drm/lima: driver for ARM Mali4xx GPUs

- Mali 4xx GPUs have two kinds of processors GP and PP. GP is for
  OpenGL vertex shader processing and PP is for fragment shader
  processing. Each processor has its own MMU so prcessors work in
  virtual address space.
- There's only one GP but multiple PP (max 4 for mali 400 and 8
  for mali 450) in the same mali 4xx GPU. All PPs are grouped
  togather to handle a single fragment shader task divided by
  FB output tiled pixels. Mali 400 user space driver is
  responsible for assign target tiled pixels to each PP, but mali
  450 has a HW module called DLBU to dynamically balance each
  PP's load.
- User space driver allocate buffer object and map into GPU
  virtual address space, upload command stream and draw data with
  CPU mmap of the buffer object, then submit task to GP/PP with
  a register frame indicating where is the command stream and misc
  settings.
- There's no command stream validation/relocation due to each user
  process has its own GPU virtual address space. GP/PP's MMU switch
  virtual address space before running two tasks from different
  user process. Error or evil user space code just get MMU fault
  or GP/PP error IRQ, then the HW/SW will be recovered.
- Use GEM+shmem for MM. Currently just alloc and pin memory when
  gem object creation. GPU vm map of the buffer is also done in
  the alloc stage in kernel space. We may delay the memory
  allocation and real GPU vm map to command submission stage in the
  furture as improvement.
- Use drm_sched for GPU task schedule. Each OpenGL context should
  have a lima context object in the kernel to distinguish tasks
  from different user. drm_sched gets task from each lima context
  in a fair way.

mesa driver can be found here before upstreamed:
https://gitlab.freedesktop.org/lima/mesa

v8:
- add comments for in_sync
- fix ctx free miss mutex unlock

v7:
- remove lima_fence_ops with default value
- move fence slab create to device probe
- check pad ioctl args to be zero
- add comments for user/kernel interface

v6:
- fix comments by checkpatch.pl

v5:
- export gp/pp version to userspace
- rebase on drm-misc-next

v4:
- use get param interface to get info
- separate context create/free ioctl
- remove unused max sched task param
- update copyright time
- use xarray instead of idr
- stop using drmP.h

v3:
- fix comments from kbuild robot
- restrict supported arch to tested ones

v2:
- fix syscall argument check
- fix job finish fence leak since kernel 5.0
- use drm syncobj to replace native fence
- move buffer object GPU va map into kernel
- reserve syscall argument space for future info
- remove kernel gem modifier
- switch TTM back to GEM+shmem MM
- use time based io poll
- use whole register name
- adopt gem reservation obj integration
- use drm_timeout_abs_to_jiffies

Cc: Eric Anholt <eric@anholt.net>
Cc: Rob Herring <robh@kernel.org>
Cc: Christian König <ckoenig.leichtzumerken@gmail.com>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Alex Deucher <alexdeucher@gmail.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Rob Clark <robdclark@gmail.com>
Cc: Dave Airlie <airlied@gmail.com>
Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Signed-off-by: Marek Vasut <marex@denx.de>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Signed-off-by: Simon Shields <simon@lineageos.org>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Rob Herring <robh@kerrnel.org>
Signed-off-by: Eric Anholt <eric@anholt.net>
Link: https://patchwork.freedesktop.org/patch/291200/

1 files changed, 404 insertions, 0 deletions

diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c
new file mode 100644
index 000000000000..97bd9c1deb87
--- /dev/null
+++ b/drivers/gpu/drm/lima/lima_sched.c
@@ -0,0 +1,404 @@
+// SPDX-License-Identifier: GPL-2.0 OR MIT
+/* Copyright 2017-2019 Qiang Yu <yuq825@gmail.com> */
+
+#include <linux/kthread.h>
+#include <linux/slab.h>
+
+#include "lima_drv.h"
+#include "lima_sched.h"
+#include "lima_vm.h"
+#include "lima_mmu.h"
+#include "lima_l2_cache.h"
+#include "lima_object.h"
+
+struct lima_fence {
+	struct dma_fence base;
+	struct lima_sched_pipe *pipe;
+};
+
+static struct kmem_cache *lima_fence_slab;
+static int lima_fence_slab_refcnt;
+
+int lima_sched_slab_init(void)
+{
+	if (!lima_fence_slab) {
+		lima_fence_slab = kmem_cache_create(
+			"lima_fence", sizeof(struct lima_fence), 0,
+			SLAB_HWCACHE_ALIGN, NULL);
+		if (!lima_fence_slab)
+			return -ENOMEM;
+	}
+
+	lima_fence_slab_refcnt++;
+	return 0;
+}
+
+void lima_sched_slab_fini(void)
+{
+	if (!--lima_fence_slab_refcnt) {
+		kmem_cache_destroy(lima_fence_slab);
+		lima_fence_slab = NULL;
+	}
+}
+
+static inline struct lima_fence *to_lima_fence(struct dma_fence *fence)
+{
+	return container_of(fence, struct lima_fence, base);
+}
+
+static const char *lima_fence_get_driver_name(struct dma_fence *fence)
+{
+	return "lima";
+}
+
+static const char *lima_fence_get_timeline_name(struct dma_fence *fence)
+{
+	struct lima_fence *f = to_lima_fence(fence);
+
+	return f->pipe->base.name;
+}
+
+static void lima_fence_release_rcu(struct rcu_head *rcu)
+{
+	struct dma_fence *f = container_of(rcu, struct dma_fence, rcu);
+	struct lima_fence *fence = to_lima_fence(f);
+
+	kmem_cache_free(lima_fence_slab, fence);
+}
+
+static void lima_fence_release(struct dma_fence *fence)
+{
+	struct lima_fence *f = to_lima_fence(fence);
+
+	call_rcu(&f->base.rcu, lima_fence_release_rcu);
+}
+
+static const struct dma_fence_ops lima_fence_ops = {
+	.get_driver_name = lima_fence_get_driver_name,
+	.get_timeline_name = lima_fence_get_timeline_name,
+	.release = lima_fence_release,
+};
+
+static struct lima_fence *lima_fence_create(struct lima_sched_pipe *pipe)
+{
+	struct lima_fence *fence;
+
+	fence = kmem_cache_zalloc(lima_fence_slab, GFP_KERNEL);
+	if (!fence)
+		return NULL;
+
+	fence->pipe = pipe;
+	dma_fence_init(&fence->base, &lima_fence_ops, &pipe->fence_lock,
+		       pipe->fence_context, ++pipe->fence_seqno);
+
+	return fence;
+}
+
+static inline struct lima_sched_task *to_lima_task(struct drm_sched_job *job)
+{
+	return container_of(job, struct lima_sched_task, base);
+}
+
+static inline struct lima_sched_pipe *to_lima_pipe(struct drm_gpu_scheduler *sched)
+{
+	return container_of(sched, struct lima_sched_pipe, base);
+}
+
+int lima_sched_task_init(struct lima_sched_task *task,
+			 struct lima_sched_context *context,
+			 struct lima_bo **bos, int num_bos,
+			 struct lima_vm *vm)
+{
+	int err, i;
+
+	task->bos = kmemdup(bos, sizeof(*bos) * num_bos, GFP_KERNEL);
+	if (!task->bos)
+		return -ENOMEM;
+
+	for (i = 0; i < num_bos; i++)
+		drm_gem_object_get(&bos[i]->gem);
+
+	err = drm_sched_job_init(&task->base, &context->base, vm);
+	if (err) {
+		kfree(task->bos);
+		return err;
+	}
+
+	task->num_bos = num_bos;
+	task->vm = lima_vm_get(vm);
+	return 0;
+}
+
+void lima_sched_task_fini(struct lima_sched_task *task)
+{
+	int i;
+
+	drm_sched_job_cleanup(&task->base);
+
+	for (i = 0; i < task->num_dep; i++)
+		dma_fence_put(task->dep[i]);
+
+	kfree(task->dep);
+
+	if (task->bos) {
+		for (i = 0; i < task->num_bos; i++)
+			drm_gem_object_put_unlocked(&task->bos[i]->gem);
+		kfree(task->bos);
+	}
+
+	lima_vm_put(task->vm);
+}
+
+int lima_sched_task_add_dep(struct lima_sched_task *task, struct dma_fence *fence)
+{
+	int i, new_dep = 4;
+
+	/* same context's fence is definitly earlier then this task */
+	if (fence->context == task->base.s_fence->finished.context) {
+		dma_fence_put(fence);
+		return 0;
+	}
+
+	if (task->dep && task->num_dep == task->max_dep)
+		new_dep = task->max_dep * 2;
+
+	if (task->max_dep < new_dep) {
+		void *dep = krealloc(task->dep, sizeof(*task->dep) * new_dep, GFP_KERNEL);
+
+		if (!dep)
+			return -ENOMEM;
+
+		task->max_dep = new_dep;
+		task->dep = dep;
+	}
+
+	for (i = 0; i < task->num_dep; i++) {
+		if (task->dep[i]->context == fence->context &&
+		    dma_fence_is_later(fence, task->dep[i])) {
+			dma_fence_put(task->dep[i]);
+			task->dep[i] = fence;
+			return 0;
+		}
+	}
+
+	task->dep[task->num_dep++] = fence;
+	return 0;
+}
+
+int lima_sched_context_init(struct lima_sched_pipe *pipe,
+			    struct lima_sched_context *context,
+			    atomic_t *guilty)
+{
+	struct drm_sched_rq *rq = pipe->base.sched_rq + DRM_SCHED_PRIORITY_NORMAL;
+
+	return drm_sched_entity_init(&context->base, &rq, 1, guilty);
+}
+
+void lima_sched_context_fini(struct lima_sched_pipe *pipe,
+			     struct lima_sched_context *context)
+{
+	drm_sched_entity_fini(&context->base);
+}
+
+struct dma_fence *lima_sched_context_queue_task(struct lima_sched_context *context,
+						struct lima_sched_task *task)
+{
+	struct dma_fence *fence = dma_fence_get(&task->base.s_fence->finished);
+
+	drm_sched_entity_push_job(&task->base, &context->base);
+	return fence;
+}
+
+static struct dma_fence *lima_sched_dependency(struct drm_sched_job *job,
+					       struct drm_sched_entity *entity)
+{
+	struct lima_sched_task *task = to_lima_task(job);
+	int i;
+
+	for (i = 0; i < task->num_dep; i++) {
+		struct dma_fence *fence = task->dep[i];
+
+		if (!task->dep[i])
+			continue;
+
+		task->dep[i] = NULL;
+
+		if (!dma_fence_is_signaled(fence))
+			return fence;
+
+		dma_fence_put(fence);
+	}
+
+	return NULL;
+}
+
+static struct dma_fence *lima_sched_run_job(struct drm_sched_job *job)
+{
+	struct lima_sched_task *task = to_lima_task(job);
+	struct lima_sched_pipe *pipe = to_lima_pipe(job->sched);
+	struct lima_fence *fence;
+	struct dma_fence *ret;
+	struct lima_vm *vm = NULL, *last_vm = NULL;
+	int i;
+
+	/* after GPU reset */
+	if (job->s_fence->finished.error < 0)
+		return NULL;
+
+	fence = lima_fence_create(pipe);
+	if (!fence)
+		return NULL;
+	task->fence = &fence->base;
+
+	/* for caller usage of the fence, otherwise irq handler
+	 * may consume the fence before caller use it
+	 */
+	ret = dma_fence_get(task->fence);
+
+	pipe->current_task = task;
+
+	/* this is needed for MMU to work correctly, otherwise GP/PP
+	 * will hang or page fault for unknown reason after running for
+	 * a while.
+	 *
+	 * Need to investigate:
+	 * 1. is it related to TLB
+	 * 2. how much performance will be affected by L2 cache flush
+	 * 3. can we reduce the calling of this function because all
+	 *    GP/PP use the same L2 cache on mali400
+	 *
+	 * TODO:
+	 * 1. move this to task fini to save some wait time?
+	 * 2. when GP/PP use different l2 cache, need PP wait GP l2
+	 *    cache flush?
+	 */
+	for (i = 0; i < pipe->num_l2_cache; i++)
+		lima_l2_cache_flush(pipe->l2_cache[i]);
+
+	if (task->vm != pipe->current_vm) {
+		vm = lima_vm_get(task->vm);
+		last_vm = pipe->current_vm;
+		pipe->current_vm = task->vm;
+	}
+
+	if (pipe->bcast_mmu)
+		lima_mmu_switch_vm(pipe->bcast_mmu, vm);
+	else {
+		for (i = 0; i < pipe->num_mmu; i++)
+			lima_mmu_switch_vm(pipe->mmu[i], vm);
+	}
+
+	if (last_vm)
+		lima_vm_put(last_vm);
+
+	pipe->error = false;
+	pipe->task_run(pipe, task);
+
+	return task->fence;
+}
+
+static void lima_sched_handle_error_task(struct lima_sched_pipe *pipe,
+					 struct lima_sched_task *task)
+{
+	drm_sched_stop(&pipe->base);
+
+	if (task)
+		drm_sched_increase_karma(&task->base);
+
+	pipe->task_error(pipe);
+
+	if (pipe->bcast_mmu)
+		lima_mmu_page_fault_resume(pipe->bcast_mmu);
+	else {
+		int i;
+
+		for (i = 0; i < pipe->num_mmu; i++)
+			lima_mmu_page_fault_resume(pipe->mmu[i]);
+	}
+
+	if (pipe->current_vm)
+		lima_vm_put(pipe->current_vm);
+
+	pipe->current_vm = NULL;
+	pipe->current_task = NULL;
+
+	drm_sched_resubmit_jobs(&pipe->base);
+	drm_sched_start(&pipe->base, true);
+}
+
+static void lima_sched_timedout_job(struct drm_sched_job *job)
+{
+	struct lima_sched_pipe *pipe = to_lima_pipe(job->sched);
+	struct lima_sched_task *task = to_lima_task(job);
+
+	DRM_ERROR("lima job timeout\n");
+
+	lima_sched_handle_error_task(pipe, task);
+}
+
+static void lima_sched_free_job(struct drm_sched_job *job)
+{
+	struct lima_sched_task *task = to_lima_task(job);
+	struct lima_sched_pipe *pipe = to_lima_pipe(job->sched);
+	struct lima_vm *vm = task->vm;
+	struct lima_bo **bos = task->bos;
+	int i;
+
+	dma_fence_put(task->fence);
+
+	for (i = 0; i < task->num_bos; i++)
+		lima_vm_bo_del(vm, bos[i]);
+
+	lima_sched_task_fini(task);
+	kmem_cache_free(pipe->task_slab, task);
+}
+
+const struct drm_sched_backend_ops lima_sched_ops = {
+	.dependency = lima_sched_dependency,
+	.run_job = lima_sched_run_job,
+	.timedout_job = lima_sched_timedout_job,
+	.free_job = lima_sched_free_job,
+};
+
+static void lima_sched_error_work(struct work_struct *work)
+{
+	struct lima_sched_pipe *pipe =
+		container_of(work, struct lima_sched_pipe, error_work);
+	struct lima_sched_task *task = pipe->current_task;
+
+	lima_sched_handle_error_task(pipe, task);
+}
+
+int lima_sched_pipe_init(struct lima_sched_pipe *pipe, const char *name)
+{
+	long timeout;
+
+	if (lima_sched_timeout_ms <= 0)
+		timeout = MAX_SCHEDULE_TIMEOUT;
+	else
+		timeout = msecs_to_jiffies(lima_sched_timeout_ms);
+
+	pipe->fence_context = dma_fence_context_alloc(1);
+	spin_lock_init(&pipe->fence_lock);
+
+	INIT_WORK(&pipe->error_work, lima_sched_error_work);
+
+	return drm_sched_init(&pipe->base, &lima_sched_ops, 1, 0, timeout, name);
+}
+
+void lima_sched_pipe_fini(struct lima_sched_pipe *pipe)
+{
+	drm_sched_fini(&pipe->base);
+}
+
+void lima_sched_pipe_task_done(struct lima_sched_pipe *pipe)
+{
+	if (pipe->error)
+		schedule_work(&pipe->error_work);
+	else {
+		struct lima_sched_task *task = pipe->current_task;
+
+		pipe->task_fini(pipe);
+		dma_fence_signal(task->fence);
+	}
+}