linux-kernel - Re: [linux-pm] [PATCH] [suspend] pci_raw_set_power_state: replace msleep by udelay in resuming procedure

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: Windows password security audit tool. GUI, reports in PDF.

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

Message-Id: <200907170433.42366.rjw@sisk.pl>
Date:	Fri, 17 Jul 2009 04:33:41 +0200
From:	"Rafael J. Wysocki" <rjw@...k.pl>
To:	wuzhangjin@...il.com
Cc:	Alan Stern <stern@...land.harvard.edu>,
	linux-pm@...ts.linux-foundation.org, len.brown@...el.com,
	yanh@...ote.com, linux-kernel@...r.kernel.org, zhangfx@...ote.com,
	pavel@....cz
Subject: Re: [linux-pm] [PATCH] [suspend] pci_raw_set_power_state: replace msleep by udelay in resuming procedure

On Friday 17 July 2009, Wu Zhangjin wrote:
> Hi,
> 
> On Thu, 2009-07-16 at 13:02 -0400, Alan Stern wrote:
> > On Thu, 16 Jul 2009, Wu Zhangjin wrote:
> > 
> > > we can not call msleep() when resuming from STR/Standby: if the
> > > current_state of the pci device is PCI_D3hot, means we are in the
> > > procedure of resuming, in this procedure, we can not re-schedule,
> > > otherwise, there will be a deadlock.
> > 
> > I don't understand.
> > 
> > First of all, why does current_state == PCI_D3hot mean the system is
> > resuming from sleep?  Isn't it possible that the PCI device is going
> > through a runtime resume?
> > 
> > Secondly, why will scheduling during a resume cause a deadlock?
> 
> Sorry, I'm stupid to make conclusion before describing the problem
> clearly, here is the problem I encountered:
> 
> when I enabled SUSPEND=y in linux-2.6.30.1(with relative patches) on a
> loongson-based machine(yeeloong laptop,loongson is mips-compatiable), I
> tried to suspend it via "echo standby > /sys/power/state", with the
> serial port debugging support, I found it enter into the standby mode
> successfully. and then, tried to wake it up via the keyboard interrupt,
> but it stopped at the "Power_up_devices:" of kernel/power/main.c.
> 
> here is a short path of this procedure:
> 
> suspend_enter:
> 	...
> 	device_power_down
> 	...
> 	arch_suspend_disable_irqs
> 	...
> 	sysdev_suspend
> 	...
> 	suspend_ops->enter  (board-specific part)
> 	...
> 	sysdev_resume
> 	...
> 	arch_suspend_enable_irqs
> 	...
> 	device_power_up	 <-----------------------stop here
> 	...
> 	
> and then I continue to trace it:
> 
> device_power_up:
> 	dpm_power_up:
> 		list_for_each_entry(dev, &dpm_list, power.entry)
> 			<<DEBUG>>
>                 if (dev->power.status > DPM_OFF) {
>                         int error;
> 
>                         dev->power.status = DPM_OFF;
>                         error = resume_device_noirq(dev, state);
>                         if (error)
>                                 pm_dev_err(dev, state, " early", error);
>                 }   
> 
> I tried to add prom_putchar() at <<DEBUG>> to print something, and This
> will make it resume from standby mode successfully. seems,
> prom_putchar() have influenced the power.status, and make some devices
> not enter into the condition statement, and make dpm_power_up return
> directly. (this is very weird, not sure why?)
> 
> so, I removed the prom_putchar() from <<DEBUG>>, and it stopped at
> resume_device_noirq, here is the following tracing path:
> 	
> resume_device_noirq:
>    --> pm_noirq_op
>    --> ops->resume_noirq (dev) <--> pci_pm_resume_noirq:
> 	   --> pci_pm_default_resume_noirq
> 	   --> pci_restore_standard_config
> 	   --> pci_set_power_state
>            --> pci_raw_set_power_state
>            --> msleep    <-----------------------[ stop here]
> 
> msleep:
> 	--> schedule_timeout_uninterruptible
> 	--> schedule_timeout
> 	     --> ... 
> 	     --> __mod_timer
>              --> ...
> 	     --> schedule ---> a new scheduling happen and never return
> 
> and then I tried to trace schedule(), and even added a prom_putchar() to
> the end of the schedule() function, it output something successfully,
> but never return to schedule_timeout(dead? no keyboard response), seems
> very weird! this is reproductive, perhaps I have missed something here.
>  
> so, to avoid this 'weird' situation, I think it's better not to
> re-schedule in the resuming procedure from standby. and here, I can not
> find another condition to judge the resuming procedure from standby, so,
> I use "current_state == PCI_D3hot"(so, my pre-expression is really wrong
> for it maybe not resume from standby as you indicated). and is there
> another condition to judge we are resuming from standby? perhaps this is
> better:
> 
>    ((current_state == PCI_D3hot) && (state == PCI_D0))
> 
> but seems this also can not indicate we are resuming from standby.

'standby' is a system sleep state, while 'current_state' and 'state' refer to
device power states.

Anyway, the fact that schedule() did not return in your testing setup indicates
that there were no timer interrupts delivered to the CPU during resume, which
should not have happened.

Perhaps it's necessary to annotate your timer interrupts appropriately so that
they are not disabled during suspend?

Best,
Rafael
--
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/