Re: [PATCH -mm] MMCONFIG: Reject a broken MCFG tables on Asus etc

[Olivier Galibert <galibert@pobox.com>]

On Sun, Jan 14, 2007 at 06:27:18AM +0900, OGAWA Hirofumi wrote:
> This rejects a broken MCFG tables on Asus etc.
> Arjan and Andi suggest this.

And I agree completely with the principle.  If you don't know the
chipset on a first-name basis, trash the MCFG unless it's squeaky
clean (or you don't have a choice).


> +static void __init pci_mmcfg_reject_broken(void)
> +{
> +	struct acpi_table_mcfg_config *cfg = &pci_mmcfg_config[0];
> +
> +	/*
> +	 * Handle more broken MCFG tables on Asus etc.
> +	 * They only contain a single entry for bus 0-0.
> +	 */
> +	if (pci_mmcfg_config_num == 1 &&
> +	    cfg->pci_segment_group_number == 0 &&
> +	    (cfg->start_bus_number | cfg->end_bus_number) == 0) {
> +		kfree(pci_mmcfg_config);
> +		pci_mmcfg_config = NULL;
> +		pci_mmcfg_config_num = 0;
> +
> +		printk(KERN_ERR "PCI: start and end of bus number is 0. "
> +		       "Rejected as broken MCFG.");
> +	}
> +}
> +

If you're going to do a MCFG validation function, and I don't have a
problem with that, you should put the e820 test in it too.

  OG.