MIME-Version: 1.0
In-Reply-To: <1335983192-23731-5-git-send-email-robert.richter@amd.com>
References: <1335983192-23731-1-git-send-email-robert.richter@amd.com>
	<1335983192-23731-5-git-send-email-robert.richter@amd.com>
Date: Mon, 7 May 2012 17:21:51 +0200
Message-ID: <CABPqkBQQcENVAdZZ3PzEbwNNRFh+r9yax10s37cAe7uuOwBPOw@mail.gmail.com>
Subject: Re: [PATCH 4/7] perf/x86-ibs: Add support for IBS pseudo events
From: Stephane Eranian <eranian@google.com>
To: Robert Richter <robert.richter@amd.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>, Ingo Molnar <mingo@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>, Jiri Olsa <jolsa@redhat.com>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset=UTF-8
Sender: linux-kernel-owner@vger.kernel.org
Content-Transfer-Encoding: 8bit
Content-Length: 35521
Lines: 631

Robert,

There is something I don't quite understand with those pseudo-events.
Is it the case that by construction, it means you can only measure
on pseudo-event at a time? Supposed I want to look at cache-misses.
For each miss, I'd like to know where it missed, any TLB impacts. All
of that in one run with no multiplexing. Can I do this with your pseudo-events?


On Wed, May 2, 2012 at 8:26 PM, Robert Richter <robert.richter@amd.com> wrote:
> This patch implements support for IBS pseudo events. Pseudo events are
> derived from an IBS sample and determined through a combination of one
> or more IBS event flags or values. See here for a full description:
>
>  Software Optimization Guide for AMD Family 15h Processors
>  Appendix F Guide to Instruction-Based Sampling on AMD Family 15h Processors
>  Advanced Micro Devices, Inc.
>  Publication No. 47414, Revision 3.06
>  January 2012
>  http://support.amd.com/us/Processor_TechDocs/47414_15h_sw_opt_guide.pdf
>
> The list of supported events is provided by perf-list. A pseudo event
> can be set up like this:
>
>  # perf record -a -e ibs_op:MISPREDICTED_BRANCH ...
>
> The filter rules for IBS samples depending on a pseudo event are also
> described in the document above. The filter is setup in the perf tool
> pmu handler and passed to the kernel via config1/config2 attr values.
> The interface is extendable to pass the pseudo events directly to the
> kernel.
>
> There are some pseudo events capable to count latencies or other
> values. Counting values of such events is not yet supported.
>
> This patch includes kernel and userland changes.
>
> Signed-off-by: Robert Richter <robert.richter@amd.com>
> ---
>  arch/x86/kernel/cpu/perf_event_amd_ibs.c |   83 ++++++-
>  tools/perf/util/pmu-ibs.c                |  433 +++++++++++++++++++++++++-----
>  2 files changed, 445 insertions(+), 71 deletions(-)
>
> diff --git a/arch/x86/kernel/cpu/perf_event_amd_ibs.c b/arch/x86/kernel/cpu/perf_event_amd_ibs.c
> index 03743ad..1675479 100644
> --- a/arch/x86/kernel/cpu/perf_event_amd_ibs.c
> +++ b/arch/x86/kernel/cpu/perf_event_amd_ibs.c
> @@ -478,6 +478,81 @@ static struct perf_ibs perf_ibs_op = {
>        .get_count              = get_ibs_op_count,
>  };
>
> +enum ibs_filter_type {
> +       IBS_NO_FILTER           = 0,
> +       IBS_MATCH_FILTER        = 1,
> +       IBS_ANY_SET_FILTER      = 2,
> +       IBS_PSEUDO_EVENT        = 0x0F,
> +};
> +
> +struct ibs_filter {
> +       struct {
> +               u16             idx             : 8;
> +               u16             reserved        : 4;
> +               u16             type            : 4;
> +       };
> +       union {
> +               struct {
> +                       u8      mask;
> +                       u8      match;
> +               };
> +               u16             any;
> +       };
> +};
> +
> +static bool
> +__perf_ibs_sample_matches(struct ibs_filter *filter, void *data, int size)
> +{
> +       int left = size;
> +
> +       switch (filter->type) {
> +       case IBS_MATCH_FILTER:
> +               left -= sizeof(u8);
> +               break;
> +       case IBS_ANY_SET_FILTER:
> +               left -= sizeof(u16);
> +               break;
> +       default:
> +               return false;
> +       }
> +
> +       left -= filter->idx;
> +       if (left < 0)
> +               return false;
> +
> +       switch (filter->type) {
> +       case IBS_MATCH_FILTER:
> +               return ((*(u8*)(data + filter->idx)) & filter->mask) == filter->match;
> +       case IBS_ANY_SET_FILTER:
> +               return (*(u16*)(data + filter->idx)) & filter->any;
> +       };
> +
> +       return false;
> +}
> +
> +static bool perf_ibs_sample_matches(struct perf_event *event,
> +                                   struct perf_ibs_data *data)
> +{
> +       int i;
> +       union {
> +               struct ibs_filter filter[4];
> +               u64     config[2];
> +       } f;
> +       struct ibs_filter *filter = f.filter;
> +
> +       f.config[0] = event->attr.config1;
> +       f.config[1] = event->attr.config2;
> +
> +       for (i = 0; i < 4; i++, filter++) {
> +               if (filter->type == IBS_NO_FILTER)
> +                       break;
> +               if (!__perf_ibs_sample_matches(filter, data->regs, data->size))
> +                       return false;
> +       }
> +
> +       return true;
> +}
> +
>  static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
>  {
>        struct cpu_perf_ibs *pcpu = this_cpu_ptr(perf_ibs->pcpu);
> @@ -487,7 +562,7 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
>        struct perf_raw_record raw;
>        struct pt_regs regs;
>        struct perf_ibs_data ibs_data;
> -       int offset, size, check_rip, offset_max, throttle = 0;
> +       int offset, size, check_rip, filter, offset_max, throttle = 0;
>        unsigned int msr;
>        u64 *buf, *config, period;
>
> @@ -517,7 +592,8 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
>        size = 1;
>        offset = 1;
>        check_rip = (perf_ibs == &perf_ibs_op && (ibs_caps & IBS_CAPS_RIPINVALIDCHK));
> -       if (event->attr.sample_type & PERF_SAMPLE_RAW)
> +       filter = (event->attr.config1 != 0);
> +       if (filter || (event->attr.sample_type & PERF_SAMPLE_RAW))
>                offset_max = perf_ibs->offset_max;
>        else if (check_rip)
>                offset_max = 2;
> @@ -532,6 +608,9 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
>        } while (offset < offset_max);
>        ibs_data.size = sizeof(u64) * size;
>
> +       if (filter && !perf_ibs_sample_matches(event, &ibs_data))
> +               goto out;
> +
>        regs = *iregs;
>        if (check_rip && (ibs_data.regs[2] & IBS_RIP_INVALID)) {
>                regs.flags &= ~PERF_EFLAGS_EXACT;
> diff --git a/tools/perf/util/pmu-ibs.c b/tools/perf/util/pmu-ibs.c
> index 07acb82..604cb8c 100644
> --- a/tools/perf/util/pmu-ibs.c
> +++ b/tools/perf/util/pmu-ibs.c
> @@ -12,84 +12,378 @@
>  #include <linux/compiler.h>
>  #include "pmu.h"
>
> -static const char *events[] = {
> -       "ibs_fetch:2M_PAGE",
> -       "ibs_fetch:4K_PAGE",
> -       "ibs_fetch:ABORTED",
> -       "ibs_fetch:ALL",
> -       "ibs_fetch:ATTEMPTED",
> -       "ibs_fetch:COMPLETED",
> -       "ibs_fetch:ICACHE_HITS",
> -       "ibs_fetch:ICACHE_MISSES",
> -       "ibs_fetch:ITLB_HITS",
> -       "ibs_fetch:KILLED",
> -       "ibs_fetch:L1_ITLB_MISSES_L2_ITLB_HITS",
> -       "ibs_fetch:L1_ITLB_MISSES_L2_ITLB_MISSES",
> -       "ibs_fetch:LATENCY",
> -       "ibs_op:ALL",
> -       "ibs_op:ALL_LOAD_STORE",
> -       "ibs_op:BANK_CONF_LOAD",
> -       "ibs_op:BANK_CONF_STORE",
> -       "ibs_op:BRANCH_RETIRED",
> -       "ibs_op:CANCELLED",
> -       "ibs_op:COMP_TO_RET",
> -       "ibs_op:DATA_CACHE_MISS",
> -       "ibs_op:DATA_HITS",
> -       "ibs_op:DC_LOAD_LAT",
> -       "ibs_op:DCUC_MEM_ACC",
> -       "ibs_op:DCWC_MEM_ACC",
> -       "ibs_op:FORWARD",
> -       "ibs_op:L1_DTLB_1G",
> -       "ibs_op:L1_DTLB_2M",
> -       "ibs_op:L1_DTLB_4K",
> -       "ibs_op:L1_DTLB_HITS",
> -       "ibs_op:L1_DTLB_MISS_L2_DTLB_HIT",
> -       "ibs_op:L1_L2_DTLB_MISS",
> -       "ibs_op:L2_DTLB_1G",
> -       "ibs_op:L2_DTLB_2M",
> -       "ibs_op:L2_DTLB_4K",
> -       "ibs_op:LOAD",
> -       "ibs_op:LOCKED",
> -       "ibs_op:MAB_HIT",
> -       "ibs_op:MISALIGNED_DATA_ACC",
> -       "ibs_op:MISPREDICTED_BRANCH",
> -       "ibs_op:MISPREDICTED_BRANCH_TAKEN",
> -       "ibs_op:MISPREDICTED_RETURNS",
> -       "ibs_op:NB_CACHE_MODIFIED",
> -       "ibs_op:NB_CACHE_OWNED",
> -       "ibs_op:NB_LOCAL_CACHE",
> -       "ibs_op:NB_LOCAL_CACHE_LAT",
> -       "ibs_op:NB_LOCAL_DRAM",
> -       "ibs_op:NB_LOCAL_L3",
> -       "ibs_op:NB_LOCAL_ONLY",
> -       "ibs_op:NB_LOCAL_OTHER",
> -       "ibs_op:NB_REMOTE_CACHE",
> -       "ibs_op:NB_REMOTE_CACHE_LAT",
> -       "ibs_op:NB_REMOTE_DRAM",
> -       "ibs_op:NB_REMOTE_ONLY",
> -       "ibs_op:NB_REMOTE_OTHER",
> -       "ibs_op:RESYNC",
> -       "ibs_op:RETURNS",
> -       "ibs_op:STORE",
> -       "ibs_op:TAG_TO_RETIRE",
> -       "ibs_op:TAKEN_BRANCH",
> -       NULL
> +enum ibs_filter_type {
> +       IBS_NO_FILTER           = 0,
> +       IBS_MATCH_FILTER        = 1,
> +       IBS_ANY_SET_FILTER      = 2,
> +       IBS_PSEUDO_EVENT        = 0x0F,
> +};
> +
> +struct ibs_filter {
> +       struct {
> +               __u16           idx             : 8;
> +               __u16           reserved        : 4;
> +               __u16           type            : 4;
> +       };
> +       union {
> +               struct {
> +                       __u8    mask;
> +                       __u8    match;
> +               };
> +               __u16           any;
> +       };
> +};
> +
> +struct ibs_event {
> +       __u16                   id;
> +       const char              *name;
> +       const char              *desc;
> +       union {
> +               __u16           pseudo_event;
> +               __u64           config;
> +               struct ibs_filter filter[2];
> +       };
> +};
> +
> +#define IBS_FETCH_CTL          0
> +#define IBS_OP_DATA            2
> +#define IBS_OP_DATA2           3
> +#define IBS_OP_DATA3           4
> +
> +#define IBS_IDX(reg, bit)      ((reg)<<3)+((bit)>>3)
> +#define IBS_MASK(bit, m)       (0xFF&m)
> +#define IBS_MASK16(bit, m)     (0xFFFF&m)
> +#define IBS_FILTER_MATCH_ANY() { { 0, 0, 0 }, { .any = 0 } }
> +
> +#define IBS_FILTER_ANY_SET(reg, bit, m)                                \
> +       {                                                       \
> +               {                                               \
> +               .type   = IBS_ANY_SET_FILTER,                   \
> +               .idx    = IBS_IDX(reg, bit),                    \
> +               .reserved = 0,                                  \
> +               },{                                             \
> +               .any    = IBS_MASK16(bit, m),                   \
> +               }                                               \
> +       },
> +
> +#define IBS_FILTER_ALL_CLEAR(reg, bit, m)                      \
> +       {                                                       \
> +               {                                               \
> +               .type   = IBS_MATCH_FILTER,                     \
> +               .idx    = IBS_IDX(reg, bit),                    \
> +               .reserved = 0,                                  \
> +               },{{                                            \
> +               .mask   = IBS_MASK(bit, m),                     \
> +               .match = 0,                                     \
> +               }}                                              \
> +       },                                                      \
> +       {                                                       \
> +               {                                               \
> +               .type   = IBS_MATCH_FILTER,                     \
> +               .idx    = IBS_IDX(reg, (bit) + 8),              \
> +               .reserved = 0,                                  \
> +               },{{                                            \
> +               .mask   = IBS_MASK(bit, (m) >> 8),              \
> +               .match  = 0,                                    \
> +               }}                                              \
> +       },
> +
> +#define IBS_FILTER_ALL_SET(reg, bit, m)                                \
> +       {                                                       \
> +               {                                               \
> +               .type   = IBS_MATCH_FILTER,                     \
> +               .idx    = IBS_IDX(reg, bit),                    \
> +               .reserved = 0,                                  \
> +               },{{                                            \
> +               .mask   = IBS_MASK(bit, m),                     \
> +               .match  = IBS_MASK(bit, m),                     \
> +               }}                                              \
> +       },
> +
> +#define IBS_FILTER_MATCH(reg, bit, m, v)                       \
> +       {                                                       \
> +               {                                               \
> +               .type   = IBS_MATCH_FILTER,                     \
> +               .idx    = IBS_IDX(reg, bit),                    \
> +               .reserved = 0,                                  \
> +               },{{                                            \
> +               .mask   = IBS_MASK(bit, m),                     \
> +               .match  = IBS_MASK(bit, v),                     \
> +               }}                                              \
> +       },
> +
> +#define IBS_FILTER_MATCH2(reg, reg2, bit, bit2, m, m2, v, v2)  \
> +       {                                                       \
> +               {                                               \
> +               .type   = IBS_MATCH_FILTER,                     \
> +               .idx    = IBS_IDX(reg, bit),                    \
> +               .reserved = 0,                                  \
> +               },{{                                            \
> +               .mask   = IBS_MASK(bit, m),                     \
> +               .match  = IBS_MASK(bit, v),                     \
> +               }}                                              \
> +       },                                                      \
> +       {                                                       \
> +               {                                               \
> +               .type   = IBS_MATCH_FILTER,                     \
> +               .idx    = IBS_IDX(reg2, bit2),                  \
> +               .reserved = 0,                                  \
> +               },{{                                            \
> +               .mask   = IBS_MASK(bit2, m2),                   \
> +               .match  = IBS_MASK(bit2, v2),                   \
> +               }}                                              \
> +       }
> +
> +#define IBS_EVENT(i, n, d)                                     \
> +       {                                                       \
> +               .id     = (i),                                  \
> +               .name   = (n),                                  \
> +               .desc   = (d),                                  \
> +               { .filter = { IBS_FILTER_##i } },               \
> +       }
> +#define IBS_FILTER(type, args...)              IBS_FILTER_##type(args)
> +
> +/*
> + * ID   Name                             Derivation
> + *
> + * F000 IBS fetch samples                Number of all IBS fetch samples
> + * F001 IBS fetch killed                 Number of killed IBS fetch samples
> + * F002 IBS fetch attempted              Number of non-killed IBS fetch samples
> + * F003 IBS fetch completed              IbsFetchComp
> + * F004 IBS fetch aborted                ~IbsFetchComp
> + * F005 IBS L1 ITLB hit                  ~IbsL1TlbMiss & IbsPhyAddrValid
> + * F006 IBS L1 ITLB miss, L2 ITLB hit    IbsL1TlbMiss & ~IbsL2TlbMiss
> + * F007 IBS L1 ITLB miss, L2 ITLB miss   IbsL1TlbMiss & IbsL2TlbMiss
> + * F008 IBS instruction cache miss       IbsIcMiss
> + * F009 IBS instruction cache hit        IbsFetchComp & ~IbsIcMiss
> + * F00A IBS 4K page translation          IbsL1TlbPgSz=0 & IbsPhyAddrValid
> + * F00B IBS 2M page translation          IbsL1TlbPgSz=1 & IbsPhyAddrValid
> + * F00C IBS 1G page translation          IbsL1TlbPgSz=2 & IbsPhyAddrValid
> + * F00D Reserved
> + * F00E IBS fetch latency                IbsfetchLat
> + */
> +#define IBS_FILTER_0xf000 IBS_FILTER(MATCH_ANY)
> +#define IBS_FILTER_0xf001 IBS_FILTER(ALL_CLEAR,        IBS_FETCH_CTL,  48, 0x019c)
> +#define IBS_FILTER_0xf002 IBS_FILTER(ANY_SET,  IBS_FETCH_CTL,  48, 0x019c)
> +#define IBS_FILTER_0xf003 IBS_FILTER(MATCH,    IBS_FETCH_CTL,  48, 0x04, 0x04)
> +#define IBS_FILTER_0xf004 IBS_FILTER(MATCH,    IBS_FETCH_CTL,  48, 0x04, 0x00)
> +#define IBS_FILTER_0xf005 IBS_FILTER(MATCH,    IBS_FETCH_CTL,  48, 0x90, 0x10)
> +#define IBS_FILTER_0xf006 IBS_FILTER(MATCH2,   IBS_FETCH_CTL, IBS_FETCH_CTL, 56, 48, 0x01, 0x80, 0x00, 0x80)
> +#define IBS_FILTER_0xf007 IBS_FILTER(MATCH2,   IBS_FETCH_CTL, IBS_FETCH_CTL, 56, 48, 0x01, 0x80, 0x01, 0x80)
> +#define IBS_FILTER_0xf008 IBS_FILTER(MATCH,    IBS_FETCH_CTL,  48, 0x08, 0x08)
> +#define IBS_FILTER_0xf009 IBS_FILTER(MATCH,    IBS_FETCH_CTL,  48, 0x0C, 0x04)
> +#define IBS_FILTER_0xf00a IBS_FILTER(MATCH,    IBS_FETCH_CTL,  48, 0x70, 0x10)
> +#define IBS_FILTER_0xf00b IBS_FILTER(MATCH,    IBS_FETCH_CTL,  48, 0x70, 0x30)
> +#define IBS_FILTER_0xf00c IBS_FILTER(MATCH,    IBS_FETCH_CTL,  48, 0x70, 0x60)
> +#if 0
> +#define IBS_FILTER_0xf00e IBS_FILTER(COUNT,    IBS_FETCH_CTL,  32, 0xffff)
> +#endif
> +
> +/*
> + * ID   Name                             Derivation
> + *
> + * F100 IBS all op samples               Number of all IBS op samples
> + * F101 IBS tag to retire cycles         Sum of all tag to retire cycles
> + * F102 ibs completion to retire cycles  Sum of all completion to retire cycles
> + * F103 IBS branch op                    IbsOpBrnRet
> + * F104 IBS mispredicted branch op       IbsOpBrnRet & IbsOpBrnMisp
> + * F105 IBS taken branch op              IbsOpBrnRet & IbsOpBrnTaken
> + * F106 IBS mispredicted taken branch op IbsOpBrnRet & IbsOpBrnTaken & IbsOpBrnMisp
> + * F107 IBS return op                    IbsOpReturn
> + * F108 IBS mispredicted return op       IbsOpReturn & IbsOpMispReturn
> + * F109 IBS resync op                    IbsOpBrnResync
> + */
> +#define IBS_FILTER_0xf100 IBS_FILTER(MATCH_ANY)
> +#if 0
> +#define IBS_FILTER_0xf101 IBS_FILTER(COUNT,    IBS_OP_DATA,    16, 0xffff)
> +#define IBS_FILTER_0xf102 IBS_FILTER(COUNT,    IBS_OP_DATA,     0, 0xffff)
> +#endif
> +#define IBS_FILTER_0xf103 IBS_FILTER(MATCH,    IBS_OP_DATA,    32, 0x20, 0x20)
> +#define IBS_FILTER_0xf104 IBS_FILTER(MATCH,    IBS_OP_DATA,    32, 0x30, 0x30)
> +#define IBS_FILTER_0xf105 IBS_FILTER(MATCH,    IBS_OP_DATA,    32, 0x28, 0x28)
> +#define IBS_FILTER_0xf106 IBS_FILTER(MATCH,    IBS_OP_DATA,    32, 0x38, 0x38)
> +#define IBS_FILTER_0xf107 IBS_FILTER(MATCH,    IBS_OP_DATA,    32, 0x04, 0x04)
> +#define IBS_FILTER_0xf108 IBS_FILTER(MATCH,    IBS_OP_DATA,    32, 0x06, 0x06)
> +#define IBS_FILTER_0xf109 IBS_FILTER(MATCH,    IBS_OP_DATA,    32, 0x01, 0x01)
> +
> +/*
> + * ID   Name                             Derivation
> + *
> + * F200 IBS All Load/Store Ops           IbsLdOp | IbsStOp
> + * F201 IBS Load Ops                     IbsLdOp
> + * F202 IBS Store Ops                    IbsStOp
> + * F203 IBS L1 DTLB Hit                  ~IbsDcL1tlbMiss & IbsDcLinAddrValid
> + * F204 IBS L1 DTLB Miss L2 DTLB Hit     IbsDcL1tlbMiss & ~IbsDcL2tlbMiss
> + * F205 IBS L1 DTLB Miss L2 DTLB Miss    IbsDcL1tlbMiss & IbsDcL2tlbMiss
> + * F206 IBS DC Miss                      IbsDcMiss
> + * F207 IBS DC Hit                       ~IbsDcMiss
> + * F208 IBS Misaligned Access            IbsDcMisAcc
> + * F209 IBS Bank Conflict On Load Op     IbsDcLdBnkCon
> + * F20A Reserved
> + * F20B IBS Store to Load Forwarded      IbsDcStToLdFwd
> + * F20C IBSStore to Load Forwarding Cancelled IbsDcStToLdCan
> + * F20D IBS UC memory access             IbsDcUcMemAcc
> + * F20E IBS WC memory access             IbsDcWcMemAcc
> + * F20F IBS locked operation             IbsDcLockedOp
> + * F210 IBS MAB hit                      IbsDcMabHit
> + * F211 IBS L1 DTLB 4K page              ~IbsDcL1tlbHit2M & ~IbsDcL1tlbHit1G &
> + *                                       IbsDcLinAddrValid
> + * F212 IBS L1 DTLB 2M page              IbsDcL1tlbHit2M & IbsDcLinAddrValid
> + * F213 IBS L1 DTLB 1G page              IbsDcL1tlbHit1G & IbsDcLinAddrValid
> + * F214 Reserved
> + * F215 IBS L2 DTLB 4K page              ~IbsDcL2tlbMiss & IbsDcL1tlbMiss &
> + *                                       ~IbsDcL1tlbHit2M & lbsDcLinAddrValid
> + * F216 IBS L2 DTLB 2M page              ~IbsDcL2tlbMiss & IbsDcL1tlbMiss &
> + *                                       IbsDcL1tlbHit2M & lbsDcLinAddrValid
> + * F217 Reserved
> + * F218 Reserved
> + * F219 IBS DC miss load latency         IbsDcMissLat when IbsLdOp & IbsDcMiss
> + */
> +#define IBS_FILTER_0xf200 IBS_FILTER(ANY_SET,  IBS_OP_DATA3,    0, 0x0003)
> +#define IBS_FILTER_0xf201 IBS_FILTER(MATCH,    IBS_OP_DATA3,    0, 0x01, 0x01)
> +#define IBS_FILTER_0xf202 IBS_FILTER(MATCH,    IBS_OP_DATA3,    0, 0x02, 0x02)
> +#define IBS_FILTER_0xf203 IBS_FILTER(MATCH2,   IBS_OP_DATA3, IBS_OP_DATA3, 16, 0, 0x02, 0x04, 0x02, 0x00)
> +#define IBS_FILTER_0xf204 IBS_FILTER(MATCH,    IBS_OP_DATA3,    0, 0x0C, 0x08)
> +#define IBS_FILTER_0xf205 IBS_FILTER(MATCH,    IBS_OP_DATA3,    0, 0x0C, 0x0C)
> +#define IBS_FILTER_0xf206 IBS_FILTER(MATCH,    IBS_OP_DATA3,    0, 0x80, 0x80)
> +#define IBS_FILTER_0xf207 IBS_FILTER(MATCH,    IBS_OP_DATA3,    0, 0x80, 0x00) IBS_FILTER(ANY_SET,   IBS_OP_DATA3,    0, 0x0003)
> +#define IBS_FILTER_0xf208 IBS_FILTER(MATCH,    IBS_OP_DATA3,    8, 0x01, 0x01)
> +#define IBS_FILTER_0xf209 IBS_FILTER(MATCH,    IBS_OP_DATA3,    8, 0x02, 0x02)
> +#define IBS_FILTER_0xf20b IBS_FILTER(MATCH,    IBS_OP_DATA3,    8, 0x08, 0x08)
> +#define IBS_FILTER_0xf20c IBS_FILTER(MATCH,    IBS_OP_DATA3,    8, 0x10, 0x10)
> +#define IBS_FILTER_0xf20d IBS_FILTER(MATCH,    IBS_OP_DATA3,    8, 0x40, 0x40)
> +#define IBS_FILTER_0xf20e IBS_FILTER(MATCH,    IBS_OP_DATA3,    8, 0x20, 0x20)
> +#define IBS_FILTER_0xf20f IBS_FILTER(MATCH,    IBS_OP_DATA3,    8, 0x80, 0x80)
> +#define IBS_FILTER_0xf210 IBS_FILTER(MATCH,    IBS_OP_DATA3,   16, 0x01, 0x01)
> +#define IBS_FILTER_0xf211 IBS_FILTER(MATCH2,   IBS_OP_DATA3, IBS_OP_DATA3, 16, 0, 0x02, 0x30, 0x02, 0x00)
> +#define IBS_FILTER_0xf212 IBS_FILTER(MATCH2,   IBS_OP_DATA3, IBS_OP_DATA3, 16, 0, 0x02, 0x10, 0x02, 0x10)
> +#define IBS_FILTER_0xf213 IBS_FILTER(MATCH2,   IBS_OP_DATA3, IBS_OP_DATA3, 16, 0, 0x02, 0x20, 0x02, 0x20)
> +#define IBS_FILTER_0xf215 IBS_FILTER(MATCH2,   IBS_OP_DATA3, IBS_OP_DATA3, 16, 0, 0x02, 0x1C, 0x02, 0x04)
> +#define IBS_FILTER_0xf216 IBS_FILTER(MATCH2,   IBS_OP_DATA3, IBS_OP_DATA3, 16, 0, 0x02, 0x1C, 0x02, 0x14)
> +#if 0
> +#define IBS_FILTER_0xf219 IBS_FILTER(COUNT,    IBS_OP_DATA3,   32, 0x00FFFF, 0x00FFFF)
> +#endif
> +
> +/*
> + * ID   Name                             Derivation
> + *
> + * F240 IBS NB local                     ~NbIbsReqDstProc
> + * F241 IBS NB remote                    NbIbsReqDstProc
> + * F242 IBS NB local L3                  NbIbsReqSrc=0x1 & ~NbIbsReqDstProc
> + * F243 IBS NB local L1/L2 (intercore)   NbIbsReqSrc=0x2 & ~NbIbsReqDstProc
> + * F244 IBS NB remote L1/L2/L3 cache     NbIbsReqSrc=0x2 & NbIbsReqDstProc
> + * F245 IBS NB local DRAM                NbIbsReqSrc=0x3 & ~NbIbsReqDstProc
> + * F246 IBS NB remote DRAM               NbIbsReqSrc=0x3 & NbIbsReqDstProc
> + * F247 IBS NB local other               NbIbsReqSrc=0x7 & ~NbIbsReqDstProc
> + * F248 IBS NB remote other              NbIbsReqSrc=0x7 & NbIbsReqDstProc
> + * F249 IBS NB cache M state             NbIbsReqSrc=0x2 & ~NbIbsReqCacheHitSt
> + * F24A IBS NB cache O state             NbIbsReqSrc=0x2 & NbIbsReqCacheHitSt
> + * F24B IBS NB local latency             IbsDcMissLat when ~NbIbsReqDstProc
> + * F24C IBS NB remote latency            IbsDcMissLat when NbIbsReqDstProc
> + */
> +#define IBS_FILTER_0xf240 IBS_FILTER(MATCH,    IBS_OP_DATA2,    0, 0x10, 0x00) IBS_FILTER(ANY_SET,     IBS_OP_DATA2,   0, 0x0007)
> +#define IBS_FILTER_0xf241 IBS_FILTER(MATCH,    IBS_OP_DATA2,    0, 0x10, 0x10) IBS_FILTER(ANY_SET,     IBS_OP_DATA2,   0, 0x0007)
> +#define IBS_FILTER_0xf242 IBS_FILTER(MATCH,    IBS_OP_DATA2,    0, 0x17, 0x01) IBS_FILTER(MATCH,       IBS_OP_DATA3,   0, 0x81, 0x81)
> +#define IBS_FILTER_0xf243 IBS_FILTER(MATCH,    IBS_OP_DATA2,    0, 0x17, 0x02) IBS_FILTER(MATCH,       IBS_OP_DATA3,   0, 0x81, 0x81)
> +#define IBS_FILTER_0xf244 IBS_FILTER(MATCH,    IBS_OP_DATA2,    0, 0x17, 0x12) IBS_FILTER(MATCH,       IBS_OP_DATA3,   0, 0x81, 0x81)
> +#define IBS_FILTER_0xf245 IBS_FILTER(MATCH,    IBS_OP_DATA2,    0, 0x17, 0x03) IBS_FILTER(MATCH,       IBS_OP_DATA3,   0, 0x81, 0x81)
> +#define IBS_FILTER_0xf246 IBS_FILTER(MATCH,    IBS_OP_DATA2,    0, 0x17, 0x13) IBS_FILTER(MATCH,       IBS_OP_DATA3,   0, 0x81, 0x81)
> +#define IBS_FILTER_0xf247 IBS_FILTER(MATCH,    IBS_OP_DATA2,    0, 0x17, 0x07) IBS_FILTER(MATCH,       IBS_OP_DATA3,   0, 0x81, 0x81)
> +#define IBS_FILTER_0xf248 IBS_FILTER(MATCH,    IBS_OP_DATA2,    0, 0x17, 0x17) IBS_FILTER(MATCH,       IBS_OP_DATA3,   0, 0x81, 0x81)
> +#define IBS_FILTER_0xf249 IBS_FILTER(MATCH,    IBS_OP_DATA2,    0, 0x27, 0x02) IBS_FILTER(MATCH,       IBS_OP_DATA3,   0, 0x81, 0x81)
> +#define IBS_FILTER_0xf24a IBS_FILTER(MATCH,    IBS_OP_DATA2,    0, 0x27, 0x22) IBS_FILTER(MATCH,       IBS_OP_DATA3,   0, 0x81, 0x81)
> +
> +static struct ibs_event events[] = {
> +       IBS_EVENT(0xf000, "ibs_fetch:ALL", "All IBS fetch samples"),
> +       IBS_EVENT(0xf001, "ibs_fetch:KILLED", "IBS fetch killed"),
> +       IBS_EVENT(0xf002, "ibs_fetch:ATTEMPTED", "IBS fetch attempted"),
> +       IBS_EVENT(0xf003, "ibs_fetch:COMPLETED", "IBS fetch completed"),
> +       IBS_EVENT(0xf004, "ibs_fetch:ABORTED", "IBS fetch aborted"),
> +       IBS_EVENT(0xf005, "ibs_fetch:ITLB_HITS", "IBS ITLB hit"),
> +       IBS_EVENT(0xf006, "ibs_fetch:L1_ITLB_MISSES_L2_ITLB_HITS", "IBS L1 ITLB misses (and L2 ITLB hits)"),
> +       IBS_EVENT(0xf007, "ibs_fetch:L1_ITLB_MISSES_L2_ITLB_MISSES", "IBS L1 L2 ITLB miss"),
> +       IBS_EVENT(0xf008, "ibs_fetch:ICACHE_MISSES", "IBS instruction cache misses"),
> +       IBS_EVENT(0xf009, "ibs_fetch:ICACHE_HITS", "IBS instruction cache hit"),
> +       IBS_EVENT(0xf00a, "ibs_fetch:4K_PAGE", "IBS 4K page translation"),
> +       IBS_EVENT(0xf00b, "ibs_fetch:2M_PAGE", "IBS 2M page translation"),
> +       IBS_EVENT(0xf00c, "ibs_fetch:1G_PAGE", "IBS 1G page translation"),
> +#if 0
> +       IBS_EVENT(0xf00e, "ibs_fetch:LATENCY", "IBS fetch latency"),
> +#endif
> +       IBS_EVENT(0xf100, "ibs_op:ALL", "All IBS op samples"),
> +#if 0
> +       IBS_EVENT(0xf101, "ibs_op:TAG_TO_RETIRE", "IBS tag-to-retire cycles"),
> +       IBS_EVENT(0xf102, "ibs_op:COMP_TO_RET", "IBS completion-to-retire cycles"),
> +#endif
> +       IBS_EVENT(0xf103, "ibs_op:BRANCH_RETIRED", "IBS branch op"),
> +       IBS_EVENT(0xf104, "ibs_op:MISPREDICTED_BRANCH", "IBS mispredicted branch op"),
> +       IBS_EVENT(0xf105, "ibs_op:TAKEN_BRANCH", "IBS taken branch op"),
> +       IBS_EVENT(0xf106, "ibs_op:MISPREDICTED_BRANCH_TAKEN", "IBS mispredicted taken branch op"),
> +       IBS_EVENT(0xf107, "ibs_op:RETURNS", "IBS return op"),
> +       IBS_EVENT(0xf108, "ibs_op:MISPREDICTED_RETURNS", "IBS mispredicted return op"),
> +       IBS_EVENT(0xf109, "ibs_op:RESYNC", "IBS resync op"),
> +       IBS_EVENT(0xf200, "ibs_op:ALL_LOAD_STORE", "IBS all load store ops"),
> +       IBS_EVENT(0xf201, "ibs_op:LOAD", "IBS load ops"),
> +       IBS_EVENT(0xf202, "ibs_op:STORE", "IBS store ops"),
> +       IBS_EVENT(0xf203, "ibs_op:L1_DTLB_HITS", "IBS L1 DTLB hit"),
> +       IBS_EVENT(0xf204, "ibs_op:L1_DTLB_MISS_L2_DTLB_HIT", "IBS L1 DTLB misses L2 hits"),
> +       IBS_EVENT(0xf205, "ibs_op:L1_L2_DTLB_MISS", "IBS L1 and L2 DTLB misses"),
> +       IBS_EVENT(0xf206, "ibs_op:DATA_CACHE_MISS", "IBS data cache misses"),
> +       IBS_EVENT(0xf207, "ibs_op:DATA_HITS", "IBS data cache hits"),
> +       IBS_EVENT(0xf208, "ibs_op:MISALIGNED_DATA_ACC", "IBS misaligned data access"),
> +       IBS_EVENT(0xf209, "ibs_op:BANK_CONF_LOAD", "IBS bank conflict on load op"),
> +#if 0
> +       IBS_EVENT(0xf20a, "ibs_op:BANK_CONF_STORE", "IBS bank conflict on store op"),
> +#endif
> +       IBS_EVENT(0xf20b, "ibs_op:FORWARD", "IBS store-to-load forwarded"),
> +       IBS_EVENT(0xf20c, "ibs_op:CANCELLED", "IBS store-to-load cancelled"),
> +       IBS_EVENT(0xf20d, "ibs_op:DCUC_MEM_ACC", "IBS UC memory access"),
> +       IBS_EVENT(0xf20e, "ibs_op:DCWC_MEM_ACC", "IBS WC memory access"),
> +       IBS_EVENT(0xf20f, "ibs_op:LOCKED", "IBS locked operation"),
> +       IBS_EVENT(0xf210, "ibs_op:MAB_HIT", "IBS MAB hit"),
> +       IBS_EVENT(0xf211, "ibs_op:L1_DTLB_4K", "IBS L1 DTLB 4K page"),
> +       IBS_EVENT(0xf212, "ibs_op:L1_DTLB_2M", "IBS L1 DTLB 2M page"),
> +       IBS_EVENT(0xf213, "ibs_op:L1_DTLB_1G", "IBS L1 DTLB 1G page"),
> +       IBS_EVENT(0xf215, "ibs_op:L2_DTLB_4K", "IBS L2 DTLB 4K page"),
> +       IBS_EVENT(0xf216, "ibs_op:L2_DTLB_2M", "IBS L2 DTLB 2M page"),
> +#if 0
> +       IBS_EVENT(0xf217, "ibs_op:L2_DTLB_1G", "IBS L2 DTLB 1G page"),
> +       IBS_EVENT(0xf219, "ibs_op:DC_LOAD_LAT", "IBS data cache miss load latency"),
> +#endif
> +       IBS_EVENT(0xf240, "ibs_op:NB_LOCAL_ONLY", "IBS Northbridge local"),
> +       IBS_EVENT(0xf241, "ibs_op:NB_REMOTE_ONLY", "IBS Northbridge remote"),
> +       IBS_EVENT(0xf242, "ibs_op:NB_LOCAL_L3", "IBS Northbridge local L3"),
> +       IBS_EVENT(0xf243, "ibs_op:NB_LOCAL_CACHE", "IBS Northbridge local core L1 or L2 cache"),
> +       IBS_EVENT(0xf244, "ibs_op:NB_REMOTE_CACHE", "IBS Northbridge local core L1, L2, L3 cache"),
> +       IBS_EVENT(0xf245, "ibs_op:NB_LOCAL_DRAM", "IBS Northbridge local DRAM"),
> +       IBS_EVENT(0xf246, "ibs_op:NB_REMOTE_DRAM", "IBS Northbridge remote DRAM"),
> +       IBS_EVENT(0xf247, "ibs_op:NB_LOCAL_OTHER", "IBS Northbridge local APIC MMIO Config PCI"),
> +       IBS_EVENT(0xf248, "ibs_op:NB_REMOTE_OTHER", "IBS Northbridge remote APIC MMIO Config PCI"),
> +       IBS_EVENT(0xf249, "ibs_op:NB_CACHE_MODIFIED", "IBS Northbridge cache modified state"),
> +       IBS_EVENT(0xf24a, "ibs_op:NB_CACHE_OWNED", "IBS Northbridge cache owned state"),
> +#if 0
> +       IBS_EVENT(0xf24b, "ibs_op:NB_LOCAL_CACHE_LAT", "IBS Northbridge local cache latency"),
> +       IBS_EVENT(0xf24c, "ibs_op:NB_REMOTE_CACHE_LAT", "IBS Northbridge remote cache latency"),
> +#endif
> +       { 0, NULL, NULL, { .filter = { IBS_FILTER_MATCH_ANY() } } }
>  };
>
>  static int ibs_parse_event(struct perf_event_attr *attr, char *sys, char *name)
>  {
> -       const char **event;
> +       struct ibs_event *event;
>
>        if (strcmp("ibs_op", sys) && strcmp("ibs_fetch", sys))
>                return -ENOENT;
>
> -       for (event = events; *event; event++) {
> -               if (!strcmp(*event + strlen(sys) + 1, name))
> +       for (event = events; event->id; event++) {
> +               if (!strcmp(event->name + strlen(sys) + 1, name))
>                        goto match;
>        }
>
>        return -EINVAL;
>  match:
> +       /* pseudo event found */
> +       attr->config1 = event->config;
>        attr->sample_type = PERF_SAMPLE_CPU;
>
>        return 0;
> @@ -97,13 +391,14 @@ match:
>
>  static void ibs_print_events(const char *sys)
>  {
> -       const char **event;
> +       struct ibs_event *event;
>
>        printf("\n");
>
> -       for (event = events; *event; event++) {
> -               if (!strncmp(sys, *event, strlen(sys)))
> -                       printf("  %-50s [PMU event: %s]\n", *event, sys);
> +       for (event = events; event->id; event++) {
> +               if (!strncmp(sys, event->name, strlen(sys)))
> +                       printf("  %-50s [PMU event: %s, id:0x%x]\n",
> +                              event->name, sys, event->id);
>        }
>  }
>
> --
> 1.7.8.4
>
>
????{.n?+???????+%?????ݶ??w??{.n?+????{??G?????{ay?ʇڙ?,j??f???h?????????z_??(?階?ݢj"???m??????G????????????&???~???iO???z??v?^?m????????????I?