Re: acpi: bad write: (was: Re: My snd_ich working well)

On 4/5/06, Nate Lawson <nate_at_root.org> wrote:
> Ben Kaduk wrote:
> > On 4/4/06, Nate Lawson <nate_at_root.org> wrote:
> >> /*
> >>   * Some BIOS vendors use AML to read/write directly to IO space.  This
> >>   * can cause a problem if such accesses interfere with the OS's access to
> >>   * the same ports.  Windows XP and newer systems block accesses to certain
> >>   * IO ports.  We print a message or block accesses based on a tunable.
> >>   */
> >> static int illegal_bios_ports[] = {
> >>         0x000, 0x00f,   /* DMA controller 1 */
> >>         0x020, 0x021,   /* PIC */
> >>         0x040, 0x043,   /* Timer 1 */
> >>         0x048, 0x04b,   /* Timer 2 failsafe */
> >>         0x070, 0x071,   /* CMOS and RTC */
> >>         0x074, 0x076,   /* Extended CMOS */
> >>         0x081, 0x083,   /* DMA1 page registers */
> >>         0x087, 0x087,   /* DMA1 ch0 low page */
> >>         0x089, 0x08b,   /* DMA2 ch2 (0x89), ch3 low page (0x8a, 0x8b) */
> >>         0x08f, 0x091,   /* DMA2 low page refresh (0x8f) */
> >>                         /* Arb ctrl port, card select feedback (0x90, 0x91) */
> >>         0x093, 0x094,   /* System board setup */
> >>         0x096, 0x097,   /* POS channel select */
> >>         0x0a0, 0x0a1,   /* PIC (cascaded) */
> >>         0x0c0, 0x0df,   /* ISA DMA */
> >>         0x4d0, 0x4d1,   /* PIC ELCR (edge/level control) */
> >>         0xcf8, 0xcff,   /* PCI config space. Microsoft adds 0xd00 also but
> >>                            that seems incorrect. */
> >>         -1, -1
> >> };
> >>
> >
> > Hi Nate,
> >
> > As posted earlier, I'm getting these acpi: bad write
> > messages spamming my console, with port 0x086 instead of Angka's
> > 0x073.  I don't see 0x086 in the above list, though, so I'm a bit
> > confused.
>
> This message would have been triggered by the off-by-one error before my
> fix (86 is just below 87, which is on the list).
>
> The check is now:
> if ((addr >= port[0] && addr <= port[1]) ||
>      (addr < port[0] && addr + (width / 8) > port[0]))
>
> Which gives (with addr = 86 and width = 8):
> 86 >= 87 ...  FALSE
> 86 < 87 (TRUE) && 86 + 1 > 87 (FALSE) ... FALSE
>
> You can try it yourself with (play around with addr and width):
>
> main()
> {
> int addr = 0x86;
> int width = 8;
> int port[2] = { 0x87, 0x87 };
>
> if ((addr >= port[0] && addr <= port[1]) ||
>      (addr < port[0] && addr + (width / 8) > port[0]))
>          printf("BUGGY BUGGY BIOS\n");
> else
>          printf("KRAD!\n");
> }
>
> So the code is correct.  Please be sure you have the right version that
> matches this code snippet above and recompile your acpi module (or
> entire kernel) and reinstall it.
>
> > I have revision 1.120 of src/sys/dev/acpica/Osd/OsdHardware.c
>
> The correct version is 1.20.
>
> Thanks,
> --
> Nate
>

Well, I went back and checked my build log, and it seems that my
sleepy fingers typed "buildkernel" instead of "kernel", so I'm running
a kernel from april 3, (with the old OsdHardware.c), instead of the
new one.
I currently have v. 1.20 (1.120 was a typo), so when I can get at the
machine later tonight, I will install the new kernel and reboot, and
report back if the problem remains (but I highly doubt that it will).

Terribly sorry about the noise -- I guess I shouldn't be so excited
about the chance to report a bug and get some sleep before sending
something off.

-Ben Kaduk