Re: kern/36838 (pci bus is disabled on ACPI kernel on Fujitsu FMV LOOX P70WN.)

[Ryo ONODERA <ryo_on%yk.rim.or.jp@localhost>]

My report is not filed in GNATS.
I will send again.

From: Ryo ONODERA <ryo_on%yk.rim.or.jp@localhost>
Subject: Re: kern/36838 (pci bus is disabled on ACPI kernel on Fujitsu FMV LOOX 
P70WN.)
Date: Tue, 02 Mar 2010 23:04:54 +0900 (JST)

> Hi,
> 
> I applied following 2 patches and investigated what disable pci2.
> 
> First, for pci_probe_device() in src/sys/dev/pci/pci.c .
> 
> --- pci.c.orig        2010-03-01 18:51:36.000000000 +0900
> +++ pci.c     2010-03-02 21:28:47.000000000 +0900
> @@ -315,10 +315,13 @@
>        * as appropriate.
>        */
>       pa.pa_flags = sc->sc_flags;
> +     aprint_verbose("Before: csr = 0x%08x, pa.pa_flags = 0x%04x, pa.pa_bus = 
> 0x%04x\n", csr, pa.pa_flags, pa.pa_bus);
>       if ((csr & PCI_COMMAND_IO_ENABLE) == 0)
>               pa.pa_flags &= ~PCI_FLAGS_IO_ENABLED;
> +     aprint_verbose("After:IO: csr = 0x%08x, pa.pa_flags = 0x%04x, pa.pa_bus 
> = 0x%04x\n", csr, pa.pa_flags, pa.pa_bus);
>       if ((csr & PCI_COMMAND_MEM_ENABLE) == 0)
>               pa.pa_flags &= ~PCI_FLAGS_MEM_ENABLED;
> +     aprint_verbose("After:MEM: csr = 0x%08x, pa.pa_flags = 0x%04x, 
> pa.pa_bus = 0x%04x\n", csr, pa.pa_flags, pa.pa_bus);
>  
>       /*
>        * If the cache line size is not configured, then
> 
> 
> 
> 
> Second, for pci_conf_read() in src/sys/arch/x86/pci/pci_machdep.c .
> 
> --- pci_machdep.c.orig        2010-03-01 18:53:17.000000000 +0900
> +++ pci_machdep.c     2010-03-02 21:58:51.000000000 +0900
> @@ -409,8 +409,10 @@
>  
>       KASSERT((reg & 0x3) == 0);
>  
> -     if (pc != NULL && pc->pc_conf_read != NULL)
> +     if (pc != NULL && pc->pc_conf_read != NULL) {
> +             aprint_verbose("RETURN from NULL check\n");
>               return (*pc->pc_conf_read)(pc, tag, reg);
> +     }
>  
>  #if defined(__i386__) && defined(XBOX)
>       if (arch_i386_is_xbox) {
> 
> 
> 
> 
> I attach two dmesg, one is enabled ACPI (dmesg-csr-ACPI), and another is 
> disabled ACPI (dmesg-csr-NOACPI).
> 
> In ACPI-enabled dmesg, pci_conf_read() returns 0x00100000.
> But in ACPI-disabled dmesg, pci_conf_read() returns 0x00100007.
> 
> Line 291, csr = pci_conf_read(pc, tag, PCI_COMMAND_STATUS_REG); exists in 
> pci_probe_device()
> in src/sys/dev/pci/pci.c, and the code
> 
>         /*
>          * Set up memory, I/O enable, and PCI command flags
>          * as appropriate.
>          */
>         pa.pa_flags = sc->sc_flags;
>         if ((csr & PCI_COMMAND_IO_ENABLE) == 0)
>                 pa.pa_flags &= ~PCI_FLAGS_IO_ENABLED;
>         if ((csr & PCI_COMMAND_MEM_ENABLE) == 0)
>                 pa.pa_flags &= ~PCI_FLAGS_MEM_ENABLED;
> 
> disables pci2 on my PC, I think.
> (0x00100000 & 0x01 is zero and 0x00100000 & 0x02 is zero.)
> 
> But, the core of pci_conf_read() in src/sys/arch/x86/pci/pci_machdep.c is 
> written in assembly language.
> I cannot understand what is done.
> 
> Can anyone analyze it? I will offer all data about this issue.
> 
> 
> P.S.
> It seems that port-i386/42676 (http://gnats.netbsd.org/42676) is similar to 
> my problem.
> In port-i386/42676, the machine uses same chipset and pci1 is disabled.
> 
> --
> Ryo ONODERA // ryo_on%yk.rim.or.jp@localhost
> PGP fingerprint = 82A2 DC91 76E0 A10A 8ABB  FD1B F404 27FA C7D1 15F3