Re: [PATCH v3 0/9] perf/mem: AMD IBS and generic tools improvements
From: Arnaldo Carvalho de Melo
Date: Mon Apr 10 2023 - 11:15:56 EST
Em Mon, Apr 10, 2023 at 07:53:57AM +0530, Ravi Bangoria escreveu:
> On 08-Apr-23 3:14 AM, Namhyung Kim wrote:
> > Hi Ravi,
> >
> > On Fri, Apr 7, 2023 at 4:25 AM Ravi Bangoria <ravi.bangoria@xxxxxxx> wrote:
> >>
> >> Kernel IBS driver wasn't using new PERF_MEM_* APIs due to some of its
> >> limitations. Mainly:
> >>
> >> 1. mem_lvl_num doesn't allow setting multiple sources whereas old API
> >> allows it. Setting multiple data sources is useful because IBS on
> >> pre-zen4 uarch doesn't provide fine granular DataSrc details (there
> >> is only one such DataSrc(2h) though).
> >> 2. perf mem sorting logic (sort__lvl_cmp()) ignores mem_lvl_num. perf
> >> c2c (c2c_decode_stats()) does not use mem_lvl_num at all. perf mem
> >> prints mem_lvl and mem_lvl_num both if both are set, which is ugly.
> >>
> >> Set mem_lvl_num, mem_remote and mem_hops for data_src via IBS. Handle
> >> first issue using mem_lvl_num = ANY_CACHE | HOPS_0. In addition to
> >> setting new API fields, convert all individual field assignments to
> >> compile time wrapper macros built using PERF_MEM_S(). Also convert
> >> DataSrc conditional code to array lookups.
> >>
> >> Interpretation of perf_mem_data_src by perf_mem__lvl_scnprintf() was
> >> non-intuitive. Make it sane.
> >
> > Looks good, but I think you need to split kernel and user patches.
>
> Patch #1 to #3 are kernel changes. Patch #4 to #9 are userspace changes.
> Arnaldo, Peter, please let me know if you wants to split the series and
> resend.
I can always use b4's -P option :-) So no need to resubmit, I can pick
the tools bits,
- Arnaldo
> >
> >>
> >> v2: https://lore.kernel.org/r/20230327130851.1565-1-ravi.bangoria%40amd.com
> >> v2->v3:
> >> - IBS: Don't club RmtNode with DataSrc=7 (IO)
> >> - Make perf_mem__lvl_scnprintf() more sane
> >> - Introduce PERF_MEM_LVLNUM_UNC, set it along with PERF_MEM_LVL_UNC
> >> and interpreat it in tool.
> >> - Add PERF_MEM_LVLNUM_NA to default data_src value
> >> - Change some of the IBS bit description according to latest PPR
> >>
> >> Namhyung Kim (1):
> >> perf/x86/ibs: Set mem_lvl_num, mem_remote and mem_hops for data_src
> >>
> >> Ravi Bangoria (8):
> >> perf/mem: Introduce PERF_MEM_LVLNUM_UNC
> >> perf/mem: Add PERF_MEM_LVLNUM_NA to PERF_MEM_NA
> >> perf headers: Sync uapi/linux/perf_event.h
> >> perf mem: Add PERF_MEM_LVLNUM_NA to PERF_MEM_DATA_SRC_NONE
> >> perf mem: Add support for printing PERF_MEM_LVLNUM_UNC
> >> perf mem: Refactor perf_mem__lvl_scnprintf()
> >> perf mem: Increase HISTC_MEM_LVL column size to 39 chars
> >> perf script ibs: Change bit description according to latest PPR
> >
> > Acked-by: Namhyung Kim <namhyung@xxxxxxxxxx>
>
> Thanks!
--
- Arnaldo