Hi Jiri, On Wed, 15 August 2012 Jiri Kosina <jkosina@xxxxxxx> wrote: > I see. Alan Stern has fixed a huge pile of things in this area in 3.6-rc1. > I have expected all of those to actually be on theoretical problems not > ever having happened in the wild, but it might be that you are actually > chasing on of those. > > Could you please retest with latest Linus' tree (or at least eb055fd0560b) > to see whether this hasn't actually been fixed already by Alan's series? I've started trying that out, it seems Alan's work improved things. For the first few attempts I have not seen SLAB corruptions, though after a few rounds I hit accumulation of the following messages: [ 297.174828] hid-picolcd 0003:04D8:C002.0003: output queue full [ 297.181098] hid-picolcd 0003:04D8:C002.0003: output queue full [ 297.187820] hid-picolcd 0003:04D8:C002.0003: output queue full [ 297.194087] hid-picolcd 0003:04D8:C002.0003: output queue full with sporadically in between: [ 292.668019] hid-picolcd 0003:04D8:C002.0003: usb_submit_urb(out) failed: -1 At first glance I think the queue filling up and never draining is caused by hid_hw_stop() stalling the queue and the time between both being just too short. Maybe me (un)binding on hid-picolcd driver level is the major cause of this and (un)binding at lower usb level would kill the queue thus preventing it just growing until full... Is there a proper way to do rate-limiting or throttling when submitting reports (which are the urbs deeper in the stack) and catch -ENODEV early while remove() is delayed for locking reasons? At least until 3.5 submitting reports provides no return value for driver which could convey errors like -ENODEV, -EBUSY. After a few bind-unbind iterations I have also hit a race in hid-picolcd with regard to fbdefio. That looks like chicken-egg release sequencing between hid-picolcd and fb_info subdev -- though I'm wondering how that happens as the only framebuffer user I know of - fbcon - should have stopped using it after unregister_framebuffer() returned and released the last reference! But that part can be improved rather easily with a new spinlock synchronizing interoperation between hid side and framebuffer side. Thanks, Bruno -- To unsubscribe from this list: send the line "unsubscribe linux-input" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
