Today I tried the kernel 2.6.35 in one of my boxes but I realized that the
box doesn't detect my SATA HD anymore. The logs show that the driver sata_sil
is correctly loaded. With 2.6.34 it worked fine.
So I tried to test the latest git to see if the issue was fixed, but that
kernel exhibits the same behaviour. Then I looked at the linux-ide mailing list
archives at http://marc.info/?t=128232284600001&r=1&w=2 and I found out that
the issue is known but not solved and since I could reliably reproduce the
problen I started to bisect.
The logs of the bisection are attached. It pointed to the commit
978c066691a49a205673672a55685305663a2554 ( libata: Remove excess delay in the
tf_load path ).
So I reverted that commit and got a bootable kernel again. I think this commit
exposed a timing bug in the sata_sil driver.
gianluca
git bisect start '--' 'drivers/ata'
# bad: [1506707a6c740db316e422239a53ae5df1727591] Linux 2.6.35.4
git bisect bad 1506707a6c740db316e422239a53ae5df1727591
# good: [e40152ee1e1c7a63f4777791863215e3faa37a86] Linus 2.6.34
git bisect good e40152ee1e1c7a63f4777791863215e3faa37a86
# bad: [c7087652e1890a3feef35b30ee1d4be68e1932cd] libata-sff: clean up BMDMA initialization
git bisect bad c7087652e1890a3feef35b30ee1d4be68e1932cd
# good: [fe6005b81e142c36f066b81ca85f80022c4ac979] pata_scc: make scc_wait_after_reset() static
git bisect good fe6005b81e142c36f066b81ca85f80022c4ac979
# bad: [978c066691a49a205673672a55685305663a2554] libata: Remove excess delay in the tf_load path
git bisect bad 978c066691a49a205673672a55685305663a2554
# good: [41dec29bcb05eb8ec396f70ce791c6e3e4ce4712] libata: introduce sff_set_devctl() method
git bisect good 41dec29bcb05eb8ec396f70ce791c6e3e4ce4712
# good: [276a47a93dbfe5d35261451925020f65cdbfcce8] pata_cmd640: don't read CFR pointlessly
git bisect good 276a47a93dbfe5d35261451925020f65cdbfcce8
# good: [95cc2c70c139936a2142bcd583da8af6f9d88efb] sata_nv: use ata_pci_sff_activate_host() instead of ata_host_activate()
git bisect good 95cc2c70c139936a2142bcd583da8af6f9d88efb
commit 978c066691a49a205673672a55685305663a2554
Author: Alan Cox <alan@xxxxxxxxxxxxxxx>
Date: Wed May 5 10:24:57 2010 +0100
libata: Remove excess delay in the tf_load path
We don't need to stall and wait after loading the task file and before
issuing a command, so don't do it. This shows up on profiles and is not
needed.
Signed-off-by: Alan Cox <alan@xxxxxxxxxxxxxxx>
Signed-off-by: Jeff Garzik <jgarzik@xxxxxxxxxx>
