drm/xe/vf: Fix VF CCS attach/detach race with in-flight BO moves

xe_bo_move() attaches VF CCS read/write batch buffers (BBs) to a BO
after it transitions NULL/SYSTEM -> TT, and detaches them after it
transitions TT -> SYSTEM. Both operations were done synchronously on
the CPU immediately after building the move's copy/clear fence,
without waiting for that fence to signal. This creates two races with
VF migration:

- Attach happens too late relative to the copy job it is meant to
  protect. If the copy job is submitted before the CCS BBs are
  attached, a VF migration event that pauses execution mid-copy can
  observe partially copied CCS metadata without the attach state
  needed to correctly save/restore it.

- Detach happens too early relative to the copy job that moves data
  out of TT. The CCS BBs are torn down right after the copy fence is
  obtained, while the actual blit may still be in flight. A VF
  migration event that pauses execution mid-copy can then race the
  save/restore path against the still-running blit, and the CCS BBs
  it would need to make sense of the paused state have already been
  removed.

Fix both races:

- Move the attach call to before the copy/clear job is submitted, so
  the CCS BBs are already registered by the time the copy runs. On
  attach failure, unwind and bail out of the move. xe_migrate_ccs_rw_copy()
  now takes the destination resource explicitly, since bo->ttm.resource
  is not updated to the new resource until after the move commits.

- Detach only after explicitly waiting for the copy fence to signal,
  instead of tearing down the CCS BBs immediately after obtaining it.

While here, also fix xe_sriov_vf_ccs_attach_bo() to properly unwind and
propagate errors: the per-context loop previously never broke out on
error, silently discarding earlier failures. Unwind by clearing each
attached context directly via xe_migrate_ccs_rw_copy_clear() instead of
reusing xe_sriov_vf_ccs_detach_bo(), which requires both contexts to be
attached before it will clean up either one.

Fixes: 864690cf4dd6 ("drm/xe/vf: Attach and detach CCS copy commands with BO")
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Cc: Satyanarayana K V P <satyanarayana.k.v.p@intel.com>
Assisted-by: GitHub_Copilot:claude-sonnet-5
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Acked-by: Satyanarayana K V P <satyanarayana.k.v.p@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patch.msgid.link/20260714062440.3421225-1-matthew.brost@intel.com
(cherry picked from commit d45ad0aa7a1eb5d7288b5ed948b05695611dc39e)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
index ddbaf4242c79471445517f821a7575d008021dd4..7ed76349075fd15bb51e1d487cc33448b86386fe 100644 (file)
--- a/drivers/gpu/drm/xe/xe_bo.c
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -1102,6 +1102,21 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
                xe_pm_runtime_get_noresume(xe);
        }
 
+       /*
+        * Attach CCS BBs before submitting the copy job below so a VF
+        * migration racing the copy sees valid, up to date attach state.
+        */
+       if (IS_VF_CCS_READY(xe) &&
+           ((move_lacks_source && new_mem->mem_type == XE_PL_TT) ||
+            (old_mem_type == XE_PL_SYSTEM && new_mem->mem_type == XE_PL_TT)) &&
+           handle_system_ccs) {
+               ret = xe_sriov_vf_ccs_attach_bo(bo, new_mem);
+               if (ret) {
+                       xe_pm_runtime_put(xe);
+                       goto out;
+               }
+       }
+
        if (move_lacks_source) {
                u32 flags = 0;
 
@@ -1139,22 +1154,19 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
                ttm_bo_move_null(ttm_bo, new_mem);
        }
 
-       dma_fence_put(fence);
-       xe_pm_runtime_put(xe);
-
        /*
-        * CCS meta data is migrated from TT -> SMEM. So, let us detach the
-        * BBs from BO as it is no longer needed.
+        * Detach must wait for the copy above to complete: a VF migration
+        * racing an in-flight copy must still see valid CCS BBs, so don't
+        * tear them down until the copy fence has signaled.
         */
        if (IS_VF_CCS_READY(xe) && old_mem_type == XE_PL_TT &&
-           new_mem->mem_type == XE_PL_SYSTEM)
+           new_mem->mem_type == XE_PL_SYSTEM) {
+               dma_fence_wait(fence, false);
                xe_sriov_vf_ccs_detach_bo(bo);
+       }
 
-       if (IS_VF_CCS_READY(xe) &&
-           ((move_lacks_source && new_mem->mem_type == XE_PL_TT) ||
-            (old_mem_type == XE_PL_SYSTEM && new_mem->mem_type == XE_PL_TT)) &&
-           handle_system_ccs)
-               ret = xe_sriov_vf_ccs_attach_bo(bo);
+       dma_fence_put(fence);
+       xe_pm_runtime_put(xe);
 
 out:
        if ((!ttm_bo->resource || ttm_bo->resource->mem_type == XE_PL_SYSTEM) &&

diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c
index 9428dd5e7760c50bc35d22d3b36f07abf0358a54..7d28290e7d1ca38f092b8a2a29ab4852b0fe0e76 100644 (file)
--- a/drivers/gpu/drm/xe/xe_migrate.c
+++ b/drivers/gpu/drm/xe/xe_migrate.c
@@ -1166,6 +1166,8 @@ static int emit_flush_invalidate(u32 *dw, int i, u32 flags)
  * @tile: Tile whose migration context to be used.
  * @q : Execution to be used along with migration context.
  * @src_bo: The buffer object @src is currently bound to.
+ * @new_mem: The (not yet committed) destination resource @src_bo is being
+ *          moved into; src_bo->ttm.resource is still the old resource.
  * @read_write : Creates BB commands for CCS read/write.
  *
  * Creates batch buffer instructions to copy CCS metadata from CCS pool to
@@ -1177,12 +1179,13 @@ static int emit_flush_invalidate(u32 *dw, int i, u32 flags)
  */
 int xe_migrate_ccs_rw_copy(struct xe_tile *tile, struct xe_exec_queue *q,
                           struct xe_bo *src_bo,
+                          struct ttm_resource *new_mem,
                           enum xe_sriov_vf_ccs_rw_ctxs read_write)
 
 {
        bool src_is_pltt = read_write == XE_SRIOV_VF_CCS_READ_CTX;
        bool dst_is_pltt = read_write == XE_SRIOV_VF_CCS_WRITE_CTX;
-       struct ttm_resource *src = src_bo->ttm.resource;
+       struct ttm_resource *src = new_mem;
        struct xe_migrate *m = tile->migrate;
        struct xe_gt *gt = tile->primary_gt;
        u32 batch_size, batch_size_allocated;

diff --git a/drivers/gpu/drm/xe/xe_migrate.h b/drivers/gpu/drm/xe/xe_migrate.h
index 965c45889c726f19604f0d7c4b21d6986f96b05a..78e5b63f3ebeafc95d59862972e6b5a566f0d34a 100644 (file)
--- a/drivers/gpu/drm/xe/xe_migrate.h
+++ b/drivers/gpu/drm/xe/xe_migrate.h
@@ -138,6 +138,7 @@ struct dma_fence *xe_migrate_resolve(struct xe_migrate *m,
 
 int xe_migrate_ccs_rw_copy(struct xe_tile *tile, struct xe_exec_queue *q,
                           struct xe_bo *src_bo,
+                          struct ttm_resource *new_mem,
                           enum xe_sriov_vf_ccs_rw_ctxs read_write);
 
 void xe_migrate_ccs_rw_copy_clear(struct xe_bo *src_bo,

diff --git a/drivers/gpu/drm/xe/xe_sriov_vf_ccs.c b/drivers/gpu/drm/xe/xe_sriov_vf_ccs.c
index 09b99fb2608bce103f8c97bde4a250ec482035b9..6787564629c65f8315e62d28eb153617f787b3a6 100644 (file)
--- a/drivers/gpu/drm/xe/xe_sriov_vf_ccs.c
+++ b/drivers/gpu/drm/xe/xe_sriov_vf_ccs.c
@@ -404,6 +404,8 @@ void xe_sriov_vf_ccs_rw_update_bb_addr(struct xe_sriov_vf_ccs_ctx *ctx)
 /**
  * xe_sriov_vf_ccs_attach_bo - Insert CCS read write commands in the BO.
  * @bo: the &buffer object to which batch buffer commands will be added.
+ * @new_mem: the (not yet committed) destination resource @bo is being moved
+ *          into; bo->ttm.resource is still the old resource at this point.
  *
  * This function shall be called only by VF. It inserts the PTEs and copy
  * command instructions in the BO by calling xe_migrate_ccs_rw_copy()
@@ -411,7 +413,7 @@ void xe_sriov_vf_ccs_rw_update_bb_addr(struct xe_sriov_vf_ccs_ctx *ctx)
  *
  * Returns: 0 if successful, negative error code on failure.
  */
-int xe_sriov_vf_ccs_attach_bo(struct xe_bo *bo)
+int xe_sriov_vf_ccs_attach_bo(struct xe_bo *bo, struct ttm_resource *new_mem)
 {
        struct xe_device *xe = xe_bo_device(bo);
        enum xe_sriov_vf_ccs_rw_ctxs ctx_id;
@@ -430,7 +432,21 @@ int xe_sriov_vf_ccs_attach_bo(struct xe_bo *bo)
                xe_assert(xe, !bb);
 
                ctx = &xe->sriov.vf.ccs.contexts[ctx_id];
-               err = xe_migrate_ccs_rw_copy(tile, ctx->mig_q, bo, ctx_id);
+               err = xe_migrate_ccs_rw_copy(tile, ctx->mig_q, bo, new_mem, ctx_id);
+               if (err)
+                       goto err_unwind;
+       }
+       return 0;
+
+err_unwind:
+       /*
+        * Clean up any contexts already attached. Can't reuse
+        * xe_sriov_vf_ccs_detach_bo() here as it requires both contexts
+        * attached before cleaning up either one.
+        */
+       for_each_ccs_rw_ctx(ctx_id) {
+               if (bo->bb_ccs[ctx_id])
+                       xe_migrate_ccs_rw_copy_clear(bo, ctx_id);
        }
        return err;
 }

diff --git a/drivers/gpu/drm/xe/xe_sriov_vf_ccs.h b/drivers/gpu/drm/xe/xe_sriov_vf_ccs.h
index 00e58b36c510aca68dd330bdf71868afc76d891a..e1034d852104069edde8a216da6be335aac3e5d0 100644 (file)
--- a/drivers/gpu/drm/xe/xe_sriov_vf_ccs.h
+++ b/drivers/gpu/drm/xe/xe_sriov_vf_ccs.h
@@ -11,11 +11,12 @@
 #include "xe_sriov_vf_ccs_types.h"
 
 struct drm_printer;
+struct ttm_resource;
 struct xe_device;
 struct xe_bo;
 
 int xe_sriov_vf_ccs_init(struct xe_device *xe);
-int xe_sriov_vf_ccs_attach_bo(struct xe_bo *bo);
+int xe_sriov_vf_ccs_attach_bo(struct xe_bo *bo, struct ttm_resource *new_mem);
 int xe_sriov_vf_ccs_detach_bo(struct xe_bo *bo);
 int xe_sriov_vf_ccs_register_context(struct xe_device *xe);
 void xe_sriov_vf_ccs_rebase(struct xe_device *xe);