* [PATCH RFC] libata-core: do something about flush cache
@ 2008-02-20 21:45 Alan Cox
0 siblings, 0 replies; only message in thread
From: Alan Cox @ 2008-02-20 21:45 UTC (permalink / raw)
To: jeff, linux-ide
The IDE layers assume a failed flush cache is a simple error, but it is
actually far more complicated. The spec says that an error is returned by
flush cache if a sector cannot be committed to disc. In that case the
failed sector is dropped from cache and the command reports the bad
sector. It does *not* automatically flush other sectors. Instead we must
re-issue the flush.
Signed-off-by: Alan Cox <alan@redhat.com>
diff -u --new-file --recursive --exclude-from /usr/src/exclude linux.vanilla-2.6.25-rc2-mm1/drivers/ata/libata-core.c linux-2.6.25-rc2-mm1/drivers/ata/libata-core.c
--- linux.vanilla-2.6.25-rc2-mm1/drivers/ata/libata-core.c 2008-02-19 11:03:26.000000000 +0000
+++ linux-2.6.25-rc2-mm1/drivers/ata/libata-core.c 2008-02-20 11:56:25.000000000 +0000
@@ -6477,10 +6477,30 @@
return 0;
}
+static void ata_flush_cache_error(u8 cmd, struct ata_device *dev,
+ struct ata_taskfile *tf)
+{
+ char *df = "";
+ u64 lba;
+
+ if (tf->command & ATA_DF)
+ df = " (device fault)";
+
+ if (cmd == ATA_CMD_FLUSH)
+ lba = ata_tf_to_lba(tf);
+ else
+ lba = ata_tf_to_lba48(tf);
+
+ ata_dev_printk(dev, KERN_WARNING, "flush of sector %lld failed%s.\n",
+ lba, df);
+}
+
int ata_flush_cache(struct ata_device *dev)
{
unsigned int err_mask;
u8 cmd;
+ struct ata_taskfile tf;
+ int i;
if (!ata_try_flush_cache(dev))
return 0;
@@ -6490,17 +6510,32 @@
else
cmd = ATA_CMD_FLUSH;
- /* This is wrong. On a failed flush we get back the LBA of the lost
- sector and we should (assuming it wasn't aborted as unknown) issue
- a further flush command to continue the writeback until it
- does not error */
- err_mask = ata_do_simple_cmd(dev, cmd);
- if (err_mask) {
- ata_dev_printk(dev, KERN_ERR, "failed to flush cache\n");
- return -EIO;
- }
+ /* 20 failed flush sectors and we abandon hope. This is a heuristic
+ we can twiddle later */
+ for (i = 0; i < 20; i++) {
+ ata_tf_init(dev, &tf);
+ tf.command = cmd;
+ tf.flags |= ATA_TFLAG_DEVICE;
+ tf.protocol = ATA_PROT_NODATA;
- return 0;
+ err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
+
+ /* Flush completed without further errors */
+ if (err_mask == 0)
+ return 0;
+ /* Non device errors cannot be fixed by a retry at this
+ level of the system */
+ if (err_mask != AC_ERR_DEV) {
+ ata_dev_printk(dev, KERN_ERR, "cache flush failed.\n");
+ return -EIO;
+ }
+ /* On a media error the bad block is reported and the flush
+ stops. It will continue from after this sector when we
+ reissue the command */
+ ata_flush_cache_error(cmd, dev, &tf);
+ }
+ ata_dev_printk(dev, KERN_ERR, "cache flush repeatedly failed.\n");
+ return -EIO;
}
#ifdef CONFIG_PM
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2008-02-20 21:54 UTC | newest]
Thread overview: (only message) (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2008-02-20 21:45 [PATCH RFC] libata-core: do something about flush cache Alan Cox
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).