drm/apple: m4: Power the display down for system sleep, while blanked and while static

drivers/gpu/drm/apple/dcp-m4-callback.h

@@ -5,6 +5,7 @@
 #include <linux/errno.h>
 #include <linux/kernel.h>
 #include <linux/string.h>
+#include <linux/timekeeping.h>
 #include <linux/types.h>
 #include <linux/unaligned.h>
 #include "dcp-m4-profile.h"
@@ -18,6 +19,8 @@ struct dcp_m4_callback_ops {
 	int (*event)(void *cookie, const char *name, const void *dictionary, size_t size);
 	int (*property)(void *cookie, unsigned int method, const void *input, void *output);
 	int (*memory)(void *cookie, unsigned int method, const void *input, void *output);
+	/* Idle caching: D117 state, D578/D579 fence, D580 cache-surface release. */
+	int (*idle)(void *cookie, unsigned int method, const void *input, void *output);
 };
 
 struct dcp_m4_callback_shape {
@@ -33,16 +36,18 @@ static inline int dcp_m4_callback_result(int ret)
 
 /* Exact qualified J713 shapes. An absent behavior hook is failure, never an ACK. */
 static const struct dcp_m4_callback_shape dcp_m4_callbacks[] = {
-	{0, 0, 4}, {1, 0, 4}, {3, 4, 60}, {6, 84, 80},
-	{100, 0, 0}, {101, 0, 4}, {102, 68, 0}, {104, 68, 0}, {107, 64, 0},
+	{0, 0, 4}, {1, 0, 4}, {2, 0, 0}, {3, 4, 60}, {6, 84, 80}, {8, 4, 0},
+	{100, 0, 0}, {101, 0, 4}, {102, 68, 0}, {104, 68, 0}, {106, 0, 4}, {107, 64, 0},
 	{108, 0, 4}, {109, 0, 4}, {110, 0, 4}, {111, 0, 4}, {112, 0, 4}, {113, 0, 4},
-	{114, 4164, 4100}, {121, 0, 4}, {122, 0, 4}, {123, 0, 4},
+	{114, 4164, 4100}, {117, 12, 0}, {121, 0, 4}, {122, 0, 4}, {123, 0, 4},
 	{125, 100, 36}, {127, 4, 4}, {128, 4104, 4}, {129, 64, 4},
-	{201, 12, 16}, {206, 0, 4}, {207, 0, 4}, {300, 16, 0},
+	{201, 12, 16}, {203, 0, 4}, {206, 0, 4}, {207, 0, 4}, {209, 0, 8}, {300, 16, 0},
 	{400, 76, 3076}, {401, 80, 12}, {406, 72, 0}, {408, 8, 8},
 	{413, 4168, 4}, {414, 80, 4}, {415, 76, 4}, {451, 20, 28}, {454, 4, 4},
 	{552, 4164, 4}, {561, 4164, 4}, {563, 76, 4}, {565, 72, 4}, {567, 128, 4},
-	{575, 88, 76}, {576, 4, 0}, {582, 8, 4}, {588, 0, 0}, {590, 1840, 0}, {599, 0, 0},
+	{572, 4, 4}, {575, 88, 76}, {576, 4, 0}, {578, 4, 4}, {579, 0, 0}, {580, 0, 0},
+	{582, 8, 4}, {588, 0, 0}, {590, 1840, 0},
+	{592, 20, 0}, {594, 4, 0}, {599, 0, 0}, {700, 4, 4},
 };
 
 static inline int dcp_m4_callback_method(u32 tag)
@@ -136,6 +141,29 @@ static inline int dcp_m4_callback_dispatch(const struct dcp_m4_callback_ops *ops
 		return 0;
 	case 300: case 406: case 576: case 588:
 		return 0;
+	/*
+	 * will_power_off_signal, set_power_trace_enable, swap_complete_intent,
+	 * enable_backlight_message_ap_gated: notifications.
+	 */
+	case 2: case 8: case 592: case 594:
+		return 0;
+	/*
+	 * set_idle_caching_state_ap, idle_fence_create, idle_fence_complete,
+	 * idle_surface_release.
+	 */
+	case 117: case 578: case 579: case 580:
+		if (!ops->idle)
+			return -EOPNOTSUPP;
+		return dcp_m4_callback_result(ops->idle(cookie, method, input, output));
+	/*
+	 * acknowledgeSetPowerState, powerUpDART, set_kernel_power_assert: success;
+	 * aot_supported_peek: false.
+	 */
+	case 106: case 572: case 700: case 203:
+		return 0;
+	case 209: /* get_calendar_time_ms */
+		put_unaligned_le64(ktime_get_real_ns() / NSEC_PER_MSEC, out);
+		return 0;
 	case 122: case 123: case 582:
 		return 0; /* False, not an unimplemented service returning success. */
 	case 101:

drivers/gpu/drm/apple/dcp-m4-frame.h

@@ -253,6 +253,22 @@ static inline int dcp_m4_swap_encode_full(void *output, size_t length, u32 swap_
 	return 0;
 }
 
+/* Swap intent that registers surface 0 as the idle-cache surface. */
+#define DCP_M4_SWAP_INTENT_IDLE_CACHE	8
+
+/*
+ * Retarget an encoded present: the DCP registers rather than presents its
+ * sole surface, which must be a distinct object (surface_id other than
+ * the primary's 1) from every surface on screen.
+ */
+static inline void dcp_m4_swap_set_intent(void *output, u32 intent, u32 surface_id)
+{
+	u8 *data = output;
+
+	put_unaligned_le32(intent, data + dcp_m4_pf->swap_intent);
+	put_unaligned_le32(surface_id, data + 0x9c);
+}
+
 static inline int dcp_m4_swap_encode_checked(void *output, size_t length, u32 swap_id,
 					     const struct dcp_m4_frame *frame,
 					     bool brightness_change, u32 nits,

drivers/gpu/drm/apple/dcp-m4-profile.h

@@ -15,6 +15,7 @@ struct dcp_m4_profile {
 	u32 swap_nullable;		/* ten nullable surface bytes */
 	u32 swap_tail;			/* final two bytes */
 	u32 swap_nits;			/* binary64 brightness */
+	u32 swap_intent;		/* swap intent, 0 for a present */
 	u32 disabled_color;		/* background with no surface */
 	u16 done_method;		/* swap completion callback */
 	u32 done_size;
@@ -28,6 +29,7 @@ static const struct dcp_m4_profile dcp_m4_profile_j713 = {
 	.panel_width = 2560, .panel_height = 1664, .boot_stride = 10240,
 	.swap_size = 0x1bd8, .swap_surface = 0x588, .swap_dva = 0xe38,
 	.swap_nullable = 0x1bcb, .swap_tail = 0x1bd6, .swap_nits = 0x35e,
+	.swap_intent = 0x17c,
 	.disabled_color = 0xff000000,
 	.done_method = 590, .done_size = 0x730, .open_method = 457,
 	.brightness_max = 500, .brightness_default = 120,

drivers/gpu/drm/apple/dcp-m4-protocol.h

@@ -119,16 +119,36 @@ static inline int dcp_m4_decode_word_ext(u64 word, struct dcp_m4_packet *packet)
 	return 0;
 }
 
+/*
+ * While an AP call is outstanding on stream 1 (bit 10, the older contexts'
+ * out-of-band +4), its reply and callbacks come on stream 1 as well, besides
+ * any stream-0 traffic.
+ */
+static inline int dcp_m4_decode_word_oob(u64 word, struct dcp_m4_packet *packet)
+{
+	u8 stream = (word >> 10) & 3;
+	int ret;
+
+	if (!stream)
+		return dcp_m4_decode_word(word, packet);
+	if (stream != 1)
+		return -EPROTO;
+	ret = dcp_m4_decode_word(word & ~0x400ULL, packet);
+	if (!ret)
+		packet->stream = 1;
+	return ret;
+}
+
 /*
  * Replies require their pending request record, not a new packet location.
- * Only callbacks may name a nonzero stream (external profile); the internal
- * decoder never produces one.
+ * Callbacks may name a nonzero stream (external profile, or stream 1 during
+ * an out-of-band call); AP calls use stream 0, or stream 1 out of band.
  */
 static inline int dcp_m4_packet_offset(const struct dcp_m4_packet *packet, bool callback)
 {
 	unsigned int base;
 
-	if (packet->reply || (callback ? packet->stream >= 4 : packet->stream) ||
+	if (packet->reply || (callback ? packet->stream >= 4 : packet->stream > 1) ||
 	    (packet->callback_flag && (!callback || !packet->remote)) ||
 	    packet->offset % DCP_M4_ALIGNMENT ||
 	    packet->offset >= DCP_M4_STREAM_SIZE || packet->length < DCP_M4_PACKET_HEADER ||

drivers/gpu/drm/apple/dcp-m4-transaction.h

@@ -32,13 +32,16 @@ struct dcp_m4_transaction {
 	unsigned int callbacks;
 	int error;
 	bool external;	/* DCPEXT profile: streams 0..3 and any flagged callback route */
+	bool oob;	/* an AP call is outstanding on stream 1 */
 };
 
 static inline int dcp_m4_transaction_decode(const struct dcp_m4_transaction *state, u64 word,
 					    struct dcp_m4_packet *packet)
 {
-	return state->external ? dcp_m4_decode_word_ext(word, packet) :
-				 dcp_m4_decode_word(word, packet);
+	if (state->external)
+		return dcp_m4_decode_word_ext(word, packet);
+	return state->oob ? dcp_m4_decode_word_oob(word, packet) :
+			    dcp_m4_decode_word(word, packet);
 }
 
 static inline int dcp_m4_transaction_fail(struct dcp_m4_transaction *state, int error)
@@ -59,14 +62,15 @@ static inline bool dcp_m4_tag(u32 tag, u8 direction)
 }
 
 /* Prepare only; the caller publishes with its DMA barrier and mailbox send. */
-static inline int dcp_m4_call_begin(struct dcp_m4_transaction *state, void *rpc,
-				  size_t rpc_size, u32 tag, bool remote,
-				  const void *input, u32 input_size, u32 output_size,
-				  u64 *word)
+/* Stream 1 is out of band: top-level calls only, such as setPowerState. */
+static inline int dcp_m4_call_begin_stream(struct dcp_m4_transaction *state, void *rpc,
+					   size_t rpc_size, u32 tag, u8 stream, bool remote,
+					   const void *input, u32 input_size, u32 output_size,
+					   u64 *word)
 {
 	struct dcp_m4_packet packet = {
 		.tag = tag, .input_size = input_size, .output_size = output_size,
-		.remote = remote,
+		.remote = remote, .stream = stream,
 	};
 	const struct dcp_m4_packet *outer = NULL;
 	u8 *data;
@@ -77,6 +81,8 @@ static inline int dcp_m4_call_begin(struct dcp_m4_transaction *state, void *rpc,
 	if (!rpc || rpc_size != DCP_M4_RPC_SIZE || !word || (input_size && !input) ||
 	    !dcp_m4_tag(tag, 'A') || state->calls >= DCP_M4_CALL_DEPTH)
 		return dcp_m4_transaction_fail(state, -EINVAL);
+	if (stream > 1 || (stream && (state->calls || state->callbacks || state->external)))
+		return dcp_m4_transaction_fail(state, -EINVAL);
 	if (state->calls) {
 		struct dcp_m4_call *pending = &state->call[state->calls - 1];
 
@@ -88,9 +94,16 @@ static inline int dcp_m4_call_begin(struct dcp_m4_transaction *state, void *rpc,
 	if (state->callbacks &&
 	    remote != state->callback[state->callbacks - 1].packet.remote)
 		return dcp_m4_transaction_fail(state, -EPROTO);
-	/* AP calls use stream 0; a callback on another stream cannot recurse. */
-	if (state->callbacks && state->callback[state->callbacks - 1].packet.stream)
-		return dcp_m4_transaction_fail(state, -EOPNOTSUPP);
+	/*
+	 * AP calls use stream 0; a callback on another stream cannot recurse,
+	 * except that calls nested in an out-of-band callback stay out of band.
+	 */
+	if (state->callbacks && state->callback[state->callbacks - 1].packet.stream) {
+		if (!state->oob || state->callback[state->callbacks - 1].packet.stream != 1)
+			return dcp_m4_transaction_fail(state, -EOPNOTSUPP);
+		stream = 1;
+		packet.stream = stream;
+	}
 	ret = dcp_m4_packet_size(input_size, output_size, &packet.length);
 	if (ret)
 		return dcp_m4_transaction_fail(state, ret);
@@ -119,11 +132,22 @@ static inline int dcp_m4_call_begin(struct dcp_m4_transaction *state, void *rpc,
 	state->call[state->calls++] = (struct dcp_m4_call) {
 		.packet = packet, .callbacks = state->callbacks,
 	};
-	*word = 2 | (remote ? 0x100 : 0) | ((u64)packet.offset << 16) |
+	*word = 2 | (remote ? 0x100 : 0) | ((u64)stream << 10) | ((u64)packet.offset << 16) |
 		((u64)packet.length << 32);
+	if (stream)
+		state->oob = true;
 	return 0;
 }
 
+static inline int dcp_m4_call_begin(struct dcp_m4_transaction *state, void *rpc,
+				    size_t rpc_size, u32 tag, bool remote,
+				    const void *input, u32 input_size, u32 output_size,
+				    u64 *word)
+{
+	return dcp_m4_call_begin_stream(state, rpc, rpc_size, tag, 0, remote, input,
+					input_size, output_size, word);
+}
+
 /* Firmware may write output, but must not change the owned packet header. */
 static inline int dcp_m4_owned_packet(const void *rpc, const struct dcp_m4_packet *owned,
 				    bool callback)
@@ -160,6 +184,7 @@ static inline int dcp_m4_call_reply(struct dcp_m4_transaction *state, const void
 		return dcp_m4_transaction_fail(state, ret);
 	call = &state->call[state->calls - 1];
 	if (!reply.reply || reply.remote != call->packet.remote ||
+	    reply.stream != call->packet.stream ||
 	    state->callbacks != call->callbacks || output_size != call->packet.output_size ||
 	    (output_size && !output))
 		return dcp_m4_transaction_fail(state, -EPROTO);
@@ -169,6 +194,9 @@ static inline int dcp_m4_call_reply(struct dcp_m4_transaction *state, const void
 	if (output_size)
 		memcpy(output, (const u8 *)rpc + offset + DCP_M4_PACKET_HEADER +
 		       call->packet.input_size, output_size);
+	/* The out-of-band call itself, not a call nested in its callbacks. */
+	if (call->packet.stream && state->calls == 1)
+		state->oob = false;
 	state->calls--;
 	return 0;
 }

drivers/gpu/drm/apple/dcp.c

@@ -6386,9 +6386,17 @@ static int dcp_platform_suspend(struct device *dev)
 	if (dcp_takeover_was_aborted(dcp))
 		return -EBUSY;
 
-	/* The retained T8132 session has no qualified system-sleep handshake. */
-	if (iomfb_is_m4(dcp))
-		return -EBUSY;
+	/*
+	 * T8132: setPowerState powers the display down; the coprocessor keeps
+	 * running and its power domains stay on, as below.
+	 */
+	if (iomfb_is_m4(dcp)) {
+		int ret = iomfb_m4_suspend(dcp);
+
+		if (!ret)
+			device_set_wakeup_path(dev);
+		return ret;
+	}
 	if (READ_ONCE(dcp->ext_dead))
 		return 0;
 	iomfb_v14_7_backlight_suspend(dcp);
@@ -6469,7 +6477,7 @@ static int dcp_platform_resume(struct device *dev)
 		return 0;
 
 	if (iomfb_is_m4(dcp))
-		return iomfb_m4_health(dcp);
+		return iomfb_m4_resume(dcp);
 	if (READ_ONCE(dcp->ext_dead))
 		return 0;
 
@@ -6492,10 +6500,26 @@ static int dcp_platform_resume(struct device *dev)
 	return 0;
 }
 
+/*
+ * Hibernation: the T8132 power-down keeps the retained session, which a
+ * hibernation image cannot carry across the boot that restores it.
+ */
+static int dcp_platform_freeze(struct device *dev)
+{
+	if (iomfb_is_m4(dev_get_drvdata(dev)))
+		return -EBUSY;
+	return dcp_platform_suspend(dev);
+}
+
 static const struct dev_pm_ops dcp_platform_pm_ops = {
 	.prepare = pm_sleep_ptr(dcp_platform_prepare),
 	.complete = pm_sleep_ptr(dcp_platform_complete),
-	SYSTEM_SLEEP_PM_OPS(dcp_platform_suspend, dcp_platform_resume)
+	.suspend = pm_sleep_ptr(dcp_platform_suspend),
+	.resume = pm_sleep_ptr(dcp_platform_resume),
+	.freeze = pm_sleep_ptr(dcp_platform_freeze),
+	.thaw = pm_sleep_ptr(dcp_platform_resume),
+	.poweroff = pm_sleep_ptr(dcp_platform_freeze),
+	.restore = pm_sleep_ptr(dcp_platform_resume),
 };
 
 static const struct apple_dcp_hw_data apple_dcp_hw_t6020 = {