| Message ID | 20230209152628.129914-1-ajones@ventanamicro.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | RISC-V: Apply Zicboz to clear_page | expand |
Hey Drew, On Thu, Feb 09, 2023 at 04:26:20PM +0100, Andrew Jones wrote: > When the Zicboz extension is available we can more rapidly zero naturally > aligned Zicboz block sized chunks of memory. As pages are always page > aligned and are larger than any Zicboz block size will be, then > clear_page() appears to be a good candidate for the extension. While cycle > count and energy consumption should also be considered, we can be pretty > certain that implementing clear_page() with the Zicboz extension is a win > by comparing the new dynamic instruction count with its current count[1]. > Doing so we see that the new count is just over a quarter of the old count > (see patch4's commit message for more details). > > For those of you who reviewed v1[2], you may be looking for the memset() > patches. As pointed out in v1, and a couple follow-up emails, it's not > clear that patching memset() is a win yet. When I get a chance to test > on real hardware with a comprehensive benchmark collection then I can > post the memset() patches separately (assuming the benchmarks show it's > worthwhile). > > Based on riscv-linux/for-next plus the dependencies listed below. > > Dependencies: > https://lore.kernel.org/all/20230108163356.3063839-1-conor@kernel.org/ > https://lore.kernel.org/all/20230105192610.1940841-1-heiko@sntech.de/ I've had a short (due to FOSDEM) & busy week since we discussed the automagic dependency collection. I'll try to get to it in the next few days. > The patches are also available here > https://github.com/jones-drew/linux/commits/riscv/zicboz-v4 > > To test over QEMU this branch may be used to enable Zicboz > https://gitlab.com/jones-drew/qemu/-/commits/riscv/zicboz > > To test running a KVM guest with Zicboz this kvmtool branch may be used > https://github.com/jones-drew/kvmtool/commits/riscv/zicboz > > Thanks, > drew > > [1] I ported the functions under test to userspace and linked them with > a test program. Then, I ran them under gdb with a script[3] which > counted instructions by single stepping. > [2] https://lore.kernel.org/all/20221027130247.31634-1-ajones@ventanamicro.com/ > [3] https://gist.github.com/jones-drew/487791c956ceca8c18adc2847eec9c60 > > v4: > - Rebased on latest for-next which allowed dropping one dependency > - Added "RISC-V: alternatives: Support patching multiple insns in assembly" > since I needed to use more than one instruction in an ALTERNATIVE call > from assembly. I can post this patch separately as a fix if desired. > - Improved the dt-binding patch commit message [Conor] > - Picked up some tags from Conor and Rob (I kept Conor's a-b on the > clear_page patch, even though there are several changes to it, because > I interpreted the a-b as "OK by me to implement a Zicboz clear_page") Yea, it was a "I am far from qualified to review your implementation, but I am okay with it existing and the remainder of the patch". Cheers, Conor.
When the Zicboz extension is available we can more rapidly zero naturally aligned Zicboz block sized chunks of memory. As pages are always page aligned and are larger than any Zicboz block size will be, then clear_page() appears to be a good candidate for the extension. While cycle count and energy consumption should also be considered, we can be pretty certain that implementing clear_page() with the Zicboz extension is a win by comparing the new dynamic instruction count with its current count[1]. Doing so we see that the new count is just over a quarter of the old count (see patch4's commit message for more details). For those of you who reviewed v1[2], you may be looking for the memset() patches. As pointed out in v1, and a couple follow-up emails, it's not clear that patching memset() is a win yet. When I get a chance to test on real hardware with a comprehensive benchmark collection then I can post the memset() patches separately (assuming the benchmarks show it's worthwhile). Based on riscv-linux/for-next plus the dependencies listed below. Dependencies: https://lore.kernel.org/all/20230108163356.3063839-1-conor@kernel.org/ https://lore.kernel.org/all/20230105192610.1940841-1-heiko@sntech.de/ The patches are also available here https://github.com/jones-drew/linux/commits/riscv/zicboz-v4 To test over QEMU this branch may be used to enable Zicboz https://gitlab.com/jones-drew/qemu/-/commits/riscv/zicboz To test running a KVM guest with Zicboz this kvmtool branch may be used https://github.com/jones-drew/kvmtool/commits/riscv/zicboz Thanks, drew [1] I ported the functions under test to userspace and linked them with a test program. Then, I ran them under gdb with a script[3] which counted instructions by single stepping. [2] https://lore.kernel.org/all/20221027130247.31634-1-ajones@ventanamicro.com/ [3] https://gist.github.com/jones-drew/487791c956ceca8c18adc2847eec9c60 v4: - Rebased on latest for-next which allowed dropping one dependency - Added "RISC-V: alternatives: Support patching multiple insns in assembly" since I needed to use more than one instruction in an ALTERNATIVE call from assembly. I can post this patch separately as a fix if desired. - Improved the dt-binding patch commit message [Conor] - Picked up some tags from Conor and Rob (I kept Conor's a-b on the clear_page patch, even though there are several changes to it, because I interpreted the a-b as "OK by me to implement a Zicboz clear_page") - Added "RISC-V: cpufeature: Put vendor_id to work", which was necessary for how I wanted to use ALTERNATIVE in clear_page. - Fallback to memset when the Zicboz block size is too big for the Zicboz implementation of clear_page [Palmer] - Use lw without la in clear_page impl [Palmer] - Implement a Duff device for clear_page using the new 'vendor_id' alternatives support [drew] v3: - CC'ed DT list - Improved commit message of DT bindings patch to point out relationship with cbom-block-size - Picked up an a-b from Conor v2: - s/blksz/block_size/, improved commit message for "RISC-V: Add Zicboz detection and block size parsing", isa ext sorting [Conor] - Added dt binding patch [Heiko] - Picked up r-b's from Conor, Heiko, and Anup - Moved config symbol and CBO_zero() introduction to "RISC-V: Use Zicboz in clear_page when available" and improved its commit message and implementation (unrolled four times) [drew] - Dropped memset() patches [drew] - Rebased on ae4d39f75308 ("Merge patch "RISC-V: fix incorrect type of ARCH_CANAAN_K210_DTB_SOURCE"") plus the dependencies Andrew Jones (8): RISC-V: alternatives: Support patching multiple insns in assembly RISC-V: Factor out body of riscv_init_cbom_blocksize loop dt-bindings: riscv: Document cboz-block-size RISC-V: Add Zicboz detection and block size parsing RISC-V: cpufeature: Put vendor_id to work RISC-V: Use Zicboz in clear_page when available RISC-V: KVM: Provide UAPI for Zicboz block size RISC-V: KVM: Expose Zicboz to the guest .../devicetree/bindings/riscv/cpus.yaml | 5 ++ arch/riscv/Kconfig | 13 ++++ arch/riscv/include/asm/alternative-macros.h | 6 +- arch/riscv/include/asm/cacheflush.h | 3 +- arch/riscv/include/asm/hwcap.h | 1 + arch/riscv/include/asm/insn-def.h | 4 ++ arch/riscv/include/asm/page.h | 6 +- arch/riscv/include/uapi/asm/kvm.h | 2 + arch/riscv/kernel/cpu.c | 1 + arch/riscv/kernel/cpufeature.c | 31 +++++++- arch/riscv/kernel/setup.c | 2 +- arch/riscv/kvm/vcpu.c | 11 +++ arch/riscv/lib/Makefile | 1 + arch/riscv/lib/clear_page.S | 71 +++++++++++++++++++ arch/riscv/mm/cacheflush.c | 64 ++++++++++------- 15 files changed, 187 insertions(+), 34 deletions(-) create mode 100644 arch/riscv/lib/clear_page.S