1 // SPDX-License-Identifier: GPL-2.0+
2 /* Copyright (C) 2015-2018 Broadcom */
4 #include <linux/delay.h>
5 #include <linux/mutex.h>
6 #include <linux/spinlock_types.h>
7 #include <linux/workqueue.h>
9 #include <drm/drm_encoder.h>
10 #include <drm/drm_gem.h>
11 #include <drm/drm_gem_shmem_helper.h>
12 #include <drm/gpu_scheduler.h>
14 #include "uapi/drm/v3d_drm.h"
18 struct platform_device;
21 #define GMP_GRANULARITY (128 * 1024)
23 /* Enum for each of the V3D queues. */
32 #define V3D_MAX_QUEUES (V3D_CACHE_CLEAN + 1)
34 struct v3d_queue_state {
35 struct drm_gpu_scheduler sched;
42 struct drm_device drm;
44 /* Short representation (e.g. 33, 41) of the V3D tech version
51 struct platform_device *pdev;
52 void __iomem *hub_regs;
53 void __iomem *core_regs[3];
54 void __iomem *bridge_regs;
55 void __iomem *gca_regs;
57 struct reset_control *reset;
59 /* Virtual and DMA addresses of the single shared page table. */
63 /* Virtual and DMA addresses of the MMU's scratch page. When
64 * a read or write is invalid in the MMU, it will be
68 dma_addr_t mmu_scratch_paddr;
69 /* virtual address bits from V3D to the MMU. */
72 /* Number of V3D cores. */
75 /* Allocator managing the address space. All units are in
81 struct work_struct overflow_mem_work;
83 struct v3d_bin_job *bin_job;
84 struct v3d_render_job *render_job;
85 struct v3d_tfu_job *tfu_job;
86 struct v3d_csd_job *csd_job;
88 struct v3d_queue_state queue[V3D_MAX_QUEUES];
90 /* Spinlock used to synchronize the overflow memory
91 * management against bin job submission.
95 /* Protects bo_stats */
98 /* Lock taken when resetting the GPU, to keep multiple
99 * processes from trying to park the scheduler threads and
102 struct mutex reset_lock;
104 /* Lock taken when creating and pushing the GPU scheduler
105 * jobs, to keep the sched-fence seqnos in order.
107 struct mutex sched_lock;
109 /* Lock taken during a cache clean and when initiating an L2
110 * flush, to keep L2 flushes from interfering with the
111 * synchronous L2 cleans.
113 struct mutex cache_clean_lock;
121 static inline struct v3d_dev *
122 to_v3d_dev(struct drm_device *dev)
124 return (struct v3d_dev *)dev->dev_private;
128 v3d_has_csd(struct v3d_dev *v3d)
130 return v3d->ver >= 41;
133 /* The per-fd struct, which tracks the MMU mappings. */
134 struct v3d_file_priv {
137 struct drm_sched_entity sched_entity[V3D_MAX_QUEUES];
141 struct drm_gem_shmem_object base;
143 struct drm_mm_node node;
145 /* List entry for the BO's position in
146 * v3d_render_job->unref_list
148 struct list_head unref_head;
151 static inline struct v3d_bo *
152 to_v3d_bo(struct drm_gem_object *bo)
154 return (struct v3d_bo *)bo;
158 struct dma_fence base;
159 struct drm_device *dev;
160 /* v3d seqno for signaled() test */
162 enum v3d_queue queue;
165 static inline struct v3d_fence *
166 to_v3d_fence(struct dma_fence *fence)
168 return (struct v3d_fence *)fence;
171 #define V3D_READ(offset) readl(v3d->hub_regs + offset)
172 #define V3D_WRITE(offset, val) writel(val, v3d->hub_regs + offset)
174 #define V3D_BRIDGE_READ(offset) readl(v3d->bridge_regs + offset)
175 #define V3D_BRIDGE_WRITE(offset, val) writel(val, v3d->bridge_regs + offset)
177 #define V3D_GCA_READ(offset) readl(v3d->gca_regs + offset)
178 #define V3D_GCA_WRITE(offset, val) writel(val, v3d->gca_regs + offset)
180 #define V3D_CORE_READ(core, offset) readl(v3d->core_regs[core] + offset)
181 #define V3D_CORE_WRITE(core, offset, val) writel(val, v3d->core_regs[core] + offset)
184 struct drm_sched_job base;
186 struct kref refcount;
190 /* This is the array of BOs that were looked up at the start
193 struct drm_gem_object **bo;
196 /* Array of struct dma_fence * to block on before submitting this job.
199 unsigned long last_dep;
201 /* v3d fence to be signaled by IRQ handler when the job is complete. */
202 struct dma_fence *irq_fence;
204 /* scheduler fence for when the job is considered complete and
205 * the BO reservations can be released.
207 struct dma_fence *done_fence;
209 /* Callback for the freeing of the job on refcount going to 0. */
210 void (*free)(struct kref *ref);
216 /* GPU virtual addresses of the start/end of the CL job. */
219 u32 timedout_ctca, timedout_ctra;
221 /* Corresponding render job, for attaching our overflow memory. */
222 struct v3d_render_job *render;
224 /* Submitted tile memory allocation start/size, tile state. */
228 struct v3d_render_job {
231 /* GPU virtual addresses of the start/end of the CL job. */
234 u32 timedout_ctca, timedout_ctra;
236 /* List of overflow BOs used in the job that need to be
237 * released once the job is complete.
239 struct list_head unref_list;
245 struct drm_v3d_submit_tfu args;
251 u32 timedout_batches;
253 struct drm_v3d_submit_csd args;
257 * __wait_for - magic wait macro
259 * Macro to help avoid open coding check/wait/timeout patterns. Note that it's
260 * important that we check the condition again after having timed out, since the
261 * timeout could be due to preemption or similar and we've never had a chance to
262 * check the condition before the timeout.
264 #define __wait_for(OP, COND, US, Wmin, Wmax) ({ \
265 const ktime_t end__ = ktime_add_ns(ktime_get_raw(), 1000ll * (US)); \
266 long wait__ = (Wmin); /* recommended min for usleep is 10 us */ \
270 const bool expired__ = ktime_after(ktime_get_raw(), end__); \
272 /* Guarantee COND check prior to timeout */ \
279 ret__ = -ETIMEDOUT; \
282 usleep_range(wait__, wait__ * 2); \
283 if (wait__ < (Wmax)) \
289 #define _wait_for(COND, US, Wmin, Wmax) __wait_for(, (COND), (US), (Wmin), \
291 #define wait_for(COND, MS) _wait_for((COND), (MS) * 1000, 10, 1000)
293 static inline unsigned long nsecs_to_jiffies_timeout(const u64 n)
295 /* nsecs_to_jiffies64() does not guard against overflow */
296 if (NSEC_PER_SEC % HZ &&
297 div_u64(n, NSEC_PER_SEC) >= MAX_JIFFY_OFFSET / HZ)
298 return MAX_JIFFY_OFFSET;
300 return min_t(u64, MAX_JIFFY_OFFSET, nsecs_to_jiffies64(n) + 1);
304 struct drm_gem_object *v3d_create_object(struct drm_device *dev, size_t size);
305 void v3d_free_object(struct drm_gem_object *gem_obj);
306 struct v3d_bo *v3d_bo_create(struct drm_device *dev, struct drm_file *file_priv,
308 int v3d_create_bo_ioctl(struct drm_device *dev, void *data,
309 struct drm_file *file_priv);
310 int v3d_mmap_bo_ioctl(struct drm_device *dev, void *data,
311 struct drm_file *file_priv);
312 int v3d_get_bo_offset_ioctl(struct drm_device *dev, void *data,
313 struct drm_file *file_priv);
314 struct drm_gem_object *v3d_prime_import_sg_table(struct drm_device *dev,
315 struct dma_buf_attachment *attach,
316 struct sg_table *sgt);
319 void v3d_debugfs_init(struct drm_minor *minor);
322 extern const struct dma_fence_ops v3d_fence_ops;
323 struct dma_fence *v3d_fence_create(struct v3d_dev *v3d, enum v3d_queue queue);
326 int v3d_gem_init(struct drm_device *dev);
327 void v3d_gem_destroy(struct drm_device *dev);
328 int v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
329 struct drm_file *file_priv);
330 int v3d_submit_tfu_ioctl(struct drm_device *dev, void *data,
331 struct drm_file *file_priv);
332 int v3d_submit_csd_ioctl(struct drm_device *dev, void *data,
333 struct drm_file *file_priv);
334 int v3d_wait_bo_ioctl(struct drm_device *dev, void *data,
335 struct drm_file *file_priv);
336 void v3d_job_put(struct v3d_job *job);
337 void v3d_reset(struct v3d_dev *v3d);
338 void v3d_invalidate_caches(struct v3d_dev *v3d);
339 void v3d_clean_caches(struct v3d_dev *v3d);
342 int v3d_irq_init(struct v3d_dev *v3d);
343 void v3d_irq_enable(struct v3d_dev *v3d);
344 void v3d_irq_disable(struct v3d_dev *v3d);
345 void v3d_irq_reset(struct v3d_dev *v3d);
348 int v3d_mmu_get_offset(struct drm_file *file_priv, struct v3d_bo *bo,
350 int v3d_mmu_set_page_table(struct v3d_dev *v3d);
351 void v3d_mmu_insert_ptes(struct v3d_bo *bo);
352 void v3d_mmu_remove_ptes(struct v3d_bo *bo);
355 int v3d_sched_init(struct v3d_dev *v3d);
356 void v3d_sched_fini(struct v3d_dev *v3d);