gpu: nvgpu: unit: increase hal.fb test coverage

This patch adds some corner cases tests to increase coverage.

JIRA NVGPU-4413

Change-Id: Ifb58c595a0953fdc91580e3d491a1f04143ad9f6
Signed-off-by: Nicolas Benech <nbenech@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2293847
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-misra <svc-mobile-misra@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit

userspace/units/fb/fb_fusa.h

@@ -175,7 +175,7 @@ int fb_mmu_fault_gv11b_init_test(struct unit_module *m, struct gk20a *g,
  * gv11b_fb_fault_buffer_size_val, gv11b_fb_read_mmu_fault_inst_lo_hi,
  * gv11b_fb_read_mmu_fault_info
  *
- * Test Type: Feature
+ * Test Type: Feature, Error injection
  *
  * Input: fb_mmu_fault_gv11b_init_test
  *
@@ -187,6 +187,8 @@ int fb_mmu_fault_gv11b_init_test(struct unit_module *m, struct gk20a *g,
  *   empty.
  * - Call the gv11b_fb_fault_buf_configure_hw HAL and enable fault buffer.
  * - Enable fault buffer again which shouldn't cause any crash.
+ * - While trying to disable the fault buffer, trigger a failure of
+ *   nvgpu_timeout_init.
  * - Disable the fault buffer.
  * - Enable fault buffer, set the busy bit in fb_mmu_fault_status_r register,
  *   disable the fault buffer which should cause an internal timeout. Ensure
@@ -273,7 +275,7 @@ int fb_mmu_fault_gv11b_handle_fault(struct unit_module *m, struct gk20a *g,
  * Targets: gv11b_fb_handle_bar2_fault, gv11b_fb_mmu_fault_info_dump,
  * gv11b_fb_fault_buf_set_state_hw
  *
- * Test Type: Feature
+ * Test Type: Feature, Error injection
  *
  * Input: fb_mmu_fault_gv11b_init_test
  *
@@ -286,6 +288,8 @@ int fb_mmu_fault_gv11b_handle_fault(struct unit_module *m, struct gk20a *g,
  *   and a pointer to the channel)
  * - Call the gv11b_fb_mmu_fault_info_dump and ensure it doesn't cause a crash.
  * - Set the fault_status to non-replayable and call gv11b_fb_handle_bar2_fault.
+ * - Set the g->ops.bus.bar2_bind HAL to report a failure and call
+ *   gv11b_fb_handle_bar2_fault again.
  * - Repeat with the fault buffer disabled.
  *
  * Output: Returns PASS if the steps above were executed successfully. FAIL

userspace/units/fb/fb_gm20b_fusa.c

@@ -169,6 +169,14 @@ int fb_gm20b_mmu_ctrl_test(struct unit_module *m, struct gk20a *g, void *args)
 			"vpr_info_fetch did not fail as expected (1)\n");
 	}
 
+	nvgpu_posix_enable_fault_injection(timer_fi, true, 1);
+	err = g->ops.fb.vpr_info_fetch(g);
+	nvgpu_posix_enable_fault_injection(timer_fi, false, 0);
+	if (err != -ETIMEDOUT) {
+		unit_return_fail(m,
+			"vpr_info_fetch did not fail as expected (2)\n");
+	}
+
 	/*
 	 * Trigger timeout in the gm20b_fb_vpr_info_fetch_wait function on
 	 * fb_mmu_vpr_info_fetch_v(val) == fb_mmu_vpr_info_fetch_false_v()
@@ -177,7 +185,7 @@ int fb_gm20b_mmu_ctrl_test(struct unit_module *m, struct gk20a *g, void *args)
 	err = g->ops.fb.vpr_info_fetch(g);
 	if (err != -ETIMEDOUT) {
 		unit_return_fail(m,
-			"vpr_info_fetch did not fail as expected (2)\n");
+			"vpr_info_fetch did not fail as expected (3)\n");
 	}
 
 	return UNIT_SUCCESS;

userspace/units/fb/fb_mmu_fault_gv11b_fusa.c

@@ -59,6 +59,11 @@ static int hal_bar2_bind_nop(struct gk20a *g, struct nvgpu_mem *bar2_inst)
 	return 0;
 }
 
+static int hal_bar2_bind_fail(struct gk20a *g, struct nvgpu_mem *bar2_inst)
+{
+	return -1;
+}
+
 static u32 hal_fifo_mmu_fault_id_to_pbdma_id(struct gk20a *g, u32 mmu_fault_id)
 {
 	return INVAL_ID;
@@ -107,6 +112,8 @@ int fb_mmu_fault_gv11b_buffer_test(struct unit_module *m, struct gk20a *g,
 	u32 get_idx;
 	u32 val;
 	u32 lo, hi;
+	struct nvgpu_posix_fault_inj *timers_fi =
+		nvgpu_timers_get_fault_injection();
 
 	if (g->ops.fb.is_fault_buf_enabled(g, 0)) {
 		unit_return_fail(m, "fault buffer not disabled as expected\n");
@@ -134,6 +141,11 @@ int fb_mmu_fault_gv11b_buffer_test(struct unit_module *m, struct gk20a *g,
 	/* Enabling again shouldn't cause an issue */
 	g->ops.fb.fault_buf_set_state_hw(g, 0, NVGPU_MMU_FAULT_BUF_ENABLED);
 
+	/* Make nvgpu_timeout_init fail during disable operation */
+	nvgpu_posix_enable_fault_injection(timers_fi, true, 0);
+	g->ops.fb.fault_buf_set_state_hw(g, 0, NVGPU_MMU_FAULT_BUF_DISABLED);
+	nvgpu_posix_enable_fault_injection(timers_fi, false, 0);
+
 	/* Disable */
 	g->ops.fb.fault_buf_set_state_hw(g, 0, NVGPU_MMU_FAULT_BUF_DISABLED);
 
@@ -247,7 +259,6 @@ int fb_mmu_fault_gv11b_handle_fault(struct unit_module *m, struct gk20a *g,
 	/* Same case but ensure fault status register is also set properly */
 	nvgpu_writel(g, fb_mmu_fault_status_r(),
 		fb_mmu_fault_status_non_replayable_overflow_m());
-	nvgpu_writel(g, fb_mmu_fault_status_r(), 0);
 	gv11b_fb_handle_mmu_fault(g, niso_intr);
 	if (!helper_is_intr_cleared(g)) {
 		unit_return_fail(m, "unhandled interrupt (5)\n");
@@ -257,7 +268,6 @@ int fb_mmu_fault_gv11b_handle_fault(struct unit_module *m, struct gk20a *g,
 	nvgpu_writel(g, fb_mmu_fault_status_r(),
 		fb_mmu_fault_status_non_replayable_overflow_m() |
 		fb_mmu_fault_status_non_replayable_getptr_corrupted_m());
-	nvgpu_writel(g, fb_mmu_fault_status_r(), 0);
 	gv11b_fb_handle_mmu_fault(g, niso_intr);
 	if (!helper_is_intr_cleared(g)) {
 		unit_return_fail(m, "unhandled interrupt (6)\n");
@@ -298,6 +308,12 @@ int fb_mmu_fault_gv11b_handle_bar2_fault(struct unit_module *m, struct gk20a *g,
 	g->ops.fb.fault_buf_set_state_hw(g, 0, NVGPU_MMU_FAULT_BUF_ENABLED);
 	gv11b_fb_handle_bar2_fault(g, &mmufault, fault_status);
 
+	/* Case where g->ops.bus.bar2_bind fails */
+	g->ops.bus.bar2_bind = hal_bar2_bind_fail;
+	g->ops.fb.fault_buf_set_state_hw(g, 0, NVGPU_MMU_FAULT_BUF_ENABLED);
+	gv11b_fb_handle_bar2_fault(g, &mmufault, fault_status);
+	g->ops.bus.bar2_bind = hal_bar2_bind_nop;
+
 	/* Case where fault buffer is not enabled */
 	g->ops.fb.fault_buf_set_state_hw(g, 0, NVGPU_MMU_FAULT_BUF_DISABLED);
 	gv11b_fb_handle_bar2_fault(g, &mmufault, fault_status);