On Sun. 12 juin 2022 at 00:31, Marc Kleine-Budde <mkl@xxxxxxxxxxxxxx> wrote: > On 10.06.2022 17:33:35, Rhett Aultman wrote: > > From: Vincent Mailhol <mailhol.vincent@xxxxxxxxxx> > > > > When allocating URB memory with kmalloc(), drivers can simply set the > > URB_FREE_BUFFER flag in urb::transfer_flags and that way, the memory > > will be freed in the background when killing the URB (for example with > > usb_kill_anchored_urbs()). > > > > However, there are no equivalent mechanism when allocating DMA memory > > (with usb_alloc_coherent()). > > > > This patch adds a new flag: URB_FREE_COHERENT. Setting this flag will > > cause the kernel to automatically call usb_free_coherent() on the > > transfer buffer when the URB is killed, similarly to how > > URB_FREE_BUFFER triggers a call to kfree(). > > > > In order to have all the flags in numerical order, URB_DIR_IN is > > renumbered from 0x0200 to 0x0400 so that URB_FREE_COHERENT can reuse > > value 0x0200. > > > > Co-developed-by: Vincent Mailhol <mailhol.vincent@xxxxxxxxxx> > > Signed-off-by: Vincent Mailhol <mailhol.vincent@xxxxxxxxxx> > > Co-developed-by: Rhett Aultman <rhett.aultman@xxxxxxxxxxx> > > Signed-off-by: Rhett Aultman <rhett.aultman@xxxxxxxxxxx> > > Reviewed-by: Vincent Mailhol <mailhol.vincent@xxxxxxxxxx> > > FWIW: > Acked-by: Marc Kleine-Budde <mkl@xxxxxxxxxxxxxx> > > This patch probably goes upstream via USB. Once this is in net I'll take > the 2nd patch. Question to Greg: can this first patch also be applied to the stable branches? Technically, this is a new feature but it will be used to solve several memory leaks on existing drivers (the gs_usb is only one example). Yours sincerely, Vincent Mailhol
