On Wed, Aug 22, 2018 at 10:54:05AM +0200, Daniel Vetter wrote:
For buffer sharing, use dma-buf instead. We can't set smem_start to 0 unconditionally since that's used by the fbdev mmap default implementation. And we have plenty of userspace which would like to keep that working.
This might break legit userspace - if it does we need to look at a case-by-cases basis how to handle that. Worst case I expect overrides for only specific drivers, since anything remotely modern should be using dma-buf/prime now (which is about 7 years old now for DRM drivers).
This issue was uncovered because Noralf's rework to implement a generic fb_probe also implements it's own fb_mmap callback. Which means smem_start didn't have to be set anymore, which blew up some blob in userspace rather badly.
Cc: Gustavo Padovan gustavo@padovan.org Cc: Maarten Lankhorst maarten.lankhorst@linux.intel.com Cc: Sean Paul sean@poorly.run Cc: David Airlie airlied@linux.ie Cc: John Stultz john.stultz@linaro.org Cc: Noralf Trønnes noralf@tronnes.org Signed-off-by: Daniel Vetter daniel.vetter@intel.com
After clarifying on IRC, I think this makes sense. The chance to break userspace requires a very specific configuration that I don't expect will be common.
Acked-by: Sean Paul seanpaul@chromium.org
drivers/gpu/drm/drm_fb_helper.c | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/drivers/gpu/drm/drm_fb_helper.c b/drivers/gpu/drm/drm_fb_helper.c index 4b0dd20bccb8..bcb78693c4f7 100644 --- a/drivers/gpu/drm/drm_fb_helper.c +++ b/drivers/gpu/drm/drm_fb_helper.c @@ -2673,6 +2673,8 @@ __drm_fb_helper_initial_config_and_unlock(struct drm_fb_helper *fb_helper,
info = fb_helper->fbdev; info->var.pixclock = 0;
/* don't leak any physical addresses to userspace */
info->flags |= FBINFO_HIDE_SMEM_START;
/* Need to drop locks to avoid recursive deadlock in
- register_framebuffer. This is ok because the only thing left to do is
-- 2.18.0