msm: kgsl: embed kgsl_context struct in adreno_context struct

Having a separate allocated struct for the device specific context
makes ownership unclear, which could lead to reference counting
problems or invalid pointers. Also, duplicate members were
starting to appear in adreno_context because there wasn't a safe
way to reach the kgsl_context from some parts of the adreno code.
This can now be done via container_of().

This change alters the lifecycle of the context->id, which is
now freed when the context reference count hits zero rather
than in kgsl_context_detach().

It also changes the context creation and destruction sequence.
The device specific code must allocate a structure containing
a struct kgsl_context and passes a pointer it to kgsl_init_context()
before doing any device specific initialization. There is also a
separate drawctxt_detach() callback for doing device specific
cleanup. This is separate from freeing memory, which is done
by the drawctxt_destroy() callback.

Change-Id: I7d238476a3bfec98fd8dbc28971cf3187a81dac2
Signed-off-by: Jeremy Gebben <jgebben@codeaurora.org>
Signed-off-by: Carter Cooper <ccooper@codeaurora.org>

drivers/gpu/msm/adreno.c

@@ -814,8 +814,6 @@ static int adreno_iommu_setstate(struct kgsl_device *device,
 
 	adreno_ctx = ADRENO_CONTEXT(context);
 
-
-
 	result = kgsl_mmu_enable_clk(&device->mmu,
 				KGSL_IOMMU_CONTEXT_USER);
 	if (result)
@@ -894,7 +892,6 @@ static int adreno_gpummu_setstate(struct kgsl_device *device,
 		context = kgsl_context_get(device, context_id);
 		if (context == NULL)
 			return -EINVAL;
-
 		adreno_ctx = ADRENO_CONTEXT(context);
 
 		if (flags & KGSL_MMUFLAGS_PTUPDATE) {
@@ -2048,8 +2045,6 @@ static bool adreno_hw_isidle(struct kgsl_device *device)
 	} else if (adreno_is_a3xx(adreno_dev)) {
 		if (!(reg_rbbm_status & 0x80000000))
 			return true;
-	} else {
-		BUG();
 	}
 
 	return false;
@@ -2100,11 +2095,12 @@ int adreno_idle(struct kgsl_device *device)
 
 	if (adreno_is_a2xx(adreno_dev))
 		kgsl_cffdump_regpoll(device,
-			adreno_dev->gpudev->reg_rbbm_status << 2, 0x110, 0x110);
+			adreno_getreg(adreno_dev, ADRENO_REG_RBBM_STATUS) << 2,
+			0x110, 0x110);
 	else
 		kgsl_cffdump_regpoll(device,
-			adreno_dev->gpudev->reg_rbbm_status << 2, 0,
-			0x80000000);
+			adreno_getreg(adreno_dev, ADRENO_REG_RBBM_STATUS) << 2,
+		        0x00000000, 0x80000000);
 
 	while (time_before(jiffies, wait)) {
 		if (adreno_isidle(device))

drivers/gpu/msm/adreno_a2xx.c

@@ -1350,7 +1350,6 @@ static int a2xx_create_gmem_shadow(struct adreno_device *adreno_dev,
 			struct adreno_context *drawctxt)
 {
 	int result;
-	struct kgsl_device *device = &adreno_dev->dev;
 
 	calc_gmemsize(&drawctxt->context_gmem_shadow, adreno_dev->gmem_size);
 	tmp_ctx.gmem_base = adreno_dev->gmem_base;
@@ -1365,7 +1364,7 @@ static int a2xx_create_gmem_shadow(struct adreno_device *adreno_dev,
 	drawctxt->flags |= CTXT_FLAGS_GMEM_SHADOW;
 
 	/* blank out gmem shadow. */
-	kgsl_sharedmem_set(device,
+	kgsl_sharedmem_set(drawctxt->base.device,
 			&drawctxt->context_gmem_shadow.gmemshadow, 0, 0,
 			drawctxt->context_gmem_shadow.size);
 
@@ -1390,7 +1389,7 @@ static int a2xx_create_gmem_shadow(struct adreno_device *adreno_dev,
 	kgsl_cache_range_op(&drawctxt->context_gmem_shadow.gmemshadow,
 			    KGSL_CACHE_OP_FLUSH);
 
-	kgsl_cffdump_syncmem(drawctxt->dev_priv,
+	kgsl_cffdump_syncmem(drawctxt->base.device,
 			&drawctxt->context_gmem_shadow.gmemshadow,
 			drawctxt->context_gmem_shadow.gmemshadow.gpuaddr,
 			drawctxt->context_gmem_shadow.gmemshadow.size, false);
@@ -1402,7 +1401,6 @@ static int a2xx_drawctxt_create(struct adreno_device *adreno_dev,
 	struct adreno_context *drawctxt)
 {
 	int ret;
-	struct kgsl_device *device = &adreno_dev->dev;
 
 	/*
 	 * Allocate memory for the GPU state and the context commands.
@@ -1417,8 +1415,8 @@ static int a2xx_drawctxt_create(struct adreno_device *adreno_dev,
 	if (ret)
 		return ret;
 
-	kgsl_sharedmem_set(device, &drawctxt->gpustate, 0,
-		0, _context_size(adreno_dev));
+	kgsl_sharedmem_set(drawctxt->base.device, &drawctxt->gpustate,
+		0, 0, _context_size(adreno_dev));
 
 	tmp_ctx.cmd = tmp_ctx.start
 	    = (unsigned int *)((char *)drawctxt->gpustate.hostptr + CMD_OFFSET);
@@ -1442,8 +1440,8 @@ static int a2xx_drawctxt_create(struct adreno_device *adreno_dev,
 	kgsl_cache_range_op(&drawctxt->gpustate,
 			    KGSL_CACHE_OP_FLUSH);
 
-	kgsl_cffdump_syncmem(drawctxt->dev_priv, &drawctxt->gpustate,
-			drawctxt->gpustate.gpuaddr,
+	kgsl_cffdump_syncmem(drawctxt->base.device,
+			&drawctxt->gpustate, drawctxt->gpustate.gpuaddr,
 			drawctxt->gpustate.size, false);
 
 done:
@@ -1518,7 +1516,7 @@ static int a2xx_drawctxt_save(struct adreno_device *adreno_dev,
 		return 0;
 
 	if (!(context->flags & CTXT_FLAGS_PREAMBLE)) {
-		kgsl_cffdump_syncmem(context->dev_priv, &context->gpustate,
+		kgsl_cffdump_syncmem(context->base.device, &context->gpustate,
 			context->reg_save[1],
 			context->reg_save[2] << 2, true);
 		/* save registers and constants. */
@@ -1530,7 +1528,7 @@ static int a2xx_drawctxt_save(struct adreno_device *adreno_dev,
 			return ret;
 
 		if (context->flags & CTXT_FLAGS_SHADER_SAVE) {
-			kgsl_cffdump_syncmem(context->dev_priv,
+			kgsl_cffdump_syncmem(context->base.device,
 				&context->gpustate,
 				context->shader_save[1],
 				context->shader_save[2] << 2, true);
@@ -1539,7 +1537,7 @@ static int a2xx_drawctxt_save(struct adreno_device *adreno_dev,
 				KGSL_CMD_FLAGS_PMODE,
 				context->shader_save, 3);
 
-			kgsl_cffdump_syncmem(context->dev_priv,
+			kgsl_cffdump_syncmem(context->base.device,
 				&context->gpustate,
 				context->shader_fixup[1],
 				context->shader_fixup[2] << 2, true);
@@ -1560,7 +1558,7 @@ static int a2xx_drawctxt_save(struct adreno_device *adreno_dev,
 
 	if ((context->flags & CTXT_FLAGS_GMEM_SAVE) &&
 	    (context->flags & CTXT_FLAGS_GMEM_SHADOW)) {
-		kgsl_cffdump_syncmem(context->dev_priv, &context->gpustate,
+		kgsl_cffdump_syncmem(context->base.device, &context->gpustate,
 			context->context_gmem_shadow.gmem_save[1],
 			context->context_gmem_shadow.gmem_save[2] << 2, true);
 		/* save gmem.
@@ -1572,7 +1570,7 @@ static int a2xx_drawctxt_save(struct adreno_device *adreno_dev,
 
 		if (ret)
 			return ret;
-		kgsl_cffdump_syncmem(context->dev_priv, &context->gpustate,
+		kgsl_cffdump_syncmem(context->base.device, &context->gpustate,
 			context->chicken_restore[1],
 			context->chicken_restore[2] << 2, true);
 
@@ -1635,7 +1633,7 @@ static int a2xx_drawctxt_restore(struct adreno_device *adreno_dev,
 	 *  (note: changes shader. shader must not already be restored.)
 	 */
 	if (context->flags & CTXT_FLAGS_GMEM_RESTORE) {
-		kgsl_cffdump_syncmem(context->dev_priv, &context->gpustate,
+		kgsl_cffdump_syncmem(context->base.device, &context->gpustate,
 			context->context_gmem_shadow.gmem_restore[1],
 			context->context_gmem_shadow.gmem_restore[2] << 2,
 			true);
@@ -1647,7 +1645,7 @@ static int a2xx_drawctxt_restore(struct adreno_device *adreno_dev,
 			return ret;
 
 		if (!(context->flags & CTXT_FLAGS_PREAMBLE)) {
-			kgsl_cffdump_syncmem(context->dev_priv,
+			kgsl_cffdump_syncmem(context->base.device,
 				&context->gpustate,
 				context->chicken_restore[1],
 				context->chicken_restore[2] << 2, true);
@@ -1664,7 +1662,7 @@ static int a2xx_drawctxt_restore(struct adreno_device *adreno_dev,
 	}
 
 	if (!(context->flags & CTXT_FLAGS_PREAMBLE)) {
-		kgsl_cffdump_syncmem(context->dev_priv, &context->gpustate,
+		kgsl_cffdump_syncmem(context->base.device, &context->gpustate,
 			context->reg_restore[1],
 			context->reg_restore[2] << 2, true);
 
@@ -1676,7 +1674,7 @@ static int a2xx_drawctxt_restore(struct adreno_device *adreno_dev,
 
 		/* restore shader instructions & partitioning. */
 		if (context->flags & CTXT_FLAGS_SHADER_RESTORE) {
-			kgsl_cffdump_syncmem(context->dev_priv,
+			kgsl_cffdump_syncmem(context->base.device,
 				&context->gpustate,
 				context->shader_restore[1],
 				context->shader_restore[2] << 2, true);

drivers/gpu/msm/adreno_a3xx.c

@@ -2426,7 +2426,7 @@ static int a3xx_drawctxt_save(struct adreno_device *adreno_dev,
 		 * already be saved.)
 		 */
 
-		kgsl_cffdump_syncmem(context->dev_priv,
+		kgsl_cffdump_syncmem(context->base.device,
 			&context->gpustate,
 			context->context_gmem_shadow.gmem_save[1],
 			context->context_gmem_shadow.gmem_save[2] << 2, true);
@@ -2487,7 +2487,7 @@ static int a3xx_drawctxt_restore(struct adreno_device *adreno_dev,
 	 */
 
 	if (context->flags & CTXT_FLAGS_GMEM_RESTORE) {
-		kgsl_cffdump_syncmem(context->dev_priv,
+		kgsl_cffdump_syncmem(context->base.device,
 			&context->gpustate,
 			context->context_gmem_shadow.gmem_restore[1],
 			context->context_gmem_shadow.gmem_restore[2] << 2,

drivers/gpu/msm/adreno_drawctxt.c

@@ -375,17 +375,14 @@ void adreno_drawctxt_invalidate(struct kgsl_device *device,
 
 /**
  * adreno_drawctxt_create - create a new adreno draw context
- * @device - KGSL device to create the context on
- * @pagetable - Pagetable for the context
- * @context- Generic KGSL context structure
- * @flags - flags for the context (passed from user space)
+ * @dev_priv: the owner of the context
+ * @flags: flags for the context (passed from user space)
  *
- * Create a new draw context for the 3D core.  Return 0 on success,
- * or error code on failure.
+ * Create and return a new draw context for the 3D core.
  */
-	struct kgsl_context *
+struct kgsl_context *
 adreno_drawctxt_create(struct kgsl_device_private *dev_priv,
-		uint32_t *flags)
+			uint32_t *flags)
 {
 	struct adreno_context *drawctxt;
 	struct kgsl_device *device = dev_priv->device;
@@ -393,7 +390,6 @@ adreno_drawctxt_create(struct kgsl_device_private *dev_priv,
 	int ret;
 
 	drawctxt = kzalloc(sizeof(struct adreno_context), GFP_KERNEL);
-
 	if (drawctxt == NULL)
 		return ERR_PTR(-ENOMEM);
 

drivers/gpu/msm/kgsl.c

@@ -410,7 +410,7 @@ static void kgsl_mem_entry_detach_process(struct kgsl_mem_entry *entry)
  * function to initialize the common members of its context struct.
  * If this function succeeds, reference counting is active in the context
  * struct and the caller should kgsl_context_put() it on error.
- * If it fails, the caller should just free the context structer
+ * If it fails, the caller should just free the context structure
  * it passed in.
  */
 int kgsl_context_init(struct kgsl_device_private *dev_priv,
@@ -454,9 +454,8 @@ int kgsl_context_init(struct kgsl_device_private *dev_priv,
 	context->pid = dev_priv->process_priv->pid;
 
 	ret = kgsl_sync_timeline_create(context);
-	if (ret) {
+	if (ret)
 		goto fail_free_id;
-	}
 
 	/* Initialize the pending event list */
 	INIT_LIST_HEAD(&context->events);
@@ -514,11 +513,12 @@ int kgsl_context_detach(struct kgsl_context *context)
 
 	/*
 	 * Cancel events after the device-specific context is
-	 * destroyed, to avoid possibly freeing memory while
+	 * detached, to avoid possibly freeing memory while
 	 * it is still in use by the GPU.
 	 */
 
 	kgsl_context_cancel_events(device, context);
+
 	kgsl_context_put(context);
 
 	return ret;

drivers/gpu/msm/kgsl_cffdump.c

@@ -414,7 +414,6 @@ void kgsl_cffdump_syncmem(struct kgsl_device *device,
 			  struct kgsl_memdesc *memdesc, uint gpuaddr,
 			  uint sizebytes, bool clean_cache)
 {
-	struct kgsl_device *device = dev_priv->device;
 	const void *src;
 
 	if (!device->cff_dump_enable)

drivers/gpu/msm/kgsl_cffdump.h

@@ -74,7 +74,7 @@ static inline void kgsl_cffdump_close(struct kgsl_device *device)
 	return;
 }
 
-static inline void kgsl_cffdump_syncmem(struct kgsl_device_private *dev_priv,
+static inline void kgsl_cffdump_syncmem(struct kgsl_device *device,
 		struct kgsl_memdesc *memdesc, uint physaddr, uint sizebytes,
 		bool clean_cache)
 {

drivers/gpu/msm/kgsl_device.h

@@ -16,6 +16,7 @@
 #include <linux/slab.h>
 #include <linux/idr.h>
 #include <linux/pm_qos.h>
+#include <linux/sched.h>
 
 #include "kgsl.h"
 #include "kgsl_mmu.h"
@@ -526,7 +527,7 @@ kgsl_context_put(struct kgsl_context *context)
  *
  * Check if a context has been destroyed by userspace and is only waiting
  * for reference counts to go away. This check is used to weed out
- * contexts that shouldn't use the gpu, so NULL is considered detached.
+ * contexts that shouldn't use the gpu so NULL is considered detached.
  */
 static inline bool kgsl_context_detached(struct kgsl_context *context)
 {