Skip to contentSkip to conversation

server gpus: fix race between buffer evict and fbdev updates

PATCH New Dave Airlie 1 participant 4 messages May 2, 2013

Subsystems (4)
lore
Share & tools
lore

Dave Airlie

author maintainer
#0 lore raw

This stores up dirty updates to the fbdev until the next update after the buffer has finished moving, otherwise you'd see a lot of failed to reserve bo during bootup with plymouth was starting.

Dave.

author maintainer
#1 lore raw

PATCH 1/3 drm/mgag200: deal with bo reserve fail in dirty update path

From: Dave Airlie <[email protected]>

On F19 testing, it was noticed we get a lot of errors in dmesg about being unable to reserve the buffer when plymouth starts, this is due to the buffer being in the process of migrating, so it makes sense we can't reserve it.

In order to deal with it, this adds delayed updates for the dirty updates, when the bo is unreservable, in the normal console case this shouldn't ever happen, its just when plymouth or X is pushing the console bo to system memory.

---
 drivers/gpu/drm/mgag200/mgag200_drv.h |  2 ++
 drivers/gpu/drm/mgag200/mgag200_fb.c  | 43 ++++++++++++++++++++++++++++++++---
 drivers/gpu/drm/mgag200/mgag200_ttm.c |  4 ++--
 3 files changed, 44 insertions(+), 5 deletions(-)
drivers/gpu/drm/mgag200/mgag200_drv.h +2 -0
index dcfc973..bf29b2f 100644
@@ -115,6 +115,8 @@ struct mga_fbdev {
 	void *sysram;
 	int size;
 	struct ttm_bo_kmap_obj mapping;
+	int x1, y1, x2, y2; /* dirty rect */
+	spinlock_t dirty_lock;
 };
 
 struct mga_crtc {
drivers/gpu/drm/mgag200/mgag200_fb.c +40 -3

72 lines not on this page · raw message

drivers/gpu/drm/mgag200/mgag200_ttm.c +2 -2
index 8fc9d92..401c989 100644
@@ -315,8 +315,8 @@ int mgag200_bo_reserve(struct mgag200_bo *bo, bool no_wait)
 
 	ret = ttm_bo_reserve(&bo->bo, true, no_wait, false, 0);
 	if (ret) {
-		if (ret != -ERESTARTSYS)
-			DRM_ERROR("reserve failed %p\n", bo);
+		if (ret != -ERESTARTSYS && ret != -EBUSY)
+			DRM_ERROR("reserve failed %p %d\n", bo, ret);
 		return ret;
 	}
 	return 0;
Signature · 1 line
1.8.1.2
author maintainer
#2 lore raw

PATCH 2/3 drm/ast: deal with bo reserve fail in dirty update path

In reply to Dave Airlie

From: Dave Airlie <[email protected]>

Port over the mgag200 fix to ast as it suffers the same issue.

On F19 testing, it was noticed we get a lot of errors in dmesg about being unable to reserve the buffer when plymouth starts, this is due to the buffer being in the process of migrating, so it makes sense we can't reserve it.

In order to deal with it, this adds delayed updates for the dirty updates, when the bo is unreservable, in the normal console case this shouldn't ever happen, its just when plymouth or X is pushing the console bo to system memory.

---
 drivers/gpu/drm/ast/ast_drv.h |  2 ++
 drivers/gpu/drm/ast/ast_fb.c  | 43 ++++++++++++++++++++++++++++++++++++++++---
 drivers/gpu/drm/ast/ast_ttm.c |  2 +-
 3 files changed, 43 insertions(+), 4 deletions(-)
drivers/gpu/drm/ast/ast_drv.h +2 -0
index 5284292..02e52d5 100644
@@ -241,6 +241,8 @@ struct ast_fbdev {
 	void *sysram;
 	int size;
 	struct ttm_bo_kmap_obj mapping;
+	int x1, y1, x2, y2; /* dirty rect */
+	spinlock_t dirty_lock;
 };
 
 #define to_ast_crtc(x) container_of(x, struct ast_crtc, base)
drivers/gpu/drm/ast/ast_fb.c +40 -3

72 lines not on this page · raw message

drivers/gpu/drm/ast/ast_ttm.c +1 -1
index 3602731..09da339 100644
@@ -316,7 +316,7 @@ int ast_bo_reserve(struct ast_bo *bo, bool no_wait)
 
 	ret = ttm_bo_reserve(&bo->bo, true, no_wait, false, 0);
 	if (ret) {
-		if (ret != -ERESTARTSYS)
+		if (ret != -ERESTARTSYS && ret != -EBUSY)
 			DRM_ERROR("reserve failed %p\n", bo);
 		return ret;
 	}
Signature · 1 line
1.8.1.2
author maintainer
#3 lore raw

PATCH 3/3 drm/cirrus: deal with bo reserve fail in dirty update path

In reply to Dave Airlie

From: Dave Airlie <[email protected]>

Port over the mgag200 fix to cirrus as it suffers the same issue.

On F19 testing, it was noticed we get a lot of errors in dmesg about being unable to reserve the buffer when plymouth starts, this is due to the buffer being in the process of migrating, so it makes sense we can't reserve it.

In order to deal with it, this adds delayed updates for the dirty updates, when the bo is unreservable, in the normal console case this shouldn't ever happen, its just when plymouth or X is pushing the console bo to system memory.

---
 drivers/gpu/drm/cirrus/cirrus_drv.h   |  2 ++
 drivers/gpu/drm/cirrus/cirrus_fbdev.c | 38 ++++++++++++++++++++++++++++++++++-
 drivers/gpu/drm/cirrus/cirrus_ttm.c   |  2 +-
 3 files changed, 40 insertions(+), 2 deletions(-)
drivers/gpu/drm/cirrus/cirrus_drv.h +2 -0
index 6e0cc72..7ca0595 100644
@@ -154,6 +154,8 @@ struct cirrus_fbdev {
 	struct list_head fbdev_list;
 	void *sysram;
 	int size;
+	int x1, y1, x2, y2; /* dirty rect */
+	spinlock_t dirty_lock;
 };
 
 struct cirrus_bo {
drivers/gpu/drm/cirrus/cirrus_fbdev.c +37 -1
index e25afcc..3541b56 100644
@@ -27,16 +27,51 @@ static void cirrus_dirty_update(struct cirrus_fbdev *afbdev,
 	int bpp = (afbdev->gfb.base.bits_per_pixel + 7)/8;
 	int ret;
 	bool unmap = false;
+	bool store_for_later = false;
+	int x2, y2;
+	unsigned long flags;
 
 	obj = afbdev->gfb.obj;
 	bo = gem_to_cirrus_bo(obj);
 
+	/*
+	 * try and reserve the BO, if we fail with busy
+	 * then the BO is being moved and we should
+	 * store up the damage until later.
+	 */
 	ret = cirrus_bo_reserve(bo, true);
 	if (ret) {
-		DRM_ERROR("failed to reserve fb bo\n");
+		if (ret != -EBUSY)
+			return;
+		store_for_later = true;
+	}
+
+	x2 = x + width - 1;
+	y2 = y + height - 1;
+	spin_lock_irqsave(&afbdev->dirty_lock, flags);
+
+	if (afbdev->y1 < y)
+		y = afbdev->y1;
+	if (afbdev->y2 > y2)
+		y2 = afbdev->y2;
+	if (afbdev->x1 < x)
+		x = afbdev->x1;
+	if (afbdev->x2 > x2)
+		x2 = afbdev->x2;
+
+	if (store_for_later) {
+		afbdev->x1 = x;
+		afbdev->x2 = x2;
+		afbdev->y1 = y;
+		afbdev->y2 = y2;
+		spin_unlock_irqrestore(&afbdev->dirty_lock, flags);
 		return;
 	}
 
+	afbdev->x1 = afbdev->y1 = INT_MAX;
+	afbdev->x2 = afbdev->y2 = 0;
+	spin_unlock_irqrestore(&afbdev->dirty_lock, flags);
+
 	if (!bo->kmap.virtual) {
 		ret = ttm_bo_kmap(&bo->bo, 0, bo->bo.num_pages, &bo->kmap);
 		if (ret) {
@@ -268,6 +303,7 @@ int cirrus_fbdev_init(struct cirrus_device *cdev)
 
 	cdev->mode_info.gfbdev = gfbdev;
 	gfbdev->helper.funcs = &cirrus_fb_helper_funcs;
+	spin_lock_init(&gfbdev->dirty_lock);
 
 	ret = drm_fb_helper_init(cdev->dev, &gfbdev->helper,
 				 cdev->num_crtc, CIRRUSFB_CONN_LIMIT);
drivers/gpu/drm/cirrus/cirrus_ttm.c +1 -1
index 1413a26..2ed8cfc 100644
@@ -321,7 +321,7 @@ int cirrus_bo_reserve(struct cirrus_bo *bo, bool no_wait)
 
 	ret = ttm_bo_reserve(&bo->bo, true, no_wait, false, 0);
 	if (ret) {
-		if (ret != -ERESTARTSYS)
+		if (ret != -ERESTARTSYS && ret != -EBUSY)
 			DRM_ERROR("reserve failed %p\n", bo);
 		return ret;
 	}
Signature · 1 line
1.8.1.2