New mm and highmem reminder

New mm and highmem reminder

[Christoph Rohland]

Hi Rik and Ingo,

Just wanted to remind you that swapping shm in highmem is still broken
in the latest patches. 

If I return a RETRY in shm_swap_core instead of FAILED for failures of
prepare_highmem_swapout it survives a little bit longer spewing lots
of 'order 0 allocation failed' and then locks up after doing some
swapping. Without this change it hardly swaps at all before lockup.

It may not be introduced by Riks vm but swapping in shm with PAE
worked quite nice until these changes.

Greetings
		Christoph

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: New mm and highmem reminder

[Benjamin C.R. LaHaise]

On 25 Oct 2000, Christoph Rohland wrote:

> Hi Rik and Ingo,
> 
> Just wanted to remind you that swapping shm in highmem is still broken
> in the latest patches. 

When did it break, in pre3?

		-ben

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: New mm and highmem reminder

[Rik van Riel]

On 25 Oct 2000, Christoph Rohland wrote:

> Just wanted to remind you that swapping shm in highmem is still
> broken in the latest patches.
> 
> If I return a RETRY in shm_swap_core instead of FAILED for
> failures of prepare_highmem_swapout it survives a little bit
> longer spewing lots of 'order 0 allocation failed' and then
> locks up after doing some swapping. Without this change it
> hardly swaps at all before lockup.

Could you test if /normal/ swapping works on highmem
machines?

If it does, it is ipc/shm.c which does the wrong thing
when no bounce buffer could be created...

regards,

Rik
--
"What you're running that piece of shit Gnome?!?!"
       -- Miguel de Icaza, UKUUG 2000

http://www.conectiva.com/		http://www.surriel.com/

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: New mm and highmem reminder

[Christoph Rohland]

Hi Rik,

Rik van Riel <riel@conectiva.com.br> writes:

> On 25 Oct 2000, Christoph Rohland wrote:
> 
> Could you test if /normal/ swapping works on highmem
> machines?

I tested this by mmaping named files instead of shm files. The machine
does not lock up and does not swap because the processes are stuck
uninteruptible and ps and vmstat do lock up on them.

I can else work on the machine

[root@ls3016 /root]# cat /proc/meminfo
         total:    used:    free:  shared: buffers:  cached: 
Mem:  4144390144 3957956608 186433536        0  9175040 3708989440 
Swap: 2048053248        0 2048053248 
MemTotal:      8241560 kB 
MemFree:        182064 kB 
MemShared:           0 kB 
Buffers:          8960 kB 
Cached:        7816364 kB 
Active:        5031424 kB 
Inact_dirty:   2793900 kB 
Inact_clean:         0 kB 
Inact_target:    21936 kB 
HighTotal:     7471104 kB 
HighFree:         2036 kB 
LowTotal:       770456 kB 
LowFree:        180028 kB 
SwapTotal:     2000052 kB 
SwapFree:      2000052 kB 


later:

Active:        5029908 kB
Inact_dirty:   2795424 kB
Inact_clean:         0 kB
Inact_target:     6528 kB                                       

Greetings
                Christoph

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: New mm and highmem reminder

[Roger Larsson]

Hi,

Christoph please check with Alt-SysRq-M if you have run out
of memory in a specific zone.

Christoph Rohland wrote:

> MemFree:        182064 kB 
>  - - -
> Inact_dirty:   2793900 kB
> Inact_clean:         0 kB

Rik, notice the imbalance between inactive dirty and
inactive clean...

In this kind of situation when will page_lauder run?

Christoph, can you put a printk in page_launder to
see if it ever runs? (There are a lot of && conditions
to fulfil before kflushd will start)

/RogerL

--
Home page:
  http://www.norran.net/nra02596/
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: New mm and highmem reminder

[Christoph Rohland]

Roger Larsson <roger.larsson@norran.net> writes:

> Christoph please check with Alt-SysRq-M if you have run out
> of memory in a specific zone.

SysRq: Show Memory
Mem-info:
Free pages:      5297816kB (4631308kB HighMem)
( Active: 381692, inactive_dirty: 328041, inactive_clean: 0, free: 1324431 (638 1276 1914) )
1*4kB 3*8kB 3*16kB 3*32kB 4*64kB 2*128kB 2*256kB 0*512kB 0*1024kB 6*2048kB = 13484kB)
9*4kB 4*8kB 0*16kB 1*32kB 1*64kB 0*128kB 0*256kB 1*512kB 1*1024kB 318*2048kB = 652964kB)
1*4kB 0*8kB 0*16kB 1*32kB 0*64kB 1*128kB 1*256kB 0*512kB 0*1024kB 2261*2048kB = 4630948kB)
Swap cache: add 0, delete 0, find 0/0
Free swap:       2000052kB
2162688 pages of RAM
1867776 pages of HIGHMEM
102298 reserved pages
1275688 pages shared
0 pages swap cached
0 pages in page table cache
Buffer memory:     4860kB
    CLEAN: 612 buffers, 2448 kbyte, 1 used (last=74), 0 locked, 0 protected, 0 dirty
   LOCKED: 211501 buffers, 846004 kbyte, 22 used (last=211325), 1364 locked, 0 protected, 0 dirty
    DIRTY: 114274 buffers, 457096 kbyte, 11 used (last=114259), 0 locked, 0 protected, 114274 dirty           

> Christoph, can you put a printk in page_launder to
> see if it ever runs? (There are a lot of && conditions
> to fulfil before kflushd will start)

No, it is not run.

Greetings
                Christoph

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/