[BUG] dmi_scan_machine uses uninitialized address on x86 / EFI

[BUG] dmi_scan_machine uses uninitialized address on x86 / EFI

[Thomas Jourdan]

[-- Attachment #1: Type: text/plain, Size: 462 bytes --]

Hi

On x86/EFI platforms, if efi_enabled is true, the dmi_scan_machine function
fails, even if valid DMI table is present. In the setup_arch function,
efi_init is called after dmi_scan_machine. But this is efi_init which will
initialize the smbios address, needed by dmi_scan_machine.

This leads to dmi_scan_machine ioremapping physical address 0, instead of
the address provided by the EFI system table. The following patch fixes the
problem.

Regards,
Thomas

[-- Attachment #2: dmi_scan.patch --]
[-- Type: application/octet-stream, Size: 989 bytes --]

On x86 / EFI platforms, the SMBIOS table address is available in the EFI
system table. During the setup_arch function, the dmi_scan_machine function
will detect valid DMI table or not. 

If CONFIG_EFI is set, the dmi_scan_table function will use the address provided
by the EFI system table. But it is inialized later, in the efi_init function. 

This patch fixes the call order.

--- linux-2.6.28.6/arch/x86/kernel/setup.c	2009-02-17 18:29:27.000000000 +0100
+++ linux-2.6.28.6/arch/x86/kernel/setup.c	2009-02-23 09:57:57.000000000 +0100
@@ -903,6 +903,9 @@
 
 	finish_e820_parsing();
 
+	if (efi_enabled)
+		efi_init();
+
 	dmi_scan_machine();
 
 	dmi_check_system(bad_bios_dmi_table);
@@ -916,9 +919,6 @@
 	insert_resource(&iomem_resource, &data_resource);
 	insert_resource(&iomem_resource, &bss_resource);
 
-	if (efi_enabled)
-		efi_init();
-
 #ifdef CONFIG_X86_32
 	if (ppro_with_ram_bug()) {
 		e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,

Re: [BUG] dmi_scan_machine uses uninitialized address on x86 / EFI

[H. Peter Anvin]

Thomas Jourdan wrote:
> 
> On x86/EFI platforms, if efi_enabled is true, the dmi_scan_machine function
> fails, even if valid DMI table is present. In the setup_arch function,
> efi_init is called after dmi_scan_machine. But this is efi_init which will
> initialize the smbios address, needed by dmi_scan_machine.
> 
> This leads to dmi_scan_machine ioremapping physical address 0, instead of
> the address provided by the EFI system table. The following patch fixes the
> problem.
> 

Hi Thomas,

What is the concrete manifestation of the bug (i.e. how critical is it?)
 We are late in the 2.6.29-rc cycle, and this patch seems to have some
risk associated with it.

Huang, Peter, Yinghai: any opinions?

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

Re: [BUG] dmi_scan_machine uses uninitialized address on x86 / EFI

[Yinghai Lu]

H. Peter Anvin wrote:
> Thomas Jourdan wrote:
>> On x86/EFI platforms, if efi_enabled is true, the dmi_scan_machine function
>> fails, even if valid DMI table is present. In the setup_arch function,
>> efi_init is called after dmi_scan_machine. But this is efi_init which will
>> initialize the smbios address, needed by dmi_scan_machine.
>>
>> This leads to dmi_scan_machine ioremapping physical address 0, instead of
>> the address provided by the EFI system table. The following patch fixes the
>> problem.
>>
> 
> Hi Thomas,
> 
> What is the concrete manifestation of the bug (i.e. how critical is it?)
>  We are late in the 2.6.29-rc cycle, and this patch seems to have some
> risk associated with it.
> 
> Huang, Peter, Yinghai: any opinions?
> 
should be ok for 2.6.29

YH

Re: [BUG] dmi_scan_machine uses uninitialized address on x86 / EFI

[H. Peter Anvin]

Yinghai Lu wrote:
> should be ok for 2.6.29

Got a feel for what the impact of deferring it to .30 would be?

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

Re: [BUG] dmi_scan_machine uses uninitialized address on x86 / EFI

[Yinghai Lu]

H. Peter Anvin wrote:
> Yinghai Lu wrote:
>> should be ok for 2.6.29
> 
> Got a feel for what the impact of deferring it to .30 would be?
> 
Not sure, wonder if someone try 2.6.29-rc6 on EFI 32/64 sytem...

YH

RE: [BUG] dmi_scan_machine uses uninitialized address on x86 / EFI

[Thomas Jourdan]

Hi guys

The bug is not critical. Vanilla kernel works fine without the patch, except for the DMI sysfs stuffs.

Without the patch, if CONFIG_DMI is enabled, I got the "DMI not present or invalid." error message on boot. Then later, if CONFIG_DMIID is enabled, DMI system identification informations won't be exported via sysfs to userspace. The /sys/class/dmi/id entry is not present.

Regards,
Thomas

-----Message d'origine-----
De : H. Peter Anvin [mailto:hpa@zytor.com]
Envoyé : mar. 24 février 2009 22:12
À : Thomas Jourdan
Cc : linux-kernel@vger.kernel.org; Huang Ying; Peter Jones; Yinghai Lu
Objet : Re: [BUG] dmi_scan_machine uses uninitialized address on x86 /
EFI


Thomas Jourdan wrote:
> 
> On x86/EFI platforms, if efi_enabled is true, the dmi_scan_machine function
> fails, even if valid DMI table is present. In the setup_arch function,
> efi_init is called after dmi_scan_machine. But this is efi_init which will
> initialize the smbios address, needed by dmi_scan_machine.
> 
> This leads to dmi_scan_machine ioremapping physical address 0, instead of
> the address provided by the EFI system table. The following patch fixes the
> problem.
> 

Hi Thomas,

What is the concrete manifestation of the bug (i.e. how critical is it?)
 We are late in the 2.6.29-rc cycle, and this patch seems to have some
risk associated with it.

Huang, Peter, Yinghai: any opinions?

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.


__________ Information NOD32 3887 (20090224) __________

Ce message a ete verifie par NOD32 Antivirus System.
http://www.nod32.com

Re: [BUG] dmi_scan_machine uses uninitialized address on x86 / EFI

[H. Peter Anvin]

Thomas Jourdan wrote:
> Hi
> 
> On x86/EFI platforms, if efi_enabled is true, the dmi_scan_machine function
> fails, even if valid DMI table is present. In the setup_arch function,
> efi_init is called after dmi_scan_machine. But this is efi_init which will
> initialize the smbios address, needed by dmi_scan_machine.
> 
> This leads to dmi_scan_machine ioremapping physical address 0, instead of
> the address provided by the EFI system table. The following patch fixes the
> problem.
> 

Hi Thomas,

Could you please resend both this patch and the phys_efi_get_time patch 
with a Signed-off-by: line so I can apply them?

Thanks,

	-hpa