It seems Bruce R. Montague wrote:
>
I'll look into it, thanks for reporting!
> Hi, the current -current ata driver panics at boot
> when using the CS5530 (National GX1, ex-"cyrix").
> This driver worked in the past on -current, likely
> up until the major rework that appears to be
> underway as of 20-Feb-2003 (that is, the creation
> of "ata-chipset.c", etc.)
>
> Routine "ata_cyrix_setmode()" in "ata-chipset.c"
> appears to be assuming that "channel->dma" has a
> valid pointer to a "struct ata_dma_funcs", and
> that "channel->r_bmio" is a valid bus resource
> id. This is not the case, both "channel->dma" and
> "channel->r_bmio" (bus master I/O supported) are
> 0, which will result in panics. The first use (and
> panic) occurs at:
>
> atadev->channel->dma->alignment = 16;
>
> "ata_cyrix_setmode+0x8b: movl $0x10,0x20(%eax)"
> on my build (%eax is 0).
>
> These panics occur regardless of the setting of
> TUNABLE_INIT() "ata_dma".
>
> Routine "ata_dmainit()", which mallocs the "struct
> ata_dma_funcs" is (likely correctly) never called.
> If required due to DMA support, it is allocated
> during the driver probe via "ctlr->dmainit(ch)"
> in "ata_pcisub_probe()" in "ata-pci.c".
>
> To make the system "come up", I replaced
> "ata_cyrix_setmode()" with the following:
>
> static void
> ata_cyrix_setmode(struct ata_device *atadev, int mode)
> {
> int error;
>
> mode = ata_limit_mode(atadev, mode, ATA_UDMA2);
>
> error = ata_command(atadev, ATA_C_SETFEATURES, 0, mode,
> ATA_C_F_SETXFER, ATA_WAIT_READY);
>
> if (bootverbose)
> ata_prtdev(atadev, "%s setting %s on Cyrix chip\n",
> (error) ? "failed" : "success", ata_mode2str(mode));
>
> atadev->mode = mode;
> }
>
>
> This seems to work, I am using the system without
> apparent problems, but it is strictly a "by guess
> and by god" fix - I haven't studied or understood
> the whole new ata driver scaffolding.
>
> What is (a) correct fix? Is there a better and
> more complete thing envisoned? Is there a tunable
> I don't understand? Or a feature flag? Is the
> current "cyrix" code in transit and untested?
> Should "ata_dma" or "r_bmio" be checked in the
> setmode codepath? Can I help assure this is fixed
> right? I can at least test, if need be.
>
> As a minor question, is the style to allocate
> malloced data structures (such as "ata_dma_funcs")
> in the probe code instead of attach code, as seems
> to be the intent in this driver, and leave permanent
> bus resource allocation until the attach? (in this
> case the "ata_pcisub_probe()" never reaches the
> allocation code because it checks and finds "r_bmio"
> is 0).
>
>
>
>
>
> - bruce
>
> To Unsubscribe: send mail to [EMAIL PROTECTED]
> with "unsubscribe freebsd-current" in the body of the message
>
-Søren
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message
