Re: linux-next: OOPS at bot time

[Michael Ellerman <michael@ellerman.id.au>]

[-- Attachment #1: Type: text/plain, Size: 2133 bytes --]

On Fri, 2010-07-23 at 02:19 +0100, Ben Hutchings wrote:
> On Fri, 2010-07-23 at 10:22 +1000, Stephen Rothwell wrote:
> > Hi all,
> > 
> > My Power7 boot test paniced like this: (next-20100722)
> > 
> > %GQLogic Fibre Channel HBA Driver: 8.03.03-k0
> > qla2xxx 0002:01:00.2: enabling device (0144 -> 0146)
> > qla2xxx 0002:01:00.2: Found an ISP8001, irq 35, iobase 0xd000080080014000
> > ------------[ cut here ]------------
> > kernel BUG at drivers/pci/msi.c:205!
> [...]
> > Call Trace:
> > [c00000000278b270] [c000000000048d9c] .rtas_setup_msi_irqs+0x1d8/0x254 (unreliable)
> > [c00000000278b360] [c00000000002a9cc] .arch_setup_msi_irqs+0x34/0x4c
> > [c00000000278b3e0] [c0000000002fd3fc] .pci_enable_msix+0x49c/0x4ac
> [...]
> > That line number is this:
> > 
> > 	BUG_ON(!(entry->msg.address_hi | entry->msg.address_lo |
> > 		 entry->msg.data));
> > 
> > in read_msi_msg_desc().  That BUG_ON was added by commit
> > 2ca1af9aa3285c6a5f103ed31ad09f7399fc65d7 ("PCI: MSI: Remove unsafe and
> > unnecessary hardware access") from the pci tree.
> 
> I wanted to assert that read_msi_msg_desc() is only used to update
> MSI/MSI-X descriptors that have already been generated by Linux.  It
> looks like you found an exception.
>
> We could make read_msi_msg() fall back to reading from the hardware, but
> I think that what the pSeries code is trying to do - save an MSI message
> generated by firmware - is different from what the other callers want.
> Instead we could add:
> 
> void save_msi_msg(unsigned int irq)
> {
> 	struct irq_desc *desc = irq_to_desc(irq);
> 	struct msi_desc *entry = get_irq_desc_msi(desc);
> 	struct msi_msg *msg = &entry->msg;
> 
> 	/* ...followed by the old implementation of read_msi_msg_desc() */
> }
> 
> Possibly conditional on something like CONFIG_ARCH_NEEDS_SAVE_MSI_MSG.

Maybe.

But then you end up with read_msi_msg(), which doesn't actually read
anything, which I think is confusing. I'd rather read_msi_msg() read the
message, from the device, and we have another routine which returns the
previously saved msg from the msi_desc.

cheers

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 198 bytes --]