On 24.06.2022 13:08, Nirmoy Das wrote:
For some platfroms we use stop_machine version of
gen8_ggtt_insert_page/gen8_ggtt_insert_entries to avoid a
concurrent GGTT access bug but this causes a circular locking
dependency warning:
Possible unsafe locking scenario:
CPU0 CPU1
---- ----
lock(&ggtt->error_mutex);
lock(dma_fence_map);
lock(&ggtt->error_mutex);
lock(cpu_hotplug_lock);
Fix this by calling gen8_ggtt_insert_page/gen8_ggtt_insert_entries
directly at error capture which is concurrent GGTT access safe because
reset path make sure of that.
v2: Fix rebase conflict and added a comment.
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/5595
Reviewed-by: Gwan-gyeong Mun <gwan-gyeong.mun@xxxxxxxxx>
Suggested-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Signed-off-by: Nirmoy Das <nirmoy.das@xxxxxxxxx>
---
drivers/gpu/drm/i915/gt/intel_ggtt.c | 10 ++++++++++
drivers/gpu/drm/i915/gt/intel_gtt.h | 9 +++++++++
drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c | 5 ++++-
drivers/gpu/drm/i915/i915_gpu_error.c | 8 ++++++--
4 files changed, 29 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
index ffff96180313..15a915bb4088 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
@@ -960,6 +960,16 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt)
if (intel_vm_no_concurrent_access_wa(i915)) {
ggtt->vm.insert_entries = bxt_vtd_ggtt_insert_entries__BKL;
ggtt->vm.insert_page = bxt_vtd_ggtt_insert_page__BKL;
+
+ /*
+ * Calling stop_machine() version of GGTT update function
+ * at error capture/reset path will raise lockdep warning.
+ * Allow calling gen8_ggtt_insert_* directly at reset path
+ * which is safe from parallel GGTT updates.
+ */
+ ggtt->vm.raw_insert_page = gen8_ggtt_insert_page;
+ ggtt->vm.raw_insert_entries = gen8_ggtt_insert_entries;
+
ggtt->vm.bind_async_flags =
I915_VMA_GLOBAL_BIND | I915_VMA_LOCAL_BIND;
}
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.h b/drivers/gpu/drm/i915/gt/intel_gtt.h
index 29fd3a9e8b2e..e639434e97fd 100644
--- a/drivers/gpu/drm/i915/gt/intel_gtt.h
+++ b/drivers/gpu/drm/i915/gt/intel_gtt.h
@@ -306,6 +306,15 @@ struct i915_address_space {
struct i915_vma_resource *vma_res,
enum i915_cache_level cache_level,
u32 flags);
+ void (*raw_insert_page)(struct i915_address_space *vm,
+ dma_addr_t addr,
+ u64 offset,
+ enum i915_cache_level cache_level,
+ u32 flags);
+ void (*raw_insert_entries)(struct i915_address_space *vm,
+ struct i915_vma_resource *vma_res,
+ enum i915_cache_level cache_level,
+ u32 flags);
I would expect rather extra flag to insert_(page|entries) instead of
extra callbacks. Anyway both should work.
Reviewed-by: Andrzej Hajda <andrzej.hajda@xxxxxxxxx>
Regards
Andrzej
void (*cleanup)(struct i915_address_space *vm);
void (*foreach)(struct i915_address_space *vm,
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c b/drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c
index d2c5c9367cc4..c06e83872c34 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c
@@ -493,7 +493,10 @@ static void uc_fw_bind_ggtt(struct intel_uc_fw *uc_fw)
if (i915_gem_object_is_lmem(obj))
pte_flags |= PTE_LM;
- ggtt->vm.insert_entries(&ggtt->vm, dummy, I915_CACHE_NONE, pte_flags);
+ if (ggtt->vm.raw_insert_entries)
+ ggtt->vm.raw_insert_entries(&ggtt->vm, dummy, I915_CACHE_NONE, pte_flags);
+ else
+ ggtt->vm.insert_entries(&ggtt->vm, dummy, I915_CACHE_NONE, pte_flags);
}
static void uc_fw_unbind_ggtt(struct intel_uc_fw *uc_fw)
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index bff8a111424a..f9b1969ed7ed 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -1104,8 +1104,12 @@ i915_vma_coredump_create(const struct intel_gt *gt,
for_each_sgt_daddr(dma, iter, vma_res->bi.pages) {
mutex_lock(&ggtt->error_mutex);
- ggtt->vm.insert_page(&ggtt->vm, dma, slot,
- I915_CACHE_NONE, 0);
+ if (ggtt->vm.raw_insert_page)
+ ggtt->vm.raw_insert_page(&ggtt->vm, dma, slot,
+ I915_CACHE_NONE, 0);
+ else
+ ggtt->vm.insert_page(&ggtt->vm, dma, slot,
+ I915_CACHE_NONE, 0);
mb();
s = io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE);
