diff options
author | Felix Kuehling <Felix.Kuehling@amd.com> | 2017-12-09 07:08:49 +0300 |
---|---|---|
committer | Oded Gabbay <oded.gabbay@gmail.com> | 2017-12-09 07:08:49 +0300 |
commit | 174de876d6d02f7159363495a3f34461ba91d3ee (patch) | |
tree | 4336dfd0390744be6d1f232e75a0018a7703b323 /drivers/gpu/drm/amd/amdkfd/kfd_topology.c | |
parent | 5108d768408abc80e4e8d99f5b406a73cb04056b (diff) | |
download | linux-174de876d6d02f7159363495a3f34461ba91d3ee.tar.xz |
drm/amdkfd: Group up CRAT related functions
Take CRAT related functions out of kfd_topology.c and place them in
kfd_crat.c. This is the initial step of supporting more CRAT features,
i.e. creating virtual CRAT table for KFD devices without CRAT.
v2: Minor cleanup that was missed previously because code moved around
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Signed-off-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_topology.c')
-rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 334 |
1 files changed, 5 insertions, 329 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c index 9b9824f97de5..b6cf785d0138 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c @@ -34,9 +34,8 @@ #include "kfd_topology.h" #include "kfd_device_queue_manager.h" -static struct list_head topology_device_list; -static int topology_crat_parsed; -static struct kfd_system_properties sys_props; +struct list_head topology_device_list; +struct kfd_system_properties sys_props; static DECLARE_RWSEM(topology_lock); @@ -76,276 +75,6 @@ struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev) return device; } -static int kfd_topology_get_crat_acpi(void *crat_image, size_t *size) -{ - struct acpi_table_header *crat_table; - acpi_status status; - - if (!size) - return -EINVAL; - - /* - * Fetch the CRAT table from ACPI - */ - status = acpi_get_table(CRAT_SIGNATURE, 0, &crat_table); - if (status == AE_NOT_FOUND) { - pr_warn("CRAT table not found\n"); - return -ENODATA; - } else if (ACPI_FAILURE(status)) { - const char *err = acpi_format_exception(status); - - pr_err("CRAT table error: %s\n", err); - return -EINVAL; - } - - if (*size >= crat_table->length && crat_image != NULL) - memcpy(crat_image, crat_table, crat_table->length); - - *size = crat_table->length; - - return 0; -} - -static void kfd_populated_cu_info_cpu(struct kfd_topology_device *dev, - struct crat_subtype_computeunit *cu) -{ - dev->node_props.cpu_cores_count = cu->num_cpu_cores; - dev->node_props.cpu_core_id_base = cu->processor_id_low; - if (cu->hsa_capability & CRAT_CU_FLAGS_IOMMU_PRESENT) - dev->node_props.capability |= HSA_CAP_ATS_PRESENT; - - pr_info("CU CPU: cores=%d id_base=%d\n", cu->num_cpu_cores, - cu->processor_id_low); -} - -static void kfd_populated_cu_info_gpu(struct kfd_topology_device *dev, - struct crat_subtype_computeunit *cu) -{ - dev->node_props.simd_id_base = cu->processor_id_low; - dev->node_props.simd_count = cu->num_simd_cores; - dev->node_props.lds_size_in_kb = cu->lds_size_in_kb; - dev->node_props.max_waves_per_simd = cu->max_waves_simd; - dev->node_props.wave_front_size = cu->wave_front_size; - dev->node_props.mem_banks_count = cu->num_banks; - dev->node_props.array_count = cu->num_arrays; - dev->node_props.cu_per_simd_array = cu->num_cu_per_array; - dev->node_props.simd_per_cu = cu->num_simd_per_cu; - dev->node_props.max_slots_scratch_cu = cu->max_slots_scatch_cu; - if (cu->hsa_capability & CRAT_CU_FLAGS_HOT_PLUGGABLE) - dev->node_props.capability |= HSA_CAP_HOT_PLUGGABLE; - pr_info("CU GPU: id_base=%d\n", cu->processor_id_low); -} - -/* kfd_parse_subtype_cu is called when the topology mutex is already acquired */ -static int kfd_parse_subtype_cu(struct crat_subtype_computeunit *cu) -{ - struct kfd_topology_device *dev; - int i = 0; - - pr_info("Found CU entry in CRAT table with proximity_domain=%d caps=%x\n", - cu->proximity_domain, cu->hsa_capability); - list_for_each_entry(dev, &topology_device_list, list) { - if (cu->proximity_domain == i) { - if (cu->flags & CRAT_CU_FLAGS_CPU_PRESENT) - kfd_populated_cu_info_cpu(dev, cu); - - if (cu->flags & CRAT_CU_FLAGS_GPU_PRESENT) - kfd_populated_cu_info_gpu(dev, cu); - break; - } - i++; - } - - return 0; -} - -/* - * kfd_parse_subtype_mem is called when the topology mutex is - * already acquired - */ -static int kfd_parse_subtype_mem(struct crat_subtype_memory *mem) -{ - struct kfd_mem_properties *props; - struct kfd_topology_device *dev; - int i = 0; - - pr_info("Found memory entry in CRAT table with proximity_domain=%d\n", - mem->promixity_domain); - list_for_each_entry(dev, &topology_device_list, list) { - if (mem->promixity_domain == i) { - props = kfd_alloc_struct(props); - if (props == NULL) - return -ENOMEM; - - if (dev->node_props.cpu_cores_count == 0) - props->heap_type = HSA_MEM_HEAP_TYPE_FB_PRIVATE; - else - props->heap_type = HSA_MEM_HEAP_TYPE_SYSTEM; - - if (mem->flags & CRAT_MEM_FLAGS_HOT_PLUGGABLE) - props->flags |= HSA_MEM_FLAGS_HOT_PLUGGABLE; - if (mem->flags & CRAT_MEM_FLAGS_NON_VOLATILE) - props->flags |= HSA_MEM_FLAGS_NON_VOLATILE; - - props->size_in_bytes = - ((uint64_t)mem->length_high << 32) + - mem->length_low; - props->width = mem->width; - - dev->mem_bank_count++; - list_add_tail(&props->list, &dev->mem_props); - - break; - } - i++; - } - - return 0; -} - -/* - * kfd_parse_subtype_cache is called when the topology mutex - * is already acquired - */ -static int kfd_parse_subtype_cache(struct crat_subtype_cache *cache) -{ - struct kfd_cache_properties *props; - struct kfd_topology_device *dev; - uint32_t id; - - id = cache->processor_id_low; - - pr_info("Found cache entry in CRAT table with processor_id=%d\n", id); - list_for_each_entry(dev, &topology_device_list, list) - if (id == dev->node_props.cpu_core_id_base || - id == dev->node_props.simd_id_base) { - props = kfd_alloc_struct(props); - if (props == NULL) - return -ENOMEM; - - props->processor_id_low = id; - props->cache_level = cache->cache_level; - props->cache_size = cache->cache_size; - props->cacheline_size = cache->cache_line_size; - props->cachelines_per_tag = cache->lines_per_tag; - props->cache_assoc = cache->associativity; - props->cache_latency = cache->cache_latency; - - if (cache->flags & CRAT_CACHE_FLAGS_DATA_CACHE) - props->cache_type |= HSA_CACHE_TYPE_DATA; - if (cache->flags & CRAT_CACHE_FLAGS_INST_CACHE) - props->cache_type |= HSA_CACHE_TYPE_INSTRUCTION; - if (cache->flags & CRAT_CACHE_FLAGS_CPU_CACHE) - props->cache_type |= HSA_CACHE_TYPE_CPU; - if (cache->flags & CRAT_CACHE_FLAGS_SIMD_CACHE) - props->cache_type |= HSA_CACHE_TYPE_HSACU; - - dev->cache_count++; - dev->node_props.caches_count++; - list_add_tail(&props->list, &dev->cache_props); - - break; - } - - return 0; -} - -/* - * kfd_parse_subtype_iolink is called when the topology mutex - * is already acquired - */ -static int kfd_parse_subtype_iolink(struct crat_subtype_iolink *iolink) -{ - struct kfd_iolink_properties *props; - struct kfd_topology_device *dev; - uint32_t i = 0; - uint32_t id_from; - uint32_t id_to; - - id_from = iolink->proximity_domain_from; - id_to = iolink->proximity_domain_to; - - pr_info("Found IO link entry in CRAT table with id_from=%d\n", id_from); - list_for_each_entry(dev, &topology_device_list, list) { - if (id_from == i) { - props = kfd_alloc_struct(props); - if (props == NULL) - return -ENOMEM; - - props->node_from = id_from; - props->node_to = id_to; - props->ver_maj = iolink->version_major; - props->ver_min = iolink->version_minor; - - /* - * weight factor (derived from CDIR), currently always 1 - */ - props->weight = 1; - - props->min_latency = iolink->minimum_latency; - props->max_latency = iolink->maximum_latency; - props->min_bandwidth = iolink->minimum_bandwidth_mbs; - props->max_bandwidth = iolink->maximum_bandwidth_mbs; - props->rec_transfer_size = - iolink->recommended_transfer_size; - - dev->io_link_count++; - dev->node_props.io_links_count++; - list_add_tail(&props->list, &dev->io_link_props); - - break; - } - i++; - } - - return 0; -} - -static int kfd_parse_subtype(struct crat_subtype_generic *sub_type_hdr) -{ - struct crat_subtype_computeunit *cu; - struct crat_subtype_memory *mem; - struct crat_subtype_cache *cache; - struct crat_subtype_iolink *iolink; - int ret = 0; - - switch (sub_type_hdr->type) { - case CRAT_SUBTYPE_COMPUTEUNIT_AFFINITY: - cu = (struct crat_subtype_computeunit *)sub_type_hdr; - ret = kfd_parse_subtype_cu(cu); - break; - case CRAT_SUBTYPE_MEMORY_AFFINITY: - mem = (struct crat_subtype_memory *)sub_type_hdr; - ret = kfd_parse_subtype_mem(mem); - break; - case CRAT_SUBTYPE_CACHE_AFFINITY: - cache = (struct crat_subtype_cache *)sub_type_hdr; - ret = kfd_parse_subtype_cache(cache); - break; - case CRAT_SUBTYPE_TLB_AFFINITY: - /* - * For now, nothing to do here - */ - pr_info("Found TLB entry in CRAT table (not processing)\n"); - break; - case CRAT_SUBTYPE_CCOMPUTE_AFFINITY: - /* - * For now, nothing to do here - */ - pr_info("Found CCOMPUTE entry in CRAT table (not processing)\n"); - break; - case CRAT_SUBTYPE_IOLINK_AFFINITY: - iolink = (struct crat_subtype_iolink *)sub_type_hdr; - ret = kfd_parse_subtype_iolink(iolink); - break; - default: - pr_warn("Unknown subtype (%d) in CRAT\n", - sub_type_hdr->type); - } - - return ret; -} - static void kfd_release_topology_device(struct kfd_topology_device *dev) { struct kfd_mem_properties *mem; @@ -380,7 +109,7 @@ static void kfd_release_topology_device(struct kfd_topology_device *dev) sys_props.num_devices--; } -static void kfd_release_live_view(void) +void kfd_release_live_view(void) { struct kfd_topology_device *dev; @@ -393,7 +122,7 @@ static void kfd_release_live_view(void) memset(&sys_props, 0, sizeof(sys_props)); } -static struct kfd_topology_device *kfd_create_topology_device(void) +struct kfd_topology_device *kfd_create_topology_device(void) { struct kfd_topology_device *dev; @@ -413,58 +142,6 @@ static struct kfd_topology_device *kfd_create_topology_device(void) return dev; } -static int kfd_parse_crat_table(void *crat_image) -{ - struct kfd_topology_device *top_dev; - struct crat_subtype_generic *sub_type_hdr; - uint16_t node_id; - int ret; - struct crat_header *crat_table = (struct crat_header *)crat_image; - uint16_t num_nodes; - uint32_t image_len; - - if (!crat_image) - return -EINVAL; - - num_nodes = crat_table->num_domains; - image_len = crat_table->length; - - pr_info("Parsing CRAT table with %d nodes\n", num_nodes); - - for (node_id = 0; node_id < num_nodes; node_id++) { - top_dev = kfd_create_topology_device(); - if (!top_dev) { - kfd_release_live_view(); - return -ENOMEM; - } - } - - sys_props.platform_id = - (*((uint64_t *)crat_table->oem_id)) & CRAT_OEMID_64BIT_MASK; - sys_props.platform_oem = *((uint64_t *)crat_table->oem_table_id); - sys_props.platform_rev = crat_table->revision; - - sub_type_hdr = (struct crat_subtype_generic *)(crat_table+1); - while ((char *)sub_type_hdr + sizeof(struct crat_subtype_generic) < - ((char *)crat_image) + image_len) { - if (sub_type_hdr->flags & CRAT_SUBTYPE_FLAGS_ENABLED) { - ret = kfd_parse_subtype(sub_type_hdr); - if (ret != 0) { - kfd_release_live_view(); - return ret; - } - } - - sub_type_hdr = (typeof(sub_type_hdr))((char *)sub_type_hdr + - sub_type_hdr->length); - } - - sys_props.generation_count++; - topology_crat_parsed = 1; - - return 0; -} - #define sysfs_show_gen_prop(buffer, fmt, ...) \ snprintf(buffer, PAGE_SIZE, "%s"fmt, buffer, __VA_ARGS__) @@ -1016,7 +693,6 @@ int kfd_topology_init(void) */ INIT_LIST_HEAD(&topology_device_list); init_rwsem(&topology_lock); - topology_crat_parsed = 0; memset(&sys_props, 0, sizeof(sys_props)); @@ -1219,7 +895,7 @@ int kfd_topology_remove_device(struct kfd_dev *gpu) up_write(&topology_lock); - if (res == 0) + if (!res) kfd_notify_gpu_change(gpu_id, 0); return res; |