Hi Hans On Tue, 7 Jul 2020 at 14:49, Hans Verkuil <hverkuil@xxxxxxxxx> wrote: > > On 07/07/2020 15:33, Dave Stevenson wrote: > > Hi Hans, Sergey, & the list. > > > > I'm running the latest v4l2-compliance (85ec2917 "v4l2-compliance: fix > > stateful encoder tests") against the Raspberry Pi vendor 5.4 kernel > > tree. That means the kernel isn't supporting > > V4L2_BUF_FLAG_NO_CACHE_[INVALIDATE|CLEAN] flags on the buffers, nor is > > anything advertising V4L2_BUF_CAP_SUPPORTS_MMAP_CACHE_HINTS. I'm > > getting failures in setupMmap. > > > > With patch "784be6a v4l2-utils: test cache_hints for MMAP queues" the > > test deliberately sets V4L2_BUF_FLAG_NO_CACHE_INVALIDATE and > > V4L2_BUF_FLAG_NO_CACHE_CLEAN on the buffers that it queues. > > Nothing in the kernel is expecting to manipulate those flags as the > > behaviour wasn't really defined before, but the test is failed if they > > aren't cleared. The v4l2 core would only clear them if the kernel > > includes the patch "f5f5fa7 media: videobuf2: handle V4L2 buffer cache > > flags", which currently means only linuxtv/master as that patch isn't > > even in 5.8-rc3. > > > > Is it really valid to look at the V4L2_BUF_FLAG_NO_CACHE_* flags at > > all if V4L2_BUF_CAP_SUPPORTS_MMAP_CACHE_HINTS isn't set? If the > > capability isn't advertised, surely the state of those bits is > > undefined. > > v4l2-compliance tests are in sync with our master and it expects that > that kernel is used. Thanks, I'd never noted that restriction. All previous times I'd used v4l2-compliance against any kernel it had performed as expected. This is the first change that causes a major failure of tests due to an older kernel. > It is possible in rare cases to add a version check (use the version > field returned by QUERYCAP) to determine if a feature is in the latest > kernel. It's not done anywhere in the current compliance test, but I'd > accept a patch that disables this test for kernels < 5.9 (since this feature > will land in 5.9). OK, I'll try to sort out a patch for you. Dave > Regards, > > Hans
