On 31 July 2018 at 18:06, Mark Brown [off-list ref] wrote:

On Tue, Jul 31, 2018 at 08:14:12AM -0700, kernelci.org bot wrote:

Today's -next fails to boot on a variety of Qualcomm 32 bit platforms:

quoted

    multi_v7_defconfig:
        qcom-apq8064-cm-qs600:
            lab-baylibre-seattle: new failure (last pass: next-20180730)
        qcom-apq8064-ifc6410:
            lab-baylibre-seattle: new failure (last pass: next-20180730)

    qcom_defconfig:
        qcom-apq8064-cm-qs600:
            lab-baylibre-seattle: new failure (last pass: next-20180730)

The logs are all somewhat similar, for example:

   https://storage.kernelci.org/next/master/next-20180731/arm/multi_v7_defconfig/lab-baylibre-seattle/boot-qcom-apq8064-cm-qs600.html

detects a DMA problem during MMCI initialization:

[    2.237566] mmci-pl18x 121c0000.sdcc: mmc2: PL180 manf 51 rev0 at 0x121c0000 irq 32,0 (pio)
[    2.244790] mmci-pl18x 121c0000.sdcc: DMA channels RX dma2chan1, TX dma2chan2
[    2.271722] mmci-pl18x 12400000.sdcc: error during DMA transfer!
[    2.271757] mmci-pl18x 12400000.sdcc: buggy DMA detected. Taking evasive action.
[    2.276798] ------------[ cut here ]------------
[    2.284185] WARNING: CPU: 0 PID: 0 at ../include/linux/dma-mapping.h:551 bam_free_chan+0x2d8/0x2e0
[    2.288772] Modules linked in:
[    2.297534] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.18.0-rc7-next-20180731 #1

then panics:

[    2.513796] ------------[ cut here ]------------
[    2.518367] kernel BUG at ../mm/vmalloc.c:1608!
[    2.522968] Internal error: Oops - BUG: 0 [#1] SMP ARM

trying to release the DMA channel.  I've not done any bisection or
anything but I do note 8bb2299d2d0b5cc (mmc: mmci: Add and implement a
->dma_setup() callback for qcom dml) and some related commits in the MMC
tree.

More details for each of the failed boots at:

  https://kernelci.org/boot/id/5b6054f559b5144b9396baa9/
  https://kernelci.org/boot/id/5b60551259b5144abb96bab6/
  https://kernelci.org/boot/id/5b6054e259b5144b1e96bab2/

including full logs, details of the build and so on.

Mark, thanks for reporting.

Problem was a simple one liner that should have been added to included
in my patch "mmc: mmci: Add and implement a ->dma_setup() callback for
qcom dml". The missing oneliner caused mmci to wrongly use dma for the
qcom variant.

I have amended the patch and published it, it should reach the next
tree as of tomorrow. Apologize for the mess it created.

Kind regards
Uffe