diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h | 289 |
1 files changed, 258 insertions, 31 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h index 86ec9d0d12c8..efd61a1ccc66 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h @@ -34,6 +34,7 @@ #include "soc15.h" #include "amdgpu_ras.h" #include "amdgpu_ring_mux.h" +#include "amdgpu_xcp.h" /* GFX current status */ #define AMDGPU_GFX_NORMAL_MODE 0x00000000L @@ -42,8 +43,11 @@ #define AMDGPU_GFX_CG_DISABLED_MODE 0x00000004L #define AMDGPU_GFX_LBPW_DISABLED_MODE 0x00000008L -#define AMDGPU_MAX_GFX_QUEUES KGD_MAX_QUEUES -#define AMDGPU_MAX_COMPUTE_QUEUES KGD_MAX_QUEUES +#define AMDGPU_MAX_GC_INSTANCES 8 +#define AMDGPU_MAX_QUEUES 128 + +#define AMDGPU_MAX_GFX_QUEUES AMDGPU_MAX_QUEUES +#define AMDGPU_MAX_COMPUTE_QUEUES AMDGPU_MAX_QUEUES enum amdgpu_gfx_pipe_priority { AMDGPU_GFX_PIPE_PRIO_NORMAL = AMDGPU_RING_PRIO_1, @@ -53,6 +57,48 @@ enum amdgpu_gfx_pipe_priority { #define AMDGPU_GFX_QUEUE_PRIORITY_MINIMUM 0 #define AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM 15 +/* 1 second timeout */ +#define GFX_PROFILE_IDLE_TIMEOUT msecs_to_jiffies(1000) + +enum amdgpu_gfx_partition { + AMDGPU_SPX_PARTITION_MODE = 0, + AMDGPU_DPX_PARTITION_MODE = 1, + AMDGPU_TPX_PARTITION_MODE = 2, + AMDGPU_QPX_PARTITION_MODE = 3, + AMDGPU_CPX_PARTITION_MODE = 4, + AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE = -1, + /* Automatically choose the right mode */ + AMDGPU_AUTO_COMPUTE_PARTITION_MODE = -2, +}; + +#define NUM_XCC(x) hweight16(x) + +enum amdgpu_gfx_ras_mem_id_type { + AMDGPU_GFX_CP_MEM = 0, + AMDGPU_GFX_GCEA_MEM, + AMDGPU_GFX_GC_CANE_MEM, + AMDGPU_GFX_GCUTCL2_MEM, + AMDGPU_GFX_GDS_MEM, + AMDGPU_GFX_LDS_MEM, + AMDGPU_GFX_RLC_MEM, + AMDGPU_GFX_SP_MEM, + AMDGPU_GFX_SPI_MEM, + AMDGPU_GFX_SQC_MEM, + AMDGPU_GFX_SQ_MEM, + AMDGPU_GFX_TA_MEM, + AMDGPU_GFX_TCC_MEM, + AMDGPU_GFX_TCA_MEM, + AMDGPU_GFX_TCI_MEM, + AMDGPU_GFX_TCP_MEM, + AMDGPU_GFX_TD_MEM, + AMDGPU_GFX_TCX_MEM, + AMDGPU_GFX_ATC_L2_MEM, + AMDGPU_GFX_UTCL2_MEM, + AMDGPU_GFX_VML2_MEM, + AMDGPU_GFX_VML2_WALKER_MEM, + AMDGPU_GFX_MEM_TYPE_NUM +}; + struct amdgpu_mec { struct amdgpu_bo *hpd_eop_obj; u64 hpd_eop_gpu_addr; @@ -64,8 +110,10 @@ struct amdgpu_mec { u32 num_mec; u32 num_pipe_per_mec; u32 num_queue_per_pipe; - void *mqd_backup[AMDGPU_MAX_COMPUTE_RINGS + 1]; + void *mqd_backup[AMDGPU_MAX_COMPUTE_RINGS * AMDGPU_MAX_GC_INSTANCES]; +}; +struct amdgpu_mec_bitmap { /* These are the resources for which amdgpu takes ownership */ DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); }; @@ -94,6 +142,10 @@ struct kiq_pm4_funcs { void (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring, uint16_t pasid, uint32_t flush_type, bool all_hub); + void (*kiq_reset_hw_queue)(struct amdgpu_ring *kiq_ring, + uint32_t queue_type, uint32_t me_id, + uint32_t pipe_id, uint32_t queue_id, + uint32_t xcc_id, uint32_t vmid); /* Packet sizes */ int set_resources_size; int map_queues_size; @@ -109,6 +161,7 @@ struct amdgpu_kiq { struct amdgpu_ring ring; struct amdgpu_irq_src irq; const struct kiq_pm4_funcs *pmf; + void *mqd_backup; }; /* @@ -117,10 +170,46 @@ struct amdgpu_kiq { #define AMDGPU_GFX_MAX_SE 4 #define AMDGPU_GFX_MAX_SH_PER_SE 2 +/** + * amdgpu_rb_config - Configure a single Render Backend (RB) + * + * Bad RBs are fused off and there is a harvest register the driver reads to + * determine which RB(s) are fused off so that the driver can configure the + * hardware state so that nothing gets sent to them. There are also user + * harvest registers that the driver can program to disable additional RBs, + * etc., for testing purposes. + */ struct amdgpu_rb_config { + /** + * @rb_backend_disable: + * + * The value captured from register RB_BACKEND_DISABLE indicates if the + * RB backend is disabled or not. + */ uint32_t rb_backend_disable; + + /** + * @user_rb_backend_disable: + * + * The value captured from register USER_RB_BACKEND_DISABLE indicates + * if the User RB backend is disabled or not. + */ uint32_t user_rb_backend_disable; + + /** + * @raster_config: + * + * To set up all of the states, it is necessary to have two registers + * to keep all of the states. This field holds the first register. + */ uint32_t raster_config; + + /** + * @raster_config_1: + * + * To set up all of the states, it is necessary to have two registers + * to keep all of the states. This field holds the second register. + */ uint32_t raster_config_1; }; @@ -168,6 +257,13 @@ struct amdgpu_gfx_config { uint32_t macrotile_mode_array[16]; struct gb_addr_config gb_addr_config_fields; + + /** + * @rb_config: + * + * Matrix that keeps all the Render Backend (color and depth buffer + * handling) configuration on the 3D engine. + */ struct amdgpu_rb_config rb_config[AMDGPU_GFX_MAX_SE][AMDGPU_GFX_MAX_SH_PER_SE]; /* gfx configure feature */ @@ -178,6 +274,8 @@ struct amdgpu_gfx_config { uint32_t num_sc_per_sh; uint32_t num_packer_per_sc; uint32_t pa_sc_tile_steering_override; + /* Whether texture coordinate truncation is conformant. */ + bool ta_cntl2_truncate_coord_mode; uint64_t tcc_disabled_mask; uint32_t gc_num_tcp_per_sa; uint32_t gc_num_sdp_interface; @@ -190,6 +288,15 @@ struct amdgpu_gfx_config { uint32_t gc_gl1c_per_sa; uint32_t gc_gl1c_size_per_instance; uint32_t gc_gl2c_per_gpu; + uint32_t gc_tcp_size_per_cu; + uint32_t gc_num_cu_per_sqc; + uint32_t gc_tcc_size; + uint32_t gc_tcp_cache_line_size; + uint32_t gc_instruction_cache_size_per_sqc; + uint32_t gc_instruction_cache_line_size; + uint32_t gc_scalar_data_cache_size_per_sqc; + uint32_t gc_scalar_data_cache_line_size; + uint32_t gc_tcc_cache_line_size; }; struct amdgpu_cu_info { @@ -203,13 +310,12 @@ struct amdgpu_cu_info { uint32_t number; uint32_t ao_cu_mask; uint32_t ao_cu_bitmap[4][4]; - uint32_t bitmap[4][4]; + uint32_t bitmap[AMDGPU_MAX_GC_INSTANCES][4][4]; }; struct amdgpu_gfx_ras { struct amdgpu_ras_block_object ras_block; void (*enable_watchdog_timer)(struct amdgpu_device *adev); - bool (*query_utcl2_poison_status)(struct amdgpu_device *adev); int (*rlc_gc_fed_irq)(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry); @@ -217,23 +323,39 @@ struct amdgpu_gfx_ras { struct amdgpu_iv_entry *entry); }; +struct amdgpu_gfx_shadow_info { + u32 shadow_size; + u32 shadow_alignment; + u32 csa_size; + u32 csa_alignment; +}; + struct amdgpu_gfx_funcs { /* get the gpu clock counter */ uint64_t (*get_gpu_clock_counter)(struct amdgpu_device *adev); void (*select_se_sh)(struct amdgpu_device *adev, u32 se_num, - u32 sh_num, u32 instance); - void (*read_wave_data)(struct amdgpu_device *adev, uint32_t simd, + u32 sh_num, u32 instance, int xcc_id); + void (*read_wave_data)(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields); - void (*read_wave_vgprs)(struct amdgpu_device *adev, uint32_t simd, + void (*read_wave_vgprs)(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t thread, uint32_t start, uint32_t size, uint32_t *dst); - void (*read_wave_sgprs)(struct amdgpu_device *adev, uint32_t simd, + void (*read_wave_sgprs)(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t start, uint32_t size, uint32_t *dst); void (*select_me_pipe_q)(struct amdgpu_device *adev, u32 me, u32 pipe, - u32 queue, u32 vmid); + u32 queue, u32 vmid, u32 xcc_id); void (*init_spm_golden)(struct amdgpu_device *adev); void (*update_perfmon_mgcg)(struct amdgpu_device *adev, bool enable); + int (*get_gfx_shadow_info)(struct amdgpu_device *adev, + struct amdgpu_gfx_shadow_info *shadow_info, + bool skip_check); + enum amdgpu_gfx_partition + (*query_partition_mode)(struct amdgpu_device *adev); + int (*switch_partition_mode)(struct amdgpu_device *adev, + int num_xccs_per_xcp); + int (*ih_node_to_logical_xcc)(struct amdgpu_device *adev, int ih_node); + int (*get_xccs_per_xcp)(struct amdgpu_device *adev); }; struct sq_work { @@ -275,6 +397,12 @@ struct amdgpu_me { DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_GFX_QUEUES); }; +struct amdgpu_isolation_work { + struct amdgpu_device *adev; + u32 xcp_id; + struct delayed_work work; +}; + struct amdgpu_gfx { struct mutex gpu_clock_mutex; struct amdgpu_gfx_config config; @@ -283,7 +411,8 @@ struct amdgpu_gfx { struct amdgpu_ce ce; struct amdgpu_me me; struct amdgpu_mec mec; - struct amdgpu_kiq kiq; + struct amdgpu_mec_bitmap mec_bitmap[AMDGPU_MAX_GC_INSTANCES]; + struct amdgpu_kiq kiq[AMDGPU_MAX_GC_INSTANCES]; struct amdgpu_imu imu; bool rs64_enable; /* firmware format */ const struct firmware *me_fw; /* ME firmware */ @@ -321,11 +450,12 @@ struct amdgpu_gfx { bool cp_fw_write_wait; struct amdgpu_ring gfx_ring[AMDGPU_MAX_GFX_RINGS]; unsigned num_gfx_rings; - struct amdgpu_ring compute_ring[AMDGPU_MAX_COMPUTE_RINGS]; + struct amdgpu_ring compute_ring[AMDGPU_MAX_COMPUTE_RINGS * AMDGPU_MAX_GC_INSTANCES]; unsigned num_compute_rings; struct amdgpu_irq_src eop_irq; struct amdgpu_irq_src priv_reg_irq; struct amdgpu_irq_src priv_inst_irq; + struct amdgpu_irq_src bad_op_irq; struct amdgpu_irq_src cp_ecc_error_irq; struct amdgpu_irq_src sq_irq; struct amdgpu_irq_src rlc_gc_fed_irq; @@ -341,6 +471,8 @@ struct amdgpu_gfx { /* reset mask */ uint32_t grbm_soft_reset; uint32_t srbm_soft_reset; + uint32_t gfx_supported_reset; + uint32_t compute_supported_reset; /* gfx off */ bool gfx_off_state; /* true: enabled, false: disabled */ @@ -362,12 +494,63 @@ struct amdgpu_gfx { struct amdgpu_ring sw_gfx_ring[AMDGPU_MAX_SW_GFX_RINGS]; struct amdgpu_ring_mux muxer; + + bool cp_gfx_shadow; /* for gfx11 */ + + uint16_t xcc_mask; + uint32_t num_xcc_per_xcp; + struct mutex partition_mutex; + bool mcbp; /* mid command buffer preemption */ + + /* IP reg dump */ + uint32_t *ip_dump_core; + uint32_t *ip_dump_compute_queues; + uint32_t *ip_dump_gfx_queues; + + struct mutex reset_sem_mutex; + + /* cleaner shader */ + struct amdgpu_bo *cleaner_shader_obj; + unsigned int cleaner_shader_size; + u64 cleaner_shader_gpu_addr; + void *cleaner_shader_cpu_ptr; + const void *cleaner_shader_ptr; + bool enable_cleaner_shader; + struct amdgpu_isolation_work enforce_isolation[MAX_XCP]; + /* Mutex for synchronizing KFD scheduler operations */ + struct mutex userq_sch_mutex; + u64 userq_sch_req_count[MAX_XCP]; + bool userq_sch_inactive[MAX_XCP]; + unsigned long enforce_isolation_jiffies[MAX_XCP]; + unsigned long enforce_isolation_time[MAX_XCP]; + + atomic_t total_submission_cnt; + struct delayed_work idle_work; + bool workload_profile_active; + struct mutex workload_profile_mutex; + + bool disable_kq; + bool disable_uq; }; +struct amdgpu_gfx_ras_reg_entry { + struct amdgpu_ras_err_status_reg_entry reg_entry; + enum amdgpu_gfx_ras_mem_id_type mem_id_type; + uint32_t se_num; +}; + +struct amdgpu_gfx_ras_mem_id_entry { + const struct amdgpu_ras_memory_id_entry *mem_id_ent; + uint32_t size; +}; + +#define AMDGPU_GFX_MEMID_ENT(x) {(x), ARRAY_SIZE(x)}, + #define amdgpu_gfx_get_gpu_clock_counter(adev) (adev)->gfx.funcs->get_gpu_clock_counter((adev)) -#define amdgpu_gfx_select_se_sh(adev, se, sh, instance) (adev)->gfx.funcs->select_se_sh((adev), (se), (sh), (instance)) -#define amdgpu_gfx_select_me_pipe_q(adev, me, pipe, q, vmid) (adev)->gfx.funcs->select_me_pipe_q((adev), (me), (pipe), (q), (vmid)) +#define amdgpu_gfx_select_se_sh(adev, se, sh, instance, xcc_id) ((adev)->gfx.funcs->select_se_sh((adev), (se), (sh), (instance), (xcc_id))) +#define amdgpu_gfx_select_me_pipe_q(adev, me, pipe, q, vmid, xcc_id) ((adev)->gfx.funcs->select_me_pipe_q((adev), (me), (pipe), (q), (vmid), (xcc_id))) #define amdgpu_gfx_init_spm_golden(adev) (adev)->gfx.funcs->init_spm_golden((adev)) +#define amdgpu_gfx_get_gfx_shadow_info(adev, si) ((adev)->gfx.funcs->get_gfx_shadow_info((adev), (si), false)) /** * amdgpu_gfx_create_bitmask - create a bitmask @@ -385,21 +568,21 @@ static inline u32 amdgpu_gfx_create_bitmask(u32 bit_width) void amdgpu_gfx_parse_disable_cu(unsigned *mask, unsigned max_se, unsigned max_sh); -int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev, - struct amdgpu_ring *ring, - struct amdgpu_irq_src *irq); +int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev, int xcc_id); void amdgpu_gfx_kiq_free_ring(struct amdgpu_ring *ring); -void amdgpu_gfx_kiq_fini(struct amdgpu_device *adev); +void amdgpu_gfx_kiq_fini(struct amdgpu_device *adev, int xcc_id); int amdgpu_gfx_kiq_init(struct amdgpu_device *adev, - unsigned hpd_size); + unsigned hpd_size, int xcc_id); int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev, - unsigned mqd_size); -void amdgpu_gfx_mqd_sw_fini(struct amdgpu_device *adev); -int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev); -int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev); + unsigned mqd_size, int xcc_id); +void amdgpu_gfx_mqd_sw_fini(struct amdgpu_device *adev, int xcc_id); +int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev, int xcc_id); +int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev, int xcc_id); +int amdgpu_gfx_disable_kgq(struct amdgpu_device *adev, int xcc_id); +int amdgpu_gfx_enable_kgq(struct amdgpu_device *adev, int xcc_id); void amdgpu_gfx_compute_queue_acquire(struct amdgpu_device *adev); void amdgpu_gfx_graphics_queue_acquire(struct amdgpu_device *adev); @@ -408,19 +591,16 @@ int amdgpu_gfx_mec_queue_to_bit(struct amdgpu_device *adev, int mec, int pipe, int queue); void amdgpu_queue_mask_bit_to_mec_queue(struct amdgpu_device *adev, int bit, int *mec, int *pipe, int *queue); -bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev, int mec, - int pipe, int queue); +bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev, int xcc_id, + int mec, int pipe, int queue); bool amdgpu_gfx_is_high_priority_compute_queue(struct amdgpu_device *adev, struct amdgpu_ring *ring); bool amdgpu_gfx_is_high_priority_graphics_queue(struct amdgpu_device *adev, struct amdgpu_ring *ring); -int amdgpu_gfx_me_queue_to_bit(struct amdgpu_device *adev, int me, - int pipe, int queue); -void amdgpu_gfx_bit_to_me_queue(struct amdgpu_device *adev, int bit, - int *me, int *pipe, int *queue); bool amdgpu_gfx_is_me_queue_enabled(struct amdgpu_device *adev, int me, int pipe, int queue); void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable); +void amdgpu_gfx_off_ctrl_immediate(struct amdgpu_device *adev, bool enable); int amdgpu_get_gfx_off_status(struct amdgpu_device *adev, uint32_t *value); int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block); void amdgpu_gfx_ras_fini(struct amdgpu_device *adev); @@ -433,12 +613,59 @@ int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev, int amdgpu_gfx_cp_ecc_error_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry); -uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg); -void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v); +uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg, uint32_t xcc_id); +void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, uint32_t xcc_id); +int amdgpu_kiq_hdp_flush(struct amdgpu_device *adev); int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev); void amdgpu_gfx_cp_init_microcode(struct amdgpu_device *adev, uint32_t ucode_id); int amdgpu_gfx_ras_sw_init(struct amdgpu_device *adev); int amdgpu_gfx_poison_consumption_handler(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry); + +bool amdgpu_gfx_is_master_xcc(struct amdgpu_device *adev, int xcc_id); +int amdgpu_gfx_sysfs_init(struct amdgpu_device *adev); +void amdgpu_gfx_sysfs_fini(struct amdgpu_device *adev); +void amdgpu_gfx_ras_error_func(struct amdgpu_device *adev, + void *ras_error_status, + void (*func)(struct amdgpu_device *adev, void *ras_error_status, + int xcc_id)); +int amdgpu_gfx_cleaner_shader_sw_init(struct amdgpu_device *adev, + unsigned int cleaner_shader_size); +void amdgpu_gfx_cleaner_shader_sw_fini(struct amdgpu_device *adev); +void amdgpu_gfx_cleaner_shader_init(struct amdgpu_device *adev, + unsigned int cleaner_shader_size, + const void *cleaner_shader_ptr); +void amdgpu_gfx_enforce_isolation_handler(struct work_struct *work); +void amdgpu_gfx_enforce_isolation_ring_begin_use(struct amdgpu_ring *ring); +void amdgpu_gfx_enforce_isolation_ring_end_use(struct amdgpu_ring *ring); + +void amdgpu_gfx_profile_idle_work_handler(struct work_struct *work); +void amdgpu_gfx_profile_ring_begin_use(struct amdgpu_ring *ring); +void amdgpu_gfx_profile_ring_end_use(struct amdgpu_ring *ring); +u32 amdgpu_gfx_csb_preamble_start(u32 *buffer); +u32 amdgpu_gfx_csb_data_parser(struct amdgpu_device *adev, u32 *buffer, u32 count); +void amdgpu_gfx_csb_preamble_end(u32 *buffer, u32 count); + +void amdgpu_debugfs_gfx_sched_mask_init(struct amdgpu_device *adev); +void amdgpu_debugfs_compute_sched_mask_init(struct amdgpu_device *adev); + +static inline const char *amdgpu_gfx_compute_mode_desc(int mode) +{ + switch (mode) { + case AMDGPU_SPX_PARTITION_MODE: + return "SPX"; + case AMDGPU_DPX_PARTITION_MODE: + return "DPX"; + case AMDGPU_TPX_PARTITION_MODE: + return "TPX"; + case AMDGPU_QPX_PARTITION_MODE: + return "QPX"; + case AMDGPU_CPX_PARTITION_MODE: + return "CPX"; + default: + return "UNKNOWN"; + } +} + #endif |