* [PATCH v2 1/2] drm/xe/drm_ras: Protect against unplugged or wedged device
2026-11-18 15:31 [PATCH v2 0/2] Protect drm_ras against unplugged or wedged device Raag Jadav
@ 2026-11-18 15:31 ` Raag Jadav
2026-11-18 15:31 ` [PATCH v2 2/2] drm/xe/wedge: Introduce xe_wedge_mode_is_no_reset() Raag Jadav
1 sibling, 0 replies; 4+ messages in thread
From: Raag Jadav @ 2026-11-18 15:31 UTC (permalink / raw)
To: intel-xe; +Cc: riana.tauro, stuart.summers, Raag Jadav
Xe driver exposes drm_ras counters to userspace through netlink interface,
which the user can access during the device lifetime. But this should not
be done in case of unplugged or wedged device. Protect against such cases.
Fixes: b40db12b542f ("drm/xe/xe_drm_ras: Add support for XE DRM RAS")
Signed-off-by: Raag Jadav <raag.jadav@intel.com>
Reviewed-by: Stuart Summers <stuart.summers@intel.com>
---
v2: Allow counter access with XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET (Riana)
---
drivers/gpu/drm/xe/xe_drm_ras.c | 93 ++++++++++++++++++++++++---------
1 file changed, 69 insertions(+), 24 deletions(-)
diff --git a/drivers/gpu/drm/xe/xe_drm_ras.c b/drivers/gpu/drm/xe/xe_drm_ras.c
index 7554803aae11..3ca6c82345dd 100644
--- a/drivers/gpu/drm/xe/xe_drm_ras.c
+++ b/drivers/gpu/drm/xe/xe_drm_ras.c
@@ -5,6 +5,7 @@
#include <linux/bitmap.h>
+#include <drm/drm_drv.h>
#include <drm/drm_managed.h>
#include <drm/drm_print.h>
#include <drm/drm_ras.h>
@@ -12,10 +13,33 @@
#include "xe_device_types.h"
#include "xe_drm_ras.h"
#include "xe_ras.h"
+#include "xe_wedge.h"
static const char * const error_components[] = DRM_XE_RAS_ERROR_COMPONENT_NAMES;
static const char * const error_severity[] = DRM_XE_RAS_ERROR_SEVERITY_NAMES;
+static int drm_ras_counter_get(struct xe_device *xe, struct xe_drm_ras_counter *info,
+ u32 error_id, int *idx)
+{
+ if (xe_wedged(xe) && xe->wedged.mode != XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET)
+ return -ECANCELED;
+
+ if (!info || !info[error_id].name)
+ return -ENOENT;
+
+ /* Must be last */
+ if (!drm_dev_enter(&xe->drm, idx))
+ return -ENODEV;
+
+ return 0;
+}
+
+static void drm_ras_counter_put(int idx)
+{
+ /* Pair with drm_ras_counter_get() */
+ drm_dev_exit(idx);
+}
+
static bool drm_ras_counter_is_valid(u32 severity, u32 component)
{
if (!in_range(severity, DRM_XE_RAS_ERR_SEV_CORRECTABLE, DRM_XE_RAS_ERR_SEV_MAX))
@@ -33,19 +57,23 @@ static int query_error_counter(struct xe_device *xe,
{
struct xe_drm_ras *ras = &xe->ras.nl_data;
struct xe_drm_ras_counter *info = ras->info[severity];
+ int idx, ret;
- if (!info || !info[error_id].name)
- return -ENOENT;
-
- *name = info[error_id].name;
+ ret = drm_ras_counter_get(xe, info, error_id, &idx);
+ if (ret)
+ return ret;
/* Fetch counter from system controller if supported */
if (xe->info.has_sysctrl)
- return xe_ras_get_counter(xe, severity, error_id, val);
+ ret = xe_ras_get_counter(xe, severity, error_id, val);
+ else
+ *val = atomic_read(&info[error_id].counter);
- *val = atomic_read(&info[error_id].counter);
+ if (!ret)
+ *name = info[error_id].name;
- return 0;
+ drm_ras_counter_put(idx);
+ return ret;
}
static int clear_error_counter(struct xe_device *xe,
@@ -54,17 +82,20 @@ static int clear_error_counter(struct xe_device *xe,
{
struct xe_drm_ras *ras = &xe->ras.nl_data;
struct xe_drm_ras_counter *info = ras->info[severity];
+ int idx, ret;
- if (!info || !info[error_id].name)
- return -ENOENT;
+ ret = drm_ras_counter_get(xe, info, error_id, &idx);
+ if (ret)
+ return ret;
/* Clear counter from system controller if supported */
if (xe->info.has_sysctrl)
- return xe_ras_clear_counter(xe, severity, error_id);
+ ret = xe_ras_clear_counter(xe, severity, error_id);
+ else
+ atomic_set(&info[error_id].counter, 0);
- atomic_set(&info[error_id].counter, 0);
-
- return 0;
+ drm_ras_counter_put(idx);
+ return ret;
}
static int query_uncorrectable_error_counter(struct drm_ras_node *ep, u32 error_id,
@@ -103,15 +134,21 @@ static int query_correctable_error_threshold(struct drm_ras_node *ep, u32 error_
struct xe_device *xe = ep->priv;
struct xe_drm_ras *ras = &xe->ras.nl_data;
struct xe_drm_ras_counter *info = ras->info[DRM_XE_RAS_ERR_SEV_CORRECTABLE];
-
- if (!info || !info[error_id].name)
- return -ENOENT;
+ int idx, ret;
if (!xe->info.has_sysctrl)
return -EOPNOTSUPP;
- *name = info[error_id].name;
- return xe_ras_get_threshold(xe, DRM_XE_RAS_ERR_SEV_CORRECTABLE, error_id, threshold);
+ ret = drm_ras_counter_get(xe, info, error_id, &idx);
+ if (ret)
+ return ret;
+
+ ret = xe_ras_get_threshold(xe, DRM_XE_RAS_ERR_SEV_CORRECTABLE, error_id, threshold);
+ if (!ret)
+ *name = info[error_id].name;
+
+ drm_ras_counter_put(idx);
+ return ret;
}
static int set_correctable_error_threshold(struct drm_ras_node *ep, u32 error_id, u32 threshold)
@@ -119,14 +156,19 @@ static int set_correctable_error_threshold(struct drm_ras_node *ep, u32 error_id
struct xe_device *xe = ep->priv;
struct xe_drm_ras *ras = &xe->ras.nl_data;
struct xe_drm_ras_counter *info = ras->info[DRM_XE_RAS_ERR_SEV_CORRECTABLE];
-
- if (!info || !info[error_id].name)
- return -ENOENT;
+ int idx, ret;
if (!xe->info.has_sysctrl)
return -EOPNOTSUPP;
- return xe_ras_set_threshold(xe, DRM_XE_RAS_ERR_SEV_CORRECTABLE, error_id, threshold);
+ ret = drm_ras_counter_get(xe, info, error_id, &idx);
+ if (ret)
+ return ret;
+
+ ret = xe_ras_set_threshold(xe, DRM_XE_RAS_ERR_SEV_CORRECTABLE, error_id, threshold);
+
+ drm_ras_counter_put(idx);
+ return ret;
}
static struct xe_drm_ras_counter *allocate_and_copy_counters(struct xe_device *xe)
@@ -244,7 +286,7 @@ void xe_drm_ras_event(struct xe_device *xe, u32 component, u32 severity, u32 val
struct xe_drm_ras *ras = &xe->ras.nl_data;
struct xe_drm_ras_counter *info;
struct drm_ras_node *node;
- int ret;
+ int idx, ret;
/* Event is supported only if drm-ras is enabled */
if (!xe->info.has_drm_ras)
@@ -256,13 +298,16 @@ void xe_drm_ras_event(struct xe_device *xe, u32 component, u32 severity, u32 val
node = &ras->node[severity];
info = ras->info[severity];
- if (!info || !info[component].name)
+ ret = drm_ras_counter_get(xe, info, component, &idx);
+ if (ret)
return;
ret = drm_ras_nl_error_event(node, component, info[component].name, value);
if (ret)
drm_err_ratelimited(&xe->drm, "drm-ras error-event failed: %d for %s %s\n", ret,
info[component].name, error_severity[severity]);
+
+ drm_ras_counter_put(idx);
}
/**
--
2.43.0
^ permalink raw reply related [flat|nested] 4+ messages in thread* [PATCH v2 2/2] drm/xe/wedge: Introduce xe_wedge_mode_is_no_reset()
2026-11-18 15:31 [PATCH v2 0/2] Protect drm_ras against unplugged or wedged device Raag Jadav
2026-11-18 15:31 ` [PATCH v2 1/2] drm/xe/drm_ras: Protect " Raag Jadav
@ 2026-11-18 15:31 ` Raag Jadav
2026-11-18 15:44 ` sashiko-bot
1 sibling, 1 reply; 4+ messages in thread
From: Raag Jadav @ 2026-11-18 15:31 UTC (permalink / raw)
To: intel-xe; +Cc: riana.tauro, stuart.summers, Raag Jadav
Convert open-coded XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET check into its
own helper.
Signed-off-by: Raag Jadav <raag.jadav@intel.com>
---
v2: Introduce xe_wedge_mode_is_no_reset()
---
drivers/gpu/drm/xe/xe_drm_ras.c | 2 +-
drivers/gpu/drm/xe/xe_guc_ads.c | 3 ++-
drivers/gpu/drm/xe/xe_guc_capture.c | 3 ++-
drivers/gpu/drm/xe/xe_guc_submit.c | 2 +-
drivers/gpu/drm/xe/xe_pci_error.c | 2 +-
drivers/gpu/drm/xe/xe_wedge.c | 7 +++----
drivers/gpu/drm/xe/xe_wedge.h | 5 +++++
7 files changed, 15 insertions(+), 9 deletions(-)
diff --git a/drivers/gpu/drm/xe/xe_drm_ras.c b/drivers/gpu/drm/xe/xe_drm_ras.c
index 3ca6c82345dd..86c589437a9f 100644
--- a/drivers/gpu/drm/xe/xe_drm_ras.c
+++ b/drivers/gpu/drm/xe/xe_drm_ras.c
@@ -21,7 +21,7 @@ static const char * const error_severity[] = DRM_XE_RAS_ERROR_SEVERITY_NAMES;
static int drm_ras_counter_get(struct xe_device *xe, struct xe_drm_ras_counter *info,
u32 error_id, int *idx)
{
- if (xe_wedged(xe) && xe->wedged.mode != XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET)
+ if (xe_wedged(xe) && !xe_wedge_mode_is_no_reset(xe))
return -ECANCELED;
if (!info || !info[error_id].name)
diff --git a/drivers/gpu/drm/xe/xe_guc_ads.c b/drivers/gpu/drm/xe/xe_guc_ads.c
index 0e1674a394b6..f333480f94e5 100644
--- a/drivers/gpu/drm/xe/xe_guc_ads.c
+++ b/drivers/gpu/drm/xe/xe_guc_ads.c
@@ -30,6 +30,7 @@
#include "xe_map.h"
#include "xe_mmio.h"
#include "xe_wa.h"
+#include "xe_wedge.h"
/* Slack of a few additional entries per engine */
#define ADS_REGSET_EXTRA_MAX 8
@@ -544,7 +545,7 @@ static void guc_policies_init(struct xe_guc_ads *ads)
ads_blob_write(ads, policies.max_num_work_items,
GLOBAL_POLICY_MAX_NUM_WI);
- if (xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET)
+ if (xe_wedge_mode_is_no_reset(xe))
global_flags |= GLOBAL_POLICY_DISABLE_ENGINE_RESET;
ads_blob_write(ads, policies.global_flags, global_flags);
diff --git a/drivers/gpu/drm/xe/xe_guc_capture.c b/drivers/gpu/drm/xe/xe_guc_capture.c
index 0fd68b41d412..e528b3e9d158 100644
--- a/drivers/gpu/drm/xe/xe_guc_capture.c
+++ b/drivers/gpu/drm/xe/xe_guc_capture.c
@@ -35,6 +35,7 @@
#include "xe_map.h"
#include "xe_mmio.h"
#include "xe_sched_job.h"
+#include "xe_wedge.h"
/*
* struct __guc_capture_bufstate
@@ -1962,7 +1963,7 @@ xe_guc_capture_get_matching_and_lock(struct xe_exec_queue *q)
xe = gt_to_xe(q->gt);
- if (xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET)
+ if (xe_wedge_mode_is_no_reset(xe))
return NULL;
if (!xe_device_uc_enabled(xe))
diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c
index b5c916a764db..fc0f50307f2f 100644
--- a/drivers/gpu/drm/xe/xe_guc_submit.c
+++ b/drivers/gpu/drm/xe/xe_guc_submit.c
@@ -1435,7 +1435,7 @@ void xe_guc_submit_wedge(struct xe_guc *guc)
if (!guc->submission_state.initialized)
return;
- if (xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET) {
+ if (xe_wedge_mode_is_no_reset(xe)) {
mutex_lock(&guc->submission_state.lock);
xa_for_each(&guc->submission_state.exec_queue_lookup, index, q)
if (xe_exec_queue_get_unless_zero(q))
diff --git a/drivers/gpu/drm/xe/xe_pci_error.c b/drivers/gpu/drm/xe/xe_pci_error.c
index d288bb291aa1..fc6ff71d84a0 100644
--- a/drivers/gpu/drm/xe/xe_pci_error.c
+++ b/drivers/gpu/drm/xe/xe_pci_error.c
@@ -179,7 +179,7 @@ static pci_ers_result_t xe_pci_error_mmio_enabled(struct pci_dev *pdev)
action = xe_ras_process_errors(xe);
/* User wants to debug the error, prevent reset */
- if (xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET) {
+ if (xe_wedge_mode_is_no_reset(xe)) {
xe_wedge_declare(xe, DRM_WEDGE_RECOVERY_NONE);
return PCI_ERS_RESULT_DISCONNECT;
}
diff --git a/drivers/gpu/drm/xe/xe_wedge.c b/drivers/gpu/drm/xe/xe_wedge.c
index 827fae058976..3b4c8bee6797 100644
--- a/drivers/gpu/drm/xe/xe_wedge.c
+++ b/drivers/gpu/drm/xe/xe_wedge.c
@@ -157,7 +157,7 @@ void xe_wedge_declare(struct xe_device *xe, unsigned long method)
* hangs, so wedge the device with 'none' recovery method and have
* it available to the user for debugging.
*/
- if (xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET)
+ if (xe_wedge_mode_is_no_reset(xe))
xe->wedged.method = DRM_WEDGE_RECOVERY_NONE;
else
xe->wedged.method = method;
@@ -184,7 +184,7 @@ static int wedge_validate_mode(struct xe_device *xe, unsigned int mode)
if (mode > XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET) {
xe_dbg(xe, "wedged_mode: invalid value (%u)\n", mode);
return -EINVAL;
- } else if (mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET && (IS_SRIOV_VF(xe) ||
+ } else if (xe_wedge_mode_is_no_reset(xe) && (IS_SRIOV_VF(xe) ||
(IS_SRIOV_PF(xe) && !IS_ENABLED(CONFIG_DRM_XE_DEBUG)))) {
xe_dbg(xe, "wedged_mode: (%u) %s mode is not supported for %s\n",
mode, wedge_mode_to_string(mode),
@@ -203,8 +203,7 @@ static bool wedge_mode_needs_policy_update(struct xe_device *xe, enum xe_wedged_
if (xe->wedged.mode == mode)
return false;
- if (xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET ||
- mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET)
+ if (xe_wedge_mode_is_no_reset(xe) || mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET)
return true;
return false;
diff --git a/drivers/gpu/drm/xe/xe_wedge.h b/drivers/gpu/drm/xe/xe_wedge.h
index f2cab855db35..edb5174c5a75 100644
--- a/drivers/gpu/drm/xe/xe_wedge.h
+++ b/drivers/gpu/drm/xe/xe_wedge.h
@@ -26,6 +26,11 @@ static inline bool xe_wedged_permanent(struct xe_device *xe)
return atomic_read(&xe->wedged.perm);
}
+static inline bool xe_wedge_mode_is_no_reset(struct xe_device *xe)
+{
+ return xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET;
+}
+
void xe_wedge_declare(struct xe_device *xe, unsigned long method);
int xe_wedge_init(struct xe_device *xe);
int xe_wedge_set_mode(struct xe_device *xe, enum xe_wedged_mode mode);
--
2.43.0
^ permalink raw reply related [flat|nested] 4+ messages in thread