drm/lima: recover task by enlarging heap buffer

Increase heap buffer backup memory when GP receive PLBU
out of memory interrupt, then resume the task.

Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Tested-by: Andreas Baierl <ichgeh@imkreisrum.de>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200116131157.13346-5-yuq825@gmail.com

5 files changed, 98 insertions, 7 deletions

diff --git a/drivers/gpu/drm/lima/lima_gp.c b/drivers/gpu/drm/lima/lima_gp.c
index ccf49faedebf..52b210f9a605 100644
--- a/drivers/gpu/drm/lima/lima_gp.c
+++ b/drivers/gpu/drm/lima/lima_gp.c
@@ -11,6 +11,8 @@
 #include "lima_device.h"
 #include "lima_gp.h"
 #include "lima_regs.h"
+#include "lima_gem.h"
+#include "lima_vm.h"
 
 #define gp_write(reg, data) writel(data, ip->iomem + reg)
 #define gp_read(reg) readl(ip->iomem + reg)
@@ -20,6 +22,7 @@ static irqreturn_t lima_gp_irq_handler(int irq, void *data)
 	struct lima_ip *ip = data;
 	struct lima_device *dev = ip->dev;
 	struct lima_sched_pipe *pipe = dev->pipe + lima_pipe_gp;
+	struct lima_sched_task *task = pipe->current_task;
 	u32 state = gp_read(LIMA_GP_INT_STAT);
 	u32 status = gp_read(LIMA_GP_STATUS);
 	bool done = false;
@@ -29,8 +32,16 @@ static irqreturn_t lima_gp_irq_handler(int irq, void *data)
 		return IRQ_NONE;
 
 	if (state & LIMA_GP_IRQ_MASK_ERROR) {
-		dev_err(dev->dev, "gp error irq state=%x status=%x\n",
-			state, status);
+		if ((state & LIMA_GP_IRQ_MASK_ERROR) ==
+		    LIMA_GP_IRQ_PLBU_OUT_OF_MEM) {
+			dev_dbg(dev->dev, "gp out of heap irq status=%x\n",
+				status);
+		} else {
+			dev_err(dev->dev, "gp error irq state=%x status=%x\n",
+				state, status);
+			if (task)
+				task->recoverable = false;
+		}
 
 		/* mask all interrupts before hard reset */
 		gp_write(LIMA_GP_INT_MASK, 0);
@@ -43,6 +54,7 @@ static irqreturn_t lima_gp_irq_handler(int irq, void *data)
 		bool active = status & (LIMA_GP_STATUS_VS_ACTIVE |
 					LIMA_GP_STATUS_PLBU_ACTIVE);
 		done = valid && !active;
+		pipe->error = false;
 	}
 
 	gp_write(LIMA_GP_INT_CLEAR, state);
@@ -121,6 +133,22 @@ static void lima_gp_task_run(struct lima_sched_pipe *pipe,
 	u32 cmd = 0;
 	int i;
 
+	/* update real heap buffer size for GP */
+	for (i = 0; i < task->num_bos; i++) {
+		struct lima_bo *bo = task->bos[i];
+
+		if (bo->heap_size &&
+		    lima_vm_get_va(task->vm, bo) ==
+		    f[LIMA_GP_PLBU_ALLOC_START_ADDR >> 2]) {
+			f[LIMA_GP_PLBU_ALLOC_END_ADDR >> 2] =
+				f[LIMA_GP_PLBU_ALLOC_START_ADDR >> 2] +
+				bo->heap_size;
+			task->recoverable = true;
+			task->heap = bo;
+			break;
+		}
+	}
+
 	if (f[LIMA_GP_VSCL_START_ADDR >> 2] !=
 	    f[LIMA_GP_VSCL_END_ADDR >> 2])
 		cmd |= LIMA_GP_CMD_START_VS;
@@ -184,6 +212,31 @@ static void lima_gp_task_mmu_error(struct lima_sched_pipe *pipe)
 	lima_sched_pipe_task_done(pipe);
 }
 
+static int lima_gp_task_recover(struct lima_sched_pipe *pipe)
+{
+	struct lima_ip *ip = pipe->processor[0];
+	struct lima_sched_task *task = pipe->current_task;
+	struct drm_lima_gp_frame *frame = task->frame;
+	u32 *f = frame->frame;
+	size_t fail_size =
+		f[LIMA_GP_PLBU_ALLOC_END_ADDR >> 2] -
+		f[LIMA_GP_PLBU_ALLOC_START_ADDR >> 2];
+
+	if (fail_size == task->heap->heap_size) {
+		int ret;
+
+		ret = lima_heap_alloc(task->heap, task->vm);
+		if (ret < 0)
+			return ret;
+	}
+
+	gp_write(LIMA_GP_INT_MASK, LIMA_GP_IRQ_MASK_USED);
+	gp_write(LIMA_GP_PLBU_ALLOC_END_ADDR,
+		 f[LIMA_GP_PLBU_ALLOC_START_ADDR >> 2] + task->heap->heap_size);
+	gp_write(LIMA_GP_CMD, LIMA_GP_CMD_UPDATE_PLBU_ALLOC);
+	return 0;
+}
+
 static void lima_gp_print_version(struct lima_ip *ip)
 {
 	u32 version, major, minor;
@@ -270,6 +323,7 @@ int lima_gp_pipe_init(struct lima_device *dev)
 	pipe->task_fini = lima_gp_task_fini;
 	pipe->task_error = lima_gp_task_error;
 	pipe->task_mmu_error = lima_gp_task_mmu_error;
+	pipe->task_recover = lima_gp_task_recover;
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/lima/lima_mmu.c b/drivers/gpu/drm/lima/lima_mmu.c
index 97ec09dee572..f79d2af427e7 100644
--- a/drivers/gpu/drm/lima/lima_mmu.c
+++ b/drivers/gpu/drm/lima/lima_mmu.c
@@ -99,6 +99,11 @@ void lima_mmu_fini(struct lima_ip *ip)
 
 }
 
+void lima_mmu_flush_tlb(struct lima_ip *ip)
+{
+	mmu_write(LIMA_MMU_COMMAND, LIMA_MMU_COMMAND_ZAP_CACHE);
+}
+
 void lima_mmu_switch_vm(struct lima_ip *ip, struct lima_vm *vm)
 {
 	struct lima_device *dev = ip->dev;
diff --git a/drivers/gpu/drm/lima/lima_mmu.h b/drivers/gpu/drm/lima/lima_mmu.h
index 8c78319bcc8e..4f8ccbebcba1 100644
--- a/drivers/gpu/drm/lima/lima_mmu.h
+++ b/drivers/gpu/drm/lima/lima_mmu.h
@@ -10,6 +10,7 @@ struct lima_vm;
 int lima_mmu_init(struct lima_ip *ip);
 void lima_mmu_fini(struct lima_ip *ip);
 
+void lima_mmu_flush_tlb(struct lima_ip *ip);
 void lima_mmu_switch_vm(struct lima_ip *ip, struct lima_vm *vm);
 void lima_mmu_page_fault_resume(struct lima_ip *ip);
 
diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c
index a31a90c380b6..5f9c54475799 100644
--- a/drivers/gpu/drm/lima/lima_sched.c
+++ b/drivers/gpu/drm/lima/lima_sched.c
@@ -312,6 +312,26 @@ static const struct drm_sched_backend_ops lima_sched_ops = {
 	.free_job = lima_sched_free_job,
 };
 
+static void lima_sched_recover_work(struct work_struct *work)
+{
+	struct lima_sched_pipe *pipe =
+		container_of(work, struct lima_sched_pipe, recover_work);
+	int i;
+
+	for (i = 0; i < pipe->num_l2_cache; i++)
+		lima_l2_cache_flush(pipe->l2_cache[i]);
+
+	if (pipe->bcast_mmu) {
+		lima_mmu_flush_tlb(pipe->bcast_mmu);
+	} else {
+		for (i = 0; i < pipe->num_mmu; i++)
+			lima_mmu_flush_tlb(pipe->mmu[i]);
+	}
+
+	if (pipe->task_recover(pipe))
+		drm_sched_fault(&pipe->base);
+}
+
 int lima_sched_pipe_init(struct lima_sched_pipe *pipe, const char *name)
 {
 	unsigned int timeout = lima_sched_timeout_ms > 0 ?
@@ -320,6 +340,8 @@ int lima_sched_pipe_init(struct lima_sched_pipe *pipe, const char *name)
 	pipe->fence_context = dma_fence_context_alloc(1);
 	spin_lock_init(&pipe->fence_lock);
 
+	INIT_WORK(&pipe->recover_work, lima_sched_recover_work);
+
 	return drm_sched_init(&pipe->base, &lima_sched_ops, 1, 0,
 			      msecs_to_jiffies(timeout), name);
 }
@@ -331,11 +353,14 @@ void lima_sched_pipe_fini(struct lima_sched_pipe *pipe)
 
 void lima_sched_pipe_task_done(struct lima_sched_pipe *pipe)
 {
-	if (pipe->error)
-		drm_sched_fault(&pipe->base);
-	else {
-		struct lima_sched_task *task = pipe->current_task;
-
+	struct lima_sched_task *task = pipe->current_task;
+
+	if (pipe->error) {
+		if (task && task->recoverable)
+			schedule_work(&pipe->recover_work);
+		else
+			drm_sched_fault(&pipe->base);
+	} else {
 		pipe->task_fini(pipe);
 		dma_fence_signal(task->fence);
 	}
diff --git a/drivers/gpu/drm/lima/lima_sched.h b/drivers/gpu/drm/lima/lima_sched.h
index 1d814fecbcc0..d64393fb50a9 100644
--- a/drivers/gpu/drm/lima/lima_sched.h
+++ b/drivers/gpu/drm/lima/lima_sched.h
@@ -20,6 +20,9 @@ struct lima_sched_task {
 	struct lima_bo **bos;
 	int num_bos;
 
+	bool recoverable;
+	struct lima_bo *heap;
+
 	/* pipe fence */
 	struct dma_fence *fence;
 };
@@ -68,6 +71,9 @@ struct lima_sched_pipe {
 	void (*task_fini)(struct lima_sched_pipe *pipe);
 	void (*task_error)(struct lima_sched_pipe *pipe);
 	void (*task_mmu_error)(struct lima_sched_pipe *pipe);
+	int (*task_recover)(struct lima_sched_pipe *pipe);
+
+	struct work_struct recover_work;
 };
 
 int lima_sched_task_init(struct lima_sched_task *task,