[BUG] 2.6.29-rc1: ata_piix boot hang due to bmdma32 change

[BUG] 2.6.29-rc1: ata_piix boot hang due to bmdma32 change

[Mikael Pettersson]

One of my machines is an Intel 815EP chipset box with
an 82801BA IDE controller (8086:244b rev 05) and two
ATAPI devices on the second channel (the boot disk is
connected to a PDC20269).

With 2.6.29-rc1 this machine hangs hard during boot, when
init starts udev. There are no messages on the console and
not even sysrq gets any reaction.

Reverting the 2.6.29-rc1 change which made ata_piix inherit
the bmdma32 port ops makes the machine boot ok again. (Patch below.)
I'm guessing the 32-bit I/O stuff doesn't work on this controller
or these ATAPI devices.

Partial dmesg from 2.6.29-rc1 with the bmdma32 change reverted:

ata_piix 0000:00:1f.1: version 2.12
ata_piix 0000:00:1f.1: setting latency timer to 64
scsi0 : ata_piix
scsi1 : ata_piix
ata1: PATA max UDMA/100 cmd 0x1f0 ctl 0x3f6 bmdma 0xf000 irq 14
ata2: PATA max UDMA/100 cmd 0x170 ctl 0x376 bmdma 0xf008 irq 15
ata2.00: ATAPI: CRD-8320B, 1.12, max MWDMA2
ata2.01: ATAPI: Hewlett-Packard CD-Writer Plus 9100, 1.0c, max UDMA/33
ata2.00: configured for MWDMA2
ata2.01: configured for UDMA/33
scsi 1:0:0:0: CD-ROM            GoldStar CD-ROM CRD-8320B 1.12 PQ: 0 ANSI: 5
scsi 1:0:1:0: CD-ROM            HP       CD-Writer+ 9100  1.0c PQ: 0 ANSI: 5

/Mikael

--- linux-2.6.29-rc1/drivers/ata/ata_piix.c.~1~	2009-01-13 23:35:23.000000000 +0100
+++ linux-2.6.29-rc1/drivers/ata/ata_piix.c	2009-01-14 22:57:37.000000000 +0100
@@ -310,7 +310,7 @@ static struct scsi_host_template piix_sh
 };
 
 static struct ata_port_operations piix_pata_ops = {
-	.inherits		= &ata_bmdma32_port_ops,
+	.inherits		= &ata_bmdma_port_ops,
 	.cable_detect		= ata_cable_40wire,
 	.set_piomode		= piix_set_piomode,
 	.set_dmamode		= piix_set_dmamode,

Re: [BUG] 2.6.29-rc1: ata_piix boot hang due to bmdma32 change

[Mikael Pettersson]

Mikael Pettersson writes:
 > One of my machines is an Intel 815EP chipset box with
 > an 82801BA IDE controller (8086:244b rev 05) and two
 > ATAPI devices on the second channel (the boot disk is
 > connected to a PDC20269).
 > 
 > With 2.6.29-rc1 this machine hangs hard during boot, when
 > init starts udev. There are no messages on the console and
 > not even sysrq gets any reaction.
 > 
 > Reverting the 2.6.29-rc1 change which made ata_piix inherit
 > the bmdma32 port ops makes the machine boot ok again. (Patch below.)
 > I'm guessing the 32-bit I/O stuff doesn't work on this controller
 > or these ATAPI devices.
 > 
 > Partial dmesg from 2.6.29-rc1 with the bmdma32 change reverted:
 > 
 > ata_piix 0000:00:1f.1: version 2.12
 > ata_piix 0000:00:1f.1: setting latency timer to 64
 > scsi0 : ata_piix
 > scsi1 : ata_piix
 > ata1: PATA max UDMA/100 cmd 0x1f0 ctl 0x3f6 bmdma 0xf000 irq 14
 > ata2: PATA max UDMA/100 cmd 0x170 ctl 0x376 bmdma 0xf008 irq 15
 > ata2.00: ATAPI: CRD-8320B, 1.12, max MWDMA2
 > ata2.01: ATAPI: Hewlett-Packard CD-Writer Plus 9100, 1.0c, max UDMA/33
 > ata2.00: configured for MWDMA2
 > ata2.01: configured for UDMA/33
 > scsi 1:0:0:0: CD-ROM            GoldStar CD-ROM CRD-8320B 1.12 PQ: 0 ANSI: 5
 > scsi 1:0:1:0: CD-ROM            HP       CD-Writer+ 9100  1.0c PQ: 0 ANSI: 5

The problem is still present in 2.6.29-rc2.

More debugging first showed that only the specific machine mentioned
above is affected, as I've tested -rc1 on several other ata_piix boxes
with various oldish disks and cdroms and none of the others triggered
the hang.

Secondly, the hang is due to the HP CD-Writer drive as having only it
on the second port suffices to produce the hang. It doesn't matter if
it's configured as master or slave.

In any event, this shows that using 32-bit I/O ops should also be a
function of the drive you're talking to, not just the controller.

/Mikael

Re: [BUG] 2.6.29-rc1: ata_piix boot hang due to bmdma32 change

[Tejun Heo]

Hello,

Mikael Pettersson wrote:
> The problem is still present in 2.6.29-rc2.
> 
> More debugging first showed that only the specific machine mentioned
> above is affected, as I've tested -rc1 on several other ata_piix boxes
> with various oldish disks and cdroms and none of the others triggered
> the hang.
> 
> Secondly, the hang is due to the HP CD-Writer drive as having only it
> on the second port suffices to produce the hang. It doesn't matter if
> it's configured as master or slave.
> 
> In any event, this shows that using 32-bit I/O ops should also be a
> function of the drive you're talking to, not just the controller.

If you connect the writer to another machine, does it make the machine
hang too?

-- 
tejun

Re: [BUG] 2.6.29-rc1: ata_piix boot hang due to bmdma32 change

[Mikael Pettersson]

Tejun Heo writes:
 > Hello,
 > 
 > Mikael Pettersson wrote:
 > > The problem is still present in 2.6.29-rc2.
 > > 
 > > More debugging first showed that only the specific machine mentioned
 > > above is affected, as I've tested -rc1 on several other ata_piix boxes
 > > with various oldish disks and cdroms and none of the others triggered
 > > the hang.
 > > 
 > > Secondly, the hang is due to the HP CD-Writer drive as having only it
 > > on the second port suffices to produce the hang. It doesn't matter if
 > > it's configured as master or slave.
 > > 
 > > In any event, this shows that using 32-bit I/O ops should also be a
 > > function of the drive you're talking to, not just the controller.
 > 
 > If you connect the writer to another machine, does it make the machine
 > hang too?

I haven't tried it since that machine is at a different geographical
location from my other machines. But I can try it next weekend.

My hang goes away with Hugh Dickins' fix for ata_sff_data_xfer32 [1]
which fixed a CD-ROM error (not hang) on his ata_piix box, and his
patch also fixed a CD-ROM error on a another person's pata_ali box [2].

The patch changes ata_sff_data_xfer32 to use 16-bit PIO instead of 32-bit
on any 1/2/3 byte "slop" after the main payload. But Alan thinks that
AMD chipsets cannot handle that [3]. (One of my test boxes is an old
AMD-chipset Athlon, I can test both my problematic CD-writer and the
possibly AMD-incompatible fix on it next weekend.)

[1] <http://marc.info/?l=linux-kernel&m=123254501314058&w=2>
[2] <http://marc.info/?l=linux-ide&m=123277213928172&w=2>
[3] <http://marc.info/?l=linux-kernel&m=123257958112434&w=2>

/Mikael

Re: [BUG] 2.6.29-rc1: ata_piix boot hang due to bmdma32 change

[Mikael Pettersson]

Mikael Pettersson writes:
 > Tejun Heo writes:
 >  > Hello,
 >  > 
 >  > Mikael Pettersson wrote:
 >  > > The problem is still present in 2.6.29-rc2.
 >  > > 
 >  > > More debugging first showed that only the specific machine mentioned
 >  > > above is affected, as I've tested -rc1 on several other ata_piix boxes
 >  > > with various oldish disks and cdroms and none of the others triggered
 >  > > the hang.
 >  > > 
 >  > > Secondly, the hang is due to the HP CD-Writer drive as having only it
 >  > > on the second port suffices to produce the hang. It doesn't matter if
 >  > > it's configured as master or slave.
 >  > > 
 >  > > In any event, this shows that using 32-bit I/O ops should also be a
 >  > > function of the drive you're talking to, not just the controller.
 >  > 
 >  > If you connect the writer to another machine, does it make the machine
 >  > hang too?
 > 
 > I haven't tried it since that machine is at a different geographical
 > location from my other machines. But I can try it next weekend.

I've now tried this CD-Writer on a couple of other boxes.

AMD756 (1022:7409 rev 03), pata_amd
2.6.29-rc3: hang as init starts udev
rc3 + Hugh's libata-sff patch: no hang and no visible issues,
despite this possibly being out of spec for the AMD IDE controller
rc3 + Hugh's patch + Alan's patch: no hang, all ok

440BX/PIIX4 (8086:7111 rev 01), ata_piix
2.6.29-rc3: hang as init starts udev
rc3 + Hugh's libata-sff patch: all ok

430HX/PIIX3 (8086:7010)
2.6.29-rc3: hang during driver init, immediately after drive was identified
rc3 + Hugh's libata-sff patch: all ok