2004-04-29 19:04:34

by Patrick Wildi

[permalink] [raw]
Subject: [RFC][PATCH] 2.4 IDE Serverworks OSB4 DMA patch

I have been using a OSB4 chipset based system with a CompactFlash
that supports PIO only and a laptop IBM/Hitachi Travelstar HDD
that supports UDMA.
For both drives, the serverworks code misconfigures the drives:

- for the CF (hooked up as /dev/hda), svwks_config_drive_xfer_rate()
will not match any tests (drive->autodma = 0, id->capability = 2,
id->field_valid = 1), but the function will then call
hwif->ide_dma_on(drive), which it should not do for this drive.
This patch moves the enabling of DMA up into the DMA section of
the code.

- for the Travelstart HDD, the settings coming into
svwks_config_drive_xfer_rate() are: drive->autodma = 32,
id->capability = 15, id->field_valid = 7, id->dma_ultra = 0x43f.
But as this is an OSB4, the hwif->ultra_mask is set to not support
UDMA. Unfortunately in that case svwks_config_drive_xfer_rate()
falls through to the end of the function, instead of trying
other DMA modes.
By adding a jump to try_dma_modes in the case that the drive
is UDMA capable but the mask does not allow UDMA, the drive
will be setup up for MWDMA.

The logic in svwks_config_drive_xfer_rate() could probably be
improved a bit, but I kept the changes to an absolute minium
to reduce the risk of breakage.

The patch is against 2.4.24, but the file has not changed in
2.4.25 and 2.4.26.

Patrick


--- linux-2.4.24-nam/drivers/ide/pci/serverworks.c.orig Thu Apr 22 17:31:28 2004
+++ linux-2.4.24-nam/drivers/ide/pci/serverworks.c Thu Apr 29 11:02:05 2004
@@ -473,7 +473,9 @@
int dma = config_chipset_for_dma(drive);
if ((id->field_valid & 2) && !dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -490,6 +492,7 @@
} else {
goto no_dma_set;
}
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
@@ -497,7 +500,7 @@
// hwif->tuneproc(drive, 5);
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}

/* This can go soon */


Subject: Re: [RFC][PATCH] 2.4 IDE Serverworks OSB4 DMA patch

On Thursday 29 of April 2004 21:04, Patrick Wildi wrote:
> I have been using a OSB4 chipset based system with a CompactFlash
> that supports PIO only and a laptop IBM/Hitachi Travelstar HDD
> that supports UDMA.
> For both drives, the serverworks code misconfigures the drives:
>
> - for the CF (hooked up as /dev/hda), svwks_config_drive_xfer_rate()
> will not match any tests (drive->autodma = 0, id->capability = 2,
> id->field_valid = 1), but the function will then call
> hwif->ide_dma_on(drive), which it should not do for this drive.
> This patch moves the enabling of DMA up into the DMA section of
> the code.

Yep, known bug, it is fixed in 2.6.

It is present in many other drivers, my 2.6 patch needs to be backported.

> - for the Travelstart HDD, the settings coming into
> svwks_config_drive_xfer_rate() are: drive->autodma = 32,
> id->capability = 15, id->field_valid = 7, id->dma_ultra = 0x43f.
> But as this is an OSB4, the hwif->ultra_mask is set to not support
> UDMA. Unfortunately in that case svwks_config_drive_xfer_rate()
> falls through to the end of the function, instead of trying
> other DMA modes.

Good catch.

It seems the same bug can be present in other drivers too (hint, hint). ;)

> By adding a jump to try_dma_modes in the case that the drive
> is UDMA capable but the mask does not allow UDMA, the drive
> will be setup up for MWDMA.
>
> The logic in svwks_config_drive_xfer_rate() could probably be
> improved a bit, but I kept the changes to an absolute minium
> to reduce the risk of breakage.

OK, thanks.

> The patch is against 2.4.24, but the file has not changed in
> 2.4.25 and 2.4.26.
>
> Patrick
>
>
> --- linux-2.4.24-nam/drivers/ide/pci/serverworks.c.orig Thu Apr 22 17:31:28
> 2004 +++ linux-2.4.24-nam/drivers/ide/pci/serverworks.c Thu Apr 29 11:02:05
> 2004 @@ -473,7 +473,9 @@
> int dma = config_chipset_for_dma(drive);
> if ((id->field_valid & 2) && !dma)
> goto try_dma_modes;
> - }
> + } else
> + /* UDMA disabled by mask, try other DMA modes */
> + goto try_dma_modes;
> } else if (id->field_valid & 2) {
> try_dma_modes:
> if ((id->dma_mword & hwif->mwdma_mask) ||
> @@ -490,6 +492,7 @@
> } else {
> goto no_dma_set;
> }
> + return hwif->ide_dma_on(drive);
> } else if ((id->capability & 8) || (id->field_valid & 2)) {
> fast_ata_pio:
> no_dma_set:
> @@ -497,7 +500,7 @@
> // hwif->tuneproc(drive, 5);
> return hwif->ide_dma_off_quietly(drive);
> }
> - return hwif->ide_dma_on(drive);
> + return 0;
> }
>
> /* This can go soon */

2004-04-29 22:11:28

by Patrick Wildi

[permalink] [raw]
Subject: Re: [RFC][PATCH] 2.4 IDE Serverworks OSB4 DMA patch

On Thu, 29 Apr 2004, Bartlomiej Zolnierkiewicz wrote:

> On Thursday 29 of April 2004 21:04, Patrick Wildi wrote:
> > I have been using a OSB4 chipset based system with a CompactFlash
> > that supports PIO only and a laptop IBM/Hitachi Travelstar HDD
> > that supports UDMA.
> > For both drives, the serverworks code misconfigures the drives:
> >
> > - for the CF (hooked up as /dev/hda), svwks_config_drive_xfer_rate()
> > will not match any tests (drive->autodma = 0, id->capability = 2,
> > id->field_valid = 1), but the function will then call
> > hwif->ide_dma_on(drive), which it should not do for this drive.
> > This patch moves the enabling of DMA up into the DMA section of
> > the code.
>
> Yep, known bug, it is fixed in 2.6.
>
> It is present in many other drivers, my 2.6 patch needs to be backported.

Are you the maintainer for 2.4 or to whom should I send the changes?

> > - for the Travelstart HDD, the settings coming into
> > svwks_config_drive_xfer_rate() are: drive->autodma = 32,
> > id->capability = 15, id->field_valid = 7, id->dma_ultra = 0x43f.
> > But as this is an OSB4, the hwif->ultra_mask is set to not support
> > UDMA. Unfortunately in that case svwks_config_drive_xfer_rate()
> > falls through to the end of the function, instead of trying
> > other DMA modes.
>
> Good catch.
>
> It seems the same bug can be present in other drivers too (hint, hint). ;)

I noticed that the piix driver uses the exact same logic. I could
replicate this part of the patch for other 2.4 drivers. I have no
way of testing them.
I can send you a combined patch for 2.4. I am not yet using 2.6.

Patrick

> > By adding a jump to try_dma_modes in the case that the drive
> > is UDMA capable but the mask does not allow UDMA, the drive
> > will be setup up for MWDMA.
> >
> > The logic in svwks_config_drive_xfer_rate() could probably be
> > improved a bit, but I kept the changes to an absolute minium
> > to reduce the risk of breakage.
>
> OK, thanks.
>
> > The patch is against 2.4.24, but the file has not changed in
> > 2.4.25 and 2.4.26.
> >
> > Patrick
> >
> >
> > --- linux-2.4.24-nam/drivers/ide/pci/serverworks.c.orig Thu Apr 22 17:31:28
> > 2004 +++ linux-2.4.24-nam/drivers/ide/pci/serverworks.c Thu Apr 29 11:02:05
> > 2004 @@ -473,7 +473,9 @@
> > int dma = config_chipset_for_dma(drive);
> > if ((id->field_valid & 2) && !dma)
> > goto try_dma_modes;
> > - }
> > + } else
> > + /* UDMA disabled by mask, try other DMA modes */
> > + goto try_dma_modes;
> > } else if (id->field_valid & 2) {
> > try_dma_modes:
> > if ((id->dma_mword & hwif->mwdma_mask) ||
> > @@ -490,6 +492,7 @@
> > } else {
> > goto no_dma_set;
> > }
> > + return hwif->ide_dma_on(drive);
> > } else if ((id->capability & 8) || (id->field_valid & 2)) {
> > fast_ata_pio:
> > no_dma_set:
> > @@ -497,7 +500,7 @@
> > // hwif->tuneproc(drive, 5);
> > return hwif->ide_dma_off_quietly(drive);
> > }
> > - return hwif->ide_dma_on(drive);
> > + return 0;
> > }
> >
> > /* This can go soon */
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>

Subject: Re: [RFC][PATCH] 2.4 IDE Serverworks OSB4 DMA patch

On Friday 30 of April 2004 00:09, Patrick Wildi wrote:
> On Thu, 29 Apr 2004, Bartlomiej Zolnierkiewicz wrote:
> > On Thursday 29 of April 2004 21:04, Patrick Wildi wrote:
> > > I have been using a OSB4 chipset based system with a CompactFlash
> > > that supports PIO only and a laptop IBM/Hitachi Travelstar HDD
> > > that supports UDMA.
> > > For both drives, the serverworks code misconfigures the drives:
> > >
> > > - for the CF (hooked up as /dev/hda), svwks_config_drive_xfer_rate()
> > > will not match any tests (drive->autodma = 0, id->capability = 2,
> > > id->field_valid = 1), but the function will then call
> > > hwif->ide_dma_on(drive), which it should not do for this drive.
> > > This patch moves the enabling of DMA up into the DMA section of
> > > the code.
> >
> > Yep, known bug, it is fixed in 2.6.
> >
> > It is present in many other drivers, my 2.6 patch needs to be backported.
>
> Are you the maintainer for 2.4 or to whom should I send the changes?

Send them to me.

> > > - for the Travelstart HDD, the settings coming into
> > > svwks_config_drive_xfer_rate() are: drive->autodma = 32,
> > > id->capability = 15, id->field_valid = 7, id->dma_ultra = 0x43f.
> > > But as this is an OSB4, the hwif->ultra_mask is set to not support
> > > UDMA. Unfortunately in that case svwks_config_drive_xfer_rate()
> > > falls through to the end of the function, instead of trying
> > > other DMA modes.
> >
> > Good catch.
> >
> > It seems the same bug can be present in other drivers too (hint, hint).
> > ;)
>
> I noticed that the piix driver uses the exact same logic. I could
> replicate this part of the patch for other 2.4 drivers. I have no
> way of testing them.
> I can send you a combined patch for 2.4. I am not yet using 2.6.

No problem. :)

Thanks!
Bartlomiej

2004-04-30 16:45:37

by Patrick Wildi

[permalink] [raw]
Subject: Re: [RFC][PATCH] 2.4 IDE Serverworks OSB4 DMA patch

On Fri, 30 Apr 2004, Bartlomiej Zolnierkiewicz wrote:
> On Friday 30 of April 2004 00:09, Patrick Wildi wrote:
> > On Thu, 29 Apr 2004, Bartlomiej Zolnierkiewicz wrote:
> > > On Thursday 29 of April 2004 21:04, Patrick Wildi wrote:
> > > > I have been using a OSB4 chipset based system with a CompactFlash
> > > > that supports PIO only and a laptop IBM/Hitachi Travelstar HDD
> > > > that supports UDMA.
> > > > For both drives, the serverworks code misconfigures the drives:
> > > >
> > > > - for the CF (hooked up as /dev/hda), svwks_config_drive_xfer_rate()
> > > > will not match any tests (drive->autodma = 0, id->capability = 2,
> > > > id->field_valid = 1), but the function will then call
> > > > hwif->ide_dma_on(drive), which it should not do for this drive.
> > > > This patch moves the enabling of DMA up into the DMA section of
> > > > the code.
> > >
> > > Yep, known bug, it is fixed in 2.6.
> > >
> > > It is present in many other drivers, my 2.6 patch needs to be backported.
> >
> > Are you the maintainer for 2.4 or to whom should I send the changes?
>
> Send them to me.
>
> > > > - for the Travelstart HDD, the settings coming into
> > > > svwks_config_drive_xfer_rate() are: drive->autodma = 32,
> > > > id->capability = 15, id->field_valid = 7, id->dma_ultra = 0x43f.
> > > > But as this is an OSB4, the hwif->ultra_mask is set to not support
> > > > UDMA. Unfortunately in that case svwks_config_drive_xfer_rate()
> > > > falls through to the end of the function, instead of trying
> > > > other DMA modes.
> > >
> > > Good catch.
> > >
> > > It seems the same bug can be present in other drivers too (hint, hint).
> > > ;)
> >
> > I noticed that the piix driver uses the exact same logic. I could
> > replicate this part of the patch for other 2.4 drivers. I have no
> > way of testing them.
> > I can send you a combined patch for 2.4. I am not yet using 2.6.
>
> No problem. :)

Below is the 2.4 patch. I believe I updated all drivers that use
similar code.

Patrick


> Thanks!
> Bartlomiej

diff -u linux-2.4.26/drivers/ide/pci/aec62xx.c linux-2.4.26-pat/drivers/ide/pci/aec62xx.c
--- linux-2.4.26/drivers/ide/pci/aec62xx.c 2003-06-19 12:37:40.000000000 -0700
+++ linux-2.4.26-pat/drivers/ide/pci/aec62xx.c 2004-04-29 17:33:08.000000000 -0700
@@ -339,7 +339,9 @@
int dma = config_chipset_for_dma(drive);
if ((id->field_valid & 2) && !dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -356,13 +358,14 @@
} else {
goto fast_ata_pio;
}
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
aec62xx_tune_drive(drive, 5);
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}

static int aec62xx_irq_timeout (ide_drive_t *drive)
diff -u linux-2.4.26/drivers/ide/pci/alim15x3.c linux-2.4.26-pat/drivers/ide/pci/alim15x3.c
--- linux-2.4.26/drivers/ide/pci/alim15x3.c 2003-08-26 13:51:43.000000000 -0700
+++ linux-2.4.26-pat/drivers/ide/pci/alim15x3.c 2004-04-29 17:05:41.000000000 -0700
@@ -539,7 +539,9 @@
int dma = config_chipset_for_dma(drive);
if ((id->field_valid & 2) && !dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -556,11 +558,12 @@
} else {
goto no_dma_set;
}
+ return hwif->ide_dma_on(drive);
} else {
no_dma_set:
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}

/**
diff -u linux-2.4.26/drivers/ide/pci/atiixp.c linux-2.4.26-pat/drivers/ide/pci/atiixp.c
--- linux-2.4.26/drivers/ide/pci/atiixp.c 2004-04-29 16:08:20.000000000 -0700
+++ linux-2.4.26-pat/drivers/ide/pci/atiixp.c 2004-04-29 17:06:08.000000000 -0700
@@ -374,7 +374,9 @@
if ((id->field_valid & 2) &&
(!atiixp_config_drive_for_dma(drive)))
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
diff -u linux-2.4.26/drivers/ide/pci/cmd64x.c linux-2.4.26-pat/drivers/ide/pci/cmd64x.c
--- linux-2.4.26/drivers/ide/pci/cmd64x.c 2003-06-19 12:37:40.000000000 -0700
+++ linux-2.4.26-pat/drivers/ide/pci/cmd64x.c 2004-04-29 17:07:06.000000000 -0700
@@ -468,7 +468,9 @@
int dma = config_chipset_for_dma(drive);
if ((id->field_valid & 2) && !dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -485,13 +487,14 @@
} else {
goto fast_ata_pio;
}
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
config_chipset_for_pio(drive, 1);
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}

static int cmd64x_alt_dma_status (struct pci_dev *dev)
diff -u linux-2.4.26/drivers/ide/pci/hpt34x.c linux-2.4.26-pat/drivers/ide/pci/hpt34x.c
--- linux-2.4.26/drivers/ide/pci/hpt34x.c 2003-06-19 12:37:40.000000000 -0700
+++ linux-2.4.26-pat/drivers/ide/pci/hpt34x.c 2004-04-29 17:08:00.000000000 -0700
@@ -199,7 +199,9 @@
int dma = config_chipset_for_dma(drive);
if ((id->field_valid & 2) && dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -216,6 +218,10 @@
} else {
goto fast_ata_pio;
}
+#ifndef CONFIG_HPT34X_AUTODMA
+ return hwif->ide_dma_off_quietly(drive);
+#endif /* CONFIG_HPT34X_AUTODMA */
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
@@ -223,10 +229,7 @@
return hwif->ide_dma_off_quietly(drive);
}

-#ifndef CONFIG_HPT34X_AUTODMA
- return hwif->ide_dma_off_quietly(drive);
-#endif /* CONFIG_HPT34X_AUTODMA */
- return hwif->ide_dma_on(drive);
+ return 0;
}

/*
diff -u linux-2.4.26/drivers/ide/pci/hpt366.c linux-2.4.26-pat/drivers/ide/pci/hpt366.c
--- linux-2.4.26/drivers/ide/pci/hpt366.c 2004-04-29 16:08:20.000000000 -0700
+++ linux-2.4.26-pat/drivers/ide/pci/hpt366.c 2004-04-29 17:08:37.000000000 -0700
@@ -551,7 +551,9 @@
int dma = config_chipset_for_dma(drive);
if ((id->field_valid & 2) && !dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if (id->dma_mword & hwif->mwdma_mask) {
@@ -567,13 +569,14 @@
} else {
goto fast_ata_pio;
}
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
hpt3xx_tune_drive(drive, 5);
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}

/*
diff -u linux-2.4.26/drivers/ide/pci/it8172.c linux-2.4.26-pat/drivers/ide/pci/it8172.c
--- linux-2.4.26/drivers/ide/pci/it8172.c 2003-06-19 12:37:40.000000000 -0700
+++ linux-2.4.26-pat/drivers/ide/pci/it8172.c 2004-04-29 17:09:26.000000000 -0700
@@ -211,7 +211,9 @@
int dma = it8172_config_chipset_for_dma(drive);
if ((id->field_valid & 2) && !dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -228,13 +230,14 @@
} else {
goto fast_ata_pio;
}
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
it8172_tune_drive(drive, 5);
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}

static unsigned int __init init_chipset_it8172 (struct pci_dev *dev, const char *name)
diff -u linux-2.4.26/drivers/ide/pci/pdc202xx_new.c linux-2.4.26-pat/drivers/ide/pci/pdc202xx_new.c
--- linux-2.4.26/drivers/ide/pci/pdc202xx_new.c 2003-06-19 12:37:40.000000000 -0700
+++ linux-2.4.26-pat/drivers/ide/pci/pdc202xx_new.c 2004-04-29 16:21:30.000000000 -0700
@@ -397,7 +397,9 @@
int dma = config_chipset_for_dma(drive);
if ((id->field_valid & 2) && !dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -415,13 +417,14 @@
} else {
goto fast_ata_pio;
}
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
hwif->tuneproc(drive, 5);
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}

static int pdcnew_quirkproc (ide_drive_t *drive)
diff -u linux-2.4.26/drivers/ide/pci/pdc202xx_old.c linux-2.4.26-pat/drivers/ide/pci/pdc202xx_old.c
--- linux-2.4.26/drivers/ide/pci/pdc202xx_old.c 2003-12-04 15:52:57.000000000 -0800
+++ linux-2.4.26-pat/drivers/ide/pci/pdc202xx_old.c 2004-04-29 17:10:14.000000000 -0700
@@ -498,7 +498,9 @@
int dma = config_chipset_for_dma(drive);
if ((id->field_valid & 2) && !dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -516,13 +518,14 @@
} else {
goto fast_ata_pio;
}
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
hwif->tuneproc(drive, 5);
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}

static int pdc202xx_quirkproc (ide_drive_t *drive)
diff -u linux-2.4.26/drivers/ide/pci/piix.c linux-2.4.26-pat/drivers/ide/pci/piix.c
--- linux-2.4.26/drivers/ide/pci/piix.c 2004-04-29 16:08:20.000000000 -0700
+++ linux-2.4.26-pat/drivers/ide/pci/piix.c 2004-04-29 16:23:13.000000000 -0700
@@ -576,7 +576,9 @@
if ((id->field_valid & 2) &&
(!piix_config_drive_for_dma(drive)))
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -593,13 +595,14 @@
} else {
goto fast_ata_pio;
}
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
hwif->tuneproc(drive, 255);
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}

/**
diff -u linux-2.4.26/drivers/ide/pci/serverworks.c linux-2.4.26-pat/drivers/ide/pci/serverworks.c
--- linux-2.4.26/drivers/ide/pci/serverworks.c 2003-12-04 15:52:57.000000000 -0800
+++ linux-2.4.26-pat/drivers/ide/pci/serverworks.c 2004-04-29 16:08:57.000000000 -0700
@@ -473,7 +473,9 @@
int dma = config_chipset_for_dma(drive);
if ((id->field_valid & 2) && !dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -490,6 +492,7 @@
} else {
goto no_dma_set;
}
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
@@ -497,7 +500,7 @@
// hwif->tuneproc(drive, 5);
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}

/* This can go soon */
diff -u linux-2.4.26/drivers/ide/pci/siimage.c linux-2.4.26-pat/drivers/ide/pci/siimage.c
--- linux-2.4.26/drivers/ide/pci/siimage.c 2004-04-29 16:08:20.000000000 -0700
+++ linux-2.4.26-pat/drivers/ide/pci/siimage.c 2004-04-29 17:10:53.000000000 -0700
@@ -499,7 +499,9 @@
int dma = config_chipset_for_dma(drive);
if ((id->field_valid & 2) && !dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -516,13 +518,14 @@
} else {
goto fast_ata_pio;
}
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
config_chipset_for_pio(drive, 1);
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}

/* returns 1 if dma irq issued, 0 otherwise */
diff -u linux-2.4.26/drivers/ide/pci/sis5513.c linux-2.4.26-pat/drivers/ide/pci/sis5513.c
--- linux-2.4.26/drivers/ide/pci/sis5513.c 2003-08-26 13:51:43.000000000 -0700
+++ linux-2.4.26-pat/drivers/ide/pci/sis5513.c 2004-04-29 16:25:07.000000000 -0700
@@ -680,7 +680,9 @@
int dma = config_chipset_for_dma(drive);
if ((id->field_valid & 2) && !dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -697,13 +699,14 @@
} else {
goto fast_ata_pio;
}
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
sis5513_tune_drive(drive, 5);
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}

/* initiates/aborts (U)DMA read/write operations on a drive. */
diff -u linux-2.4.26/drivers/ide/pci/slc90e66.c linux-2.4.26-pat/drivers/ide/pci/slc90e66.c
--- linux-2.4.26/drivers/ide/pci/slc90e66.c 2003-06-19 12:37:40.000000000 -0700
+++ linux-2.4.26-pat/drivers/ide/pci/slc90e66.c 2004-04-29 17:11:29.000000000 -0700
@@ -285,7 +285,9 @@
int dma = slc90e66_config_drive_for_dma(drive);
if ((id->field_valid & 2) && !dma)
goto try_dma_modes;
- }
+ } else
+ /* UDMA disabled by mask, try other DMA modes */
+ goto try_dma_modes;
} else if (id->field_valid & 2) {
try_dma_modes:
if ((id->dma_mword & hwif->mwdma_mask) ||
@@ -302,13 +304,14 @@
} else {
goto fast_ata_pio;
}
+ return hwif->ide_dma_on(drive);
} else if ((id->capability & 8) || (id->field_valid & 2)) {
fast_ata_pio:
no_dma_set:
hwif->tuneproc(drive, 5);
return hwif->ide_dma_off_quietly(drive);
}
- return hwif->ide_dma_on(drive);
+ return 0;
}
#endif /* CONFIG_BLK_DEV_IDEDMA */

Subject: Re: [RFC][PATCH] 2.4 IDE Serverworks OSB4 DMA patch

On Friday 30 of April 2004 18:44, Patrick Wildi wrote:
> On Fri, 30 Apr 2004, Bartlomiej Zolnierkiewicz wrote:
> > On Friday 30 of April 2004 00:09, Patrick Wildi wrote:
> > > On Thu, 29 Apr 2004, Bartlomiej Zolnierkiewicz wrote:
> > > > On Thursday 29 of April 2004 21:04, Patrick Wildi wrote:
> > > > > I have been using a OSB4 chipset based system with a CompactFlash
> > > > > that supports PIO only and a laptop IBM/Hitachi Travelstar HDD
> > > > > that supports UDMA.
> > > > > For both drives, the serverworks code misconfigures the drives:
> > > > >
> > > > > - for the CF (hooked up as /dev/hda),
> > > > > svwks_config_drive_xfer_rate() will not match any tests
> > > > > (drive->autodma = 0, id->capability = 2, id->field_valid = 1), but
> > > > > the function will then call
> > > > > hwif->ide_dma_on(drive), which it should not do for this drive.
> > > > > This patch moves the enabling of DMA up into the DMA section of
> > > > > the code.
> > > >
> > > > Yep, known bug, it is fixed in 2.6.
> > > >
> > > > It is present in many other drivers, my 2.6 patch needs to be
> > > > backported.
> > >
> > > Are you the maintainer for 2.4 or to whom should I send the changes?
> >
> > Send them to me.
> >
> > > > > - for the Travelstart HDD, the settings coming into
> > > > > svwks_config_drive_xfer_rate() are: drive->autodma = 32,
> > > > > id->capability = 15, id->field_valid = 7, id->dma_ultra = 0x43f.
> > > > > But as this is an OSB4, the hwif->ultra_mask is set to not
> > > > > support UDMA. Unfortunately in that case
> > > > > svwks_config_drive_xfer_rate() falls through to the end of the
> > > > > function, instead of trying other DMA modes.
> > > >
> > > > Good catch.
> > > >
> > > > It seems the same bug can be present in other drivers too (hint,
> > > > hint). ;)
> > >
> > > I noticed that the piix driver uses the exact same logic. I could
> > > replicate this part of the patch for other 2.4 drivers. I have no
> > > way of testing them.
> > > I can send you a combined patch for 2.4. I am not yet using 2.6.
> >
> > No problem. :)
>
> Below is the 2.4 patch. I believe I updated all drivers that use
> similar code.

This is not needed. I've just checked all drivers:

- no IORDY fix is not necessary for alim15x3.c

- hwif->ultra_mask fix is only needed for OSB4

[ I discovered another bug in the process - some drivers are using
hwif->ultra_mask == 0x80 to indicate that UDMA is unsupported
but bit 0x80 means UDMA7 (UDMA150) now, oh well. ]

Therefore I will send Marcelo backport of 2.6 patch for no IORDY support
+ your OSB4 fix and Linus will get only your OSB4 fix. Thanks!

Cheers,
Bartlomiej