Re: [PATCH 1/2] hw/misc/bcm2835_property: Fix framebuffer with recent RPi kernels

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Sun, 2021-10-17 at 17:08 +0200, Philippe Mathieu-Daudé wrote:
> Hi Benjamin,
> 
> On 10/17/21 09:48, Benjamin Herrenschmidt wrote:
> > The framebuffer driver fails to initialize with recent Raspberry Pi
> > kernels, such as the ones shipped in the current RaspiOS images
> > (with the out of tree bcm2708_fb.c driver)
> 
> Which particular version?

The one I dug out of 2021-05-07-raspios-buster-arm64-lite.img (note
that this then fails to boot some time after the fb is setup, even
after the fix, in the vchip driver init (before serial is up even).

That said, the same fb problem happens with 5.10.60-v8+ from raspbian.

I'm not sure your fix will work on these, see below:

> +        case 0x00040013: /* Get number of displays */
> +            stl_le_phys(&s->dma_as, value + 12, 0 /* old fw: unique display */);
> +            resplen = 4;
> +            break;
> 
This should have been equivalen to not having the property. However,
the failure path in the driver looks like this (note the mismatch
between the comment and the code.. this is rpi 5.10.73 from the rpi
repo :

	ret = rpi_firmware_property(fw,
				    RPI_FIRMWARE_FRAMEBUFFER_GET_NUM_DISPLAYS,
				    &num_displays, sizeof(u32));

	/* If we fail to get the number of displays, or it returns 0, then
	 * assume old firmware that doesn't have the mailbox call, so just
	 * set one display
	 */
	if (ret || num_displays == 0) {
		dev_err(&dev->dev,
			"Unable to determine number of FBs. Disabling driver.\n");
		return -ENOENT;
	} else {
		fbdev->firmware_supports_multifb = 1;
	}

So it appears that the intention at some stage was to set only one display but
the code as written will fail to initialize the drive if the properly is absent
*or* if it returns 0.

I've just checked the rpi-5.15.y branch and it's the same.

Cheers,
Ben.