Vojtech's version of drivers/ide/piix.c which went into 2.5.7
oopses with a divide-by-zero exception when initialising older
pre-UDMA chips, like in the following 430HX chipset:
00:00.0 Host bridge: Intel Corporation 430HX - 82439HX TXC [Triton II] (rev 03)
00:07.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II] (rev 01)
00:07.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton II]
(PCI IDs 8086:1250, 8086:7000, and 8086:7010, respectively)
The error occurs in piix.c:piix_set_drive() line 334, shown below.
The 82371SB has PIIX_UDMA_NONE in the piix_ide_chips[] array,
so piix_config->flags & PIIX_UDMA is zero, which makes "umul" zero,
which causes the divide-by-zero on line 334.
317 static int piix_set_drive(ide_drive_t *drive, unsigned char speed)
318 {
319 ide_drive_t *peer = HWIF(drive)->drives + (~drive->dn & 1);
320 struct ata_timing t, p;
321 int err, T, UT, umul;
322
323 if (speed != XFER_PIO_SLOW && speed != drive->current_speed)
324 if ((err = ide_config_drive_speed(drive, speed)))
325 return err;
326
327 umul = min((speed > XFER_UDMA_4) ? 4 : ((speed > XFER_UDMA_2) ? 2 : 1),
328 piix_config->flags & PIIX_UDMA);
329
330 if (piix_config->flags & PIIX_VICTORY)
331 umul = 2;
332
333 T = 1000000000 / piix_clock;
334 UT = T / umul;
/Mikael
Mikael Pettersson wrote:
> Vojtech's version of drivers/ide/piix.c which went into 2.5.7
> oopses with a divide-by-zero exception when initialising older
> pre-UDMA chips, like in the following 430HX chipset:
>
> 00:00.0 Host bridge: Intel Corporation 430HX - 82439HX TXC [Triton II] (rev 03)
> 00:07.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II] (rev 01)
> 00:07.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton II]
> (PCI IDs 8086:1250, 8086:7000, and 8086:7010, respectively)
>
> The error occurs in piix.c:piix_set_drive() line 334, shown below.
> The 82371SB has PIIX_UDMA_NONE in the piix_ide_chips[] array,
> so piix_config->flags & PIIX_UDMA is zero, which makes "umul" zero,
> which causes the divide-by-zero on line 334.
>
> 317 static int piix_set_drive(ide_drive_t *drive, unsigned char speed)
> 318 {
> 319 ide_drive_t *peer = HWIF(drive)->drives + (~drive->dn & 1);
> 320 struct ata_timing t, p;
> 321 int err, T, UT, umul;
> 322
> 323 if (speed != XFER_PIO_SLOW && speed != drive->current_speed)
> 324 if ((err = ide_config_drive_speed(drive, speed)))
> 325 return err;
> 326
> 327 umul = min((speed > XFER_UDMA_4) ? 4 : ((speed > XFER_UDMA_2) ? 2 : 1),
> 328 piix_config->flags & PIIX_UDMA);
> 329
> 330 if (piix_config->flags & PIIX_VICTORY)
> 331 umul = 2;
> 332
> 333 T = 1000000000 / piix_clock;
> 334 UT = T / umul;
I think that it should be just sufficient to add the
following test just in front of the offending calculartion.
if (umul == 0)
++umul;
Vojtech is this right?
On Fri, Mar 29, 2002 at 03:00:24PM +0100, Martin Dalecki wrote:
> Mikael Pettersson wrote:
> > Vojtech's version of drivers/ide/piix.c which went into 2.5.7
> > oopses with a divide-by-zero exception when initialising older
> > pre-UDMA chips, like in the following 430HX chipset:
> >
> > 00:00.0 Host bridge: Intel Corporation 430HX - 82439HX TXC [Triton II] (rev 03)
> > 00:07.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II] (rev 01)
> > 00:07.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton II]
> > (PCI IDs 8086:1250, 8086:7000, and 8086:7010, respectively)
> >
> > The error occurs in piix.c:piix_set_drive() line 334, shown below.
> > The 82371SB has PIIX_UDMA_NONE in the piix_ide_chips[] array,
> > so piix_config->flags & PIIX_UDMA is zero, which makes "umul" zero,
> > which causes the divide-by-zero on line 334.
> >
> > 317 static int piix_set_drive(ide_drive_t *drive, unsigned char speed)
> > 318 {
> > 319 ide_drive_t *peer = HWIF(drive)->drives + (~drive->dn & 1);
> > 320 struct ata_timing t, p;
> > 321 int err, T, UT, umul;
> > 322
> > 323 if (speed != XFER_PIO_SLOW && speed != drive->current_speed)
> > 324 if ((err = ide_config_drive_speed(drive, speed)))
> > 325 return err;
> > 326
> > 327 umul = min((speed > XFER_UDMA_4) ? 4 : ((speed > XFER_UDMA_2) ? 2 : 1),
> > 328 piix_config->flags & PIIX_UDMA);
> > 329
> > 330 if (piix_config->flags & PIIX_VICTORY)
> > 331 umul = 2;
> > 332
> > 333 T = 1000000000 / piix_clock;
> > 334 UT = T / umul;
>
> I think that it should be just sufficient to add the
> following test just in front of the offending calculartion.
>
> if (umul == 0)
> ++umul;
>
> Vojtech is this right?
Probably yes, the value of UT isn't used in that case, because we will
never try to set an UDMA mode on such a chipset. I've sent you a
slightly different patch, though.
--
Vojtech Pavlik
SuSE Labs
On Fri, Mar 29, 2002 at 01:39:56PM +0100, Mikael Pettersson wrote:
> Vojtech's version of drivers/ide/piix.c which went into 2.5.7
> oopses with a divide-by-zero exception when initialising older
> pre-UDMA chips, like in the following 430HX chipset:
>
> 00:00.0 Host bridge: Intel Corporation 430HX - 82439HX TXC [Triton II] (rev 03)
> 00:07.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II] (rev 01)
> 00:07.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton II]
> (PCI IDs 8086:1250, 8086:7000, and 8086:7010, respectively)
>
> The error occurs in piix.c:piix_set_drive() line 334, shown below.
> The 82371SB has PIIX_UDMA_NONE in the piix_ide_chips[] array,
> so piix_config->flags & PIIX_UDMA is zero, which makes "umul" zero,
> which causes the divide-by-zero on line 334.
>
> 317 static int piix_set_drive(ide_drive_t *drive, unsigned char speed)
> 318 {
> 319 ide_drive_t *peer = HWIF(drive)->drives + (~drive->dn & 1);
> 320 struct ata_timing t, p;
> 321 int err, T, UT, umul;
> 322
> 323 if (speed != XFER_PIO_SLOW && speed != drive->current_speed)
> 324 if ((err = ide_config_drive_speed(drive, speed)))
> 325 return err;
> 326
> 327 umul = min((speed > XFER_UDMA_4) ? 4 : ((speed > XFER_UDMA_2) ? 2 : 1),
> 328 piix_config->flags & PIIX_UDMA);
> 329
> 330 if (piix_config->flags & PIIX_VICTORY)
> 331 umul = 2;
> 332
> 333 T = 1000000000 / piix_clock;
> 334 UT = T / umul;
Thanks for the bug report! The attached patch should fix that.
Martin, please apply this patch.
--
Vojtech Pavlik
SuSE Labs
On Fri, 29 Mar 2002, Martin Dalecki wrote:
> > 333 T = 1000000000 / piix_clock;
> > 334 UT = T / umul;
>
> I think that it should be just sufficient to add the
> following test just in front of the offending calculartion.
>
> if (umul == 0)
> ++umul;
- 334 UT = T / umul;
+ 334 UT = T / (umul || 1);
--
bill davidsen <[email protected]>
CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.
On Fri, 29 Mar 2002 16:08:18 +0100, Vojtech Pavlik wrote:
>diff -urN linux-2.5.7/drivers/ide/piix.c linux-2.5.7-fixpiix/drivers/ide/piix.c
>--- linux-2.5.7/drivers/ide/piix.c Fri Mar 29 16:01:51 2002
>+++ linux-2.5.7-fixpiix/drivers/ide/piix.c Fri Mar 29 16:07:05 2002
...
>@@ -331,7 +331,7 @@
> umul = 2;
>
> T = 1000000000 / piix_clock;
>- UT = T / umul;
>+ UT = umul ? (T / umul) : 0;
That did indeed fix the problem. Thanks.
/Mikael
Bill Davidsen wrote:
> On Fri, 29 Mar 2002, Martin Dalecki wrote:
>
>
>>> 333 T = 1000000000 / piix_clock;
>>> 334 UT = T / umul;
>>
>>I think that it should be just sufficient to add the
>>following test just in front of the offending calculartion.
>>
>>if (umul == 0)
>> ++umul;
>
>
> - 334 UT = T / umul;
> + 334 UT = T / (umul || 1);
That is ceratily wrong. becouse umul || 1 == 1 independently
from umul.