diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu')
29 files changed, 2017 insertions, 166 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile index 616dfd4a1398..908360584e4d 100644 --- a/drivers/gpu/drm/amd/amdgpu/Makefile +++ b/drivers/gpu/drm/amd/amdgpu/Makefile @@ -71,6 +71,12 @@ amdgpu-y += \ amdgpu_vce.o \ vce_v3_0.o +# add amdkfd interfaces +amdgpu-y += \ + amdgpu_amdkfd.o \ + amdgpu_amdkfd_gfx_v7.o \ + amdgpu_amdkfd_gfx_v8.o + amdgpu-$(CONFIG_COMPAT) += amdgpu_ioc32.o amdgpu-$(CONFIG_VGA_SWITCHEROO) += amdgpu_atpx_handler.o amdgpu-$(CONFIG_ACPI) += amdgpu_acpi.o diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 01657830b470..baefa635169a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1130,6 +1130,9 @@ struct amdgpu_gfx { uint32_t me_feature_version; uint32_t ce_feature_version; uint32_t pfp_feature_version; + uint32_t rlc_feature_version; + uint32_t mec_feature_version; + uint32_t mec2_feature_version; struct amdgpu_ring gfx_ring[AMDGPU_MAX_GFX_RINGS]; unsigned num_gfx_rings; struct amdgpu_ring compute_ring[AMDGPU_MAX_COMPUTE_RINGS]; @@ -1614,6 +1617,9 @@ struct amdgpu_uvd { #define AMDGPU_MAX_VCE_HANDLES 16 #define AMDGPU_VCE_FIRMWARE_OFFSET 256 +#define AMDGPU_VCE_HARVEST_VCE0 (1 << 0) +#define AMDGPU_VCE_HARVEST_VCE1 (1 << 1) + struct amdgpu_vce { struct amdgpu_bo *vcpu_bo; uint64_t gpu_addr; @@ -1626,6 +1632,7 @@ struct amdgpu_vce { const struct firmware *fw; /* VCE firmware */ struct amdgpu_ring ring[AMDGPU_MAX_VCE_RINGS]; struct amdgpu_irq_src irq; + unsigned harvest_config; }; /* @@ -1635,6 +1642,7 @@ struct amdgpu_sdma { /* SDMA firmware */ const struct firmware *fw; uint32_t fw_version; + uint32_t feature_version; struct amdgpu_ring ring; }; @@ -1862,6 +1870,12 @@ typedef void (*amdgpu_wreg_t)(struct amdgpu_device*, uint32_t, uint32_t); typedef uint32_t (*amdgpu_block_rreg_t)(struct amdgpu_device*, uint32_t, uint32_t); typedef void (*amdgpu_block_wreg_t)(struct amdgpu_device*, uint32_t, uint32_t, uint32_t); +struct amdgpu_ip_block_status { + bool valid; + bool sw; + bool hw; +}; + struct amdgpu_device { struct device *dev; struct drm_device *ddev; @@ -2004,13 +2018,16 @@ struct amdgpu_device { const struct amdgpu_ip_block_version *ip_blocks; int num_ip_blocks; - bool *ip_block_enabled; + struct amdgpu_ip_block_status *ip_block_status; struct mutex mn_lock; DECLARE_HASHTABLE(mn_hash, 7); /* tracking pinned memory */ u64 vram_pin_size; u64 gart_pin_size; + + /* amdkfd interface */ + struct kfd_dev *kfd; }; bool amdgpu_device_is_px(struct drm_device *dev); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c new file mode 100644 index 000000000000..bc763e0c8f4c --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -0,0 +1,267 @@ +/* + * 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 "amdgpu_amdkfd.h" +#include "amdgpu_family.h" +#include <drm/drmP.h> +#include "amdgpu.h" +#include <linux/module.h> + +const struct kfd2kgd_calls *kfd2kgd; +const struct kgd2kfd_calls *kgd2kfd; +bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); + +bool amdgpu_amdkfd_init(void) +{ +#if defined(CONFIG_HSA_AMD_MODULE) + bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); + + kgd2kfd_init_p = symbol_request(kgd2kfd_init); + + if (kgd2kfd_init_p == NULL) + return false; +#endif + return true; +} + +bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev) +{ +#if defined(CONFIG_HSA_AMD_MODULE) + bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); +#endif + + switch (rdev->asic_type) { + case CHIP_KAVERI: + kfd2kgd = amdgpu_amdkfd_gfx_7_get_functions(); + break; + case CHIP_CARRIZO: + kfd2kgd = amdgpu_amdkfd_gfx_8_0_get_functions(); + break; + default: + return false; + } + +#if defined(CONFIG_HSA_AMD_MODULE) + kgd2kfd_init_p = symbol_request(kgd2kfd_init); + + if (kgd2kfd_init_p == NULL) { + kfd2kgd = NULL; + return false; + } + + if (!kgd2kfd_init_p(KFD_INTERFACE_VERSION, &kgd2kfd)) { + symbol_put(kgd2kfd_init); + kfd2kgd = NULL; + kgd2kfd = NULL; + + return false; + } + + return true; +#elif defined(CONFIG_HSA_AMD) + if (!kgd2kfd_init(KFD_INTERFACE_VERSION, &kgd2kfd)) { + kfd2kgd = NULL; + kgd2kfd = NULL; + return false; + } + + return true; +#else + kfd2kgd = NULL; + return false; +#endif +} + +void amdgpu_amdkfd_fini(void) +{ + if (kgd2kfd) { + kgd2kfd->exit(); + symbol_put(kgd2kfd_init); + } +} + +void amdgpu_amdkfd_device_probe(struct amdgpu_device *rdev) +{ + if (kgd2kfd) + rdev->kfd = kgd2kfd->probe((struct kgd_dev *)rdev, + rdev->pdev, kfd2kgd); +} + +void amdgpu_amdkfd_device_init(struct amdgpu_device *rdev) +{ + if (rdev->kfd) { + struct kgd2kfd_shared_resources gpu_resources = { + .compute_vmid_bitmap = 0xFF00, + + .first_compute_pipe = 1, + .compute_pipe_count = 4 - 1, + }; + + amdgpu_doorbell_get_kfd_info(rdev, + &gpu_resources.doorbell_physical_address, + &gpu_resources.doorbell_aperture_size, + &gpu_resources.doorbell_start_offset); + + kgd2kfd->device_init(rdev->kfd, &gpu_resources); + } +} + +void amdgpu_amdkfd_device_fini(struct amdgpu_device *rdev) +{ + if (rdev->kfd) { + kgd2kfd->device_exit(rdev->kfd); + rdev->kfd = NULL; + } +} + +void amdgpu_amdkfd_interrupt(struct amdgpu_device *rdev, + const void *ih_ring_entry) +{ + if (rdev->kfd) + kgd2kfd->interrupt(rdev->kfd, ih_ring_entry); +} + +void amdgpu_amdkfd_suspend(struct amdgpu_device *rdev) +{ + if (rdev->kfd) + kgd2kfd->suspend(rdev->kfd); +} + +int amdgpu_amdkfd_resume(struct amdgpu_device *rdev) +{ + int r = 0; + + if (rdev->kfd) + r = kgd2kfd->resume(rdev->kfd); + + return r; +} + +u32 pool_to_domain(enum kgd_memory_pool p) +{ + switch (p) { + case KGD_POOL_FRAMEBUFFER: return AMDGPU_GEM_DOMAIN_VRAM; + default: return AMDGPU_GEM_DOMAIN_GTT; + } +} + +int alloc_gtt_mem(struct kgd_dev *kgd, size_t size, + void **mem_obj, uint64_t *gpu_addr, + void **cpu_ptr) +{ + struct amdgpu_device *rdev = (struct amdgpu_device *)kgd; + struct kgd_mem **mem = (struct kgd_mem **) mem_obj; + int r; + + BUG_ON(kgd == NULL); + BUG_ON(gpu_addr == NULL); + BUG_ON(cpu_ptr == NULL); + + *mem = kmalloc(sizeof(struct kgd_mem), GFP_KERNEL); + if ((*mem) == NULL) + return -ENOMEM; + + r = amdgpu_bo_create(rdev, size, PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_GTT, + AMDGPU_GEM_CREATE_CPU_GTT_USWC, NULL, &(*mem)->bo); + if (r) { + dev_err(rdev->dev, + "failed to allocate BO for amdkfd (%d)\n", r); + return r; + } + + /* map the buffer */ + r = amdgpu_bo_reserve((*mem)->bo, true); + if (r) { + dev_err(rdev->dev, "(%d) failed to reserve bo for amdkfd\n", r); + goto allocate_mem_reserve_bo_failed; + } + + r = amdgpu_bo_pin((*mem)->bo, AMDGPU_GEM_DOMAIN_GTT, + &(*mem)->gpu_addr); + if (r) { + dev_err(rdev->dev, "(%d) failed to pin bo for amdkfd\n", r); + goto allocate_mem_pin_bo_failed; + } + *gpu_addr = (*mem)->gpu_addr; + + r = amdgpu_bo_kmap((*mem)->bo, &(*mem)->cpu_ptr); + if (r) { + dev_err(rdev->dev, + "(%d) failed to map bo to kernel for amdkfd\n", r); + goto allocate_mem_kmap_bo_failed; + } + *cpu_ptr = (*mem)->cpu_ptr; + + amdgpu_bo_unreserve((*mem)->bo); + + return 0; + +allocate_mem_kmap_bo_failed: + amdgpu_bo_unpin((*mem)->bo); +allocate_mem_pin_bo_failed: + amdgpu_bo_unreserve((*mem)->bo); +allocate_mem_reserve_bo_failed: + amdgpu_bo_unref(&(*mem)->bo); + + return r; +} + +void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj) +{ + struct kgd_mem *mem = (struct kgd_mem *) mem_obj; + + BUG_ON(mem == NULL); + + amdgpu_bo_reserve(mem->bo, true); + amdgpu_bo_kunmap(mem->bo); + amdgpu_bo_unpin(mem->bo); + amdgpu_bo_unreserve(mem->bo); + amdgpu_bo_unref(&(mem->bo)); + kfree(mem); +} + +uint64_t get_vmem_size(struct kgd_dev *kgd) +{ + struct amdgpu_device *rdev = + (struct amdgpu_device *)kgd; + + BUG_ON(kgd == NULL); + + return rdev->mc.real_vram_size; +} + +uint64_t get_gpu_clock_counter(struct kgd_dev *kgd) +{ + struct amdgpu_device *rdev = (struct amdgpu_device *)kgd; + + if (rdev->asic_funcs->get_gpu_clock_counter) + return rdev->asic_funcs->get_gpu_clock_counter(rdev); + return 0; +} + +uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd) +{ + struct amdgpu_device *rdev = (struct amdgpu_device *)kgd; + + /* The sclk is in quantas of 10kHz */ + return rdev->pm.dpm.dyn_state.max_clock_voltage_on_ac.sclk / 100; +} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h new file mode 100644 index 000000000000..a8be765542e6 --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -0,0 +1,65 @@ +/* + * 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. + */ + +/* amdgpu_amdkfd.h defines the private interface between amdgpu and amdkfd. */ + +#ifndef AMDGPU_AMDKFD_H_INCLUDED +#define AMDGPU_AMDKFD_H_INCLUDED + +#include <linux/types.h> +#include <kgd_kfd_interface.h> + +struct amdgpu_device; + +struct kgd_mem { + struct amdgpu_bo *bo; + uint64_t gpu_addr; + void *cpu_ptr; +}; + +bool amdgpu_amdkfd_init(void); +void amdgpu_amdkfd_fini(void); + +bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev); + +void amdgpu_amdkfd_suspend(struct amdgpu_device *rdev); +int amdgpu_amdkfd_resume(struct amdgpu_device *rdev); +void amdgpu_amdkfd_interrupt(struct amdgpu_device *rdev, + const void *ih_ring_entry); +void amdgpu_amdkfd_device_probe(struct amdgpu_device *rdev); +void amdgpu_amdkfd_device_init(struct amdgpu_device *rdev); +void amdgpu_amdkfd_device_fini(struct amdgpu_device *rdev); + +struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void); +struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void); + +/* Shared API */ +int alloc_gtt_mem(struct kgd_dev *kgd, size_t size, + void **mem_obj, uint64_t *gpu_addr, + void **cpu_ptr); +void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj); +uint64_t get_vmem_size(struct kgd_dev *kgd); +uint64_t get_gpu_clock_counter(struct kgd_dev *kgd); + +uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd); + +#endif /* AMDGPU_AMDKFD_H_INCLUDED */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c new file mode 100644 index 000000000000..dd2037bc0b4a --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c @@ -0,0 +1,670 @@ +/* + * 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/fdtable.h> +#include <linux/uaccess.h> +#include <linux/firmware.h> +#include <drm/drmP.h> +#include "amdgpu.h" +#include "amdgpu_amdkfd.h" +#include "cikd.h" +#include "cik_sdma.h" +#include "amdgpu_ucode.h" +#include "gca/gfx_7_2_d.h" +#include "gca/gfx_7_2_enum.h" +#include "gca/gfx_7_2_sh_mask.h" +#include "oss/oss_2_0_d.h" +#include "oss/oss_2_0_sh_mask.h" +#include "gmc/gmc_7_1_d.h" +#include "gmc/gmc_7_1_sh_mask.h" +#include "cik_structs.h" + +#define CIK_PIPE_PER_MEC (4) + +enum { + MAX_TRAPID = 8, /* 3 bits in the bitfield. */ + MAX_WATCH_ADDRESSES = 4 +}; + +enum { + ADDRESS_WATCH_REG_ADDR_HI = 0, + ADDRESS_WATCH_REG_ADDR_LO, + ADDRESS_WATCH_REG_CNTL, + ADDRESS_WATCH_REG_MAX +}; + +/* not defined in the CI/KV reg file */ +enum { + ADDRESS_WATCH_REG_CNTL_ATC_BIT = 0x10000000UL, + ADDRESS_WATCH_REG_CNTL_DEFAULT_MASK = 0x00FFFFFF, + ADDRESS_WATCH_REG_ADDLOW_MASK_EXTENSION = 0x03000000, + /* extend the mask to 26 bits to match the low address field */ + ADDRESS_WATCH_REG_ADDLOW_SHIFT = 6, + ADDRESS_WATCH_REG_ADDHIGH_MASK = 0xFFFF +}; + +static const uint32_t watchRegs[MAX_WATCH_ADDRESSES * ADDRESS_WATCH_REG_MAX] = { + mmTCP_WATCH0_ADDR_H, mmTCP_WATCH0_ADDR_L, mmTCP_WATCH0_CNTL, + mmTCP_WATCH1_ADDR_H, mmTCP_WATCH1_ADDR_L, mmTCP_WATCH1_CNTL, + mmTCP_WATCH2_ADDR_H, mmTCP_WATCH2_ADDR_L, mmTCP_WATCH2_CNTL, + mmTCP_WATCH3_ADDR_H, mmTCP_WATCH3_ADDR_L, mmTCP_WATCH3_CNTL +}; + +union TCP_WATCH_CNTL_BITS { + struct { + uint32_t mask:24; + uint32_t vmid:4; + uint32_t atc:1; + uint32_t mode:2; + uint32_t valid:1; + } bitfields, bits; + uint32_t u32All; + signed int i32All; + float f32All; +}; + +/* + * Register access functions + */ + +static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid, + uint32_t sh_mem_config, uint32_t sh_mem_ape1_base, + uint32_t sh_mem_ape1_limit, uint32_t sh_mem_bases); + +static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid, + unsigned int vmid); + +static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id, + uint32_t hpd_size, uint64_t hpd_gpu_addr); +static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id); +static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id, + uint32_t queue_id, uint32_t __user *wptr); +static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd); +static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address, + uint32_t pipe_id, uint32_t queue_id); + +static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type, + unsigned int timeout, uint32_t pipe_id, + uint32_t queue_id); +static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd); +static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, + unsigned int timeout); +static int kgd_address_watch_disable(struct kgd_dev *kgd); +static int kgd_address_watch_execute(struct kgd_dev *kgd, + unsigned int watch_point_id, + uint32_t cntl_val, + uint32_t addr_hi, + uint32_t addr_lo); +static int kgd_wave_control_execute(struct kgd_dev *kgd, + uint32_t gfx_index_val, + uint32_t sq_cmd); +static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd, + unsigned int watch_point_id, + unsigned int reg_offset); + +static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd, uint8_t vmid); +static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd, + uint8_t vmid); +static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid); + +static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type); + +static const struct kfd2kgd_calls kfd2kgd = { + .init_gtt_mem_allocation = alloc_gtt_mem, + .free_gtt_mem = free_gtt_mem, + .get_vmem_size = get_vmem_size, + .get_gpu_clock_counter = get_gpu_clock_counter, + .get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz, + .program_sh_mem_settings = kgd_program_sh_mem_settings, + .set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping, + .init_pipeline = kgd_init_pipeline, + .init_interrupts = kgd_init_interrupts, + .hqd_load = kgd_hqd_load, + .hqd_sdma_load = kgd_hqd_sdma_load, + .hqd_is_occupied = kgd_hqd_is_occupied, + .hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied, + .hqd_destroy = kgd_hqd_destroy, + .hqd_sdma_destroy = kgd_hqd_sdma_destroy, + .address_watch_disable = kgd_address_watch_disable, + .address_watch_execute = kgd_address_watch_execute, + .wave_control_execute = kgd_wave_control_execute, + .address_watch_get_offset = kgd_address_watch_get_offset, + .get_atc_vmid_pasid_mapping_pasid = get_atc_vmid_pasid_mapping_pasid, + .get_atc_vmid_pasid_mapping_valid = get_atc_vmid_pasid_mapping_valid, + .write_vmid_invalidate_request = write_vmid_invalidate_request, + .get_fw_version = get_fw_version +}; + +struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions() +{ + return (struct kfd2kgd_calls *)&kfd2kgd; +} + +static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd) +{ + return (struct amdgpu_device *)kgd; +} + +static void lock_srbm(struct kgd_dev *kgd, uint32_t mec, uint32_t pipe, + uint32_t queue, uint32_t vmid) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + uint32_t value = PIPEID(pipe) | MEID(mec) | VMID(vmid) | QUEUEID(queue); + + mutex_lock(&adev->srbm_mutex); + WREG32(mmSRBM_GFX_CNTL, value); +} + +static void unlock_srbm(struct kgd_dev *kgd) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + + WREG32(mmSRBM_GFX_CNTL, 0); + mutex_unlock(&adev->srbm_mutex); +} + +static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id, + uint32_t queue_id) +{ + uint32_t mec = (++pipe_id / CIK_PIPE_PER_MEC) + 1; + uint32_t pipe = (pipe_id % CIK_PIPE_PER_MEC); + + lock_srbm(kgd, mec, pipe, queue_id, 0); +} + +static void release_queue(struct kgd_dev *kgd) +{ + unlock_srbm(kgd); +} + +static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid, + uint32_t sh_mem_config, + uint32_t sh_mem_ape1_base, + uint32_t sh_mem_ape1_limit, + uint32_t sh_mem_bases) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + + lock_srbm(kgd, 0, 0, 0, vmid); + + WREG32(mmSH_MEM_CONFIG, sh_mem_config); + WREG32(mmSH_MEM_APE1_BASE, sh_mem_ape1_base); + WREG32(mmSH_MEM_APE1_LIMIT, sh_mem_ape1_limit); + WREG32(mmSH_MEM_BASES, sh_mem_bases); + + unlock_srbm(kgd); +} + +static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid, + unsigned int vmid) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + + /* + * We have to assume that there is no outstanding mapping. + * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because + * a mapping is in progress or because a mapping finished and the + * SW cleared it. So the protocol is to always wait & clear. + */ + uint32_t pasid_mapping = (pasid == 0) ? 0 : (uint32_t)pasid | + ATC_VMID0_PASID_MAPPING__VALID_MASK; + + WREG32(mmATC_VMID0_PASID_MAPPING + vmid, pasid_mapping); + + while (!(RREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS) & (1U << vmid))) + cpu_relax(); + WREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS, 1U << vmid); + + /* Mapping vmid to pasid also for IH block */ + WREG32(mmIH_VMID_0_LUT + vmid, pasid_mapping); + + return 0; +} + +static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id, + uint32_t hpd_size, uint64_t hpd_gpu_addr) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + + uint32_t mec = (++pipe_id / CIK_PIPE_PER_MEC) + 1; + uint32_t pipe = (pipe_id % CIK_PIPE_PER_MEC); + + lock_srbm(kgd, mec, pipe, 0, 0); + WREG32(mmCP_HPD_EOP_BASE_ADDR, lower_32_bits(hpd_gpu_addr >> 8)); + WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(hpd_gpu_addr >> 8)); + WREG32(mmCP_HPD_EOP_VMID, 0); + WREG32(mmCP_HPD_EOP_CONTROL, hpd_size); + unlock_srbm(kgd); + + return 0; +} + +static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + uint32_t mec; + uint32_t pipe; + + mec = (pipe_id / CIK_PIPE_PER_MEC) + 1; + pipe = (pipe_id % CIK_PIPE_PER_MEC); + + lock_srbm(kgd, mec, pipe, 0, 0); + + WREG32(mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK | + CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK); + + unlock_srbm(kgd); + + return 0; +} + +static inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m) +{ + uint32_t retval; + + retval = m->sdma_engine_id * SDMA1_REGISTER_OFFSET + + m->sdma_queue_id * KFD_CIK_SDMA_QUEUE_OFFSET; + + pr_debug("kfd: sdma base address: 0x%x\n", retval); + + return retval; +} + +static inline struct cik_mqd *get_mqd(void *mqd) +{ + return (struct cik_mqd *)mqd; +} + +static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd) +{ + return (struct cik_sdma_rlc_registers *)mqd; +} + +static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id, + uint32_t queue_id, uint32_t __user *wptr) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + uint32_t wptr_shadow, is_wptr_shadow_valid; + struct cik_mqd *m; + + m = get_mqd(mqd); + + is_wptr_shadow_valid = !get_user(wptr_shadow, wptr); + + acquire_queue(kgd, pipe_id, queue_id); + WREG32(mmCP_MQD_BASE_ADDR, m->cp_mqd_base_addr_lo); + WREG32(mmCP_MQD_BASE_ADDR_HI, m->cp_mqd_base_addr_hi); + WREG32(mmCP_MQD_CONTROL, m->cp_mqd_control); + + WREG32(mmCP_HQD_PQ_BASE, m->cp_hqd_pq_base_lo); + WREG32(mmCP_HQD_PQ_BASE_HI, m->cp_hqd_pq_base_hi); + WREG32(mmCP_HQD_PQ_CONTROL, m->cp_hqd_pq_control); + + WREG32(mmCP_HQD_IB_CONTROL, m->cp_hqd_ib_control); + WREG32(mmCP_HQD_IB_BASE_ADDR, m->cp_hqd_ib_base_addr_lo); + WREG32(mmCP_HQD_IB_BASE_ADDR_HI, m->cp_hqd_ib_base_addr_hi); + + WREG32(mmCP_HQD_IB_RPTR, m->cp_hqd_ib_rptr); + + WREG32(mmCP_HQD_PERSISTENT_STATE, m->cp_hqd_persistent_state); + WREG32(mmCP_HQD_SEMA_CMD, m->cp_hqd_sema_cmd); + WREG32(mmCP_HQD_MSG_TYPE, m->cp_hqd_msg_type); + + WREG32(mmCP_HQD_ATOMIC0_PREOP_LO, m->cp_hqd_atomic0_preop_lo); + WREG32(mmCP_HQD_ATOMIC0_PREOP_HI, m->cp_hqd_atomic0_preop_hi); + WREG32(mmCP_HQD_ATOMIC1_PREOP_LO, m->cp_hqd_atomic1_preop_lo); + WREG32(mmCP_HQD_ATOMIC1_PREOP_HI, m->cp_hqd_atomic1_preop_hi); + + WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR, m->cp_hqd_pq_rptr_report_addr_lo); + WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI, + m->cp_hqd_pq_rptr_report_addr_hi); + + WREG32(mmCP_HQD_PQ_RPTR, m->cp_hqd_pq_rptr); + + WREG32(mmCP_HQD_PQ_WPTR_POLL_ADDR, m->cp_hqd_pq_wptr_poll_addr_lo); + WREG32(mmCP_HQD_PQ_WPTR_POLL_ADDR_HI, m->cp_hqd_pq_wptr_poll_addr_hi); + + WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, m->cp_hqd_pq_doorbell_control); + + WREG32(mmCP_HQD_VMID, m->cp_hqd_vmid); + + WREG32(mmCP_HQD_QUANTUM, m->cp_hqd_quantum); + + WREG32(mmCP_HQD_PIPE_PRIORITY, m->cp_hqd_pipe_priority); + WREG32(mmCP_HQD_QUEUE_PRIORITY, m->cp_hqd_queue_priority); + + WREG32(mmCP_HQD_IQ_RPTR, m->cp_hqd_iq_rptr); + + if (is_wptr_shadow_valid) + WREG32(mmCP_HQD_PQ_WPTR, wptr_shadow); + + WREG32(mmCP_HQD_ACTIVE, m->cp_hqd_active); + release_queue(kgd); + + return 0; +} + +static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + struct cik_sdma_rlc_registers *m; + uint32_t sdma_base_addr; + + m = get_sdma_mqd(mqd); + sdma_base_addr = get_sdma_base_addr(m); + + WREG32(sdma_base_addr + mmSDMA0_RLC0_VIRTUAL_ADDR, + m->sdma_rlc_virtual_addr); + + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, + m->sdma_rlc_rb_base); + + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE_HI, + m->sdma_rlc_rb_base_hi); + + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_LO, + m->sdma_rlc_rb_rptr_addr_lo); + + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI, + m->sdma_rlc_rb_rptr_addr_hi); + + WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, + m->sdma_rlc_doorbell); + + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, + m->sdma_rlc_rb_cntl); + + return 0; +} + +static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address, + uint32_t pipe_id, uint32_t queue_id) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + uint32_t act; + bool retval = false; + uint32_t low, high; + + acquire_queue(kgd, pipe_id, queue_id); + act = RREG32(mmCP_HQD_ACTIVE); + if (act) { + low = lower_32_bits(queue_address >> 8); + high = upper_32_bits(queue_address >> 8); + + if (low == RREG32(mmCP_HQD_PQ_BASE) && + high == RREG32(mmCP_HQD_PQ_BASE_HI)) + retval = true; + } + release_queue(kgd); + return retval; +} + +static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + struct cik_sdma_rlc_registers *m; + uint32_t sdma_base_addr; + uint32_t sdma_rlc_rb_cntl; + + m = get_sdma_mqd(mqd); + sdma_base_addr = get_sdma_base_addr(m); + + sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); + + if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK) + return true; + + return false; +} + +static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type, + unsigned int timeout, uint32_t pipe_id, + uint32_t queue_id) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + uint32_t temp; + + acquire_queue(kgd, pipe_id, queue_id); + WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, 0); + + WREG32(mmCP_HQD_DEQUEUE_REQUEST, reset_type); + + while (true) { + temp = RREG32(mmCP_HQD_ACTIVE); + if (temp & CP_HQD_ACTIVE__ACTIVE_MASK) + break; + if (timeout == 0) { + pr_err("kfd: cp queue preemption time out (%dms)\n", + temp); + release_queue(kgd); + return -ETIME; + } + msleep(20); + timeout -= 20; + } + + release_queue(kgd); + return 0; +} + +static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, + unsigned int timeout) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + struct cik_sdma_rlc_registers *m; + uint32_t sdma_base_addr; + uint32_t temp; + + m = get_sdma_mqd(mqd); + sdma_base_addr = get_sdma_base_addr(m); + + temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); + temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK; + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp); + + while (true) { + temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS); + if (temp & SDMA0_STATUS_REG__RB_CMD_IDLE__SHIFT) + break; + if (timeout == 0) + return -ETIME; + msleep(20); + timeout -= 20; + } + + WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0); + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0); + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0); + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, 0); + + return 0; +} + +static int kgd_address_watch_disable(struct kgd_dev *kgd) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + union TCP_WATCH_CNTL_BITS cntl; + unsigned int i; + + cntl.u32All = 0; + + cntl.bitfields.valid = 0; + cntl.bitfields.mask = ADDRESS_WATCH_REG_CNTL_DEFAULT_MASK; + cntl.bitfields.atc = 1; + + /* Turning off this address until we set all the registers */ + for (i = 0; i < MAX_WATCH_ADDRESSES; i++) + WREG32(watchRegs[i * ADDRESS_WATCH_REG_MAX + + ADDRESS_WATCH_REG_CNTL], cntl.u32All); + + return 0; +} + +static int kgd_address_watch_execute(struct kgd_dev *kgd, + unsigned int watch_point_id, + uint32_t cntl_val, + uint32_t addr_hi, + uint32_t addr_lo) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + union TCP_WATCH_CNTL_BITS cntl; + + cntl.u32All = cntl_val; + + /* Turning off this watch point until we set all the registers */ + cntl.bitfields.valid = 0; + WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX + + ADDRESS_WATCH_REG_CNTL], cntl.u32All); + + WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX + + ADDRESS_WATCH_REG_ADDR_HI], addr_hi); + + WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX + + ADDRESS_WATCH_REG_ADDR_LO], addr_lo); + + /* Enable the watch point */ + cntl.bitfields.valid = 1; + + WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX + + ADDRESS_WATCH_REG_CNTL], cntl.u32All); + + return 0; +} + +static int kgd_wave_control_execute(struct kgd_dev *kgd, + uint32_t gfx_index_val, + uint32_t sq_cmd) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + uint32_t data; + + mutex_lock(&adev->grbm_idx_mutex); + + WREG32(mmGRBM_GFX_INDEX, gfx_index_val); + WREG32(mmSQ_CMD, sq_cmd); + + /* Restore the GRBM_GFX_INDEX register */ + + data = GRBM_GFX_INDEX__INSTANCE_BROADCAST_WRITES_MASK | + GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK | + GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK; + + WREG32(mmGRBM_GFX_INDEX, data); + + mutex_unlock(&adev->grbm_idx_mutex); + + return 0; +} + +static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd, + unsigned int watch_point_id, + unsigned int reg_offset) +{ + return watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX + reg_offset]; +} + +static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd, + uint8_t vmid) +{ + uint32_t reg; + struct amdgpu_device *adev = (struct amdgpu_device *) kgd; + + reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid); + return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK; +} + +static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd, + uint8_t vmid) +{ + uint32_t reg; + struct amdgpu_device *adev = (struct amdgpu_device *) kgd; + + reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid); + return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK; +} + +static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid) +{ + struct amdgpu_device *adev = (struct amdgpu_device *) kgd; + + WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid); +} + +static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type) +{ + struct amdgpu_device *adev = (struct amdgpu_device *) kgd; + const union amdgpu_firmware_header *hdr; + + BUG_ON(kgd == NULL); + + switch (type) { + case KGD_ENGINE_PFP: + hdr = (const union amdgpu_firmware_header *) + adev->gfx.pfp_fw->data; + break; + + case KGD_ENGINE_ME: + hdr = (const union amdgpu_firmware_header *) + adev->gfx.me_fw->data; + break; + + case KGD_ENGINE_CE: + hdr = (const union amdgpu_firmware_header *) + adev->gfx.ce_fw->data; + break; + + case KGD_ENGINE_MEC1: + hdr = (const union amdgpu_firmware_header *) + adev->gfx.mec_fw->data; + break; + + case KGD_ENGINE_MEC2: + hdr = (const union amdgpu_firmware_header *) + adev->gfx.mec2_fw->data; + break; + + case KGD_ENGINE_RLC: + hdr = (const union amdgpu_firmware_header *) + adev->gfx.rlc_fw->data; + break; + + case KGD_ENGINE_SDMA1: + hdr = (const union amdgpu_firmware_header *) + adev->sdma[0].fw->data; + break; + + case KGD_ENGINE_SDMA2: + hdr = (const union amdgpu_firmware_header *) + adev->sdma[1].fw->data; + break; + + default: + return 0; + } + + if (hdr == NULL) + return 0; + + /* Only 12 bit in use*/ + return hdr->common.ucode_version; +} + diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c new file mode 100644 index 000000000000..dfd1d503bccf --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c @@ -0,0 +1,543 @@ +/* + * 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/module.h> +#include <linux/fdtable.h> +#include <linux/uaccess.h> +#include <linux/firmware.h> +#include <drm/drmP.h> +#include "amdgpu.h" +#include "amdgpu_amdkfd.h" +#include "amdgpu_ucode.h" +#include "gca/gfx_8_0_sh_mask.h" +#include "gca/gfx_8_0_d.h" +#include "gca/gfx_8_0_enum.h" +#include "oss/oss_3_0_sh_mask.h" +#include "oss/oss_3_0_d.h" +#include "gmc/gmc_8_1_sh_mask.h" +#include "gmc/gmc_8_1_d.h" +#include "vi_structs.h" +#include "vid.h" + +#define VI_PIPE_PER_MEC (4) + +struct cik_sdma_rlc_registers; + +/* + * Register access functions + */ + +static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid, + uint32_t sh_mem_config, + uint32_t sh_mem_ape1_base, uint32_t sh_mem_ape1_limit, + uint32_t sh_mem_bases); +static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid, + unsigned int vmid); +static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id, + uint32_t hpd_size, uint64_t hpd_gpu_addr); +static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id); +static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id, + uint32_t queue_id, uint32_t __user *wptr); +static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd); +static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address, + uint32_t pipe_id, uint32_t queue_id); +static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd); +static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type, + unsigned int timeout, uint32_t pipe_id, + uint32_t queue_id); +static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, + unsigned int timeout); +static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid); +static int kgd_address_watch_disable(struct kgd_dev *kgd); +static int kgd_address_watch_execute(struct kgd_dev *kgd, + unsigned int watch_point_id, + uint32_t cntl_val, + uint32_t addr_hi, + uint32_t addr_lo); +static int kgd_wave_control_execute(struct kgd_dev *kgd, + uint32_t gfx_index_val, + uint32_t sq_cmd); +static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd, + unsigned int watch_point_id, + unsigned int reg_offset); + +static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd, + uint8_t vmid); +static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd, + uint8_t vmid); +static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid); +static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type); + +static const struct kfd2kgd_calls kfd2kgd = { + .init_gtt_mem_allocation = alloc_gtt_mem, + .free_gtt_mem = free_gtt_mem, + .get_vmem_size = get_vmem_size, + .get_gpu_clock_counter = get_gpu_clock_counter, + .get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz, + .program_sh_mem_settings = kgd_program_sh_mem_settings, + .set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping, + .init_pipeline = kgd_init_pipeline, + .init_interrupts = kgd_init_interrupts, + .hqd_load = kgd_hqd_load, + .hqd_sdma_load = kgd_hqd_sdma_load, + .hqd_is_occupied = kgd_hqd_is_occupied, + .hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied, + .hqd_destroy = kgd_hqd_destroy, + .hqd_sdma_destroy = kgd_hqd_sdma_destroy, + .address_watch_disable = kgd_address_watch_disable, + .address_watch_execute = kgd_address_watch_execute, + .wave_control_execute = kgd_wave_control_execute, + .address_watch_get_offset = kgd_address_watch_get_offset, + .get_atc_vmid_pasid_mapping_pasid = + get_atc_vmid_pasid_mapping_pasid, + .get_atc_vmid_pasid_mapping_valid = + get_atc_vmid_pasid_mapping_valid, + .write_vmid_invalidate_request = write_vmid_invalidate_request, + .get_fw_version = get_fw_version +}; + +struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions() +{ + return (struct kfd2kgd_calls *)&kfd2kgd; +} + +static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd) +{ + return (struct amdgpu_device *)kgd; +} + +static void lock_srbm(struct kgd_dev *kgd, uint32_t mec, uint32_t pipe, + uint32_t queue, uint32_t vmid) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + uint32_t value = PIPEID(pipe) | MEID(mec) | VMID(vmid) | QUEUEID(queue); + + mutex_lock(&adev->srbm_mutex); + WREG32(mmSRBM_GFX_CNTL, value); +} + +static void unlock_srbm(struct kgd_dev *kgd) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + + WREG32(mmSRBM_GFX_CNTL, 0); + mutex_unlock(&adev->srbm_mutex); +} + +static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id, + uint32_t queue_id) +{ + uint32_t mec = (++pipe_id / VI_PIPE_PER_MEC) + 1; + uint32_t pipe = (pipe_id % VI_PIPE_PER_MEC); + + lock_srbm(kgd, mec, pipe, queue_id, 0); +} + +static void release_queue(struct kgd_dev *kgd) +{ + unlock_srbm(kgd); +} + +static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid, + uint32_t sh_mem_config, + uint32_t sh_mem_ape1_base, + uint32_t sh_mem_ape1_limit, + uint32_t sh_mem_bases) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + + lock_srbm(kgd, 0, 0, 0, vmid); + + WREG32(mmSH_MEM_CONFIG, sh_mem_config); + WREG32(mmSH_MEM_APE1_BASE, sh_mem_ape1_base); + WREG32(mmSH_MEM_APE1_LIMIT, sh_mem_ape1_limit); + WREG32(mmSH_MEM_BASES, sh_mem_bases); + + unlock_srbm(kgd); +} + +static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid, + unsigned int vmid) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + + /* + * We have to assume that there is no outstanding mapping. + * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because + * a mapping is in progress or because a mapping finished + * and the SW cleared it. + * So the protocol is to always wait & clear. + */ + uint32_t pasid_mapping = (pasid == 0) ? 0 : (uint32_t)pasid | + ATC_VMID0_PASID_MAPPING__VALID_MASK; + + WREG32(mmATC_VMID0_PASID_MAPPING + vmid, pasid_mapping); + + while (!(RREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS) & (1U << vmid))) + cpu_relax(); + WREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS, 1U << vmid); + + /* Mapping vmid to pasid also for IH block */ + WREG32(mmIH_VMID_0_LUT + vmid, pasid_mapping); + + return 0; +} + +static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id, + uint32_t hpd_size, uint64_t hpd_gpu_addr) +{ + return 0; +} + +static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + uint32_t mec; + uint32_t pipe; + + mec = (++pipe_id / VI_PIPE_PER_MEC) + 1; + pipe = (pipe_id % VI_PIPE_PER_MEC); + + lock_srbm(kgd, mec, pipe, 0, 0); + + WREG32(mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK); + + unlock_srbm(kgd); + + return 0; +} + +static inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m) +{ + return 0; +} + +static inline struct vi_mqd *get_mqd(void *mqd) +{ + return (struct vi_mqd *)mqd; +} + +static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd) +{ + return (struct cik_sdma_rlc_registers *)mqd; +} + +static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id, + uint32_t queue_id, uint32_t __user *wptr) +{ + struct vi_mqd *m; + uint32_t shadow_wptr, valid_wptr; + struct amdgpu_device *adev = get_amdgpu_device(kgd); + + m = get_mqd(mqd); + + valid_wptr = copy_from_user(&shadow_wptr, wptr, sizeof(shadow_wptr)); + acquire_queue(kgd, pipe_id, queue_id); + + WREG32(mmCP_MQD_CONTROL, m->cp_mqd_control); + WREG32(mmCP_MQD_BASE_ADDR, m->cp_mqd_base_addr_lo); + WREG32(mmCP_MQD_BASE_ADDR_HI, m->cp_mqd_base_addr_hi); + + WREG32(mmCP_HQD_VMID, m->cp_hqd_vmid); + WREG32(mmCP_HQD_PERSISTENT_STATE, m->cp_hqd_persistent_state); + WREG32(mmCP_HQD_PIPE_PRIORITY, m->cp_hqd_pipe_priority); + WREG32(mmCP_HQD_QUEUE_PRIORITY, m->cp_hqd_queue_priority); + WREG32(mmCP_HQD_QUANTUM, m->cp_hqd_quantum); + WREG32(mmCP_HQD_PQ_BASE, m->cp_hqd_pq_base_lo); + WREG32(mmCP_HQD_PQ_BASE_HI, m->cp_hqd_pq_base_hi); + WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR, m->cp_hqd_pq_rptr_report_addr_lo); + WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI, + m->cp_hqd_pq_rptr_report_addr_hi); + + if (valid_wptr > 0) + WREG32(mmCP_HQD_PQ_WPTR, shadow_wptr); + + WREG32(mmCP_HQD_PQ_CONTROL, m->cp_hqd_pq_control); + WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, m->cp_hqd_pq_doorbell_control); + + WREG32(mmCP_HQD_EOP_BASE_ADDR, m->cp_hqd_eop_base_addr_lo); + WREG32(mmCP_HQD_EOP_BASE_ADDR_HI, m->cp_hqd_eop_base_addr_hi); + WREG32(mmCP_HQD_EOP_CONTROL, m->cp_hqd_eop_control); + WREG32(mmCP_HQD_EOP_RPTR, m->cp_hqd_eop_rptr); + WREG32(mmCP_HQD_EOP_WPTR, m->cp_hqd_eop_wptr); + WREG32(mmCP_HQD_EOP_EVENTS, m->cp_hqd_eop_done_events); + + WREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO, m->cp_hqd_ctx_save_base_addr_lo); + WREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI, m->cp_hqd_ctx_save_base_addr_hi); + WREG32(mmCP_HQD_CTX_SAVE_CONTROL, m->cp_hqd_ctx_save_control); + WREG32(mmCP_HQD_CNTL_STACK_OFFSET, m->cp_hqd_cntl_stack_offset); + WREG32(mmCP_HQD_CNTL_STACK_SIZE, m->cp_hqd_cntl_stack_size); + WREG32(mmCP_HQD_WG_STATE_OFFSET, m->cp_hqd_wg_state_offset); + WREG32(mmCP_HQD_CTX_SAVE_SIZE, m->cp_hqd_ctx_save_size); + + WREG32(mmCP_HQD_IB_CONTROL, m->cp_hqd_ib_control); + + WREG32(mmCP_HQD_DEQUEUE_REQUEST, m->cp_hqd_dequeue_request); + WREG32(mmCP_HQD_ERROR, m->cp_hqd_error); + WREG32(mmCP_HQD_EOP_WPTR_MEM, m->cp_hqd_eop_wptr_mem); + WREG32(mmCP_HQD_EOP_DONES, m->cp_hqd_eop_dones); + + WREG32(mmCP_HQD_ACTIVE, m->cp_hqd_active); + + release_queue(kgd); + + return 0; +} + +static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd) +{ + return 0; +} + +static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address, + uint32_t pipe_id, uint32_t queue_id) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + uint32_t act; + bool retval = false; + uint32_t low, high; + + acquire_queue(kgd, pipe_id, queue_id); + act = RREG32(mmCP_HQD_ACTIVE); + if (act) { + low = lower_32_bits(queue_address >> 8); + high = upper_32_bits(queue_address >> 8); + + if (low == RREG32(mmCP_HQD_PQ_BASE) && + high == RREG32(mmCP_HQD_PQ_BASE_HI)) + retval = true; + } + release_queue(kgd); + return retval; +} + +static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + struct cik_sdma_rlc_registers *m; + uint32_t sdma_base_addr; + uint32_t sdma_rlc_rb_cntl; + + m = get_sdma_mqd(mqd); + sdma_base_addr = get_sdma_base_addr(m); + + sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); + + if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK) + return true; + + return false; +} + +static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type, + unsigned int timeout, uint32_t pipe_id, + uint32_t queue_id) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + uint32_t temp; + + acquire_queue(kgd, pipe_id, queue_id); + + WREG32(mmCP_HQD_DEQUEUE_REQUEST, reset_type); + + while (true) { + temp = RREG32(mmCP_HQD_ACTIVE); + if (temp & CP_HQD_ACTIVE__ACTIVE_MASK) + break; + if (timeout == 0) { + pr_err("kfd: cp queue preemption time out (%dms)\n", + temp); + release_queue(kgd); + return -ETIME; + } + msleep(20); + timeout -= 20; + } + + release_queue(kgd); + return 0; +} + +static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, + unsigned int timeout) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + struct cik_sdma_rlc_registers *m; + uint32_t sdma_base_addr; + uint32_t temp; + + m = get_sdma_mqd(mqd); + sdma_base_addr = get_sdma_base_addr(m); + + temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); + temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK; + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp); + + while (true) { + temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS); + if (temp & SDMA0_STATUS_REG__RB_CMD_IDLE__SHIFT) + break; + if (timeout == 0) + return -ETIME; + msleep(20); + timeout -= 20; + } + + WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0); + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0); + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0); + WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, 0); + + return 0; +} + +static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd, + uint8_t vmid) +{ + uint32_t reg; + struct amdgpu_device *adev = (struct amdgpu_device *) kgd; + + reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid); + return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK; +} + +static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd, + uint8_t vmid) +{ + uint32_t reg; + struct amdgpu_device *adev = (struct amdgpu_device *) kgd; + + reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid); + return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK; +} + +static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid) +{ + struct amdgpu_device *adev = (struct amdgpu_device *) kgd; + + WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid); +} + +static int kgd_address_watch_disable(struct kgd_dev *kgd) +{ + return 0; +} + +static int kgd_address_watch_execute(struct kgd_dev *kgd, + unsigned int watch_point_id, + uint32_t cntl_val, + uint32_t addr_hi, + uint32_t addr_lo) +{ + return 0; +} + +static int kgd_wave_control_execute(struct kgd_dev *kgd, + uint32_t gfx_index_val, + uint32_t sq_cmd) +{ + struct amdgpu_device *adev = get_amdgpu_device(kgd); + uint32_t data = 0; + + mutex_lock(&adev->grbm_idx_mutex); + + WREG32(mmGRBM_GFX_INDEX, gfx_index_val); + WREG32(mmSQ_CMD, sq_cmd); + + data = REG_SET_FIELD(data, GRBM_GFX_INDEX, + INSTANCE_BROADCAST_WRITES, 1); + data = REG_SET_FIELD(data, GRBM_GFX_INDEX, + SH_BROADCAST_WRITES, 1); + data = REG_SET_FIELD(data, GRBM_GFX_INDEX, + SE_BROADCAST_WRITES, 1); + + WREG32(mmGRBM_GFX_INDEX, data); + mutex_unlock(&adev->grbm_idx_mutex); + + return 0; +} + +static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd, + unsigned int watch_point_id, + unsigned int reg_offset) +{ + return 0; +} + +static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type) +{ + struct amdgpu_device *adev = (struct amdgpu_device *) kgd; + const union amdgpu_firmware_header *hdr; + + BUG_ON(kgd == NULL); + + switch (type) { + case KGD_ENGINE_PFP: + hdr = (const union amdgpu_firmware_header *) + adev->gfx.pfp_fw->data; + break; + + case KGD_ENGINE_ME: + hdr = (const union amdgpu_firmware_header *) + adev->gfx.me_fw->data; + break; + + case KGD_ENGINE_CE: + hdr = (const union amdgpu_firmware_header *) + adev->gfx.ce_fw->data; + break; + + case KGD_ENGINE_MEC1: + hdr = (const union amdgpu_firmware_header *) + adev->gfx.mec_fw->data; + break; + + case KGD_ENGINE_MEC2: + hdr = (const union amdgpu_firmware_header *) + adev->gfx.mec2_fw->data; + break; + + case KGD_ENGINE_RLC: + hdr = (const union amdgpu_firmware_header *) + adev->gfx.rlc_fw->data; + break; + + case KGD_ENGINE_SDMA1: + hdr = (const union amdgpu_firmware_header *) + adev->sdma[0].fw->data; + break; + + case KGD_ENGINE_SDMA2: + hdr = (const union amdgpu_firmware_header *) + adev->sdma[1].fw->data; + break; + + default: + return 0; + } + + if (hdr == NULL) + return 0; + + /* Only 12 bit in use*/ + return hdr->common.ucode_version; +} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index d63135bf29c0..1f040d85ac47 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -669,6 +669,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev, static int amdgpu_cs_dependencies(struct amdgpu_device *adev, struct amdgpu_cs_parser *p) { + struct amdgpu_fpriv *fpriv = p->filp->driver_priv; struct amdgpu_ib *ib; int i, j, r; @@ -694,6 +695,7 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev, for (j = 0; j < num_deps; ++j) { struct amdgpu_fence *fence; struct amdgpu_ring *ring; + struct amdgpu_ctx *ctx; r = amdgpu_cs_get_ring(adev, deps[j].ip_type, deps[j].ip_instance, @@ -701,14 +703,21 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev, if (r) return r; + ctx = amdgpu_ctx_get(fpriv, deps[j].ctx_id); + if (ctx == NULL) + return -EINVAL; + r = amdgpu_fence_recreate(ring, p->filp, deps[j].handle, &fence); - if (r) + if (r) { + amdgpu_ctx_put(ctx); return r; + } amdgpu_sync_fence(&ib->sync, fence); amdgpu_fence_unref(&fence); + amdgpu_ctx_put(ctx); } } @@ -808,12 +817,16 @@ int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data, r = amdgpu_cs_get_ring(adev, wait->in.ip_type, wait->in.ip_instance, wait->in.ring, &ring); - if (r) + if (r) { + amdgpu_ctx_put(ctx); return r; + } r = amdgpu_fence_recreate(ring, filp, wait->in.handle, &fence); - if (r) + if (r) { + amdgpu_ctx_put(ctx); return r; + } r = fence_wait_timeout(&fence->base, true, timeout); amdgpu_fence_unref(&fence); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index ba46be361c9b..99f158e1baff 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -1191,8 +1191,9 @@ static int amdgpu_early_init(struct amdgpu_device *adev) return -EINVAL; } - adev->ip_block_enabled = kcalloc(adev->num_ip_blocks, sizeof(bool), GFP_KERNEL); - if (adev->ip_block_enabled == NULL) + adev->ip_block_status = kcalloc(adev->num_ip_blocks, + sizeof(struct amdgpu_ip_block_status), GFP_KERNEL); + if (adev->ip_block_status == NULL) return -ENOMEM; if (adev->ip_blocks == NULL) { @@ -1203,14 +1204,19 @@ static int amdgpu_early_init(struct amdgpu_device *adev) for (i = 0; i < adev->num_ip_blocks; i++) { if ((amdgpu_ip_block_mask & (1 << i)) == 0) { DRM_ERROR("disabled ip block: %d\n", i); - adev->ip_block_enabled[i] = false; + adev->ip_block_status[i].valid = false; } else { if (adev->ip_blocks[i].funcs->early_init) { r = adev->ip_blocks[i].funcs->early_init((void *)adev); - if (r) + if (r == -ENOENT) + adev->ip_block_status[i].valid = false; + else if (r) return r; + else + adev->ip_block_status[i].valid = true; + } else { + adev->ip_block_status[i].valid = true; } - adev->ip_block_enabled[i] = true; } } @@ -1222,11 +1228,12 @@ static int amdgpu_init(struct amdgpu_device *adev) int i, r; for (i = 0; i < adev->num_ip_blocks; i++) { - if (!adev->ip_block_enabled[i]) + if (!adev->ip_block_status[i].valid) continue; r = adev->ip_blocks[i].funcs->sw_init((void *)adev); if (r) return r; + adev->ip_block_status[i].sw = true; /* need to do gmc hw init early so we can allocate gpu mem */ if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) { r = amdgpu_vram_scratch_init(adev); @@ -1238,11 +1245,12 @@ static int amdgpu_init(struct amdgpu_device *adev) r = amdgpu_wb_init(adev); if (r) return r; + adev->ip_block_status[i].hw = true; } } for (i = 0; i < adev->num_ip_blocks; i++) { - if (!adev->ip_block_enabled[i]) + if (!adev->ip_block_status[i].sw) continue; /* gmc hw init is done early */ if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) @@ -1250,6 +1258,7 @@ static int amdgpu_init(struct amdgpu_device *adev) r = adev->ip_blocks[i].funcs->hw_init((void *)adev); if (r) return r; + adev->ip_block_status[i].hw = true; } return 0; @@ -1260,7 +1269,7 @@ static int amdgpu_late_init(struct amdgpu_device *adev) int i = 0, r; for (i = 0; i < adev->num_ip_blocks; i++) { - if (!adev->ip_block_enabled[i]) + if (!adev->ip_block_status[i].valid) continue; /* enable clockgating to save power */ r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, @@ -1282,7 +1291,7 @@ static int amdgpu_fini(struct amdgpu_device *adev) int i, r; for (i = adev->num_ip_blocks - 1; i >= 0; i--) { - if (!adev->ip_block_enabled[i]) + if (!adev->ip_block_status[i].hw) continue; if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) { amdgpu_wb_fini(adev); @@ -1295,14 +1304,16 @@ static int amdgpu_fini(struct amdgpu_device *adev) return r; r = adev->ip_blocks[i].funcs->hw_fini((void *)adev); /* XXX handle errors */ + adev->ip_block_status[i].hw = false; } for (i = adev->num_ip_blocks - 1; i >= 0; i--) { - if (!adev->ip_block_enabled[i]) + if (!adev->ip_block_status[i].sw) continue; r = adev->ip_blocks[i].funcs->sw_fini((void *)adev); /* XXX handle errors */ - adev->ip_block_enabled[i] = false; + adev->ip_block_status[i].sw = false; + adev->ip_block_status[i].valid = false; } return 0; @@ -1313,7 +1324,7 @@ static int amdgpu_suspend(struct amdgpu_device *adev) int i, r; for (i = adev->num_ip_blocks - 1; i >= 0; i--) { - if (!adev->ip_block_enabled[i]) + if (!adev->ip_block_status[i].valid) continue; /* ungate blocks so that suspend can properly shut them down */ r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, @@ -1331,7 +1342,7 @@ static int amdgpu_resume(struct amdgpu_device *adev) int i, r; for (i = 0; i < adev->num_ip_blocks; i++) { - if (!adev->ip_block_enabled[i]) + if (!adev->ip_block_status[i].valid) continue; r = adev->ip_blocks[i].funcs->resume(adev); if (r) @@ -1577,8 +1588,8 @@ void amdgpu_device_fini(struct amdgpu_device *adev) amdgpu_fence_driver_fini(adev); amdgpu_fbdev_fini(adev); r = amdgpu_fini(adev); - kfree(adev->ip_block_enabled); - adev->ip_block_enabled = NULL; + kfree(adev->ip_block_status); + adev->ip_block_status = NULL; adev->accel_working = false; /* free i2c buses */ amdgpu_i2c_fini(adev); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 56da962231fc..115906f5fda0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -44,6 +44,8 @@ #include "amdgpu.h" #include "amdgpu_irq.h" +#include "amdgpu_amdkfd.h" + /* * KMS wrapper. * - 3.0.0 - initial driver @@ -527,12 +529,15 @@ static int __init amdgpu_init(void) driver->num_ioctls = amdgpu_max_kms_ioctl; amdgpu_register_atpx_handler(); + amdgpu_amdkfd_init(); + /* let modprobe override vga console setting */ return drm_pci_init(driver, pdriver); } static void __exit amdgpu_exit(void) { + amdgpu_amdkfd_fini(); drm_pci_exit(driver, pdriver); amdgpu_unregister_atpx_handler(); } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c index c1645d21f8e2..81b821247dde 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c @@ -53,9 +53,9 @@ static struct fb_ops amdgpufb_ops = { .owner = THIS_MODULE, .fb_check_var = drm_fb_helper_check_var, .fb_set_par = drm_fb_helper_set_par, - .fb_fillrect = cfb_fillrect, - .fb_copyarea = cfb_copyarea, - .fb_imageblit = cfb_imageblit, + .fb_fillrect = drm_fb_helper_cfb_fillrect, + .fb_copyarea = drm_fb_helper_cfb_copyarea, + .fb_imageblit = drm_fb_helper_cfb_imageblit, .fb_pan_display = drm_fb_helper_pan_display, .fb_blank = drm_fb_helper_blank, .fb_setcmap = drm_fb_helper_setcmap, @@ -179,7 +179,6 @@ static int amdgpufb_create(struct drm_fb_helper *helper, struct drm_mode_fb_cmd2 mode_cmd; struct drm_gem_object *gobj = NULL; struct amdgpu_bo *rbo = NULL; - struct device *device = &adev->pdev->dev; int ret; unsigned long tmp; @@ -201,9 +200,9 @@ static int amdgpufb_create(struct drm_fb_helper *helper, rbo = gem_to_amdgpu_bo(gobj); /* okay we have an object now allocate the framebuffer */ - info = framebuffer_alloc(0, device); - if (info == NULL) { - ret = -ENOMEM; + info = drm_fb_helper_alloc_fbi(helper); + if (IS_ERR(info)) { + ret = PTR_ERR(info); goto out_unref; } @@ -212,14 +211,13 @@ static int amdgpufb_create(struct drm_fb_helper *helper, ret = amdgpu_framebuffer_init(adev->ddev, &rfbdev->rfb, &mode_cmd, gobj); if (ret) { DRM_ERROR("failed to initialize framebuffer %d\n", ret); - goto out_unref; + goto out_destroy_fbi; } fb = &rfbdev->rfb.base; /* setup helper */ rfbdev->helper.fb = fb; - rfbdev->helper.fbdev = info; memset_io(rbo->kptr, 0x0, amdgpu_bo_size(rbo)); @@ -239,11 +237,6 @@ static int amdgpufb_create(struct drm_fb_helper *helper, drm_fb_helper_fill_var(info, &rfbdev->helper, sizes->fb_width, sizes->fb_height); /* setup aperture base/size for vesafb takeover */ - info->apertures = alloc_apertures(1); - if (!info->apertures) { - ret = -ENOMEM; - goto out_unref; - } info->apertures->ranges[0].base = adev->ddev->mode_config.fb_base; info->apertures->ranges[0].size = adev->mc.aper_size; @@ -251,13 +244,7 @@ static int amdgpufb_create(struct drm_fb_helper *helper, if (info->screen_base == NULL) { ret = -ENOSPC; - goto out_unref; - } - - ret = fb_alloc_cmap(&info->cmap, 256, 0); - if (ret) { - ret = -ENOMEM; - goto out_unref; + goto out_destroy_fbi; } DRM_INFO("fb mappable at 0x%lX\n", info->fix.smem_start); @@ -269,6 +256,8 @@ static int amdgpufb_create(struct drm_fb_helper *helper, vga_switcheroo_client_fb_set(adev->ddev->pdev, info); return 0; +out_destroy_fbi: + drm_fb_helper_release_fbi(helper); out_unref: if (rbo) { @@ -290,17 +279,10 @@ void amdgpu_fb_output_poll_changed(struct amdgpu_device *adev) static int amdgpu_fbdev_destroy(struct drm_device *dev, struct amdgpu_fbdev *rfbdev) { - struct fb_info *info; struct amdgpu_framebuffer *rfb = &rfbdev->rfb; - if (rfbdev->helper.fbdev) { - info = rfbdev->helper.fbdev; - - unregister_framebuffer(info); - if (info->cmap.len) - fb_dealloc_cmap(&info->cmap); - framebuffer_release(info); - } + drm_fb_helper_unregister_fbi(&rfbdev->helper); + drm_fb_helper_release_fbi(&rfbdev->helper); if (rfb->obj) { amdgpufb_destroy_pinned_object(rfb->obj); @@ -395,7 +377,8 @@ void amdgpu_fbdev_fini(struct amdgpu_device *adev) void amdgpu_fbdev_set_suspend(struct amdgpu_device *adev, int state) { if (adev->mode_info.rfbdev) - fb_set_suspend(adev->mode_info.rfbdev->helper.fbdev, state); + drm_fb_helper_set_suspend(&adev->mode_info.rfbdev->helper, + state); } int amdgpu_fbdev_total_size(struct amdgpu_device *adev) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c index 975edb1000a2..4afc507820c0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c @@ -352,7 +352,7 @@ unsigned long amdgpu_gem_timeout(uint64_t timeout_ns) if (((int64_t)timeout_ns) < 0) return MAX_SCHEDULE_TIMEOUT; - timeout = ktime_sub_ns(ktime_get(), timeout_ns); + timeout = ktime_sub(ns_to_ktime(timeout_ns), ktime_get()); if (ktime_to_ns(timeout) < 0) return 0; @@ -449,7 +449,7 @@ out: * vital here, so they are not reported back to userspace. */ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev, - struct amdgpu_bo_va *bo_va) + struct amdgpu_bo_va *bo_va, uint32_t operation) { struct ttm_validate_buffer tv, *entry; struct amdgpu_bo_list_entry *vm_bos; @@ -485,7 +485,9 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev, if (r) goto error_unlock; - r = amdgpu_vm_bo_update(adev, bo_va, &bo_va->bo->tbo.mem); + + if (operation == AMDGPU_VA_OP_MAP) + r = amdgpu_vm_bo_update(adev, bo_va, &bo_va->bo->tbo.mem); error_unlock: mutex_unlock(&bo_va->vm->mutex); @@ -580,7 +582,7 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, } if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE)) - amdgpu_gem_va_update_vm(adev, bo_va); + amdgpu_gem_va_update_vm(adev, bo_va, args->operation); drm_gem_object_unreference_unlocked(gobj); return r; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index 52dff75aac6f..bc0fac618a3f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -180,16 +180,16 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs, if (vm) { /* do context switch */ amdgpu_vm_flush(ring, vm, ib->sync.last_vm_update); - } - if (vm && ring->funcs->emit_gds_switch) - amdgpu_ring_emit_gds_switch(ring, ib->vm->ids[ring->idx].id, - ib->gds_base, ib->gds_size, - ib->gws_base, ib->gws_size, - ib->oa_base, ib->oa_size); + if (ring->funcs->emit_gds_switch) + amdgpu_ring_emit_gds_switch(ring, ib->vm->ids[ring->idx].id, + ib->gds_base, ib->gds_size, + ib->gws_base, ib->gws_size, + ib->oa_base, ib->oa_size); - if (ring->funcs->emit_hdp_flush) - amdgpu_ring_emit_hdp_flush(ring); + if (ring->funcs->emit_hdp_flush) + amdgpu_ring_emit_hdp_flush(ring); + } old_ctx = ring->current_ctx; for (i = 0; i < num_ibs; ++i) { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c index db5422e65ec5..fb44dd2231b1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c @@ -24,6 +24,7 @@ #include <drm/drmP.h> #include "amdgpu.h" #include "amdgpu_ih.h" +#include "amdgpu_amdkfd.h" /** * amdgpu_ih_ring_alloc - allocate memory for the IH ring @@ -199,6 +200,12 @@ restart_ih: rmb(); while (adev->irq.ih.rptr != wptr) { + u32 ring_index = adev->irq.ih.rptr >> 2; + + /* Before dispatching irq to IP blocks, send it to amdkfd */ + amdgpu_amdkfd_interrupt(adev, + (const void *) &adev->irq.ih.ring[ring_index]); + amdgpu_ih_decode_iv(adev, &entry); adev->irq.ih.rptr &= adev->irq.ih.ptr_mask; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c index 5533434c7a8f..93000af92283 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c @@ -34,6 +34,7 @@ #include <linux/vga_switcheroo.h> #include <linux/slab.h> #include <linux/pm_runtime.h> +#include "amdgpu_amdkfd.h" #if defined(CONFIG_VGA_SWITCHEROO) bool amdgpu_has_atpx(void); @@ -61,6 +62,8 @@ int amdgpu_driver_unload_kms(struct drm_device *dev) pm_runtime_get_sync(dev->dev); + amdgpu_amdkfd_device_fini(adev); + amdgpu_acpi_fini(adev); amdgpu_device_fini(adev); @@ -118,6 +121,10 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags) "Error during ACPI methods call\n"); } + amdgpu_amdkfd_load_interface(adev); + amdgpu_amdkfd_device_probe(adev); + amdgpu_amdkfd_device_init(adev); + if (amdgpu_device_is_px(dev)) { pm_runtime_use_autosuspend(dev->dev); pm_runtime_set_autosuspend_delay(dev->dev, 5000); @@ -235,7 +242,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file for (i = 0; i < adev->num_ip_blocks; i++) { if (adev->ip_blocks[i].type == type && - adev->ip_block_enabled[i]) { + adev->ip_block_status[i].valid) { ip.hw_ip_version_major = adev->ip_blocks[i].major; ip.hw_ip_version_minor = adev->ip_blocks[i].minor; ip.capabilities_flags = 0; @@ -274,7 +281,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file for (i = 0; i < adev->num_ip_blocks; i++) if (adev->ip_blocks[i].type == type && - adev->ip_block_enabled[i] && + adev->ip_block_status[i].valid && count < AMDGPU_HW_IP_INSTANCE_MAX_COUNT) count++; @@ -317,16 +324,17 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file break; case AMDGPU_INFO_FW_GFX_RLC: fw_info.ver = adev->gfx.rlc_fw_version; - fw_info.feature = 0; + fw_info.feature = adev->gfx.rlc_feature_version; break; case AMDGPU_INFO_FW_GFX_MEC: - if (info->query_fw.index == 0) + if (info->query_fw.index == 0) { fw_info.ver = adev->gfx.mec_fw_version; - else if (info->query_fw.index == 1) + fw_info.feature = adev->gfx.mec_feature_version; + } else if (info->query_fw.index == 1) { fw_info.ver = adev->gfx.mec2_fw_version; - else + fw_info.feature = adev->gfx.mec2_feature_version; + } else return -EINVAL; - fw_info.feature = 0; break; case AMDGPU_INFO_FW_SMC: fw_info.ver = adev->pm.fw_version; @@ -336,7 +344,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file if (info->query_fw.index >= 2) return -EINVAL; fw_info.ver = adev->sdma[info->query_fw.index].fw_version; - fw_info.feature = 0; + fw_info.feature = adev->sdma[info->query_fw.index].feature_version; break; default: return -EINVAL; @@ -416,7 +424,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file return n ? -EFAULT : 0; } case AMDGPU_INFO_DEV_INFO: { - struct drm_amdgpu_info_device dev_info; + struct drm_amdgpu_info_device dev_info = {}; struct amdgpu_cu_info cu_info; dev_info.device_id = dev->pdev->device; @@ -459,6 +467,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file memcpy(&dev_info.cu_bitmap[0], &cu_info.bitmap[0], sizeof(cu_info.bitmap)); dev_info.vram_type = adev->mc.vram_type; dev_info.vram_bit_width = adev->mc.vram_width; + dev_info.vce_harvest_config = adev->vce.harvest_config; return copy_to_user(out, &dev_info, min((size_t)size, sizeof(dev_info))) ? -EFAULT : 0; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c index 2f7a5efa21c2..f5c22556ec2c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c @@ -374,7 +374,7 @@ static int amdgpu_uvd_cs_msg_decode(uint32_t *msg, unsigned buf_sizes[]) unsigned height_in_mb = ALIGN(height / 16, 2); unsigned fs_in_mb = width_in_mb * height_in_mb; - unsigned image_size, tmp, min_dpb_size, num_dpb_buffer; + unsigned image_size, tmp, min_dpb_size, num_dpb_buffer, min_ctx_size; image_size = width * height; image_size += image_size / 2; @@ -466,6 +466,8 @@ static int amdgpu_uvd_cs_msg_decode(uint32_t *msg, unsigned buf_sizes[]) num_dpb_buffer = (le32_to_cpu(msg[59]) & 0xff) + 2; min_dpb_size = image_size * num_dpb_buffer; + min_ctx_size = ((width + 255) / 16) * ((height + 255) / 16) + * 16 * num_dpb_buffer + 52 * 1024; break; default: @@ -486,6 +488,7 @@ static int amdgpu_uvd_cs_msg_decode(uint32_t *msg, unsigned buf_sizes[]) buf_sizes[0x1] = dpb_size; buf_sizes[0x2] = image_size; + buf_sizes[0x4] = min_ctx_size; return 0; } @@ -628,6 +631,13 @@ static int amdgpu_uvd_cs_pass2(struct amdgpu_uvd_cs_ctx *ctx) return -EINVAL; } + } else if (cmd == 0x206) { + if ((end - start) < ctx->buf_sizes[4]) { + DRM_ERROR("buffer (%d) to small (%d / %d)!\n", cmd, + (unsigned)(end - start), + ctx->buf_sizes[4]); + return -EINVAL; + } } else if ((cmd != 0x100) && (cmd != 0x204)) { DRM_ERROR("invalid UVD command %X!\n", cmd); return -EINVAL; @@ -755,9 +765,10 @@ int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx) struct amdgpu_uvd_cs_ctx ctx = {}; unsigned buf_sizes[] = { [0x00000000] = 2048, - [0x00000001] = 32 * 1024 * 1024, - [0x00000002] = 2048 * 1152 * 3, + [0x00000001] = 0xFFFFFFFF, + [0x00000002] = 0xFFFFFFFF, [0x00000003] = 2048, + [0x00000004] = 0xFFFFFFFF, }; struct amdgpu_ib *ib = &parser->ibs[ib_idx]; int r; diff --git a/drivers/gpu/drm/amd/amdgpu/cik.c b/drivers/gpu/drm/amd/amdgpu/cik.c index 341c56681841..b3b66a0d5ff7 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik.c +++ b/drivers/gpu/drm/amd/amdgpu/cik.c @@ -64,6 +64,8 @@ #include "oss/oss_2_0_d.h" #include "oss/oss_2_0_sh_mask.h" +#include "amdgpu_amdkfd.h" + /* * Indirect registers accessor */ @@ -2448,14 +2450,21 @@ static int cik_common_suspend(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + amdgpu_amdkfd_suspend(adev); + return cik_common_hw_fini(adev); } static int cik_common_resume(void *handle) { + int r; struct amdgpu_device *adev = (struct amdgpu_device *)handle; - return cik_common_hw_init(adev); + r = cik_common_hw_init(adev); + if (r) + return r; + + return amdgpu_amdkfd_resume(adev); } static bool cik_common_is_idle(void *handle) diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c index ab83cc1ca4cc..15df46c93f0a 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c +++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c @@ -500,6 +500,7 @@ static int cik_sdma_load_microcode(struct amdgpu_device *adev) amdgpu_ucode_print_sdma_hdr(&hdr->header); fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version); + adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version); fw_data = (const __le32 *) (adev->sdma[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0); diff --git a/drivers/gpu/drm/amd/amdgpu/cikd.h b/drivers/gpu/drm/amd/amdgpu/cikd.h index d19085a97064..a3e3dfaa01a4 100644 --- a/drivers/gpu/drm/amd/amdgpu/cikd.h +++ b/drivers/gpu/drm/amd/amdgpu/cikd.h @@ -552,6 +552,12 @@ #define VCE_CMD_IB_AUTO 0x00000005 #define VCE_CMD_SEMAPHORE 0x00000006 +/* if PTR32, these are the bases for scratch and lds */ +#define PRIVATE_BASE(x) ((x) << 0) /* scratch */ +#define SHARED_BASE(x) ((x) << 16) /* LDS */ + +#define KFD_CIK_SDMA_QUEUE_OFFSET 0x200 + /* valid for both DEFAULT_MTYPE and APE1_MTYPE */ enum { MTYPE_CACHED = 0, diff --git a/drivers/gpu/drm/amd/amdgpu/cz_dpm.c b/drivers/gpu/drm/amd/amdgpu/cz_dpm.c index f75a31df30bd..ace870afc7d4 100644 --- a/drivers/gpu/drm/amd/amdgpu/cz_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/cz_dpm.c @@ -494,29 +494,67 @@ static void cz_dpm_fini(struct amdgpu_device *adev) amdgpu_free_extended_power_table(adev); } +#define ixSMUSVI_NB_CURRENTVID 0xD8230044 +#define CURRENT_NB_VID_MASK 0xff000000 +#define CURRENT_NB_VID__SHIFT 24 +#define ixSMUSVI_GFX_CURRENTVID 0xD8230048 +#define CURRENT_GFX_VID_MASK 0xff000000 +#define CURRENT_GFX_VID__SHIFT 24 + static void cz_dpm_debugfs_print_current_performance_level(struct amdgpu_device *adev, struct seq_file *m) { + struct cz_power_info *pi = cz_get_pi(adev); struct amdgpu_clock_voltage_dependency_table *table = &adev->pm.dpm.dyn_state.vddc_dependency_on_sclk; - u32 current_index = - (RREG32_SMC(ixTARGET_AND_CURRENT_PROFILE_INDEX) & - TARGET_AND_CURRENT_PROFILE_INDEX__CURR_SCLK_INDEX_MASK) >> - TARGET_AND_CURRENT_PROFILE_INDEX__CURR_SCLK_INDEX__SHIFT; - u32 sclk, tmp; - u16 vddc; - - if (current_index >= NUM_SCLK_LEVELS) { - seq_printf(m, "invalid dpm profile %d\n", current_index); + struct amdgpu_uvd_clock_voltage_dependency_table *uvd_table = + &adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table; + struct amdgpu_vce_clock_voltage_dependency_table *vce_table = + &adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table; + u32 sclk_index = REG_GET_FIELD(RREG32_SMC(ixTARGET_AND_CURRENT_PROFILE_INDEX), + TARGET_AND_CURRENT_PROFILE_INDEX, CURR_SCLK_INDEX); + u32 uvd_index = REG_GET_FIELD(RREG32_SMC(ixTARGET_AND_CURRENT_PROFILE_INDEX_2), + TARGET_AND_CURRENT_PROFILE_INDEX_2, CURR_UVD_INDEX); + u32 vce_index = REG_GET_FIELD(RREG32_SMC(ixTARGET_AND_CURRENT_PROFILE_INDEX_2), + TARGET_AND_CURRENT_PROFILE_INDEX_2, CURR_VCE_INDEX); + u32 sclk, vclk, dclk, ecclk, tmp; + u16 vddnb, vddgfx; + + if (sclk_index >= NUM_SCLK_LEVELS) { + seq_printf(m, "invalid sclk dpm profile %d\n", sclk_index); } else { - sclk = table->entries[current_index].clk; - tmp = (RREG32_SMC(ixSMU_VOLTAGE_STATUS) & - SMU_VOLTAGE_STATUS__SMU_VOLTAGE_CURRENT_LEVEL_MASK) >> - SMU_VOLTAGE_STATUS__SMU_VOLTAGE_CURRENT_LEVEL__SHIFT; - vddc = cz_convert_8bit_index_to_voltage(adev, (u16)tmp); - seq_printf(m, "power level %d sclk: %u vddc: %u\n", - current_index, sclk, vddc); + sclk = table->entries[sclk_index].clk; + seq_printf(m, "%u sclk: %u\n", sclk_index, sclk); + } + + tmp = (RREG32_SMC(ixSMUSVI_NB_CURRENTVID) & + CURRENT_NB_VID_MASK) >> CURRENT_NB_VID__SHIFT; + vddnb = cz_convert_8bit_index_to_voltage(adev, (u16)tmp); + tmp = (RREG32_SMC(ixSMUSVI_GFX_CURRENTVID) & + CURRENT_GFX_VID_MASK) >> CURRENT_GFX_VID__SHIFT; + vddgfx = cz_convert_8bit_index_to_voltage(adev, (u16)tmp); + seq_printf(m, "vddnb: %u vddgfx: %u\n", vddnb, vddgfx); + + seq_printf(m, "uvd %sabled\n", pi->uvd_power_gated ? "dis" : "en"); + if (!pi->uvd_power_gated) { + if (uvd_index >= CZ_MAX_HARDWARE_POWERLEVELS) { + seq_printf(m, "invalid uvd dpm level %d\n", uvd_index); + } else { + vclk = uvd_table->entries[uvd_index].vclk; + dclk = uvd_table->entries[uvd_index].dclk; + seq_printf(m, "%u uvd vclk: %u dclk: %u\n", uvd_index, vclk, dclk); + } + } + + seq_printf(m, "vce %sabled\n", pi->vce_power_gated ? "dis" : "en"); + if (!pi->vce_power_gated) { + if (vce_index >= CZ_MAX_HARDWARE_POWERLEVELS) { + seq_printf(m, "invalid vce dpm level %d\n", vce_index); + } else { + ecclk = vce_table->entries[vce_index].ecclk; + seq_printf(m, "%u vce ecclk: %u\n", vce_index, ecclk); + } } } @@ -1679,25 +1717,31 @@ static int cz_dpm_unforce_dpm_levels(struct amdgpu_device *adev) if (ret) return ret; - DRM_INFO("DPM unforce state min=%d, max=%d.\n", - pi->sclk_dpm.soft_min_clk, - pi->sclk_dpm.soft_max_clk); + DRM_DEBUG("DPM unforce state min=%d, max=%d.\n", + pi->sclk_dpm.soft_min_clk, + pi->sclk_dpm.soft_max_clk); return 0; } static int cz_dpm_force_dpm_level(struct amdgpu_device *adev, - enum amdgpu_dpm_forced_level level) + enum amdgpu_dpm_forced_level level) { int ret = 0; switch (level) { case AMDGPU_DPM_FORCED_LEVEL_HIGH: + ret = cz_dpm_unforce_dpm_levels(adev); + if (ret) + return ret; ret = cz_dpm_force_highest(adev); if (ret) return ret; break; case AMDGPU_DPM_FORCED_LEVEL_LOW: + ret = cz_dpm_unforce_dpm_levels(adev); + if (ret) + return ret; ret = cz_dpm_force_lowest(adev); if (ret) return ret; @@ -1711,6 +1755,8 @@ static int cz_dpm_force_dpm_level(struct amdgpu_device *adev, break; } + adev->pm.dpm.forced_level = level; + return ret; } diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c index 5cde635978f9..e70a26f587a0 100644 --- a/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c @@ -2632,6 +2632,7 @@ static void dce_v10_0_crtc_dpms(struct drm_crtc *crtc, int mode) struct drm_device *dev = crtc->dev; struct amdgpu_device *adev = dev->dev_private; struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); + unsigned type; switch (mode) { case DRM_MODE_DPMS_ON: @@ -2640,6 +2641,9 @@ static void dce_v10_0_crtc_dpms(struct drm_crtc *crtc, int mode) dce_v10_0_vga_enable(crtc, true); amdgpu_atombios_crtc_blank(crtc, ATOM_DISABLE); dce_v10_0_vga_enable(crtc, false); + /* Make sure VBLANK interrupt is still enabled */ + type = amdgpu_crtc_idx_to_irq_type(adev, amdgpu_crtc->crtc_id); + amdgpu_irq_update(adev, &adev->crtc_irq, type); drm_vblank_post_modeset(dev, amdgpu_crtc->crtc_id); dce_v10_0_crtc_load_lut(crtc); break; @@ -3403,19 +3407,25 @@ static int dce_v10_0_crtc_irq(struct amdgpu_device *adev, switch (entry->src_data) { case 0: /* vblank */ - if (disp_int & interrupt_status_offsets[crtc].vblank) { + if (disp_int & interrupt_status_offsets[crtc].vblank) dce_v10_0_crtc_vblank_int_ack(adev, crtc); - if (amdgpu_irq_enabled(adev, source, irq_type)) { - drm_handle_vblank(adev->ddev, crtc); - } - DRM_DEBUG("IH: D%d vblank\n", crtc + 1); + else + DRM_DEBUG("IH: IH event w/o asserted irq bit?\n"); + + if (amdgpu_irq_enabled(adev, source, irq_type)) { + drm_handle_vblank(adev->ddev, crtc); } + DRM_DEBUG("IH: D%d vblank\n", crtc + 1); + break; case 1: /* vline */ - if (disp_int & interrupt_status_offsets[crtc].vline) { + if (disp_int & interrupt_status_offsets[crtc].vline) dce_v10_0_crtc_vline_int_ack(adev, crtc); - DRM_DEBUG("IH: D%d vline\n", crtc + 1); - } + else + DRM_DEBUG("IH: IH event w/o asserted irq bit?\n"); + + DRM_DEBUG("IH: D%d vline\n", crtc + 1); + break; default: DRM_DEBUG("Unhandled interrupt: %d %d\n", entry->src_id, entry->src_data); diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c index 95efd98b202d..dcb402ee048a 100644 --- a/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c @@ -2631,6 +2631,7 @@ static void dce_v11_0_crtc_dpms(struct drm_crtc *crtc, int mode) struct drm_device *dev = crtc->dev; struct amdgpu_device *adev = dev->dev_private; struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); + unsigned type; switch (mode) { case DRM_MODE_DPMS_ON: @@ -2639,6 +2640,9 @@ static void dce_v11_0_crtc_dpms(struct drm_crtc *crtc, int mode) dce_v11_0_vga_enable(crtc, true); amdgpu_atombios_crtc_blank(crtc, ATOM_DISABLE); dce_v11_0_vga_enable(crtc, false); + /* Make sure VBLANK interrupt is still enabled */ + type = amdgpu_crtc_idx_to_irq_type(adev, amdgpu_crtc->crtc_id); + amdgpu_irq_update(adev, &adev->crtc_irq, type); drm_vblank_post_modeset(dev, amdgpu_crtc->crtc_id); dce_v11_0_crtc_load_lut(crtc); break; @@ -3402,19 +3406,25 @@ static int dce_v11_0_crtc_irq(struct amdgpu_device *adev, switch (entry->src_data) { case 0: /* vblank */ - if (disp_int & interrupt_status_offsets[crtc].vblank) { + if (disp_int & interrupt_status_offsets[crtc].vblank) dce_v11_0_crtc_vblank_int_ack(adev, crtc); - if (amdgpu_irq_enabled(adev, source, irq_type)) { - drm_handle_vblank(adev->ddev, crtc); - } - DRM_DEBUG("IH: D%d vblank\n", crtc + 1); + else + DRM_DEBUG("IH: IH event w/o asserted irq bit?\n"); + + if (amdgpu_irq_enabled(adev, source, irq_type)) { + drm_handle_vblank(adev->ddev, crtc); } + DRM_DEBUG("IH: D%d vblank\n", crtc + 1); + break; case 1: /* vline */ - if (disp_int & interrupt_status_offsets[crtc].vline) { + if (disp_int & interrupt_status_offsets[crtc].vline) dce_v11_0_crtc_vline_int_ack(adev, crtc); - DRM_DEBUG("IH: D%d vline\n", crtc + 1); - } + else + DRM_DEBUG("IH: IH event w/o asserted irq bit?\n"); + + DRM_DEBUG("IH: D%d vline\n", crtc + 1); + break; default: DRM_DEBUG("Unhandled interrupt: %d %d\n", entry->src_id, entry->src_data); diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c index aaca8d663f2c..cc050a329c49 100644 --- a/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c @@ -2566,6 +2566,7 @@ static void dce_v8_0_crtc_dpms(struct drm_crtc *crtc, int mode) struct drm_device *dev = crtc->dev; struct amdgpu_device *adev = dev->dev_private; struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); + unsigned type; switch (mode) { case DRM_MODE_DPMS_ON: @@ -2574,6 +2575,9 @@ static void dce_v8_0_crtc_dpms(struct drm_crtc *crtc, int mode) dce_v8_0_vga_enable(crtc, true); amdgpu_atombios_crtc_blank(crtc, ATOM_DISABLE); dce_v8_0_vga_enable(crtc, false); + /* Make sure VBLANK interrupt is still enabled */ + type = amdgpu_crtc_idx_to_irq_type(adev, amdgpu_crtc->crtc_id); + amdgpu_irq_update(adev, &adev->crtc_irq, type); drm_vblank_post_modeset(dev, amdgpu_crtc->crtc_id); dce_v8_0_crtc_load_lut(crtc); break; @@ -3237,19 +3241,25 @@ static int dce_v8_0_crtc_irq(struct amdgpu_device *adev, switch (entry->src_data) { case 0: /* vblank */ - if (disp_int & interrupt_status_offsets[crtc].vblank) { + if (disp_int & interrupt_status_offsets[crtc].vblank) WREG32(mmLB_VBLANK_STATUS + crtc_offsets[crtc], LB_VBLANK_STATUS__VBLANK_ACK_MASK); - if (amdgpu_irq_enabled(adev, source, irq_type)) { - drm_handle_vblank(adev->ddev, crtc); - } - DRM_DEBUG("IH: D%d vblank\n", crtc + 1); + else + DRM_DEBUG("IH: IH event w/o asserted irq bit?\n"); + + if (amdgpu_irq_enabled(adev, source, irq_type)) { + drm_handle_vblank(adev->ddev, crtc); } + DRM_DEBUG("IH: D%d vblank\n", crtc + 1); + break; case 1: /* vline */ - if (disp_int & interrupt_status_offsets[crtc].vline) { + if (disp_int & interrupt_status_offsets[crtc].vline) WREG32(mmLB_VLINE_STATUS + crtc_offsets[crtc], LB_VLINE_STATUS__VLINE_ACK_MASK); - DRM_DEBUG("IH: D%d vline\n", crtc + 1); - } + else + DRM_DEBUG("IH: IH event w/o asserted irq bit?\n"); + + DRM_DEBUG("IH: D%d vline\n", crtc + 1); + break; default: DRM_DEBUG("Unhandled interrupt: %d %d\n", entry->src_id, entry->src_data); diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c index 2c188fb9fd22..0d8bf2cb1956 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c @@ -2561,7 +2561,7 @@ static bool gfx_v7_0_ring_emit_semaphore(struct amdgpu_ring *ring, * sheduling on the ring. This function schedules the IB * on the gfx ring for execution by the GPU. */ -static void gfx_v7_0_ring_emit_ib(struct amdgpu_ring *ring, +static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { bool need_ctx_switch = ring->current_ctx != ib->ctx; @@ -2569,15 +2569,10 @@ static void gfx_v7_0_ring_emit_ib(struct amdgpu_ring *ring, u32 next_rptr = ring->wptr + 5; /* drop the CE preamble IB for the same context */ - if ((ring->type == AMDGPU_RING_TYPE_GFX) && - (ib->flags & AMDGPU_IB_FLAG_PREAMBLE) && - !need_ctx_switch) + if ((ib->flags & AMDGPU_IB_FLAG_PREAMBLE) && !need_ctx_switch) return; - if (ring->type == AMDGPU_RING_TYPE_COMPUTE) - control |= INDIRECT_BUFFER_VALID; - - if (need_ctx_switch && ring->type == AMDGPU_RING_TYPE_GFX) + if (need_ctx_switch) next_rptr += 2; next_rptr += 4; @@ -2588,7 +2583,7 @@ static void gfx_v7_0_ring_emit_ib(struct amdgpu_ring *ring, amdgpu_ring_write(ring, next_rptr); /* insert SWITCH_BUFFER packet before first IB in the ring frame */ - if (need_ctx_switch && ring->type == AMDGPU_RING_TYPE_GFX) { + if (need_ctx_switch) { amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); amdgpu_ring_write(ring, 0); } @@ -2611,6 +2606,35 @@ static void gfx_v7_0_ring_emit_ib(struct amdgpu_ring *ring, amdgpu_ring_write(ring, control); } +static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring, + struct amdgpu_ib *ib) +{ + u32 header, control = 0; + u32 next_rptr = ring->wptr + 5; + + control |= INDIRECT_BUFFER_VALID; + next_rptr += 4; + amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); + amdgpu_ring_write(ring, WRITE_DATA_DST_SEL(5) | WR_CONFIRM); + amdgpu_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc); + amdgpu_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff); + amdgpu_ring_write(ring, next_rptr); + + header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); + + control |= ib->length_dw | + (ib->vm ? (ib->vm->ids[ring->idx].id << 24) : 0); + + amdgpu_ring_write(ring, header); + amdgpu_ring_write(ring, +#ifdef __BIG_ENDIAN + (2 << 0) | +#endif + (ib->gpu_addr & 0xFFFFFFFC)); + amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF); + amdgpu_ring_write(ring, control); +} + /** * gfx_v7_0_ring_test_ib - basic ring IB test * @@ -3056,6 +3080,8 @@ static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev) mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; amdgpu_ucode_print_gfx_hdr(&mec_hdr->header); adev->gfx.mec_fw_version = le32_to_cpu(mec_hdr->header.ucode_version); + adev->gfx.mec_feature_version = le32_to_cpu( + mec_hdr->ucode_feature_version); gfx_v7_0_cp_compute_enable(adev, false); @@ -3078,6 +3104,8 @@ static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev) mec2_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; amdgpu_ucode_print_gfx_hdr(&mec2_hdr->header); adev->gfx.mec2_fw_version = le32_to_cpu(mec2_hdr->header.ucode_version); + adev->gfx.mec2_feature_version = le32_to_cpu( + mec2_hdr->ucode_feature_version); /* MEC2 */ fw_data = (const __le32 *) @@ -4042,6 +4070,8 @@ static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev) hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data; amdgpu_ucode_print_rlc_hdr(&hdr->header); adev->gfx.rlc_fw_version = le32_to_cpu(hdr->header.ucode_version); + adev->gfx.rlc_feature_version = le32_to_cpu( + hdr->ucode_feature_version); gfx_v7_0_rlc_stop(adev); @@ -5098,7 +5128,7 @@ static void gfx_v7_0_print_status(void *handle) dev_info(adev->dev, " CP_HPD_EOP_CONTROL=0x%08X\n", RREG32(mmCP_HPD_EOP_CONTROL)); - for (queue = 0; queue < 8; i++) { + for (queue = 0; queue < 8; queue++) { cik_srbm_select(adev, me, pipe, queue, 0); dev_info(adev->dev, " queue: %d\n", queue); dev_info(adev->dev, " CP_PQ_WPTR_POLL_CNTL=0x%08X\n", @@ -5555,7 +5585,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = { .get_wptr = gfx_v7_0_ring_get_wptr_gfx, .set_wptr = gfx_v7_0_ring_set_wptr_gfx, .parse_cs = NULL, - .emit_ib = gfx_v7_0_ring_emit_ib, + .emit_ib = gfx_v7_0_ring_emit_ib_gfx, .emit_fence = gfx_v7_0_ring_emit_fence_gfx, .emit_semaphore = gfx_v7_0_ring_emit_semaphore, .emit_vm_flush = gfx_v7_0_ring_emit_vm_flush, @@ -5571,7 +5601,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = { .get_wptr = gfx_v7_0_ring_get_wptr_compute, .set_wptr = gfx_v7_0_ring_set_wptr_compute, .parse_cs = NULL, - .emit_ib = gfx_v7_0_ring_emit_ib, + .emit_ib = gfx_v7_0_ring_emit_ib_compute, .emit_fence = gfx_v7_0_ring_emit_fence_compute, .emit_semaphore = gfx_v7_0_ring_emit_semaphore, .emit_vm_flush = gfx_v7_0_ring_emit_vm_flush, diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 7b683fb2173c..20e2cfd521d5 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -587,6 +587,7 @@ static int gfx_v8_0_init_microcode(struct amdgpu_device *adev) int err; struct amdgpu_firmware_info *info = NULL; const struct common_firmware_header *header = NULL; + const struct gfx_firmware_header_v1_0 *cp_hdr; DRM_DEBUG("\n"); @@ -611,6 +612,9 @@ static int gfx_v8_0_init_microcode(struct amdgpu_device *adev) err = amdgpu_ucode_validate(adev->gfx.pfp_fw); if (err) goto out; + cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; + adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); + adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name); err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev); @@ -619,6 +623,9 @@ static int gfx_v8_0_init_microcode(struct amdgpu_device *adev) err = amdgpu_ucode_validate(adev->gfx.me_fw); if (err) goto out; + cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; + adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); + adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name); err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev); @@ -627,12 +634,18 @@ static int gfx_v8_0_init_microcode(struct amdgpu_device *adev) err = amdgpu_ucode_validate(adev->gfx.ce_fw); if (err) goto out; + cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; + adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); + adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name); err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev); if (err) goto out; err = amdgpu_ucode_validate(adev->gfx.rlc_fw); + cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.rlc_fw->data; + adev->gfx.rlc_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); + adev->gfx.rlc_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name); err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev); @@ -641,6 +654,9 @@ static int gfx_v8_0_init_microcode(struct amdgpu_device *adev) err = amdgpu_ucode_validate(adev->gfx.mec_fw); if (err) goto out; + cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; + adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); + adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name); err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev); @@ -648,6 +664,12 @@ static int gfx_v8_0_init_microcode(struct amdgpu_device *adev) err = amdgpu_ucode_validate(adev->gfx.mec2_fw); if (err) goto out; + cp_hdr = (const struct gfx_firmware_header_v1_0 *) + adev->gfx.mec2_fw->data; + adev->gfx.mec2_fw_version = le32_to_cpu( + cp_hdr->header.ucode_version); + adev->gfx.mec2_feature_version = le32_to_cpu( + cp_hdr->ucode_feature_version); } else { err = 0; adev->gfx.mec2_fw = NULL; @@ -1813,10 +1835,7 @@ static u32 gfx_v8_0_get_rb_disabled(struct amdgpu_device *adev, u32 data, mask; data = RREG32(mmCC_RB_BACKEND_DISABLE); - if (data & 1) - data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK; - else - data = 0; + data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK; data |= RREG32(mmGC_USER_RB_BACKEND_DISABLE); @@ -1986,6 +2005,7 @@ static void gfx_v8_0_gpu_init(struct amdgpu_device *adev) adev->gfx.config.max_shader_engines = 1; adev->gfx.config.max_tile_pipes = 2; adev->gfx.config.max_sh_per_se = 1; + adev->gfx.config.max_backends_per_se = 2; switch (adev->pdev->revision) { case 0xc4: @@ -1994,7 +2014,6 @@ static void gfx_v8_0_gpu_init(struct amdgpu_device *adev) case 0xcc: /* B10 */ adev->gfx.config.max_cu_per_sh = 8; - adev->gfx.config.max_backends_per_se = 2; break; case 0xc5: case 0x81: @@ -2003,14 +2022,12 @@ static void gfx_v8_0_gpu_init(struct amdgpu_device *adev) case 0xcd: /* B8 */ adev->gfx.config.max_cu_per_sh = 6; - adev->gfx.config.max_backends_per_se = 2; break; case 0xc6: case 0xca: case 0xce: /* B6 */ adev->gfx.config.max_cu_per_sh = 6; - adev->gfx.config.max_backends_per_se = 2; break; case 0xc7: case 0x87: @@ -2018,7 +2035,6 @@ static void gfx_v8_0_gpu_init(struct amdgpu_device *adev) default: /* B4 */ adev->gfx.config.max_cu_per_sh = 4; - adev->gfx.config.max_backends_per_se = 1; break; } @@ -2278,7 +2294,6 @@ static int gfx_v8_0_rlc_load_microcode(struct amdgpu_device *adev) hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data; amdgpu_ucode_print_rlc_hdr(&hdr->header); - adev->gfx.rlc_fw_version = le32_to_cpu(hdr->header.ucode_version); fw_data = (const __le32 *)(adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); @@ -2364,12 +2379,6 @@ static int gfx_v8_0_cp_gfx_load_microcode(struct amdgpu_device *adev) amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header); amdgpu_ucode_print_gfx_hdr(&ce_hdr->header); amdgpu_ucode_print_gfx_hdr(&me_hdr->header); - adev->gfx.pfp_fw_version = le32_to_cpu(pfp_hdr->header.ucode_version); - adev->gfx.ce_fw_version = le32_to_cpu(ce_hdr->header.ucode_version); - adev->gfx.me_fw_version = le32_to_cpu(me_hdr->header.ucode_version); - adev->gfx.me_feature_version = le32_to_cpu(me_hdr->ucode_feature_version); - adev->gfx.ce_feature_version = le32_to_cpu(ce_hdr->ucode_feature_version); - adev->gfx.pfp_feature_version = le32_to_cpu(pfp_hdr->ucode_feature_version); gfx_v8_0_cp_gfx_enable(adev, false); @@ -2625,7 +2634,6 @@ static int gfx_v8_0_cp_compute_load_microcode(struct amdgpu_device *adev) mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; amdgpu_ucode_print_gfx_hdr(&mec_hdr->header); - adev->gfx.mec_fw_version = le32_to_cpu(mec_hdr->header.ucode_version); fw_data = (const __le32 *) (adev->gfx.mec_fw->data + @@ -2644,7 +2652,6 @@ static int gfx_v8_0_cp_compute_load_microcode(struct amdgpu_device *adev) mec2_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; amdgpu_ucode_print_gfx_hdr(&mec2_hdr->header); - adev->gfx.mec2_fw_version = le32_to_cpu(mec2_hdr->header.ucode_version); fw_data = (const __le32 *) (adev->gfx.mec2_fw->data + @@ -3128,7 +3135,7 @@ static int gfx_v8_0_cp_compute_resume(struct amdgpu_device *adev) WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, AMDGPU_DOORBELL_KIQ << 2); WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, - 0x7FFFF << 2); + AMDGPU_DOORBELL_MEC_RING7 << 2); } tmp = RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL); tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, @@ -3756,7 +3763,7 @@ static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) amdgpu_ring_write(ring, 0x20); /* poll interval */ } -static void gfx_v8_0_ring_emit_ib(struct amdgpu_ring *ring, +static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { bool need_ctx_switch = ring->current_ctx != ib->ctx; @@ -3764,15 +3771,10 @@ static void gfx_v8_0_ring_emit_ib(struct amdgpu_ring *ring, u32 next_rptr = ring->wptr + 5; /* drop the CE preamble IB for the same context */ - if ((ring->type == AMDGPU_RING_TYPE_GFX) && - (ib->flags & AMDGPU_IB_FLAG_PREAMBLE) && - !need_ctx_switch) + if ((ib->flags & AMDGPU_IB_FLAG_PREAMBLE) && !need_ctx_switch) return; - if (ring->type == AMDGPU_RING_TYPE_COMPUTE) - control |= INDIRECT_BUFFER_VALID; - - if (need_ctx_switch && ring->type == AMDGPU_RING_TYPE_GFX) + if (need_ctx_switch) next_rptr += 2; next_rptr += 4; @@ -3783,7 +3785,7 @@ static void gfx_v8_0_ring_emit_ib(struct amdgpu_ring *ring, amdgpu_ring_write(ring, next_rptr); /* insert SWITCH_BUFFER packet before first IB in the ring frame */ - if (need_ctx_switch && ring->type == AMDGPU_RING_TYPE_GFX) { + if (need_ctx_switch) { amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); amdgpu_ring_write(ring, 0); } @@ -3806,6 +3808,36 @@ static void gfx_v8_0_ring_emit_ib(struct amdgpu_ring *ring, amdgpu_ring_write(ring, control); } +static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring, + struct amdgpu_ib *ib) +{ + u32 header, control = 0; + u32 next_rptr = ring->wptr + 5; + + control |= INDIRECT_BUFFER_VALID; + + next_rptr += 4; + amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); + amdgpu_ring_write(ring, WRITE_DATA_DST_SEL(5) | WR_CONFIRM); + amdgpu_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc); + amdgpu_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff); + amdgpu_ring_write(ring, next_rptr); + + header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); + + control |= ib->length_dw | + (ib->vm ? (ib->vm->ids[ring->idx].id << 24) : 0); + + amdgpu_ring_write(ring, header); + amdgpu_ring_write(ring, +#ifdef __BIG_ENDIAN + (2 << 0) | +#endif + (ib->gpu_addr & 0xFFFFFFFC)); + amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF); + amdgpu_ring_write(ring, control); +} + static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr, u64 seq, unsigned flags) { @@ -4227,7 +4259,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = { .get_wptr = gfx_v8_0_ring_get_wptr_gfx, .set_wptr = gfx_v8_0_ring_set_wptr_gfx, .parse_cs = NULL, - .emit_ib = gfx_v8_0_ring_emit_ib, + .emit_ib = gfx_v8_0_ring_emit_ib_gfx, .emit_fence = gfx_v8_0_ring_emit_fence_gfx, .emit_semaphore = gfx_v8_0_ring_emit_semaphore, .emit_vm_flush = gfx_v8_0_ring_emit_vm_flush, @@ -4243,7 +4275,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = { .get_wptr = gfx_v8_0_ring_get_wptr_compute, .set_wptr = gfx_v8_0_ring_set_wptr_compute, .parse_cs = NULL, - .emit_ib = gfx_v8_0_ring_emit_ib, + .emit_ib = gfx_v8_0_ring_emit_ib_compute, .emit_fence = gfx_v8_0_ring_emit_fence_compute, .emit_semaphore = gfx_v8_0_ring_emit_semaphore, .emit_vm_flush = gfx_v8_0_ring_emit_vm_flush, diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c index d7895885fe0c..a988dfb1d394 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c @@ -121,6 +121,7 @@ static int sdma_v2_4_init_microcode(struct amdgpu_device *adev) int err, i; struct amdgpu_firmware_info *info = NULL; const struct common_firmware_header *header = NULL; + const struct sdma_firmware_header_v1_0 *hdr; DRM_DEBUG("\n"); @@ -142,6 +143,9 @@ static int sdma_v2_4_init_microcode(struct amdgpu_device *adev) err = amdgpu_ucode_validate(adev->sdma[i].fw); if (err) goto out; + hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; + adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version); + adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version); if (adev->firmware.smu_load) { info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i]; @@ -541,8 +545,6 @@ static int sdma_v2_4_load_microcode(struct amdgpu_device *adev) hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; amdgpu_ucode_print_sdma_hdr(&hdr->header); fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; - adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version); - fw_data = (const __le32 *) (adev->sdma[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c index 7bb37b93993f..2b86569b18d3 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c @@ -159,6 +159,7 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev) int err, i; struct amdgpu_firmware_info *info = NULL; const struct common_firmware_header *header = NULL; + const struct sdma_firmware_header_v1_0 *hdr; DRM_DEBUG("\n"); @@ -183,6 +184,9 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev) err = amdgpu_ucode_validate(adev->sdma[i].fw); if (err) goto out; + hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; + adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version); + adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version); if (adev->firmware.smu_load) { info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i]; @@ -630,8 +634,6 @@ static int sdma_v3_0_load_microcode(struct amdgpu_device *adev) hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; amdgpu_ucode_print_sdma_hdr(&hdr->header); fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; - adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version); - fw_data = (const __le32 *) (adev->sdma[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c index d62c4002e39c..d1064ca3670e 100644 --- a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c @@ -35,6 +35,8 @@ #include "oss/oss_2_0_d.h" #include "oss/oss_2_0_sh_mask.h" #include "gca/gfx_8_0_d.h" +#include "smu/smu_7_1_2_d.h" +#include "smu/smu_7_1_2_sh_mask.h" #define GRBM_GFX_INDEX__VCE_INSTANCE__SHIFT 0x04 #define GRBM_GFX_INDEX__VCE_INSTANCE_MASK 0x10 @@ -112,6 +114,10 @@ static int vce_v3_0_start(struct amdgpu_device *adev) mutex_lock(&adev->grbm_idx_mutex); for (idx = 0; idx < 2; ++idx) { + + if (adev->vce.harvest_config & (1 << idx)) + continue; + if(idx == 0) WREG32_P(mmGRBM_GFX_INDEX, 0, ~GRBM_GFX_INDEX__VCE_INSTANCE_MASK); @@ -190,10 +196,52 @@ static int vce_v3_0_start(struct amdgpu_device *adev) return 0; } +#define ixVCE_HARVEST_FUSE_MACRO__ADDRESS 0xC0014074 +#define VCE_HARVEST_FUSE_MACRO__SHIFT 27 +#define VCE_HARVEST_FUSE_MACRO__MASK 0x18000000 + +static unsigned vce_v3_0_get_harvest_config(struct amdgpu_device *adev) +{ + u32 tmp; + unsigned ret; + + if (adev->flags & AMDGPU_IS_APU) + tmp = (RREG32_SMC(ixVCE_HARVEST_FUSE_MACRO__ADDRESS) & + VCE_HARVEST_FUSE_MACRO__MASK) >> + VCE_HARVEST_FUSE_MACRO__SHIFT; + else + tmp = (RREG32_SMC(ixCC_HARVEST_FUSES) & + CC_HARVEST_FUSES__VCE_DISABLE_MASK) >> + CC_HARVEST_FUSES__VCE_DISABLE__SHIFT; + + switch (tmp) { + case 1: + ret = AMDGPU_VCE_HARVEST_VCE0; + break; + case 2: + ret = AMDGPU_VCE_HARVEST_VCE1; + break; + case 3: + ret = AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1; + break; + default: + ret = 0; + } + + return ret; +} + static int vce_v3_0_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + adev->vce.harvest_config = vce_v3_0_get_harvest_config(adev); + + if ((adev->vce.harvest_config & + (AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1)) == + (AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1)) + return -ENOENT; + vce_v3_0_set_ring_funcs(adev); vce_v3_0_set_irq_funcs(adev); diff --git a/drivers/gpu/drm/amd/amdgpu/vi.c b/drivers/gpu/drm/amd/amdgpu/vi.c index fa5a4448531d..68552da40287 100644 --- a/drivers/gpu/drm/amd/amdgpu/vi.c +++ b/drivers/gpu/drm/amd/amdgpu/vi.c @@ -122,6 +122,32 @@ static void vi_smc_wreg(struct amdgpu_device *adev, u32 reg, u32 v) spin_unlock_irqrestore(&adev->smc_idx_lock, flags); } +/* smu_8_0_d.h */ +#define mmMP0PUB_IND_INDEX 0x180 +#define mmMP0PUB_IND_DATA 0x181 + +static u32 cz_smc_rreg(struct amdgpu_device *adev, u32 reg) +{ + unsigned long flags; + u32 r; + + spin_lock_irqsave(&adev->smc_idx_lock, flags); + WREG32(mmMP0PUB_IND_INDEX, (reg)); + r = RREG32(mmMP0PUB_IND_DATA); + spin_unlock_irqrestore(&adev->smc_idx_lock, flags); + return r; +} + +static void cz_smc_wreg(struct amdgpu_device *adev, u32 reg, u32 v) +{ + unsigned long flags; + + spin_lock_irqsave(&adev->smc_idx_lock, flags); + WREG32(mmMP0PUB_IND_INDEX, (reg)); + WREG32(mmMP0PUB_IND_DATA, (v)); + spin_unlock_irqrestore(&adev->smc_idx_lock, flags); +} + static u32 vi_uvd_ctx_rreg(struct amdgpu_device *adev, u32 reg) { unsigned long flags; @@ -1222,8 +1248,13 @@ static int vi_common_early_init(void *handle) bool smc_enabled = false; struct amdgpu_device *adev = (struct amdgpu_device *)handle; - adev->smc_rreg = &vi_smc_rreg; - adev->smc_wreg = &vi_smc_wreg; + if (adev->flags & AMDGPU_IS_APU) { + adev->smc_rreg = &cz_smc_rreg; + adev->smc_wreg = &cz_smc_wreg; + } else { + adev->smc_rreg = &vi_smc_rreg; + adev->smc_wreg = &vi_smc_wreg; + } adev->pcie_rreg = &vi_pcie_rreg; adev->pcie_wreg = &vi_pcie_wreg; adev->uvd_ctx_rreg = &vi_uvd_ctx_rreg; diff --git a/drivers/gpu/drm/amd/amdgpu/vid.h b/drivers/gpu/drm/amd/amdgpu/vid.h index 31bb89452e12..d98aa9d82fa1 100644 --- a/drivers/gpu/drm/amd/amdgpu/vid.h +++ b/drivers/gpu/drm/amd/amdgpu/vid.h @@ -66,6 +66,11 @@ #define AMDGPU_NUM_OF_VMIDS 8 +#define PIPEID(x) ((x) << 0) +#define MEID(x) ((x) << 2) +#define VMID(x) ((x) << 4) +#define QUEUEID(x) ((x) << 8) + #define RB_BITMAP_WIDTH_PER_SH 2 #define MC_SEQ_MISC0__MT__MASK 0xf0000000 |