We expose engine properties under sysfs so that the sysadmin can
configure the driver according to their requirements. We can also use
udev rules to then apply that configuration anytime a device is
reloaded. Include a udev snippet provided by Joonas as an example.
v2: Update the snippet
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Cc: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx>
---
drivers/gpu/drm/i915/gt/sysfs_engines.c | 16 ++++++++++++++++
1 file changed, 16 insertions(+)
diff --git a/drivers/gpu/drm/i915/gt/sysfs_engines.c b/drivers/gpu/drm/i915/gt/sysfs_engines.c
index 9cf943b6a44b..37f078e69a42 100644
--- a/drivers/gpu/drm/i915/gt/sysfs_engines.c
+++ b/drivers/gpu/drm/i915/gt/sysfs_engines.c
@@ -11,6 +11,22 @@
#include "intel_engine_heartbeat.h"
#include "sysfs_engines.h"
+/*
+ * The sysfs provides a means for configuring each engine for the intended
+ * usecase, and by utilising a udev the configuration can be made persistent
+ * across reboots and device rebinding.
+ *
+ * An example udev rule to run a custom sysadmin script would be,
+ * /etc/udev/rules.d/50-intel-gpu.rules:
+ *
+ * ACTION=="bind|add",SUBSYSTEM=="pci",DRIVER=="i915",RUN+="/usr/local/libexec/i915-disable-preempt-timeout.sh"
+ *
+ * where the script receives the device name and can open the sysfs, e.g.:
+ * for ENGINE in $(find /sys/$DEVPATH/drm/card?/engine -maxdepth 1) do
+ * echo 0 > $ENGINE/preempt_timeout_ms # Disable reset on preemption failure
+ * done
+ */
+
struct kobj_engine {
struct kobject base;
struct intel_engine_cs *engine;
--
2.20.1
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
