Re: today's linux-next fails to boot

[Takashi Iwai]

At Tue, 15 Jul 2008 04:11:26 +0200,
Frédéric Weisbecker wrote:
> 
> Török Edwin a écrit :
> > Hi,
> >
> > Today's linux-next tree (commit
> > 93847083e4791567931bd17c039cc35881cdad29) fails to boot:
> > [built with gcc-4.2.4-3]
> >
> > BUG: Int 14: CR2 b0049dea
> >      EDI 00000082 ESI 00000000 EBP c059be88 ESP c059be5c
> >      EBX f000ec62 EDX 0000000e ECX c0595480 EAX f000ec62
> >      err 00000000 EIP c0181ca0  CS 00000060 flg 00010082
> > Stack:   00000040 c06a2ba0 000080d0 c0595480 c0000f19c c000f180 c0581120
> > c059bea8
> >          c02bf19b 00000000 00000080 c059beb8 c0000f194 c000f180 0000000a
> > c059beb8
> >          c03a1059 00000000 00000000 c059bed8 c05c4c7c  0009efff 00000000
> > c04f4df4
> >
> > I get this as soon as I boot from grub2, strangely the error message is
> > at the bottom of the screen, and I can't see the full message (scrolling
> > won't work).
> >
> > The last kernel I built & booted was 2.6.26-rc8 from Linus's tree. I
> > will try to built&boot 2.6.26-rc9, and then bisect.
> >
> > This happens on 32-bit Dell Inspiron 6400 (Intel Core Duo T2300 @1.66
> > Ghz CPU),  Intel ICH-7 chipset, and a seagate SATA drive. 
> > I will provide  full hardware details once I bisected the problem.
> >
> > Meanwhile, if somebody has an idea as to what is wrong?
> >
> > Best regards,
> > --Edwin
> > --
> > 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/
> >   
> 
> I got the same problem on an x86-PC and after looking at the stack, the 
> problem comes from firmware_map_add_early() in drivers/firmware/memmap.c
> 
> The backtrace is the following:
> 
> kzalloc()
> verify_dynamic_kobject_allocation()
> kobject_init()
> firmware_map_add_entry()
> firmware_map_add_early()
> e820_reserve_resources()
> setup_arch() (in x86)
> 
> The problem is that verify_dynamic_kobject_allocation() calls kzalloc 
> assuming that kmem_cache_init() has already been called. But it's not 
> the case and it's too soon to call the kmalloc/kzalloc functions.
> 
> I don't know what is the real problem: the fact that kobject_init is 
> called too soon or verify_dynamic_kobject_allocation() which calls 
> kzalloc without making any assumption about its current context.
> 
> So here is just a patch to temporarily disable 
> verify_dynamic_kobject_allocation() This function just checks the sanity 
> of the code.

Confirmed that this fixes the boot problem on my machine, too.
(It explains why this happens only on x86-32...)

Added Bernhard to Cc.  Maybe we should defer firmware_map_add*()?


thanks,

Takashi


> [2 disable_verify_dyn_kobject.diff <text/x-patch (7bit)>]
> --- a/lib/kobject.c	2008-07-15 02:55:08.000000000 +0200
> +++ b/lib/kobject.c	2008-07-15 04:01:10.000000000 +0200
> @@ -335,7 +335,6 @@
>  		       "object, something is seriously wrong.\n", kobj);
>  		dump_stack();
>  	}
> -	verify_dynamic_kobject_allocation(kobj);
>  
>  	kobject_init_internal(kobj);
>  	kobj->ktype = ktype;
--
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/