Merge tag 'drm-intel-fixes-2023-03-30' of git://anongit.freedesktop.org/drm/drm-intel into drm-fixes

drm/i915 fixes for v6.3-rc5:
- Fix PMU support by reusing functions with sysfs
- Fix a number of issues related to color, PSR and arm/noarm
- Fix state check related to ICL PHY ownership check in TC-cold state
- Flush lmem contents after construction
- Fix hibernate oops related to DPT BO
- Fix perf stream error path wakeref balance

Signed-off-by: Daniel Vetter <[email protected]>
From: Jani Nikula <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]

Author: danvet

drivers/gpu/drm/i915/display/intel_color.c

@@ -46,6 +46,11 @@ struct intel_color_funcs {
 	 * registers involved with the same commit.
 	 */
 	void (*color_commit_arm)(const struct intel_crtc_state *crtc_state);
+	/*
+	 * Perform any extra tasks needed after all the
+	 * double buffered registers have been latched.
+	 */
+	void (*color_post_update)(const struct intel_crtc_state *crtc_state);
 	/*
 	 * Load LUTs (and other single buffered color management
 	 * registers). Will (hopefully) be called during the vblank
@@ -614,9 +619,33 @@ static void ilk_lut_12p4_pack(struct drm_color_lut *entry, u32 ldw, u32 udw)
 
 static void icl_color_commit_noarm(const struct intel_crtc_state *crtc_state)
 {
+	/*
+	 * Despite Wa_1406463849, ICL no longer suffers from the SKL
+	 * DC5/PSR CSC black screen issue (see skl_color_commit_noarm()).
+	 * Possibly due to the extra sticky CSC arming
+	 * (see icl_color_post_update()).
+	 *
+	 * On TGL+ all CSC arming issues have been properly fixed.
+	 */
 	icl_load_csc_matrix(crtc_state);
 }
 
+static void skl_color_commit_noarm(const struct intel_crtc_state *crtc_state)
+{
+	/*
+	 * Possibly related to display WA #1184, SKL CSC loses the latched
+	 * CSC coeff/offset register values if the CSC registers are disarmed
+	 * between DC5 exit and PSR exit. This will cause the plane(s) to
+	 * output all black (until CSC_MODE is rearmed and properly latched).
+	 * Once PSR exit (and proper register latching) has occurred the
+	 * danger is over. Thus when PSR is enabled the CSC coeff/offset
+	 * register programming will be peformed from skl_color_commit_arm()
+	 * which is called after PSR exit.
+	 */
+	if (!crtc_state->has_psr)
+		ilk_load_csc_matrix(crtc_state);
+}
+
 static void ilk_color_commit_noarm(const struct intel_crtc_state *crtc_state)
 {
 	ilk_load_csc_matrix(crtc_state);
@@ -659,6 +688,9 @@ static void skl_color_commit_arm(const struct intel_crtc_state *crtc_state)
 	enum pipe pipe = crtc->pipe;
 	u32 val = 0;
 
+	if (crtc_state->has_psr)
+		ilk_load_csc_matrix(crtc_state);
+
 	/*
 	 * We don't (yet) allow userspace to control the pipe background color,
 	 * so force it to black, but apply pipe gamma and CSC appropriately
@@ -677,6 +709,47 @@ static void skl_color_commit_arm(const struct intel_crtc_state *crtc_state)
 			  crtc_state->csc_mode);
 }
 
+static void icl_color_commit_arm(const struct intel_crtc_state *crtc_state)
+{
+	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
+	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
+	enum pipe pipe = crtc->pipe;
+
+	/*
+	 * We don't (yet) allow userspace to control the pipe background color,
+	 * so force it to black.
+	 */
+	intel_de_write(i915, SKL_BOTTOM_COLOR(pipe), 0);
+
+	intel_de_write(i915, GAMMA_MODE(crtc->pipe),
+		       crtc_state->gamma_mode);
+
+	intel_de_write_fw(i915, PIPE_CSC_MODE(crtc->pipe),
+			  crtc_state->csc_mode);
+}
+
+static void icl_color_post_update(const struct intel_crtc_state *crtc_state)
+{
+	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
+	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
+
+	/*
+	 * Despite Wa_1406463849, ICL CSC is no longer disarmed by
+	 * coeff/offset register *writes*. Instead, once CSC_MODE
+	 * is armed it stays armed, even after it has been latched.
+	 * Afterwards the coeff/offset registers become effectively
+	 * self-arming. That self-arming must be disabled before the
+	 * next icl_color_commit_noarm() tries to write the next set
+	 * of coeff/offset registers. Fortunately register *reads*
+	 * do still disarm the CSC. Naturally this must not be done
+	 * until the previously written CSC registers have actually
+	 * been latched.
+	 *
+	 * TGL+ no longer need this workaround.
+	 */
+	intel_de_read_fw(i915, PIPE_CSC_PREOFF_HI(crtc->pipe));
+}
+
 static struct drm_property_blob *
 create_linear_lut(struct drm_i915_private *i915, int lut_size)
 {
@@ -1373,6 +1446,14 @@ void intel_color_commit_arm(const struct intel_crtc_state *crtc_state)
 	i915->display.funcs.color->color_commit_arm(crtc_state);
 }
 
+void intel_color_post_update(const struct intel_crtc_state *crtc_state)
+{
+	struct drm_i915_private *i915 = to_i915(crtc_state->uapi.crtc->dev);
+
+	if (i915->display.funcs.color->color_post_update)
+		i915->display.funcs.color->color_post_update(crtc_state);
+}
+
 void intel_color_prepare_commit(struct intel_crtc_state *crtc_state)
 {
 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
@@ -3064,18 +3145,28 @@ static const struct intel_color_funcs i9xx_color_funcs = {
 	.lut_equal = i9xx_lut_equal,
 };
 
+static const struct intel_color_funcs tgl_color_funcs = {
+	.color_check = icl_color_check,
+	.color_commit_noarm = icl_color_commit_noarm,
+	.color_commit_arm = icl_color_commit_arm,
+	.load_luts = icl_load_luts,
+	.read_luts = icl_read_luts,
+	.lut_equal = icl_lut_equal,
+};
+
 static const struct intel_color_funcs icl_color_funcs = {
 	.color_check = icl_color_check,
 	.color_commit_noarm = icl_color_commit_noarm,
-	.color_commit_arm = skl_color_commit_arm,
+	.color_commit_arm = icl_color_commit_arm,
+	.color_post_update = icl_color_post_update,
 	.load_luts = icl_load_luts,
 	.read_luts = icl_read_luts,
 	.lut_equal = icl_lut_equal,
 };
 
 static const struct intel_color_funcs glk_color_funcs = {
 	.color_check = glk_color_check,
-	.color_commit_noarm = ilk_color_commit_noarm,
+	.color_commit_noarm = skl_color_commit_noarm,
 	.color_commit_arm = skl_color_commit_arm,
 	.load_luts = glk_load_luts,
 	.read_luts = glk_read_luts,
@@ -3084,7 +3175,7 @@ static const struct intel_color_funcs glk_color_funcs = {
 
 static const struct intel_color_funcs skl_color_funcs = {
 	.color_check = ivb_color_check,
-	.color_commit_noarm = ilk_color_commit_noarm,
+	.color_commit_noarm = skl_color_commit_noarm,
 	.color_commit_arm = skl_color_commit_arm,
 	.load_luts = bdw_load_luts,
 	.read_luts = bdw_read_luts,
@@ -3180,7 +3271,9 @@ void intel_color_init_hooks(struct drm_i915_private *i915)
 		else
 			i915->display.funcs.color = &i9xx_color_funcs;
 	} else {
-		if (DISPLAY_VER(i915) >= 11)
+		if (DISPLAY_VER(i915) >= 12)
+			i915->display.funcs.color = &tgl_color_funcs;
+		else if (DISPLAY_VER(i915) == 11)
 			i915->display.funcs.color = &icl_color_funcs;
 		else if (DISPLAY_VER(i915) == 10)
 			i915->display.funcs.color = &glk_color_funcs;

drivers/gpu/drm/i915/display/intel_color.h

@@ -21,6 +21,7 @@ void intel_color_prepare_commit(struct intel_crtc_state *crtc_state);
 void intel_color_cleanup_commit(struct intel_crtc_state *crtc_state);
 void intel_color_commit_noarm(const struct intel_crtc_state *crtc_state);
 void intel_color_commit_arm(const struct intel_crtc_state *crtc_state);
+void intel_color_post_update(const struct intel_crtc_state *crtc_state);
 void intel_color_load_luts(const struct intel_crtc_state *crtc_state);
 void intel_color_get_config(struct intel_crtc_state *crtc_state);
 bool intel_color_lut_equal(const struct intel_crtc_state *crtc_state,

drivers/gpu/drm/i915/display/intel_display.c

@@ -1209,6 +1209,9 @@ static void intel_post_plane_update(struct intel_atomic_state *state,
 	if (needs_cursorclk_wa(old_crtc_state) &&
 	    !needs_cursorclk_wa(new_crtc_state))
 		icl_wa_cursorclkgating(dev_priv, pipe, false);
+
+	if (intel_crtc_needs_color_update(new_crtc_state))
+		intel_color_post_update(new_crtc_state);
 }
 
 static void intel_crtc_enable_flip_done(struct intel_atomic_state *state,
@@ -7091,6 +7094,8 @@ static void intel_update_crtc(struct intel_atomic_state *state,
 
 	intel_fbc_update(state, crtc);
 
+	drm_WARN_ON(&i915->drm, !intel_display_power_is_enabled(i915, POWER_DOMAIN_DC_OFF));
+
 	if (!modeset &&
 	    intel_crtc_needs_color_update(new_crtc_state))
 		intel_color_commit_noarm(new_crtc_state);
@@ -7458,8 +7463,28 @@ static void intel_atomic_commit_tail(struct intel_atomic_state *state)
 	drm_atomic_helper_wait_for_dependencies(&state->base);
 	drm_dp_mst_atomic_wait_for_dependencies(&state->base);
 
-	if (state->modeset)
-		wakeref = intel_display_power_get(dev_priv, POWER_DOMAIN_MODESET);
+	/*
+	 * During full modesets we write a lot of registers, wait
+	 * for PLLs, etc. Doing that while DC states are enabled
+	 * is not a good idea.
+	 *
+	 * During fastsets and other updates we also need to
+	 * disable DC states due to the following scenario:
+	 * 1. DC5 exit and PSR exit happen
+	 * 2. Some or all _noarm() registers are written
+	 * 3. Due to some long delay PSR is re-entered
+	 * 4. DC5 entry -> DMC saves the already written new
+	 *    _noarm() registers and the old not yet written
+	 *    _arm() registers
+	 * 5. DC5 exit -> DMC restores a mixture of old and
+	 *    new register values and arms the update
+	 * 6. PSR exit -> hardware latches a mixture of old and
+	 *    new register values -> corrupted frame, or worse
+	 * 7. New _arm() registers are finally written
+	 * 8. Hardware finally latches a complete set of new
+	 *    register values, and subsequent frames will be OK again
+	 */
+	wakeref = intel_display_power_get(dev_priv, POWER_DOMAIN_DC_OFF);
 
 	intel_atomic_prepare_plane_clear_colors(state);
 
@@ -7608,8 +7633,8 @@ static void intel_atomic_commit_tail(struct intel_atomic_state *state)
 		 * the culprit.
 		 */
 		intel_uncore_arm_unclaimed_mmio_detection(&dev_priv->uncore);
-		intel_display_power_put(dev_priv, POWER_DOMAIN_MODESET, wakeref);
 	}
+	intel_display_power_put(dev_priv, POWER_DOMAIN_DC_OFF, wakeref);
 	intel_runtime_pm_put(&dev_priv->runtime_pm, state->wakeref);
 
 	/*

drivers/gpu/drm/i915/display/intel_dpt.c

@@ -301,6 +301,7 @@ intel_dpt_create(struct intel_framebuffer *fb)
 	vm->pte_encode = gen8_ggtt_pte_encode;
 
 	dpt->obj = dpt_obj;
+	dpt->obj->is_dpt = true;
 
 	return &dpt->vm;
 }
@@ -309,5 +310,6 @@ void intel_dpt_destroy(struct i915_address_space *vm)
 {
 	struct i915_dpt *dpt = i915_vm_to_dpt(vm);
 
+	dpt->obj->is_dpt = false;
 	i915_vm_put(&dpt->vm);
 }

drivers/gpu/drm/i915/display/intel_tc.c

@@ -418,9 +418,9 @@ static bool icl_tc_phy_is_owned(struct intel_digital_port *dig_port)
 	val = intel_de_read(i915, PORT_TX_DFLEXDPCSSS(dig_port->tc_phy_fia));
 	if (val == 0xffffffff) {
 		drm_dbg_kms(&i915->drm,
-			    "Port %s: PHY in TCCOLD, assume safe mode\n",
+			    "Port %s: PHY in TCCOLD, assume not owned\n",
 			    dig_port->tc_port_name);
-		return true;
+		return false;
 	}
 
 	return val & DP_PHY_MODE_STATUS_NOT_SAFE(dig_port->tc_phy_fia_idx);

drivers/gpu/drm/i915/gem/i915_gem_lmem.c

@@ -127,7 +127,8 @@ i915_gem_object_create_lmem_from_data(struct drm_i915_private *i915,
 
 	memcpy(map, data, size);
 
-	i915_gem_object_unpin_map(obj);
+	i915_gem_object_flush_map(obj);
+	__i915_gem_object_release_map(obj);
 
 	return obj;
 }

drivers/gpu/drm/i915/gem/i915_gem_object.h

@@ -303,7 +303,7 @@ i915_gem_object_never_mmap(const struct drm_i915_gem_object *obj)
 static inline bool
 i915_gem_object_is_framebuffer(const struct drm_i915_gem_object *obj)
 {
-	return READ_ONCE(obj->frontbuffer);
+	return READ_ONCE(obj->frontbuffer) || obj->is_dpt;
 }
 
 static inline unsigned int

drivers/gpu/drm/i915/gem/i915_gem_object_types.h

@@ -491,6 +491,9 @@ struct drm_i915_gem_object {
 	 */
 	unsigned int cache_dirty:1;
 
+	/* @is_dpt: Object houses a display page table (DPT) */
+	unsigned int is_dpt:1;
+
 	/**
 	 * @read_domains: Read memory domains.
 	 *

drivers/gpu/drm/i915/gt/intel_rps.c

@@ -2075,16 +2075,6 @@ void intel_rps_sanitize(struct intel_rps *rps)
 		rps_disable_interrupts(rps);
 }
 
-u32 intel_rps_read_rpstat_fw(struct intel_rps *rps)
-{
-	struct drm_i915_private *i915 = rps_to_i915(rps);
-	i915_reg_t rpstat;
-
-	rpstat = (GRAPHICS_VER(i915) >= 12) ? GEN12_RPSTAT1 : GEN6_RPSTAT1;
-
-	return intel_uncore_read_fw(rps_to_gt(rps)->uncore, rpstat);
-}
-
 u32 intel_rps_read_rpstat(struct intel_rps *rps)
 {
 	struct drm_i915_private *i915 = rps_to_i915(rps);
@@ -2095,7 +2085,7 @@ u32 intel_rps_read_rpstat(struct intel_rps *rps)
 	return intel_uncore_read(rps_to_gt(rps)->uncore, rpstat);
 }
 
-u32 intel_rps_get_cagf(struct intel_rps *rps, u32 rpstat)
+static u32 intel_rps_get_cagf(struct intel_rps *rps, u32 rpstat)
 {
 	struct drm_i915_private *i915 = rps_to_i915(rps);
 	u32 cagf;
@@ -2118,33 +2108,42 @@ u32 intel_rps_get_cagf(struct intel_rps *rps, u32 rpstat)
 	return cagf;
 }
 
-static u32 read_cagf(struct intel_rps *rps)
+static u32 __read_cagf(struct intel_rps *rps, bool take_fw)
 {
 	struct drm_i915_private *i915 = rps_to_i915(rps);
 	struct intel_uncore *uncore = rps_to_uncore(rps);
+	i915_reg_t r = INVALID_MMIO_REG;
 	u32 freq;
 
 	/*
 	 * For Gen12+ reading freq from HW does not need a forcewake and
 	 * registers will return 0 freq when GT is in RC6
 	 */
 	if (GRAPHICS_VER_FULL(i915) >= IP_VER(12, 70)) {
-		freq = intel_uncore_read(uncore, MTL_MIRROR_TARGET_WP1);
+		r = MTL_MIRROR_TARGET_WP1;
 	} else if (GRAPHICS_VER(i915) >= 12) {
-		freq = intel_uncore_read(uncore, GEN12_RPSTAT1);
+		r = GEN12_RPSTAT1;
 	} else if (IS_VALLEYVIEW(i915) || IS_CHERRYVIEW(i915)) {
 		vlv_punit_get(i915);
 		freq = vlv_punit_read(i915, PUNIT_REG_GPU_FREQ_STS);
 		vlv_punit_put(i915);
 	} else if (GRAPHICS_VER(i915) >= 6) {
-		freq = intel_uncore_read(uncore, GEN6_RPSTAT1);
+		r = GEN6_RPSTAT1;
 	} else {
-		freq = intel_uncore_read(uncore, MEMSTAT_ILK);
+		r = MEMSTAT_ILK;
 	}
 
+	if (i915_mmio_reg_valid(r))
+		freq = take_fw ? intel_uncore_read(uncore, r) : intel_uncore_read_fw(uncore, r);
+
 	return intel_rps_get_cagf(rps, freq);
 }
 
+static u32 read_cagf(struct intel_rps *rps)
+{
+	return __read_cagf(rps, true);
+}
+
 u32 intel_rps_read_actual_frequency(struct intel_rps *rps)
 {
 	struct intel_runtime_pm *rpm = rps_to_uncore(rps)->rpm;
@@ -2157,7 +2156,12 @@ u32 intel_rps_read_actual_frequency(struct intel_rps *rps)
 	return freq;
 }
 
-u32 intel_rps_read_punit_req(struct intel_rps *rps)
+u32 intel_rps_read_actual_frequency_fw(struct intel_rps *rps)
+{
+	return intel_gpu_freq(rps, __read_cagf(rps, false));
+}
+
+static u32 intel_rps_read_punit_req(struct intel_rps *rps)
 {
 	struct intel_uncore *uncore = rps_to_uncore(rps);
 	struct intel_runtime_pm *rpm = rps_to_uncore(rps)->rpm;