diff options
| author | Al Viro <viro@ZenIV.linux.org.uk> | 2008-01-12 14:16:14 +0000 |
|---|---|---|
| committer | Jeff Garzik <jeff@garzik.org> | 2008-01-15 15:58:36 -0500 |
| commit | b50e56d81e0df964e9b28001d792021b109cf4f1 (patch) | |
| tree | e76679a3230265d1484882fd914e1b7b208004ba /drivers/ata/pata_winbond.c | |
| parent | 38ad9aebe70dc72df08851bbd1620d89329129ba (diff) | |
| download | linux-b50e56d81e0df964e9b28001d792021b109cf4f1.tar.bz2 | |
libata fixes for sparse-found problems
In pata_legacy and pata_winbond we've got bugs - cpu_to_le16() instead
of cpu_to_le32(). Fortunately, both affected suckers are VLB, thus
l-e-only, so we might get away with that unless we hit it with slop == 3
(hadn't checked if playing with badly aligned sg could trigger that).
Still buggy... Moreover, pata_legacy, pata_winbond and pata_qdi forgot to
initialize pad on the write side of 32bit case in their ->data_xfer().
Hopefully the hardware does't care, but still, sending uninitialized
data to it...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Diffstat (limited to 'drivers/ata/pata_winbond.c')
| -rw-r--r-- | drivers/ata/pata_winbond.c | 8 |
1 files changed, 3 insertions, 5 deletions
diff --git a/drivers/ata/pata_winbond.c b/drivers/ata/pata_winbond.c index 311cdb3a5566..7116a9e7a8b2 100644 --- a/drivers/ata/pata_winbond.c +++ b/drivers/ata/pata_winbond.c @@ -104,14 +104,12 @@ static void winbond_data_xfer(struct ata_device *adev, unsigned char *buf, unsig ioread32_rep(ap->ioaddr.data_addr, buf, buflen >> 2); if (unlikely(slop)) { - u32 pad; + __le32 pad = 0; if (write_data) { memcpy(&pad, buf + buflen - slop, slop); - pad = le32_to_cpu(pad); - iowrite32(pad, ap->ioaddr.data_addr); + iowrite32(le32_to_cpu(pad), ap->ioaddr.data_addr); } else { - pad = ioread32(ap->ioaddr.data_addr); - pad = cpu_to_le16(pad); + pad = cpu_to_le32(ioread32(ap->ioaddr.data_addr)); memcpy(buf + buflen - slop, &pad, slop); } } |