public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
* [PATCH v2] x86: Don't recursively acquire rtc_lock
@ 2011-08-30 16:12 Matt Fleming
  2011-08-30 16:17 ` Matt Fleming
  0 siblings, 1 reply; 6+ messages in thread
From: Matt Fleming @ 2011-08-30 16:12 UTC (permalink / raw)
  To: Ingo Molnar, Thomas Gleixner
  Cc: Jeremy Fitzhardinge, H. Peter Anvin, Marcelo Tosatti, johnstultz,
	Matthew Garrett, linux-kernel, Matt Fleming

From: Matt Fleming <matt.fleming@intel.com>

A deadlock was introduced on x86 in commit ef68c8f87ed1 ("x86:
Serialize EFI time accesses on rtc_lock") because efi_get_time() and
friends can be called with rtc_lock already held by
read_persistent_time(), e.g.

timekeeping_init()
    read_persistent_clock()     <-- acquire rtc_lock
        efi_get_time()
            phys_efi_get_time() <-- acquire rtc_lock <DEADLOCK>

To fix this let's push the locking down into the get_wallclock() and
set_wallclock() implementations. Only the clock implementations that
access the x86 RTC directly need to acquire rtc_lock, so it makes
sense to push the locking down into the rtc, vrtc and efi code.

The virtualization implementations don't require rtc_lock to be held
because they provide their own serialization.

Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Acked-by: Jan Beulich <jbeulich@novell.com>
Acked-by: Avi Kivity <avi@redhat.com> [for the virtualization aspect]
---
 arch/x86/kernel/rtc.c         |   23 ++++++++++++-----------
 arch/x86/platform/mrst/vrtc.c |    9 +++++++++
 2 files changed, 21 insertions(+), 11 deletions(-)

diff --git a/arch/x86/kernel/rtc.c b/arch/x86/kernel/rtc.c
index 3f2ad26..ccdbc16 100644
--- a/arch/x86/kernel/rtc.c
+++ b/arch/x86/kernel/rtc.c
@@ -42,8 +42,11 @@ int mach_set_rtc_mmss(unsigned long nowtime)
 {
 	int real_seconds, real_minutes, cmos_minutes;
 	unsigned char save_control, save_freq_select;
+	unsigned long flags;
 	int retval = 0;
 
+	spin_lock_irqsave(&rtc_lock, flags);
+
 	 /* tell the clock it's being set */
 	save_control = CMOS_READ(RTC_CONTROL);
 	CMOS_WRITE((save_control|RTC_SET), RTC_CONTROL);
@@ -93,12 +96,17 @@ int mach_set_rtc_mmss(unsigned long nowtime)
 	CMOS_WRITE(save_control, RTC_CONTROL);
 	CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
 
+	spin_unlock_irqrestore(&rtc_lock, flags);
+
 	return retval;
 }
 
 unsigned long mach_get_cmos_time(void)
 {
 	unsigned int status, year, mon, day, hour, min, sec, century = 0;
+	unsigned long flags;
+
+	spin_lock_irqsave(&rtc_lock, flags);
 
 	/*
 	 * If UIP is clear, then we have >= 244 microseconds before
@@ -125,6 +133,8 @@ unsigned long mach_get_cmos_time(void)
 	status = CMOS_READ(RTC_CONTROL);
 	WARN_ON_ONCE(RTC_ALWAYS_BCD && (status & RTC_DM_BINARY));
 
+	spin_unlock_irqrestore(&rtc_lock, flags);
+
 	if (RTC_ALWAYS_BCD || !(status & RTC_DM_BINARY)) {
 		sec = bcd2bin(sec);
 		min = bcd2bin(min);
@@ -169,24 +179,15 @@ EXPORT_SYMBOL(rtc_cmos_write);
 
 int update_persistent_clock(struct timespec now)
 {
-	unsigned long flags;
-	int retval;
-
-	spin_lock_irqsave(&rtc_lock, flags);
-	retval = x86_platform.set_wallclock(now.tv_sec);
-	spin_unlock_irqrestore(&rtc_lock, flags);
-
-	return retval;
+	return x86_platform.set_wallclock(now.tv_sec);
 }
 
 /* not static: needed by APM */
 void read_persistent_clock(struct timespec *ts)
 {
-	unsigned long retval, flags;
+	unsigned long retval;
 
-	spin_lock_irqsave(&rtc_lock, flags);
 	retval = x86_platform.get_wallclock();
-	spin_unlock_irqrestore(&rtc_lock, flags);
 
 	ts->tv_sec = retval;
 	ts->tv_nsec = 0;
diff --git a/arch/x86/platform/mrst/vrtc.c b/arch/x86/platform/mrst/vrtc.c
index 73d70d6..6d5dbcd 100644
--- a/arch/x86/platform/mrst/vrtc.c
+++ b/arch/x86/platform/mrst/vrtc.c
@@ -58,8 +58,11 @@ EXPORT_SYMBOL_GPL(vrtc_cmos_write);
 unsigned long vrtc_get_time(void)
 {
 	u8 sec, min, hour, mday, mon;
+	unsigned long flags;
 	u32 year;
 
+	spin_lock_irqsave(&rtc_lock, flags);
+
 	while ((vrtc_cmos_read(RTC_FREQ_SELECT) & RTC_UIP))
 		cpu_relax();
 
@@ -70,6 +73,8 @@ unsigned long vrtc_get_time(void)
 	mon = vrtc_cmos_read(RTC_MONTH);
 	year = vrtc_cmos_read(RTC_YEAR);
 
+	spin_unlock_irqrestore(&rtc_lock, flags);
+
 	/* vRTC YEAR reg contains the offset to 1960 */
 	year += 1960;
 
@@ -83,8 +88,10 @@ unsigned long vrtc_get_time(void)
 int vrtc_set_mmss(unsigned long nowtime)
 {
 	int real_sec, real_min;
+	unsigned long flags;
 	int vrtc_min;
 
+	spin_lock_irqsave(&rtc_lock, flags);
 	vrtc_min = vrtc_cmos_read(RTC_MINUTES);
 
 	real_sec = nowtime % 60;
@@ -95,6 +102,8 @@ int vrtc_set_mmss(unsigned long nowtime)
 
 	vrtc_cmos_write(real_sec, RTC_SECONDS);
 	vrtc_cmos_write(real_min, RTC_MINUTES);
+	spin_unlock_irqrestore(&rtc_lock, flags);
+
 	return 0;
 }
 
-- 
1.7.4.4


^ permalink raw reply related	[flat|nested] 6+ messages in thread

* Re: [PATCH v2] x86: Don't recursively acquire rtc_lock
  2011-08-30 16:12 [PATCH v2] x86: Don't recursively acquire rtc_lock Matt Fleming
@ 2011-08-30 16:17 ` Matt Fleming
  2011-09-30  7:26   ` Matt Fleming
  0 siblings, 1 reply; 6+ messages in thread
From: Matt Fleming @ 2011-08-30 16:17 UTC (permalink / raw)
  To: Ingo Molnar
  Cc: Thomas Gleixner, Jeremy Fitzhardinge, H. Peter Anvin,
	Marcelo Tosatti, johnstultz, Matthew Garrett, linux-kernel,
	Jan Beulich, Avi Kivity

(Adding Jan and Avi, apparently git send-email doesn't grok Acked-by's)

On Tue, 2011-08-30 at 17:12 +0100, Matt Fleming wrote:
> From: Matt Fleming <matt.fleming@intel.com>
> 
> A deadlock was introduced on x86 in commit ef68c8f87ed1 ("x86:
> Serialize EFI time accesses on rtc_lock") because efi_get_time() and
> friends can be called with rtc_lock already held by
> read_persistent_time(), e.g.
> 
> timekeeping_init()
>     read_persistent_clock()     <-- acquire rtc_lock
>         efi_get_time()
>             phys_efi_get_time() <-- acquire rtc_lock <DEADLOCK>
> 
> To fix this let's push the locking down into the get_wallclock() and
> set_wallclock() implementations. Only the clock implementations that
> access the x86 RTC directly need to acquire rtc_lock, so it makes
> sense to push the locking down into the rtc, vrtc and efi code.
> 
> The virtualization implementations don't require rtc_lock to be held
> because they provide their own serialization.
> 
> Signed-off-by: Matt Fleming <matt.fleming@intel.com>
> Acked-by: Jan Beulich <jbeulich@novell.com>
> Acked-by: Avi Kivity <avi@redhat.com> [for the virtualization aspect]
> ---
>  arch/x86/kernel/rtc.c         |   23 ++++++++++++-----------
>  arch/x86/platform/mrst/vrtc.c |    9 +++++++++
>  2 files changed, 21 insertions(+), 11 deletions(-)
> 
> diff --git a/arch/x86/kernel/rtc.c b/arch/x86/kernel/rtc.c
> index 3f2ad26..ccdbc16 100644
> --- a/arch/x86/kernel/rtc.c
> +++ b/arch/x86/kernel/rtc.c
> @@ -42,8 +42,11 @@ int mach_set_rtc_mmss(unsigned long nowtime)
>  {
>  	int real_seconds, real_minutes, cmos_minutes;
>  	unsigned char save_control, save_freq_select;
> +	unsigned long flags;
>  	int retval = 0;
>  
> +	spin_lock_irqsave(&rtc_lock, flags);
> +
>  	 /* tell the clock it's being set */
>  	save_control = CMOS_READ(RTC_CONTROL);
>  	CMOS_WRITE((save_control|RTC_SET), RTC_CONTROL);
> @@ -93,12 +96,17 @@ int mach_set_rtc_mmss(unsigned long nowtime)
>  	CMOS_WRITE(save_control, RTC_CONTROL);
>  	CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
>  
> +	spin_unlock_irqrestore(&rtc_lock, flags);
> +
>  	return retval;
>  }
>  
>  unsigned long mach_get_cmos_time(void)
>  {
>  	unsigned int status, year, mon, day, hour, min, sec, century = 0;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&rtc_lock, flags);
>  
>  	/*
>  	 * If UIP is clear, then we have >= 244 microseconds before
> @@ -125,6 +133,8 @@ unsigned long mach_get_cmos_time(void)
>  	status = CMOS_READ(RTC_CONTROL);
>  	WARN_ON_ONCE(RTC_ALWAYS_BCD && (status & RTC_DM_BINARY));
>  
> +	spin_unlock_irqrestore(&rtc_lock, flags);
> +
>  	if (RTC_ALWAYS_BCD || !(status & RTC_DM_BINARY)) {
>  		sec = bcd2bin(sec);
>  		min = bcd2bin(min);
> @@ -169,24 +179,15 @@ EXPORT_SYMBOL(rtc_cmos_write);
>  
>  int update_persistent_clock(struct timespec now)
>  {
> -	unsigned long flags;
> -	int retval;
> -
> -	spin_lock_irqsave(&rtc_lock, flags);
> -	retval = x86_platform.set_wallclock(now.tv_sec);
> -	spin_unlock_irqrestore(&rtc_lock, flags);
> -
> -	return retval;
> +	return x86_platform.set_wallclock(now.tv_sec);
>  }
>  
>  /* not static: needed by APM */
>  void read_persistent_clock(struct timespec *ts)
>  {
> -	unsigned long retval, flags;
> +	unsigned long retval;
>  
> -	spin_lock_irqsave(&rtc_lock, flags);
>  	retval = x86_platform.get_wallclock();
> -	spin_unlock_irqrestore(&rtc_lock, flags);
>  
>  	ts->tv_sec = retval;
>  	ts->tv_nsec = 0;
> diff --git a/arch/x86/platform/mrst/vrtc.c b/arch/x86/platform/mrst/vrtc.c
> index 73d70d6..6d5dbcd 100644
> --- a/arch/x86/platform/mrst/vrtc.c
> +++ b/arch/x86/platform/mrst/vrtc.c
> @@ -58,8 +58,11 @@ EXPORT_SYMBOL_GPL(vrtc_cmos_write);
>  unsigned long vrtc_get_time(void)
>  {
>  	u8 sec, min, hour, mday, mon;
> +	unsigned long flags;
>  	u32 year;
>  
> +	spin_lock_irqsave(&rtc_lock, flags);
> +
>  	while ((vrtc_cmos_read(RTC_FREQ_SELECT) & RTC_UIP))
>  		cpu_relax();
>  
> @@ -70,6 +73,8 @@ unsigned long vrtc_get_time(void)
>  	mon = vrtc_cmos_read(RTC_MONTH);
>  	year = vrtc_cmos_read(RTC_YEAR);
>  
> +	spin_unlock_irqrestore(&rtc_lock, flags);
> +
>  	/* vRTC YEAR reg contains the offset to 1960 */
>  	year += 1960;
>  
> @@ -83,8 +88,10 @@ unsigned long vrtc_get_time(void)
>  int vrtc_set_mmss(unsigned long nowtime)
>  {
>  	int real_sec, real_min;
> +	unsigned long flags;
>  	int vrtc_min;
>  
> +	spin_lock_irqsave(&rtc_lock, flags);
>  	vrtc_min = vrtc_cmos_read(RTC_MINUTES);
>  
>  	real_sec = nowtime % 60;
> @@ -95,6 +102,8 @@ int vrtc_set_mmss(unsigned long nowtime)
>  
>  	vrtc_cmos_write(real_sec, RTC_SECONDS);
>  	vrtc_cmos_write(real_min, RTC_MINUTES);
> +	spin_unlock_irqrestore(&rtc_lock, flags);
> +
>  	return 0;
>  }
>  




^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v2] x86: Don't recursively acquire rtc_lock
  2011-08-30 16:17 ` Matt Fleming
@ 2011-09-30  7:26   ` Matt Fleming
  2011-10-10  6:45     ` Ingo Molnar
  0 siblings, 1 reply; 6+ messages in thread
From: Matt Fleming @ 2011-09-30  7:26 UTC (permalink / raw)
  To: Ingo Molnar, Thomas Gleixner
  Cc: Jeremy Fitzhardinge, H. Peter Anvin, Marcelo Tosatti, johnstultz,
	Matthew Garrett, linux-kernel, Jan Beulich, Avi Kivity,
	Zhang, Rui

On Tue, 2011-08-30 at 17:17 +0100, Matt Fleming wrote:
> (Adding Jan and Avi, apparently git send-email doesn't grok Acked-by's)
> 
> On Tue, 2011-08-30 at 17:12 +0100, Matt Fleming wrote:
> > From: Matt Fleming <matt.fleming@intel.com>
> > 
> > A deadlock was introduced on x86 in commit ef68c8f87ed1 ("x86:
> > Serialize EFI time accesses on rtc_lock") because efi_get_time() and
> > friends can be called with rtc_lock already held by
> > read_persistent_time(), e.g.
> > 
> > timekeeping_init()
> >     read_persistent_clock()     <-- acquire rtc_lock
> >         efi_get_time()
> >             phys_efi_get_time() <-- acquire rtc_lock <DEADLOCK>
> > 
> > To fix this let's push the locking down into the get_wallclock() and
> > set_wallclock() implementations. Only the clock implementations that
> > access the x86 RTC directly need to acquire rtc_lock, so it makes
> > sense to push the locking down into the rtc, vrtc and efi code.
> > 
> > The virtualization implementations don't require rtc_lock to be held
> > because they provide their own serialization.
> > 
> > Signed-off-by: Matt Fleming <matt.fleming@intel.com>
> > Acked-by: Jan Beulich <jbeulich@novell.com>
> > Acked-by: Avi Kivity <avi@redhat.com> [for the virtualization aspect]

Ping? It's -rc8 and 32-bit EFI machines still don't boot.

-- 
Matt Fleming, Intel Open Source Technology Center


^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v2] x86: Don't recursively acquire rtc_lock
  2011-09-30  7:26   ` Matt Fleming
@ 2011-10-10  6:45     ` Ingo Molnar
  2011-10-10  7:15       ` Jan Beulich
  0 siblings, 1 reply; 6+ messages in thread
From: Ingo Molnar @ 2011-10-10  6:45 UTC (permalink / raw)
  To: Matt Fleming
  Cc: Thomas Gleixner, Jeremy Fitzhardinge, H. Peter Anvin,
	Marcelo Tosatti, johnstultz, Matthew Garrett, linux-kernel,
	Jan Beulich, Avi Kivity, Zhang, Rui


* Matt Fleming <matt@console-pimps.org> wrote:

> On Tue, 2011-08-30 at 17:17 +0100, Matt Fleming wrote:
> > (Adding Jan and Avi, apparently git send-email doesn't grok Acked-by's)
> > 
> > On Tue, 2011-08-30 at 17:12 +0100, Matt Fleming wrote:
> > > From: Matt Fleming <matt.fleming@intel.com>
> > > 
> > > A deadlock was introduced on x86 in commit ef68c8f87ed1 ("x86:
> > > Serialize EFI time accesses on rtc_lock") because efi_get_time() and
> > > friends can be called with rtc_lock already held by
> > > read_persistent_time(), e.g.
> > > 
> > > timekeeping_init()
> > >     read_persistent_clock()     <-- acquire rtc_lock
> > >         efi_get_time()
> > >             phys_efi_get_time() <-- acquire rtc_lock <DEADLOCK>
> > > 
> > > To fix this let's push the locking down into the get_wallclock() and
> > > set_wallclock() implementations. Only the clock implementations that
> > > access the x86 RTC directly need to acquire rtc_lock, so it makes
> > > sense to push the locking down into the rtc, vrtc and efi code.
> > > 
> > > The virtualization implementations don't require rtc_lock to be held
> > > because they provide their own serialization.
> > > 
> > > Signed-off-by: Matt Fleming <matt.fleming@intel.com>
> > > Acked-by: Jan Beulich <jbeulich@novell.com>
> > > Acked-by: Avi Kivity <avi@redhat.com> [for the virtualization aspect]
> 
> Ping? It's -rc8 and 32-bit EFI machines still don't boot.

Don't know the status of this - Thomas?

Thanks,

	Ingo

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v2] x86: Don't recursively acquire rtc_lock
  2011-10-10  6:45     ` Ingo Molnar
@ 2011-10-10  7:15       ` Jan Beulich
  2011-10-10  7:54         ` Ingo Molnar
  0 siblings, 1 reply; 6+ messages in thread
From: Jan Beulich @ 2011-10-10  7:15 UTC (permalink / raw)
  To: Ingo Molnar
  Cc: Jeremy Fitzhardinge, Matt Fleming, Rui Zhang, Thomas Gleixner,
	H. Peter Anvin, Avi Kivity, Matthew Garrett, Marcelo Tosatti,
	johnstultz, linux-kernel

>>> On 10.10.11 at 08:45, Ingo Molnar <mingo@elte.hu> wrote:
> * Matt Fleming <matt@console-pimps.org> wrote:
>> Ping? It's -rc8 and 32-bit EFI machines still don't boot.
> 
> Don't know the status of this - Thomas?

This went into 3.1-rc9 (with your signed-off being last).

Jan

> Thanks,
> 
> 	Ingo




^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v2] x86: Don't recursively acquire rtc_lock
  2011-10-10  7:15       ` Jan Beulich
@ 2011-10-10  7:54         ` Ingo Molnar
  0 siblings, 0 replies; 6+ messages in thread
From: Ingo Molnar @ 2011-10-10  7:54 UTC (permalink / raw)
  To: Jan Beulich
  Cc: Jeremy Fitzhardinge, Matt Fleming, Rui Zhang, Thomas Gleixner,
	H. Peter Anvin, Avi Kivity, Matthew Garrett, Marcelo Tosatti,
	johnstultz, linux-kernel


* Jan Beulich <JBeulich@suse.com> wrote:

> >>> On 10.10.11 at 08:45, Ingo Molnar <mingo@elte.hu> wrote:
> > * Matt Fleming <matt@console-pimps.org> wrote:
> >> Ping? It's -rc8 and 32-bit EFI machines still don't boot.
> > 
> > Don't know the status of this - Thomas?
> 
> This went into 3.1-rc9 (with your signed-off being last).

Heh, I typoed the search for the commit and thought it's still 
unmerged and there's still some problem left with this patch.

Thanks,

	Ingo

^ permalink raw reply	[flat|nested] 6+ messages in thread

end of thread, other threads:[~2011-10-10  7:56 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2011-08-30 16:12 [PATCH v2] x86: Don't recursively acquire rtc_lock Matt Fleming
2011-08-30 16:17 ` Matt Fleming
2011-09-30  7:26   ` Matt Fleming
2011-10-10  6:45     ` Ingo Molnar
2011-10-10  7:15       ` Jan Beulich
2011-10-10  7:54         ` Ingo Molnar

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox