This is a note to let you know that I've just added the patch titled
drm/amdkfd: Update cache info reporting for GFX v9.4.3
to the 6.5-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
drm-amdkfd-update-cache-info-reporting-for-gfx-v9.4..patch
and it can be found in the queue-6.5 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree,
please let <stable@xxxxxxxxxxxxxxx> know about it.
commit 7d8488befbe225e24c545a004ec4c81dd256c902
Author: Mukul Joshi <mukul.joshi@xxxxxxx>
Date: Fri Aug 25 12:18:06 2023 -0400
drm/amdkfd: Update cache info reporting for GFX v9.4.3
[ Upstream commit 0752e66e91fa86fa5481b04b22053363833ffb85 ]
Update cache info reporting in sysfs to report the correct
number of CUs and associated cache information based on
different spatial partitioning modes.
Signed-off-by: Mukul Joshi <mukul.joshi@xxxxxxx>
Reviewed-by: Felix Kuehling <Felix.Kuehling@xxxxxxx>
Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_crat.h b/drivers/gpu/drm/amd/amdkfd/kfd_crat.h
index fc719389b5d65..4684711aa695a 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_crat.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_crat.h
@@ -79,6 +79,10 @@ struct crat_header {
#define CRAT_SUBTYPE_IOLINK_AFFINITY 5
#define CRAT_SUBTYPE_MAX 6
+/*
+ * Do not change the value of CRAT_SIBLINGMAP_SIZE from 32
+ * as it breaks the ABI.
+ */
#define CRAT_SIBLINGMAP_SIZE 32
/*
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index ea67a353beb00..5582191022106 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -1650,14 +1650,17 @@ static int fill_in_l1_pcache(struct kfd_cache_properties **props_ext,
static int fill_in_l2_l3_pcache(struct kfd_cache_properties **props_ext,
struct kfd_gpu_cache_info *pcache_info,
struct kfd_cu_info *cu_info,
- int cache_type, unsigned int cu_processor_id)
+ int cache_type, unsigned int cu_processor_id,
+ struct kfd_node *knode)
{
unsigned int cu_sibling_map_mask;
int first_active_cu;
- int i, j, k;
+ int i, j, k, xcc, start, end;
struct kfd_cache_properties *pcache = NULL;
- cu_sibling_map_mask = cu_info->cu_bitmap[0][0][0];
+ start = ffs(knode->xcc_mask) - 1;
+ end = start + NUM_XCC(knode->xcc_mask);
+ cu_sibling_map_mask = cu_info->cu_bitmap[start][0][0];
cu_sibling_map_mask &=
((1 << pcache_info[cache_type].num_cu_shared) - 1);
first_active_cu = ffs(cu_sibling_map_mask);
@@ -1692,16 +1695,18 @@ static int fill_in_l2_l3_pcache(struct kfd_cache_properties **props_ext,
cu_sibling_map_mask = cu_sibling_map_mask >> (first_active_cu - 1);
k = 0;
- for (i = 0; i < cu_info->num_shader_engines; i++) {
- for (j = 0; j < cu_info->num_shader_arrays_per_engine; j++) {
- pcache->sibling_map[k] = (uint8_t)(cu_sibling_map_mask & 0xFF);
- pcache->sibling_map[k+1] = (uint8_t)((cu_sibling_map_mask >> 8) & 0xFF);
- pcache->sibling_map[k+2] = (uint8_t)((cu_sibling_map_mask >> 16) & 0xFF);
- pcache->sibling_map[k+3] = (uint8_t)((cu_sibling_map_mask >> 24) & 0xFF);
- k += 4;
-
- cu_sibling_map_mask = cu_info->cu_bitmap[0][i % 4][j + i / 4];
- cu_sibling_map_mask &= ((1 << pcache_info[cache_type].num_cu_shared) - 1);
+ for (xcc = start; xcc < end; xcc++) {
+ for (i = 0; i < cu_info->num_shader_engines; i++) {
+ for (j = 0; j < cu_info->num_shader_arrays_per_engine; j++) {
+ pcache->sibling_map[k] = (uint8_t)(cu_sibling_map_mask & 0xFF);
+ pcache->sibling_map[k+1] = (uint8_t)((cu_sibling_map_mask >> 8) & 0xFF);
+ pcache->sibling_map[k+2] = (uint8_t)((cu_sibling_map_mask >> 16) & 0xFF);
+ pcache->sibling_map[k+3] = (uint8_t)((cu_sibling_map_mask >> 24) & 0xFF);
+ k += 4;
+
+ cu_sibling_map_mask = cu_info->cu_bitmap[xcc][i % 4][j + i / 4];
+ cu_sibling_map_mask &= ((1 << pcache_info[cache_type].num_cu_shared) - 1);
+ }
}
}
pcache->sibling_map_size = k;
@@ -1719,7 +1724,7 @@ static int fill_in_l2_l3_pcache(struct kfd_cache_properties **props_ext,
static void kfd_fill_cache_non_crat_info(struct kfd_topology_device *dev, struct kfd_node *kdev)
{
struct kfd_gpu_cache_info *pcache_info = NULL;
- int i, j, k;
+ int i, j, k, xcc, start, end;
int ct = 0;
unsigned int cu_processor_id;
int ret;
@@ -1753,37 +1758,42 @@ static void kfd_fill_cache_non_crat_info(struct kfd_topology_device *dev, struct
* then it will consider only one CU from
* the shared unit
*/
+ start = ffs(kdev->xcc_mask) - 1;
+ end = start + NUM_XCC(kdev->xcc_mask);
+
for (ct = 0; ct < num_of_cache_types; ct++) {
cu_processor_id = gpu_processor_id;
if (pcache_info[ct].cache_level == 1) {
- for (i = 0; i < pcu_info->num_shader_engines; i++) {
- for (j = 0; j < pcu_info->num_shader_arrays_per_engine; j++) {
- for (k = 0; k < pcu_info->num_cu_per_sh; k += pcache_info[ct].num_cu_shared) {
-
- ret = fill_in_l1_pcache(&props_ext, pcache_info, pcu_info,
- pcu_info->cu_bitmap[0][i % 4][j + i / 4], ct,
- cu_processor_id, k);
-
- if (ret < 0)
- break;
-
- if (!ret) {
- num_of_entries++;
- list_add_tail(&props_ext->list, &dev->cache_props);
+ for (xcc = start; xcc < end; xcc++) {
+ for (i = 0; i < pcu_info->num_shader_engines; i++) {
+ for (j = 0; j < pcu_info->num_shader_arrays_per_engine; j++) {
+ for (k = 0; k < pcu_info->num_cu_per_sh; k += pcache_info[ct].num_cu_shared) {
+
+ ret = fill_in_l1_pcache(&props_ext, pcache_info, pcu_info,
+ pcu_info->cu_bitmap[xcc][i % 4][j + i / 4], ct,
+ cu_processor_id, k);
+
+ if (ret < 0)
+ break;
+
+ if (!ret) {
+ num_of_entries++;
+ list_add_tail(&props_ext->list, &dev->cache_props);
+ }
+
+ /* Move to next CU block */
+ num_cu_shared = ((k + pcache_info[ct].num_cu_shared) <=
+ pcu_info->num_cu_per_sh) ?
+ pcache_info[ct].num_cu_shared :
+ (pcu_info->num_cu_per_sh - k);
+ cu_processor_id += num_cu_shared;
}
-
- /* Move to next CU block */
- num_cu_shared = ((k + pcache_info[ct].num_cu_shared) <=
- pcu_info->num_cu_per_sh) ?
- pcache_info[ct].num_cu_shared :
- (pcu_info->num_cu_per_sh - k);
- cu_processor_id += num_cu_shared;
}
}
}
} else {
ret = fill_in_l2_l3_pcache(&props_ext, pcache_info,
- pcu_info, ct, cu_processor_id);
+ pcu_info, ct, cu_processor_id, kdev);
if (ret < 0)
break;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.h b/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
index cba2cd5ed9d19..46927263e014d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
@@ -86,7 +86,7 @@ struct kfd_mem_properties {
struct attribute attr;
};
-#define CACHE_SIBLINGMAP_SIZE 64
+#define CACHE_SIBLINGMAP_SIZE 128
struct kfd_cache_properties {
struct list_head list;
