On Fri, 8 Dec 2017, Geert Uytterhoeven wrote: > Hi Alan, > > On Thu, Dec 7, 2017 at 10:26 PM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote: > > On Thu, 7 Dec 2017, Dan Carpenter wrote: > >> The standard is to treat them like errors and try press forward in a > >> degraded mode but don't print a message. Checkpatch.pl complains if you > >> print a warning for allocation failures. A lot of low level functions > >> handle their own messages pretty well but especially kmalloc() does. > > > > Which brings us back to my original objection. If an allocation > > failure has localized effects, but the low-level warning is unable to > > specify what will be affected, how is the user supposed to connect the > > effect to the cause? > > The backtrace would include usb_hub_clear_tt_buffer, so the user will > know USB is affected. > Note that the cause of the memory exhaustion is usually not the caller. > Unless it requests a real big block of memory, in which case that will be > mentioned in the backtrace, too. > > In this particular case, the driver uses GFP_ATOMIC, so allocation failures > aren't that rare, and I think the driver should be prepared for that, and try > to recover gracefully. > > The comment > > /* FIXME recover somehow ... RESET_TT? */ > > makes me think it isn't. > > As this is a pretty small allocation, perhaps it can be done beforehand, without > GFP_ATOMIC? I can't see how to make that work. We don't know beforehand how many structures will be needed at any time. Alan Stern -- To unsubscribe from this list: send the line "unsubscribe kernel-janitors" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
