masahiro Yamada <masahiroy@kernel.org> wrote: > > > > On Sun, Nov 3, 2024 at 2:51 AM Rong Xu <xur@google.com> wrote: > > > > > > Hi, > > > > > > This patch series is to integrate AutoFDO and Propeller support into > > > the Linux kernel. AutoFDO is a profile-guided optimization technique > > > that leverages hardware sampling to enhance binary performance. > > > Unlike Instrumentation-based FDO (iFDO), AutoFDO offers a user-friendly > > > and straightforward application process. While iFDO generally yields > > > superior profile quality and performance, our findings reveal that > > > AutoFDO achieves remarkable effectiveness, bringing performance close > > > to iFDO for benchmark applications. > > > > > > Propeller is a profile-guided, post-link optimizer that improves > > > the performance of large-scale applications compiled with LLVM. It > > > operates by relinking the binary based on an additional round of runtime > > > profiles, enabling precise optimizations that are not possible at > > > compile time. Similar to AutoFDO, Propeller too utilizes hardware > > > sampling to collect profiles and apply post-link optimizations to improve > > > the benchmark’s performance over and above AutoFDO. > > > > > > Our empirical data demonstrates significant performance improvements > > > with AutoFDO and Propeller, up to 10% on microbenchmarks and up to 5% > > > on large warehouse-scale benchmarks. This makes a strong case for their > > > inclusion as supported features in the upstream kernel. > > > > > > Background > > > > > > A significant fraction of fleet processing cycles (excluding idle time) > > > from data center workloads are attributable to the kernel. Ware-house > > > scale workloads maximize performance by optimizing the production kernel > > > using iFDO (a.k.a instrumented PGO, Profile Guided Optimization). > > > > > > iFDO can significantly enhance application performance but its use > > > within the kernel has raised concerns. AutoFDO is a variant of FDO that > > > uses the hardware’s Performance Monitoring Unit (PMU) to collect > > > profiling data. While AutoFDO typically yields smaller performance > > > gains than iFDO, it presents unique benefits for optimizing kernels. > > > > > > AutoFDO eliminates the need for instrumented kernels, allowing a single > > > optimized kernel to serve both execution and profile collection. It also > > > minimizes slowdown during profile collection, potentially yielding > > > higher-fidelity profiling, especially for time-sensitive code, compared > > > to iFDO. Additionally, AutoFDO profiles can be obtained from production > > > environments via the hardware’s PMU whereas iFDO profiles require > > > carefully curated load tests that are representative of real-world > > > traffic. > > > > > > AutoFDO facilitates profile collection across diverse targets. > > > Preliminary studies indicate significant variation in kernel hot spots > > > within Google’s infrastructure, suggesting potential performance gains > > > through target-specific kernel customization. > > > > > > Furthermore, other advanced compiler optimization techniques, including > > > ThinLTO and Propeller can be stacked on top of AutoFDO, similar to iFDO. > > > ThinLTO achieves better runtime performance through whole-program > > > analysis and cross module optimizations. The main difference between > > > traditional LTO and ThinLTO is that the latter is scalable in time and > > > memory. > > > > > > This patch series adds AutoFDO and Propeller support to the kernel. The > > > actual solution comes in six parts: > > > > > > [P 1] Add the build support for using AutoFDO in Clang > > > > > > Add the basic support for AutoFDO build and provide the > > > instructions for using AutoFDO. > > > > > > [P 2] Fix objtool for bogus warnings when -ffunction-sections is enabled > > > > > > [P 3] Adjust symbol ordering in text output sections > > > > > > [P 4] Add markers for text_unlikely and text_hot sections > > > > > > [P 5] Enable –ffunction-sections for the AutoFDO build > > > > > > [P 6] Enable Machine Function Split (MFS) optimization for AutoFDO > > > > > > [P 7] Add Propeller configuration to the kernel build > > > > > > Patch 1 provides basic AutoFDO build support. Patches 2 to 6 further > > > enhance the performance of AutoFDO builds and are functionally dependent > > > on Patch 1. Patch 7 enables support for Propeller and is dependent on > > > patch 2 to patch 4. > > > > > > Caveats > > > > > > AutoFDO is compatible with both GCC and Clang, but the patches in this > > > series are exclusively applicable to LLVM 17 or newer for AutoFDO and > > > LLVM 19 or newer for Propeller. For profile conversion, two different > > > tools could be used, llvm_profgen or create_llvm_prof. llvm_profgen > > > needs to be the LLVM 19 or newer, or just the LLVM trunk. Alternatively, > > > create_llvm_prof v0.30.1 or newer can be used instead of llvm-profgen. > > > > > > Additionally, the build is only supported on x86 platforms equipped > > > with PMU capabilities, such as LBR on Intel machines. More > > > specifically: > > > * Intel platforms: works on every platform that supports LBR; > > > we have tested on Skylake. > > > * AMD platforms: tested on AMD Zen3 with the BRS feature. The kernel > > > needs to be configured with “CONFIG_PERF_EVENTS_AMD_BRS=y", To > > > check, use > > > $ cat /proc/cpuinfo | grep “ brs” > > > For the AMD Zen4, AMD LBRV2 is supported, but we suspect a bug with > > > AMD LBRv2 implementation in Genoa which blocks the usage. > > > > > > For ARM, we plan to send patches for SPE-based Propeller when > > > AutoFDO for Arm is ready. > > > > > > Experiments and Results > > > > > > Experiments were conducted to compare the performance of AutoFDO-optimized > > > kernel images (version 6.9.x) against default builds.. The evaluation > > > encompassed both open source microbenchmarks and real-world production > > > services from Google and Meta. The selected microbenchmarks included Neper, > > > a network subsystem benchmark, and UnixBench which is a comprehensive suite > > > for assessing various kernel operations. > > > > > > For Neper, AutoFDO optimization resulted in a 6.1% increase in throughput > > > and a 10.6% reduction in latency. UnixBench saw a 2.2% improvement in its > > > index score under low system load and a 2.6% improvement under high system > > > load. > > > > > > For further details on the improvements observed in Google and Meta's > > > production services, please refer to the LLVM discourse post: > > > https://discourse.llvm.org/t/optimizing-the-linux-kernel-with-autofdo-including-thinlto-and-propeller/79108 > > > > > > Thanks, > > > > > > Rong Xu and Han Shen > > > > > > I applied this series to linux-kbuild. > > > > Thanks for taking the patch! > > > As I mentioned before, I do not like #ifdef because > > it hides (not fixes) issues only for default cases. > > We followed the suggestion and removed most of the #if (or #ifdef) in > the linker script. > I just checked: there are two #ifdef remaining: > (1) in the propeller patch for .llvm_bb_addr_map > (2) in linker script patch for arch/sparc/kernel/vmlinux.lds.S. > > I think it's likely safe to remove the checks for head_64.o in > non-SPARC64 builds and .llvm_bb_addr_map symbols in non-propeller builds. > > SPARC64 builds should always produce head_64.o, and non-SPARC64 > builds shouldn't. > > Propeller builds always generate .llvm_bb_addr_map symbols, and the > linker will omit the section if it's empty in non-propeller builds. > > Keeping the checks is harmless and might slightly reduce linker > workload for matching. > But If you'd prefer to remove them, I'm happy to provide a patch. I am talking about the #ifdef in include/asm-generic/vmlinux.lds.h Yeah, it is me who (reluctantly) accepted cb87481ee89d. Now, the #ifdef has become a little more complicated. The default case is safe, but there are hidden issues. Some issues are easy to fix, so I sent some patches. https://lore.kernel.org/linux-kbuild/20241106161445.189399-1-masahiroy@kernel.org/T/#t https://lore.kernel.org/linux-kbuild/20241106161445.189399-1-masahiroy@kernel.org/T/#m4e4fa70386696e903b68d3fe1d7277e9a63fbefe https://lore.kernel.org/linux-kbuild/20241107111519.GA15424@willie-the-truck/T/#mccf6d49ddd11c90dcc583d7a68934bb3311da880 For example, see e41f501d3912. When CONFIG_LD_DEAD_CODE_DATA_ELIMINATION=y or CONFIG_LTO_CLANG=y or CONFIG_AUTOFDO_CLANG=y or CONFIG_PROPELLER_CLANG=y, the .text.startup sections will go to TEXT_MAIN instead of INIT_TEXT. This is not a fatal issue, but we cannot reuse memory for .text.startup sections. Removing the #ifdef (i.e. reverting cb87481ee89d) is more difficult because we need to take a closer look at potential impacts for all architectures. I understood you did not want to take a risk to break random architectures, so I decided to postpone the #ifdef issue and accept your patch set. -- Best Regards Masahiro Yamada
Thanks for the explanation. On Thu, Nov 7, 2024 at 6:58 AM Masahiro Yamada <masahiroy@kernel.org> wrote: > > On Thu, Nov 7, 2024 at 4:00 AM Rong Xu <xur@google.com> wrote: > > > > On Wed, Nov 6, 2024 at 8:09 AM Masahiro Yamada <masahiroy@kernel.org> wrote: > > > > > > On Sun, Nov 3, 2024 at 2:51 AM Rong Xu <xur@google.com> wrote: > > > > > > > > Hi, > > > > > > > > This patch series is to integrate AutoFDO and Propeller support into > > > > the Linux kernel. AutoFDO is a profile-guided optimization technique > > > > that leverages hardware sampling to enhance binary performance. > > > > Unlike Instrumentation-based FDO (iFDO), AutoFDO offers a user-friendly > > > > and straightforward application process. While iFDO generally yields > > > > superior profile quality and performance, our findings reveal that > > > > AutoFDO achieves remarkable effectiveness, bringing performance close > > > > to iFDO for benchmark applications. > > > > > > > > Propeller is a profile-guided, post-link optimizer that improves > > > > the performance of large-scale applications compiled with LLVM. It > > > > operates by relinking the binary based on an additional round of runtime > > > > profiles, enabling precise optimizations that are not possible at > > > > compile time. Similar to AutoFDO, Propeller too utilizes hardware > > > > sampling to collect profiles and apply post-link optimizations to improve > > > > the benchmark’s performance over and above AutoFDO. > > > > > > > > Our empirical data demonstrates significant performance improvements > > > > with AutoFDO and Propeller, up to 10% on microbenchmarks and up to 5% > > > > on large warehouse-scale benchmarks. This makes a strong case for their > > > > inclusion as supported features in the upstream kernel. > > > > > > > > Background > > > > > > > > A significant fraction of fleet processing cycles (excluding idle time) > > > > from data center workloads are attributable to the kernel. Ware-house > > > > scale workloads maximize performance by optimizing the production kernel > > > > using iFDO (a.k.a instrumented PGO, Profile Guided Optimization). > > > > > > > > iFDO can significantly enhance application performance but its use > > > > within the kernel has raised concerns. AutoFDO is a variant of FDO that > > > > uses the hardware’s Performance Monitoring Unit (PMU) to collect > > > > profiling data. While AutoFDO typically yields smaller performance > > > > gains than iFDO, it presents unique benefits for optimizing kernels. > > > > > > > > AutoFDO eliminates the need for instrumented kernels, allowing a single > > > > optimized kernel to serve both execution and profile collection. It also > > > > minimizes slowdown during profile collection, potentially yielding > > > > higher-fidelity profiling, especially for time-sensitive code, compared > > > > to iFDO. Additionally, AutoFDO profiles can be obtained from production > > > > environments via the hardware’s PMU whereas iFDO profiles require > > > > carefully curated load tests that are representative of real-world > > > > traffic. > > > > > > > > AutoFDO facilitates profile collection across diverse targets. > > > > Preliminary studies indicate significant variation in kernel hot spots > > > > within Google’s infrastructure, suggesting potential performance gains > > > > through target-specific kernel customization. > > > > > > > > Furthermore, other advanced compiler optimization techniques, including > > > > ThinLTO and Propeller can be stacked on top of AutoFDO, similar to iFDO. > > > > ThinLTO achieves better runtime performance through whole-program > > > > analysis and cross module optimizations. The main difference between > > > > traditional LTO and ThinLTO is that the latter is scalable in time and > > > > memory. > > > > > > > > This patch series adds AutoFDO and Propeller support to the kernel. The > > > > actual solution comes in six parts: > > > > > > > > [P 1] Add the build support for using AutoFDO in Clang > > > > > > > > Add the basic support for AutoFDO build and provide the > > > > instructions for using AutoFDO. > > > > > > > > [P 2] Fix objtool for bogus warnings when -ffunction-sections is enabled > > > > > > > > [P 3] Adjust symbol ordering in text output sections > > > > > > > > [P 4] Add markers for text_unlikely and text_hot sections > > > > > > > > [P 5] Enable –ffunction-sections for the AutoFDO build > > > > > > > > [P 6] Enable Machine Function Split (MFS) optimization for AutoFDO > > > > > > > > [P 7] Add Propeller configuration to the kernel build > > > > > > > > Patch 1 provides basic AutoFDO build support. Patches 2 to 6 further > > > > enhance the performance of AutoFDO builds and are functionally dependent > > > > on Patch 1. Patch 7 enables support for Propeller and is dependent on > > > > patch 2 to patch 4. > > > > > > > > Caveats > > > > > > > > AutoFDO is compatible with both GCC and Clang, but the patches in this > > > > series are exclusively applicable to LLVM 17 or newer for AutoFDO and > > > > LLVM 19 or newer for Propeller. For profile conversion, two different > > > > tools could be used, llvm_profgen or create_llvm_prof. llvm_profgen > > > > needs to be the LLVM 19 or newer, or just the LLVM trunk. Alternatively, > > > > create_llvm_prof v0.30.1 or newer can be used instead of llvm-profgen. > > > > > > > > Additionally, the build is only supported on x86 platforms equipped > > > > with PMU capabilities, such as LBR on Intel machines. More > > > > specifically: > > > > * Intel platforms: works on every platform that supports LBR; > > > > we have tested on Skylake. > > > > * AMD platforms: tested on AMD Zen3 with the BRS feature. The kernel > > > > needs to be configured with “CONFIG_PERF_EVENTS_AMD_BRS=y", To > > > > check, use > > > > $ cat /proc/cpuinfo | grep “ brs” > > > > For the AMD Zen4, AMD LBRV2 is supported, but we suspect a bug with > > > > AMD LBRv2 implementation in Genoa which blocks the usage. > > > > > > > > For ARM, we plan to send patches for SPE-based Propeller when > > > > AutoFDO for Arm is ready. > > > > > > > > Experiments and Results > > > > > > > > Experiments were conducted to compare the performance of AutoFDO-optimized > > > > kernel images (version 6.9.x) against default builds.. The evaluation > > > > encompassed both open source microbenchmarks and real-world production > > > > services from Google and Meta. The selected microbenchmarks included Neper, > > > > a network subsystem benchmark, and UnixBench which is a comprehensive suite > > > > for assessing various kernel operations. > > > > > > > > For Neper, AutoFDO optimization resulted in a 6.1% increase in throughput > > > > and a 10.6% reduction in latency. UnixBench saw a 2.2% improvement in its > > > > index score under low system load and a 2.6% improvement under high system > > > > load. > > > > > > > > For further details on the improvements observed in Google and Meta's > > > > production services, please refer to the LLVM discourse post: > > > > https://discourse.llvm.org/t/optimizing-the-linux-kernel-with-autofdo-including-thinlto-and-propeller/79108 > > > > > > > > Thanks, > > > > > > > > Rong Xu and Han Shen > > > > > > > > > I applied this series to linux-kbuild. > > > > > > > Thanks for taking the patch! > > > > > As I mentioned before, I do not like #ifdef because > > > it hides (not fixes) issues only for default cases. > > > > We followed the suggestion and removed most of the #if (or #ifdef) in > > the linker script. > > I just checked: there are two #ifdef remaining: > > (1) in the propeller patch for .llvm_bb_addr_map > > (2) in linker script patch for arch/sparc/kernel/vmlinux.lds.S. > > > > I think it's likely safe to remove the checks for head_64.o in > > non-SPARC64 builds and .llvm_bb_addr_map symbols in non-propeller builds. > > > > SPARC64 builds should always produce head_64.o, and non-SPARC64 > > builds shouldn't. > > > > Propeller builds always generate .llvm_bb_addr_map symbols, and the > > linker will omit the section if it's empty in non-propeller builds. > > > > Keeping the checks is harmless and might slightly reduce linker > > workload for matching. > > But If you'd prefer to remove them, I'm happy to provide a patch. > > > I am talking about the #ifdef in include/asm-generic/vmlinux.lds.h > > > Yeah, it is me who (reluctantly) accepted cb87481ee89d. > > Now, the #ifdef has become a little more complicated. > The default case is safe, but there are hidden issues. > > Some issues are easy to fix, so I sent some patches. > https://lore.kernel.org/linux-kbuild/20241106161445.189399-1-masahiroy@kernel.org/T/#t > https://lore.kernel.org/linux-kbuild/20241106161445.189399-1-masahiroy@kernel.org/T/#m4e4fa70386696e903b68d3fe1d7277e9a63fbefe > https://lore.kernel.org/linux-kbuild/20241107111519.GA15424@willie-the-truck/T/#mccf6d49ddd11c90dcc583d7a68934bb3311da880 I did notice the issues for .data.* -- that is one of the reasons we separated text from data in our patch. > > For example, see e41f501d3912. > > When CONFIG_LD_DEAD_CODE_DATA_ELIMINATION=y or > CONFIG_LTO_CLANG=y or CONFIG_AUTOFDO_CLANG=y or > CONFIG_PROPELLER_CLANG=y, the .text.startup sections > will go to TEXT_MAIN instead of INIT_TEXT. > This is not a fatal issue, but we cannot reuse memory for .text.startup > sections. > > Removing the #ifdef (i.e. reverting cb87481ee89d) is more difficult > because we need to take a closer look at potential impacts for all > architectures. I'm not sure if there is a naming convention for section names in the kernel. For special sections, we should avoid using .text.* or .data.*, instead, using "..', or use other prefixes. The compiler can generate sections names like .text.hot.*", ".text.unknown.*", ".text.unlikely.*", ".text.split.*", ".text.startup." or ".text.exit. It seems we've addressed most of them except .text.startup and .text.exit. For text.startup and .text.exit, have you considered renaming the sections within the linker script -- they are fixed strings and should be able to be renamed. > > I understood you did not want to take a risk to break random architectures, > so I decided to postpone the #ifdef issue and accept your patch set. Thanks for the understanding! > > -- > Best Regards > Masahiro Yamada
© 2016 - 2024 Red Hat, Inc.