Re: [BUG] 2.6.25-rc2-git3 broke cdrecord

[scsi and ide list cc's added]

On Sun, 2008-03-02 at 21:53 +0100, Mikael Pettersson wrote:

On my main workstation (Intel P965 chipset, ICH8 SATA driven by ahci,
Samsung/Toshiba SH-203P SATA DVD writer, FC5 user-space), cdrecord
stopped working on the 2.6.25-rc3 kernel, with the complaint:

Warning: controller returns zero sized CD capabilities page.
Warning: controller returns wrong page 0 for CD capabilities page (2A).

There's nothing in dmesg to suggest that the kernel has experienced
some problem.

2.6.25-rc2-git2 was the last working kernel, 2.6.25-rc2-git3 up to
today's 2.6.25-rc3-git3 are all non-working.

I've narrowed the cause down to these two commits in 2.6.25-rc2-git3:

40b01b9bbdf51ae543a04744283bf2d56c4a6afa
block: update bio according to DMA alignment padding

6b00769fe1502b4ad97bb327ef7ac971b208bfb5
block: add request->raw_data_len

Four or so subsequent commits depend on the second of these, so it's
easier to test them by applying them to -git2 than to revert them.

Also, these two need to be treated as a unit. With only the first
applied, I get a massive delay as init tries to start udev, and then
if I try cdrecord it hangs in an unkillable state.

I've put some information in
<http://user.it.uu.se/~mikpe/linux/bug-2.6.25-rc3-cdrecord/>:
- the kernel's .config
- cdrecord -VV dev=/dev/scd0 blank=fast output for 2.6.25-rc2-git2 and -git3
- strace -v logs for the above cdrecord runs

I think this is another manifestation of this problem:

http://marc.info/?t=120358360500002

The fix for which is being actively discussed here:

http://marc.info/?t=120418867400002

James