From mboxrd@z Thu Jan 1 00:00:00 1970 From: Floris Bos Date: Wed, 12 Aug 2015 23:47:01 +0200 Subject: [Buildroot] [PATCH] wf111: Add mdev dependency In-Reply-To: References: <1439277422-26111-1-git-send-email-cdhmanning@gmail.com> <55CB2019.5030304@je-eigen-domein.nl> Message-ID: <55CBBED5.80108@je-eigen-domein.nl> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: buildroot@busybox.net On 08/12/2015 10:49 PM, Charles Manning wrote: > On Wed, Aug 12, 2015 at 10:29 PM, Floris Bos wrote: >> On 08/11/2015 09:17 AM, Charles Manning wrote: >>> wf111 really needs mdev >> >> Problem I have seen with mdev in its default configuration is that it >> doesn't just handle firmware load requests, but also takes it upon itself to >> create/delete the device nodes (even when that already is being handled by >> devtmpfs), and does a bad job with that. >> >> It doesn't preserve sequencing by default, and events are processed >> concurrently, and get out-of-order. >> >> Becomes a problem if your software for example does anything that causes the >> kernel to re-read the partition table of your storage. >> In that case the kernel sends events to delete all devices nodes of the >> storage device, and add them again, right after each other. >> However depending on your luck, they can get processed in reverse order: >> >> == >> mdev[598]: 9.191594 ACTION:add SUBSYSTEM:block DEVNAME:mmcblk0p6 >> DEVPATH:/devices/platform/soc/3f300000.mmc/mmc_host/mmc0/mmc0:1234/block/mmcblk0/mmcblk0p6 >> mdev[598]: dev 179,6 >> [...] >> mdev[593]: 9.192168 ACTION:remove SUBSYSTEM:block DEVNAME:mmcblk0p6 >> DEVPATH:/devices/platform/soc/3f300000.mmc/mmc_host/mmc0/mmc0:1234/block/mmcblk0/mmcblk0p6 >> mdev[593]: rule matched, line -1 >> mdev[593]: unlink: mmcblk0p6 >> == >> >> Add is processed before remove, with the end result that /dev/mmcblk0p6 goes >> missing... >> >> >> So please don't pull in mdev automatically just for firmware loading, >> because it does more than just that, and can have serious side effects. > Ok, now I'm stumped. People are telling me "don't use mdev", but > nobody is providing a viable alternative. > > The situation is this: > * I did not write wf111, it is not as if I am submitting a new package. > * We have an existing package wf111. > * The existing package does not work (ie. does not do fw loading) > unless there is mdev or udev. Without fw loading wf111 is broken. Suggest you double check there isn't a newer kernel version available for your target device. Pretty sure newer kernels can also load the firmware straight from /lib/firmware without userspace helper. Yours sincerely, Floris Bos