2.6.26: x86/kernel/pci_dma.c: gfp |= __GFP_NORETRY ?

[Miquel van Smoorenburg <miquels@cistron.nl>]

I've recently switched some of my boxes from a 32 to a
64 bit kernel. These are usenet server boxes that do
a lot of I/O. They are running 2.6.24 / 2.6.25

Every 15 minutes a cronjob calls a management utility, tw_cli,
 to read the raid status of the 3ware disk arrays. That
often fails with a segmentation violation .. 

tw_cli: page allocation failure. order:0, mode:0x10d0
Pid: 9296, comm: tw_cli Not tainted 2.6.25.4 #2

Call Trace:
 [<ffffffff802604b6>] __alloc_pages+0x336/0x390
 [<ffffffff80210ff4>] dma_alloc_pages+0x24/0xa0
 [<ffffffff80211113>] dma_alloc_coherent+0xa3/0x2e0
 [<ffffffff8804a58f>] :3w_9xxx:twa_chrdev_ioctl+0x11f/0x810
 [<ffffffff802826c0>] chrdev_open+0x0/0x1c0
 [<ffffffff8027d997>] __dentry_open+0x197/0x210
 [<ffffffff8028c4ed>] vfs_ioctl+0x7d/0xa0
 [<ffffffff8028c584>] do_vfs_ioctl+0x74/0x2d0
 [<ffffffff8028c829>] sys_ioctl+0x49/0x80
 [<ffffffff8020b29b>] system_call_after_swapgs+0x7b/0x80

Mem-info:
DMA per-cpu:
CPU    0: hi:    0, btch:   1 usd:   0
CPU    1: hi:    0, btch:   1 usd:   0
CPU    2: hi:    0, btch:   1 usd:   0
CPU    3: hi:    0, btch:   1 usd:   0
DMA32 per-cpu:
CPU    0: hi:  186, btch:  31 usd:  60
CPU    1: hi:  186, btch:  31 usd: 185
CPU    2: hi:  186, btch:  31 usd: 176
CPU    3: hi:  186, btch:  31 usd: 165
Normal per-cpu:
CPU    0: hi:  186, btch:  31 usd: 120
CPU    1: hi:  186, btch:  31 usd: 164
CPU    2: hi:  186, btch:  31 usd: 177
CPU    3: hi:  186, btch:  31 usd: 182
Active:265929 inactive:1657355 dirty:663189 writeback:62890 unstable:0
 free:49079 slab:65923 mapped:1238 pagetables:927 bounce:0
DMA free:12308kB min:184kB low:228kB high:276kB active:0kB inactive:0kB present:11816kB pages_scanned:0 all_unreclaimable? yes
lowmem_reserve[]: 0 3255 8053 8053
DMA32 free:94200kB min:52912kB low:66140kB high:79368kB active:440616kB inactive:2505772kB present:3333792kB pages_scanned:0 all_unreclaimable? no
lowmem_reserve[]: 0 0 4797 4797
Normal free:86792kB min:77968kB low:97460kB high:116952kB active:623100kB inactive:4126872kB present:4912640kB pages_scanned:32 all_unreclaimable? no
lowmem_reserve[]: 0 0 0 0
DMA: 3*4kB 5*8kB 2*16kB 6*32kB 4*64kB 4*128kB 0*256kB 0*512kB 1*1024kB 1*2048kB 2*4096kB = 12308kB
DMA32: 150*4kB 5*8kB 2299*16kB 120*32kB 0*64kB 0*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB 13*4096kB = 94512kB
Normal: 462*4kB 3803*8kB 123*16kB 24*32kB 2*64kB 1*128kB 1*256kB 1*512kB 0*1024kB 0*2048kB 18*4096kB = 109760kB
1653409 total pagecache pages
Swap cache: add 5748, delete 5411, find 4317/4852
Free swap  = 4588488kB
Total swap = 4594580kB
Free swap:       4588488kB
2293760 pages of RAM
249225 reserved pages
1658761 pages shared
337 pages swap cached

(this is easily reproducible by pinning a lot of memory with
 mmap/mlock, say 6 GB on an 8 GB box, while running
 cat /dev/zero > filename, then invoking tw_cli)

Now this appears to happen because dma_alloc_coherent() in
pci-dma_64.c does this:

        /* Don't invoke OOM killer */
        gfp |= __GFP_NORETRY;

However, if you read mm/page_alloc.c you can see that this not only
prevents invoking the OOM killer, it also does what it says:
no retries when allocating memory.

That means that dma_alloc_coherent(..., GFP_KERNEL) can become
unreliable. Bad news.

pci-dma_32 does not do this.

And in 2.6.26-rc1, pci-dma_32.c and pci-dma_64.c were merged,
so now the 32 bit kernel has the same problem.

Does anyone know why this was added on x86_64 ?

If not I think this patch should go into 2.6.26:

diff -ruN linux-2.6.26-rc3.orig/arch/x86/kernel/pci-dma.c linux-2.6.26-rc3/arch/x86/kernel/pci-dma.c
--- linux-2.6.26-rc3.orig/arch/x86/kernel/pci-dma.c	2008-05-18 23:36:41.000000000 +0200
+++ linux-2.6.26-rc3/arch/x86/kernel/pci-dma.c	2008-05-21 13:15:54.000000000 +0200
@@ -397,9 +397,6 @@
 	if (dev->dma_mask == NULL)
 		return NULL;
 
-	/* Don't invoke OOM killer */
-	gfp |= __GFP_NORETRY;
-
 #ifdef CONFIG_X86_64
 	/* Why <=? Even when the mask is smaller than 4GB it is often
 	   larger than 16MB and in this case we have a chance of


Ideas ? Maybe a __GFP_NO_OOMKILLER ? 

Mike.