Hi Alan, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> ezt írta (időpont: 2020. júl. 20., H, 19:29): > 4.19 is a very old kernel. You'd be a lot better off testing under a > recent kernel, like 5.7. It also would be more convenient to do your > development on a regular PC instead of a Raspberry Pi. You can use > dummy-hcd for testing gadget drivers. Okay, I put together a Slackware-current based VM to conduct the development work on. I compiled a patched version of kernel 5.7.9 with all the necessary config parameters to have USB gadget support with debugging enabled, and also the dummy HCD. At first loaded the g_mass_storage module with a smaller ~700 MB iso file, and it loaded seemingly fine, and functionality-wise it also seemed to work fine, although there were frequent periodical debug messages about the gadget returning command-failure status: https://gist.github.com/jszigetvari/5bc4cdc8c55588907b71832558cb00a7 After that as a test I tried to dd all the data from /dev/sr1 (that's the device file the gadget was assigned) to /dev/null. The process completed without any user-visible errors, although the failure status related messages kept on appearing: https://gist.github.com/jszigetvari/ffa19850c94f12510c0b78ed5aaba2e1 Finally I tried to load the large, 11 GB CentOS iso file as the second phase of the test. On the Raspberry Pi (with kernel 4.19) this was when the crash happened, however this time it did not. Originally Tiziano Bacocco's patch came around to be the time of the 3.x kernels. The patch needed minimal adjustments to apply to 4.19 and 5.4 and also to 5.7.9 (though there was a one line offset in f_mass_storage.c). https://gist.github.com/jszigetvari/5a9796c8af8a01c0edba6a8696540029 Although I have to admit that on the Pi Zero I was using the g_multi kernel module, and not g_mass_storage I assumed they share the same code base, so they should work the same way. On the Pi zero I need the extra serial and Ethernet interface to have access to the Pi Zero, as it doesn't have a wireless NIC, and that's why I chose g_multi.ko. In the end I did a dd with the 11 GB iso, and to my surprise it worked well. dd returned without any errors and it did show the true 11 GB size when it finished. In the kernel logs most log entries were about block reads, but the failure status messages were still around. If there was no activity on the device, then they accounted for the majority of the logs generated: https://gist.github.com/jszigetvari/a9225fa3173e6c7b7b50673a494d9c37 I have to admit, I stand puzzled why it worked on an x86_64-based machine and crashed on the Pi Zero. I guess the next step will be either: * to try using g_multi instead of g_mass_storage on the test machine * upgrade the Pi Zero to run kernel 5.4.x (AFAIK that's the latest one can get from the Raspberry Pi Foundation's git repo), and recompile that kernel version with the patch applied. (This will likely take a few days.) Any further suggestions would also be welcome! Best Regards, János -- Janos SZIGETVARI RHCE, License no. 150-053-692 LinkedIn: linkedin.com/in/janosszigetvari __@__˚V˚ Make the switch to open (source) applications, protocols, formats now: - windows -> Linux, iexplore -> Firefox, msoffice -> LibreOffice - msn -> jabber protocol (Pidgin, Google Talk) - mp3 -> ogg, wmv -> ogg, jpg -> png, doc/xls/ppt -> odt/ods/odp
