[lkml]   [2005]   [Nov]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectVIA SATA Raid needs a long time to recover from suspend
I have been debugging a power management problem for a few days now, and 
I believe I have finally solved the problem. Because it involved
patching the kernel, I felt I should share the fix here in hopes that it
can be improved and/or integrated into future kernels. Right now I am
running on amd64, compiled myself, with the ubuntu breezy amd64

First I'll state the fix. It involved changing two lines in

static inline u8 ata_busy_wait(struct ata_port *ap, unsigned int bits,
unsigned int max)
u8 status;

do {
udelay(100); <-- changed to 100
from 10
status = ata_chk_status(ap);
} while ((status & bits) && (max > 0));

return status;


static inline u8 ata_wait_idle(struct ata_port *ap)
u8 status = ata_busy_wait(ap, ATA_BUSY | ATA_DRQ,
10000); <-- changed to 10,000 from 1,000

if (status & (ATA_BUSY | ATA_DRQ)) {
unsigned long l = ap->ioaddr.status_addr;
"ATA: abnormal status 0x%X on port 0x%lX\n",
status, l);

return status;

The problem seems to be that my VIA SATA raid controller requires more
time to recover from being suspended. It looks like the code in
sata_via.c restores the task file after a resume, then calls
ata_wait_idle to wait for the busy bit to clear. The problem was that
this function timed out before the busy bit cleared, resulting in
messages like this:

ATA: abnormal status 0x80 on port 0xE007

Then if there was an IO request made immediately after resuming, it
would timeout and fail, because it was issued before the hardware was
ready. Changing the timeout resolved this. I tried changing both the
udelay and ata_busy_wait lines to increase the timeout, and it did not
seem to matter which I changed, as long as the total timeout was
increased by a factor of 100.

Since increasing the maximum timeout, suspend and hibernate work great
for me. While experiencing this bug, it may have exposed another bug,
which I will mention now in passing. As I said before, after a resume,
if there was an IO request made immediately ( before the busy bit
finally did clear ) it would timeout and fail. It seemed the kernel
filled the buffer cache for the requested block with garbage rather than
retry the read. It seems to me that at some point, the read should have
been retried. The symptoms of this were:

1) When called immediately after the echo mem >
/sys/power/state line, then on resume, the read would fail in a block in
the resierfs tree that was required to lookup the file. This
caused reiserfs to complain about errors in the node, and the script
failed to execute. Further attempts to touch the script, even with ls
-al /etc/acpi/ failed with EPERM. I would think that at worst,
this should fail with EIO or something, not EPERM.

2) At one point I tried running echo mem > /sys/power/state ; df. After
the resume, the IO read failed when trying to load df, and I got an
error message saying the kernel could not execute the binary file.
Further attempts to run df failed also. Other IO at this point was fine.

This leads me to think that when the IO failed, rather than inform the
calling code of the failure, for example, with an EIO status, the buffer
cache got filled with junk, and this should not happen. Either the
operation should succeed, and the correct data be returned, or it should
fail, and the caller should be informed of the failure, and not given
incorrect data.

When the first IO immediately following the suspend failed, I got these

[ 32.013538] ata1: command 0x35 timeout, stat 0x50 host_stat 0x1
[ 32.045510] ata2: command 0x35 timeout, stat 0x50 host_stat 0x1

As long as no IO was immediately requested after the resume ( i.e. if I
echo mem > /sys/power/state on an otherwise idle system, rather than
using ) then these errors did not happen, only the abnormal
status messages did.

For reference, my system is configured as follows:

Motherboard: Asus K8V Deluxe
CPU: AMD Athlon 64 3200+
RAM: 1 GB of Corsair low latency pc3200 ddr sdram
Video: ATI Radeon 9800 Pro with a Samsung 930B 19 inch LCD display
Disks: 2 WD 36 gig SATA 10,000 rpm raptors in a raid0 configuration on
the via sata raid controller

/dev/mapper/via_hfciifae1: 40 gig winxp NTFS partition
/dev/mapper/via_hfciifae3: 10 gig experimental partition
/dev/mapper/via_hfciifae5: 50 meg ext2 /boot partition
/dev/mapper/via_hfciifae6: 1 gig swap partition
/dev/mapper/via_hfciifae7: 22 gig reiserfs root partition

If anyone has any suggestions of further tests I can perform to narrow
down the problem, or a better solution for it, you have my full
cooperation. If this fix seems acceptable, then I hope it can be merged
in the next kernel release.

PS> Please CC me on any replies, as I am not subscribed to this list

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-11-16 04:40    [W:0.089 / U:8.644 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site