Re: MCE boot crash in qemu

From: Ingo Molnar
Date: Wed Jun 17 2009 - 02:57:48 EST



* Pekka Enberg <penberg@xxxxxxxxxxxxxx> wrote:

> On Mon, 2009-06-15 at 16:22 +0300, Pekka Enberg wrote:
> > On Mon, 2009-06-15 at 14:52 +0200, Andi Kleen wrote:
> > > x86: mce: Handle banks == 0 case in K7 quirk
> > >
> > > This happens on QEMU which reports MCA capability, but no banks.
> > > Without this patch there is a buffer overrun and boot ops because the code
> > > would try to initialize the 0 element of a zero length kmalloc()
> > > buffer.
> > >
> > > Signed-off-by: Andi Kleen <ak@xxxxxxxxxxxxxxx>
> >
> > This fixes the bug for me!
> >
> > Tested-by: Pekka Enberg <penberg@xxxxxxxxxxxxxx>
>
> Ingo, I hit this again in my testing after rebasing to
> linus/master so I really would like this in mainline.

yep, i've tidied up the changelog and have committed it to
x86/urgent.

But the bank[] code is quirky and butt-ugly and that needs to be
cleaned up - it's no wonder that bugs like this slip in.

- There's zero description about the hw model it represents
and how it relates to the bank[] array - what do the banks mean,
how are they organized.

- It's full of magic constants and implicitly-assumed size
calculations with little explanation and little extensibility:

...
if (c->x86 == 15 && banks > 4) {
/*
* disable GART TBL walk error reporting, which
* trips off incorrectly with the IOMMU & 3ware
* & Cerberus:
*/
clear_bit(10, (unsigned long *)&bank[4]);
}
...
bank = kmalloc(banks * sizeof(u64), GFP_KERNEL);
...
memset(bank, 0xff, banks * sizeof(u64));
...

- There's lots of bitmaps, arrays, flags interacting, creating a
maze of logic.

Instead of this messy code, the proper approach is to introduce an
abstract data structure representing the attributes of an MCE bank
register:

struct mce_bank_register {
int enabled;
int polled;
int dont_init;
int msr_idx;
};

( There's lots of other structural problems with the MCE code too -
but now that it's unified lets first fix the most obvious ones... )

Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/