Re: lockup booting 8.0-CURRENT-200811 snap image

So the real fix is that we need to disable memory and I/O decoding int  
the PCI command register when messing with the BARs. One thing to be  
careful with is we can't do any console I/O (i.e. printfs) while the  
BAR is disabled. I will come up with a proper patch when I get back to  
a real computer.

-- 
John Baldwin

On Dec 13, 2008, at 10:51 PM, Tor Egge <Tor.Egge_at_cvsup.no.freebsd.org>  
wrote:

>> Well, we probably need to come up with a better way to determine  
>> which
>> machines to actually use it on.  It's a shame that MCFG is  
>> apparently busted
>> on so many machines (either that or we are not doing something  
>> correctly).
>
> pci_add_map() writes 0xffffffff to the map register in order to  
> probe for
> writable bits.  This had the immediate side effect of hanging my  
> machine.
> Since I didn't know about the "Add hw.pci.mcfg=0 to /boot/ 
> loader.conf" at that
> time, I started using the enclosed kludgy patch (specific for Intel  
> 82Q33
> hardware).  Only the GMADDR map register was a problem.
>
> It looks like the graphics memory aperture was mapped on top of the  
> memory
> mapped pci config registers, thus only IO access to the pci config  
> registers
> worked to undo the supposedly temporary change.
>
> - Tor Egge
> Index: sys/conf/options
> ===================================================================
> RCS file: /home/ncvs/src/sys/conf/options,v
> retrieving revision 1.651
> diff -u -r1.651 options
> --- sys/conf/options    11 Dec 2008 16:13:17 -0000    1.651
> +++ sys/conf/options    12 Dec 2008 20:00:11 -0000
> _at__at_ -786,3 +786,7 _at__at_
> # Virtualize the network stack
> VIMAGE            opt_global.h
> VIMAGE_GLOBALS        opt_global.h
> +
> +# Kludge to work around hangs during probing of Intel 82Q33 PCI  
> devices
> +# (vendor 0x8086, device 0x29d2).
> +PCI_INTEL_82Q33_KLUDGE    opt_kludge.h
> Index: sys/dev/pci/pci.c
> ===================================================================
> RCS file: /home/ncvs/src/sys/dev/pci/pci.c,v
> retrieving revision 1.366
> diff -u -r1.366 pci.c
> --- sys/dev/pci/pci.c    13 Nov 2008 19:57:33 -0000    1.366
> +++ sys/dev/pci/pci.c    14 Nov 2008 18:27:31 -0000
> _at__at_ -30,6 +30,7 _at__at_
> __FBSDID("$FreeBSD: src/sys/dev/pci/pci.c,v 1.366 2008/11/13  
> 19:57:33 mav Exp $");
>
> #include "opt_bus.h"
> +#include "opt_kludge.h"
>
> #include <sys/param.h>
> #include <sys/systm.h>
> _at__at_ -2289,11 +2290,52 _at__at_
>    int type;
>    int barlen;
>    struct resource *res;
> +#ifdef PCI_INTEL_82Q33_KLUDGE
> +    struct pci_devinfo *dinfo;
> +    pcicfgregs *cfg;
> +    uint32_t msacval;
> +#endif
>
>    map = PCIB_READ_CONFIG(pcib, b, s, f, reg, 4);
> +#ifdef PCI_INTEL_82Q33_KLUDGE
> +    dinfo = device_get_ivars(dev);
> +    cfg = &dinfo->cfg;
> +    /*
> +     * The test write to probe for writable address bits hangs
> +     * some machines.  Kludge around this by faking the test based
> +     * on the MSAC register at offset 0x62.  Intel doc is slightly
> +     * inconsistent about which bits in MSAC controls the aperture
> +     * size.
> +     */
> +    if (cfg->vendor == 0x8086 && cfg->device == 0x29d2 && reg ==  
> 24) {
> +        printf("Kludge: "
> +               "faking test to avoid hang for 82Q33 GMADDR config 
> \n");
> +        msacval = PCIB_READ_CONFIG(pcib, b, s, f, 0x62, 1);
> +        printf("82Q33 MSAC (offset 0x62) val is 0x%02x\n", msacval);
> +        switch ((msacval >> 1) & 3) {
> +        case 0:
> +            testval = 0xf8000008u;
> +            break;
> +        case 1:
> +            testval = 0xf0000008u;
> +            break;
> +        case 3:
> +            testval = 0xe0000008u;
> +            break;
> +        default:
> +            printf("BAD MSAC val\n");
> +            testval = 0xe0000008u;
> +            break;
> +        }
> +        goto kludgedone;
> +    }
> +#endif
>    PCIB_WRITE_CONFIG(pcib, b, s, f, reg, 0xffffffff, 4);
>    testval = PCIB_READ_CONFIG(pcib, b, s, f, reg, 4);
>    PCIB_WRITE_CONFIG(pcib, b, s, f, reg, map, 4);
> +#ifdef PCI_INTEL_82Q33_KLUDGE
> +kludgedone:
> +#endif
>
>    if (PCI_BAR_MEM(map))
>        type = SYS_RES_MEMORY;