Skip to contentSkip to conversation

x86, efi: Only print warning when desc_size is smaller than defined one.

PATCH In review Yinghai Lu 3 participants 4 messages Apr 26, 2011 – Apr 28, 2011

Subsystems (2)
x86 EFI
lore
Share & tools
lore

Yinghai Lu

author
#0 lore raw

PATCH 1/1 x86, efi: Only print warning when desc_size is smaller than defined one.

One system get warning:
Kernel-defined memdesc doesn't match the one from EFI!

UEFI spec allows GetMemoryMap() return bigger desc_size and desc_ver for future extension.
And desc_size is already used to get offset of next memory entry.

Only need to check if desc_size is smaller than defined struct.

Also print the size in warning if it happens.

Signed-off-by Yinghai Lu <[email protected]>
---
 arch/x86/platform/efi/efi.c |    5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
arch/x86/platform/efi/efi.c +3 -2
Index: linux-2.6/arch/x86/platform/efi/efi.c
===================================================================
@@ -448,9 +448,10 @@ void __init efi_init(void)
 		printk(KERN_ERR "Could not map the EFI memory map!\n");
 	memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
 
-	if (memmap.desc_size != sizeof(efi_memory_desc_t))
+	if (memmap.desc_size < sizeof(efi_memory_desc_t))
 		printk(KERN_WARNING
-		  "Kernel-defined memdesc doesn't match the one from EFI!\n");
+		  "Kernel-defined memdesc size is bigger than the one from EFI! %ld > %ld\n",
+			 sizeof(efi_memory_desc_t), memmap.desc_size);
 
 	if (add_efi_memmap)
 		do_add_efi_memmap();
#1 lore raw
Quoted 23 lines · On Tue, 26 Apr 2011 15:42:24 -0700 Yinghai Lu <[email protected]> wrote:

One system get warning:
Kernel-defined memdesc doesn't match the one from EFI!

UEFI spec allows GetMemoryMap() return bigger desc_size and desc_ver for future extension.
And desc_size is already used to get offset of next memory entry.

Only need to check if desc_size is smaller than defined struct.

Also print the size in warning if it happens.

Signed-off-by Yinghai Lu <[email protected]>
---
 arch/x86/platform/efi/efi.c |    5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

Index: linux-2.6/arch/x86/platform/efi/efi.c
===================================================================

arch/x86/platform/efi/efi.c void __init efi_init(void) · line 454

Quoted 9 lines
arch/x86/platform/efi/efi.c +3 -2
@@ -448,9 +448,10 @@ void __init efi_init(void)
 		printk(KERN_ERR "Could not map the EFI memory map!\n");
 	memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
 
-	if (memmap.desc_size != sizeof(efi_memory_desc_t))
+	if (memmap.desc_size < sizeof(efi_memory_desc_t))
 		printk(KERN_WARNING
-		  "Kernel-defined memdesc doesn't match the one from EFI!\n");
+		  "Kernel-defined memdesc size is bigger than the one from EFI! %ld > %ld\n",
+			 sizeof(efi_memory_desc_t), memmap.desc_size);

Does it make sense for this to just be a warning? If sizeof(efi_memory_desec_t) > sizeof(memmap.desc_size) aren't we going to run into some pretty serious problems when we start accessing memory descriptors?

Would it not make more sense for this to be a BUG_ON()? In the unlikely event that the kernel's efi_memory_desc_t is too large I really don't see how we can recover from that.

Signature · 1 line
Matt Fleming, Intel Open Source Technology Center
author
#2 lore raw
Quoted 41 lines · Matt Fleming
Quoted 23 lines · On Tue, 26 Apr 2011 15:42:24 -0700 Yinghai Lu <[email protected]> wrote:

One system get warning:
Kernel-defined memdesc doesn't match the one from EFI!

UEFI spec allows GetMemoryMap() return bigger desc_size and desc_ver for future extension.
And desc_size is already used to get offset of next memory entry.

Only need to check if desc_size is smaller than defined struct.

Also print the size in warning if it happens.

Signed-off-by Yinghai Lu <[email protected]>
---
 arch/x86/platform/efi/efi.c |    5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

Index: linux-2.6/arch/x86/platform/efi/efi.c
===================================================================

arch/x86/platform/efi/efi.c void __init efi_init(void) · line 454

Quoted 9 lines
arch/x86/platform/efi/efi.c +3 -2
@@ -448,9 +448,10 @@ void __init efi_init(void)
 		printk(KERN_ERR "Could not map the EFI memory map!\n");
 	memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
 
-	if (memmap.desc_size != sizeof(efi_memory_desc_t))
+	if (memmap.desc_size < sizeof(efi_memory_desc_t))
 		printk(KERN_WARNING
-		  "Kernel-defined memdesc doesn't match the one from EFI!\n");
+		  "Kernel-defined memdesc size is bigger than the one from EFI! %ld > %ld\n",
+			 sizeof(efi_memory_desc_t), memmap.desc_size);

Does it make sense for this to just be a warning? If sizeof(efi_memory_desec_t) > sizeof(memmap.desc_size) aren't we going to run into some pretty serious problems when we start accessing memory descriptors?

Quoted 3 lines

Would it not make more sense for this to be a BUG_ON()? In the unlikely event that the kernel's efi_memory_desc_t is too large I really don't see how we can recover from that.

Using WARNING is ok here.

1. bootloader aka grub.efi will convert efi memmap to e820 memmap and pass it in boot params. and current grub.efi does not compare desc_size and struct size.
2. kernel efi.c will only need to use desc_size when
a. do_add_efi_memmap when "add_efi_memmap" is appended. it will add more entries in case bootloader miss some. b. for print_efi_memmap() debug purpose.

So if that efi_memmap is broken, those code can not be reached.

other case: if efi_memmap is broken or bootloader does not convert it properly. user could specify mem= or memmap= to pass info to boot. if they can boot, print_efi_memmap could print out the messed up memmap for them.
if you change to BUG_ON, they can not boot anymore.

Thanks

Yinghai Lu

#3 lore raw
Quoted 2 lines · Yinghai Lu

1. bootloader aka grub.efi will convert efi memmap to e820 memmap and pass it in boot params. and current grub.efi does not compare desc_size and struct size.

So we'll already have crashed and burned before we got here, and so this check will never fire. I sent a patch to just remove the message on the 28th of March.

There's no way someody could ship a system with a desc_size that's smaller than sizeof(memdesc). It wouldn't boot anything.

Signature · 1 line
Matthew Garrett | [email protected]