On 09-11-2016 20:28, Emil Velikov wrote: > From: Emil Velikov <[email protected]> > > Currently the drmGetDevice[s] API parses the config sysfs file to > retrieve the revision field. > > That is required since there's no separate file (nor a libudev call > afaict) that can be used. At the same time doing so causes the device to > be awaken, so if an application creates/destroys a GL context multiple > times they can observe noticeable delays. 2-3s in the case of > firefox/thunderbird/chromium + radeon module. > > There's a kernel patch on the PCI mailing list, but until then we can > use drmDeviceUseRevisionFile() which will use separate sysfs files and > default to 0 if the revision one is missing. > > Since we don't care about the revision id, we can use that. > > v2: Add second invocation deep within loader_get_user_preferred_fd() > > Cc: "13.0" <[email protected]> > Cc: Mauro Santos <[email protected]> > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98502 > Signed-off-by: Emil Velikov <[email protected]> > --- > Just because some platforms use loader_get_user_preferred_fd while > others do not. That in itself happens prior to any other loader_get_* > call > --- > configure.ac | 2 +- > src/loader/loader.c | 2 ++ > 2 files changed, 3 insertions(+), 1 deletion(-) > > diff --git a/configure.ac b/configure.ac > index d215b63..f529f2cf 100644 > --- a/configure.ac > +++ b/configure.ac > @@ -68,7 +68,7 @@ OPENCL_VERSION=1 > AC_SUBST([OPENCL_VERSION]) > > dnl Versions for external dependencies > -LIBDRM_REQUIRED=2.4.66 > +LIBDRM_REQUIRED=2.4.72 > LIBDRM_RADEON_REQUIRED=2.4.56 > LIBDRM_AMDGPU_REQUIRED=2.4.63 > LIBDRM_INTEL_REQUIRED=2.4.61 > diff --git a/src/loader/loader.c b/src/loader/loader.c > index fe90307..a69d583 100644 > --- a/src/loader/loader.c > +++ b/src/loader/loader.c > @@ -145,6 +145,7 @@ static char *drm_get_id_path_tag_for_fd(int fd) > drmDevicePtr device; > char *tag; > > + drmDeviceUseRevisionFile(); > if (drmGetDevice(fd, &device) != 0) > return NULL; > > @@ -275,6 +276,7 @@ drm_get_pci_id_for_fd(int fd, int *vendor_id, int > *chip_id) > drmDevicePtr device; > int ret; > > + drmDeviceUseRevisionFile(); > if (drmGetDevice(fd, &device) == 0) { > if (device->bustype == DRM_BUS_PCI) { > *vendor_id = device->deviceinfo.pci->vendor_id; >
I've omitted the version hunk and as far as I can tell this one works fine. I see no delays when starting glxgears/etc. With a patched kernel drmdevice returns the correct revision, with an unpatched kernel it returns 0 as expected. -- Mauro Santos _______________________________________________ mesa-dev mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/mesa-dev
