Re: sata_sil24 memory fragmentation issues

[Robert Hancock]

On 09/02/2010 11:02 AM, Jonathan Haws wrote:
> I am having some issues with the sata_sil24 driver.  It appears that when memory gets fragmented enough, bad things start to happen.  However, this only occurs when I am receiving large amounts of data over the network as well.
>
> Here is my test setup:  I am running an AMCC 405EX processor on their Kilauea development board.  I have a PCIe SATA controller based on the 3531 single port chip (which uses the sata_sil24 driver).  I have a program that simply dumps data out to disk.  When I am running that program, I am also running ping -s 8500<some-ip>.
>
> Here is the output:
>
> 8508 bytes from 172.31.22.21: seq=137 ttl=128 time=1.306 ms
>    CNT:        129 WRIT:   35651584 RATE: 34.00000 MB/s READ:          0 RATE: 0.00000 MB/s AVG WR: 34.17188 MB/s AVG RD: 0.00000 MB/s
> 8508 bytes from 172.31.22.21: seq=138 ttl=128 time=1.254 ms
>    CNT:        130 WRIT:   34603008 RATE: 33.00000 MB/s READ:          0 RATE: 0.00000 MB/s AVG WR: 34.16279 MB/s AVG RD: 0.00000 MB/s
> 8508 bytes from 172.31.22.21: seq=139 ttl=128 time=1.291 ms
>    CNT:        131 WRIT:   34603008 RATE: 33.00000 MB/s READ:          0 RATE: 0.00000 MB/s AVG WR: 34.15385 MB/s AVG RD: 0.00000 MB/s
> 8508 bytes from 172.31.22.21: seq=140 ttl=128 time=1.254 ms
>    CNT:        132 WRIT:   35651584 RATE: 34.00000 MB/s READ:          0 RATE: 0.00000 MB/s AVG WR: 34.15267 MB/s AVG RD: 0.00000 MB/s
> sata: page allocation failure. order:0, mode:0x22

sata is the process name, I believe? Not sure the SATA driver is 
involved here at all.

> Call Trace:
> [ccad9a10] [c0006ef0] show_stack+0x44/0x16c (unreliable)
> [ccad9a50] [c006f9f0] __alloc_pages_nodemask+0x38c/0x4f8
> [ccad9ad0] [c01a7e3c] emac_poll_rx+0x5ac/0x768
> [ccad9b10] [c01a28e4] mal_poll+0xa8/0x1ec
> [ccad9b40] [c01d3eec] net_rx_action+0x9c/0x1b4
> [ccad9b70] [c003b3c0] __do_softirq+0xc4/0x148
> [ccad9bb0] [c0004d18] do_softirq+0x78/0x80
> [ccad9bc0] [c003afac] irq_exit+0x64/0x7c
> [ccad9bd0] [c0005210] do_IRQ+0x9c/0xb4
> [ccad9bf0] [c000fa7c] ret_from_except+0x0/0x18
> [ccad9cb0] [00000001] 0x1
> [ccad9cd0] [c00c2a68] generic_write_end+0x24/0xe0
> [ccad9d00] [c0069cc0] generic_file_buffered_write+0x18c/0x304
> [ccad9d90] [c006a38c] __generic_file_aio_write_nolock+0x288/0x4fc
> [ccad9e00] [c006a8e4] generic_file_aio_write+0x68/0xf8
> [ccad9e30] [c009a9b4] do_sync_write+0xc4/0x138
> [ccad9ef0] [c009b41c] vfs_write+0xb4/0x158
> [ccad9f10] [c009ba4c] sys_write+0x4c/0x90
> [ccad9f40] [c000f434] ret_from_syscall+0x0/0x3c
> Mem-Info:
> DMA per-cpu:
> CPU    0: hi:   90, btch:  15 usd:  53
> Active_anon:554 active_file:1753 inactive_anon:608
>   inactive_file:49910 unevictable:0 dirty:6020 writeback:0 unstable:0
>   free:192 slab:2573 mapped:445 pagetables:25 bounce:0
> DMA free:768kB min:2036kB low:2544kB high:3052kB active_anon:2216kB inactive_anon:2432kB active_file:7012kB inactive_file:199640kB unevictable:0kB present:260096kB pages_scanned:0 all_unreclaimable? no
> lowmem_reserve[]: 0 0 0
> DMA: 2*4kB 1*8kB 1*16kB 1*32kB 1*64kB 1*128kB 0*256kB 1*512kB 0*1024kB 0*2048kB 0*4096kB = 768kB
> 51673 total pagecache pages
> 0 pages in swap cache
> Swap cache stats: add 0, delete 0, find 0/0
> Free swap  = 0kB
> Total swap = 0kB
> 65536 pages RAM
> 1400 pages reserved
> 51949 pages shared
> 12273 pages non-shared
> emac_alloc_rx_skb: allocating new page after consumption
>
> I am seeing a few different issues - one, the SATA driver is failing to allocate memory, even though there is memory available.  Also, the ibm_newemac driver is failing to allocate memory even though there is memory available.
>
> I have already modified the ibm_newemac driver to only ever allocate single pages.  Before, it would allocate based on MTU size - so if I had the MTU set to 9000, it would try and allocate 4 contiguous pages, which it would fail because there were no contiguous 16k pages available.
>
> Can anyone help me with this issue?  I have been trying to move to Linux on this hardware for a long time, but just cannot seem to get past this issue.  I am very new to kernel development and am not very familiar with the ATA system at all, let alone the memory manager.  I had a hard enough time getting to understand the networking stack (which I still feel like I don't understand completely).
>
> Thanks for the help.
>
> Jonathan

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/