linux-kernel - Re: Intel graphics CPU usage

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Mon, 16 Aug 2010 00:06:03 -0400
From:	Andy Lutomirski <luto@...ealbox.com>
To:	Linus Torvalds <torvalds@...ux-foundation.org>
CC:	Dave Airlie <airlied@...ux.ie>,
	Chris Wilson <chris@...is-wilson.co.uk>,
	Zhao Yakui <yakui.zhao@...el.com>,
	Jesse Barnes <jbarnes@...tuousgeek.org>,
	Adam Jackson <ajax@...hat.com>,
	DRI mailing list <dri-devel@...ts.freedesktop.org>,
	linux-kernel@...r.kernel.org
Subject: Re: Intel graphics CPU usage - SDVO detect bogosity?

Linus Torvalds wrote:
> I started wondering why 'top' was showing an otherwise idle system as
> having a load average of 0.5+, and worker threads constantly using the
> CPU.
> 
> So I did a system-wide profile, and got the attached output (look at
> it in a really wide terminal).
> 
> There seems to be something _seriously_ wrong with i915 SDVO detect.
> This is on an Apple Mac Mini (hey, your favorite problem child!), and
> apparently it spends 20% of its non-idle CPU time just doing udelay's
> for the i2c SDVO connection detection.

You might be hitting the infamous hotplug storm [1].  The symptoms vary 
by kernel version.

2.6.34 and before: udevadm --monitor shows craploads of events and, as 
long as X is running, X keeps reprobing the outputs which (depending on 
the particular bug) can suck cpu in the i2c code or cause more hotplug 
events.  It also makes X oddly laggy.

2.6.35 and newer: The kernel is smart enough to probe outputs itself 
before telling X, so the events never hit userspace.  But things still 
can get a bit laggy.

Anyone know why merely *reading* /sys/class/drm/whatever/status causes 
the output to get probed?  (I see it in the code, but I have no idea why 
that code's still there after most of the rest of the hotplug code got 
cleaned up in 2.6.36).

Once I find some free time, I plan on trying to at least fix the issue 
that causes this bug for me.  (It's apparently quite nontrivial due to 
silliness in the way dock/undock (!) works on some laptops.)

[1] for example: 
http://www.mail-archive.com/intel-gfx@lists.freedesktop.org/msg00921.html

> 
> That sounds a bit wrong, doesn't it?
> 
> I don't know how recent this is - it might have been going on for some
> time without me noticing. It's the wife's computer, and the same thing
> doesn't seem to happen on my Core i5 desktop
> 
> Any ideas? Any information I can give about the machine?

If I'm right, the outputs of intel_bios_dumper and intel_bios_reader 
could be instructive (both are in intel-gpu-tools).

You could also try intel_reg_write 0x61110 0x0 and see if the problem 
stops (at least until a suspend/resume cycle).  That command turns off 
output hotplug on the card, which has the side effect that the kernel 
will stop acting on bogus interrupts.

--Andy
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/