[PATCH v2 08/12] EDAC/versalnet: Guard against NULL drvdata in rpmsg_cb()

From: Shubhrajyoti Datta

Date: Wed Nov 18 2026 - 09:24:23 EST


The rpmsg core creates its own endpoint bound to rpmsg_cb() in
rpmsg_dev_probe(), before rpmsg_probe() runs and attaches mc_priv via
dev_set_drvdata(). A message dispatched to that endpoint in that
window would dereference a NULL mc_priv and crash.

The same race exists on the teardown side: rpmsg_dev_remove() calls
rpmsg_remove() first, which clears drvdata back to NULL, and only
destroys the core's default endpoint afterwards. A message arriving
in that window reaches rpmsg_cb() with drvdata already NULL.

Check for a NULL mc_priv in rpmsg_cb() and drop the message instead
of crashing, in either window.

Fixes: d5fe2fec6c40 ("EDAC: Add a driver for the AMD Versal NET DDR controller")
Signed-off-by: Shubhrajyoti Datta <shubhrajyoti.datta@xxxxxxx>
---

Changes in v2:
Patch addition for the rpmsg drvdata race

drivers/edac/versalnet_edac.c | 6 ++++++
1 file changed, 6 insertions(+)

diff --git a/drivers/edac/versalnet_edac.c b/drivers/edac/versalnet_edac.c
index 046293d54d40..1a327a2dbda7 100644
--- a/drivers/edac/versalnet_edac.c
+++ b/drivers/edac/versalnet_edac.c
@@ -639,6 +639,12 @@ static int rpmsg_cb(struct rpmsg_device *rpdev, void *data,
const char *err_str;
u32 *adec_data;

+ if (!mc_priv) {
+ dev_warn_ratelimited(&rpdev->dev,
+ "Dropping rpmsg message, driver not ready or being removed\n");
+ return 0;
+ }
+
if (*(u8 *)data == MCDI_RESPONSE) {
cdx_mcdi_process_cmd(mc_priv->mcdi, (struct cdx_dword *)data, len);
return 0;
--
2.44.4