gadgetfs has had a module refcount bug in all recent kernels, from v3.10 up through present. I realize that gadgetfs is deprecated in favor of functionfs nowadays, but probably you will want to fix it anyways since gadgetfs is still in the tree. The refcount bug can be reproduced with the following test case (tried on both ARM and x86-64): * build gadgetfs as module, build gadgetfs example app from http://www.linux-usb.org/gadget/usb.c * run test script modprobe dummy_hcd modprobe gadgetfs mkdir -p /dev/gadget mount -t gadgetfs none /dev/gadget lsmod | grep gadgetfs # this will show usage 1, as expected ./usb lsmod | grep gadgetfs # still shows usage 1 * now when example app is killed, gadgetfs usage count goes from 1 down to 0 (despite still being mounted) * in fact, the usage count is decremented every time a process exits after using gadgetfs (meaning it goes negative on the 2nd exit) Some poking around showed fops_put() in __fput() being the guilty party for decrementing the refcount on process exit, which was an unbalanced module put because running the process didn't do a module get. More time spent with git-bisect showed that commit 3273097ee9c077 "gadgetfs: don't bother with fops->owner" was the one that caused the regression. Simply reverting that commit restores old behavior with balanced gets and puts: opening gadgetfs files increments the module refcount, closing them decrements it. -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
