On Fri, Jan 13, 2017 at 10:21:15AM -0700, Stephen Warren wrote:
On 01/13/2017 04:48 AM, Brian Masney wrote:
On Thu, Jan 12, 2017 at 11:47:48AM -0700, Stephen Warren wrote:
On 01/12/2017 11:32 AM, Brian Masney wrote:
On Thu, Jan 12, 2017 at 11:02:14AM -0700, Stephen Warren wrote:
On 01/12/2017 01:57 AM, Brian Masney wrote:
The bcm2835 driver polls the monitor and selects the highest resolution that is available. This patch allows optionally setting the video-mode environment variable so that a different video resolution can be used. If the environment variable is not specified, then it will fall back to using the old behavior of using the maximum allowable resolution.
This patch is needed to fix an issue booting an upstream Linux kernel on a Raspberry Pi 2 with a Pi Top screen. Previously, the bcm2835 would select the 1366x768 resolution (which is a supported resolution), however the screen would be unreadable. (See https://www.flickr.com/photos/masneyb/30942037416/ for picture). Using this patch, the resolution 1024x768 can be selected and is readable on the screen.
Doesn't this mean that the RPi firmware is reporting the wrong resolution? If so, isn't the correct fix to get an updated firmware that reports the correct resolution, rather than patching each piece of SW to ignore the FW-reported resolution? Or, if this is caused by incorrect EDID in the Pi Top, then fix the EDID EEPROM on that.
Perhaps there are other use-cases for using a non-default resolution, but to support that, you'd need to make a call into the FW to request and configure that non-default resolution, not just ignore what resolution the FW programmed.
Hi Stephen, The Pi Top screen works correctly with the 1366x768 resolution when booting the 4.4 kernel provided by the Raspberry Pi foundation in stock Raspbian (no u-boot). (There are no outside provided drivers from Pi Top.) When booting with u-boot, I can't use the 1366x768 resolution, even when setting the resolution manually using my patch. When auto detection is in place, u-boot correctly detects the 1366x768 resolution according to debugging messages that I see on the serial console. 1024x768 is the highest resolution that I can get a working display with the Pi Top and u-boot. I also tried changing the display depth.
I tried booting u-boot using the latest Raspberry Pi firmware and the older firmware provided with the Raspbian 4.4 kernel. In both cases, the screen correctly displays the rainbow square at the top left when the GPU is booting, but then the screen becomes garbled when it gets to u-boot and the bcm2835 code sets the resolution.
I tried going through the Pi Top vendor for help but didn't get very far with them. I'm open to other suggestions to try.
Is the bad image that you get static/fixed, or does it move about randomly?
If it's static/fixed, I wonder if the issue is with the memory pitch calculation. What value does bcm2835_pitch have without your patch? (and with your patch, I think it's uninitialized?).
I wonder if you round the value of bcm2835_pitch up to a multiple of 256 (or something like that), then perhaps the issue may be solved? If you change the pitch value, and you notice the angle of the diagonal edges in the image change, the issue is almost certainly related to this pitch value.
I can't recall how the mainline kernel driver works now. If it also uses the property mailbox to configure the display, and you're just using the dumb simplefb driver, perhaps you can compare the memory layout parameters the kernel uses when it's working to what U-Boot uses when it's not working.
The image is fixed. I can see when the Linux Kernel boots and the console messages scroll across the screen.
Without my patch, u-boot detects the screen resolution as 1366x768 with a pitch of 5504. With my patch, 1024x768 uses a pitch of 2048.
I reverted my u-boot patch and tried hard coding the pitch to the values 5376, 5632, 2048, 4096, and 6144 with no success. Once I read what the pitch value does, I also tried 5464 (1366 x 4 (for 32bpp)) and 2732.
Is this related?
https://www.riscosopen.org/forum/forums/4/topics/6400?posts_per_page=25&...
(See "I think I've just fixed that")
https://www.riscosopen.org/viewer/revisions/logs?ident=1471703957-443671.htm...
(See the diff in the BCMVideo file)
Another thing to try might be to remove the SET_* operations from U-Boot's property mailbox calls (I think the FW always initializes video out anyway, so U-Boot doesn't need to set anything up), and simply invoke the relevant GET_* operations to query where the buffer is and its size. IIRC I didn't do that because you can only query certain information as a side-effect of asking the FW to allocate a frame-buffer though (i.e. the info comes back in a SET_xxx/ALLOCATE_xxx response message, but there's no GET_xxx to query the data without modifying the configuration or re-allocating anything).
I attached a patch (uboot-bcm2835-release-fb.patch) that releases the frame buffer before the other mailbox properties are set. This did not fix my display problem.
I also tried removing all of the SET_ operations (uboot-bcm2835-remote-mbox-sets.patch). The screen stays on the rainbow screen when the GPU is initialized. I can see through the serial console that the pi boots normally.
Applying both patches gives a blank screen so it appears that releasing the frame buffer works.
In the BCMVideo diff that you referenced, there are two rows between ARM2VC_Tag_FBRelease and ARM2VC_Tag_End with zeros:
+tagrel DCD tagrellen + DCD 0 + DCD ARM2VC_Tag_FBRelease + DCD 0 + DCD 0 + DCD ARM2VC_Tag_End +tagrellen * . - tagrel
Are these parameters? I don't see places to put those in bcm2835_mbox_tag_release_buffer.
From arch/arm/mach-bcm283x/include/mach/mbox.h:
struct bcm2835_mbox_tag_release_buffer { struct bcm2835_mbox_tag_hdr tag_hdr; union { struct { } req; struct { } resp; } body; };
I also forgot to mention in my last email that changing the value of bcm2835_pitch does not cause the output to move.
Brian