On 09/08/2019 11:07, Christoph Hellwig wrote:

On Fri, Aug 09, 2019 at 09:40:32AM +0300, Tomi Valkeinen wrote:

...

We do call dma_set_coherent_mask() in omapdrm's probe() (in omap_drv.c), but apparently that's not enough anymore. Changing that call to dma_coerce_mask_and_coherent() removes the WARN. I can create a patch for that, or Christoph can respin this one.

Oh, yes - that actually is the right thing to do here. If you already have it please just send it out.

...

I am not too familiar with the dma mask handling, so maybe someone can educate:

dma_coerce_mask_and_coherent() overwrites dev->dma_mask. Isn't that a bad thing? What if the platform has set dev->dma_mask, and the driver overwrites it with its value? Or who is supposed to set dev->dma_mask?

->dma_mask is a complete mess. It is a pointer when it really should just be a u64, and that means every driver layer has to allocate space for it. We don't really do that for platform_devices, as that breaks horribly assumptions in the usb code. That is why dma_coerce_mask_and_coherent exists as a nasty workaround that sets the dma_mask to the coherent_dma_mask for devices that don't have space for ->dma_mask allocated, which works as long as the device doesn't have differnet addressing requirements for both.

I'm actually working to fix that mess up at the moment, but it is going to take a few cycles until everything falls into place.

Alright, thanks for the clarification!

Here's my version.

From c258309e36fc86076db155aead03a3900b96c3d4 Mon Sep 17 00:00:00 2001

From: Tomi Valkeinen tomi.valkeinen@ti.com Date: Fri, 9 Aug 2019 09:54:49 +0300 Subject: [PATCH] drm/omap: ensure we have a valid dma_mask

The omapdrm driver uses dma_set_coherent_mask(), but that's not enough anymore when LPAE is enabled.

From Christoph Hellwig hch@lst.de:

The traditional arm DMA code ignores, but the generic dma-direct/swiotlb has stricter checks and thus fails mappings without a DMA mask. As we use swiotlb for arm with LPAE now, omapdrm needs to catch up and actually set a DMA mask.

Change the dma_set_coherent_mask() call to dma_coerce_mask_and_coherent() so that the dev->dma_mask is also set.

Fixes: ad3c7b18c5b3 ("arm: use swiotlb for bounce buffering on LPAE configs") Reported-by: "H. Nikolaus Schaller" hns@goldelico.com Signed-off-by: Tomi Valkeinen tomi.valkeinen@ti.com

diff --git a/drivers/gpu/drm/omapdrm/omap_drv.c b/drivers/gpu/drm/omapdrm/omap_drv.c index 288c59dae56a..1bad0a2cc5c6 100644 --- a/drivers/gpu/drm/omapdrm/omap_drv.c +++ b/drivers/gpu/drm/omapdrm/omap_drv.c @@ -669,7 +669,7 @@ static int pdev_probe(struct platform_device *pdev) if (omapdss_is_initialized() == false) return -EPROBE_DEFER;

- ret = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32)); + ret = dma_coerce_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); if (ret) { dev_err(&pdev->dev, "Failed to set the DMA mask\n"); return ret;