Skip to contentSkip to conversation

arch/i386/boot/video.S hang

DISCUSSION Simon Richard Grint 3 participants 4 messages Jun 15, 2005 – Jun 18, 2005

lore
Share & tools
lore
#0 lore raw

I'm trying to upgrade an old AMD K6-2 machine (aladdin bios/gigabyte GA-5AX motherboard, latest version of the bios) to 2.6.11. Unfortunately kernels later than 2.6.9 hang very early in the boot process just after the vga= mode selection, but before the kernel announces "uncompressing linux".

I have narrowed the problem down to the store_edid function in arch/i386/boot/video.S where the edid block is obtained and stored before entering protected mode. The exact patch which seems to cause me problems is http://www.ussg.iu.edu/hypermail/linux/kernel/0409.3/1786.html

Storing the edid block at 0x140 causes this machine to hang, whereas backing this patch out and instead using 0x440 (or even 0x160) seems to work fine.

Is this problem just because of an old and buggy bios or is there another reason?

Thanks for your help

sr

#1 lore raw
Quoted 11 lines · Simon Richard Grint

I'm trying to upgrade an old AMD K6-2 machine (aladdin bios/gigabyte GA-5AX motherboard, latest version of the bios) to 2.6.11. Unfortunately kernels later than 2.6.9 hang very early in the boot process just after the vga= mode selection, but before the kernel announces "uncompressing linux".

I have narrowed the problem down to the store_edid function in arch/i386/boot/video.S where the edid block is obtained and stored before entering protected mode. The exact patch which seems to cause me problems is http://www.ussg.iu.edu/hypermail/linux/kernel/0409.3/1786.html

Quoted 6 lines

Storing the edid block at 0x140 causes this machine to hang, whereas backing this patch out and instead using 0x440 (or even 0x160) seems to work fine.

Is this problem just because of an old and buggy bios or is there another reason?

What boot loader are you using. grub/lilo?

Does it work with CONFIG_VIDEO_SELECT disabled in your kernel CONFIG?

Thanks,
Venki

#2 lore raw
Quoted 1 line · Venkatesh Pallipadi

What boot loader are you using. grub/lilo?

The same thing happens with either grub or lilo, but I'm using grub at present

Quoted 1 line

Does it work with CONFIG_VIDEO_SELECT disabled in your kernel CONFIG?

It works fine if CONFIG_VIDEO_SELECT is disabled. Even with CONFIG_VIDEO_SELECT enabled, the problem only arises if I pass a vga= parameter to the kernel

Thanks for your help

sr

#3 lore raw
Quoted 6 lines · Simon Richard Grint
Quoted 1 line · Venkatesh Pallipadi

What boot loader are you using. grub/lilo?

The same thing happens with either grub or lilo, but I'm using grub at present

Quoted 8 lines
Quoted 1 line

Does it work with CONFIG_VIDEO_SELECT disabled in your kernel CONFIG?

It works fine if CONFIG_VIDEO_SELECT is disabled. Even with CONFIG_VIDEO_SELECT enabled, the problem only arises if I pass a vga= parameter to the kernel

Thanks for your help

This one continues to be mysterious.

One reason I could think of: VBE call 4f15:0s is storing more than 128 bytes on this platform. With the base address at 0x440, it can write much longer, without affecting anything else in zero page. But, at 0x140 it overwrites some other fields.

But, that doesn't explain why it only fails when vga=<num> is passed. For the above theory it should fail whenever VIDEO_SELECT is enabled.

Can you try some other address higher than 0x440 and less than 0x600 for and see whether it works? Or you can also try and print complete boot_params[] at some place in arch/i386/kernel/setup.c:setup_arch() in the case where it runs fine (with 0x440) and send me the log.

Thanks,
Venki