Re: Regression: 2.6.34 boot fails on E5405 system, bisected: de08e2c26

[Jesse Barnes <jbarnes@virtuousgeek.org>]

On Tue, 13 Jul 2010 18:17:18 -0700
Ben Greear <greearb@candelatech.com> wrote:

> On 07/13/2010 05:36 PM, Ben Greear wrote:
> > We're seeing boot failures on multiple machines, running FC8 and
> > F11. I bisected on an FC8 32-bit system. Newer hardware works,
> > but these older ones do not.
> >
> > A console log of the hang is found later in this email.
> >
> > Please let me know if you would like any additional information,
> > and I will be happy to test patches.
> >
> > The same failure happens in 2.6.34.1, so the fix does not appear to
> > be in the stable tree yet.
> 
> 
> I added some printks to the offending code.  It seems the problem
> is that the fixed_bar_cap method in arch/x86/pci/mrst.c loops forever:
> 
> # Endless loop of this spewing to console...
> 
> pcie_cap: 268435456Checking vendor..
> pos after shift: 256
> Before read..
> pcie_cap: 268435456Checking vendor..
> pos after shift: 256
> Before read..
> pcie_cap: 268435456Checking vendor..
> pos after shift: 256
> Before read..
> pcie_cap: 268435456Checking vendor..
> pos after shift: 256
> Before read..
> pcie_cap: 268435456Checking vendor..
> pos after shift: 256
> Before read..
> pcie_cap: 268435456Checking vendor..
> 
> 
> static int fixed_bar_cap(struct pci_bus *bus, unsigned int devfn)
> {
> 	int pos;
> 	u32 pcie_cap = 0, cap_data;
> 	printk("fixed_bar_cap, bus: %p  devfn: %u\n", bus, devfn);
> 	pos = PCIE_CAP_OFFSET;
> 	while (pos) {
> 		printk("Before read..\n");
> 		if (raw_pci_ext_ops->read(pci_domain_nr(bus), bus->number,
> 					  devfn, pos, 4, &pcie_cap))
> 			return 0;
> 		printk("pcie_cap: %u", pcie_cap);
> 		
> 		if (pcie_cap == 0xffffffff)
> 			return 0;
> 
> 		printk("Checking vendor..\n");
> 		if (PCI_EXT_CAP_ID(pcie_cap) == PCI_EXT_CAP_ID_VNDR) {
> 			printk("reading domain_nr\n");
> 			raw_pci_ext_ops->read(pci_domain_nr(bus), bus->number,
> 					      devfn, pos + 4, 4, &cap_data);
> 			printk("cap_data: %u\n", cap_data);
> 			if ((cap_data & 0xffff) == PCIE_VNDR_CAP_ID_FIXED_BAR)
> 				return pos;
> 		}
> 
> 		pos = pcie_cap >> 20;
> 		printk("pos after shift: %i\n", pos);
> 	}
> 
> 	printk("Returning from fixed_bar_cap\n");
> 	return 0;
> }
> 
> 

I thought a related bug was fixed already; the code should be returning
all zeros for non-existent BAR reads.

-- 
Jesse Barnes, Intel Open Source Technology Center