1 files changed, 255 insertions, 48 deletions

diff --git a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
index 035bd13dc8b..942e09d898a 100644
--- a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
@@ -15,6 +15,10 @@
  * this program.  If not, see <http://www.gnu.org/licenses/>.
  */
 
+#ifdef CONFIG_MSM_OCMEM
+#  include <mach/ocmem.h>
+#endif
+
 #include "a3xx_gpu.h"
 
 #define A3XX_INT0_MASK \
@@ -31,7 +35,11 @@
 	 A3XX_INT0_CP_AHB_ERROR_HALT |     \
 	 A3XX_INT0_UCHE_OOB_ACCESS)
 
-static struct platform_device *a3xx_pdev;
+
+static bool hang_debug = false;
+MODULE_PARM_DESC(hang_debug, "Dump registers when hang is detected (can be slow!)");
+module_param_named(hang_debug, hang_debug, bool, 0600);
+static void a3xx_dump(struct msm_gpu *gpu);
 
 static void a3xx_me_init(struct msm_gpu *gpu)
 {
@@ -63,6 +71,7 @@ static void a3xx_me_init(struct msm_gpu *gpu)
 static int a3xx_hw_init(struct msm_gpu *gpu)
 {
 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
+	struct a3xx_gpu *a3xx_gpu = to_a3xx_gpu(adreno_gpu);
 	uint32_t *ptr, len;
 	int i, ret;
 
@@ -105,6 +114,21 @@ static int a3xx_hw_init(struct msm_gpu *gpu)
 		gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x000000ff);
 		gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT_CONF, 0x000000a4);
 
+	} else if (adreno_is_a330v2(adreno_gpu)) {
+		/*
+		 * Most of the VBIF registers on 8974v2 have the correct
+		 * values at power on, so we won't modify those if we don't
+		 * need to
+		 */
+		/* Enable 1k sort: */
+		gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x0001003f);
+		gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT_CONF, 0x000000a4);
+		/* Enable WR-REQ: */
+		gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x00003f);
+		gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303);
+		/* Set up VBIF_ROUND_ROBIN_QOS_ARB: */
+		gpu_write(gpu, REG_A3XX_VBIF_ROUND_ROBIN_QOS_ARB, 0x0003);
+
 	} else if (adreno_is_a330(adreno_gpu)) {
 		/* Set up 16 deep read/write request queues: */
 		gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF0, 0x18181818);
@@ -121,10 +145,10 @@ static int a3xx_hw_init(struct msm_gpu *gpu)
 		/* Set up VBIF_ROUND_ROBIN_QOS_ARB: */
 		gpu_write(gpu, REG_A3XX_VBIF_ROUND_ROBIN_QOS_ARB, 0x0001);
 		/* Set up AOOO: */
-		gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO_EN, 0x0000ffff);
-		gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO, 0xffffffff);
+		gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO_EN, 0x0000003f);
+		gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO, 0x003f003f);
 		/* Enable 1K sort: */
-		gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x0001ffff);
+		gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x0001003f);
 		gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT_CONF, 0x000000a4);
 		/* Disable VBIF clock gating. This is to enable AXI running
 		 * higher frequency than GPU:
@@ -162,23 +186,32 @@ static int a3xx_hw_init(struct msm_gpu *gpu)
 	gpu_write(gpu, REG_A3XX_UCHE_CACHE_MODE_CONTROL_REG, 0x00000001);
 
 	/* Enable Clock gating: */
-	gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xbfffffff);
-
-	/* Set the OCMEM base address for A330 */
-//TODO:
-//	if (adreno_is_a330(adreno_gpu)) {
-//		gpu_write(gpu, REG_A3XX_RB_GMEM_BASE_ADDR,
-//			(unsigned int)(a3xx_gpu->ocmem_base >> 14));
-//	}
+	if (adreno_is_a320(adreno_gpu))
+		gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xbfffffff);
+	else if (adreno_is_a330v2(adreno_gpu))
+		gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xaaaaaaaa);
+	else if (adreno_is_a330(adreno_gpu))
+		gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xbffcffff);
+
+	if (adreno_is_a330v2(adreno_gpu))
+		gpu_write(gpu, REG_A3XX_RBBM_GPR0_CTL, 0x05515455);
+	else if (adreno_is_a330(adreno_gpu))
+		gpu_write(gpu, REG_A3XX_RBBM_GPR0_CTL, 0x00000000);
+
+	/* Set the OCMEM base address for A330, etc */
+	if (a3xx_gpu->ocmem_hdl) {
+		gpu_write(gpu, REG_A3XX_RB_GMEM_BASE_ADDR,
+			(unsigned int)(a3xx_gpu->ocmem_base >> 14));
+	}
 
 	/* Turn on performance counters: */
 	gpu_write(gpu, REG_A3XX_RBBM_PERFCTR_CTL, 0x01);
 
-	/* Set SP perfcounter 7 to count SP_FS_FULL_ALU_INSTRUCTIONS
-	 * we will use this to augment our hang detection:
-	 */
-	gpu_write(gpu, REG_A3XX_SP_PERFCOUNTER7_SELECT,
-			SP_FS_FULL_ALU_INSTRUCTIONS);
+	/* Enable the perfcntrs that we use.. */
+	for (i = 0; i < gpu->num_perfcntrs; i++) {
+		const struct msm_gpu_perfcntr *perfcntr = &gpu->perfcntrs[i];
+		gpu_write(gpu, perfcntr->select_reg, perfcntr->select_val);
+	}
 
 	gpu_write(gpu, REG_A3XX_RBBM_INT_0_MASK, A3XX_INT0_MASK);
 
@@ -219,7 +252,7 @@ static int a3xx_hw_init(struct msm_gpu *gpu)
 	/* Load PM4: */
 	ptr = (uint32_t *)(adreno_gpu->pm4->data);
 	len = adreno_gpu->pm4->size / 4;
-	DBG("loading PM4 ucode version: %u", ptr[0]);
+	DBG("loading PM4 ucode version: %x", ptr[1]);
 
 	gpu_write(gpu, REG_AXXX_CP_DEBUG,
 			AXXX_CP_DEBUG_DYNAMIC_CLK_DISABLE |
@@ -231,19 +264,26 @@ static int a3xx_hw_init(struct msm_gpu *gpu)
 	/* Load PFP: */
 	ptr = (uint32_t *)(adreno_gpu->pfp->data);
 	len = adreno_gpu->pfp->size / 4;
-	DBG("loading PFP ucode version: %u", ptr[0]);
+	DBG("loading PFP ucode version: %x", ptr[5]);
 
 	gpu_write(gpu, REG_A3XX_CP_PFP_UCODE_ADDR, 0);
 	for (i = 1; i < len; i++)
 		gpu_write(gpu, REG_A3XX_CP_PFP_UCODE_DATA, ptr[i]);
 
 	/* CP ROQ queue sizes (bytes) - RB:16, ST:16, IB1:32, IB2:64 */
-	if (adreno_is_a305(adreno_gpu) || adreno_is_a320(adreno_gpu))
+	if (adreno_is_a305(adreno_gpu) || adreno_is_a320(adreno_gpu)) {
 		gpu_write(gpu, REG_AXXX_CP_QUEUE_THRESHOLDS,
 				AXXX_CP_QUEUE_THRESHOLDS_CSQ_IB1_START(2) |
 				AXXX_CP_QUEUE_THRESHOLDS_CSQ_IB2_START(6) |
 				AXXX_CP_QUEUE_THRESHOLDS_CSQ_ST_START(14));
-
+	} else if (adreno_is_a330(adreno_gpu)) {
+		/* NOTE: this (value take from downstream android driver)
+		 * includes some bits outside of the known bitfields.  But
+		 * A330 has this "MERCIU queue" thing too, which might
+		 * explain a new bitfield or reshuffling:
+		 */
+		gpu_write(gpu, REG_AXXX_CP_QUEUE_THRESHOLDS, 0x003e2008);
+	}
 
 	/* clear ME_HALT to start micro engine */
 	gpu_write(gpu, REG_AXXX_CP_ME_CNTL, 0);
@@ -253,6 +293,17 @@ static int a3xx_hw_init(struct msm_gpu *gpu)
 	return 0;
 }
 
+static void a3xx_recover(struct msm_gpu *gpu)
+{
+	/* dump registers before resetting gpu, if enabled: */
+	if (hang_debug)
+		a3xx_dump(gpu);
+	gpu_write(gpu, REG_A3XX_RBBM_SW_RESET_CMD, 1);
+	gpu_read(gpu, REG_A3XX_RBBM_SW_RESET_CMD);
+	gpu_write(gpu, REG_A3XX_RBBM_SW_RESET_CMD, 0);
+	adreno_recover(gpu);
+}
+
 static void a3xx_destroy(struct msm_gpu *gpu)
 {
 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
@@ -261,27 +312,24 @@ static void a3xx_destroy(struct msm_gpu *gpu)
 	DBG("%s", gpu->name);
 
 	adreno_gpu_cleanup(adreno_gpu);
-	put_device(&a3xx_gpu->pdev->dev);
+
+#ifdef CONFIG_MSM_OCMEM
+	if (a3xx_gpu->ocmem_base)
+		ocmem_free(OCMEM_GRAPHICS, a3xx_gpu->ocmem_hdl);
+#endif
+
 	kfree(a3xx_gpu);
 }
 
 static void a3xx_idle(struct msm_gpu *gpu)
 {
-	unsigned long t;
-
 	/* wait for ringbuffer to drain: */
 	adreno_idle(gpu);
 
-	t = jiffies + ADRENO_IDLE_TIMEOUT;
-
 	/* then wait for GPU to finish: */
-	do {
-		uint32_t rbbm_status = gpu_read(gpu, REG_A3XX_RBBM_STATUS);
-		if (!(rbbm_status & A3XX_RBBM_STATUS_GPU_BUSY))
-			return;
-	} while(time_before(jiffies, t));
-
-	DRM_ERROR("timeout waiting for %s to idle!\n", gpu->name);
+	if (spin_until(!(gpu_read(gpu, REG_A3XX_RBBM_STATUS) &
+			A3XX_RBBM_STATUS_GPU_BUSY)))
+		DRM_ERROR("%s: timeout waiting for GPU to idle!\n", gpu->name);
 
 	/* TODO maybe we need to reset GPU here to recover from hang? */
 }
@@ -302,7 +350,6 @@ static irqreturn_t a3xx_irq(struct msm_gpu *gpu)
 	return IRQ_HANDLED;
 }
 
-#ifdef CONFIG_DEBUG_FS
 static const unsigned int a3xx_registers[] = {
 	0x0000, 0x0002, 0x0010, 0x0012, 0x0018, 0x0018, 0x0020, 0x0027,
 	0x0029, 0x002b, 0x002e, 0x0033, 0x0040, 0x0042, 0x0050, 0x005c,
@@ -342,11 +389,18 @@ static const unsigned int a3xx_registers[] = {
 	0x303c, 0x303c, 0x305e, 0x305f,
 };
 
+#ifdef CONFIG_DEBUG_FS
 static void a3xx_show(struct msm_gpu *gpu, struct seq_file *m)
 {
+	struct drm_device *dev = gpu->dev;
 	int i;
 
 	adreno_show(gpu, m);
+
+	mutex_lock(&dev->struct_mutex);
+
+	gpu->funcs->pm_resume(gpu);
+
 	seq_printf(m, "status:   %08x\n",
 			gpu_read(gpu, REG_A3XX_RBBM_STATUS));
 
@@ -362,16 +416,43 @@ static void a3xx_show(struct msm_gpu *gpu, struct seq_file *m)
 			seq_printf(m, "IO:R %08x %08x\n", addr<<2, val);
 		}
 	}
+
+	gpu->funcs->pm_suspend(gpu);
+
+	mutex_unlock(&dev->struct_mutex);
 }
 #endif
 
+/* would be nice to not have to duplicate the _show() stuff with printk(): */
+static void a3xx_dump(struct msm_gpu *gpu)
+{
+	int i;
+
+	adreno_dump(gpu);
+	printk("status:   %08x\n",
+			gpu_read(gpu, REG_A3XX_RBBM_STATUS));
+
+	/* dump these out in a form that can be parsed by demsm: */
+	printk("IO:region %s 00000000 00020000\n", gpu->name);
+	for (i = 0; i < ARRAY_SIZE(a3xx_registers); i += 2) {
+		uint32_t start = a3xx_registers[i];
+		uint32_t end   = a3xx_registers[i+1];
+		uint32_t addr;
+
+		for (addr = start; addr <= end; addr++) {
+			uint32_t val = gpu_read(gpu, addr);
+			printk("IO:R %08x %08x\n", addr<<2, val);
+		}
+	}
+}
+
 static const struct adreno_gpu_funcs funcs = {
 	.base = {
 		.get_param = adreno_get_param,
 		.hw_init = a3xx_hw_init,
 		.pm_suspend = msm_gpu_pm_suspend,
 		.pm_resume = msm_gpu_pm_resume,
-		.recover = adreno_recover,
+		.recover = a3xx_recover,
 		.last_fence = adreno_last_fence,
 		.submit = adreno_submit,
 		.flush = adreno_flush,
@@ -384,11 +465,20 @@ static const struct adreno_gpu_funcs funcs = {
 	},
 };
 
+static const struct msm_gpu_perfcntr perfcntrs[] = {
+	{ REG_A3XX_SP_PERFCOUNTER6_SELECT, REG_A3XX_RBBM_PERFCTR_SP_6_LO,
+			SP_ALU_ACTIVE_CYCLES, "ALUACTIVE" },
+	{ REG_A3XX_SP_PERFCOUNTER7_SELECT, REG_A3XX_RBBM_PERFCTR_SP_7_LO,
+			SP_FS_FULL_ALU_INSTRUCTIONS, "ALUFULL" },
+};
+
 struct msm_gpu *a3xx_gpu_init(struct drm_device *dev)
 {
 	struct a3xx_gpu *a3xx_gpu = NULL;
+	struct adreno_gpu *adreno_gpu;
 	struct msm_gpu *gpu;
-	struct platform_device *pdev = a3xx_pdev;
+	struct msm_drm_private *priv = dev->dev_private;
+	struct platform_device *pdev = priv->gpu_pdev;
 	struct adreno_platform_config *config;
 	int ret;
 
@@ -406,24 +496,57 @@ struct msm_gpu *a3xx_gpu_init(struct drm_device *dev)
 		goto fail;
 	}
 
-	gpu = &a3xx_gpu->base.base;
+	adreno_gpu = &a3xx_gpu->base;
+	gpu = &adreno_gpu->base;
 
-	get_device(&pdev->dev);
 	a3xx_gpu->pdev = pdev;
 
 	gpu->fast_rate = config->fast_rate;
 	gpu->slow_rate = config->slow_rate;
 	gpu->bus_freq  = config->bus_freq;
+#ifdef CONFIG_MSM_BUS_SCALING
+	gpu->bus_scale_table = config->bus_scale_table;
+#endif
 
 	DBG("fast_rate=%u, slow_rate=%u, bus_freq=%u",
 			gpu->fast_rate, gpu->slow_rate, gpu->bus_freq);
 
-	ret = adreno_gpu_init(dev, pdev, &a3xx_gpu->base,
-			&funcs, config->rev);
+	gpu->perfcntrs = perfcntrs;
+	gpu->num_perfcntrs = ARRAY_SIZE(perfcntrs);
+
+	ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, config->rev);
 	if (ret)
 		goto fail;
 
-	return &a3xx_gpu->base.base;
+	/* if needed, allocate gmem: */
+	if (adreno_is_a330(adreno_gpu)) {
+#ifdef CONFIG_MSM_OCMEM
+		/* TODO this is different/missing upstream: */
+		struct ocmem_buf *ocmem_hdl =
+				ocmem_allocate(OCMEM_GRAPHICS, adreno_gpu->gmem);
+
+		a3xx_gpu->ocmem_hdl = ocmem_hdl;
+		a3xx_gpu->ocmem_base = ocmem_hdl->addr;
+		adreno_gpu->gmem = ocmem_hdl->len;
+		DBG("using %dK of OCMEM at 0x%08x", adreno_gpu->gmem / 1024,
+				a3xx_gpu->ocmem_base);
+#endif
+	}
+
+	if (!gpu->mmu) {
+		/* TODO we think it is possible to configure the GPU to
+		 * restrict access to VRAM carveout.  But the required
+		 * registers are unknown.  For now just bail out and
+		 * limp along with just modesetting.  If it turns out
+		 * to not be possible to restrict access, then we must
+		 * implement a cmdstream validator.
+		 */
+		dev_err(dev->dev, "No memory protection without IOMMU\n");
+		ret = -ENXIO;
+		goto fail;
+	}
+
+	return gpu;
 
 fail:
 	if (a3xx_gpu)
@@ -436,19 +559,66 @@ fail:
  * The a3xx device:
  */
 
-static int a3xx_probe(struct platform_device *pdev)
+#if defined(CONFIG_MSM_BUS_SCALING) && !defined(CONFIG_OF)
+#  include <mach/kgsl.h>
+#endif
+
+static void set_gpu_pdev(struct drm_device *dev,
+		struct platform_device *pdev)
+{
+	struct msm_drm_private *priv = dev->dev_private;
+	priv->gpu_pdev = pdev;
+}
+
+static int a3xx_bind(struct device *dev, struct device *master, void *data)
 {
 	static struct adreno_platform_config config = {};
 #ifdef CONFIG_OF
-	/* TODO */
+	struct device_node *child, *node = dev->of_node;
+	u32 val;
+	int ret;
+
+	ret = of_property_read_u32(node, "qcom,chipid", &val);
+	if (ret) {
+		dev_err(dev, "could not find chipid: %d\n", ret);
+		return ret;
+	}
+
+	config.rev = ADRENO_REV((val >> 24) & 0xff,
+			(val >> 16) & 0xff, (val >> 8) & 0xff, val & 0xff);
+
+	/* find clock rates: */
+	config.fast_rate = 0;
+	config.slow_rate = ~0;
+	for_each_child_of_node(node, child) {
+		if (of_device_is_compatible(child, "qcom,gpu-pwrlevels")) {
+			struct device_node *pwrlvl;
+			for_each_child_of_node(child, pwrlvl) {
+				ret = of_property_read_u32(pwrlvl, "qcom,gpu-freq", &val);
+				if (ret) {
+					dev_err(dev, "could not find gpu-freq: %d\n", ret);
+					return ret;
+				}
+				config.fast_rate = max(config.fast_rate, val);
+				config.slow_rate = min(config.slow_rate, val);
+			}
+		}
+	}
+
+	if (!config.fast_rate) {
+		dev_err(dev, "could not find clk rates\n");
+		return -ENXIO;
+	}
+
 #else
+	struct kgsl_device_platform_data *pdata = dev->platform_data;
 	uint32_t version = socinfo_get_version();
 	if (cpu_is_apq8064ab()) {
 		config.fast_rate = 450000000;
 		config.slow_rate = 27000000;
 		config.bus_freq  = 4;
 		config.rev = ADRENO_REV(3, 2, 1, 0);
-	} else if (cpu_is_apq8064() || cpu_is_msm8960ab()) {
+	} else if (cpu_is_apq8064()) {
 		config.fast_rate = 400000000;
 		config.slow_rate = 27000000;
 		config.bus_freq  = 4;
@@ -461,6 +631,16 @@ static int a3xx_probe(struct platform_device *pdev)
 		else
 			config.rev = ADRENO_REV(3, 2, 0, 0);
 
+	} else if (cpu_is_msm8960ab()) {
+		config.fast_rate = 400000000;
+		config.slow_rate = 320000000;
+		config.bus_freq  = 4;
+
+		if (SOCINFO_VERSION_MINOR(version) == 0)
+			config.rev = ADRENO_REV(3, 2, 1, 0);
+		else
+			config.rev = ADRENO_REV(3, 2, 1, 1);
+
 	} else if (cpu_is_msm8930()) {
 		config.fast_rate = 400000000;
 		config.slow_rate = 27000000;
@@ -473,22 +653,49 @@ static int a3xx_probe(struct platform_device *pdev)
 			config.rev = ADRENO_REV(3, 0, 5, 0);
 
 	}
+#  ifdef CONFIG_MSM_BUS_SCALING
+	config.bus_scale_table = pdata->bus_scale_table;
+#  endif
 #endif
-	pdev->dev.platform_data = &config;
-	a3xx_pdev = pdev;
+	dev->platform_data = &config;
+	set_gpu_pdev(dev_get_drvdata(master), to_platform_device(dev));
 	return 0;
 }
 
+static void a3xx_unbind(struct device *dev, struct device *master,
+		void *data)
+{
+	set_gpu_pdev(dev_get_drvdata(master), NULL);
+}
+
+static const struct component_ops a3xx_ops = {
+		.bind   = a3xx_bind,
+		.unbind = a3xx_unbind,
+};
+
+static int a3xx_probe(struct platform_device *pdev)
+{
+	return component_add(&pdev->dev, &a3xx_ops);
+}
+
 static int a3xx_remove(struct platform_device *pdev)
 {
-	a3xx_pdev = NULL;
+	component_del(&pdev->dev, &a3xx_ops);
 	return 0;
 }
 
+static const struct of_device_id dt_match[] = {
+	{ .compatible = "qcom,kgsl-3d0" },
+	{}
+};
+
 static struct platform_driver a3xx_driver = {
 	.probe = a3xx_probe,
 	.remove = a3xx_remove,
-	.driver.name = "kgsl-3d0",
+	.driver = {
+		.name = "kgsl-3d0",
+		.of_match_table = dt_match,
+	},
 };
 
 void __init a3xx_register(void)