2016-12-21 13:11:45

by Petr Oros

Subject: [PATCH] x86/efi Fix regression in efi_arch_mem_reserve

Booting on EFI with ESRT table has been stop since commit:
8e80632 efi/esrt: Use efi_mem_reserve() and avoid a kmalloc()

This is caused by this commit:
816e761 efi: Allow drivers to reserve boot services forever

Problem is, that efi_memmap_insert need memory aligned
on EFI_PAGE_SIZE. If memory not aligned, efi_memmap_insert
just return and let efi.memmap in inconsistent state.
This breaking boot.

Tested in my machine, which stop booting
after upgrade to 4.9

Signed-off-by: Petr Oros <[email protected]>
---
arch/x86/platform/efi/quirks.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/platform/efi/quirks.c b/arch/x86/platform/efi/quirks.c
index 10aca63..7678857 100644
--- a/arch/x86/platform/efi/quirks.c
+++ b/arch/x86/platform/efi/quirks.c
@@ -214,7 +214,7 @@ void __init efi_arch_mem_reserve(phys_addr_t addr, u64 size)

new_size = efi.memmap.desc_size * num_entries;

- new_phys = memblock_alloc(new_size, 0);
+ new_phys = memblock_alloc(new_size, EFI_PAGE_SIZE);
if (!new_phys) {
pr_err("Could not allocate boot services memmap\n");
return;
--
2.10.2

2016-12-21 22:30:59

by Matt Fleming

Subject: Re: [PATCH] x86/efi Fix regression in efi_arch_mem_reserve

On Wed, 21 Dec, at 02:11:38PM, Petr Oros wrote:
> Booting on EFI with ESRT table has been stop since commit:
> 8e80632 efi/esrt: Use efi_mem_reserve() and avoid a kmalloc()
>
> This is caused by this commit:
> 816e761 efi: Allow drivers to reserve boot services forever
>
> Problem is, that efi_memmap_insert need memory aligned
> on EFI_PAGE_SIZE. If memory not aligned, efi_memmap_insert
> just return and let efi.memmap in inconsistent state.
> This breaking boot.
>
> Tested in my machine, which stop booting
> after upgrade to 4.9
>
> Signed-off-by: Petr Oros <[email protected]>
> ---
> arch/x86/platform/efi/quirks.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)

Could you provide some more information? Why does efi_memmap_insert()
require this alignment? How does booting "break"? If you see an Oops,
please post it here.

2016-12-22 10:21:53

by Petr Oros

Subject: Re: [PATCH] x86/efi Fix regression in efi_arch_mem_reserve

Matt Fleming píše v Wed 21. 12. 2016 v 22:30 +0000:
> On Wed, 21 Dec, at 02:11:38PM, Petr Oros wrote:
> >
> >   Booting on EFI with ESRT table has been stop since commit:
> >     8e80632 efi/esrt: Use efi_mem_reserve() and avoid a kmalloc()
> >
> >   This is caused by this commit:
> >     816e761 efi: Allow drivers to reserve boot services forever
> >
> >   Problem is, that efi_memmap_insert need memory aligned
> >   on EFI_PAGE_SIZE. If memory not aligned, efi_memmap_insert
> >   just return and let efi.memmap in inconsistent state.
> >   This breaking boot.
> >
> >   Tested in my machine, which stop booting
> >   after upgrade to 4.9
> >
> > Signed-off-by: Petr Oros <[email protected]>
> > ---
> >  arch/x86/platform/efi/quirks.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
>
> Could you provide some more information? Why does efi_memmap_insert()
> require this alignment? How does booting "break"? If you see an Oops,
> please post it here.

Ooops, sorry, please ignore this patch. I overlooked efi_memmap_insert
argument order. I bisected kernel and this patch breaking kernel boot
but from other reason. Provided patch fixed kernel booting issue by
mistake.

-Petr