diff options
Diffstat (limited to 'src')
| -rwxr-xr-x | src/cache_types.h | 4 | ||||
| -rwxr-xr-x | src/cachepc.c | 52 | ||||
| -rwxr-xr-x | src/cachepc.h | 10 | ||||
| -rw-r--r-- | src/util.c | 4 |
4 files changed, 49 insertions, 21 deletions
diff --git a/src/cache_types.h b/src/cache_types.h index 4bc112c..6bba73d 100755 --- a/src/cache_types.h +++ b/src/cache_types.h @@ -8,8 +8,8 @@ #define REMOVE_PAGE_OFFSET(ptr) ((void *) (((uintptr_t) ptr) & PAGE_MASK)) -#define GET_BIT(b, i) (((b & (1 << i)) >> i) & 1) -#define SET_BIT(b, i) (b | (1 << i)) +#define GET_BIT(b, i) (((b) >> (i)) & 1) +#define SET_BIT(b, i) ((b) | (1 << (i))) /* Operate cacheline flags * Used flags: diff --git a/src/cachepc.c b/src/cachepc.c index bd5f4d2..b5f8589 100755 --- a/src/cachepc.c +++ b/src/cachepc.c @@ -26,16 +26,16 @@ cachepc_init_counters(void) event = event_no | (event_mask << 8); event |= (1<< 17); /* OsUserMode bit */ event |= (1 << 22); /* enable performance counter */ - printk(KERN_INFO "Writing to msr event %d", event); + printk(KERN_WARNING "CachePC: Initialized event %d\n", event); asm volatile ("wrmsr" : : "c"(reg_addr), "a"(event), "d"(0x00)); reg_addr = 0xc0010202; event_no = 0x64; event_mask = 0xC8; event = event_no | (event_mask << 8); - event |= (1<< 17); /* OsUserMode bit */ + event |= (1 << 17); /* OsUserMode bit */ event |= (1 << 22); /* enable performance counter */ - printk(KERN_INFO "Writing to msr event %d", event); + printk(KERN_WARNING "CachePC: Initialized event %d\n", event); asm volatile ("wrmsr" : : "c"(reg_addr), "a"(event), "d"(0x00)); } @@ -44,11 +44,12 @@ cachepc_get_ctx(cache_level cache_level) { cache_ctx *ctx; - printk(KERN_INFO "CACHEPC_GET_CTX"); + // printk(KERN_WARNING "CachePC: Getting ctx..\n"); ctx = kzalloc(sizeof(cache_ctx), GFP_KERNEL); BUG_ON(ctx == NULL); + BUG_ON(cache_level != L1); if (cache_level == L1) { ctx->addressing = L1_ADDRESSING; ctx->sets = L1_SETS; @@ -68,6 +69,8 @@ cachepc_get_ctx(cache_level cache_level) ctx->set_size = CACHELINE_SIZE * ctx->associativity; ctx->cache_size = ctx->sets * ctx->set_size; + // printk(KERN_WARNING "CachePC: Getting ctx done\n"); + return ctx; } @@ -80,12 +83,14 @@ cachepc_prepare_ds(cache_ctx *ctx) cacheline **cacheline_ptr_arr; cacheline *cache_ds; - printk(KERN_INFO "CACHEPC_BUILD_CACHE_DS"); + //printk(KERN_WARNING "CachePC: Preparing ds..\n"); cacheline_ptr_arr = allocate_cache_ds(ctx); cache_ds = build_cache_ds(ctx, cacheline_ptr_arr); kfree(cacheline_ptr_arr); + // printk(KERN_WARNING "CachePC: Preparing ds done\n"); + return cache_ds; } @@ -106,7 +111,7 @@ cachepc_print_msrmts(cacheline *head) curr_cl = head; do { if (IS_FIRST(curr_cl->flags)) { - printk(KERN_WARNING "Count for cache set %i: %llu\n", + printk(KERN_WARNING "CachePC: Count for cache set %i: %llu\n", curr_cl->cache_set, curr_cl->count); } @@ -114,6 +119,24 @@ cachepc_print_msrmts(cacheline *head) } while (curr_cl != head); } +void * +remove_cache_set(cache_ctx *ctx, void *ptr) +{ + return (void *) (((uintptr_t) ptr) & ~SET_MASK(ctx->sets)); +} + +void +cachepc_release_ds(cache_ctx *ctx, cacheline *ds) +{ + kfree(remove_cache_set(ctx, ds)); +} + +void +cachepc_release_ctx(cache_ctx *ctx) +{ + kfree(ctx); +} + /* * Create a randomized doubly linked list with the following structure: * set A <--> set B <--> ... <--> set X <--> set A @@ -136,13 +159,13 @@ cacheline *build_cache_ds(cache_ctx *ctx, cacheline **cl_ptr_arr) { idx_per_set = kzalloc(ctx->sets * sizeof(uint32_t), GFP_KERNEL); BUG_ON(idx_per_set == NULL); - cl_ptr_arr_sorted = kmalloc(ctx->nr_of_cachelines * sizeof(cacheline *), GFP_KERNEL); + cl_ptr_arr_sorted = kzalloc(ctx->nr_of_cachelines * sizeof(cacheline *), GFP_KERNEL); BUG_ON(cl_ptr_arr_sorted == NULL); set_len = ctx->associativity; for (i = 0; i < ctx->nr_of_cachelines; ++i) { - set_offset = cl_ptr_arr[i]->cache_set * set_len; - idx_curr_set = idx_per_set[cl_ptr_arr[i]->cache_set]; + set_offset = cl_ptr_arr[i]->cache_set * set_len; + idx_curr_set = idx_per_set[cl_ptr_arr[i]->cache_set]; cl_ptr_arr_sorted[set_offset + idx_curr_set] = cl_ptr_arr[i]; idx_per_set[cl_ptr_arr[i]->cache_set] += 1; @@ -198,8 +221,8 @@ void build_randomized_list_for_cache_set(cache_ctx *ctx, cacheline **cacheline_p curr_cl->prev = cacheline_ptr_arr[idx_map[(len - 1 + i) % len]]; curr_cl->count = 0; - if (curr_cl == cacheline_ptr_arr[0]) { - curr_cl->flags = SET_FIRST(DEFAULT_FLAGS); + if (idx_map[i] == 0) { + curr_cl->flags = SET_FIRST(DEFAULT_FLAGS); curr_cl->prev->flags = SET_LAST(DEFAULT_FLAGS); } else { curr_cl->flags = curr_cl->flags | DEFAULT_FLAGS; @@ -219,13 +242,13 @@ allocate_cache_ds(cache_ctx *ctx) cacheline **cl_ptr_arr, *cl_arr; uint32_t i; - cl_ptr_arr = (cacheline **) kzalloc(ctx->nr_of_cachelines * sizeof(cacheline *), GFP_KERNEL); + cl_ptr_arr = kzalloc(ctx->nr_of_cachelines * sizeof(cacheline *), GFP_KERNEL); BUG_ON(cl_ptr_arr == NULL); BUG_ON(ctx->addressing != VIRTUAL); // For virtual addressing, allocating a consecutive chunk of memory is enough - cl_arr = (cacheline *) aligned_alloc(PAGE_SIZE, ctx->cache_size); + cl_arr = aligned_alloc(PAGE_SIZE, ctx->cache_size); BUG_ON(cl_arr == NULL); for (i = 0; i < ctx->nr_of_cachelines; ++i) { @@ -249,10 +272,9 @@ aligned_alloc(size_t alignment, size_t size) if (size % alignment != 0) size = size - (size % alignment) + alignment; - p = kmalloc(size, GFP_KERNEL); + p = kzalloc(size, GFP_KERNEL); BUG_ON(((uintptr_t) p) % alignment != 0); return p; } - diff --git a/src/cachepc.h b/src/cachepc.h index 06f85f0..c40fc10 100755 --- a/src/cachepc.h +++ b/src/cachepc.h @@ -17,6 +17,8 @@ cache_ctx *cachepc_get_ctx(cache_level cl); cacheline *cachepc_prepare_ds(cache_ctx *ctx); void cachepc_save_msrmts(cacheline *head, const char *prefix, int index); void cachepc_print_msrmts(cacheline *head); +void cachepc_release_ds(cache_ctx *ctx, cacheline *ds); +void cachepc_release_ctx(cache_ctx *ctx); __attribute__((always_inline)) static inline cacheline *cachepc_prime(cacheline *head); @@ -39,7 +41,7 @@ cachepc_prime(cacheline *head) { cacheline *curr_cl; - printk(KERN_WARNING "PROBE"); + //printk(KERN_WARNING "CachePC: Priming..\n"); cachepc_cpuid(); curr_cl = head; @@ -49,6 +51,8 @@ cachepc_prime(cacheline *head) } while(curr_cl != head); cachepc_cpuid(); + //printk(KERN_WARNING "CachePC: Priming done\n"); + return curr_cl->prev; } @@ -128,13 +132,15 @@ cachepc_probe(cacheline *head) { cacheline *curr_cs; - printk(KERN_WARNING "PROBE"); + //printk(KERN_WARNING "CachePC: Probing.."); curr_cs = head; do { curr_cs = cachepc_probe_set(curr_cs); } while (__builtin_expect(curr_cs != head, 1)); + //printk(KERN_WARNING "CachePC: Probing done"); + return curr_cs->next; } @@ -12,8 +12,8 @@ random_perm(uint32_t *arr, uint32_t arr_len) idx = idx % i; tmp = arr[idx]; - arr[i] = arr[idx]; - arr[idx] = tmp; + arr[idx] = arr[i]; + arr[i] = tmp; } } |