Re: Issue with panic handling and ipmi

[Corey Minyard]

On Thu, Sep 16, 2021 at 04:53:00PM +0200, Anton Lundin wrote:
> Hi.
> 
> I've just done a upgrade of the kernel we're using in a product from
> 4.19 to 5.10 and I noted a issue.
> 
> It started that with that we didn't get panic and oops dumps in our erst
> backed pstore, and when debugging that I noted that the reboot on panic
> timer didn't work either.
> 
> I've bisected it down to 2033f6858970 ("ipmi: Free receive messages when
> in an oops").

Hmm.  Unfortunately removing that will break other things.  Can you try
the following patch?  It's a good idea, in general, to do as little as
possible in the panic path, this should cover a multitude of issues.

Thanks for the report.

-corey

> 
> I tested just reverting that and both dumps to pstore and the panic
> reboot timer started working again.
> 
> 
> //Anton

commit e28aa211190b7d3a1135f051f0c30b0195016489
Author: Corey Minyard <cminyard@xxxxxxxxxx>
Date:   Thu Sep 16 11:36:20 2021 -0500

    ipmi: Disable some operations during a panic

    Don't do kfree or other risky things when oops_in_progress is set.

    Reported-by: Anton Lundin <glance@xxxxxxxxxx>
    Fixes: 2033f6858970 ("ipmi: Free receive messages when > in an oops")
    Signed-off-by: Corey Minyard <cminyard@xxxxxxxxxx>

diff --git a/drivers/char/ipmi/ipmi_msghandler.c b/drivers/char/ipmi/ipmi_msghandler.c
index e96cb5c4f97a..a08f53f208bf 100644
--- a/drivers/char/ipmi/ipmi_msghandler.c
+++ b/drivers/char/ipmi/ipmi_msghandler.c
@@ -4789,7 +4789,9 @@ static atomic_t recv_msg_inuse_count = ATOMIC_INIT(0);
 static void free_smi_msg(struct ipmi_smi_msg *msg)
 {
 	atomic_dec(&smi_msg_inuse_count);
-	kfree(msg);
+	/* Try to keep as much stuff out of the panic path as possible. */
+	if (!oops_in_progress)
+		kfree(msg);
 }

 struct ipmi_smi_msg *ipmi_alloc_smi_msg(void)
@@ -4808,7 +4810,9 @@ EXPORT_SYMBOL(ipmi_alloc_smi_msg);
 static void free_recv_msg(struct ipmi_recv_msg *msg)
 {
 	atomic_dec(&recv_msg_inuse_count);
-	kfree(msg);
+	/* Try to keep as much stuff out of the panic path as possible. */
+	if (!oops_in_progress)
+		kfree(msg);
 }

 static struct ipmi_recv_msg *ipmi_alloc_recv_msg(void)
@@ -4826,7 +4830,7 @@ static struct ipmi_recv_msg *ipmi_alloc_recv_msg(void)

 void ipmi_free_recv_msg(struct ipmi_recv_msg *msg)
 {
-	if (msg->user)
+	if (msg->user && !oops_in_progress)
 		kref_put(&msg->user->refcount, free_user);
 	msg->done(msg);
 }