summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_topology.c')
-rw-r--r--drivers/gpu/drm/amd/amdkfd/kfd_topology.c135
1 files changed, 112 insertions, 23 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index 6ed2ec381aaa..ceb9fb475ef1 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -31,6 +31,7 @@
#include <linux/log2.h>
#include <linux/dmi.h>
#include <linux/atomic.h>
+#include <linux/crc16.h>
#include "kfd_priv.h"
#include "kfd_crat.h"
@@ -291,6 +292,8 @@ static ssize_t iolink_show(struct kobject *kobj, struct attribute *attr,
iolink->max_bandwidth);
sysfs_show_32bit_prop(buffer, offs, "recommended_transfer_size",
iolink->rec_transfer_size);
+ sysfs_show_32bit_prop(buffer, offs, "recommended_sdma_engine_id_mask",
+ iolink->rec_sdma_eng_id_mask);
sysfs_show_32bit_prop(buffer, offs, "flags", iolink->flags);
return offs;
@@ -958,8 +961,7 @@ static void kfd_update_system_properties(void)
dev = list_last_entry(&topology_device_list,
struct kfd_topology_device, list);
if (dev) {
- sys_props.platform_id =
- (*((uint64_t *)dev->oem_id)) & CRAT_OEMID_64BIT_MASK;
+ sys_props.platform_id = dev->oem_id64;
sys_props.platform_oem = *((uint64_t *)dev->oem_table_id);
sys_props.platform_rev = dev->oem_revision;
}
@@ -1092,14 +1094,17 @@ void kfd_topology_shutdown(void)
static uint32_t kfd_generate_gpu_id(struct kfd_node *gpu)
{
- uint32_t hashout;
+ uint32_t gpu_id;
uint32_t buf[8];
uint64_t local_mem_size;
- int i;
+ struct kfd_topology_device *dev;
+ bool is_unique;
+ uint8_t *crc_buf;
if (!gpu)
return 0;
+ crc_buf = (uint8_t *)&buf;
local_mem_size = gpu->local_mem_info.local_mem_size_private +
gpu->local_mem_info.local_mem_size_public;
buf[0] = gpu->adev->pdev->devfn;
@@ -1112,10 +1117,34 @@ static uint32_t kfd_generate_gpu_id(struct kfd_node *gpu)
buf[6] = upper_32_bits(local_mem_size);
buf[7] = (ffs(gpu->xcc_mask) - 1) | (NUM_XCC(gpu->xcc_mask) << 16);
- for (i = 0, hashout = 0; i < 8; i++)
- hashout ^= hash_32(buf[i], KFD_GPU_ID_HASH_WIDTH);
+ gpu_id = crc16(0, crc_buf, sizeof(buf)) &
+ ((1 << KFD_GPU_ID_HASH_WIDTH) - 1);
- return hashout;
+ /* There is a very small possibility when generating a
+ * 16 (KFD_GPU_ID_HASH_WIDTH) bit value from 8 word buffer
+ * that the value could be 0 or non-unique. So, check if
+ * it is unique and non-zero. If not unique increment till
+ * unique one is found. In case of overflow, restart from 1
+ */
+
+ down_read(&topology_lock);
+ do {
+ is_unique = true;
+ if (!gpu_id)
+ gpu_id = 1;
+ list_for_each_entry(dev, &topology_device_list, list) {
+ if (dev->gpu && dev->gpu_id == gpu_id) {
+ is_unique = false;
+ break;
+ }
+ }
+ if (unlikely(!is_unique))
+ gpu_id = (gpu_id + 1) &
+ ((1 << KFD_GPU_ID_HASH_WIDTH) - 1);
+ } while (!is_unique);
+ up_read(&topology_lock);
+
+ return gpu_id;
}
/* kfd_assign_gpu - Attach @gpu to the correct kfd topology device. If
* the GPU device is not already present in the topology device
@@ -1238,6 +1267,54 @@ static void kfd_set_iolink_non_coherent(struct kfd_topology_device *to_dev,
}
}
+#define REC_SDMA_NUM_GPU 8
+static const int rec_sdma_eng_map[REC_SDMA_NUM_GPU][REC_SDMA_NUM_GPU] = {
+ { -1, 14, 12, 2, 4, 8, 10, 6 },
+ { 14, -1, 2, 10, 8, 4, 6, 12 },
+ { 10, 2, -1, 12, 14, 6, 4, 8 },
+ { 2, 12, 10, -1, 6, 14, 8, 4 },
+ { 4, 8, 14, 6, -1, 10, 12, 2 },
+ { 8, 4, 6, 14, 12, -1, 2, 10 },
+ { 10, 6, 4, 8, 12, 2, -1, 14 },
+ { 6, 12, 8, 4, 2, 10, 14, -1 }};
+
+static void kfd_set_recommended_sdma_engines(struct kfd_topology_device *to_dev,
+ struct kfd_iolink_properties *outbound_link,
+ struct kfd_iolink_properties *inbound_link)
+{
+ struct kfd_node *gpu = outbound_link->gpu;
+ struct amdgpu_device *adev = gpu->adev;
+ int num_xgmi_nodes = adev->gmc.xgmi.num_physical_nodes;
+ bool support_rec_eng = !amdgpu_sriov_vf(adev) && to_dev->gpu &&
+ adev->aid_mask && num_xgmi_nodes && gpu->kfd->num_nodes == 1 &&
+ kfd_get_num_xgmi_sdma_engines(gpu) >= 14 &&
+ (!(adev->flags & AMD_IS_APU) && num_xgmi_nodes == 8);
+
+ if (support_rec_eng) {
+ int src_socket_id = adev->gmc.xgmi.physical_node_id;
+ int dst_socket_id = to_dev->gpu->adev->gmc.xgmi.physical_node_id;
+
+ outbound_link->rec_sdma_eng_id_mask =
+ 1 << rec_sdma_eng_map[src_socket_id][dst_socket_id];
+ inbound_link->rec_sdma_eng_id_mask =
+ 1 << rec_sdma_eng_map[dst_socket_id][src_socket_id];
+ } else {
+ int num_sdma_eng = kfd_get_num_sdma_engines(gpu);
+ int i, eng_offset = 0;
+
+ if (outbound_link->iolink_type == CRAT_IOLINK_TYPE_XGMI &&
+ kfd_get_num_xgmi_sdma_engines(gpu) && to_dev->gpu) {
+ eng_offset = num_sdma_eng;
+ num_sdma_eng = kfd_get_num_xgmi_sdma_engines(gpu);
+ }
+
+ for (i = 0; i < num_sdma_eng; i++) {
+ outbound_link->rec_sdma_eng_id_mask |= (1 << (i + eng_offset));
+ inbound_link->rec_sdma_eng_id_mask |= (1 << (i + eng_offset));
+ }
+ }
+}
+
static void kfd_fill_iolink_non_crat_info(struct kfd_topology_device *dev)
{
struct kfd_iolink_properties *link, *inbound_link;
@@ -1276,6 +1353,7 @@ static void kfd_fill_iolink_non_crat_info(struct kfd_topology_device *dev)
inbound_link->flags = CRAT_IOLINK_FLAGS_ENABLED;
kfd_set_iolink_no_atomics(peer_dev, dev, inbound_link);
kfd_set_iolink_non_coherent(peer_dev, link, inbound_link);
+ kfd_set_recommended_sdma_engines(peer_dev, link, inbound_link);
}
}
@@ -1564,6 +1642,7 @@ static int fill_in_l1_pcache(struct kfd_cache_properties **props_ext,
pcache->processor_id_low = cu_processor_id + (first_active_cu - 1);
pcache->cache_level = pcache_info[cache_type].cache_level;
pcache->cache_size = pcache_info[cache_type].cache_size;
+ pcache->cacheline_size = pcache_info[cache_type].cache_line_size;
if (pcache_info[cache_type].flags & CRAT_CACHE_FLAGS_DATA_CACHE)
pcache->cache_type |= HSA_CACHE_TYPE_DATA;
@@ -1632,8 +1711,11 @@ static int fill_in_l2_l3_pcache(struct kfd_cache_properties **props_ext,
pcache->processor_id_low = cu_processor_id
+ (first_active_cu - 1);
pcache->cache_level = pcache_info[cache_type].cache_level;
+ pcache->cacheline_size = pcache_info[cache_type].cache_line_size;
- if (KFD_GC_VERSION(knode) == IP_VERSION(9, 4, 3))
+ if (KFD_GC_VERSION(knode) == IP_VERSION(9, 4, 3) ||
+ KFD_GC_VERSION(knode) == IP_VERSION(9, 4, 4) ||
+ KFD_GC_VERSION(knode) == IP_VERSION(9, 5, 0))
mode = adev->gmc.gmc_funcs->query_mem_partition_mode(adev);
else
mode = UNKNOWN_MEMORY_PARTITION_MODE;
@@ -1695,7 +1777,7 @@ static void kfd_fill_cache_non_crat_info(struct kfd_topology_device *dev, struct
struct amdgpu_cu_info *cu_info = &kdev->adev->gfx.cu_info;
struct amdgpu_gfx_config *gfx_info = &kdev->adev->gfx.config;
int gpu_processor_id;
- struct kfd_cache_properties *props_ext;
+ struct kfd_cache_properties *props_ext = NULL;
int num_of_entries = 0;
int num_of_cache_types = 0;
struct kfd_gpu_cache_info cache_info[KFD_MAX_CACHE_TYPES];
@@ -1703,6 +1785,7 @@ static void kfd_fill_cache_non_crat_info(struct kfd_topology_device *dev, struct
gpu_processor_id = dev->node_props.simd_id_base;
+ memset(cache_info, 0, sizeof(cache_info));
pcache_info = cache_info;
num_of_cache_types = kfd_get_gpu_cache_info(kdev, &pcache_info);
if (!num_of_cache_types) {
@@ -1769,7 +1852,7 @@ static void kfd_fill_cache_non_crat_info(struct kfd_topology_device *dev, struct
pr_debug("Added [%d] GPU cache entries\n", num_of_entries);
}
-static int kfd_topology_add_device_locked(struct kfd_node *gpu, uint32_t gpu_id,
+static int kfd_topology_add_device_locked(struct kfd_node *gpu,
struct kfd_topology_device **dev)
{
int proximity_domain = ++topology_crat_proximity_domain;
@@ -1782,8 +1865,7 @@ static int kfd_topology_add_device_locked(struct kfd_node *gpu, uint32_t gpu_id,
COMPUTE_UNIT_GPU, gpu,
proximity_domain);
if (res) {
- pr_err("Error creating VCRAT for GPU (ID: 0x%x)\n",
- gpu_id);
+ dev_err(gpu->adev->dev, "Error creating VCRAT\n");
topology_crat_proximity_domain--;
goto err;
}
@@ -1794,8 +1876,7 @@ static int kfd_topology_add_device_locked(struct kfd_node *gpu, uint32_t gpu_id,
&temp_topology_device_list,
proximity_domain);
if (res) {
- pr_err("Error parsing VCRAT for GPU (ID: 0x%x)\n",
- gpu_id);
+ dev_err(gpu->adev->dev, "Error parsing VCRAT\n");
topology_crat_proximity_domain--;
goto err;
}
@@ -1821,8 +1902,8 @@ static int kfd_topology_add_device_locked(struct kfd_node *gpu, uint32_t gpu_id,
if (!res)
sys_props.generation_count++;
else
- pr_err("Failed to update GPU (ID: 0x%x) to sysfs topology. res=%d\n",
- gpu_id, res);
+ dev_err(gpu->adev->dev, "Failed to update GPU to sysfs topology. res=%d\n",
+ res);
err:
kfd_destroy_crat_image(crat_image);
@@ -1905,7 +1986,8 @@ static void kfd_topology_set_capabilities(struct kfd_topology_device *dev)
dev->node_props.debug_prop |= HSA_DBG_DISPATCH_INFO_ALWAYS_VALID;
if (KFD_GC_VERSION(dev->gpu) < IP_VERSION(10, 0, 0)) {
- if (KFD_GC_VERSION(dev->gpu) == IP_VERSION(9, 4, 3))
+ if (KFD_GC_VERSION(dev->gpu) == IP_VERSION(9, 4, 3) ||
+ KFD_GC_VERSION(dev->gpu) == IP_VERSION(9, 4, 4))
dev->node_props.debug_prop |=
HSA_DBG_WATCH_ADDR_MASK_LO_BIT_GFX9_4_3 |
HSA_DBG_WATCH_ADDR_MASK_HI_BIT_GFX9_4_3;
@@ -1917,6 +1999,8 @@ static void kfd_topology_set_capabilities(struct kfd_topology_device *dev)
if (KFD_GC_VERSION(dev->gpu) >= IP_VERSION(9, 4, 2))
dev->node_props.capability |=
HSA_CAP_TRAP_DEBUG_PRECISE_MEMORY_OPERATIONS_SUPPORTED;
+
+ dev->node_props.capability |= HSA_CAP_PER_QUEUE_RESET_SUPPORTED;
} else {
dev->node_props.debug_prop |= HSA_DBG_WATCH_ADDR_MASK_LO_BIT_GFX10 |
HSA_DBG_WATCH_ADDR_MASK_HI_BIT;
@@ -1924,6 +2008,10 @@ static void kfd_topology_set_capabilities(struct kfd_topology_device *dev)
if (KFD_GC_VERSION(dev->gpu) >= IP_VERSION(11, 0, 0))
dev->node_props.capability |=
HSA_CAP_TRAP_DEBUG_PRECISE_MEMORY_OPERATIONS_SUPPORTED;
+
+ if (KFD_GC_VERSION(dev->gpu) >= IP_VERSION(12, 0, 0))
+ dev->node_props.capability |=
+ HSA_CAP_TRAP_DEBUG_PRECISE_ALU_OPERATIONS_SUPPORTED;
}
kfd_topology_set_dbg_firmware_support(dev);
@@ -1939,14 +2027,12 @@ int kfd_topology_add_device(struct kfd_node *gpu)
struct amdgpu_gfx_config *gfx_info = &gpu->adev->gfx.config;
struct amdgpu_cu_info *cu_info = &gpu->adev->gfx.cu_info;
- gpu_id = kfd_generate_gpu_id(gpu);
if (gpu->xcp && !gpu->xcp->ddev) {
dev_warn(gpu->adev->dev,
- "Won't add GPU (ID: 0x%x) to topology since it has no drm node assigned.",
- gpu_id);
+ "Won't add GPU to topology since it has no drm node assigned.");
return 0;
} else {
- pr_debug("Adding new GPU (ID: 0x%x) to topology\n", gpu_id);
+ dev_dbg(gpu->adev->dev, "Adding new GPU to topology\n");
}
/* Check to see if this gpu device exists in the topology_device_list.
@@ -1958,11 +2044,12 @@ int kfd_topology_add_device(struct kfd_node *gpu)
down_write(&topology_lock);
dev = kfd_assign_gpu(gpu);
if (!dev)
- res = kfd_topology_add_device_locked(gpu, gpu_id, &dev);
+ res = kfd_topology_add_device_locked(gpu, &dev);
up_write(&topology_lock);
if (res)
return res;
+ gpu_id = kfd_generate_gpu_id(gpu);
dev->gpu_id = gpu_id;
gpu->id = gpu_id;
@@ -1994,7 +2081,7 @@ int kfd_topology_add_device(struct kfd_node *gpu)
HSA_CAP_ASIC_REVISION_MASK);
dev->node_props.location_id = pci_dev_id(gpu->adev->pdev);
- if (KFD_GC_VERSION(dev->gpu->kfd) == IP_VERSION(9, 4, 3))
+ if (gpu->kfd->num_nodes > 1)
dev->node_props.location_id |= dev->gpu->node_id;
dev->node_props.domain = pci_domain_nr(gpu->adev->pdev->bus);
@@ -2087,6 +2174,8 @@ int kfd_topology_add_device(struct kfd_node *gpu)
dev->gpu->adev->gmc.xgmi.connected_to_cpu)
dev->node_props.capability |= HSA_CAP_FLAGS_COHERENTHOSTACCESS;
+ kfd_queue_ctx_save_restore_size(dev);
+
kfd_debug_print_topology();
kfd_notify_gpu_change(gpu_id, 1);
generated by cgit (git 2.34.1) at 2025-07-25 20:10:25 +0000