Re: [linux-pm] Re: swsusp: Not enough free pages

Re: [linux-pm] Re: swsusp: Not enough free pages

[Yu, Luming]

> > > So far, yes. I just tried 2 times.
> >
> > always. (I check that swap dev is on)
> >
> > Sometimes, my ia32 laptop free 0 pages too.
> > I think we should always free some pages
> > from various caches.
>
> Try this hack... it is basically mm problem I don't know how to fix,
> but this seems to help.
> 								Pavel

Thanks Pavel, this hack works.
..
Freeing memory...  ^Hdone (0 pages freed)
Freeing memory...  ^H-^Hdone (4636 pages freed)
Freeing memory...  ^Hdone (0 pages freed)
Freeing memory...  ^H-^Hdone (914 pages freed)
Freeing memory...  ^Hdone (0 pages freed)
Freezing CPUs (at 0)...ok

Any mm guru know how to fix this?

>
> Index: kernel/power/disk.c
> ===================================================================
> --- 805a02ec2bcff3671d7b1e701bd1981ad2fa196c/kernel/power/disk.c 
> (mode:100644) +++
> ecd8559cc08319bb16a42aac06cf7d664157643a/kernel/power/disk.c  (mode:100644)
> @@ -88,23 +92,25 @@
>
>  static void free_some_memory(void)
>  {
> -	unsigned int i = 0;
> -	unsigned int tmp;
> -	unsigned long pages = 0;
> -	char *p = "-\\|/";
> -
> -	printk("Freeing memory...  ");
> -	while ((tmp = shrink_all_memory(10000))) {
> -		pages += tmp;
> -		printk("\b%c", p[i]);
> -		i++;
> -		if (i > 3)
> -			i = 0;
> +	int i;
> +	for (i=0; i<5; i++) {
> +		int i = 0, tmp;
> +		long pages = 0;
> +		char *p = "-\\|/";
> +
> +		printk("Freeing memory...  ");
> +		while ((tmp = shrink_all_memory(10000))) {
> +			pages += tmp;
> +			printk("\b%c", p[i]);
> +			i++;
> +			if (i > 3)
> +				i = 0;
> +		}
> +		printk("\bdone (%li pages freed)\n", pages);
> +		msleep_interruptible(200);
>  	}
> -	printk("\bdone (%li pages freed)\n", pages);
>  }
>
> -
>  static inline void platform_finish(void)
>  {
>  	if (pm_disk_mode == PM_DISK_PLATFORM) {

Re: [linux-pm] Re: swsusp: Not enough free pages

[Pavel Machek]

Hi!

> > > > So far, yes. I just tried 2 times.
> > >
> > > always. (I check that swap dev is on)
> > >
> > > Sometimes, my ia32 laptop free 0 pages too.
> > > I think we should always free some pages
> > > from various caches.
> >
> > Try this hack... it is basically mm problem I don't know how to fix,
> > but this seems to help.
> > 								Pavel
> 
> Thanks Pavel, this hack works.
> ..
> Freeing memory...  ^Hdone (0 pages freed)
> Freeing memory...  ^H-^Hdone (4636 pages freed)
> Freeing memory...  ^Hdone (0 pages freed)
> Freeing memory...  ^H-^Hdone (914 pages freed)
> Freeing memory...  ^Hdone (0 pages freed)
> Freezing CPUs (at 0)...ok
> 
> Any mm guru know how to fix this?

Andrew, can you help? It seems free_some_memory does not really free
all reclaimable memory in recent kernels. In fact, it likes to free
nothing on first invocations....

Pausing and trying few times helps, but is *very* ugly.
								Pavel

> > Index: kernel/power/disk.c
> > ===================================================================
> > --- 805a02ec2bcff3671d7b1e701bd1981ad2fa196c/kernel/power/disk.c 
> > (mode:100644) +++
> > ecd8559cc08319bb16a42aac06cf7d664157643a/kernel/power/disk.c  (mode:100644)
> > @@ -88,23 +92,25 @@
> >
> >  static void free_some_memory(void)
> >  {
> > -	unsigned int i = 0;
> > -	unsigned int tmp;
> > -	unsigned long pages = 0;
> > -	char *p = "-\\|/";
> > -
> > -	printk("Freeing memory...  ");
> > -	while ((tmp = shrink_all_memory(10000))) {
> > -		pages += tmp;
> > -		printk("\b%c", p[i]);
> > -		i++;
> > -		if (i > 3)
> > -			i = 0;
> > +	int i;
> > +	for (i=0; i<5; i++) {
> > +		int i = 0, tmp;
> > +		long pages = 0;
> > +		char *p = "-\\|/";
> > +
> > +		printk("Freeing memory...  ");
> > +		while ((tmp = shrink_all_memory(10000))) {
> > +			pages += tmp;
> > +			printk("\b%c", p[i]);
> > +			i++;
> > +			if (i > 3)
> > +				i = 0;
> > +		}
> > +		printk("\bdone (%li pages freed)\n", pages);
> > +		msleep_interruptible(200);
> >  	}
> > -	printk("\bdone (%li pages freed)\n", pages);
> >  }
> >
> > -
> >  static inline void platform_finish(void)
> >  {
> >  	if (pm_disk_mode == PM_DISK_PLATFORM) {
> 

> _______________________________________________
> linux-pm mailing list
> linux-pm@lists.osdl.org
> http://lists.osdl.org/mailman/listinfo/linux-pm


-- 

Re: [linux-pm] Re: swsusp: Not enough free pages

[Rafael J. Wysocki]

Hi,

On Monday, 6 of June 2005 12:39, Pavel Machek wrote:
> Hi!
> 
> > > > > So far, yes. I just tried 2 times.
> > > >
> > > > always. (I check that swap dev is on)
> > > >
> > > > Sometimes, my ia32 laptop free 0 pages too.
> > > > I think we should always free some pages
> > > > from various caches.
> > >
> > > Try this hack... it is basically mm problem I don't know how to fix,
> > > but this seems to help.
> > > 								Pavel
> > 
> > Thanks Pavel, this hack works.
> > ..
> > Freeing memory...  ^Hdone (0 pages freed)
> > Freeing memory...  ^H-^Hdone (4636 pages freed)
> > Freeing memory...  ^Hdone (0 pages freed)
> > Freeing memory...  ^H-^Hdone (914 pages freed)
> > Freeing memory...  ^Hdone (0 pages freed)
> > Freezing CPUs (at 0)...ok
> > 
> > Any mm guru know how to fix this?
> 
> Andrew, can you help? It seems free_some_memory does not really free
> all reclaimable memory in recent kernels. In fact, it likes to free
> nothing on first invocations....

Actually, on (my) x86-64 it seems to work.  It frees even more memory than
I'd like it to (there's 80-90% of RAM free after it's finished). ;-)

If I had to guess, I'd say the problem is related to PAGE_SIZE != 4096.

Greets,
Rafael


> Pausing and trying few times helps, but is *very* ugly.
> 								Pavel
> 
> > > Index: kernel/power/disk.c
> > > ===================================================================
> > > --- 805a02ec2bcff3671d7b1e701bd1981ad2fa196c/kernel/power/disk.c 
> > > (mode:100644) +++
> > > ecd8559cc08319bb16a42aac06cf7d664157643a/kernel/power/disk.c  (mode:100644)
> > > @@ -88,23 +92,25 @@
> > >
> > >  static void free_some_memory(void)
> > >  {
> > > -	unsigned int i = 0;
> > > -	unsigned int tmp;
> > > -	unsigned long pages = 0;
> > > -	char *p = "-\\|/";
> > > -
> > > -	printk("Freeing memory...  ");
> > > -	while ((tmp = shrink_all_memory(10000))) {
> > > -		pages += tmp;
> > > -		printk("\b%c", p[i]);
> > > -		i++;
> > > -		if (i > 3)
> > > -			i = 0;
> > > +	int i;
> > > +	for (i=0; i<5; i++) {
> > > +		int i = 0, tmp;
> > > +		long pages = 0;
> > > +		char *p = "-\\|/";
> > > +
> > > +		printk("Freeing memory...  ");
> > > +		while ((tmp = shrink_all_memory(10000))) {
> > > +			pages += tmp;
> > > +			printk("\b%c", p[i]);
> > > +			i++;
> > > +			if (i > 3)
> > > +				i = 0;
> > > +		}
> > > +		printk("\bdone (%li pages freed)\n", pages);
> > > +		msleep_interruptible(200);
> > >  	}
> > > -	printk("\bdone (%li pages freed)\n", pages);
> > >  }
> > >
> > > -
> > >  static inline void platform_finish(void)
> > >  {
> > >  	if (pm_disk_mode == PM_DISK_PLATFORM) {
> > 
> 
> > _______________________________________________
> > linux-pm mailing list
> > linux-pm@lists.osdl.org
> > http://lists.osdl.org/mailman/listinfo/linux-pm
> 
> 
> -- 
> 

-- 
- Would you tell me, please, which way I ought to go from here?
- That depends a good deal on where you want to get to.
		-- Lewis Carroll "Alice's Adventures in Wonderland"

Re: [linux-pm] Re: swsusp: Not enough free pages

[Pavel Machek]

Hi!

> > > > > > So far, yes. I just tried 2 times.
> > > > >
> > > > > always. (I check that swap dev is on)
> > > > >
> > > > > Sometimes, my ia32 laptop free 0 pages too.
> > > > > I think we should always free some pages
> > > > > from various caches.
> > > >
> > > > Try this hack... it is basically mm problem I don't know how to fix,
> > > > but this seems to help.
> > > > 								Pavel
> > > 
> > > Thanks Pavel, this hack works.
> > > ..
> > > Freeing memory...  ^Hdone (0 pages freed)
> > > Freeing memory...  ^H-^Hdone (4636 pages freed)
> > > Freeing memory...  ^Hdone (0 pages freed)
> > > Freeing memory...  ^H-^Hdone (914 pages freed)
> > > Freeing memory...  ^Hdone (0 pages freed)
> > > Freezing CPUs (at 0)...ok
> > > 
> > > Any mm guru know how to fix this?
> > 
> > Andrew, can you help? It seems free_some_memory does not really free
> > all reclaimable memory in recent kernels. In fact, it likes to free
> > nothing on first invocations....
> 
> Actually, on (my) x86-64 it seems to work.  It frees even more memory than
> I'd like it to (there's 80-90% of RAM free after it's finished). ;-)
> 
> If I had to guess, I'd say the problem is related to PAGE_SIZE !=
>4096.

No, I see it on i386, too. Try patch below; if it frees some after
first pass, you have that problem, too.

								Pavel

> > > > Index: kernel/power/disk.c
> > > > ===================================================================
> > > > --- 805a02ec2bcff3671d7b1e701bd1981ad2fa196c/kernel/power/disk.c 
> > > > (mode:100644) +++
> > > > ecd8559cc08319bb16a42aac06cf7d664157643a/kernel/power/disk.c  (mode:100644)
> > > > @@ -88,23 +92,25 @@
> > > >
> > > >  static void free_some_memory(void)
> > > >  {
> > > > -	unsigned int i = 0;
> > > > -	unsigned int tmp;
> > > > -	unsigned long pages = 0;
> > > > -	char *p = "-\\|/";
> > > > -
> > > > -	printk("Freeing memory...  ");
> > > > -	while ((tmp = shrink_all_memory(10000))) {
> > > > -		pages += tmp;
> > > > -		printk("\b%c", p[i]);
> > > > -		i++;
> > > > -		if (i > 3)
> > > > -			i = 0;
> > > > +	int i;
> > > > +	for (i=0; i<5; i++) {
> > > > +		int i = 0, tmp;
> > > > +		long pages = 0;
> > > > +		char *p = "-\\|/";
> > > > +
> > > > +		printk("Freeing memory...  ");
> > > > +		while ((tmp = shrink_all_memory(10000))) {
> > > > +			pages += tmp;
> > > > +			printk("\b%c", p[i]);
> > > > +			i++;
> > > > +			if (i > 3)
> > > > +				i = 0;
> > > > +		}
> > > > +		printk("\bdone (%li pages freed)\n", pages);
> > > > +		msleep_interruptible(200);
> > > >  	}
> > > > -	printk("\bdone (%li pages freed)\n", pages);
> > > >  }
> > > >
> > > > -
> > > >  static inline void platform_finish(void)
> > > >  {
> > > >  	if (pm_disk_mode == PM_DISK_PLATFORM) {
> > > 
> > 
> > > _______________________________________________
> > > linux-pm mailing list
> > > linux-pm@lists.osdl.org
> > > http://lists.osdl.org/mailman/listinfo/linux-pm
> > 
> > 
> > -- 
> > 
> 

-- 

Re: [linux-pm] Re: swsusp: Not enough free pages

[Rafael J. Wysocki]

Hi,

On Monday, 6 of June 2005 19:14, Pavel Machek wrote:
> Hi!
> 
> > > > > > > So far, yes. I just tried 2 times.
> > > > > >
> > > > > > always. (I check that swap dev is on)
> > > > > >
> > > > > > Sometimes, my ia32 laptop free 0 pages too.
> > > > > > I think we should always free some pages
> > > > > > from various caches.
> > > > >
> > > > > Try this hack... it is basically mm problem I don't know how to fix,
> > > > > but this seems to help.
> > > > > 								Pavel
> > > > 
> > > > Thanks Pavel, this hack works.
> > > > ..
> > > > Freeing memory...  ^Hdone (0 pages freed)
> > > > Freeing memory...  ^H-^Hdone (4636 pages freed)
> > > > Freeing memory...  ^Hdone (0 pages freed)
> > > > Freeing memory...  ^H-^Hdone (914 pages freed)
> > > > Freeing memory...  ^Hdone (0 pages freed)
> > > > Freezing CPUs (at 0)...ok
> > > > 
> > > > Any mm guru know how to fix this?
> > > 
> > > Andrew, can you help? It seems free_some_memory does not really free
> > > all reclaimable memory in recent kernels. In fact, it likes to free
> > > nothing on first invocations....
> > 
> > Actually, on (my) x86-64 it seems to work.  It frees even more memory than
> > I'd like it to (there's 80-90% of RAM free after it's finished). ;-)
> > 
> > If I had to guess, I'd say the problem is related to PAGE_SIZE !=
> >4096.
> 
> No, I see it on i386, too. Try patch below; if it frees some after
> first pass, you have that problem, too.

I've run it once and the result is this:

Freeing memory... done (75876 pages freed)
Freeing memory... done (1536 pages freed)
Freeing memory... done (0 pages freed)
Freeing memory... done (1792 pages freed)
Freeing memory... done (0 pages freed)

It does free some pages after the first pass, but this is only a small fraction
of all pages freed.  I wouldn't call it a bad result ...

Greets,
Rafael


-- 
- Would you tell me, please, which way I ought to go from here?
- That depends a good deal on where you want to get to.
		-- Lewis Carroll "Alice's Adventures in Wonderland"

Re: [linux-pm] Re: swsusp: Not enough free pages

[Pavel Machek]

Hi!

> > No, I see it on i386, too. Try patch below; if it frees some after
> > first pass, you have that problem, too.
> 
> I've run it once and the result is this:
> 
> Freeing memory... done (75876 pages freed)
> Freeing memory... done (1536 pages freed)
> Freeing memory... done (0 pages freed)
> Freeing memory... done (1792 pages freed)
> Freeing memory... done (0 pages freed)
> 
> It does free some pages after the first pass, but this is only a small fraction
> of all pages freed.  I wouldn't call it a bad result ...

Well, it still did not free all memory it should have freed, and you
were lucky. Apparently for some people it does not that well (and that
includes me, I see 0 in first pass quite often).
								Pavel

Re: [linux-pm] Re: swsusp: Not enough free pages

[Rafael J. Wysocki]

Hi,

On Monday, 6 of June 2005 23:58, Pavel Machek wrote:
> Hi!
> 
> > > No, I see it on i386, too. Try patch below; if it frees some after
> > > first pass, you have that problem, too.
> > 
> > I've run it once and the result is this:
> > 
> > Freeing memory... done (75876 pages freed)
> > Freeing memory... done (1536 pages freed)
> > Freeing memory... done (0 pages freed)
> > Freeing memory... done (1792 pages freed)
> > Freeing memory... done (0 pages freed)
> > 
> > It does free some pages after the first pass, but this is only a small fraction
> > of all pages freed.  I wouldn't call it a bad result ...
> 
> Well, it still did not free all memory it should have freed, and you
> were lucky.

This is a reproducible behavior.  Here goes the result for another suspend:

Freeing memory... done (136611 pages freed)
Freeing memory... done (200 pages freed)
Freeing memory... done (128 pages freed)
Freeing memory... done (0 pages freed)
Freeing memory... done (2353 pages freed)

and it is always like that.  It usually frees more than 100000 pages
in the first pass and about 5% more in the next passes together.

> Apparently for some people it does not that well (and that 
> includes me, I see 0 in first pass quite often).

On 2.6.12-rc3+ I have never seen 0 in the first pass.  In fact, with X running
I have never seen less than 60000. :-)

Perhaps there's a bug that does not hit x86-64 for some reason.  I'll try to
run it on my second box later today and see what happens.

Anyway, I think we need to collect some statistics.

Greets,
Rafael


-- 
- Would you tell me, please, which way I ought to go from here?
- That depends a good deal on where you want to get to.
		-- Lewis Carroll "Alice's Adventures in Wonderland"

Re: [linux-pm] Re: swsusp: Not enough free pages

[Rafael J. Wysocki]

Hi,

On Tuesday, 7 of June 2005 12:39, Rafael J. Wysocki wrote: 
> On Monday, 6 of June 2005 23:58, Pavel Machek wrote:
> > Hi!
> > 
> > > > No, I see it on i386, too. Try patch below; if it frees some after
> > > > first pass, you have that problem, too.
> > > 
> > > I've run it once and the result is this:
> > > 
> > > Freeing memory... done (75876 pages freed)
> > > Freeing memory... done (1536 pages freed)
> > > Freeing memory... done (0 pages freed)
> > > Freeing memory... done (1792 pages freed)
> > > Freeing memory... done (0 pages freed)
> > > 
> > > It does free some pages after the first pass, but this is only a small fraction
> > > of all pages freed.  I wouldn't call it a bad result ...
> > 
> > Well, it still did not free all memory it should have freed, and you
> > were lucky.
> 
> This is a reproducible behavior.  Here goes the result for another suspend:
> 
> Freeing memory... done (136611 pages freed)
> Freeing memory... done (200 pages freed)
> Freeing memory... done (128 pages freed)
> Freeing memory... done (0 pages freed)
> Freeing memory... done (2353 pages freed)
> 
> and it is always like that.  It usually frees more than 100000 pages
> in the first pass and about 5% more in the next passes together.
> 
> > Apparently for some people it does not that well (and that 
> > includes me, I see 0 in first pass quite often).
> 
> On 2.6.12-rc3+ I have never seen 0 in the first pass.  In fact, with X running
> I have never seen less than 60000. :-)
> 
> Perhaps there's a bug that does not hit x86-64 for some reason.  I'll try to
> run it on my second box later today and see what happens.

This is the worst result from the second box:

Freeing memory...  done (54641 pages freed)
Freeing memory...  done (0 pages freed)
Freeing memory...  done (5120 pages freed)
Freeing memory...  done (1952 pages freed)
Freeing memory...  done (2304 pages freed)

Still, there are 5x more pages freed in the first pass (80% of RAM was
empty anyway before suspend), and usually it is 10-20x more or so.

AFAICT, on x86-64 shrink_all_memory(10000) is good enough.

Greets,
Rafael


-- 
- Would you tell me, please, which way I ought to go from here?
- That depends a good deal on where you want to get to.
		-- Lewis Carroll "Alice's Adventures in Wonderland"

Re: [linux-pm] Re: swsusp: Not enough free pages

[Pavel Machek]

Hi!

> > > > > No, I see it on i386, too. Try patch below; if it frees some after
> > > > > first pass, you have that problem, too.
> > > > 
> > > > I've run it once and the result is this:
> > > > 
> > > > Freeing memory... done (75876 pages freed)
> > > > Freeing memory... done (1536 pages freed)
> > > > Freeing memory... done (0 pages freed)
> > > > Freeing memory... done (1792 pages freed)
> > > > Freeing memory... done (0 pages freed)
> > > > 
> > > > It does free some pages after the first pass, but this is only a small fraction
> > > > of all pages freed.  I wouldn't call it a bad result ...
> > > 
> > > Well, it still did not free all memory it should have freed, and you
> > > were lucky.
> > 
> > This is a reproducible behavior.  Here goes the result for another suspend:
> > 
> > Freeing memory... done (136611 pages freed)
> > Freeing memory... done (200 pages freed)
> > Freeing memory... done (128 pages freed)
> > Freeing memory... done (0 pages freed)
> > Freeing memory... done (2353 pages freed)
> > 
> > and it is always like that.  It usually frees more than 100000 pages
> > in the first pass and about 5% more in the next passes together.
> > 
> > > Apparently for some people it does not that well (and that 
> > > includes me, I see 0 in first pass quite often).
> > 
> > On 2.6.12-rc3+ I have never seen 0 in the first pass.  In fact, with X running
> > I have never seen less than 60000. :-)
> > 
> > Perhaps there's a bug that does not hit x86-64 for some reason.  I'll try to
> > run it on my second box later today and see what happens.
> 
> This is the worst result from the second box:
> 
> Freeing memory...  done (54641 pages freed)
> Freeing memory...  done (0 pages freed)
> Freeing memory...  done (5120 pages freed)
> Freeing memory...  done (1952 pages freed)
> Freeing memory...  done (2304 pages freed)
> 
> Still, there are 5x more pages freed in the first pass (80% of RAM was
> empty anyway before suspend), and usually it is 10-20x more or so.

I have seen 0 freed on i386 machine with preempt -rc6-mm1, today...
Something is definitely wrong there.
				Pavel
-- 
64 bytes from 195.113.31.123: icmp_seq=28 ttl=51 time=448769.1 ms         

Re: [linux-pm] Re: swsusp: Not enough free pages

[Rafael J. Wysocki]

Hi,

On Wednesday, 8 of June 2005 18:27, Pavel Machek wrote:
> Hi!
> 
]--snip--[
> > 
> > This is the worst result from the second box:
> > 
> > Freeing memory...  done (54641 pages freed)
> > Freeing memory...  done (0 pages freed)
> > Freeing memory...  done (5120 pages freed)
> > Freeing memory...  done (1952 pages freed)
> > Freeing memory...  done (2304 pages freed)
> > 
> > Still, there are 5x more pages freed in the first pass (80% of RAM was
> > empty anyway before suspend), and usually it is 10-20x more or so.
> 
> I have seen 0 freed on i386 machine with preempt -rc6-mm1, today...
> Something is definitely wrong there.

Well, I have compiled the kernel with preempt and retested (on -rc6) but it
doesn't want to get worse. :-)

The problem seems to be arch-dependent or at least configuration-dependent ... 

Hm, how much RAM is there in your box?

Rafael


-- 
- Would you tell me, please, which way I ought to go from here?
- That depends a good deal on where you want to get to.
		-- Lewis Carroll "Alice's Adventures in Wonderland"

Re: [linux-pm] Re: swsusp: Not enough free pages

[Pavel Machek]

Hi!

> > > This is the worst result from the second box:
> > > 
> > > Freeing memory...  done (54641 pages freed)
> > > Freeing memory...  done (0 pages freed)
> > > Freeing memory...  done (5120 pages freed)
> > > Freeing memory...  done (1952 pages freed)
> > > Freeing memory...  done (2304 pages freed)
> > > 
> > > Still, there are 5x more pages freed in the first pass (80% of RAM was
> > > empty anyway before suspend), and usually it is 10-20x more or so.
> > 
> > I have seen 0 freed on i386 machine with preempt -rc6-mm1, today...
> > Something is definitely wrong there.
> 
> Well, I have compiled the kernel with preempt and retested (on -rc6) but it
> doesn't want to get worse. :-)
> 
> The problem seems to be arch-dependent or at least configuration-dependent ... 
> 
> Hm, how much RAM is there in your box?

This was on 2GB box...
							Pavel