Re: 2.6.19-rc3: known unfixed regressions (v3)

[Linus Torvalds]

On Mon, 30 Oct 2006, Jun'ichi Nomura wrote:
> 
> The code is related to bd_claim_by_disk which is called when
> device-mapper or md tries to mark the underlying devices
> for exclusive use and creates symlinks from/to the devices
> in sysfs. The patch added error handlings which weren't in
> the original code.

Actually, looking closer at the code, the patch seems to add _incorrect_ 
error handling.

For example, look at bd_claim_by_kobject(): if the "bd_claim()" inside of 
it succeeds, we used to always return success. Now, we don't necessarily 
do that: we may have done a _successful_ "bd_claim()" call, but then we 
return an error because something else failed, and now we're returning 
with from bd_claim_by_kobject() with the bd_claim() done, but with an 
error return (so the caller will _not_ call "bd_release()", and the 
block_device will forever stay exclusive).

No?

Now, exactly why acpi stops working as a result, I don't know, but maybe 
something else tries to get exclusive access to a swap partition, for 
example, and now fails, causing some acpi sequence to not be set up? 
Dunno.

So I suspect it should be reverted, but maybe somebody can see exactly 
what goes wrong here.

		Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/