VBE DDC bios call stalls boot
RFC 1/1 VBE DDC bios call stalls boot
In my setup I have a monitor connected via a DLink KVM (DKVM 2K). The boot stalls forever during early boot phase until I physically disconnect the KVM from the video port at my CPU box and reconnect it. I was able to trace it to the VBE/DDC call in video.S. The bios call does not seem to return. When I disconnect the video, the bios call returns with failure and boot picks it back from there. When I directly connect my monitor bypassing the KVM, this issue does not happen.
There is a bios call to check if read EDID is supported. My first thought was that before doing read EDID, video.S should first check to see if the hardware supports the read EDID feature. Unfortunately that bios call too ends up in the woods until I physically disconnect/reconnect my CPU video output that feeds into the KVM. I see the same issue with user space programs such as read-edid and xwindows that also attempt to read EDID. They simply stall at the DDC call until I physically disconnect/reconnect my video.
I don't think boot should stall for any reason. Moreover given that user space programs are doing these VBE/DDC calls, I don't see any compelling reason why this code need to exist is such early boot stage. If it absolutely needs to be called in the kernel, at least if invoked sometime later, we could time out this call as a workaround.
Kindly suggest the best approach. The following patch works for me:
Thanks.
arch/i386/boot/video.S +0 -7
@@ -1946,13 +1946,6 @@ store_edid: rep stosl - movw $0x4f15, %ax # do VBE/DDC - movw $0x01, %bx - movw $0x00, %cx - movw $0x00, %dx - movw $0x140, %di - int $0x10 - popw %di # restore all registers popw %dx popw %cx
Quoted 5 lines · Yum Rayan
I don't think boot should stall for any reason. Moreover given that user space programs are doing these VBE/DDC calls, I don't see any compelling reason why this code need to exist is such early boot stage. If it absolutely needs to be called in the kernel, at least if invoked sometime later, we could time out this call as a workaround.
Don't configure CONFIG_FB_FIRMWARE_EDID, if it doesn't work on your hardware.
Just disable "Enable firmware EDID" in
"Device Drivers"->"Graphics support"->"Support for frame buffer devices" of your menuconfig dialog. The help mentions, that this will not always work.
That will make your hardware work flawlessly again.
Regards
Ingo Oeser
Quoted 9 lines · Ingo Oeser
Quoted 5 lines · Yum Rayan
I don't think boot should stall for any reason. Moreover given that user space programs are doing these VBE/DDC calls, I don't see any compelling reason why this code need to exist is such early boot stage. If it absolutely needs to be called in the kernel, at least if invoked sometime later, we could time out this call as a workaround.
Don't configure
CONFIG_FB_FIRMWARE_EDID, if it doesn't work on your hardware.
This option is new and not yet available in 2.6.15, so you may have to use your patch temporarily. If you can, set CONFIG_VIDEO_SELECT = n in Device Drivers -> Graphics support -> Console Drivers -> Video mode selection support. However, this is automatically selected by FB, PCI, and X86_32, so it may not be changeable.
The "vga=" option will not be available if VIDEO_SELECT is not set.
Or just apply this changeset:
to make CONFIG_FB_FIRMWARE_EDID available.
Tony
Quoted 5 lines · Yum Rayan
There is a bios call to check if read EDID is supported. My first thought was that before doing read EDID,
video.Sshould first check to see if the hardware supports the read EDID feature. Unfortunately that bios call too ends up in the woods until I physically disconnect/reconnect my CPU video output that feeds into the KVM.
Is this how your patch looked when you tried this ?
Dave
arch/i386/boot/video.S renamed from arch/i386/boot/video.S~ +11 -2
@@ -1951,9 +1951,18 @@ store_edid: stosl movw $0x4f15, %ax # do VBE/DDC - movw $0x01, %bx + movw $0x00, %bx # INSTALLATION CHECK / CAPABILITIES movw $0x00, %cx - movw $0x00, %dx + movw $0x00, %dx + movw $0x140, %di + int $0x10 + cmpb $0x01, %ah + je no_edid + + movw $0x4f15, %ax # do VBE/DDC + movw $0x01, %bx # READ_EDID + movw $0x00, %cx + movw $0x00, %dx movw $0x140, %di int $0x10
arch/i386/boot/video.S renamed from arch/i386/boot/video.S~ +1 -0
@@ -1966,6 +1966,7 @@ store_edid: movw $0x140, %di int $0x10 +no_edid: popw %di # restore all registers popw %dx popw %cx
Signature · 1 line
http://www.codemonkey.org.uk
Quoted 9 lines · Dave Jones
Quoted 5 lines · Yum Rayan
There is a bios call to check if read EDID is supported. My first thought was that before doing read EDID,
video.Sshould first check to see if the hardware supports the read EDID feature. Unfortunately that bios call too ends up in the woods until I physically disconnect/reconnect my CPU video output that feeds into the KVM.Is this how your patch looked when you tried this ?
I've discovered that checking for availability does not give any added benefit, but this is the correct approach and I have to agree with this patch.
Tony
Quoted 4 lines
Quoted 1 line
Is this how your patch looked when you tried this ?
I've discovered that checking for availability does not give any added benefit, but this is the correct approach and I have to agree with this patch.
My patch was similar. I referenced
http://www.vesa.org/public/VBE/VBEDDC11.PDF. Differences were the finer look up of the ambiguous status code and setting up ES:DI to be NULL before call to INT 10h, as follows:
arch/i386/boot/video.S +17 -0
@@ -1946,6 +1946,22 @@ store_edid: rep stosl + pushw %es + movw $0x4f15, %ax # report VBE/DDC capabilities + xorw %bx, %bx + xorw %cx, %cx + xorw %di, %di + pushw %di + popw %es + int $0x10 + popw %es + + cmpb $0x0, %ah # function call failed + jne no_edid + + cmpb $0x4f, %al # function not supported + jne no_edid + movw $0x4f15, %ax # do VBE/DDC movw $0x01, %bx movw $0x00, %cx
@@ -1953,6 +1969,7 @@ store_edid: movw $0x140, %di int $0x10 +no_edid: popw %di # restore all registers popw %dx popw %cx
Quoted 2 lines
This option is new and not yet available in 2.6.15, so you may have to use your patch temporarily.
I was able to download a more recent kernel and configure CONFIG_FB_FIRMWARE_EDID. Thanks.