From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758621Ab0EBUGp (ORCPT ); Sun, 2 May 2010 16:06:45 -0400 Received: from legolas.restena.lu ([158.64.1.34]:58297 "EHLO legolas.restena.lu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758543Ab0EBUGm convert rfc822-to-8bit (ORCPT ); Sun, 2 May 2010 16:06:42 -0400 Date: Sun, 2 May 2010 22:06:28 +0200 From: Bruno =?UTF-8?B?UHLDqW1vbnQ=?= To: Alan Stern Cc: dri-devel@lists.freedesktop.org, "Rafael J. Wysocki" , , Subject: Re: s2ram slow (radeon) / failing (usb) Message-ID: <20100502220628.0728d8d4@neptune.home> In-Reply-To: References: <20100502155653.68fc3643@neptune.home> X-Mailer: Claws Mail 3.7.5 (GTK+ 2.18.6; i686-pc-linux-gnu) Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8BIT Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Sun, 02 May 2010 Alan Stern wrote: > On Sun, 2 May 2010, Bruno [UTF-8] Prémont wrote: > > On a IEI Kino 690S1 I'm having a hard time to get s2ram running. > > It freezes during device suspend (unless I rmmod everything USB > > related) - usb fails even in pm_test case 'devices'. > > > > When the system is able to suspend it takes an eternity (more than 3 > > minutes to wake-up, the radeon apparently being responsible for quite > > a big share of that slowness. > > > > > > During resume early it looks like every PCI access needs about a second, > > and there are a few cases where during lots of seconds nothing seems to > > happen and the first event following is related to radeon. > > > > The kernel used is todays Linus's tree at commit be1066bbcd443a65df312fdecea7e4959adedb45 > > with Dave's drm-linus and drm-radeon-testing applied on top. > > > > Note, I've not been able to suspend to RAM properly recently (last one > > that worked correctly but resumed without graphics was some-when during > > 2.6.2x, before KMS) > > Since then the system would either fail suspend or resume. > > > > Manual changes I applied in order to find out some context information: > > - add a few debugging printk's to ata/ahci as that was the last entry > > on serial console for freezing suspends (that one succeeded but > > following step never completed, from suspend_prepare that would have > > been USB => unload usb before suspend) > > - strip "if EMBEDED" from CONFIG_SERIAL_8250_PNP and disabled it so serial > > console would continue working as long as possible and output suspend > > progress (resume output happens only very late) > > > > Is there some additional information I could gather in order do help > > improving s2ram on this system? > > - get it to suspend with usb loaded (ohci + ehci) > > - get it to resume a reasonable speed > > There's no way to fix the USB problem without knowing what goes wrong. > Let's see how far you get before the system freezes on a kernel with > CONFIG_USB_DEBUG enabled. Am I missing something? I've enabled CONFIG_USB_DEBUG but don't see any additional module parameter nor anything extra to toggle and I don't get more output than without it. Device suspend (pm_test = device) works well when there is no USB device connected, but with USB keyboard I get the freeze (though the keyboard is still usable, e.g. CAPS key works and I can issue SYSRQ commands). When I issue sysreq-t, I find the following suspicious entry: [ 669.112505] usbhid_resume D ffff88007a085fd8 0 1145 2 0x00000000 [ 669.112505] ffff88007a085e20 0000000000000046 ffff88007a085fd8 ffff88007c536820 [ 669.112505] ffff88007a085fd8 ffff88007a085fd8 00000000000129c0 00000000000129c0 [ 669.112505] ffff88007c536820 ffff88007cf3f040 ffff88007a085fd8 ffff88007a085fd8 [ 669.112505] Call Trace: [ 669.112505] [] refrigerator+0x95/0xf0 [ 669.112505] [] worker_thread+0xc6/0x1e0 [ 669.112505] [] ? autoremove_wake_function+0x0/0x40 [ 669.112505] [] ? worker_thread+0x0/0x1e0 [ 669.112505] [] kthread+0x8e/0xa0 [ 669.112505] [] kernel_thread_helper+0x4/0x10 [ 669.112505] [] ? kthread+0x0/0xa0 [ 669.112505] [] ? kernel_thread_helper+0x0/0x10 Except for that one there are a few async/* tasks waiting. Thanks, Bruno