misc/123887: PDC20262 does not support 48 bit DMA access
J. Su
minichocobo at gmail.com
Thu May 22 07:10:02 UTC 2008
>Number: 123887
>Category: misc
>Synopsis: PDC20262 does not support 48 bit DMA access
>Confidential: no
>Severity: serious
>Priority: medium
>Responsible: freebsd-bugs
>State: open
>Quarter:
>Keywords:
>Date-Required:
>Class: sw-bug
>Submitter-Id: current-users
>Arrival-Date: Thu May 22 07:10:01 UTC 2008
>Closed-Date:
>Last-Modified:
>Originator: J. Su
>Release: 6.3-RELEASE
>Organization:
>Environment:
FreeBSD rasputin.eecs.harvard.edu 6.3-RELEASE-p2 FreeBSD 6.3-RELEASE-p2 #2: Thu May 22 00:30:06 EDT 2008 root at somewhere.net:/usr/src/sys/i386/compile/MYKERNEL i386
>Description:
My system has a Promise Ultra66 and two 400GB hard disks on it.
If I enable IDE DMA, everything works fine until my system start to access something above 137GB boundary. It keeps produce error messages like:
kernel: ad6: FAILURE - READ_DMA48 timed out LBA=781422764
kernel: ad6: FAILURE - READ_DMA timed out LBA=0
kernel: ad6: FAILURE - READ_DMA48 timed out LBA=781422767
kernel: ad6: TIMEOUT - READ_DMA retrying (1 retry left) LBA=1
kernel: ad6: FAILURE - READ_DMA48 timed out LBA=781422705
If I disable DMA ( hw.ata.ata_dma=0), accessing is fine but accessing speed is very low.
I checked the linux kernel mailing list. It seems that Ultra66 does not support the hardware hack of LBA48 support used by Ultra 100, as claimed by Hank Yang. http://lwn.net/Articles/7253/
Linux 2.4.20 kernel source also indicate that the DMA hack is disabled for Ultra66 ( chip: PDC20262 ) as well. check http://lxr-itec.uni-klu.ac.at/um-linux-2.4.20/source/drivers/ide/pdc202xx.c
I tried to disable the DMA hack for PDC20262. It will generate the same error messages as shown above. I end up making another path disable the DMA support for LBA48. Now everything looks fine now.
>How-To-Repeat:
Pug in a Ultra 66 and put a >137GB hard drive on it. Do a "dd if=/dev/zero of=/dev/ad4"
>Fix:
1. Disable ata DMA support.
2. Use the attached patch, which disables 48bit DMA access.
Patch attached with submission follows:
*** sys/dev/ata/ata-chipset.c.orig Wed May 21 21:31:49 2008
--- sys/dev/ata/ata-chipset.c Thu May 22 00:23:25 2008
***************
*** 3264,3273 ****
--- 3264,3279 ----
/* FALLTHROUGH */
case PROLD:
/* enable burst mode */
ATA_OUTB(ctlr->r_res1, 0x1f, ATA_INB(ctlr->r_res1, 0x1f) | 0x01);
+
+ if (ctlr->chip->chipid == ATA_PDC20262)
+ device_printf(dev,
+ "using PIO transfers above 137GB as workaround for "
+ "48bit DMA access bug, expect reduced performance\n");
+
ctlr->allocate = ata_promise_allocate;
ctlr->setmode = ata_promise_setmode;
return 0;
case PRTX:
***************
*** 3387,3401 ****
}
static int
ata_promise_allocate(device_t dev)
{
struct ata_channel *ch = device_get_softc(dev);
!
if (ata_pci_allocate(dev))
return ENXIO;
ch->hw.status = ata_promise_status;
return 0;
}
static int
--- 3393,3411 ----
}
static int
ata_promise_allocate(device_t dev)
{
+ struct ata_pci_controller *ctlr = device_get_softc(device_get_parent(dev));
struct ata_channel *ch = device_get_softc(dev);
!
if (ata_pci_allocate(dev))
return ENXIO;
+ if (ctlr->chip->chipid == ATA_PDC20262)
+ ch->flags |= ATA_NO_48BIT_DMA;
+
ch->hw.status = ata_promise_status;
return 0;
}
static int
>Release-Note:
>Audit-Trail:
>Unformatted:
More information about the freebsd-bugs
mailing list