On 21/07/14 01:37, Jerome Glisse wrote: > On Thu, Jul 17, 2014 at 04:29:17PM +0300, Oded Gabbay wrote: >> From: Evgeny Pinchuk <evgeny.pinchuk at amd.com> >> >> This patch adds the topology module to the driver. The topology is exposed to >> userspace through the sysfs. >> >> The calls to add and remove a device to/from topology are done by the radeon >> driver. > > So overall we already said that we do not want to see the cpu architecture > re-expose by hsa in its own format. This pacth is NACK. Only expose additional > non existent information and also follow the number one rules of sysfs which > is one value -> one file. > > I understand the temptation to rexpose the cpu topology in your own way to > make life simpler but there is already api for this so please use what exist > today and if there is short coming than i am sure they can be fixed. > > See : > > /sys/devices/system/cpu/cpu* > Agreed and we will change the code. Hopefully it will be in v3, although we may release v3 early this week and release v4 with this change next week.
Oded >> >> Signed-off-by: Evgeny Pinchuk <evgeny.pinchuk at amd.com> >> Signed-off-by: Oded Gabbay <oded.gabbay at amd.com> >> --- >> drivers/gpu/drm/radeon/amdkfd/Makefile | 2 +- >> drivers/gpu/drm/radeon/amdkfd/kfd_crat.h | 294 +++++++ >> drivers/gpu/drm/radeon/amdkfd/kfd_device.c | 7 + >> drivers/gpu/drm/radeon/amdkfd/kfd_module.c | 7 + >> drivers/gpu/drm/radeon/amdkfd/kfd_priv.h | 17 + >> drivers/gpu/drm/radeon/amdkfd/kfd_topology.c | 1207 >> ++++++++++++++++++++++++++ >> drivers/gpu/drm/radeon/amdkfd/kfd_topology.h | 168 ++++ >> 7 files changed, 1701 insertions(+), 1 deletion(-) >> create mode 100644 drivers/gpu/drm/radeon/amdkfd/kfd_crat.h >> create mode 100644 drivers/gpu/drm/radeon/amdkfd/kfd_topology.c >> create mode 100644 drivers/gpu/drm/radeon/amdkfd/kfd_topology.h >> >> diff --git a/drivers/gpu/drm/radeon/amdkfd/Makefile >> b/drivers/gpu/drm/radeon/amdkfd/Makefile >> index 9564e75..08ecfcd 100644 >> --- a/drivers/gpu/drm/radeon/amdkfd/Makefile >> +++ b/drivers/gpu/drm/radeon/amdkfd/Makefile >> @@ -4,6 +4,6 @@ >> >> ccflags-y := -Iinclude/drm >> >> -amdkfd-y := kfd_module.o kfd_device.o kfd_chardev.o >> +amdkfd-y := kfd_module.o kfd_device.o kfd_chardev.o kfd_topology.o >> >> obj-$(CONFIG_HSA_RADEON) += amdkfd.o >> diff --git a/drivers/gpu/drm/radeon/amdkfd/kfd_crat.h >> b/drivers/gpu/drm/radeon/amdkfd/kfd_crat.h >> new file mode 100644 >> index 0000000..a374fa3 >> --- /dev/null >> +++ b/drivers/gpu/drm/radeon/amdkfd/kfd_crat.h >> @@ -0,0 +1,294 @@ >> +/* >> + * Copyright 2014 Advanced Micro Devices, Inc. >> + * >> + * Permission is hereby granted, free of charge, to any person obtaining a >> + * copy of this software and associated documentation files (the >> "Software"), >> + * to deal in the Software without restriction, including without limitation >> + * the rights to use, copy, modify, merge, publish, distribute, sublicense, >> + * and/or sell copies of the Software, and to permit persons to whom the >> + * Software is furnished to do so, subject to the following conditions: >> + * >> + * The above copyright notice and this permission notice shall be included >> in >> + * all copies or substantial portions of the Software. >> + * >> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS >> OR >> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, >> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL >> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR >> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, >> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR >> + * OTHER DEALINGS IN THE SOFTWARE. >> + */ >> + >> +#ifndef KFD_CRAT_H_INCLUDED >> +#define KFD_CRAT_H_INCLUDED >> + >> +#include <linux/types.h> >> + >> +#pragma pack(1) > > No pragma > >> + >> +/* >> + * 4CC signature values for the CRAT and CDIT ACPI tables >> + */ >> + >> +#define CRAT_SIGNATURE "CRAT" >> +#define CDIT_SIGNATURE "CDIT" >> + >> +/* >> + * Component Resource Association Table (CRAT) >> + */ >> + >> +#define CRAT_OEMID_LENGTH 6 >> +#define CRAT_OEMTABLEID_LENGTH 8 >> +#define CRAT_RESERVED_LENGTH 6 >> + >> +#define CRAT_OEMID_64BIT_MASK ((1ULL << (CRAT_OEMID_LENGTH * 8)) - 1) >> + >> +struct crat_header { >> + uint32_t signature; >> + uint32_t length; >> + uint8_t revision; >> + uint8_t checksum; >> + uint8_t oem_id[CRAT_OEMID_LENGTH]; >> + uint8_t oem_table_id[CRAT_OEMTABLEID_LENGTH]; >> + uint32_t oem_revision; >> + uint32_t creator_id; >> + uint32_t creator_revision; >> + uint32_t total_entries; >> + uint16_t num_domains; >> + uint8_t reserved[CRAT_RESERVED_LENGTH]; >> +}; >> + >> +/* >> + * The header structure is immediately followed by total_entries of the >> + * data definitions >> + */ >> + >> +/* >> + * The currently defined subtype entries in the CRAT >> + */ >> +#define CRAT_SUBTYPE_COMPUTEUNIT_AFFINITY 0 >> +#define CRAT_SUBTYPE_MEMORY_AFFINITY 1 >> +#define CRAT_SUBTYPE_CACHE_AFFINITY 2 >> +#define CRAT_SUBTYPE_TLB_AFFINITY 3 >> +#define CRAT_SUBTYPE_CCOMPUTE_AFFINITY 4 >> +#define CRAT_SUBTYPE_IOLINK_AFFINITY 5 >> +#define CRAT_SUBTYPE_MAX 6 >> + >> +#define CRAT_SIBLINGMAP_SIZE 32 >> + >> +/* >> + * ComputeUnit Affinity structure and definitions >> + */ >> +#define CRAT_CU_FLAGS_ENABLED 0x00000001 >> +#define CRAT_CU_FLAGS_HOT_PLUGGABLE 0x00000002 >> +#define CRAT_CU_FLAGS_CPU_PRESENT 0x00000004 >> +#define CRAT_CU_FLAGS_GPU_PRESENT 0x00000008 >> +#define CRAT_CU_FLAGS_IOMMU_PRESENT 0x00000010 >> +#define CRAT_CU_FLAGS_RESERVED 0xffffffe0 >> + >> +#define CRAT_COMPUTEUNIT_RESERVED_LENGTH 4 >> + >> +struct crat_subtype_computeunit { >> + uint8_t type; >> + uint8_t length; >> + uint16_t reserved; >> + uint32_t flags; >> + uint32_t proximity_domain; >> + uint32_t processor_id_low; >> + uint16_t num_cpu_cores; >> + uint16_t num_simd_cores; >> + uint16_t max_waves_simd; >> + uint16_t io_count; >> + uint16_t hsa_capability; >> + uint16_t lds_size_in_kb; >> + uint8_t wave_front_size; >> + uint8_t num_banks; >> + uint16_t micro_engine_id; >> + uint8_t num_arrays; >> + uint8_t num_cu_per_array; >> + uint8_t num_simd_per_cu; >> + uint8_t max_slots_scatch_cu; >> + uint8_t reserved2[CRAT_COMPUTEUNIT_RESERVED_LENGTH]; >> +}; >> + >> +/* >> + * HSA Memory Affinity structure and definitions >> + */ >> +#define CRAT_MEM_FLAGS_ENABLED 0x00000001 >> +#define CRAT_MEM_FLAGS_HOT_PLUGGABLE 0x00000002 >> +#define CRAT_MEM_FLAGS_NON_VOLATILE 0x00000004 >> +#define CRAT_MEM_FLAGS_RESERVED 0xfffffff8 >> + >> +#define CRAT_MEMORY_RESERVED_LENGTH 8 >> + >> +struct crat_subtype_memory { >> + uint8_t type; >> + uint8_t length; >> + uint16_t reserved; >> + uint32_t flags; >> + uint32_t promixity_domain; >> + uint32_t base_addr_low; >> + uint32_t base_addr_high; >> + uint32_t length_low; >> + uint32_t length_high; >> + uint32_t width; >> + uint8_t reserved2[CRAT_MEMORY_RESERVED_LENGTH]; >> +}; >> + >> +/* >> + * HSA Cache Affinity structure and definitions >> + */ >> +#define CRAT_CACHE_FLAGS_ENABLED 0x00000001 >> +#define CRAT_CACHE_FLAGS_DATA_CACHE 0x00000002 >> +#define CRAT_CACHE_FLAGS_INST_CACHE 0x00000004 >> +#define CRAT_CACHE_FLAGS_CPU_CACHE 0x00000008 >> +#define CRAT_CACHE_FLAGS_SIMD_CACHE 0x00000010 >> +#define CRAT_CACHE_FLAGS_RESERVED 0xffffffe0 >> + >> +#define CRAT_CACHE_RESERVED_LENGTH 8 >> + >> +struct crat_subtype_cache { >> + uint8_t type; >> + uint8_t length; >> + uint16_t reserved; >> + uint32_t flags; >> + uint32_t processor_id_low; >> + uint8_t sibling_map[CRAT_SIBLINGMAP_SIZE]; >> + uint32_t cache_size; >> + uint8_t cache_level; >> + uint8_t lines_per_tag; >> + uint16_t cache_line_size; >> + uint8_t associativity; >> + uint8_t cache_properties; >> + uint16_t cache_latency; >> + uint8_t reserved2[CRAT_CACHE_RESERVED_LENGTH]; >> +}; >> + >> +/* >> + * HSA TLB Affinity structure and definitions >> + */ >> +#define CRAT_TLB_FLAGS_ENABLED 0x00000001 >> +#define CRAT_TLB_FLAGS_DATA_TLB 0x00000002 >> +#define CRAT_TLB_FLAGS_INST_TLB 0x00000004 >> +#define CRAT_TLB_FLAGS_CPU_TLB 0x00000008 >> +#define CRAT_TLB_FLAGS_SIMD_TLB 0x00000010 >> +#define CRAT_TLB_FLAGS_RESERVED 0xffffffe0 >> + >> +#define CRAT_TLB_RESERVED_LENGTH 4 >> + >> +struct crat_subtype_tlb { >> + uint8_t type; >> + uint8_t length; >> + uint16_t reserved; >> + uint32_t flags; >> + uint32_t processor_id_low; >> + uint8_t sibling_map[CRAT_SIBLINGMAP_SIZE]; >> + uint32_t tlb_level; >> + uint8_t data_tlb_associativity_2mb; >> + uint8_t data_tlb_size_2mb; >> + uint8_t instruction_tlb_associativity_2mb; >> + uint8_t instruction_tlb_size_2mb; >> + uint8_t data_tlb_associativity_4k; >> + uint8_t data_tlb_size_4k; >> + uint8_t instruction_tlb_associativity_4k; >> + uint8_t instruction_tlb_size_4k; >> + uint8_t data_tlb_associativity_1gb; >> + uint8_t data_tlb_size_1gb; >> + uint8_t instruction_tlb_associativity_1gb; >> + uint8_t instruction_tlb_size_1gb; >> + uint8_t reserved2[CRAT_TLB_RESERVED_LENGTH]; >> +}; >> + >> +/* >> + * HSA CCompute/APU Affinity structure and definitions >> + */ >> +#define CRAT_CCOMPUTE_FLAGS_ENABLED 0x00000001 >> +#define CRAT_CCOMPUTE_FLAGS_RESERVED 0xfffffffe >> + >> +#define CRAT_CCOMPUTE_RESERVED_LENGTH 16 >> + >> +struct crat_subtype_ccompute { >> + uint8_t type; >> + uint8_t length; >> + uint16_t reserved; >> + uint32_t flags; >> + uint32_t processor_id_low; >> + uint8_t sibling_map[CRAT_SIBLINGMAP_SIZE]; >> + uint32_t apu_size; >> + uint8_t reserved2[CRAT_CCOMPUTE_RESERVED_LENGTH]; >> +}; >> + >> +/* >> + * HSA IO Link Affinity structure and definitions >> + */ >> +#define CRAT_IOLINK_FLAGS_ENABLED 0x00000001 >> +#define CRAT_IOLINK_FLAGS_COHERENCY 0x00000002 >> +#define CRAT_IOLINK_FLAGS_RESERVED 0xfffffffc >> + >> +/* >> + * IO interface types >> + */ >> +#define CRAT_IOLINK_TYPE_UNDEFINED 0 >> +#define CRAT_IOLINK_TYPE_HYPERTRANSPORT 1 >> +#define CRAT_IOLINK_TYPE_PCIEXPRESS 2 >> +#define CRAT_IOLINK_TYPE_OTHER 3 >> +#define CRAT_IOLINK_TYPE_MAX 255 >> + >> +#define CRAT_IOLINK_RESERVED_LENGTH 24 >> + >> +struct crat_subtype_iolink { >> + uint8_t type; >> + uint8_t length; >> + uint16_t reserved; >> + uint32_t flags; >> + uint32_t proximity_domain_from; >> + uint32_t proximity_domain_to; >> + uint8_t io_interface_type; >> + uint8_t version_major; >> + uint16_t version_minor; >> + uint32_t minimum_latency; >> + uint32_t maximum_latency; >> + uint32_t minimum_bandwidth_mbs; >> + uint32_t maximum_bandwidth_mbs; >> + uint32_t recommended_transfer_size; >> + uint8_t reserved2[CRAT_IOLINK_RESERVED_LENGTH]; >> +}; >> + >> +/* >> + * HSA generic sub-type header >> + */ >> + >> +#define CRAT_SUBTYPE_FLAGS_ENABLED 0x00000001 >> + >> +struct crat_subtype_generic { >> + uint8_t type; >> + uint8_t length; >> + uint16_t reserved; >> + uint32_t flags; >> +}; >> + >> +/* >> + * Component Locality Distance Information Table (CDIT) >> + */ >> +#define CDIT_OEMID_LENGTH 6 >> +#define CDIT_OEMTABLEID_LENGTH 8 >> + >> +struct cdit_header { >> + uint32_t signature; >> + uint32_t length; >> + uint8_t revision; >> + uint8_t checksum; >> + uint8_t oem_id[CDIT_OEMID_LENGTH]; >> + uint8_t oem_table_id[CDIT_OEMTABLEID_LENGTH]; >> + uint32_t oem_revision; >> + uint32_t creator_id; >> + uint32_t creator_revision; >> + uint32_t total_entries; >> + uint16_t num_domains; >> + uint8_t entry[1]; >> +}; >> + >> +#pragma pack() >> + >> +#endif /* KFD_CRAT_H_INCLUDED */ >> diff --git a/drivers/gpu/drm/radeon/amdkfd/kfd_device.c >> b/drivers/gpu/drm/radeon/amdkfd/kfd_device.c >> index dd63ce09..4138694 100644 >> --- a/drivers/gpu/drm/radeon/amdkfd/kfd_device.c >> +++ b/drivers/gpu/drm/radeon/amdkfd/kfd_device.c >> @@ -100,6 +100,9 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, >> { >> kfd->shared_resources = *gpu_resources; >> >> + if (kfd_topology_add_device(kfd) != 0) >> + return false; >> + >> kfd->init_complete = true; >> dev_info(kfd_device, "added device (%x:%x)\n", kfd->pdev->vendor, >> kfd->pdev->device); >> @@ -109,6 +112,10 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, >> >> void kgd2kfd_device_exit(struct kfd_dev *kfd) >> { >> + int err = kfd_topology_remove_device(kfd); >> + >> + BUG_ON(err != 0); >> + >> kfree(kfd); >> } >> >> diff --git a/drivers/gpu/drm/radeon/amdkfd/kfd_module.c >> b/drivers/gpu/drm/radeon/amdkfd/kfd_module.c >> index c7faac6..c51f981 100644 >> --- a/drivers/gpu/drm/radeon/amdkfd/kfd_module.c >> +++ b/drivers/gpu/drm/radeon/amdkfd/kfd_module.c >> @@ -73,16 +73,23 @@ static int __init kfd_module_init(void) >> if (err < 0) >> goto err_ioctl; >> >> + err = kfd_topology_init(); >> + if (err < 0) >> + goto err_topology; >> + >> dev_info(kfd_device, "Initialized module\n"); >> >> return 0; >> >> +err_topology: >> + kfd_chardev_exit(); >> err_ioctl: >> return err; >> } >> >> static void __exit kfd_module_exit(void) >> { >> + kfd_topology_shutdown(); >> kfd_chardev_exit(); >> dev_info(kfd_device, "Removed module\n"); >> } >> diff --git a/drivers/gpu/drm/radeon/amdkfd/kfd_priv.h >> b/drivers/gpu/drm/radeon/amdkfd/kfd_priv.h >> index 05e892f..b391e24 100644 >> --- a/drivers/gpu/drm/radeon/amdkfd/kfd_priv.h >> +++ b/drivers/gpu/drm/radeon/amdkfd/kfd_priv.h >> @@ -32,6 +32,14 @@ >> #include <linux/spinlock.h> >> #include "../radeon_kfd.h" >> >> +#define KFD_SYSFS_FILE_MODE 0444 >> + >> +/* GPU ID hash width in bits */ >> +#define KFD_GPU_ID_HASH_WIDTH 16 >> + >> +/* Macro for allocating structures */ >> +#define kfd_alloc_struct(ptr_to_struct) ((typeof(ptr_to_struct)) >> kzalloc(sizeof(*ptr_to_struct), GFP_KERNEL)) >> + >> struct kfd_device_info { >> const struct kfd_scheduler_class *scheduler_class; >> unsigned int max_pasid_bits; >> @@ -71,6 +79,15 @@ struct kfd_process { >> >> extern struct device *kfd_device; >> >> +/* Topology */ >> +int kfd_topology_init(void); >> +void kfd_topology_shutdown(void); >> +int kfd_topology_add_device(struct kfd_dev *gpu); >> +int kfd_topology_remove_device(struct kfd_dev *gpu); >> +struct kfd_dev *kfd_device_by_id(uint32_t gpu_id); >> +struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev); >> +struct kfd_dev *kfd_topology_enum_kfd_devices(uint8_t idx); >> + >> /* Interrupts */ >> void kgd2kfd_interrupt(struct kfd_dev *dev, const void *ih_ring_entry); >> >> diff --git a/drivers/gpu/drm/radeon/amdkfd/kfd_topology.c >> b/drivers/gpu/drm/radeon/amdkfd/kfd_topology.c >> new file mode 100644 >> index 0000000..30da4c3 >> --- /dev/null >> +++ b/drivers/gpu/drm/radeon/amdkfd/kfd_topology.c >> @@ -0,0 +1,1207 @@ >> +/* >> + * Copyright 2014 Advanced Micro Devices, Inc. >> + * >> + * Permission is hereby granted, free of charge, to any person obtaining a >> + * copy of this software and associated documentation files (the >> "Software"), >> + * to deal in the Software without restriction, including without limitation >> + * the rights to use, copy, modify, merge, publish, distribute, sublicense, >> + * and/or sell copies of the Software, and to permit persons to whom the >> + * Software is furnished to do so, subject to the following conditions: >> + * >> + * The above copyright notice and this permission notice shall be included >> in >> + * all copies or substantial portions of the Software. >> + * >> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS >> OR >> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, >> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL >> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR >> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, >> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR >> + * OTHER DEALINGS IN THE SOFTWARE. >> + */ >> + >> +#include <linux/types.h> >> +#include <linux/kernel.h> >> +#include <linux/pci.h> >> +#include <linux/errno.h> >> +#include <linux/acpi.h> >> +#include <linux/hash.h> >> +#include <linux/cpufreq.h> >> + >> +#include "kfd_priv.h" >> +#include "kfd_crat.h" >> +#include "kfd_topology.h" >> + >> +static struct list_head topology_device_list; >> +static int topology_crat_parsed; >> +static struct kfd_system_properties sys_props; >> + >> +static DECLARE_RWSEM(topology_lock); >> + >> +struct kfd_dev *kfd_device_by_id(uint32_t gpu_id) >> +{ >> + struct kfd_topology_device *top_dev; >> + struct kfd_dev *device = NULL; >> + >> + down_read(&topology_lock); >> + >> + list_for_each_entry(top_dev, &topology_device_list, list) >> + if (top_dev->gpu_id == gpu_id) { >> + device = top_dev->gpu; >> + break; >> + } >> + >> + up_read(&topology_lock); >> + >> + return device; >> +} >> + >> +struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev) >> +{ >> + struct kfd_topology_device *top_dev; >> + struct kfd_dev *device = NULL; >> + >> + down_read(&topology_lock); >> + >> + list_for_each_entry(top_dev, &topology_device_list, list) >> + if (top_dev->gpu->pdev == pdev) { >> + device = top_dev->gpu; >> + break; >> + } >> + >> + up_read(&topology_lock); >> + >> + 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 != 0) >> + 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) >> +{ >> + BUG_ON(!dev); >> + BUG_ON(!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) >> +{ >> + BUG_ON(!dev); >> + BUG_ON(!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: simds=%d id_base=%d\n", cu->num_simd_cores, >> + 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; >> + >> + BUG_ON(!cu); >> + >> + 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; >> + >> + BUG_ON(!mem); >> + >> + 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 == 0) >> + 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; >> + >> + BUG_ON(!cache); >> + >> + 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 == 0) >> + 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; >> + >> + BUG_ON(!iolink); >> + >> + 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 == 0) >> + 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; >> + >> + BUG_ON(!sub_type_hdr); >> + >> + 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; >> + struct kfd_cache_properties *cache; >> + struct kfd_iolink_properties *iolink; >> + >> + BUG_ON(!dev); >> + >> + list_del(&dev->list); >> + >> + while (dev->mem_props.next != &dev->mem_props) { >> + mem = container_of(dev->mem_props.next, >> + struct kfd_mem_properties, list); >> + list_del(&mem->list); >> + kfree(mem); >> + } >> + >> + while (dev->cache_props.next != &dev->cache_props) { >> + cache = container_of(dev->cache_props.next, >> + struct kfd_cache_properties, list); >> + list_del(&cache->list); >> + kfree(cache); >> + } >> + >> + while (dev->io_link_props.next != &dev->io_link_props) { >> + iolink = container_of(dev->io_link_props.next, >> + struct kfd_iolink_properties, list); >> + list_del(&iolink->list); >> + kfree(iolink); >> + } >> + >> + kfree(dev); >> + >> + sys_props.num_devices--; >> +} >> + >> +static void kfd_release_live_view(void) >> +{ >> + struct kfd_topology_device *dev; >> + >> + while (topology_device_list.next != &topology_device_list) { >> + dev = container_of(topology_device_list.next, >> + struct kfd_topology_device, list); >> + kfd_release_topology_device(dev); >> +} >> + >> + memset(&sys_props, 0, sizeof(sys_props)); >> +} >> + >> +static struct kfd_topology_device *kfd_create_topology_device(void) >> +{ >> + struct kfd_topology_device *dev; >> + >> + dev = kfd_alloc_struct(dev); >> + if (dev == 0) { >> + pr_err("No memory to allocate a topology device"); >> + return 0; >> + } >> + >> + INIT_LIST_HEAD(&dev->mem_props); >> + INIT_LIST_HEAD(&dev->cache_props); >> + INIT_LIST_HEAD(&dev->io_link_props); >> + >> + list_add_tail(&dev->list, &topology_device_list); >> + sys_props.num_devices++; >> + >> + 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__) >> +#define sysfs_show_32bit_prop(buffer, name, value) \ >> + sysfs_show_gen_prop(buffer, "%s %u\n", name, value) >> +#define sysfs_show_64bit_prop(buffer, name, value) \ >> + sysfs_show_gen_prop(buffer, "%s %llu\n", name, value) >> +#define sysfs_show_32bit_val(buffer, value) \ >> + sysfs_show_gen_prop(buffer, "%u\n", value) >> +#define sysfs_show_str_val(buffer, value) \ >> + sysfs_show_gen_prop(buffer, "%s\n", value) >> + >> +static ssize_t sysprops_show(struct kobject *kobj, struct attribute *attr, >> + char *buffer) >> +{ >> + ssize_t ret; >> + >> + /* Making sure that the buffer is an empty string */ >> + buffer[0] = 0; >> + >> + if (attr == &sys_props.attr_genid) { >> + ret = sysfs_show_32bit_val(buffer, sys_props.generation_count); >> + } else if (attr == &sys_props.attr_props) { >> + sysfs_show_64bit_prop(buffer, "platform_oem", >> + sys_props.platform_oem); >> + sysfs_show_64bit_prop(buffer, "platform_id", >> + sys_props.platform_id); >> + ret = sysfs_show_64bit_prop(buffer, "platform_rev", >> + sys_props.platform_rev); >> + } else { >> + ret = -EINVAL; >> + } >> + >> + return ret; >> +} >> + >> +static const struct sysfs_ops sysprops_ops = { >> + .show = sysprops_show, >> +}; >> + >> +static struct kobj_type sysprops_type = { >> + .sysfs_ops = &sysprops_ops, >> +}; >> + >> +static ssize_t iolink_show(struct kobject *kobj, struct attribute *attr, >> + char *buffer) >> +{ >> + ssize_t ret; >> + struct kfd_iolink_properties *iolink; >> + >> + /* Making sure that the buffer is an empty string */ >> + buffer[0] = 0; >> + >> + iolink = container_of(attr, struct kfd_iolink_properties, attr); >> + sysfs_show_32bit_prop(buffer, "type", iolink->iolink_type); >> + sysfs_show_32bit_prop(buffer, "version_major", iolink->ver_maj); >> + sysfs_show_32bit_prop(buffer, "version_minor", iolink->ver_min); >> + sysfs_show_32bit_prop(buffer, "node_from", iolink->node_from); >> + sysfs_show_32bit_prop(buffer, "node_to", iolink->node_to); >> + sysfs_show_32bit_prop(buffer, "weight", iolink->weight); >> + sysfs_show_32bit_prop(buffer, "min_latency", iolink->min_latency); >> + sysfs_show_32bit_prop(buffer, "max_latency", iolink->max_latency); >> + sysfs_show_32bit_prop(buffer, "min_bandwidth", iolink->min_bandwidth); >> + sysfs_show_32bit_prop(buffer, "max_bandwidth", iolink->max_bandwidth); >> + sysfs_show_32bit_prop(buffer, "recommended_transfer_size", >> + iolink->rec_transfer_size); >> + ret = sysfs_show_32bit_prop(buffer, "flags", iolink->flags); >> + >> + return ret; >> +} >> + >> +static const struct sysfs_ops iolink_ops = { >> + .show = iolink_show, >> +}; >> + >> +static struct kobj_type iolink_type = { >> + .sysfs_ops = &iolink_ops, >> +}; >> + >> +static ssize_t mem_show(struct kobject *kobj, struct attribute *attr, >> + char *buffer) >> +{ >> + ssize_t ret; >> + struct kfd_mem_properties *mem; >> + >> + /* Making sure that the buffer is an empty string */ >> + buffer[0] = 0; >> + >> + mem = container_of(attr, struct kfd_mem_properties, attr); >> + sysfs_show_32bit_prop(buffer, "heap_type", mem->heap_type); >> + sysfs_show_64bit_prop(buffer, "size_in_bytes", mem->size_in_bytes); >> + sysfs_show_32bit_prop(buffer, "flags", mem->flags); >> + sysfs_show_32bit_prop(buffer, "width", mem->width); >> + ret = sysfs_show_32bit_prop(buffer, "mem_clk_max", mem->mem_clk_max); >> + >> + return ret; >> +} >> + >> +static const struct sysfs_ops mem_ops = { >> + .show = mem_show, >> +}; >> + >> +static struct kobj_type mem_type = { >> + .sysfs_ops = &mem_ops, >> +}; >> + >> +static ssize_t kfd_cache_show(struct kobject *kobj, struct attribute *attr, >> + char *buffer) >> +{ >> + ssize_t ret; >> + uint32_t i; >> + struct kfd_cache_properties *cache; >> + >> + /* Making sure that the buffer is an empty string */ >> + buffer[0] = 0; >> + >> + cache = container_of(attr, struct kfd_cache_properties, attr); >> + sysfs_show_32bit_prop(buffer, "processor_id_low", >> + cache->processor_id_low); >> + sysfs_show_32bit_prop(buffer, "level", cache->cache_level); >> + sysfs_show_32bit_prop(buffer, "size", cache->cache_size); >> + sysfs_show_32bit_prop(buffer, "cache_line_size", cache->cacheline_size); >> + sysfs_show_32bit_prop(buffer, "cache_lines_per_tag", >> + cache->cachelines_per_tag); >> + sysfs_show_32bit_prop(buffer, "association", cache->cache_assoc); >> + sysfs_show_32bit_prop(buffer, "latency", cache->cache_latency); >> + sysfs_show_32bit_prop(buffer, "type", cache->cache_type); >> + snprintf(buffer, PAGE_SIZE, "%ssibling_map ", buffer); >> + for (i = 0; i < KFD_TOPOLOGY_CPU_SIBLINGS; i++) >> + ret = snprintf(buffer, PAGE_SIZE, "%s%d%s", >> + buffer, cache->sibling_map[i], >> + (i == KFD_TOPOLOGY_CPU_SIBLINGS-1) ? >> + "\n" : ","); >> + >> + return ret; >> +} >> + >> +static const struct sysfs_ops cache_ops = { >> + .show = kfd_cache_show, >> +}; >> + >> +static struct kobj_type cache_type = { >> + .sysfs_ops = &cache_ops, >> +}; >> + >> +static ssize_t node_show(struct kobject *kobj, struct attribute *attr, >> + char *buffer) >> +{ >> + ssize_t ret; >> + struct kfd_topology_device *dev; >> + char public_name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE]; >> + uint32_t i; >> + >> + /* Making sure that the buffer is an empty string */ >> + buffer[0] = 0; >> + >> + if (strcmp(attr->name, "gpu_id") == 0) { >> + dev = container_of(attr, struct kfd_topology_device, >> + attr_gpuid); >> + ret = sysfs_show_32bit_val(buffer, dev->gpu_id); >> + } else if (strcmp(attr->name, "name") == 0) { >> + dev = container_of(attr, struct kfd_topology_device, >> + attr_name); >> + for (i = 0; i < KFD_TOPOLOGY_PUBLIC_NAME_SIZE; i++) { >> + public_name[i] = >> + (char)dev->node_props.marketing_name[i]; >> + if (dev->node_props.marketing_name[i] == 0) >> + break; >> + } >> + public_name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE-1] = 0x0; >> + ret = sysfs_show_str_val(buffer, public_name); >> + } else { >> + dev = container_of(attr, struct kfd_topology_device, >> + attr_props); >> + sysfs_show_32bit_prop(buffer, "cpu_cores_count", >> + dev->node_props.cpu_cores_count); >> + sysfs_show_32bit_prop(buffer, "simd_count", >> + dev->node_props.simd_count); >> + sysfs_show_32bit_prop(buffer, "mem_banks_count", >> + dev->node_props.mem_banks_count); >> + sysfs_show_32bit_prop(buffer, "caches_count", >> + dev->node_props.caches_count); >> + sysfs_show_32bit_prop(buffer, "io_links_count", >> + dev->node_props.io_links_count); >> + sysfs_show_32bit_prop(buffer, "cpu_core_id_base", >> + dev->node_props.cpu_core_id_base); >> + sysfs_show_32bit_prop(buffer, "simd_id_base", >> + dev->node_props.simd_id_base); >> + sysfs_show_32bit_prop(buffer, "capability", >> + dev->node_props.capability); >> + sysfs_show_32bit_prop(buffer, "max_waves_per_simd", >> + dev->node_props.max_waves_per_simd); >> + sysfs_show_32bit_prop(buffer, "lds_size_in_kb", >> + dev->node_props.lds_size_in_kb); >> + sysfs_show_32bit_prop(buffer, "gds_size_in_kb", >> + dev->node_props.gds_size_in_kb); >> + sysfs_show_32bit_prop(buffer, "wave_front_size", >> + dev->node_props.wave_front_size); >> + sysfs_show_32bit_prop(buffer, "array_count", >> + dev->node_props.array_count); >> + sysfs_show_32bit_prop(buffer, "simd_arrays_per_engine", >> + dev->node_props.simd_arrays_per_engine); >> + sysfs_show_32bit_prop(buffer, "cu_per_simd_array", >> + dev->node_props.cu_per_simd_array); >> + sysfs_show_32bit_prop(buffer, "simd_per_cu", >> + dev->node_props.simd_per_cu); >> + sysfs_show_32bit_prop(buffer, "max_slots_scratch_cu", >> + dev->node_props.max_slots_scratch_cu); >> + sysfs_show_32bit_prop(buffer, "engine_id", >> + dev->node_props.engine_id); >> + sysfs_show_32bit_prop(buffer, "vendor_id", >> + dev->node_props.vendor_id); >> + sysfs_show_32bit_prop(buffer, "device_id", >> + dev->node_props.device_id); >> + sysfs_show_32bit_prop(buffer, "location_id", >> + dev->node_props.location_id); >> + sysfs_show_32bit_prop(buffer, "max_engine_clk_fcompute", >> + kfd2kgd->get_max_engine_clock_in_mhz( >> + dev->gpu->kgd)); >> + sysfs_show_64bit_prop(buffer, "local_mem_size", >> + kfd2kgd->get_vmem_size(dev->gpu->kgd)); >> + ret = sysfs_show_32bit_prop(buffer, "max_engine_clk_ccompute", >> + cpufreq_quick_get_max(0)/1000); >> + } >> + >> + return ret; >> +} >> + >> +static const struct sysfs_ops node_ops = { >> + .show = node_show, >> +}; >> + >> +static struct kobj_type node_type = { >> + .sysfs_ops = &node_ops, >> +}; >> + >> +static void kfd_remove_sysfs_file(struct kobject *kobj, struct attribute >> *attr) >> +{ >> + sysfs_remove_file(kobj, attr); >> + kobject_del(kobj); >> + kobject_put(kobj); >> +} >> + >> +static void kfd_remove_sysfs_node_entry(struct kfd_topology_device *dev) >> +{ >> + struct kfd_iolink_properties *iolink; >> + struct kfd_cache_properties *cache; >> + struct kfd_mem_properties *mem; >> + >> + BUG_ON(!dev); >> + >> + if (dev->kobj_iolink) { >> + list_for_each_entry(iolink, &dev->io_link_props, list) >> + if (iolink->kobj) { >> + kfd_remove_sysfs_file(iolink->kobj, >> &iolink->attr); >> + iolink->kobj = 0; >> + } >> + kobject_del(dev->kobj_iolink); >> + kobject_put(dev->kobj_iolink); >> + dev->kobj_iolink = 0; >> + } >> + >> + if (dev->kobj_cache) { >> + list_for_each_entry(cache, &dev->cache_props, list) >> + if (cache->kobj) { >> + kfd_remove_sysfs_file(cache->kobj, >> &cache->attr); >> + cache->kobj = 0; >> + } >> + kobject_del(dev->kobj_cache); >> + kobject_put(dev->kobj_cache); >> + dev->kobj_cache = 0; >> + } >> + >> + if (dev->kobj_mem) { >> + list_for_each_entry(mem, &dev->mem_props, list) >> + if (mem->kobj) { >> + kfd_remove_sysfs_file(mem->kobj, &mem->attr); >> + mem->kobj = 0; >> + } >> + kobject_del(dev->kobj_mem); >> + kobject_put(dev->kobj_mem); >> + dev->kobj_mem = 0; >> + } >> + >> + if (dev->kobj_node) { >> + sysfs_remove_file(dev->kobj_node, &dev->attr_gpuid); >> + sysfs_remove_file(dev->kobj_node, &dev->attr_name); >> + sysfs_remove_file(dev->kobj_node, &dev->attr_props); >> + kobject_del(dev->kobj_node); >> + kobject_put(dev->kobj_node); >> + dev->kobj_node = 0; >> + } >> +} >> + >> +static int kfd_build_sysfs_node_entry(struct kfd_topology_device *dev, >> + uint32_t id) >> +{ >> + struct kfd_iolink_properties *iolink; >> + struct kfd_cache_properties *cache; >> + struct kfd_mem_properties *mem; >> + int ret; >> + uint32_t i; >> + >> + BUG_ON(!dev); >> + >> + /* >> + * Creating the sysfs folders >> + */ >> + BUG_ON(dev->kobj_node); >> + dev->kobj_node = kfd_alloc_struct(dev->kobj_node); >> + if (!dev->kobj_node) >> + return -ENOMEM; >> + >> + ret = kobject_init_and_add(dev->kobj_node, &node_type, >> + sys_props.kobj_nodes, "%d", id); >> + if (ret < 0) >> + return ret; >> + >> + dev->kobj_mem = kobject_create_and_add("mem_banks", dev->kobj_node); >> + if (!dev->kobj_mem) >> + return -ENOMEM; >> + >> + dev->kobj_cache = kobject_create_and_add("caches", dev->kobj_node); >> + if (!dev->kobj_cache) >> + return -ENOMEM; >> + >> + dev->kobj_iolink = kobject_create_and_add("io_links", dev->kobj_node); >> + if (!dev->kobj_iolink) >> + return -ENOMEM; >> + >> + /* >> + * Creating sysfs files for node properties >> + */ >> + dev->attr_gpuid.name = "gpu_id"; >> + dev->attr_gpuid.mode = KFD_SYSFS_FILE_MODE; >> + sysfs_attr_init(&dev->attr_gpuid); >> + dev->attr_name.name = "name"; >> + dev->attr_name.mode = KFD_SYSFS_FILE_MODE; >> + sysfs_attr_init(&dev->attr_name); >> + dev->attr_props.name = "properties"; >> + dev->attr_props.mode = KFD_SYSFS_FILE_MODE; >> + sysfs_attr_init(&dev->attr_props); >> + ret = sysfs_create_file(dev->kobj_node, &dev->attr_gpuid); >> + if (ret < 0) >> + return ret; >> + ret = sysfs_create_file(dev->kobj_node, &dev->attr_name); >> + if (ret < 0) >> + return ret; >> + ret = sysfs_create_file(dev->kobj_node, &dev->attr_props); >> + if (ret < 0) >> + return ret; >> + >> + i = 0; >> + list_for_each_entry(mem, &dev->mem_props, list) { >> + mem->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL); >> + if (!mem->kobj) >> + return -ENOMEM; >> + ret = kobject_init_and_add(mem->kobj, &mem_type, >> + dev->kobj_mem, "%d", i); >> + if (ret < 0) >> + return ret; >> + >> + mem->attr.name = "properties"; >> + mem->attr.mode = KFD_SYSFS_FILE_MODE; >> + sysfs_attr_init(&mem->attr); >> + ret = sysfs_create_file(mem->kobj, &mem->attr); >> + if (ret < 0) >> + return ret; >> + i++; >> + } >> + >> + i = 0; >> + list_for_each_entry(cache, &dev->cache_props, list) { >> + cache->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL); >> + if (!cache->kobj) >> + return -ENOMEM; >> + ret = kobject_init_and_add(cache->kobj, &cache_type, >> + dev->kobj_cache, "%d", i); >> + if (ret < 0) >> + return ret; >> + >> + cache->attr.name = "properties"; >> + cache->attr.mode = KFD_SYSFS_FILE_MODE; >> + sysfs_attr_init(&cache->attr); >> + ret = sysfs_create_file(cache->kobj, &cache->attr); >> + if (ret < 0) >> + return ret; >> + i++; >> + } >> + >> + i = 0; >> + list_for_each_entry(iolink, &dev->io_link_props, list) { >> + iolink->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL); >> + if (!iolink->kobj) >> + return -ENOMEM; >> + ret = kobject_init_and_add(iolink->kobj, &iolink_type, >> + dev->kobj_iolink, "%d", i); >> + if (ret < 0) >> + return ret; >> + >> + iolink->attr.name = "properties"; >> + iolink->attr.mode = KFD_SYSFS_FILE_MODE; >> + sysfs_attr_init(&iolink->attr); >> + ret = sysfs_create_file(iolink->kobj, &iolink->attr); >> + if (ret < 0) >> + return ret; >> + i++; >> +} >> + >> + return 0; >> +} >> + >> +static int kfd_build_sysfs_node_tree(void) >> +{ >> + struct kfd_topology_device *dev; >> + int ret; >> + uint32_t i = 0; >> + >> + list_for_each_entry(dev, &topology_device_list, list) { >> + ret = kfd_build_sysfs_node_entry(dev, 0); >> + if (ret < 0) >> + return ret; >> + i++; >> + } >> + >> + return 0; >> +} >> + >> +static void kfd_remove_sysfs_node_tree(void) >> +{ >> + struct kfd_topology_device *dev; >> + >> + list_for_each_entry(dev, &topology_device_list, list) >> + kfd_remove_sysfs_node_entry(dev); >> +} >> + >> +static int kfd_topology_update_sysfs(void) >> +{ >> + int ret; >> + >> + pr_info("Creating topology SYSFS entries\n"); >> + if (sys_props.kobj_topology == 0) { >> + sys_props.kobj_topology = >> kfd_alloc_struct(sys_props.kobj_topology); >> + if (!sys_props.kobj_topology) >> + return -ENOMEM; >> + >> + ret = kobject_init_and_add(sys_props.kobj_topology, >> + &sysprops_type, &kfd_device->kobj, >> + "topology"); >> + if (ret < 0) >> + return ret; >> + >> + sys_props.kobj_nodes = kobject_create_and_add("nodes", >> + sys_props.kobj_topology); >> + if (!sys_props.kobj_nodes) >> + return -ENOMEM; >> + >> + sys_props.attr_genid.name = "generation_id"; >> + sys_props.attr_genid.mode = KFD_SYSFS_FILE_MODE; >> + sysfs_attr_init(&sys_props.attr_genid); >> + ret = sysfs_create_file(sys_props.kobj_topology, >> + &sys_props.attr_genid); >> + if (ret < 0) >> + return ret; >> + >> + sys_props.attr_props.name = "system_properties"; >> + sys_props.attr_props.mode = KFD_SYSFS_FILE_MODE; >> + sysfs_attr_init(&sys_props.attr_props); >> + ret = sysfs_create_file(sys_props.kobj_topology, >> + &sys_props.attr_props); >> + if (ret < 0) >> + return ret; >> + } >> + >> + kfd_remove_sysfs_node_tree(); >> + >> + return kfd_build_sysfs_node_tree(); >> +} >> + >> +static void kfd_topology_release_sysfs(void) >> +{ >> + kfd_remove_sysfs_node_tree(); >> + if (sys_props.kobj_topology) { >> + sysfs_remove_file(sys_props.kobj_topology, >> + &sys_props.attr_genid); >> + sysfs_remove_file(sys_props.kobj_topology, >> + &sys_props.attr_props); >> + if (sys_props.kobj_nodes) { >> + kobject_del(sys_props.kobj_nodes); >> + kobject_put(sys_props.kobj_nodes); >> + sys_props.kobj_nodes = 0; >> + } >> + kobject_del(sys_props.kobj_topology); >> + kobject_put(sys_props.kobj_topology); >> + sys_props.kobj_topology = 0; >> + } >> +} >> + >> +int kfd_topology_init(void) >> +{ >> + void *crat_image = 0; >> + size_t image_size = 0; >> + int ret; >> + >> + /* >> + * Initialize the head for the topology device list >> + */ >> + INIT_LIST_HEAD(&topology_device_list); >> + init_rwsem(&topology_lock); >> + topology_crat_parsed = 0; >> + >> + memset(&sys_props, 0, sizeof(sys_props)); >> + >> + /* >> + * Get the CRAT image from the ACPI >> + */ >> + ret = kfd_topology_get_crat_acpi(crat_image, &image_size); >> + if (ret == 0 && image_size > 0) { >> + pr_info("Found CRAT image with size=%zd\n", image_size); >> + crat_image = kmalloc(image_size, GFP_KERNEL); >> + if (!crat_image) { >> + ret = -ENOMEM; >> + pr_err("No memory for allocating CRAT image\n"); >> + goto err; >> + } >> + ret = kfd_topology_get_crat_acpi(crat_image, &image_size); >> + >> + if (ret == 0) { >> + down_write(&topology_lock); >> + ret = kfd_parse_crat_table(crat_image); >> + if (ret == 0) >> + ret = kfd_topology_update_sysfs(); >> + up_write(&topology_lock); >> + } else { >> + pr_err("Couldn't get CRAT table size from ACPI\n"); >> + } >> + kfree(crat_image); >> + } else if (ret == -ENODATA) { >> + ret = 0; >> + } else { >> + pr_err("Couldn't get CRAT table size from ACPI\n"); >> + } >> + >> +err: >> + pr_info("Finished initializing topology ret=%d\n", ret); >> + return ret; >> +} >> + >> +void kfd_topology_shutdown(void) >> +{ >> + kfd_topology_release_sysfs(); >> + kfd_release_live_view(); >> +} >> + >> +static void kfd_debug_print_topology(void) >> +{ >> + struct kfd_topology_device *dev; >> + uint32_t i = 0; >> + >> + pr_info("DEBUG PRINT OF TOPOLOGY:"); >> + list_for_each_entry(dev, &topology_device_list, list) { >> + pr_info("Node: %d\n", i); >> + pr_info("\tGPU assigned: %s\n", (dev->gpu ? "yes" : "no")); >> + pr_info("\tCPU count: %d\n", dev->node_props.cpu_cores_count); >> + pr_info("\tSIMD count: %d", dev->node_props.simd_count); >> + i++; >> + } >> +} >> + >> +static uint32_t kfd_generate_gpu_id(struct kfd_dev *gpu) >> +{ >> + uint32_t hashout; >> + uint32_t buf[7]; >> + int i; >> + >> + if (!gpu) >> + return 0; >> + >> + buf[0] = gpu->pdev->devfn; >> + buf[1] = gpu->pdev->subsystem_vendor; >> + buf[2] = gpu->pdev->subsystem_device; >> + buf[3] = gpu->pdev->device; >> + buf[4] = gpu->pdev->bus->number; >> + buf[5] = (uint32_t)(kfd2kgd->get_vmem_size(gpu->kgd) & 0xffffffff); >> + buf[6] = (uint32_t)(kfd2kgd->get_vmem_size(gpu->kgd) >> 32); >> + >> + for (i = 0, hashout = 0; i < 7; i++) >> + hashout ^= hash_32(buf[i], KFD_GPU_ID_HASH_WIDTH); >> + >> + return hashout; >> +} >> + >> +static struct kfd_topology_device *kfd_assign_gpu(struct kfd_dev *gpu) >> +{ >> + struct kfd_topology_device *dev; >> + struct kfd_topology_device *out_dev = 0; >> + >> + BUG_ON(!gpu); >> + >> + list_for_each_entry(dev, &topology_device_list, list) >> + if (dev->gpu == 0 && dev->node_props.simd_count > 0) { >> + dev->gpu = gpu; >> + out_dev = dev; >> + break; >> + } >> + >> + return out_dev; >> +} >> + >> +static void kfd_notify_gpu_change(uint32_t gpu_id, int arrival) >> +{ >> + /* >> + * TODO: Generate an event for thunk about the arrival/removal >> + * of the GPU >> + */ >> +} >> + >> +int kfd_topology_add_device(struct kfd_dev *gpu) >> +{ >> + uint32_t gpu_id; >> + struct kfd_topology_device *dev; >> + int res; >> + >> + BUG_ON(!gpu); >> + >> + gpu_id = kfd_generate_gpu_id(gpu); >> + >> + pr_debug("kfd: Adding new GPU (ID: 0x%x) to topology\n", gpu_id); >> + >> + down_write(&topology_lock); >> + /* >> + * Try to assign the GPU to existing topology device (generated from >> + * CRAT table >> + */ >> + dev = kfd_assign_gpu(gpu); >> + if (!dev) { >> + pr_info("GPU was not found in the current topology. >> Extending.\n"); >> + kfd_debug_print_topology(); >> + dev = kfd_create_topology_device(); >> + if (!dev) { >> + res = -ENOMEM; >> + goto err; >> + } >> + dev->gpu = gpu; >> + >> + /* >> + * TODO: Make a call to retrieve topology information from the >> + * GPU vBIOS >> + */ >> + >> + /* >> + * Update the SYSFS tree, since we added another topology device >> + */ >> + if (kfd_topology_update_sysfs() < 0) >> + kfd_topology_release_sysfs(); >> + >> + } >> + >> + dev->gpu_id = gpu_id; >> + gpu->id = gpu_id; >> + dev->node_props.vendor_id = gpu->pdev->vendor; >> + dev->node_props.device_id = gpu->pdev->device; >> + dev->node_props.location_id = (gpu->pdev->bus->number << 24) + >> + (gpu->pdev->devfn & 0xffffff); >> + /* >> + * TODO: Retrieve max engine clock values from KGD >> + */ >> + >> + res = 0; >> + >> +err: >> + up_write(&topology_lock); >> + >> + if (res == 0) >> + kfd_notify_gpu_change(gpu_id, 1); >> + >> + return res; >> +} >> + >> +int kfd_topology_remove_device(struct kfd_dev *gpu) >> +{ >> + struct kfd_topology_device *dev; >> + uint32_t gpu_id; >> + int res = -ENODEV; >> + >> + BUG_ON(!gpu); >> + >> + down_write(&topology_lock); >> + >> + list_for_each_entry(dev, &topology_device_list, list) >> + if (dev->gpu == gpu) { >> + gpu_id = dev->gpu_id; >> + kfd_remove_sysfs_node_entry(dev); >> + kfd_release_topology_device(dev); >> + res = 0; >> + if (kfd_topology_update_sysfs() < 0) >> + kfd_topology_release_sysfs(); >> + break; >> + } >> + >> + up_write(&topology_lock); >> + >> + if (res == 0) >> + kfd_notify_gpu_change(gpu_id, 0); >> + >> + return res; >> +} >> + >> +/* >> + * When idx is out of bounds, the function will return NULL >> + */ >> +struct kfd_dev *kfd_topology_enum_kfd_devices(uint8_t idx) >> +{ >> + >> + struct kfd_topology_device *top_dev; >> + struct kfd_dev *device = NULL; >> + uint8_t device_idx = 0; >> + >> + down_read(&topology_lock); >> + >> + list_for_each_entry(top_dev, &topology_device_list, list) { >> + if (device_idx == idx) { >> + device = top_dev->gpu; >> + break; >> + } >> + >> + device_idx++; >> + } >> + >> + up_read(&topology_lock); >> + >> + return device; >> + >> +} >> diff --git a/drivers/gpu/drm/radeon/amdkfd/kfd_topology.h >> b/drivers/gpu/drm/radeon/amdkfd/kfd_topology.h >> new file mode 100644 >> index 0000000..989624b >> --- /dev/null >> +++ b/drivers/gpu/drm/radeon/amdkfd/kfd_topology.h >> @@ -0,0 +1,168 @@ >> +/* >> + * Copyright 2014 Advanced Micro Devices, Inc. >> + * >> + * Permission is hereby granted, free of charge, to any person obtaining a >> + * copy of this software and associated documentation files (the >> "Software"), >> + * to deal in the Software without restriction, including without limitation >> + * the rights to use, copy, modify, merge, publish, distribute, sublicense, >> + * and/or sell copies of the Software, and to permit persons to whom the >> + * Software is furnished to do so, subject to the following conditions: >> + * >> + * The above copyright notice and this permission notice shall be included >> in >> + * all copies or substantial portions of the Software. >> + * >> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS >> OR >> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, >> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL >> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR >> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, >> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR >> + * OTHER DEALINGS IN THE SOFTWARE. >> + */ >> + >> +#ifndef __KFD_TOPOLOGY_H__ >> +#define __KFD_TOPOLOGY_H__ >> + >> +#include <linux/types.h> >> +#include <linux/list.h> >> +#include "kfd_priv.h" >> + >> +#define KFD_TOPOLOGY_PUBLIC_NAME_SIZE 128 >> + >> +#define HSA_CAP_HOT_PLUGGABLE 0x00000001 >> +#define HSA_CAP_ATS_PRESENT 0x00000002 >> +#define HSA_CAP_SHARED_WITH_GRAPHICS 0x00000004 >> +#define HSA_CAP_QUEUE_SIZE_POW2 0x00000008 >> +#define HSA_CAP_QUEUE_SIZE_32BIT 0x00000010 >> +#define HSA_CAP_QUEUE_IDLE_EVENT 0x00000020 >> +#define HSA_CAP_VA_LIMIT 0x00000040 >> +#define HSA_CAP_WATCH_POINTS_SUPPORTED 0x00000080 >> +#define HSA_CAP_WATCH_POINTS_TOTALBITS_MASK 0x00000f00 >> +#define HSA_CAP_WATCH_POINTS_TOTALBITS_SHIFT 8 >> +#define HSA_CAP_RESERVED 0xfffff000 >> + >> +struct kfd_node_properties { >> + uint32_t cpu_cores_count; >> + uint32_t simd_count; >> + uint32_t mem_banks_count; >> + uint32_t caches_count; >> + uint32_t io_links_count; >> + uint32_t cpu_core_id_base; >> + uint32_t simd_id_base; >> + uint32_t capability; >> + uint32_t max_waves_per_simd; >> + uint32_t lds_size_in_kb; >> + uint32_t gds_size_in_kb; >> + uint32_t wave_front_size; >> + uint32_t array_count; >> + uint32_t simd_arrays_per_engine; >> + uint32_t cu_per_simd_array; >> + uint32_t simd_per_cu; >> + uint32_t max_slots_scratch_cu; >> + uint32_t engine_id; >> + uint32_t vendor_id; >> + uint32_t device_id; >> + uint32_t location_id; >> + uint32_t max_engine_clk_fcompute; >> + uint32_t max_engine_clk_ccompute; >> + uint16_t marketing_name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE]; >> +}; >> + >> +#define HSA_MEM_HEAP_TYPE_SYSTEM 0 >> +#define HSA_MEM_HEAP_TYPE_FB_PUBLIC 1 >> +#define HSA_MEM_HEAP_TYPE_FB_PRIVATE 2 >> +#define HSA_MEM_HEAP_TYPE_GPU_GDS 3 >> +#define HSA_MEM_HEAP_TYPE_GPU_LDS 4 >> +#define HSA_MEM_HEAP_TYPE_GPU_SCRATCH 5 >> + >> +#define HSA_MEM_FLAGS_HOT_PLUGGABLE 0x00000001 >> +#define HSA_MEM_FLAGS_NON_VOLATILE 0x00000002 >> +#define HSA_MEM_FLAGS_RESERVED 0xfffffffc >> + >> +struct kfd_mem_properties { >> + struct list_head list; >> + uint32_t heap_type; >> + uint64_t size_in_bytes; >> + uint32_t flags; >> + uint32_t width; >> + uint32_t mem_clk_max; >> + struct kobject *kobj; >> + struct attribute attr; >> +}; >> + >> +#define KFD_TOPOLOGY_CPU_SIBLINGS 256 >> + >> +#define HSA_CACHE_TYPE_DATA 0x00000001 >> +#define HSA_CACHE_TYPE_INSTRUCTION 0x00000002 >> +#define HSA_CACHE_TYPE_CPU 0x00000004 >> +#define HSA_CACHE_TYPE_HSACU 0x00000008 >> +#define HSA_CACHE_TYPE_RESERVED 0xfffffff0 >> + >> +struct kfd_cache_properties { >> + struct list_head list; >> + uint32_t processor_id_low; >> + uint32_t cache_level; >> + uint32_t cache_size; >> + uint32_t cacheline_size; >> + uint32_t cachelines_per_tag; >> + uint32_t cache_assoc; >> + uint32_t cache_latency; >> + uint32_t cache_type; >> + uint8_t sibling_map[KFD_TOPOLOGY_CPU_SIBLINGS]; >> + struct kobject *kobj; >> + struct attribute attr; >> +}; >> + >> +struct kfd_iolink_properties { >> + struct list_head list; >> + uint32_t iolink_type; >> + uint32_t ver_maj; >> + uint32_t ver_min; >> + uint32_t node_from; >> + uint32_t node_to; >> + uint32_t weight; >> + uint32_t min_latency; >> + uint32_t max_latency; >> + uint32_t min_bandwidth; >> + uint32_t max_bandwidth; >> + uint32_t rec_transfer_size; >> + uint32_t flags; >> + struct kobject *kobj; >> + struct attribute attr; >> +}; >> + >> +struct kfd_topology_device { >> + struct list_head list; >> + uint32_t gpu_id; >> + struct kfd_node_properties node_props; >> + uint32_t mem_bank_count; >> + struct list_head mem_props; >> + uint32_t cache_count; >> + struct list_head cache_props; >> + uint32_t io_link_count; >> + struct list_head io_link_props; >> + struct kfd_dev *gpu; >> + struct kobject *kobj_node; >> + struct kobject *kobj_mem; >> + struct kobject *kobj_cache; >> + struct kobject *kobj_iolink; >> + struct attribute attr_gpuid; >> + struct attribute attr_name; >> + struct attribute attr_props; >> +}; >> + >> +struct kfd_system_properties { >> + uint32_t num_devices; /* Number of H-NUMA nodes */ >> + uint32_t generation_count; >> + uint64_t platform_oem; >> + uint64_t platform_id; >> + uint64_t platform_rev; >> + struct kobject *kobj_topology; >> + struct kobject *kobj_nodes; >> + struct attribute attr_genid; >> + struct attribute attr_props; >> +}; >> + >> + >> + >> +#endif /* __KFD_TOPOLOGY_H__ */ >> -- >> 1.9.1 >>