diff --git a/drivers/gpu/nvgpu/include/nvgpu/preempt.h b/drivers/gpu/nvgpu/include/nvgpu/preempt.h
index 863d6ca44..2cfa7cd38 100644
--- a/drivers/gpu/nvgpu/include/nvgpu/preempt.h
+++ b/drivers/gpu/nvgpu/include/nvgpu/preempt.h
@@ -24,13 +24,52 @@
 
 #include <nvgpu/types.h>
 
+/**
+ * @file
+ *
+ * Preemption interface.
+ */
 struct gk20a;
 struct nvgpu_channel;
 struct nvgpu_tsg;
 
+/**
+ * @brief Get preemption timeout (ms). This timeout is defined by s/w.
+ *
+ * @param g[in]		The GPU driver struct to query preempt timeout for.
+ *
+ * @return Maximum amount of time in ms to wait for preemption completion,
+ *         i.e. context non resident on PBDMAs and engines.
+ */
 u32 nvgpu_preempt_get_timeout(struct gk20a *g);
+
+/**
+ * @brief Preempts TSG if channel is bound to TSG.
+ *
+ * @param g[in]		The GPU driver struct which owns this channel.
+ * @param ch[in]	Pointer to channel to be preempted.
+ *
+ * Preempts TSG if channel is bound to TSG. Preemption implies that the
+ * context's state is saved out and also that the context cannot remain parked
+ * either in Host or in any engine.
+ *
+ * After triggering a preempt request for channel's TSG, pbdmas and engines
+ * are polled to make sure preemption completed, i.e. context is not loaded
+ * on any pbdma or engine.
+ *
+ * @return 0 in case of success, <0 in case of failure
+ * @retval 0 if channel was not bound to TSG.
+ * @retval 0 if TSG was not loaded on pbdma or engine.
+ * @retval 0 if TSG was loaded (pbdma or engine) and could be preempted.
+ * @retval non-zero value if preemption did not complete within s/w defined
+ *         timeout.
+ */
 int nvgpu_preempt_channel(struct gk20a *g, struct nvgpu_channel *ch);
 
+/**
+ * Called from recovery handling for volta onwards. This will
+ * not be part of safety build after recovery is not supported in safety build.
+ */
 void nvgpu_preempt_poll_tsg_on_pbdma(struct gk20a *g,
 		struct nvgpu_tsg *tsg);
 #endif /* NVGPU_PREEMPT_H */