On Sat, Jun 14, 2008 at 07:29:00PM +0200, Bartlomiej Zolnierkiewicz wrote:

Hi,

quoted text
> 
> Hi,
> 
> On Thursday 12 June 2008, Borislav Petkov wrote:
> > Hi Bart,
> > 
> > here are some generic ide conversion patches. The first 12 are what i thought
> > you suggested :) concerning prepping the ide-cd code for the generic layer. The
> > remaining 6 are what i've done so far wrt removing ide_atapi_pc from the ide
> > drivers. It is obvious that this is not trivial and I basically tiptoe around the
> 
> Thanks.
> 
> I applied patches #1-2, #5-12 and #14-15.
> 
> I skipped patches #3-4, #13 and #16-18 for now
> (more details in replies to corresponding mails).
> 
> > landmines in the IRQ handler and request issue paths :). This raises also
> > several questions:
> > 
> > 1. ide-cd uses the cdrom_info struct for e.g. dma status notification, the other
> > ide drivers use it per packet command in pc->flags. Well the last are kinda too
> > much to carry for _each_ command and i'm thinking maybe put all that in the
> > ide_drive_t and make it generic enough to fit all drivers. This way
> > pc->callback() can also be harboured there. One concern might be when a flag is
> > strictly per packet command which could be put somewhere else (maybe in the
> > struct request since it already has members when it is used as a packet
> > command).
> 
> Some pc->flags describe device's properties and thus should be moved to
> ide_drive_t (->dev_flags) while some other correspond to the queued command
> and moving them to ide_drive_t would be a bad idea IMO.

With "the last" i meant pc->flags and not "per packet" so i completely agree:
per-packet commands go somewhere in rq and the others land in ide_drive_t.

quoted text
> For ATA commands I was planning to put taskfile flags into rq->special field
> (well, I actually implemented it in draft patch and it looks OK) so maybe we
> can do something similar for packet commands.

Yep, i was looking for a field to put all those per-command flags into so
rq->special sounds good. Will prepare some patches for that later.

quoted text
> > 2. Can all that pc->xferred, pc->b_count, pc->errors and pc->cur_pos accounting
> > be safely mapped to a rq? I see some discrepancies like is pc->buf_size ==
> > rq->data_len, what about pc->req_xfer? I'll have a more detailed look at those
> > when i have more spare time later.
> 
> If you ask if they can be mapped 'directly' then the answer is: "probably no"
> but if the question is whether it is possible to do it after some changes then
> the answer is: "probably yes". :)
> 
> [ However it may be necessary to convert ATAPI drivers to use scatterlists
>   instead of open-coded ->bio walking for PIO transfers first. ]

Looking into it right now...

Thanks.

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