[igt-dev] [PATCH i-g-t] benchmarks/gem_wsim: introduce bb_size field

[Tvrtko Ursulin]

On 27/10/2023 13:28, Marcin Bernatowicz wrote:
> In certain scenarios (ATS-M when using LMEM), PAGE_SIZE=4096
> is insufficient for Xe, necessitating alignment considerations.
> This change ensures that 'bb_size' aligns properly,
> preventing VM BIND failures in cases where the size is not aligned
> to the minimum page size of the region.
> 
> Additionally, 'alloc_bo' for i915 has been updated to
> accommodate page-aligned allocated size.
> 
> Signed-off-by: Marcin Bernatowicz <marcin.bernatowicz at intel.com>
> ---
>   benchmarks/gem_wsim.c | 41 +++++++++++++++++++++++++----------------
>   1 file changed, 25 insertions(+), 16 deletions(-)
> 
> diff --git a/benchmarks/gem_wsim.c b/benchmarks/gem_wsim.c
> index 28b809520..01973ade6 100644
> --- a/benchmarks/gem_wsim.c
> +++ b/benchmarks/gem_wsim.c
> @@ -201,6 +201,7 @@ struct w_step {
>   			struct drm_xe_sync *syncs;
>   		} xe;
>   	};
> +	unsigned long bb_size;
>   	uint32_t bb_handle;
>   };
>   
> @@ -1503,7 +1504,7 @@ static unsigned int create_bb(struct w_step *w, int self)
>   	gem_set_domain(fd, w->bb_handle,
>   		       I915_GEM_DOMAIN_WC, I915_GEM_DOMAIN_WC);
>   
> -	cs = ptr = gem_mmap__wc(fd, w->bb_handle, 0, 4096, PROT_WRITE);
> +	cs = ptr = gem_mmap__wc(fd, w->bb_handle, 0, w->bb_size, PROT_WRITE);
>   
>   	/* Store initial 64b timestamp: start */
>   	*cs++ = MI_LOAD_REGISTER_IMM(1) | MI_CS_MMIO_DST;
> @@ -1653,9 +1654,16 @@ xe_get_vm(struct workload *wrk, const struct w_step *w)
>   	return wrk->xe.vm_list;
>   }
>   
> -static uint32_t alloc_bo(int i915, unsigned long size)
> +static uint32_t alloc_bo(int i915, unsigned long *size)
>   {
> -	return gem_create(i915, size);
> +	uint32_t handle;
> +	uint64_t sz = *size;
> +
> +	igt_assert_eq(__gem_create(i915, &sz, &handle), 0);
> +	igt_assert(sz <= ULONG_MAX);
> +	*size = sz;
> +
> +	return handle;
>   }
>   
>   static void
> @@ -1669,7 +1677,8 @@ alloc_step_batch(struct workload *wrk, struct w_step *w)
>   	w->i915.obj = calloc(nr_obj, sizeof(*w->i915.obj));
>   	igt_assert(w->i915.obj);
>   
> -	w->i915.obj[j].handle = alloc_bo(fd, 4096);
> +	w->bb_size = PAGE_SIZE;
> +	w->i915.obj[j].handle = alloc_bo(fd, &w->bb_size);
>   	w->i915.obj[j].flags = EXEC_OBJECT_WRITE;
>   	j++;
>   	igt_assert(j < nr_obj);
> @@ -1706,7 +1715,7 @@ alloc_step_batch(struct workload *wrk, struct w_step *w)
>   		igt_assert(j < nr_obj);
>   	}
>   
> -	w->bb_handle = w->i915.obj[j].handle = gem_create(fd, 4096);
> +	w->bb_handle = w->i915.obj[j].handle = alloc_bo(fd, &w->bb_size);
>   	w->i915.obj[j].relocation_count = create_bb(w, j);
>   	igt_assert(w->i915.obj[j].relocation_count <= ARRAY_SIZE(w->i915.reloc));
>   	w->i915.obj[j].relocs_ptr = to_user_pointer(&w->i915.reloc);
> @@ -1734,13 +1743,13 @@ xe_alloc_step_batch(struct workload *wrk, struct w_step *w)
>   	struct dep_entry *dep;
>   	int i;
>   
> -	w->bb_handle = xe_bo_create_flags(fd, vm->id, PAGE_SIZE,
> +	w->bb_size = ALIGN(PAGE_SIZE + xe_cs_prefetch_size(fd),
> +			   xe_get_default_alignment(fd));
> +	w->bb_handle = xe_bo_create_flags(fd, vm->id, w->bb_size,
>   				visible_vram_if_possible(fd, eq->hwe_list[0].gt_id));
> -	w->xe.data = xe_bo_map(fd, w->bb_handle, PAGE_SIZE);
> -	w->xe.exec.address =
> -		intel_allocator_alloc_with_strategy(vm->ahnd, w->bb_handle, PAGE_SIZE,
> -						    0, ALLOC_STRATEGY_LOW_TO_HIGH);
> -	xe_vm_bind_sync(fd, vm->id, w->bb_handle, 0, w->xe.exec.address, PAGE_SIZE);
> +	w->xe.data = xe_bo_map(fd, w->bb_handle, w->bb_size);
> +	w->xe.exec.address = get_offset(vm->ahnd, w->bb_handle, w->bb_size, 0);

Apart from replacing intel_allocator_alloc_with_strategy with 
get_offset, which I don't see is a local, although get_offset suggests 
it should be, the rest looks fine to me.

Regards,

Tvrtko

> +	xe_vm_bind_sync(fd, vm->id, w->bb_handle, 0, w->xe.exec.address, w->bb_size);
>   	xe_spin_init_opts(&w->xe.data->spin, .addr = w->xe.exec.address,
>   				   .preempt = (w->preempt_us > 0),
>   				   .ctx_ticks = duration_to_ctx_ticks(fd, eq->hwe_list[0].gt_id,
> @@ -1936,7 +1945,7 @@ allocate_working_set(struct workload *wrk, struct working_set *set)
>   
>   	for (i = 0; i < set->nr; i++) {
>   		set->sizes[i].size = get_buffer_size(wrk, &set->sizes[i]);
> -		set->handles[i] = alloc_bo(fd, set->sizes[i].size);
> +		set->handles[i] = alloc_bo(fd, &set->sizes[i].size);
>   		total += set->sizes[i].size;
>   	}
>   
> @@ -1971,7 +1980,7 @@ static void measure_active_set(struct workload *wrk)
>   		if (w->type != BATCH)
>   			continue;
>   
> -		batch_sizes += 4096;
> +		batch_sizes += w->bb_size;
>   
>   		if (is_xe)
>   			continue;
> @@ -1990,7 +1999,7 @@ static void measure_active_set(struct workload *wrk)
>   
>   			if (!find_dep(deps, nr, _dep)) {
>   				if (dep->working_set == -1) {
> -					total += 4096;
> +					total += w->bb_size;
>   				} else {
>   					struct working_set *set;
>   
> @@ -2834,8 +2843,8 @@ static void *run_workload(void *data)
>   				syncobj_destroy(fd, w->xe.syncs[0].handle);
>   				free(w->xe.syncs);
>   				xe_vm_unbind_sync(fd, xe_get_vm(wrk, w)->id, 0, w->xe.exec.address,
> -						  PAGE_SIZE);
> -				gem_munmap(w->xe.data, PAGE_SIZE);
> +						  w->bb_size);
> +				gem_munmap(w->xe.data, w->bb_size);
>   				gem_close(fd, w->bb_handle);
>   			} else if (w->type == SW_FENCE) {
>   				syncobj_destroy(fd, w->xe.syncs[0].handle);