Skip to contentSkip to conversation

EDAC/device_sysfs: Reject poll_msec value 0

PATCH In review Jad Keskes 2 participants 3 messages Jul 10 – Jul 30 EDAC-core

lore
Share & tools
lore

Jad Keskes

author
#0 lore raw

PATCH v2 1/2 EDAC/device_sysfs: Reject poll_msec value 0

The poll_msec store allows writing 0, which sets the workqueue delay to 0 jiffies. This causes the poll work to spin without any delay, consuming 100% CPU on the polling kworker.

The comment has said "must be at least one millisecond" since the file was first added, but no validation was ever enforced. Add a check for value < 1 and return -EINVAL.

Fixes e27e3dac6517 ("drivers/edac: add edac_device class")
---
 drivers/edac/edac_device_sysfs.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/edac/edac_device_sysfs.c +3 -0
index b1c2717cd023..89adfd3091e0 100644
@@ -96,6 +96,9 @@ static ssize_t edac_device_ctl_poll_msec_store(struct edac_device_ctl_info
 	 * and set a new one.
 	 */
 	value = simple_strtoul(data, NULL, 0);
+	if (value < 1)
+		return -EINVAL;
+
 	edac_device_reset_delay_period(ctl_info, value);
 
 	return count;
Signature · 1 line
2.55.0
author
#1 lore raw

PATCH v2 2/2 EDAC/device: Don't re-arm workqueue during teardown

edac_device_del_device() stops the workqueue then removes the sysfs attributes. However, writing to poll_msec triggers edac_device_reset_delay_period(), which unconditionally re-arms the workqueue. If writes occur during that window, the re-armed work runs after edac_dev has been freed.

Fix this by checking op_state in reset_delay_period(). del_device() already sets it to OP_OFFLINE before tearing down the workqueue, so the new guard causes reset_delay_period() to skip mod_work() when the device is being removed.

---
 drivers/edac/edac_device.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/edac/edac_device.c +4 -0
index cf0d3c2dfc04..50ae7f1e5152 100644
@@ -354,6 +354,10 @@ void edac_device_reset_delay_period(struct edac_device_ctl_info *edac_dev,
 	edac_dev->poll_msec = msec;
 	edac_dev->delay	    = msecs_to_jiffies(msec);
 
+	/* Don't re-arm the workqueue if the device is being torn down */
+	if (edac_dev->op_state == OP_OFFLINE)
+		return;
+
 	/* See comment in edac_device_workq_setup() above */
 	if (edac_dev->poll_msec == DEFAULT_POLL_INTERVAL)
 		edac_mod_work(&edac_dev->work, round_jiffies_relative(edac_dev->delay));
Signature · 1 line
2.55.0
maintainer
#2 lore raw
In reply to Jad Keskes
Quoted 8 lines · Jad Keskes

The poll_msec store allows writing 0, which sets the workqueue delay to 0 jiffies. This causes the poll work to spin without any delay, consuming 100% CPU on the polling kworker.

The comment has said "must be at least one millisecond" since the file was first added, but no validation was ever enforced. Add a check for value < 1 and return -EINVAL.

Quoted 5 lines
Fixes e27e3dac6517 ("drivers/edac: add edac_device class")
---
 drivers/edac/edac_device_sysfs.c | 3 +++
 1 file changed, 3 insertions(+)

Thanks for those, definitely better.

However, Sashiko found more crud:

https://sashiko.dev/#/patchset/20260710113548.46266-1-inasj268%40gmail.com

That thing in edac_device_ctl_poll_msec_store() should not do unsigned longs and then stuff them into a uint. I guess we didn't have the kstrtox-etc helpers back then...

I think you should convert it to kstrtouint() directly and then deal with the value in the proper range.

And then the other reported issues would be good to address too.

Thx.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette