[PATCH] drm/xe/pf: Release all VFs configs on device removal

Michal Wajdeczko michal.wajdeczko at intel.com
Mon Feb 10 22:59:44 UTC 2025


If we try to manually provision VFs using debugfs and then we
try to unload the driver, we will see complains like:

 [ ] Memory manager not clean during takedown.
 [ ] RIP: 0010:drm_mm_takedown+0x3f/0x100
 [ ] [drm:drm_mm_takedown] *ERROR* node [fedff000 + 00001000]: inserted at
      drm_mm_insert_node_in_range+0x2bd/0x520
      xe_ggtt_node_insert+0x52/0x90 [xe]
      pf_provision_vf_ggtt+0x1fa/0xac0 [xe]
      xe_gt_sriov_pf_config_set_ggtt+0x79/0x7a0 [xe]
      ggtt_set+0x53/0x80 [xe]
      simple_attr_write_xsigned.isra.0+0xd2/0x150
      simple_attr_write+0x14/0x30
      debugfs_attr_write+0x4e/0x80

 [ ] xe 0000:00:02.0: [drm] *ERROR* GT0: GUC ID manager unclean (1/65535)
 [ ] xe 0000:00:02.0: [drm] GT0:      total 65535
 [ ] xe 0000:00:02.0: [drm] GT0:      used 1
 [ ] xe 0000:00:02.0: [drm] GT0:      range 65534..65534 (1)

 [ ] xe 0000:00:02.0: [drm] *ERROR* GT0: GuC doorbells manager unclean (1/256)
 [ ] xe 0000:00:02.0: [drm] GT0:      count: 256
 [ ] xe 0000:00:02.0: [drm] GT0:      available range: 1..255 (255)
 [ ] xe 0000:00:02.0: [drm] GT0:      available total: 255
 [ ] xe 0000:00:02.0: [drm] GT0:      reserved range: 0..0 (1)
 [ ] xe 0000:00:02.0: [drm] GT0:      reserved total: 1

This could be easily fixed by adding config release action.

Signed-off-by: Michal Wajdeczko <michal.wajdeczko at intel.com>
---
 drivers/gpu/drm/xe/xe_gt_sriov_pf.c        |  6 ++++++
 drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c | 25 ++++++++++++++++++++++
 drivers/gpu/drm/xe/xe_gt_sriov_pf_config.h |  1 +
 3 files changed, 32 insertions(+)

diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf.c b/drivers/gpu/drm/xe/xe_gt_sriov_pf.c
index d66478deab98..c08efca6420e 100644
--- a/drivers/gpu/drm/xe/xe_gt_sriov_pf.c
+++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf.c
@@ -89,6 +89,12 @@ int xe_gt_sriov_pf_init_early(struct xe_gt *gt)
  */
 int xe_gt_sriov_pf_init(struct xe_gt *gt)
 {
+	int err;
+
+	err = xe_gt_sriov_pf_config_init(gt);
+	if (err)
+		return err;
+
 	return xe_gt_sriov_pf_migration_init(gt);
 }
 
diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c b/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c
index 88bd9d97ba5c..3e2446277e48 100644
--- a/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c
+++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c
@@ -2356,6 +2356,31 @@ int xe_gt_sriov_pf_config_restore(struct xe_gt *gt, unsigned int vfid,
 	return err;
 }
 
+static void fini_config(void *arg)
+{
+	struct xe_gt *gt = arg;
+	struct xe_device *xe = gt_to_xe(gt);
+	unsigned int n, total_vfs = xe_sriov_pf_get_totalvfs(xe);
+
+	mutex_lock(xe_gt_sriov_pf_master_mutex(gt));
+	for (n = 1; n <= total_vfs; n++)
+		pf_release_vf_config(gt, n);
+	mutex_unlock(xe_gt_sriov_pf_master_mutex(gt));
+}
+
+/**
+ * xe_gt_sriov_pf_config_init - Initialize SR-IOV configuration data.
+ * @gt: the &xe_gt
+ *
+ * This function can only be called on PF.
+ *
+ * Return: 0 on success or a negative error code on failure.
+ */
+int xe_gt_sriov_pf_config_init(struct xe_gt *gt)
+{
+	return devm_add_action_or_reset(gt_to_xe(gt)->drm.dev, fini_config, gt);
+}
+
 /**
  * xe_gt_sriov_pf_config_restart - Restart SR-IOV configurations after a GT reset.
  * @gt: the &xe_gt
diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.h b/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.h
index f894e9d4abba..513e6512a575 100644
--- a/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.h
+++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.h
@@ -63,6 +63,7 @@ int xe_gt_sriov_pf_config_restore(struct xe_gt *gt, unsigned int vfid,
 
 bool xe_gt_sriov_pf_config_is_empty(struct xe_gt *gt, unsigned int vfid);
 
+int xe_gt_sriov_pf_config_init(struct xe_gt *gt);
 void xe_gt_sriov_pf_config_restart(struct xe_gt *gt);
 
 int xe_gt_sriov_pf_config_print_ggtt(struct xe_gt *gt, struct drm_printer *p);
-- 
2.47.1



More information about the Intel-xe mailing list