[PATCH] pata_ali: Lots of problems still showing up with small ATAPI DMA

[PATCH] pata_ali: Lots of problems still showing up with small ATAPI DMA

[Alan Cox]

Hopefully there is a better long term solution but for now lets favour
reliability.

Signed-off-by: Alan Cox <alan@redhat.com>

diff -u --new-file --recursive --exclude-from /usr/src/exclude linux.vanilla-2.6.24-rc2-mm1/drivers/ata/pata_ali.c linux-2.6.24-rc2-mm1/drivers/ata/pata_ali.c
--- linux.vanilla-2.6.24-rc2-mm1/drivers/ata/pata_ali.c	2007-11-16 17:55:11.000000000 +0000
+++ linux-2.6.24-rc2-mm1/drivers/ata/pata_ali.c	2007-11-16 18:12:21.000000000 +0000
@@ -282,6 +285,21 @@
 	adev->max_sectors = 255;
 }
 
+/**
+ *	ali_check_atapi_dma	-	DMA check for most ALi controllers
+ *	@adev: Device
+ *
+ *	Called to decide whether commands should be sent by DMA or PIO
+ */
+ 
+static int ali_check_atapi_dma(struct ata_queued_cmd *qc)
+{
+	/* If its not a media command, its not worth it */
+	if (qc->nbytes < 2048)
+		return -EOPNOTSUPP;
+	return 0;
+}
+
 static struct scsi_host_template ali_sht = {
 	.module			= THIS_MODULE,
 	.name			= DRV_NAME,
@@ -378,6 +396,7 @@
 	.mode_filter	= ata_pci_default_filter,
 	.tf_load	= ata_tf_load,
 	.tf_read	= ata_tf_read,
+	.check_atapi_dma = ali_check_atapi_dma,
 	.check_status 	= ata_check_status,
 	.exec_command	= ata_exec_command,
 	.dev_select 	= ata_std_dev_select,
@@ -415,6 +434,7 @@
 	.mode_filter	= ata_pci_default_filter,
 	.tf_load	= ata_tf_load,
 	.tf_read	= ata_tf_read,
+	.check_atapi_dma = ali_check_atapi_dma,
 	.check_status 	= ata_check_status,
 	.exec_command	= ata_exec_command,
 	.dev_select 	= ata_std_dev_select,

Re: [PATCH] pata_ali: Lots of problems still showing up with small ATAPI DMA

[Jeff Garzik]

Alan Cox wrote:
> Hopefully there is a better long term solution but for now lets favour
> reliability.
> 
> Signed-off-by: Alan Cox <alan@redhat.com>

applied #upstream-fixes

Re: [PATCH] pata_ali: Lots of problems still showing up with small ATAPI DMA

[Tejun Heo]

Alan Cox wrote:
> Hopefully there is a better long term solution but for now lets favour
> reliability.

I'm getting a lot of reports on pata_ali w/ misc ATAPI commands too.  Do
you know whether it's a controller problem or drive one?  All the
reports I got are laptops and I can't really ask them to disassemble
their laptops.  :-)

Thanks.

-- 
tejun

Re: [PATCH] pata_ali: Lots of problems still showing up with small ATAPI DMA

[Alan Cox]

On Wed, 28 Nov 2007 10:36:15 +0900
Tejun Heo <htejun@gmail.com> wrote:

> Alan Cox wrote:
> > Hopefully there is a better long term solution but for now lets favour
> > reliability.
> 
> I'm getting a lot of reports on pata_ali w/ misc ATAPI commands too.  Do
> you know whether it's a controller problem or drive one?  All the
> reports I got are laptops and I can't really ask them to disassemble
> their laptops.  :-)

ALi is one that most definitely appears to be controller related. Thats
why I sent Jeff a patch to use DMA for 2K+ transfers only. Some ALi also
required the patch to set the xfer lengths.

Its only some revs/combinations annoyingly - a lot of them (including the
ones I have direct access too) just work.

Alan

Re: [PATCH] pata_ali: Lots of problems still showing up with small ATAPI DMA

[Tejun Heo]

Alan Cox wrote:
> On Wed, 28 Nov 2007 10:36:15 +0900
> Tejun Heo <htejun@gmail.com> wrote:
> 
>> Alan Cox wrote:
>>> Hopefully there is a better long term solution but for now lets favour
>>> reliability.
>> I'm getting a lot of reports on pata_ali w/ misc ATAPI commands too.  Do
>> you know whether it's a controller problem or drive one?  All the
>> reports I got are laptops and I can't really ask them to disassemble
>> their laptops.  :-)
> 
> ALi is one that most definitely appears to be controller related. Thats
> why I sent Jeff a patch to use DMA for 2K+ transfers only. Some ALi also
> required the patch to set the xfer lengths.
> 
> Its only some revs/combinations annoyingly - a lot of them (including the
> ones I have direct access too) just work.

Does READ_CD work fine via DMA?  2K+ filtering filters most if not all
misc commands so usually only sector (2K) aligned READ and WRITEs are
done via DMA.  Interesting exceptions are READ_CD[_MSF] and
READ/WRITE_BUFFER which usually exceed 2K but aren't aligned to 2K.  I
guess it's about time to write a test case program which issues a bunch
of problematic commands and see which works.

Thanks.

-- 
tejun

Re: [PATCH] pata_ali: Lots of problems still showing up with small ATAPI DMA

[Alan Cox]

O> Does READ_CD work fine via DMA?  2K+ filtering filters most if not all
> misc commands so usually only sector (2K) aligned READ and WRITEs are
> done via DMA.  Interesting exceptions are READ_CD[_MSF] and
> READ/WRITE_BUFFER which usually exceed 2K but aren't aligned to 2K.  I
> guess it's about time to write a test case program which issues a bunch
> of problematic commands and see which works.

I'm not 100% sure but the reports for both IT821x and ALi with that are
that it does and the IT821x case I'm pretty certain of.

Getting a tester to move a CD drive between the ALi and a PIIX
controller shows the behaviour is controller specific

Alan