lkml.org 
[lkml]   [2014]   [Oct]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 1/4] perf,x86: De-obfuscate HSW offcore bits
On Thu, Oct 23, 2014 at 12:51:20PM +0200, Peter Zijlstra wrote:
> Andi introduced the HSW cache events array, but used magic constants
> against convention as set by all the other uarchs. Try and deobfuscate
> these a bit.
>
> This patch should not change the actual values generated; however
> weird they seems.
>
> In that patch Andi also said there were differences between the
> SNB/IVB and HSW/BDW offcore tables but failed to specify which and
> why.
>
> Fixes: 86a349a28b24 ("perf/x86/intel: Add Broadwell core support")
> Cc: Andi Kleen <ak@linux.intel.com>
> Cc: Stephane Eranian <eranian@google.com>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> Link: http://lkml.kernel.org/n/tip-@git.kernel.org

This link looks funny.

> ---
> arch/x86/kernel/cpu/perf_event_intel.c | 48 +++++++++++++++++++++++++++------
> 1 file changed, 40 insertions(+), 8 deletions(-)
>
> --- a/arch/x86/kernel/cpu/perf_event_intel.c
> +++ b/arch/x86/kernel/cpu/perf_event_intel.c
> @@ -247,6 +247,7 @@ static u64 intel_pmu_event_map(int hw_ev
> #define SNB_BUS_LOCKS (1ULL << 10)
> #define SNB_STRM_ST (1ULL << 11)
> #define SNB_OTHER (1ULL << 15)
> +
> #define SNB_RESP_ANY (1ULL << 16)
> #define SNB_NO_SUPP (1ULL << 17)
> #define SNB_LLC_HITM (1ULL << 18)
> @@ -255,6 +256,7 @@ static u64 intel_pmu_event_map(int hw_ev
> #define SNB_LLC_HITF (1ULL << 21)
> #define SNB_LOCAL (1ULL << 22)
> #define SNB_REMOTE (0xffULL << 23)
> +
> #define SNB_SNP_NONE (1ULL << 31)
> #define SNB_SNP_NOT_NEEDED (1ULL << 32)
> #define SNB_SNP_MISS (1ULL << 33)
> @@ -519,6 +521,40 @@ static __initconst const u64 hsw_hw_cach
> },
> };
>
> +/* HSW Request type */
> +#define HSW_DMND_DATA_RD (1ULL << 0)
> +#define HSW_DMND_RFO (1ULL << 1)
> +#define HSW_DMND_IFETCH (1ULL << 2)
> +
> +#define HSW_PF_DATA_RD (1ULL << 4)
> +#define HSW_PF_RFO (1ULL << 5)
> +#define HSW_PF_IFETCH (1ULL << 6)
> +
> +#define HSW_OTHER (1ULL << 15)
> +
> +/* HSW Supplier info := SNB Supplier info */
> +/* HSW Snoop Info := SNB Snoop Info */
> +
> +/*
> + * SNB_LLC_* is specified as 'reserved' in the SDM
> + * but Andi added it anyhow, suggesting HSW/BDW have it just fine
> + *
> + * XXX NHM/WSM/SNB/IVB don't have the PF/IFETCH bits set
> + */
> +#define HSW_DMND_READ (HSW_DMND_DATA_RD|HSW_DMND_IFETCH| \
> + HSW_PF_DATA_RD|HSW_PF_IFETCH| \
> + SNB_LLC_DATA_RD|SNB_LLC_IFETCH)
> +
> +/* XXX NHM/WSM/SNB/IVB dont have the PF bits set */
> +#define HSW_DMND_WRITE (HSW_DMND_RFO|HSW_PF_RFO|SNB_LLC_RFO)
> +
> +#define HSW_DMND_PREFETCH (0) /* XXX broken ? */
> +
> +#define HSW_DRAM_ANY (SNB_NO_SUPP|SNB_SNP_ANY|(0x78ULL << 23)) /* WTF */
> +
> +#define HSW_L3_ACCESS (0) /* XXX no supplier! */
> +#define NSW_L3_MISS (HSW_DRAM_ANY|SNB_NON_DRAM)

You probably mean HSW and not NSW here...

--
Regards/Gruss,
Boris.

Sent from a fat crate under my desk. Formatting is fine.
--


\
 
 \ /
  Last update: 2014-10-23 15:01    [W:0.083 / U:0.064 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site