Messages in this thread | | | Date | Wed, 2 Jan 2008 12:49:32 -0800 (PST) | From | Linus Torvalds <> | Subject | Re: [patch] scsi: revert "[SCSI] Get rid of scsi_cmnd->done" |
| |
On Wed, 2 Jan 2008, Christoph Hellwig wrote: > > I think you misunderstood Matthew here. REQ_TYPE_BLOCK_PC is indeed > used by any kind of SG_IO or similar passthrough no matter where it > originates. And exactly because REQ_TYPE_BLOCK_PC are entirely passthru > the actual driver (sd, sr or sg) is not doing the actual command > completion but it's handled in the scsi layer because it's exactly > the same no matter what driver it came on.
You say that, but you then ignore that something *did* change.
Maybe it's not that one suspicious test. Maybe it's somethign else. But that commit was confirmed to break something, almost two months ago. You guys seem to be in denial, and saying "it didn't change anything".
And no, waiting for more reporters when one reporter has already narrowed it down to the exact (smallish) commit, is simply not good. Either you can fix it by looking at the source, or it gets reverted.
I was hoping somebody in SCSI-land would actually look at the commit and try to find out what's wrong, instead of all of you apparently trying to say "nothing is wrong".
I hate the excuses of "but, but, but.. it *should* work". It doesn't. Face that, *then* you can argue about why.
Linus
| |