| Message ID | 20231012141456.4078-1-jszhang@kernel.org (mailing list archive) |
|---|---|
| Headers | show |
| Series | riscv: errata: thead: use riscv_nonstd_cache_ops for CMO | expand |
On Thu, Oct 12, 2023 at 10:14:54PM +0800, Jisheng Zhang wrote: > Previously, we use alternative mechanism to dynamically patch > the CMO operations for THEAD C906/C910 during boot for performance > reason. But as pointed out by Arnd, "there is already a significant > cost in accessing the invalidated cache lines afterwards, which is > likely going to be much higher than the cost of an indirect branch". > And indeed, there's no performance difference with GMAC and EMMC per > my test on Sipeed Lichee Pi 4A board. > > Use riscv_nonstd_cache_ops for THEAD C906/C910 CMO to simplify > the alternative code, and to acchieve Arnd's goal -- "I think > moving the THEAD ops at the same level as all nonstandard operations > makes sense, but I'd still leave CMO as an explicit fast path that > avoids the indirect branch. This seems like the right thing to do both > for readability and for platforms on which the indirect branch has a > noticeable overhead." > > To make bisect easy, I use two patches here: patch1 does the conversion > which just mimics current CMO behavior via. riscv_nonstd_cache_ops, I > assume no functionalities changes. patch2 uses T-HEAD PA based CMO > instructions so that we don't need to covert PA to VA. > > Hi Guo, > > I didn't use wback_inv for wback as you suggested during v1 reviewing, > this can be left as future optimizations. > > Thanks > > since v2: > - collect Reviewed-by tag Oh, I missed the tag collection, but I know maintainers are using b4 which can collect and apply tags automatically ;). let me know if want a new version. > - fix typo > > since v1: > - collect Tested-by tag > - add patch2 to use T-HEAD PA based CMO instructions. > > Jisheng Zhang (2): > riscv: errata: thead: use riscv_nonstd_cache_ops for CMO > riscv: errata: thead: use pa based instructions for CMO > > arch/riscv/Kconfig.errata | 1 + > arch/riscv/errata/thead/errata.c | 69 +++++++++++++++++++++++++++- > arch/riscv/include/asm/errata_list.h | 50 +++----------------- > 3 files changed, 74 insertions(+), 46 deletions(-) > > -- > 2.40.1 > > > _______________________________________________ > linux-riscv mailing list > linux-riscv@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-riscv
On Thu, Oct 12, 2023 at 10:21:08PM +0800, Jisheng Zhang wrote: > On Thu, Oct 12, 2023 at 10:14:54PM +0800, Jisheng Zhang wrote: > > Previously, we use alternative mechanism to dynamically patch > > the CMO operations for THEAD C906/C910 during boot for performance > > reason. But as pointed out by Arnd, "there is already a significant > > cost in accessing the invalidated cache lines afterwards, which is > > likely going to be much higher than the cost of an indirect branch". > > And indeed, there's no performance difference with GMAC and EMMC per > > my test on Sipeed Lichee Pi 4A board. > > > > Use riscv_nonstd_cache_ops for THEAD C906/C910 CMO to simplify > > the alternative code, and to acchieve Arnd's goal -- "I think > > moving the THEAD ops at the same level as all nonstandard operations > > makes sense, but I'd still leave CMO as an explicit fast path that > > avoids the indirect branch. This seems like the right thing to do both > > for readability and for platforms on which the indirect branch has a > > noticeable overhead." > > > > To make bisect easy, I use two patches here: patch1 does the conversion > > which just mimics current CMO behavior via. riscv_nonstd_cache_ops, I > > assume no functionalities changes. patch2 uses T-HEAD PA based CMO > > instructions so that we don't need to covert PA to VA. > > > > Hi Guo, > > > > I didn't use wback_inv for wback as you suggested during v1 reviewing, > > this can be left as future optimizations. > > > > Thanks > > > > since v2: > > - collect Reviewed-by tag > > Oh, I missed the tag collection, but I know maintainers are using b4 which can > collect and apply tags automatically ;). let me know if want a new > version. It doesn't collect tags (AFAIU) from earlier revisions though.
On Thu, Oct 12, 2023 at 03:36:28PM +0100, Conor Dooley wrote: > On Thu, Oct 12, 2023 at 10:21:08PM +0800, Jisheng Zhang wrote: > > On Thu, Oct 12, 2023 at 10:14:54PM +0800, Jisheng Zhang wrote: > > > Previously, we use alternative mechanism to dynamically patch > > > the CMO operations for THEAD C906/C910 during boot for performance > > > reason. But as pointed out by Arnd, "there is already a significant > > > cost in accessing the invalidated cache lines afterwards, which is > > > likely going to be much higher than the cost of an indirect branch". > > > And indeed, there's no performance difference with GMAC and EMMC per > > > my test on Sipeed Lichee Pi 4A board. > > > > > > Use riscv_nonstd_cache_ops for THEAD C906/C910 CMO to simplify > > > the alternative code, and to acchieve Arnd's goal -- "I think > > > moving the THEAD ops at the same level as all nonstandard operations > > > makes sense, but I'd still leave CMO as an explicit fast path that > > > avoids the indirect branch. This seems like the right thing to do both > > > for readability and for platforms on which the indirect branch has a > > > noticeable overhead." > > > > > > To make bisect easy, I use two patches here: patch1 does the conversion > > > which just mimics current CMO behavior via. riscv_nonstd_cache_ops, I > > > assume no functionalities changes. patch2 uses T-HEAD PA based CMO > > > instructions so that we don't need to covert PA to VA. > > > > > > Hi Guo, > > > > > > I didn't use wback_inv for wback as you suggested during v1 reviewing, > > > this can be left as future optimizations. > > > > > > Thanks > > > > > > since v2: > > > - collect Reviewed-by tag > > > > Oh, I missed the tag collection, but I know maintainers are using b4 which can > > collect and apply tags automatically ;). let me know if want a new > > version. > > It doesn't collect tags (AFAIU) from earlier revisions though. oops I didn't know this before, just sent out v4 with real tag collection to make the merging progress smooth.