2012.12.20. 12:06 keltezéssel, Stanislaw Gruszka írta: > On Thu, Dec 20, 2012 at 10:58:31AM +0200, Kalle Valo wrote: >> Gabor Juhos <juhosg@xxxxxxxxxxx> writes: >> >>>> Since we use completion here, why we can not just use normal synchronous >>>> version of request_firmware? I heard of request_firmware drawbacks, so >>>> this approach can be correct. Just want to know if we do not complicate >>>> things not necessarily here. >>> >>> If the driver is built into the kernel, then the synchronous version >>> would fail because user-space is not up during probe time. >> >> Didn't udev/systemd guys finally fix their software? At least I recall >> seeing such claims on g+. > > Udev broke and later fixed request_firmware when driver is compiled as > module. Here we avoid problem when driver is compiled into kernel > (problem which was there from very beginning) - no userspace available > so request_firmware fail. But since everyone are using modules, nobody > cares. > > Openwrt must use modules too, since it use compat-wireless, right? Yes, we are using modules. > So I'm not sure if it worth to use request_firmware_nowait. Especially like > that with completion, which block ->probe. As long kernel does not probes > devices in parallel, we block whole boot, and deadlock since probe wait for > userspace. > > Gabor, did you test your patches ? I have tested the patches, but only agains compat-wireless. I did not try the rt2x00 driver built into the kernel. Although we are using 3.6.11 currently, but the deadlock should happen on that as well. I will try it. -Gabor -- To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
