| 1 |
|
| 2 | |
| 3 | * Copyright (C) 2013 Red Hat
|
| 4 | * Author: Rob Clark <[email protected]>
|
| 5 |
|
| 6 |
|
| 7 | #ifndef __MSM_RINGBUFFER_H__
|
| 8 | #define __MSM_RINGBUFFER_H__
|
| 9 |
|
| 10 | #include "drm/gpu_scheduler.h"
|
| 11 | #include "msm_drv.h"
|
| 12 |
|
| 13 | #define rbmemptr(ring, member) \
|
| 14 | ((ring)->memptrs_iova + offsetof(struct msm_rbmemptrs, member))
|
| 15 |
|
| 16 | #define rbmemptr_stats(ring, index, member) \
|
| 17 | (rbmemptr((ring), stats) + \
|
| 18 | ((index) * sizeof(struct msm_gpu_submit_stats)) + \
|
| 19 | offsetof(struct msm_gpu_submit_stats, member))
|
| 20 |
|
| 21 | struct msm_gpu_submit_stats {
|
| 22 | u64 cpcycles_start;
|
| 23 | u64 cpcycles_end;
|
| 24 | u64 alwayson_start;
|
| 25 | u64 alwayson_end;
|
| 26 | };
|
| 27 |
|
| 28 | #define MSM_GPU_SUBMIT_STATS_COUNT 64
|
| 29 |
|
| 30 | struct msm_rbmemptrs {
|
| 31 | volatile uint32_t rptr;
|
| 32 | volatile uint32_t fence;
|
| 33 |
|
| 34 | volatile uint32_t bv_rptr;
|
| 35 | volatile uint32_t bv_fence;
|
| 36 |
|
| 37 | volatile struct msm_gpu_submit_stats stats[MSM_GPU_SUBMIT_STATS_COUNT];
|
| 38 | volatile u64 ttbr0;
|
| 39 | volatile u32 context_idr;
|
| 40 |
|
| 41 | volatile u32 perfcntr_fence;
|
| 42 | };
|
| 43 |
|
| 44 | struct msm_cp_state {
|
| 45 | uint64_t ib1_base, ib2_base;
|
| 46 | uint32_t ib1_rem, ib2_rem;
|
| 47 | };
|
| 48 |
|
| 49 | struct msm_ringbuffer {
|
| 50 | struct msm_gpu *gpu;
|
| 51 | int id;
|
| 52 | struct drm_gem_object *bo;
|
| 53 | uint32_t *start, *end, *cur, *next;
|
| 54 |
|
| 55 | |
| 56 | * The job scheduler for this ring.
|
| 57 |
|
| 58 | struct rcu_head rcu;
|
| 59 | struct drm_gpu_scheduler sched;
|
| 60 | bool sched_initialized;
|
| 61 |
|
| 62 | |
| 63 | * List of in-flight submits on this ring. Protected by submit_lock.
|
| 64 | *
|
| 65 | * Currently just submits that are already written into the ring, not
|
| 66 | * submits that are still in drm_gpu_scheduler's queues. At a later
|
| 67 | * step we could probably move to letting drm_gpu_scheduler manage
|
| 68 | * hangcheck detection and keep track of submit jobs that are in-
|
| 69 | * flight.
|
| 70 |
|
| 71 | struct list_head submits;
|
| 72 | spinlock_t submit_lock;
|
| 73 |
|
| 74 | uint64_t iova;
|
| 75 | uint32_t hangcheck_fence;
|
| 76 | struct msm_rbmemptrs *memptrs;
|
| 77 | uint64_t memptrs_iova;
|
| 78 | struct msm_fence_context *fctx;
|
| 79 |
|
| 80 | |
| 81 | * hangcheck_progress_retries:
|
| 82 | *
|
| 83 | * The number of extra hangcheck duration cycles that we have given
|
| 84 | * due to it appearing that the GPU is making forward progress.
|
| 85 | *
|
| 86 | * For GPU generations which support progress detection (see.
|
| 87 | * msm_gpu_funcs::progress()), if the GPU appears to be making progress
|
| 88 | * (ie. the CP has advanced in the command stream, we'll allow up to
|
| 89 | * DRM_MSM_HANGCHECK_PROGRESS_RETRIES expirations of the hangcheck timer
|
| 90 | * before killing the job. But to detect progress we need two sample
|
| 91 | * points, so the duration of the hangcheck timer is halved. In other
|
| 92 | * words we'll let the submit run for up to:
|
| 93 | *
|
| 94 | * (DRM_MSM_HANGCHECK_DEFAULT_PERIOD / 2) * (DRM_MSM_HANGCHECK_PROGRESS_RETRIES + 1)
|
| 95 |
|
| 96 | int hangcheck_progress_retries;
|
| 97 |
|
| 98 | |
| 99 | * last_cp_state: The state of the CP at the last call to gpu->progress()
|
| 100 |
|
| 101 | struct msm_cp_state last_cp_state;
|
| 102 |
|
| 103 | |
| 104 | * preempt_lock protects preemption and serializes wptr updates against
|
| 105 | * preemption. Can be aquired from irq context.
|
| 106 |
|
| 107 | spinlock_t preempt_lock;
|
| 108 |
|
| 109 | |
| 110 | * Whether we skipped writing wptr and it needs to be updated in the
|
| 111 | * future when the ring becomes current.
|
| 112 |
|
| 113 | bool restore_wptr;
|
| 114 |
|
| 115 | |
| 116 | * cur_ctx_seqno:
|
| 117 | *
|
| 118 | * The ctx->seqno value of the last context to submit to this ring
|
| 119 | * Tracked by seqno rather than pointer value to avoid dangling
|
| 120 | * pointers, and cases where a ctx can be freed and a new one created
|
| 121 | * with the same address.
|
| 122 |
|
| 123 | int cur_ctx_seqno;
|
| 124 | };
|
| 125 |
|
| 126 | struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int id,
|
| 127 | void *memptrs, uint64_t memptrs_iova);
|
| 128 | void msm_ringbuffer_destroy(struct msm_ringbuffer *ring);
|
| 129 |
|
| 130 |
|
| 131 |
|
| 132 | static inline void
|
| 133 | OUT_RING(struct msm_ringbuffer *ring, uint32_t data)
|
| 134 | {
|
| 135 | |
| 136 | * ring->next points to the current command being written - it won't be
|
| 137 | * committed as ring->cur until the flush
|
| 138 |
|
| 139 | if (ring->next == ring->end)
|
| 140 | ring->next = ring->start;
|
| 141 | *(ring->next++) = data;
|
| 142 | }
|
| 143 |
|
| 144 | #endif
|