Re: 3.12 to 3.13 boot regression bisected - still applies to 3.16

[Bruno PrÃmont]

On Tue, 5 Aug 2014 09:45:42 +0100 Matt Fleming wrote:
> On Tue, 05 Aug, at 10:02:42AM, Bruno PrÃmont wrote:
> > 
> > I tried in setup_arch(), but system still keeps rebooting.
> > 
> > Working backwards I got to x86_64_start_kernel() in
> > arch/x86/kernel/head64.c but system is still rebooting.
>  
> Thanks for doing this. I'm sure it was a major PITA ;-)

Fortunately building the kernels on a separate system and being
able to build a dozen of them to try from efi shell it's survivable.

> > Not sure what happens before x86_64_start_kernel() is called, it seems
> > to be called from ASM code in arch/x86/kernel/head_64.S.
>  
> Yep. Roughly the code flow goes like this (chronologically),
> 
>   efi_pe_entry()	[arch/x86/boot/compressed/head_64.S]
>     efi_main()		[arch/x86/boot/compressed/eboot.c]

I get at least to just before
  status = efi_call_early(exit_boot_services, handle, key);
in eboot.c on line 1310. A efi_printk inserted there is displayed.

>   startup_64		[arch/x86/kernel/head_64.S]
>   secondary_startup64	[arch/x86/kernel/head_64.S]
>   x86_64_start_kernel()	[arch/x86/kernel/head64.c]
> 
> > > Meanwhile I'm going to go and stare at the EFI boot stub code and
> > > instrument OVMF to check for more memory corruption bugs like the one
> > > Michael found in commit c7fb93ec51d4 ("x86/efi: Include a .bss section
> > > within the PE/COFF headers").
> > 
> > If there are places between exit_boot() in
> > arch/x86/boot/compressed/eboot.c and x86_64_start_kernel() where I
> > should include such loops, please tell!
> 
> I guess we need to verify efi_main() actually exits correctly. So a
> while (1); loop at the end of that function would be useful.
> 
> Assuming that does actually hang, you get the fun of rummaging around in
> the early assembly code, where you can use something like this,
> 
> bruno:
> 	hlt
> 	jmp bruno
> 
> to try and force a hang.

Will spin a few attempts and see what I get.

> Could you also attach your .config? In particular I'm wondering whether
> you've got CONFIG_RELOCATBLE enabled.

Config attached (gzipped). CONFIG_RELOCATBLE is not enabled.

Bruno

Attachment: config.gz
Description: application/gzip