Hallo GRUB and NETBOOT/ETHERBOOT hackers !
In my oppinion it is not a good design to have the LOM as last and not
as first device. But we have to live with this.
One idea to solve the problem is such special situations, it can be
a good idea to have a compiler otion to pass pci specific data like
slot number, device number and function number to select that card,
which should be used. This concept violates the concept having auto-
probing, but this is a special situation.
Another possibility is to use different network cards from the LOM.
So I have to enable only the LOM nic in GRUB and so this is the only
used adapter.
I have not checked the neboot code, if it is easy possible to send
bootp/dhacp requests to all interfaces.
With friendly regards
Christoph Plattner
OKUJI Yoshinori wrote:
>
> I forward this message to netboot, since the master code is in
> Etherboot. Please tell us how you feel...
>
> ------------------------------------------------------------------------
>
> Subject: [PATCH] Kludge for Intel's PXE scan order
> Resent-Date: Thu, 8 Jun 2000 22:42:36 -0400
> Resent-From: bug-grub@gnu.org
> Date: Thu, 8 Jun 2000 19:14:09 -0700
> From: Chip Salzenberg <chip@valinux.com>
> To: OKUJI Yoshinori <okuji@kuicr.kyoto-u.ac.jp>
> CC: bug-grub@gnu.org
>
> (Warning: This patch is ugly. But the issue it addresses is important.)
>
> Many of the systems with which we will be using GRUB have multiple
> Ethernet adapters. As it happens, all of them are Intel EEPro100s
> with Intel's PXE BIOS for netbooting. Unfortunately, PXE insists on
> always preferring the LOM (LAN On Motherboard), if any; it is
> unable/unwilling to detect which adapter is actually connected. Even
> worse, Intel's LOMs almost always show up *last* in a PCI scan, so
> GRUB never uses it.
>
> The upshot of this is that, on a multi-adapter system, PXE insists on
> loading from the motherboard port, while GRUB insists on using the
> non-motherboard port. So you can't netboot without two cables.
>
> The below patch is the workaround I will use unless/until a better
> solution is found. It makes a special case of Intel adapters and
> diskless booting; in that case, it effectively reverses the scan.
>
> A much better solution, IMO, would be to give GRUB the ability to send
> DHCP requests out on *all* adapters *simultaneously*. The Linux
> kernel does this for its boot-time DHCP support, and it's a wonderful
> way to prevent headaches for users.
>
> PS: The variable "pci_ioaddr" in pci.c seemed unused, so I removed it.
>
> 2000-06-08 Chip Salzenberg <chip@valinux.com>
>
> Adjust scan order for Intel adapters under diskless boot.
> * netboot/pci.h (eth_pci_prefer_last): Declare new variable.
> * netboot/pci.c (eth_pci_prefer_last): Define it, default -1.
> (scan_bus): Compare it to each adapter's vendor; if it matches,
> then keep scanning, so the last one found will be used.
> Also tweak debugging messages.
> (pci_ioaddr): Eliminate unused variable.
> * stage2/common.c (setup_diskless_environment): Set
> eth_pci_prefer_last to PCI_VENDOR_ID_INTEL.
>
> Index: netboot/pci.h
> diff -u -2 -r1.6 pci.h
> --- netboot/pci.h 2000/05/17 13:55:18 1.6
> +++ netboot/pci.h 2000/06/09 01:55:48
> @@ -146,4 +146,6 @@
> };
>
> +extern unsigned short eth_pci_prefer_last;
> +
> extern void eth_pci_init(struct pci_device *);
>
>
> Index: netboot/pci.c
> diff -u -2 -r1.5 pci.c
> --- netboot/pci.c 2000/05/31 02:15:13 1.5
> +++ netboot/pci.c 2000/06/09 01:55:48
> @@ -23,6 +23,7 @@
>
> /*#define DEBUG 1*/
> -static unsigned int pci_ioaddr = 0;
>
> +unsigned short eth_pci_prefer_last = 0xFFFF; /* no such vendor */
> +
> #ifdef CONFIG_PCI_DIRECT
> #define PCIBIOS_SUCCESSFUL 0x00
> @@ -406,5 +407,4 @@
> int i, reg;
>
> - pci_ioaddr = 0;
> buses=1;
> for (bus = 0; bus < buses; ++bus) {
> @@ -430,5 +430,5 @@
>
> #if DEBUG
> - printf("bus %x, function %x, vendor %x, device %x\n",
> + printf("bus %x, func %x: vendor %x, device %x\n",
> bus, devfn, vendor, device);
> #endif
> @@ -437,4 +437,8 @@
> || device != pcidev[i].dev_id)
> continue;
> +#if DEBUG
> + printf("bus %x, func %x: '%s'\n",
> + bus, devfn, pcidev[i].name);
> +#endif
> for (reg = PCI_BASE_ADDRESS_0; reg <= PCI_BASE_ADDRESS_5; reg += 4) {
> pcibios_read_config_dword(bus, devfn, reg, &ioaddr);
> @@ -453,13 +457,18 @@
> pcidev[i].name, ioaddr, romaddr);
> /* Take the first one or the one that matches in boot ROM address */
> - if (pci_ioaddr == 0 || romaddr == ((unsigned long) rom.rom_segment << 4)) {
> - pcidev[i].membase = membase;
> - pcidev[i].ioaddr = ioaddr;
> - pcidev[i].devfn = devfn;
> + pcidev[i].membase = membase;
> + pcidev[i].ioaddr = ioaddr;
> + pcidev[i].devfn = devfn;
> +
> + /* Should we prefer the last eth? */
> + if (vendor == eth_pci_prefer_last)
> + goto next_devfn;
>
> - return;
> - }
> + /* No, first one is good enough */
> + return;
> }
> }
> +
> + next_devfn: ;
> }
> }
>
> Index: stage2/common.c
> diff -u -2 -r1.12 common.c
> --- stage2/common.c 2000/05/29 16:11:08 1.12
> +++ stage2/common.c 2000/06/09 01:55:48
> @@ -24,4 +24,5 @@
> #ifdef SUPPORT_DISKLESS
> # include <etherboot.h>
> +# include <pci.h>
> #endif
>
> @@ -136,4 +137,12 @@
> setup_diskless_environment (void)
> {
> + /*
> + * Intel's PXE prefers the LOM (LAN on motherboard).
> + * If we booted with PXE, then that must be the one connected.
> + * As it happens, the LOM always shows up last in sequence.
> + * So preferring the last Intel adapter found is a good strategy.
> + */
> + eth_pci_prefer_last = PCI_VENDOR_ID_INTEL;
> +
> /* For now, there is no difference between BOOTP and DHCP in GRUB. */
> if (! bootp ())
>
> --
> Chip Salzenberg - a.k.a. - <chip@valinux.com>
> "I wanted to play hopscotch with the impenetrable mystery of existence,
> but he stepped in a wormhole and had to go in early." // MST3K
--
+--------V--------+ Christoph.Plattner@alcatel.at
| A L C A T E L | -----------------------------
+-----------------+ Phone: +43 1 27722 3706
T A S Fax: +43 1 27722 3955
===========================================================================
This Mail was sent to netboot mailing list by:
Christoph Plattner <christoph.plattner@alcatel.at>
To get help about this list, send a mail with 'help' as the only string in
it's body to majordomo@baghira.han.de. If you have problems with this list,
send a mail to netboot-owner@baghira.han.de.
For requests or suggestions regarding this mailing list archive please write to netboot@gkminix.han.de.