drm/panfrost: Fix races between perfcnt and reset sequence

Formerly, the reset sequence would race with panfrost_mmu_as_put()
when tearing down a perfcnt session. On top of that, poking GPU
registers to program a perfcnt session or obtaining a dump might lead to
undefined behaviour when done at the same time a reset was ongoing.

Use the reset r/w semaphore when disabling and re-enabling perfcnt
configuration, and also in the sections inside the 'enable'and 'dump'
ioctls where device registers are being accessed.

On top of that, expand the DRM uAPI for the perfcnt DUMP operation
so that user space can be made aware of a reset having happened,
whether it succeeded or failed to restore the original configuration.
UM needs to know about this condition because counter data is inaccurate
after a reset, so the best approach might be simply to try again.

Finally, update driver uAPI documentation to explain the meaning of the
new perfcnt dump ioctl's state flags, and bump DRM driver minor number
to reflect the new DUMP IOCTL req field.

Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Fixes: 73e467f60acd ("drm/panfrost: Consolidate reset handling")
Fixes: 7786fd108777 ("drm/panfrost: Expose performance counters through unstable ioctls")
Reviewed-by: Steven Price <steven.price@arm.com>
Link: https://patch.msgid.link/20261012-claude-fixes-v15-15-bc0a6c498553@collabora.com
Signed-off-by: Adrián Larumbe <adrian.larumbe@collabora.com>

1 files changed, 28 insertions, 1 deletions

diff --git a/include/uapi/drm/panfrost_drm.h b/include/uapi/drm/panfrost_drm.h
index 50d5337f35ef..7831c48c59c0 100644
--- a/include/uapi/drm/panfrost_drm.h
+++ b/include/uapi/drm/panfrost_drm.h
@@ -47,7 +47,7 @@ extern "C" {
  * them for anything but debugging purpose.
  */
 #define DRM_IOCTL_PANFROST_PERFCNT_ENABLE	DRM_IOW(DRM_COMMAND_BASE + DRM_PANFROST_PERFCNT_ENABLE, struct drm_panfrost_perfcnt_enable)
-#define DRM_IOCTL_PANFROST_PERFCNT_DUMP		DRM_IOW(DRM_COMMAND_BASE + DRM_PANFROST_PERFCNT_DUMP, struct drm_panfrost_perfcnt_dump)
+#define DRM_IOCTL_PANFROST_PERFCNT_DUMP		DRM_IOWR(DRM_COMMAND_BASE + DRM_PANFROST_PERFCNT_DUMP, struct drm_panfrost_perfcnt_dump)
 
 #define PANFROST_JD_REQ_FS (1 << 0)
 #define PANFROST_JD_REQ_CYCLE_COUNT (1 << 1)
@@ -270,8 +270,35 @@ struct drm_panfrost_perfcnt_enable {
 	__u32 counterset;
 };
 
+/*
+ * The next two flags describe the state of a perfcnt dump request
+ * as influenced by a device reset. They are the only values the
+ * perfcnt_dump ioctl state field can take on.
+ * Only certain state and ioctl retval combinations are legitimate.
+ */
+
+/* A new perfcnt_enable ioctl should be issued before requesting
+ * more dumps, because a HW reset failed to recreate perfcnt's
+ * original state. Otherwise further perfcnt_dump's will fail.
+ * This flag being set means ioctl's retval is always -EIO.
+ */
+#define PANFROST_PERFCNT_SESSION_DEAD (1 << 0)
+
+/* A HW reset happened before or during a sample request, and
+ * perfcnt's internal state was successfully restored. There
+ * are two possible outcomes depending on the ioctl's retval:
+ *	0: A reset happened before a dump was requested, but did
+ * nonetheless succeed. Counter values are relative to last reset.
+ *	-EAGAIN: A reset happened when a counter values sampling
+ * request was ongoing. Values are undefined so a new dump ioctl
+ * should be issued.
+ */
+#define PANFROST_PERFCNT_SESSION_INTERRUPTED_BY_RESET (1 << 1)
+
 struct drm_panfrost_perfcnt_dump {
 	__u64 buf_ptr;
+	__u32 state;
+	__u32 pad;		/* MBZ */
 };
 
 /* madvise provides a way to tell the kernel in case a buffers contents