2016q2 update based on drm-intel kernel 4.3.0

Signed-off-by: libo zhu <libo.zhu@intel.com>

arch/x86/kvm/mmu.c

@@ -2518,6 +2518,7 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
 
 #ifdef CONFIG_KVMGT
 		if (kvmgt_gfn_is_write_protected(vcpu->kvm, gfn)) {
+			spte &= ~(PT_WRITABLE_MASK | SPTE_MMU_WRITEABLE);
 			ret = 1;
 			goto set_pte;
 		}

drivers/gpu/drm/i915/i915_gem_vgtbuffer.c

@@ -185,7 +185,7 @@ static int vgt_decode_information(struct drm_device *dev,
 		return -EINVAL;
 	}
 
-	args->size = (((args->width * args->height * args->bpp) / 8) +
+	args->size = (((args->stride * args->height * args->bpp) / 8) +
 		      (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 
 	if (args->start & (PAGE_SIZE - 1)) {

drivers/gpu/drm/i915/intel_pm.c

@@ -4219,38 +4219,74 @@ static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val)
 		return;
 
 	/* Note the units here are not exactly 1us, but 1280ns. */
-	switch (new_power) {
-	case LOW_POWER:
-		/* Upclock if more than 95% busy over 16ms */
-		ei_up = 16000;
-		threshold_up = 95;
+	if (!i915_host_mediate) {
+		switch (new_power) {
+		case LOW_POWER:
+			/* Upclock if more than 95% busy over 16ms */
+			ei_up = 16000;
+			threshold_up = 95;
+
+			/* Downclock if less than 85% busy over 32ms */
+			ei_down = 32000;
+			threshold_down = 85;
+			break;
 
-		/* Downclock if less than 85% busy over 32ms */
-		ei_down = 32000;
-		threshold_down = 85;
-		break;
+		case BETWEEN:
+			/* Upclock if more than 90% busy over 13ms */
+			ei_up = 13000;
+			threshold_up = 90;
 
-	case BETWEEN:
-		/* Upclock if more than 90% busy over 13ms */
-		ei_up = 13000;
-		threshold_up = 90;
+			/* Downclock if less than 75% busy over 32ms */
+			ei_down = 32000;
+			threshold_down = 75;
+			break;
 
-		/* Downclock if less than 75% busy over 32ms */
-		ei_down = 32000;
-		threshold_down = 75;
-		break;
+		case HIGH_POWER:
+			/* Upclock if more than 85% busy over 10ms */
+			ei_up = 10000;
+			threshold_up = 85;
 
-	case HIGH_POWER:
-		/* Upclock if more than 85% busy over 10ms */
-		ei_up = 10000;
-		threshold_up = 85;
+			/* Downclock if less than 60% busy over 32ms */
+			ei_down = 32000;
+			threshold_down = 60;
+			break;
+		}
 
-		/* Downclock if less than 60% busy over 32ms */
-		ei_down = 32000;
-		threshold_down = 60;
-		break;
-	}
+	} else {
+		/* add separated rps threshold policy in GVT-g host */
+		switch (new_power) {
+		case LOW_POWER:
+			/* Upclock if more than 80% busy over 16ms */
+			ei_up = 16000;
+			threshold_up = 80;
+
+			/* Downclock if less than 60% busy over 32ms */
+			ei_down = 32000;
+			threshold_down = 60;
+			break;
+
+		case BETWEEN:
+			/* Upclock if more than 75% busy over 13ms */
+			ei_up = 13000;
+			threshold_up = 75;
+
+			/* Downclock if less than 55% busy over 32ms */
+			ei_down = 32000;
+			threshold_down = 55;
+			break;
 
+		case HIGH_POWER:
+			/* Upclock if more than 70% busy over 10ms */
+			ei_up = 10000;
+			threshold_up = 70;
+
+			/* Downclock if less than 40% busy over 32ms */
+			ei_down = 32000;
+			threshold_down = 40;
+			break;
+		}
+
+	}
 	I915_WRITE(GEN6_RP_UP_EI,
 		GT_INTERVAL_FROM_US(dev_priv, ei_up));
 	I915_WRITE(GEN6_RP_UP_THRESHOLD,

drivers/gpu/drm/i915/vgt/aperture_gm.c

@@ -150,6 +150,11 @@ void rsvd_aperture_free(struct pgt_device *pdev, unsigned long start, unsigned l
 	}
 }
 
+void rsvd_aperture_runout_handler(struct pgt_device *pdev)
+{
+	pdev->dummy_vm_switch = true;
+}
+
 ssize_t get_avl_vm_aperture_gm_and_fence(struct pgt_device *pdev, char *buf,
 		ssize_t buf_sz)
 {

drivers/gpu/drm/i915/vgt/cmd_parser.c

@@ -128,6 +128,8 @@ static inline int add_patch_entry(struct parser_exec_state *s,
 	struct cmd_patch_info *patch;
 	int next;
 
+	ASSERT(s->shadow != INDIRECT_CTX_SHADOW);
+
 	if (addr == NULL) {
 		vgt_err("VM(%d) CMD_SCAN: NULL address to be patched\n",
 				s->vgt->vgt_id);
@@ -160,6 +162,8 @@ static inline int add_post_handle_entry(struct parser_exec_state *s,
 	struct cmd_handler_info *entry;
 	int next;
 
+	ASSERT(s->shadow != INDIRECT_CTX_SHADOW);
+
 	next = get_next_entry(list);
 	if (next == list->count) {
 		vgt_err("CMD_SCAN: no free post-handle entry\n");
@@ -191,6 +195,8 @@ static int add_tail_entry(struct parser_exec_state *s,
 	struct cmd_tail_info *entry;
 	int next;
 
+	ASSERT(s->shadow != INDIRECT_CTX_SHADOW);
+
 	next = get_next_entry(list);
 	if (next == list->count) {
 		vgt_err("CMD_SCAN: no free tail entry\n");
@@ -994,10 +1000,20 @@ static inline unsigned long vgt_get_gma_from_bb_start(
 	if (g_gm_is_valid(vgt, ip_gma)) {
 		bb_start_gma = 0;
 		va = vgt_gma_to_va(vgt->gtt.ggtt_mm, ip_gma);
+		if (va == NULL) {
+			vgt_err("VM-%d(ring %d>: Failed to get va of guest gma 0x%lx!\n",
+				vgt->vm_id, ring_id, ip_gma);
+			return 0;
+		}
 		hypervisor_read_va(vgt, va, &cmd, 4, 1);
 		opcode = vgt_get_opcode(cmd, ring_id);
 		ASSERT(opcode == OP_MI_BATCH_BUFFER_START);
 		va = vgt_gma_to_va(vgt->gtt.ggtt_mm, ip_gma + 4);
+		if (va == NULL) {
+			vgt_err("VM-%d(ring %d>: Failed to get va of guest gma 0x%lx!\n",
+				vgt->vm_id, ring_id, ip_gma + 4);
+			return 0;
+		}
 		hypervisor_read_va(vgt, va, &bb_start_gma, 4, 1);
 	} else if (g_gm_is_reserved(vgt, ip_gma)) {
 		va = v_aperture(vgt->pdev, ip_gma);
@@ -1357,8 +1373,7 @@ static int vgt_handle_mi_display_flip(struct parser_exec_state *s)
 	}
 
 	{
-		if (!vgt_flip_parameter_check(s, plane_code, stride_val, surf_val))
-			goto wrong_command;
+		vgt_flip_parameter_check(s, plane_code, stride_val, surf_val);
 
 		GET_INFO_FOR_FLIP(pipe, plane,
 			ctrl_reg, surf_reg, stride_reg, stride_mask);
@@ -1418,7 +1433,7 @@ static int vgt_handle_mi_display_flip(struct parser_exec_state *s)
 			((value & ~plane_select_mask) |
 			 (real_plane_code << plane_select_shift)));
 
-		vgt_inject_flip_done(s->vgt, pipe);
+		vgt_inject_flip_done(s->vgt, pipe, plane);
 
 		return 0;
 	}
@@ -1432,7 +1447,7 @@ static int vgt_handle_mi_display_flip(struct parser_exec_state *s)
 
 	rc |= add_patch_entry(s, cmd_ptr(s, 0), MI_NOOP);
 
-	vgt_inject_flip_done(s->vgt, pipe);
+	vgt_inject_flip_done(s->vgt, pipe, plane);
 
 	return rc;
 
@@ -1771,6 +1786,7 @@ static int batch_buffer_needs_scan(struct parser_exec_state *s)
 	return 1;
 }
 
+#define LITE_RESTORE_FLOOD_THRESHOLD 1000
 static int vgt_perform_bb_shadow(struct parser_exec_state *s)
 {
 	struct vgt_device *vgt = s->vgt;
@@ -1835,7 +1851,9 @@ static int vgt_perform_bb_shadow(struct parser_exec_state *s)
 		s_cmd_page->guest_gma = bb_guest_gma;
 		s_cmd_page->bound_gma = shadow_gma;
 
-		s->el_ctx->shadow_priv_bb.n_pages ++;
+		if (s->el_ctx->shadow_priv_bb.n_pages++ > LITE_RESTORE_FLOOD_THRESHOLD)
+			rsvd_aperture_runout_handler(vgt->pdev);
+
 		list_add_tail(&s_cmd_page->list,
 			      &s->el_ctx->shadow_priv_bb.pages);
 
@@ -2783,7 +2801,7 @@ static inline bool gma_out_of_range(unsigned long gma, unsigned long gma_head, u
 #define MAX_PARSER_ERROR_NUM	10
 
 static int __vgt_scan_vring(struct vgt_device *vgt, int ring_id, vgt_reg_t head,
-			vgt_reg_t tail, vgt_reg_t base, vgt_reg_t size, bool shadow)
+			vgt_reg_t tail, vgt_reg_t base, vgt_reg_t size, cmd_shadow_t shadow)
 {
 	unsigned long gma_head, gma_tail, gma_bottom;
 	struct parser_exec_state s;
@@ -2869,7 +2887,7 @@ static int __vgt_scan_vring(struct vgt_device *vgt, int ring_id, vgt_reg_t head,
 		}
 	}
 
-	if (!rc) {
+	if (!rc && shadow != INDIRECT_CTX_SHADOW) {
 		/*
 		 * Set flag to indicate the command buffer is end with user interrupt,
 		 * and save the instruction's offset in ring buffer.
@@ -3061,7 +3079,8 @@ int vgt_scan_vring(struct vgt_device *vgt, int ring_id)
 	if (ret == 0) {
 		ret = __vgt_scan_vring(vgt, ring_id, rs->last_scan_head,
 			vring->tail & RB_TAIL_OFF_MASK, rb_base,
-			_RING_CTL_BUF_SIZE(vring->ctl), shadow_cmd_buffer);
+			_RING_CTL_BUF_SIZE(vring->ctl),
+			shadow_cmd_buffer ? NORMAL_CMD_SHADOW : NO_CMD_SHADOW);
 
 		rs->last_scan_head = vring->tail;
 	}
@@ -3084,7 +3103,7 @@ int vgt_scan_vring(struct vgt_device *vgt, int ring_id)
 			if (ret)
 				goto err;
 			if (!__vgt_scan_vring(vgt, ring_id, 0, ctx_tail,
-				ctx_base, dummy_ctx_size, true)) {
+				ctx_base, dummy_ctx_size, INDIRECT_CTX_SHADOW)) {
 				vgt_get_bb_per_ctx_shadow_base(vgt, rs->el_ctx);
 			} else {
 				ret = -1;

drivers/gpu/drm/i915/vgt/cmd_parser.h

@@ -439,6 +439,12 @@ typedef enum{
 	PPGTT_BUFFER
 }gtt_addr_t;
 
+typedef enum {
+	NO_CMD_SHADOW = 0,
+	NORMAL_CMD_SHADOW = 1,
+	INDIRECT_CTX_SHADOW = 2,
+} cmd_shadow_t;
+
 struct parser_exec_state {
 	struct vgt_device *vgt;
 	int ring_id;
@@ -487,7 +493,7 @@ struct parser_exec_state {
 	void *ip_buf;
 
 	struct execlist_context *el_ctx;
-	bool shadow;
+	cmd_shadow_t shadow;
 };
 
 #define CMD_TAIL_NUM	1024

drivers/gpu/drm/i915/vgt/debugfs.c

@@ -685,7 +685,7 @@ static void vgt_dump_dpy_mmio(struct seq_file *m, struct pgt_device *pdev,
 		seq_printf(m, "\t\teDP select: %s\n", str);
 	}
 	seq_printf(m, "\n");
-	
+
 	if (is_current_display_owner(vgt)) {
 		return;
 	}
@@ -1397,6 +1397,12 @@ int vgt_create_debugfs(struct vgt_device *vgt)
 		debugfs_create_u64_node ("gpt_find_miss_cnt", 0444, perf_dir_entry, &(vgt->stat.gpt_find_miss_cnt));
 		debugfs_create_u64_node ("gpt_find_miss_cycles", 0444, perf_dir_entry, &(vgt->stat.gpt_find_miss_cycles));
 		debugfs_create_u64_node ("skip_bb_cnt", 0444, perf_dir_entry, &(vgt->stat.skip_bb_cnt));
+		debugfs_create_u64_node ("shadow_last_level_page_cnt", 0444, perf_dir_entry, &(vgt->stat.shadow_last_level_page_cnt));
+		debugfs_create_u64_node ("shadow_last_level_page_cycles", 0444, perf_dir_entry, &(vgt->stat.shadow_last_level_page_cycles));
+		debugfs_create_u64_node ("oos_page_cnt", 0444, perf_dir_entry, &(vgt->stat.oos_page_cnt));
+		debugfs_create_u64_node ("oos_page_cycles", 0444, perf_dir_entry, &(vgt->stat.oos_page_cycles));
+		debugfs_create_u64_node ("oos_pte_cnt", 0444, perf_dir_entry, &(vgt->stat.oos_pte_cnt));
+		debugfs_create_u64_node ("oos_pte_cycles", 0444, perf_dir_entry, &(vgt->stat.oos_pte_cycles));
 
 		/* cmd statistics for ring/batch buffers */
 		cmdstat_dir_entry = debugfs_create_dir("ring", perf_dir_entry);
@@ -1411,6 +1417,38 @@ int vgt_create_debugfs(struct vgt_device *vgt)
 	return 0;
 }
 
+#define VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(runtime_module_param, runtime_dir_entry, target_module_param) \
+	debugfs_create_symlink(#runtime_module_param, runtime_dir_entry, "/sys/module/i915/parameters/" #target_module_param)
+
+void vgt_debugfs_symlink_module_param(void)
+{
+	struct dentry *runtime_dir_entry = debugfs_create_dir("runtime", d_vgt_debug);
+	if (!runtime_dir_entry) {
+		vgt_err("Failed to create debugfs directory: runtime\n");
+		return ;
+	}
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(vgt_debug, runtime_dir_entry, debug);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(tbs_period_ms, runtime_dir_entry, tbs_period_ms);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(render_engine_reset, runtime_dir_entry, render_engine_reset);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(propagate_monitor_to_guest, runtime_dir_entry, propagate_monitor_to_guest);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(preallocated_shadow_pages, runtime_dir_entry, preallocated_shadow_pages);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(preallocated_oos_pages, runtime_dir_entry, preallocated_oos_pages);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(spt_out_of_sync, runtime_dir_entry, spt_out_of_sync);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(enable_video_switch, runtime_dir_entry, enable_video_switch);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(bypass_scan_mask, runtime_dir_entry, bypass_scan);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(bypass_dom0_addr_check, runtime_dir_entry, bypass_dom0_addr_check);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(enable_panel_fitting, runtime_dir_entry, enable_panel_fitting);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(enable_reset, runtime_dir_entry, enable_reset);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(preemption_policy, runtime_dir_entry, preemption_policy);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(reset_count_threshold, runtime_dir_entry, reset_count_threshold);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(reset_dur_threshold, runtime_dir_entry, reset_dur_threshold);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(reset_max_threshold, runtime_dir_entry, reset_max_threshold);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(shadow_ctx_check, runtime_dir_entry, shadow_ctx_check);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(shadow_indirect_ctx_bb, runtime_dir_entry, shadow_indirect_ctx_bb);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(vgt_cmd_audit, runtime_dir_entry, vgt_cmd_audit);
+	VGT_CREATE_SYMLINK_FOR_MODULE_PARAM(vgt_hold_forcewake, runtime_dir_entry, vgt_hold_forcewake);
+}
+
 /* debugfs_remove_recursive has no return value, this fuction
  * also return nothing */
 void vgt_destroy_debugfs(struct vgt_device *vgt)