[Intel-xe] [PATCH] drm/xe: fix pvc unload issue

* [Intel-xe] [PATCH] drm/xe: fix pvc unload issue
@ 2023-04-03 22:20 Chang, Bruce
  2023-04-03 22:31 ` [Intel-xe] ✓ CI.Patch_applied: success for drm/xe: fix pvc unload issue (rev2) Patchwork
                   ` (5 more replies)
  0 siblings, 6 replies; 13+ messages in thread
From: Chang, Bruce @ 2023-04-03 22:20 UTC (permalink / raw)
  To: intel-xe; +Cc: Stuart Summers, Rodrigo, Chang, Bruce, Vivi

Currently, unload pvc driver will generate a null dereference
and the call stack is as below.

[ 4850.618000] Call Trace:
[ 4850.620740]  <TASK>
[ 4850.623134]  ttm_bo_cleanup_memtype_use+0x3f/0x50 [ttm]
[ 4850.628661]  ttm_bo_release+0x154/0x2c0 [ttm]
[ 4850.633317]  ? drm_buddy_fini+0x62/0x80 [drm_buddy]
[ 4850.638487]  ? __kmem_cache_free+0x27d/0x2c0
[ 4850.643054]  ttm_bo_put+0x38/0x60 [ttm]
[ 4850.647190]  xe_gem_object_free+0x1f/0x30 [xe]
[ 4850.651945]  drm_gem_object_free+0x1e/0x30 [drm]
[ 4850.656904]  ggtt_fini_noalloc+0x9d/0xe0 [xe]
[ 4850.661574]  drm_managed_release+0xb5/0x150 [drm]
[ 4850.666617]  drm_dev_release+0x30/0x50 [drm]
[ 4850.671209]  devm_drm_dev_init_release+0x3c/0x60 [drm]

There are a couple issues, but the main one is due to TTM has only
one TTM_PL_TT region, but since pvc has 2 tiles and tries to setup
1 TTM_PL_TT each tile. The second will overwrite the first one.

During unload time, the first tile will reset the TTM_PL_TT manger
and when the second tile is trying to free Bo and it will generate
the null reference since the TTM manage is already got reset to 0.

The fix is to use one global TTM_PL_TT manager.

v2: make gtt mgr global and change the name to sys_mgr

Cc: Stuart Summers <stuart.summers@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Vivi, Rodrigo <rodrigo.vivi@intel.com>
Signed-off-by: Bruce Chang <yu.bruce.chang@intel.com>
---
 drivers/gpu/drm/xe/Makefile                   |  2 +-
 drivers/gpu/drm/xe/xe_device.c                |  2 +
 drivers/gpu/drm/xe/xe_device.h                |  1 +
 drivers/gpu/drm/xe/xe_device_types.h          |  2 +
 drivers/gpu/drm/xe/xe_gt.c                    | 18 -----
 drivers/gpu/drm/xe/xe_gt_types.h              |  2 -
 drivers/gpu/drm/xe/xe_ttm_gtt_mgr.h           | 16 -----
 drivers/gpu/drm/xe/xe_ttm_gtt_mgr_types.h     | 18 -----
 .../xe/{xe_ttm_gtt_mgr.c => xe_ttm_sys_mgr.c} | 67 +++++++------------
 9 files changed, 31 insertions(+), 97 deletions(-)
 delete mode 100644 drivers/gpu/drm/xe/xe_ttm_gtt_mgr.h
 delete mode 100644 drivers/gpu/drm/xe/xe_ttm_gtt_mgr_types.h
 rename drivers/gpu/drm/xe/{xe_ttm_gtt_mgr.c => xe_ttm_sys_mgr.c} (52%)

diff --git a/drivers/gpu/drm/xe/Makefile b/drivers/gpu/drm/xe/Makefile
index 669c7b6f552c..7e19a15af3dc 100644
--- a/drivers/gpu/drm/xe/Makefile
+++ b/drivers/gpu/drm/xe/Makefile
@@ -83,7 +83,7 @@ xe-y += xe_bb.o \
 	xe_step.o \
 	xe_sync.o \
 	xe_trace.o \
-	xe_ttm_gtt_mgr.o \
+	xe_ttm_sys_mgr.o \
 	xe_ttm_stolen_mgr.o \
 	xe_ttm_vram_mgr.o \
 	xe_tuning.o \
diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
index a79f934e3d2d..4fb9aff27686 100644
--- a/drivers/gpu/drm/xe/xe_device.c
+++ b/drivers/gpu/drm/xe/xe_device.c
@@ -279,6 +279,8 @@ int xe_device_probe(struct xe_device *xe)
 	if (err)
 		goto err_irq_shutdown;
 
+	xe_ttm_sys_mgr_init(xe);
+
 	for_each_gt(gt, xe, id) {
 		err = xe_gt_init_noalloc(gt);
 		if (err)
diff --git a/drivers/gpu/drm/xe/xe_device.h b/drivers/gpu/drm/xe/xe_device.h
index d277f8985f7b..d9d1b09a8e38 100644
--- a/drivers/gpu/drm/xe/xe_device.h
+++ b/drivers/gpu/drm/xe/xe_device.h
@@ -116,4 +116,5 @@ static inline bool xe_device_has_flat_ccs(struct xe_device *xe)
 }
 
 u32 xe_device_ccs_bytes(struct xe_device *xe, u64 size);
+int xe_ttm_sys_mgr_init(struct xe_device *xe);
 #endif
diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
index 88f863edc41c..6fb53a54913f 100644
--- a/drivers/gpu/drm/xe/xe_device_types.h
+++ b/drivers/gpu/drm/xe/xe_device_types.h
@@ -195,6 +195,8 @@ struct xe_device {
 			/** @mapping: pointer to VRAM mappable space */
 			void *__iomem mapping;
 		} vram;
+		/** @sys_mgr: system TTM manager */
+		struct ttm_resource_manager sys_mgr;
 	} mem;
 
 	/** @usm: unified memory state */
diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
index fd7a5b43ba3e..a6ea138a7877 100644
--- a/drivers/gpu/drm/xe/xe_gt.c
+++ b/drivers/gpu/drm/xe/xe_gt.c
@@ -36,7 +36,6 @@
 #include "xe_ring_ops.h"
 #include "xe_sa.h"
 #include "xe_sched_job.h"
-#include "xe_ttm_gtt_mgr.h"
 #include "xe_ttm_vram_mgr.h"
 #include "xe_tuning.h"
 #include "xe_uc.h"
@@ -77,16 +76,11 @@ int xe_gt_alloc(struct xe_device *xe, struct xe_gt *gt)
 		if (!gt->mem.vram_mgr)
 			return -ENOMEM;
 
-		gt->mem.gtt_mgr = drmm_kzalloc(drm, sizeof(*gt->mem.gtt_mgr),
-					       GFP_KERNEL);
-		if (!gt->mem.gtt_mgr)
-			return -ENOMEM;
 	} else {
 		struct xe_gt *full_gt = xe_find_full_gt(gt);
 
 		gt->mem.ggtt = full_gt->mem.ggtt;
 		gt->mem.vram_mgr = full_gt->mem.vram_mgr;
-		gt->mem.gtt_mgr = full_gt->mem.gtt_mgr;
 	}
 
 	gt->ordered_wq = alloc_ordered_workqueue("gt-ordered-wq", 0);
@@ -98,26 +92,14 @@ static int gt_ttm_mgr_init(struct xe_gt *gt)
 {
 	struct xe_device *xe = gt_to_xe(gt);
 	int err;
-	struct sysinfo si;
-	u64 gtt_size;
-
-	si_meminfo(&si);
-	gtt_size = (u64)si.totalram * si.mem_unit * 3/4;
 
 	if (gt->mem.vram.size) {
 		err = xe_ttm_vram_mgr_init(gt, gt->mem.vram_mgr);
 		if (err)
 			return err;
-		gtt_size = min(max((XE_DEFAULT_GTT_SIZE_MB << 20),
-				   (u64)gt->mem.vram.size),
-			       gtt_size);
 		xe->info.mem_region_mask |= BIT(gt->info.vram_id) << 1;
 	}
 
-	err = xe_ttm_gtt_mgr_init(gt, gt->mem.gtt_mgr, gtt_size);
-	if (err)
-		return err;
-
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/xe/xe_gt_types.h b/drivers/gpu/drm/xe/xe_gt_types.h
index 8f29aba455e0..9d3117fad2e4 100644
--- a/drivers/gpu/drm/xe/xe_gt_types.h
+++ b/drivers/gpu/drm/xe/xe_gt_types.h
@@ -162,8 +162,6 @@ struct xe_gt {
 		} vram;
 		/** @vram_mgr: VRAM TTM manager */
 		struct xe_ttm_vram_mgr *vram_mgr;
-		/** @gtt_mr: GTT TTM manager */
-		struct xe_ttm_gtt_mgr *gtt_mgr;
 		/** @ggtt: Global graphics translation table */
 		struct xe_ggtt *ggtt;
 	} mem;
diff --git a/drivers/gpu/drm/xe/xe_ttm_gtt_mgr.h b/drivers/gpu/drm/xe/xe_ttm_gtt_mgr.h
deleted file mode 100644
index d1d57cb9c2b8..000000000000
--- a/drivers/gpu/drm/xe/xe_ttm_gtt_mgr.h
+++ /dev/null
@@ -1,16 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-/*
- * Copyright © 2022 Intel Corporation
- */
-
-#ifndef _XE_TTGM_GTT_MGR_H_
-#define _XE_TTGM_GTT_MGR_H_
-
-#include "xe_ttm_gtt_mgr_types.h"
-
-struct xe_gt;
-
-int xe_ttm_gtt_mgr_init(struct xe_gt *gt, struct xe_ttm_gtt_mgr *mgr,
-			u64 gtt_size);
-
-#endif
diff --git a/drivers/gpu/drm/xe/xe_ttm_gtt_mgr_types.h b/drivers/gpu/drm/xe/xe_ttm_gtt_mgr_types.h
deleted file mode 100644
index c66737488326..000000000000
--- a/drivers/gpu/drm/xe/xe_ttm_gtt_mgr_types.h
+++ /dev/null
@@ -1,18 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-/*
- * Copyright © 2022 Intel Corporation
- */
-
-#ifndef _XE_TTM_GTT_MGR_TYPES_H_
-#define _XE_TTM_GTT_MGR_TYPES_H_
-
-#include <drm/ttm/ttm_device.h>
-
-struct xe_gt;
-
-struct xe_ttm_gtt_mgr {
-	struct xe_gt *gt;
-	struct ttm_resource_manager manager;
-};
-
-#endif
diff --git a/drivers/gpu/drm/xe/xe_ttm_gtt_mgr.c b/drivers/gpu/drm/xe/xe_ttm_sys_mgr.c
similarity index 52%
rename from drivers/gpu/drm/xe/xe_ttm_gtt_mgr.c
rename to drivers/gpu/drm/xe/xe_ttm_sys_mgr.c
index 8075781070f2..cf5f4f73d4dc 100644
--- a/drivers/gpu/drm/xe/xe_ttm_gtt_mgr.c
+++ b/drivers/gpu/drm/xe/xe_ttm_sys_mgr.c
@@ -12,31 +12,24 @@
 
 #include "xe_bo.h"
 #include "xe_gt.h"
-#include "xe_ttm_gtt_mgr.h"
 
-struct xe_ttm_gtt_node {
+struct xe_ttm_sys_node {
 	struct ttm_buffer_object *tbo;
 	struct ttm_range_mgr_node base;
 };
 
-static inline struct xe_ttm_gtt_mgr *
-to_gtt_mgr(struct ttm_resource_manager *man)
+static inline struct xe_ttm_sys_node *
+to_xe_ttm_sys_node(struct ttm_resource *res)
 {
-	return container_of(man, struct xe_ttm_gtt_mgr, manager);
+	return container_of(res, struct xe_ttm_sys_node, base.base);
 }
 
-static inline struct xe_ttm_gtt_node *
-to_xe_ttm_gtt_node(struct ttm_resource *res)
-{
-	return container_of(res, struct xe_ttm_gtt_node, base.base);
-}
-
-static int xe_ttm_gtt_mgr_new(struct ttm_resource_manager *man,
+static int xe_ttm_sys_mgr_new(struct ttm_resource_manager *man,
 			      struct ttm_buffer_object *tbo,
 			      const struct ttm_place *place,
 			      struct ttm_resource **res)
 {
-	struct xe_ttm_gtt_node *node;
+	struct xe_ttm_sys_node *node;
 	int r;
 
 	node = kzalloc(struct_size(node, base.mm_nodes, 1), GFP_KERNEL);
@@ -66,32 +59,31 @@ static int xe_ttm_gtt_mgr_new(struct ttm_resource_manager *man,
 	return r;
 }
 
-static void xe_ttm_gtt_mgr_del(struct ttm_resource_manager *man,
+static void xe_ttm_sys_mgr_del(struct ttm_resource_manager *man,
 			       struct ttm_resource *res)
 {
-	struct xe_ttm_gtt_node *node = to_xe_ttm_gtt_node(res);
+	struct xe_ttm_sys_node *node = to_xe_ttm_sys_node(res);
 
 	ttm_resource_fini(man, res);
 	kfree(node);
 }
 
-static void xe_ttm_gtt_mgr_debug(struct ttm_resource_manager *man,
+static void xe_ttm_sys_mgr_debug(struct ttm_resource_manager *man,
 				 struct drm_printer *printer)
 {
 
 }
 
-static const struct ttm_resource_manager_func xe_ttm_gtt_mgr_func = {
-	.alloc = xe_ttm_gtt_mgr_new,
-	.free = xe_ttm_gtt_mgr_del,
-	.debug = xe_ttm_gtt_mgr_debug
+static const struct ttm_resource_manager_func xe_ttm_sys_mgr_func = {
+	.alloc = xe_ttm_sys_mgr_new,
+	.free = xe_ttm_sys_mgr_del,
+	.debug = xe_ttm_sys_mgr_debug
 };
 
-static void ttm_gtt_mgr_fini(struct drm_device *drm, void *arg)
+static void ttm_sys_mgr_fini(struct drm_device *drm, void *arg)
 {
-	struct xe_ttm_gtt_mgr *mgr = arg;
-	struct xe_device *xe = gt_to_xe(mgr->gt);
-	struct ttm_resource_manager *man = &mgr->manager;
+	struct xe_device *xe = (struct xe_device *)arg;
+	struct ttm_resource_manager *man = &xe->mem.sys_mgr;
 	int err;
 
 	ttm_resource_manager_set_used(man, false);
@@ -104,27 +96,18 @@ static void ttm_gtt_mgr_fini(struct drm_device *drm, void *arg)
 	ttm_set_driver_manager(&xe->ttm, XE_PL_TT, NULL);
 }
 
-int xe_ttm_gtt_mgr_init(struct xe_gt *gt, struct xe_ttm_gtt_mgr *mgr,
-			u64 gtt_size)
+int xe_ttm_sys_mgr_init(struct xe_device *xe)
 {
-	struct xe_device *xe = gt_to_xe(gt);
-	struct ttm_resource_manager *man = &mgr->manager;
-	int err;
-
-	XE_BUG_ON(xe_gt_is_media_type(gt));
+	struct ttm_resource_manager *man = &xe->mem.sys_mgr;
+	struct sysinfo si;
+	u64 gtt_size;
 
-	mgr->gt = gt;
+	si_meminfo(&si);
+	gtt_size = (u64)si.totalram * si.mem_unit * 3/4;
 	man->use_tt = true;
-	man->func = &xe_ttm_gtt_mgr_func;
-
+	man->func = &xe_ttm_sys_mgr_func;
 	ttm_resource_manager_init(man, &xe->ttm, gtt_size >> PAGE_SHIFT);
-
-	ttm_set_driver_manager(&xe->ttm, XE_PL_TT, &mgr->manager);
+	ttm_set_driver_manager(&xe->ttm, XE_PL_TT, man);
 	ttm_resource_manager_set_used(man, true);
-
-	err = drmm_add_action_or_reset(&xe->drm, ttm_gtt_mgr_fini, mgr);
-	if (err)
-		return err;
-
-	return 0;
+	return drmm_add_action_or_reset(&xe->drm, ttm_sys_mgr_fini, xe);
 }
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 13+ messages in thread