Hi, On Sun, Feb 24, 2019 at 08:03:10PM +0100, Stefan Wahren wrote: > > Aaro Koskinen <aaro.koskinen@xxxxxx> hat am 24. Februar 2019 um 16:07 geschrieben: > > After upgrading from v4.20 -> v5.0-rc7, I'm getting MMC issues on > > RPi3. Below is one example, system just ends up being completely freezed > > after couple hours into GCC bootstrap (passed fine with v4.20). > > > > Happened three times already (also after replacing the MMC card just > > in case). > > > > thanks for your report. > > Can you give us a more detailed test scenario? Well, I have minimal arm64 rootfs + swap on MMC. Then I extract GCC 8.3.0 sources and try to do the compiler bootstrap: mkdir build cd build ../gcc-8.3.0/configure <long list of options here> make -j4 Then just wait. :) > How long does it take until the issue occur? I would say 1-2 hours. The serial log shows ~6500 seconds, and I started the build right after the boot. My guess is that the issue occurs during the stage in the build where there is some really heavy swapping going on (almost 1 GB swap used). > Did you already tested v4.20.12 or only v4.20? Only v4.20, and it was already few weeks ago when I did the GCC 8.2.0 bootstrap. It went fine (5 hours build + 12 hours testsuite) without any errors in dmesg. > Does the issue occur in v5.0-rc7 if you make sdhost use PIO mode > (CONFIG_DMA_BCM2835=m)? I have currently test running where I moved the file system and swap from SD card slot behind an USB SD card adapter (still the same physical SD card). And that has been running fine for 4 hours already. I want to first see if that one finishes properly, then I can start a new test using sdhost and PIO mode. A.
