Linux kernel 4.1.9 completely lockups the machine

atelszewski.versades · 10-01-2015, 03:00 AM

Hi,

I updated my kernel to 4.1.9 and I experience total lock-ups.

As always, my first suspicion is Intel graphics, so I disabled RC6 power saving and for now it seems to work the treat. We'll see in a few hours (minutes?).

Some more reports here.

--
Best regards,
Andrzej Telszewski

atelszewski.versades · 10-01-2015, 05:24 AM

Hi,

Well, seems like "Series 4800" failure here.
It's not about RC6, 4.1.9 still locks up.
No ping replies, no keyboard nor mouse response.
Interestingly, I had terminal emulator opened (under X) and the cursor was blinking. I guess it's the beauty of hardware cursor

Reverted back to 4.1.8.

--
Best regards,
Andrzej Telszewski

1337_powerslacker · 10-01-2015, 12:22 PM

Quote:

Originally Posted by atelszewski.versades

Hi,

Well, seems like "Series 4800" failure here.
It's not about RC6, 4.1.9 still locks up.
No ping replies, no keyboard nor mouse response.
Interestingly, I had terminal emulator opened (under X) and the cursor was blinking. I guess it's the beauty of hardware cursor

Reverted back to 4.1.8.

--
Best regards,
Andrzej Telszewski

You're not the only one to experience hard locks. Seems 4.1.9 has introduced a regression which interacts badly with X. I,too, have retreated to an earlier version, and will wait until this bug is fixed.

cwizardone · 10-01-2015, 01:37 PM

I've had some problems with the 4.1.6 kernel when using Flash with Opera or Chromium with the Widevine plugin. In both cases it appears the CPU overheats and the machine locks up.
No problems with Firefox and the pipelight plugins to view Netflix. Regardless, I've gone back to the 3.18.11 kernel.

1337_powerslacker · 10-01-2015, 02:55 PM

Quote:

Originally Posted by cwizardone

I've had some problems with the 4.1.6 kernel when using Flash with Opera or Chromium with the Widevine plugin. In both cases it appears the CPU overheats and the machine locks up.
No problems with Firefox and the pipelight plugins to view Netflix. Regardless, I've gone back to the 3.18.11 kernel.

I'd say that going with any LTS kernel would be a good bet, but personally, I wouldn't go back that far. Nevertheless, each use case differs (and user preference), so whatever works best for you. I've had pretty good mileage with all kernels I've tried on all hardware I've owned; with only one exception (the 14.0 kernel - 3.2.10 on an HP Pavilion notebook - forget the model name at the moment - which resulted in kernel panics upon boot), they've all worked.

lazardo · 10-02-2015, 12:57 AM

4.1.9 on up-to-date 14.1 base, zero issues.

GeForce GT730 + v340.93 nvidia driver, Asus M4A88T-M, Phenom(tm) II X4 B55

1337_powerslacker · 10-02-2015, 09:38 AM

Quote:

Originally Posted by lazardo

4.1.9 on up-to-date 14.1 base, zero issues.

GeForce GT730 + v340.93 nvidia driver, Asus M4A88T-M, Phenom(tm) II X4 B55

Could be specific to some hardware configurations, I suppose. All I know is that I didn't have much luck with it.

gus3 · 10-03-2015, 03:51 PM

The reports started coming in very quickly about the lockups.

http://lwn.net/Articles/658749/

There is a fix, but it didn't make it into 4.1.10.

1337_powerslacker · 10-04-2015, 07:44 AM

Quote:

Originally Posted by gus3

The reports started coming in very quickly about the lockups.

http://lwn.net/Articles/658749/

There is a fix, but it didn't make it into 4.1.10.

Thanks, gus3, for linking to the fix. I'm building 4.1.10 now; we'll see whether the one-liner fixes the issue.

1337_powerslacker · 10-04-2015, 08:30 AM

For those who don't want to chase links & manually copy/paste into a .patch file, I've uploaded the patch so that others interested in the fix can get it easily. Nevertheless, if you want an explanation for the lockup, do click on the link provided in the above post. As always, remove the .txt suffix before applying the patch.

EDIT: The patch appears to have fixed the issue. Success!

Pearlseattle · 10-04-2015, 03:56 PM

Any clue about what caused the lockup?
(read the contents of that link but didn't understand anything)

1337_powerslacker · 10-04-2015, 04:58 PM

Quote:

Originally Posted by Pearlseattle

Any clue about what caused the lockup?
(read the contents of that link but didn't understand anything)

I'm no kernel developer, but I do have some experience in programming, so my interpretation of the developer's comments is that a subroutine was being called under inappropriate test conditions, causing the deadlock condition (lockup). He added another test condition so that the subroutine would be called only when two conditions were met, instead of just one.

davleema · 10-23-2015, 06:32 AM

Quote:

Originally Posted by gus3

The reports started coming in very quickly about the lockups.

http://lwn.net/Articles/658749/

There is a fix, but it didn't make it into 4.1.10.

4.1.11 is out. Can anybody tell me if the fix made it into this one?

Thx,

David

atelszewski.versades · 10-23-2015, 06:42 AM

Hi,

Looking at the source code from 4.1.11:

Code:

static bool reqsk_queue_unlink(struct request_sock_queue *queue,
			       struct request_sock *req)
{
	struct listen_sock *lopt = queue->listen_opt;
	struct request_sock **prev;
	bool found = false;

	spin_lock(&queue->syn_wait_lock);

	for (prev = &lopt->syn_table[req->rsk_hash]; *prev != NULL;
	     prev = &(*prev)->dl_next) {
		if (*prev == req) {
			*prev = req->dl_next;
			found = true;
			break;
		}
	}

	spin_unlock(&queue->syn_wait_lock);
	if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
		reqsk_put(req);
	return found;
}

it seems to be fixed.

I'm running 4.1.11 for 5 hours now without hanging.

--
Best regards,
Andrzej Telszewski

davleema · 10-23-2015, 07:20 AM

Quote:

Originally Posted by atelszewski.versades

Hi,

Looking at the source code from 4.1.11:

Code:

static bool reqsk_queue_unlink(struct request_sock_queue *queue,
			       struct request_sock *req)
{
	struct listen_sock *lopt = queue->listen_opt;
	struct request_sock **prev;
	bool found = false;

	spin_lock(&queue->syn_wait_lock);

	for (prev = &lopt->syn_table[req->rsk_hash]; *prev != NULL;
	     prev = &(*prev)->dl_next) {
		if (*prev == req) {
			*prev = req->dl_next;
			found = true;
			break;
		}
	}

	spin_unlock(&queue->syn_wait_lock);
	if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
		reqsk_put(req);
	return found;
}

it seems to be fixed.

I'm running 4.1.11 for 5 hours now without hanging.

--
Best regards,
Andrzej Telszewski

Thx for the info.

David