Re: [PATCH v2 07/12] EDAC/versalnet: Initialize MCDI before RPMsg registration

From: Pandey, Radhey Shyam

Date: Sat Nov 21 2026 - 10:33:52 EST


On 11/18/2026 7:50 PM, Shubhrajyoti Datta wrote:
setup_mcdi() currently allocates the MCDI instance, assigns the RPMsg
endpoint, and retrieves DDR configuration data. The DDR configuration
path invokes cdx_mcdi_rpc(), which requires a functional RPMsg endpoint.

Split setup_mcdi() so that only MCDI allocation and initialization are
performed before RPMsg registration. Update the error paths to match
the new initialization order, ensuring resources are released in
reverse order of acquisition.


All three from the earlier review. setup_mcdi() now precedes register_rpmsg_driver(), so rpmsg_probe() can't deref a NULL priv->mcdi. The get_ddr_config() loop moved below the completion wait, so the endpoint is live before first use. err_init unregisters the rpmsg driver before cdx_mcdi_finish()/kfree(), so rpmsg_remove() can't write through freed memory.

I suggest squashing this into the previous patch so that the series
does not introduce a bug that is fixed by a subsequent patch.

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

Changes in v2:
Update the labels
this fixes the sashiko comment
"register_rpmsg_driver() exposes the callback endpoint before setup_mcdi()
initializes the mcdi pointer. Could this lead to a NULL pointer dereference
if a response arrives early?"

drivers/edac/versalnet_edac.c | 27 +++++++++++++--------------
1 file changed, 13 insertions(+), 14 deletions(-)

diff --git a/drivers/edac/versalnet_edac.c b/drivers/edac/versalnet_edac.c
index 90918293f76d..046293d54d40 100644
--- a/drivers/edac/versalnet_edac.c
+++ b/drivers/edac/versalnet_edac.c
@@ -605,7 +605,7 @@ static void get_ddr_config(u32 index, u32 *buffer, struct cdx_mcdi *amd_mcdi)
static int setup_mcdi(struct mc_priv *mc_priv)
{
struct cdx_mcdi *amd_mcdi;
- int ret, i;
+ int ret;
amd_mcdi = kzalloc_obj(*amd_mcdi);
if (!amd_mcdi)
@@ -618,12 +618,8 @@ static int setup_mcdi(struct mc_priv *mc_priv)
return ret;
}
- amd_mcdi->ept = mc_priv->ept;
mc_priv->mcdi = amd_mcdi;
- for (i = 0; i < NUM_CONTROLLERS; i++)
- get_ddr_config(i, &mc_priv->adec[ADEC_NUM * i], amd_mcdi);
-
return 0;
}
@@ -978,7 +974,7 @@ static int mc_probe(struct platform_device *pdev)
{
struct mc_priv *priv;
struct rproc *rp;
- int rc;
+ int rc, i;
struct device_node *r5_core_node __free(device_node) =
of_parse_phandle(pdev->dev.of_node, "amd,rproc", 0);
@@ -1008,16 +1004,16 @@ static int mc_probe(struct platform_device *pdev)
init_completion(&priv->rpmsg_ready);
amd_rpmsg_id_table[0].driver_data = (kernel_ulong_t)priv;
+ rc = setup_mcdi(priv);
+ if (rc)
+ goto err_alloc;
+
rc = register_rpmsg_driver(&amd_rpmsg_driver);
if (rc) {
edac_printk(KERN_ERR, EDAC_MC, "Failed to register RPMsg driver: %d\n", rc);
- goto err_alloc;
+ goto err_reg;
}
- rc = setup_mcdi(priv);
- if (rc)
- goto err_unreg;
-
if (!wait_for_completion_timeout(&priv->rpmsg_ready, MCDI_RPC_TIMEOUT)) {
dev_err(&pdev->dev, "Timed out waiting for RPMsg endpoint\n");
rc = -ENODEV;
@@ -1026,6 +1022,9 @@ static int mc_probe(struct platform_device *pdev)
priv->mcdi->r5_rproc = rp;
+ for (i = 0; i < NUM_CONTROLLERS; i++)
+ get_ddr_config(i, &priv->adec[ADEC_NUM * i], priv->mcdi);
+
rc = init_versalnet(priv, pdev);
if (rc)
goto err_init;
@@ -1040,12 +1039,12 @@ static int mc_probe(struct platform_device *pdev)
return 0;
err_init:
+ unregister_rpmsg_driver(&amd_rpmsg_driver);
+
+err_reg:
cdx_mcdi_finish(priv->mcdi);
kfree(priv->mcdi);
-err_unreg:
- unregister_rpmsg_driver(&amd_rpmsg_driver);
-
err_alloc:
rproc_shutdown(rp);