Re: r8169 chips on some Intel D945GSEJT boards fail to work after PXE boot

[Simon Farnsworth <simon.farnsworth@onelan.com>]

Francois Romieu wrote:
> Simon Farnsworth <simon.farnsworth@onelan.com> :
> [...]
>> Some boards are good, and just work, whether I boot via PXE or boot from
>> the local disk; dmesg.working and lspci.working are from a good board.
>>
>> Some boards are bad; they work fine if I boot from local disk (including
>> network), but the kernel cannot detect link, or send or receive data if
>> I PXE boot. dmesg.broken and lspci.broken are from a bad board.
> 
> No cunning theroy in sight but does reducing the amount of memory on a
> bad board from 1 Go to 512 Mo turn it into a good one ?
> 
We've tried this, and we've tried 2GB and 1GB modules; the failure to
boot sticks with the board, not with the memory module. On my most
recent attempt, the failing board isn't showing a correctable error
status, so I've not yet tried your patch, on the assumption that it just
clears the error status.

Is my assumption wrong? If not, is there anything else I can do that
would help you diagnose this?

> The failing board exhibits a correctable error status bit. Clearing it
> is the least we can do.
> 
> diff --git a/drivers/net/r8169.c b/drivers/net/r8169.c
> index 50c6a3c..79bc4ab 100644
> --- a/drivers/net/r8169.c
> +++ b/drivers/net/r8169.c
> @@ -2200,6 +2200,11 @@ rtl8169_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
>  	tp->pcie_cap = pci_find_capability(pdev, PCI_CAP_ID_EXP);
>  	if (!tp->pcie_cap && netif_msg_probe(tp))
>  		dev_info(&pdev->dev, "no PCI Express capability\n");
> +	else {
> +		pci_write_config_word(pdev, tp->pcie_cap + PCI_EXP_DEVSTA,
> +				      PCI_EXP_DEVSTA_CED | PCI_EXP_DEVSTA_NFED |
> +				      PCI_EXP_DEVSTA_FED | PCI_EXP_DEVSTA_URD);
> +	}
>  
>  	RTL_W16(IntrMask, 0x0000);
>  

-- 
Simon Farnsworth