Re: [PATCH 1/4] perf: Add memory load/store events generic code

[Lin Ming]

On Wed, 2011-07-06 at 21:51 +0800, Peter Zijlstra wrote:
> On Wed, 2011-07-06 at 13:53 +0800, Lin Ming wrote:
> 
> > Do you mean to use the "impossible combinations" to express the inverse?
> 
> Nah, impossible would be things like having neither LOAD nor STORE set.
> 
> > MEM_STORE_DCU_MISS as: store-l2-l3-dram
> > MEM_STORE_STLB_MISS as: store-itlb-dtlb
> > 
> > How about below code?
> 
> Right, something like that. Robert can the IBS data source data be
> mapped onto this as well?
> 
> > #define PERF_MEM_LOAD                   (1ULL << 0)
> > #define PERF_MEM_STORE                  (1ULL << 1)
> > #define PERF_MEM_ATOMIC                 (1ULL << 2)
> > #define PERF_MEM_L1                     (1ULL << 3)
> > #define PERF_MEM_L2                     (1ULL << 4)
> > #define PERF_MEM_L3                     (1ULL << 5)
> > #define PERF_MEM_RAM                    (1ULL << 6)
> > #define PERF_MEM_UNKNOWN                (1ULL << 7)
> > #define PERF_MEM_IO                     (1ULL << 8)
> > #define PERF_MEM_UNCACHED               (1ULL << 9)
> > #define PERF_MEM_RESERVED               (1ULL << 10)
> > #define PERF_MEM_LOCAL                  (1ULL << 11)
> > #define PERF_MEM_REMOTE                 (1ULL << 12)
> > #define PERF_MEM_SNOOP                  (1ULL << 13)
> > #define PERF_MEM_MODIFIED               (1ULL << 14)
> > #define PERF_MEM_EXCLUSIVE              (1ULL << 15)
> > #define PERF_MEM_SHARED                 (1ULL << 16)
> > #define PERF_MEM_INVALID                (1ULL << 17)
> 
> > #define PERF_MEM_ITLB                   (1ULL << 18)
> > #define PERF_MEM_DTLB                   (1ULL << 19)
> > #define PERF_MEM_STLB                   (1ULL << 20)
> 
> Are these TLB hit or miss?

I meant hit, but that's not correct.
How about more complete definition as below?

#define PERF_MEM_L1D_HIT           
#define PERF_MEM_L1D_MISS               
#define PERF_MEM_L1I_HIT                
#define PERF_MEM_L1I_MISS              
#define PERF_MEM_L2_HIT                
#define PERF_MEM_L2_MISS                
#define PERF_MEM_L3_HIT                 
#define PERF_MEM_L3_MISS                

#define PERF_MEM_ITLB_HIT               
#define PERF_MEM_ITLB_MISS              
#define PERF_MEM_DTLB_HIT               
#define PERF_MEM_DTLB_MISS              
#define PERF_MEM_STLB_HIT               
#define PERF_MEM_STLB_MISS              

#define PERF_MEM_STORE_L1D_HIT  \
        (PERF_MEM_STORE | PERF_MEM_L1D_HIT)

#define PERF_MEM_STORE_L1D_MISS \
        (PERF_MEM_STORE | PERF_MEM_L1D_MISS)

#define PERF_MEM_STORE_STLB_HIT \
        (PERF_MEM_STORE | PERF_MEM_STLB_HIT)

#define PERF_MEM_STORE_STLB_MISS \
        (PERF_MEM_STORE | PERF_MEM_STLB_MISS)

#define PERF_MEM_STORE_ATOMIC \
        (PERF_MEM_STORE | PERF_MEM_ATOMIC)

#define PERF_MEM_LOAD_STLB_HIT  \
        (PERF_MEM_LOAD | PERF_MEM_STLB_HIT)

#define PERF_MEM_LOAD_STLB_MISS \
        (PERF_MEM_LOAD | PERF_MEM_STLB_MISS)

#define PERF_MEM_LOAD_ATOMIC \
        (PERF_MEM_LOAD | PERF_MEM_ATOMIC)

Lin Ming

> 
> > #define PERF_MEM_STORE_L1D_HIT  \
> >         (PERF_MEM_STORE | PERF_MEM_L1)
> > 
> > #define PERF_MEM_STORE_L1D_MISS \
> >         (PERF_MEM_STORE | PERF_MEM_L2 | PERF_MEM_L3 | PERF_MEM_RAM)
> > 
> > #define PERF_MEM_STORE_STLB_HIT \
> >         (PERF_MEM_STORE | PERF_MEM_STLB)
> >       
> > #define PERF_MEM_STORE_STLB_MISS \
> >         (PERF_MEM_STORE | PERF_MEM_ITLB | PERF_MEM_DTLB)
> 
> Going by the definition in table 30-22 neither of these seem correct, a
> STLB_HIT was defined as DTLB|STLB whereas a STLB_MISS was missing both
> (resulting in a full page-table walk I presume).
> 
> > #define PERF_MEM_STORE_ATOMIC \
> >         (PERF_MEM_STORE | PERF_MEM_ATOMIC)
> > 
> > #define PERF_MEM_LOAD_STLB_HIT  \
> >         (PERF_MEM_LOAD | PERF_MEM_STLB)
> >    
> > #define PERF_MEM_LOAD_STLB_MISS \
> >         (PERF_MEM_LOAD | PERF_MEM_ITLB | PERF_MEM_DTLB)
> 
> idem
> 
> > #define PERF_MEM_LOAD_ATOMIC \
> >         (PERF_MEM_LOAD | PERF_MEM_ATOMIC)
> > 
> 


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/