drm/msm: gpu: Add support for the GPMU

Most 5XX targets have GPMU (Graphics Power Management Unit) that
handles a lot of the heavy lifting for power management including
thermal and limits management and dynamic power collapse. While
the GPMU itself is optional, it is usually nessesary to hit
aggressive power targets.

The GPMU firmware needs to be loaded into the GPMU at init time via a
shared hardware block of registers. Using the GPU to write the microcode
is more efficient than using the CPU so at first load create an indirect
buffer that can be executed during subsequent initalization sequences.

After loading the GPMU gets initalized through a shared register
interface and then we mostly get out of its way and let it do
its thing.
Signed-off-by: NJordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: NRob Clark <robdclark@gmail.com>

drivers/gpu/drm/msm/Makefile

@@ -7,6 +7,7 @@ msm-y := \
 	adreno/a3xx_gpu.o \
 	adreno/a4xx_gpu.o \
 	adreno/a5xx_gpu.o \
+	adreno/a5xx_power.o \
 	hdmi/hdmi.o \
 	hdmi/hdmi_audio.o \
 	hdmi/hdmi_bridge.o \

drivers/gpu/drm/msm/adreno/a5xx_gpu.c

@@ -450,6 +450,9 @@ static int a5xx_hw_init(struct msm_gpu *gpu)
 		REG_A5XX_RBBM_SECVID_TSB_TRUSTED_BASE_HI, 0x00000000);
 	gpu_write(gpu, REG_A5XX_RBBM_SECVID_TSB_TRUSTED_SIZE, 0x00000000);
 
+	/* Load the GPMU firmware before starting the HW init */
+	a5xx_gpmu_ucode_init(gpu);
+
 	ret = adreno_hw_init(gpu);
 	if (ret)
 		return ret;
@@ -467,8 +470,9 @@ static int a5xx_hw_init(struct msm_gpu *gpu)
 	if (ret)
 		return ret;
 
-	/* Put the GPU into insecure mode */
-	gpu_write(gpu, REG_A5XX_RBBM_SECVID_TRUST_CNTL, 0x0);
+	ret = a5xx_power_init(gpu);
+	if (ret)
+		return ret;
 
 	/*
 	 * Send a pipeline event stat to get misbehaving counters to start
@@ -483,6 +487,9 @@ static int a5xx_hw_init(struct msm_gpu *gpu)
 			return -EINVAL;
 	}
 
+	/* Put the GPU into unsecure mode */
+	gpu_write(gpu, REG_A5XX_RBBM_SECVID_TRUST_CNTL, 0x0);
+
 	return 0;
 }
 
@@ -525,6 +532,12 @@ static void a5xx_destroy(struct msm_gpu *gpu)
 		drm_gem_object_unreference_unlocked(a5xx_gpu->pfp_bo);
 	}
 
+	if (a5xx_gpu->gpmu_bo) {
+		if (a5xx_gpu->gpmu_bo)
+			msm_gem_put_iova(a5xx_gpu->gpmu_bo, gpu->id);
+		drm_gem_object_unreference_unlocked(a5xx_gpu->gpmu_bo);
+	}
+
 	adreno_gpu_cleanup(adreno_gpu);
 	kfree(a5xx_gpu);
 }
@@ -748,11 +761,54 @@ static void a5xx_dump(struct msm_gpu *gpu)
 
 static int a5xx_pm_resume(struct msm_gpu *gpu)
 {
-	return msm_gpu_pm_resume(gpu);
+	int ret;
+
+	/* Turn on the core power */
+	ret = msm_gpu_pm_resume(gpu);
+	if (ret)
+		return ret;
+
+	/* Turn the RBCCU domain first to limit the chances of voltage droop */
+	gpu_write(gpu, REG_A5XX_GPMU_RBCCU_POWER_CNTL, 0x778000);
+
+	/* Wait 3 usecs before polling */
+	udelay(3);
+
+	ret = spin_usecs(gpu, 20, REG_A5XX_GPMU_RBCCU_PWR_CLK_STATUS,
+		(1 << 20), (1 << 20));
+	if (ret) {
+		DRM_ERROR("%s: timeout waiting for RBCCU GDSC enable: %X\n",
+			gpu->name,
+			gpu_read(gpu, REG_A5XX_GPMU_RBCCU_PWR_CLK_STATUS));
+		return ret;
+	}
+
+	/* Turn on the SP domain */
+	gpu_write(gpu, REG_A5XX_GPMU_SP_POWER_CNTL, 0x778000);
+	ret = spin_usecs(gpu, 20, REG_A5XX_GPMU_SP_PWR_CLK_STATUS,
+		(1 << 20), (1 << 20));
+	if (ret)
+		DRM_ERROR("%s: timeout waiting for SP GDSC enable\n",
+			gpu->name);
+
+	return ret;
 }
 
 static int a5xx_pm_suspend(struct msm_gpu *gpu)
 {
+	/* Clear the VBIF pipe before shutting down */
+	gpu_write(gpu, REG_A5XX_VBIF_XIN_HALT_CTRL0, 0xF);
+	spin_until((gpu_read(gpu, REG_A5XX_VBIF_XIN_HALT_CTRL1) & 0xF) == 0xF);
+
+	gpu_write(gpu, REG_A5XX_VBIF_XIN_HALT_CTRL0, 0);
+
+	/*
+	 * Reset the VBIF before power collapse to avoid issue with FIFO
+	 * entries
+	 */
+	gpu_write(gpu, REG_A5XX_RBBM_BLOCK_SW_RESET_CMD, 0x003C0000);
+	gpu_write(gpu, REG_A5XX_RBBM_BLOCK_SW_RESET_CMD, 0x00000000);
+
 	return msm_gpu_pm_suspend(gpu);
 }
 
@@ -820,6 +876,8 @@ struct msm_gpu *a5xx_gpu_init(struct drm_device *dev)
 	adreno_gpu->registers = a5xx_registers;
 	adreno_gpu->reg_offsets = a5xx_register_offsets;
 
+	a5xx_gpu->lm_leakage = 0x4E001A;
+
 	ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs);
 	if (ret) {
 		a5xx_destroy(&(a5xx_gpu->base.base));

drivers/gpu/drm/msm/adreno/a5xx_gpu.h

@@ -30,8 +30,31 @@ struct a5xx_gpu {
 
 	struct drm_gem_object *pfp_bo;
 	uint64_t pfp_iova;
+
+	struct drm_gem_object *gpmu_bo;
+	uint64_t gpmu_iova;
+	uint32_t gpmu_dwords;
+
+	uint32_t lm_leakage;
 };
 
 #define to_a5xx_gpu(x) container_of(x, struct a5xx_gpu, base)
 
+int a5xx_power_init(struct msm_gpu *gpu);
+void a5xx_gpmu_ucode_init(struct msm_gpu *gpu);
+
+static inline int spin_usecs(struct msm_gpu *gpu, uint32_t usecs,
+		uint32_t reg, uint32_t mask, uint32_t value)
+{
+	while (usecs--) {
+		udelay(1);
+		if ((gpu_read(gpu, reg) & mask) == value)
+			return 0;
+		cpu_relax();
+	}
+
+	return -ETIMEDOUT;
+}
+
+
 #endif /* __A5XX_GPU_H__ */

drivers/gpu/drm/msm/adreno/a5xx_power.c

+/* Copyright (c) 2016 The Linux Foundation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 and
+ * only version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ */
+
+#include <linux/pm_opp.h>
+#include "a5xx_gpu.h"
+
+/*
+ * The GPMU data block is a block of shared registers that can be used to
+ * communicate back and forth. These "registers" are by convention with the GPMU
+ * firwmare and not bound to any specific hardware design
+ */
+
+#define AGC_INIT_BASE REG_A5XX_GPMU_DATA_RAM_BASE
+#define AGC_INIT_MSG_MAGIC (AGC_INIT_BASE + 5)
+#define AGC_MSG_BASE (AGC_INIT_BASE + 7)
+
+#define AGC_MSG_STATE (AGC_MSG_BASE + 0)
+#define AGC_MSG_COMMAND (AGC_MSG_BASE + 1)
+#define AGC_MSG_PAYLOAD_SIZE (AGC_MSG_BASE + 3)
+#define AGC_MSG_PAYLOAD(_o) ((AGC_MSG_BASE + 5) + (_o))
+
+#define AGC_POWER_CONFIG_PRODUCTION_ID 1
+#define AGC_INIT_MSG_VALUE 0xBABEFACE
+
+static struct {
+	uint32_t reg;
+	uint32_t value;
+} a5xx_sequence_regs[] = {
+	{ 0xB9A1, 0x00010303 },
+	{ 0xB9A2, 0x13000000 },
+	{ 0xB9A3, 0x00460020 },
+	{ 0xB9A4, 0x10000000 },
+	{ 0xB9A5, 0x040A1707 },
+	{ 0xB9A6, 0x00010000 },
+	{ 0xB9A7, 0x0E000904 },
+	{ 0xB9A8, 0x10000000 },
+	{ 0xB9A9, 0x01165000 },
+	{ 0xB9AA, 0x000E0002 },
+	{ 0xB9AB, 0x03884141 },
+	{ 0xB9AC, 0x10000840 },
+	{ 0xB9AD, 0x572A5000 },
+	{ 0xB9AE, 0x00000003 },
+	{ 0xB9AF, 0x00000000 },
+	{ 0xB9B0, 0x10000000 },
+	{ 0xB828, 0x6C204010 },
+	{ 0xB829, 0x6C204011 },
+	{ 0xB82A, 0x6C204012 },
+	{ 0xB82B, 0x6C204013 },
+	{ 0xB82C, 0x6C204014 },
+	{ 0xB90F, 0x00000004 },
+	{ 0xB910, 0x00000002 },
+	{ 0xB911, 0x00000002 },
+	{ 0xB912, 0x00000002 },
+	{ 0xB913, 0x00000002 },
+	{ 0xB92F, 0x00000004 },
+	{ 0xB930, 0x00000005 },
+	{ 0xB931, 0x00000005 },
+	{ 0xB932, 0x00000005 },
+	{ 0xB933, 0x00000005 },
+	{ 0xB96F, 0x00000001 },
+	{ 0xB970, 0x00000003 },
+	{ 0xB94F, 0x00000004 },
+	{ 0xB950, 0x0000000B },
+	{ 0xB951, 0x0000000B },
+	{ 0xB952, 0x0000000B },
+	{ 0xB953, 0x0000000B },
+	{ 0xB907, 0x00000019 },
+	{ 0xB927, 0x00000019 },
+	{ 0xB947, 0x00000019 },
+	{ 0xB967, 0x00000019 },
+	{ 0xB987, 0x00000019 },
+	{ 0xB906, 0x00220001 },
+	{ 0xB926, 0x00220001 },
+	{ 0xB946, 0x00220001 },
+	{ 0xB966, 0x00220001 },
+	{ 0xB986, 0x00300000 },
+	{ 0xAC40, 0x0340FF41 },
+	{ 0xAC41, 0x03BEFED0 },
+	{ 0xAC42, 0x00331FED },
+	{ 0xAC43, 0x021FFDD3 },
+	{ 0xAC44, 0x5555AAAA },
+	{ 0xAC45, 0x5555AAAA },
+	{ 0xB9BA, 0x00000008 },
+};
+
+/*
+ * Get the actual voltage value for the operating point at the specified
+ * frequency
+ */
+static inline uint32_t _get_mvolts(struct msm_gpu *gpu, uint32_t freq)
+{
+	struct drm_device *dev = gpu->dev;
+	struct msm_drm_private *priv = dev->dev_private;
+	struct platform_device *pdev = priv->gpu_pdev;
+	struct dev_pm_opp *opp;
+
+	opp = dev_pm_opp_find_freq_exact(&pdev->dev, freq, true);
+
+	return (!IS_ERR(opp)) ? dev_pm_opp_get_voltage(opp) / 1000 : 0;
+}
+
+/* Setup thermal limit management */
+static void a5xx_lm_setup(struct msm_gpu *gpu)
+{
+	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
+	struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu);
+	unsigned int i;
+
+	/* Write the block of sequence registers */
+	for (i = 0; i < ARRAY_SIZE(a5xx_sequence_regs); i++)
+		gpu_write(gpu, a5xx_sequence_regs[i].reg,
+			a5xx_sequence_regs[i].value);
+
+	/* Hard code the A530 GPU thermal sensor ID for the GPMU */
+	gpu_write(gpu, REG_A5XX_GPMU_TEMP_SENSOR_ID, 0x60007);
+	gpu_write(gpu, REG_A5XX_GPMU_DELTA_TEMP_THRESHOLD, 0x01);
+	gpu_write(gpu, REG_A5XX_GPMU_TEMP_SENSOR_CONFIG, 0x01);
+
+	/* Until we get clock scaling 0 is always the active power level */
+	gpu_write(gpu, REG_A5XX_GPMU_GPMU_VOLTAGE, 0x80000000 | 0);
+
+	gpu_write(gpu, REG_A5XX_GPMU_BASE_LEAKAGE, a5xx_gpu->lm_leakage);
+
+	/* The threshold is fixed at 6000 for A530 */
+	gpu_write(gpu, REG_A5XX_GPMU_GPMU_PWR_THRESHOLD, 0x80000000 | 6000);
+
+	gpu_write(gpu, REG_A5XX_GPMU_BEC_ENABLE, 0x10001FFF);
+	gpu_write(gpu, REG_A5XX_GDPM_CONFIG1, 0x00201FF1);
+
+	/* Write the voltage table */
+	gpu_write(gpu, REG_A5XX_GPMU_BEC_ENABLE, 0x10001FFF);
+	gpu_write(gpu, REG_A5XX_GDPM_CONFIG1, 0x201FF1);
+
+	gpu_write(gpu, AGC_MSG_STATE, 1);
+	gpu_write(gpu, AGC_MSG_COMMAND, AGC_POWER_CONFIG_PRODUCTION_ID);
+
+	/* Write the max power - hard coded to 5448 for A530 */
+	gpu_write(gpu, AGC_MSG_PAYLOAD(0), 5448);
+	gpu_write(gpu, AGC_MSG_PAYLOAD(1), 1);
+
+	/*
+	 * For now just write the one voltage level - we will do more when we
+	 * can do scaling
+	 */
+	gpu_write(gpu, AGC_MSG_PAYLOAD(2), _get_mvolts(gpu, gpu->fast_rate));
+	gpu_write(gpu, AGC_MSG_PAYLOAD(3), gpu->fast_rate / 1000000);
+
+	gpu_write(gpu, AGC_MSG_PAYLOAD_SIZE, 4 * sizeof(uint32_t));
+	gpu_write(gpu, AGC_INIT_MSG_MAGIC, AGC_INIT_MSG_VALUE);
+}
+
+/* Enable SP/TP cpower collapse */
+static void a5xx_pc_init(struct msm_gpu *gpu)
+{
+	gpu_write(gpu, REG_A5XX_GPMU_PWR_COL_INTER_FRAME_CTRL, 0x7F);
+	gpu_write(gpu, REG_A5XX_GPMU_PWR_COL_BINNING_CTRL, 0);
+	gpu_write(gpu, REG_A5XX_GPMU_PWR_COL_INTER_FRAME_HYST, 0xA0080);
+	gpu_write(gpu, REG_A5XX_GPMU_PWR_COL_STAGGER_DELAY, 0x600040);
+}
+
+/* Enable the GPMU microcontroller */
+static int a5xx_gpmu_init(struct msm_gpu *gpu)
+{
+	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
+	struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu);
+	struct msm_ringbuffer *ring = gpu->rb;
+
+	if (!a5xx_gpu->gpmu_dwords)
+		return 0;
+
+	/* Turn off protected mode for this operation */
+	OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1);
+	OUT_RING(ring, 0);
+
+	/* Kick off the IB to load the GPMU microcode */
+	OUT_PKT7(ring, CP_INDIRECT_BUFFER_PFE, 3);
+	OUT_RING(ring, lower_32_bits(a5xx_gpu->gpmu_iova));
+	OUT_RING(ring, upper_32_bits(a5xx_gpu->gpmu_iova));
+	OUT_RING(ring, a5xx_gpu->gpmu_dwords);
+
+	/* Turn back on protected mode */
+	OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1);
+	OUT_RING(ring, 1);
+
+	gpu->funcs->flush(gpu);
+
+	if (!gpu->funcs->idle(gpu)) {
+		DRM_ERROR("%s: Unable to load GPMU firmware. GPMU will not be active\n",
+			gpu->name);
+		return -EINVAL;
+	}
+
+	gpu_write(gpu, REG_A5XX_GPMU_WFI_CONFIG, 0x4014);
+
+	/* Kick off the GPMU */
+	gpu_write(gpu, REG_A5XX_GPMU_CM3_SYSRESET, 0x0);
+
+	/*
+	 * Wait for the GPMU to respond. It isn't fatal if it doesn't, we just
+	 * won't have advanced power collapse.
+	 */
+	if (spin_usecs(gpu, 25, REG_A5XX_GPMU_GENERAL_0, 0xFFFFFFFF,
+		0xBABEFACE))
+		DRM_ERROR("%s: GPMU firmware initialization timed out\n",
+			gpu->name);
+
+	return 0;
+}
+
+/* Enable limits management */
+static void a5xx_lm_enable(struct msm_gpu *gpu)
+{
+	gpu_write(gpu, REG_A5XX_GDPM_INT_MASK, 0x0);
+	gpu_write(gpu, REG_A5XX_GDPM_INT_EN, 0x0A);
+	gpu_write(gpu, REG_A5XX_GPMU_GPMU_VOLTAGE_INTR_EN_MASK, 0x01);
+	gpu_write(gpu, REG_A5XX_GPMU_TEMP_THRESHOLD_INTR_EN_MASK, 0x50000);
+	gpu_write(gpu, REG_A5XX_GPMU_THROTTLE_UNMASK_FORCE_CTRL, 0x30000);
+
+	gpu_write(gpu, REG_A5XX_GPMU_CLOCK_THROTTLE_CTRL, 0x011);
+}
+
+int a5xx_power_init(struct msm_gpu *gpu)
+{
+	int ret;
+
+	/* Set up the limits management */
+	a5xx_lm_setup(gpu);
+
+	/* Set up SP/TP power collpase */
+	a5xx_pc_init(gpu);
+
+	/* Start the GPMU */
+	ret = a5xx_gpmu_init(gpu);
+	if (ret)
+		return ret;
+
+	/* Start the limits management */
+	a5xx_lm_enable(gpu);
+
+	return 0;
+}
+
+void a5xx_gpmu_ucode_init(struct msm_gpu *gpu)
+{
+	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
+	struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu);
+	struct drm_device *drm = gpu->dev;
+	const struct firmware *fw;
+	uint32_t dwords = 0, offset = 0, bosize;
+	unsigned int *data, *ptr, *cmds;
+	unsigned int cmds_size;
+
+	if (a5xx_gpu->gpmu_bo)
+		return;
+
+	/* Get the firmware */
+	if (request_firmware(&fw, adreno_gpu->info->gpmufw, drm->dev)) {
+		DRM_ERROR("%s: Could not get GPMU firmware. GPMU will not be active\n",
+			gpu->name);
+		return;
+	}
+
+	data = (unsigned int *) fw->data;
+
+	/*
+	 * The first dword is the size of the remaining data in dwords. Use it
+	 * as a checksum of sorts and make sure it matches the actual size of
+	 * the firmware that we read
+	 */
+
+	if (fw->size < 8 || (data[0] < 2) || (data[0] >= (fw->size >> 2)))
+		goto out;
+
+	/* The second dword is an ID - look for 2 (GPMU_FIRMWARE_ID) */
+	if (data[1] != 2)
+		goto out;
+
+	cmds = data + data[2] + 3;
+	cmds_size = data[0] - data[2] - 2;
+
+	/*
+	 * A single type4 opcode can only have so many values attached so
+	 * add enough opcodes to load the all the commands
+	 */
+	bosize = (cmds_size + (cmds_size / TYPE4_MAX_PAYLOAD) + 1) << 2;
+
+	mutex_lock(&drm->struct_mutex);
+	a5xx_gpu->gpmu_bo = msm_gem_new(drm, bosize, MSM_BO_UNCACHED);
+	mutex_unlock(&drm->struct_mutex);
+
+	if (IS_ERR(a5xx_gpu->gpmu_bo))
+		goto err;
+
+	if (msm_gem_get_iova(a5xx_gpu->gpmu_bo, gpu->id, &a5xx_gpu->gpmu_iova))
+		goto err;
+
+	ptr = msm_gem_get_vaddr(a5xx_gpu->gpmu_bo);
+	if (!ptr)
+		goto err;
+
+	while (cmds_size > 0) {
+		int i;
+		uint32_t _size = cmds_size > TYPE4_MAX_PAYLOAD ?
+			TYPE4_MAX_PAYLOAD : cmds_size;
+
+		ptr[dwords++] = PKT4(REG_A5XX_GPMU_INST_RAM_BASE + offset,
+			_size);
+
+		for (i = 0; i < _size; i++)
+			ptr[dwords++] = *cmds++;
+
+		offset += _size;
+		cmds_size -= _size;
+	}
+
+	msm_gem_put_vaddr(a5xx_gpu->gpmu_bo);
+	a5xx_gpu->gpmu_dwords = dwords;
+
+	goto out;
+
+err:
+	if (a5xx_gpu->gpmu_iova)
+		msm_gem_put_iova(a5xx_gpu->gpmu_bo, gpu->id);
+	if (a5xx_gpu->gpmu_bo)
+		drm_gem_object_unreference_unlocked(a5xx_gpu->gpmu_bo);
+
+	a5xx_gpu->gpmu_bo = NULL;
+	a5xx_gpu->gpmu_iova = 0;
+	a5xx_gpu->gpmu_dwords = 0;
+
+out:
+	/* No need to keep that firmware laying around anymore */
+	release_firmware(fw);
+}

drivers/gpu/drm/msm/adreno/adreno_device.c

@@ -82,6 +82,7 @@ static const struct adreno_info gpulist[] = {
 		.pfpfw = "a530_pfp.fw",
 		.gmem = SZ_1M,
 		.init = a5xx_gpu_init,
+		.gpmufw = "a530v3_gpmu.fw2",
 	},
 };
 

drivers/gpu/drm/msm/adreno/adreno_gpu.h

@@ -73,6 +73,7 @@ struct adreno_info {
 	uint32_t revn;
 	const char *name;
 	const char *pm4fw, *pfpfw;
+	const char *gpmufw;
 	uint32_t gmem;
 	struct msm_gpu *(*init)(struct drm_device *dev);
 };