Re: Patch to add support for SGI's IOC4 chipset

[Jeremy Higdon]

On Tue, Oct 21, 2003 at 04:39:28PM +0200, Bartlomiej Zolnierkiewicz wrote:
> On Tuesday 21 of October 2003 08:35, Jeremy Higdon wrote:
> > > - defining IDE_ARCH_ACK_INTR and ide_ack_intr() in sgiioc4.c is a no-op,
> > >   it should be done <asm/ide.h> to make it work
> > >   (I think the same problem is present in 2.4.x)
> >
> > The definition in <include/linux/ide.h> is only used if IDE_ARCH_ACK_INTR
> > is not defined.  sgiioc4.c defines IDE_ARCH_ACK_INTR before including that
> > file, so I believe we get the definition we want without touching ide.h,
> > don't we?
> 
> ide_ack_intr() is used by ide-io.c.  If IDE_ARCH_ACK_INTR is not defined
> in ide.h (and it won't be cause you are doing this only in sgiioc4.c
> /sgiioc4.h in 2.4.x case/ about which ide-io.c has abolutely no idea)
> ide_ack_intr() will turn into no-op and hwif->ack_intr() won't be called.

I see what you mean.  Thanks for spotting and fixing this.

I've run into a problem in testing.  For some reason, I've started to get
ide timeouts, and the error recovery is not working correctly, due to a
problem in the driver.

In sgiioc4_ide_dma_stop(), sgiioc4_ide_dma_end(), and sgiioc4_clearirq(),
there are calls to xide_delay(), which uses schedule_timeout() to sleep.
Since all of these sgiioc4_ functions can be called from interrupt context,
that's an obvious problem.

In sgiioc4_clearirq(), the delay function is while we're waiting for the
interrupt to clear.

In sgiioc4_ide_dma_stop(), we're waiting for the DMA bit to clear.

In sgiioc4_ide_dma_end(), we're waiting for another DMA to finish.

I believe that the right answer is to use udelay() and give up after
a short period of time.  My question is what does the ide layer
expect?  That is, if you call the dma_end function and the hardware
driver can't succeed, what would you like us to do?  Is there a way
to return error, or should we just fail and the ide infrastructure
will pick it up later and reset things?

I am new to Linux IDE, so forgive these questions if the answers should
be obvious.

thanks

jeremy
-
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/