Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 58ea104d authored by Shubhraprakash Das's avatar Shubhraprakash Das
Browse files

msm: kgsl: Add hang interrupt and change HW fault generation 



Add a sysfs property to dynamically control HW hang interrupt and
enable it for A4XX by default. Also, HW fault only needs to
be generated for fatal interrupts only.

Change-Id: Ib50c57476e54cccfef0ae18a0cbb42fce29457c1
Signed-off-by: default avatarShubhraprakash Das <sadas@codeaurora.org>
parent d3394510

drivers/gpu/msm/a4xx_reg.h

+12 −0
Original line number Diff line number Diff line
@@ -14,6 +14,18 @@ 
#ifndef _A4XX_REG_H

#define _A4XX_REG_H



/* A4XX interrupt bit that are different from A3XX */

#define A4XX_INT_RBBM_ETS_MS_TIMEOUT		5

#define A4XX_INT_RBBM_ASYNC_OVERFLOW		6

#define A4XX_INT_RBBM_GPC_ERR			7

#define A4XX_INT_CP_OPCODE_ERR			9

#define A4XX_INT_CP_SW				8

#define A4XX_INT_RBBM_ATB_BUS_OVERFLOW		22

#define A4XX_INT_RBBM_DPM_CALC_ERR		28

#define A4XX_INT_RBBM_DPM_EPOCH_ERR		29

#define A4XX_INT_RBBM_DPM_THERMAL_YELLOW_ERR	30

#define A4XX_INT_RBBM_DPM_THERMAL_RED_ERR	31



/* RB registers */

#define A4XX_RB_GMEM_BASE_ADDR		0xcc0

drivers/gpu/msm/adreno.c

+93 −1
Original line number Diff line number Diff line
@@ -1807,6 +1807,13 @@ static int adreno_init(struct kgsl_device *device) 
	for (i = 6; i < FT_DETECT_REGS_COUNT; i++)

		ft_detect_regs[i] = 0;



	/* turn on hang interrupt for A4XX by default */

	if (adreno_is_a4xx(adreno_dev))

		set_bit(ADRENO_DEVICE_HANG_INTR, &adreno_dev->priv);



	if (adreno_dev->gpudev->irq_setup)

		adreno_dev->gpudev->irq_setup(adreno_dev);



	ret = adreno_perfcounter_init(device);



	/* Power down the device */
@@ -2015,7 +2022,7 @@ int adreno_reset(struct kgsl_device *device) 
 * _ft_sysfs_store() -  Common routine to write to FT sysfs files

 * @buf: value to write

 * @count: size of the value to write

 * @sysfs_cfg: KGSL FT sysfs config to write

 * @ptr: pointer to config to write

 *

 * This is a common routine to write to FT sysfs files.

 */
@@ -2242,6 +2249,89 @@ static ssize_t _ft_long_ib_detect_show(struct device *dev, 
}





/**

 * _ft_hang_intr_status_store -  Routine to enable/disable h/w hang interrupt

 * @dev: device ptr

 * @attr: Device attribute

 * @buf: value to write

 * @count: size of the value to write

 */

static ssize_t _ft_hang_intr_status_store(struct device *dev,

				struct device_attribute *attr,

				const char *buf, size_t count)

{

	unsigned int new_setting, old_setting;

	struct kgsl_device *device = kgsl_device_from_dev(dev);

	struct adreno_device *adreno_dev;

	int ret;

	if (device == NULL)

		return 0;

	adreno_dev = ADRENO_DEVICE(device);



	mutex_lock(&device->mutex);

	ret = _ft_sysfs_store(buf, count, &new_setting);

	if (ret != count)

		goto done;

	if (new_setting)

		new_setting = 1;

	old_setting =

		(test_bit(ADRENO_DEVICE_HANG_INTR, &adreno_dev->priv) ? 1 : 0);

	if (new_setting != old_setting) {

		if (new_setting)

			set_bit(ADRENO_DEVICE_HANG_INTR, &adreno_dev->priv);

		else

			clear_bit(ADRENO_DEVICE_HANG_INTR, &adreno_dev->priv);

		/* Set the new setting based on device state */

		switch (device->state) {

		case KGSL_STATE_NAP:

		case KGSL_STATE_SLEEP:

			kgsl_pwrctrl_clk(device, KGSL_PWRFLAGS_ON,

					device->state);

		case KGSL_STATE_ACTIVE:

			adreno_dev->gpudev->irq_control(adreno_dev, 1);

		/*

		 * For following states setting will be picked up on device

		 * start. Still need them in switch statement to differentiate

		 * from default

		 */

		case KGSL_STATE_SLUMBER:

		case KGSL_STATE_SUSPEND:

			break;

		default:

			ret = -EACCES;

			/* reset back to old setting on error */

			if (new_setting)

				clear_bit(ADRENO_DEVICE_HANG_INTR,

					&adreno_dev->priv);

			else

				set_bit(ADRENO_DEVICE_HANG_INTR,

					&adreno_dev->priv);

			goto done;

		}

	}

done:

	mutex_unlock(&device->mutex);

	return ret;

}



/**

 * _ft_hang_intr_status_show() -  Routine to read hardware hang interrupt

 * enablement

 * @dev: device ptr

 * @attr: Device attribute

 * @buf: value read

 */

static ssize_t _ft_hang_intr_status_show(struct device *dev,

					struct device_attribute *attr,

					char *buf)

{

	struct adreno_device *adreno_dev = _get_adreno_dev(dev);

	if (adreno_dev == NULL)

		return 0;

	return snprintf(buf, PAGE_SIZE, "%d\n",

		test_bit(ADRENO_DEVICE_HANG_INTR, &adreno_dev->priv) ? 1 : 0);

}



#define FT_DEVICE_ATTR(name) \

	DEVICE_ATTR(name, 0644,	_ ## name ## _show, _ ## name ## _store);
@@ -2249,6 +2339,7 @@ FT_DEVICE_ATTR(ft_policy); 
FT_DEVICE_ATTR(ft_pagefault_policy);

FT_DEVICE_ATTR(ft_fast_hang_detect);

FT_DEVICE_ATTR(ft_long_ib_detect);

FT_DEVICE_ATTR(ft_hang_intr_status);





const struct device_attribute *ft_attr_list[] = {
@@ -2256,6 +2347,7 @@ const struct device_attribute *ft_attr_list[] = { 
	&dev_attr_ft_pagefault_policy,

	&dev_attr_ft_fast_hang_detect,

	&dev_attr_ft_long_ib_detect,

	&dev_attr_ft_hang_intr_status,

	NULL,

};

drivers/gpu/msm/adreno.h

+14 −0
Original line number Diff line number Diff line
@@ -197,6 +197,7 @@ enum adreno_device_flags { 
	ADRENO_DEVICE_PWRON_FIXUP = 1,

	ADRENO_DEVICE_INITIALIZED = 2,

	ADRENO_DEVICE_CORESIGHT = 3,

	ADRENO_DEVICE_HANG_INTR = 4,

};



#define PERFCOUNTER_FLAG_NONE 0x0
@@ -412,6 +413,17 @@ struct adreno_coresight { 
};





struct adreno_irq_funcs {

	void (*func)(struct adreno_device *, int);

};

#define ADRENO_IRQ_CALLBACK(_c) { .func = _c }



struct adreno_irq {

	unsigned int mask;

	struct adreno_irq_funcs *funcs;

	int funcs_count;

};



struct adreno_gpudev {

	/*

	 * These registers are in a different location on different devices,
@@ -427,11 +439,13 @@ struct adreno_gpudev { 


	struct adreno_coresight *coresight;



	struct adreno_irq *irq;

	/* GPU specific function hooks */

	int (*ctxt_create)(struct adreno_device *, struct adreno_context *);

	irqreturn_t (*irq_handler)(struct adreno_device *);

	void (*irq_control)(struct adreno_device *, int);

	unsigned int (*irq_pending)(struct adreno_device *);

	void (*irq_setup)(struct adreno_device *);

	void * (*snapshot)(struct adreno_device *, void *, int *, int);

	int (*rb_init)(struct adreno_device *, struct adreno_ringbuffer *);

	int (*perfcounter_init)(struct adreno_device *);

drivers/gpu/msm/adreno_a3xx.c

+128 −52
Original line number Diff line number Diff line
@@ -3134,7 +3134,13 @@ int a3xx_rb_init(struct adreno_device *adreno_dev, 
	return 0;

}



static void a3xx_err_callback(struct adreno_device *adreno_dev, int bit)

/*

 * a3xx_a4xx_err_callback() - Common interrupts shared between A4XX

 * and A3XX

 * @adreno_dev: Pointer to device

 * @bit: Interrupt bit

 */

void a3xx_a4xx_err_callback(struct adreno_device *adreno_dev, int bit)

{

	struct kgsl_device *device = &adreno_dev->dev;

	const char *err = "";
@@ -3175,17 +3181,8 @@ static void a3xx_err_callback(struct adreno_device *adreno_dev, int bit) 
	case A3XX_INT_RBBM_PFP_MS_TIMEOUT:

		err = "RBBM: PFP master split timeout";

		break;

	case A3XX_INT_RBBM_ATB_BUS_OVERFLOW:

		err = "RBBM: ATB bus oveflow";

		break;

	case A3XX_INT_VFD_ERROR:

		err = "VFD: Out of bounds access";

		break;

	case A3XX_INT_CP_T0_PACKET_IN_IB:

		err = "ringbuffer TO packet in IB interrupt";

		break;

	case A3XX_INT_CP_OPCODE_ERROR:

		err = "ringbuffer opcode error interrupt";

	case A3XX_INT_UCHE_OOB_ACCESS:

		err = "UCHE:  Out of bounds access";

		break;

	case A3XX_INT_CP_RESERVED_BIT_ERROR:

		err = "ringbuffer reserved bit error interrupt";
@@ -3209,21 +3206,64 @@ static void a3xx_err_callback(struct adreno_device *adreno_dev, int bit) 
	case A3XX_INT_MISC_HANG_DETECT:

		err = "MISC: GPU hang detected";

		break;

	case A3XX_INT_UCHE_OOB_ACCESS:

		err = "UCHE:  Out of bounds access";

	}

done:

	KGSL_DRV_CRIT(device, "%s\n", err);

}



static void a3xx_err_callback(struct adreno_device *adreno_dev, int bit)

{

	struct kgsl_device *device = &adreno_dev->dev;

	const char *err = "";



	switch (bit) {

	case A3XX_INT_RBBM_ATB_BUS_OVERFLOW:

		err = "RBBM: ATB bus oveflow";

		break;

	case A3XX_INT_VFD_ERROR:

		err = "VFD: Out of bounds access";

		break;

	case A3XX_INT_CP_OPCODE_ERROR:

		err = "ringbuffer opcode error interrupt";

		break;

	case A3XX_INT_CP_T0_PACKET_IN_IB:

		err = "ringbuffer TO packet in IB interrupt";

		break;

	default:

		return;

	}

	KGSL_DRV_CRIT(device, "%s\n", err);

}



/*

 * a3xx_fatal_err_callback() - Isr for fatal interrupts that hang GPU

 * @adreno_dev: Pointer to device

 * @bit: Interrupt bit

 *

 * Called for both A4XX and A3XX

 */

void a3xx_fatal_err_callback(struct adreno_device *adreno_dev, int bit)

{

	struct kgsl_device *device = &adreno_dev->dev;



	/* Call the other error routines to get the error print */

	switch (bit) {

	case A3XX_INT_RBBM_ATB_BUS_OVERFLOW:

	case A3XX_INT_VFD_ERROR:

	case A3XX_INT_CP_T0_PACKET_IN_IB:

		a3xx_err_callback(adreno_dev, bit);

		break;

	default:

		a3xx_a4xx_err_callback(adreno_dev, bit);

	}



	kgsl_pwrctrl_irq(device, KGSL_PWRFLAGS_OFF);



done:

	/* Trigger a fault in the dispatcher - this will effect a restart */

	adreno_dispatcher_irq_fault(device);

}



static void a3xx_gpu_idle_callback(struct adreno_device *adreno_dev,

void a3xx_gpu_idle_callback(struct adreno_device *adreno_dev,

					int irq)

{

	struct kgsl_device *device = &adreno_dev->dev;
@@ -3245,7 +3285,7 @@ static void a3xx_gpu_idle_callback(struct adreno_device *adreno_dev, 
 * Handle the cp interrupt generated by GPU, common function between a3xx and

 * a4xx devices

 */

static void a3xx_cp_callback(struct adreno_device *adreno_dev, int irq)

void a3xx_cp_callback(struct adreno_device *adreno_dev, int irq)

{

	struct kgsl_device *device = &adreno_dev->dev;
@@ -3699,8 +3739,6 @@ static void a3xx_perfcounter_restore(struct adreno_device *adreno_dev) 


}



#define A3XX_IRQ_CALLBACK(_c) { .func = _c }



#define A3XX_INT_MASK \

	((1 << A3XX_INT_RBBM_GPU_IDLE) |         \

	 (1 << A3XX_INT_RBBM_AHB_ERROR) |        \
@@ -3716,38 +3754,70 @@ static void a3xx_perfcounter_restore(struct adreno_device *adreno_dev) 
	 (1 << A3XX_INT_CP_AHB_ERROR_HALT) |     \

	 (1 << A3XX_INT_UCHE_OOB_ACCESS))



static struct {

	void (*func)(struct adreno_device *, int);

} a3xx_irq_funcs[] = {

	A3XX_IRQ_CALLBACK(a3xx_gpu_idle_callback), /* 0 - RBBM_GPU_IDLE */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 1 - RBBM_AHB_ERROR */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 2 - RBBM_REG_TIMEOUT */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 3 - RBBM_ME_MS_TIMEOUT */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 4 - RBBM_PFP_MS_TIMEOUT */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 5 - RBBM_ATB_BUS_OVERFLOW */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 6 - RBBM_VFD_ERROR */

	A3XX_IRQ_CALLBACK(NULL),	       /* 7 - CP_SW */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 8 - CP_T0_PACKET_IN_IB */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 9 - CP_OPCODE_ERROR */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 10 - CP_RESERVED_BIT_ERROR */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 11 - CP_HW_FAULT */

	A3XX_IRQ_CALLBACK(NULL),	       /* 12 - CP_DMA */

	A3XX_IRQ_CALLBACK(a3xx_cp_callback),   /* 13 - CP_IB2_INT */

	A3XX_IRQ_CALLBACK(a3xx_cp_callback),   /* 14 - CP_IB1_INT */

	A3XX_IRQ_CALLBACK(a3xx_cp_callback),   /* 15 - CP_RB_INT */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 16 - CP_REG_PROTECT_FAULT */

	A3XX_IRQ_CALLBACK(NULL),	       /* 17 - CP_RB_DONE_TS */

	A3XX_IRQ_CALLBACK(NULL),	       /* 18 - CP_VS_DONE_TS */

	A3XX_IRQ_CALLBACK(NULL),	       /* 19 - CP_PS_DONE_TS */

	A3XX_IRQ_CALLBACK(NULL),	       /* 20 - CP_CACHE_FLUSH_TS */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 21 - CP_AHB_ERROR_FAULT */

	A3XX_IRQ_CALLBACK(NULL),	       /* 22 - Unused */

	A3XX_IRQ_CALLBACK(NULL),	       /* 23 - Unused */

	A3XX_IRQ_CALLBACK(NULL),	       /* 24 - MISC_HANG_DETECT */

	A3XX_IRQ_CALLBACK(a3xx_err_callback),  /* 25 - UCHE_OOB_ACCESS */

static struct adreno_irq_funcs a3xx_irq_funcs[] = {

	ADRENO_IRQ_CALLBACK(a3xx_gpu_idle_callback), /* 0 - RBBM_GPU_IDLE */

	ADRENO_IRQ_CALLBACK(a3xx_a4xx_err_callback),  /* 1 - RBBM_AHB_ERROR */

	ADRENO_IRQ_CALLBACK(a3xx_a4xx_err_callback),  /* 2 - RBBM_REG_TIMEOUT */

	/* * 3 - RBBM_ME_MS_TIMEOUT */

	ADRENO_IRQ_CALLBACK(a3xx_a4xx_err_callback),

	/* 4 - RBBM_PFP_MS_TIMEOUT */

	ADRENO_IRQ_CALLBACK(a3xx_a4xx_err_callback),

	ADRENO_IRQ_CALLBACK(a3xx_err_callback),  /* 5 - RBBM_ATB_BUS_OVERFLOW */

	ADRENO_IRQ_CALLBACK(a3xx_err_callback),  /* 6 - RBBM_VFD_ERROR */

	ADRENO_IRQ_CALLBACK(NULL),	       /* 7 - CP_SW */

	/* 8 - CP_T0_PACKET_IN_IB */

	ADRENO_IRQ_CALLBACK(a3xx_fatal_err_callback),

	ADRENO_IRQ_CALLBACK(a3xx_fatal_err_callback),  /* 9 - CP_OPCODE_ERROR */

	/* 10 - CP_RESERVED_BIT_ERROR */

	ADRENO_IRQ_CALLBACK(a3xx_fatal_err_callback),

	ADRENO_IRQ_CALLBACK(a3xx_fatal_err_callback),  /* 11 - CP_HW_FAULT */

	ADRENO_IRQ_CALLBACK(NULL),	       /* 12 - CP_DMA */

	ADRENO_IRQ_CALLBACK(a3xx_cp_callback),   /* 13 - CP_IB2_INT */

	ADRENO_IRQ_CALLBACK(a3xx_cp_callback),   /* 14 - CP_IB1_INT */

	ADRENO_IRQ_CALLBACK(a3xx_cp_callback),   /* 15 - CP_RB_INT */

	/* 16 - CP_REG_PROTECT_FAULT */

	ADRENO_IRQ_CALLBACK(a3xx_fatal_err_callback),

	ADRENO_IRQ_CALLBACK(NULL),	       /* 17 - CP_RB_DONE_TS */

	ADRENO_IRQ_CALLBACK(NULL),	       /* 18 - CP_VS_DONE_TS */

	ADRENO_IRQ_CALLBACK(NULL),	       /* 19 - CP_PS_DONE_TS */

	ADRENO_IRQ_CALLBACK(NULL),	       /* 20 - CP_CACHE_FLUSH_TS */

	/* 21 - CP_AHB_ERROR_FAULT */

	ADRENO_IRQ_CALLBACK(a3xx_fatal_err_callback),

	ADRENO_IRQ_CALLBACK(NULL),	       /* 22 - Unused */

	ADRENO_IRQ_CALLBACK(NULL),	       /* 23 - Unused */

	ADRENO_IRQ_CALLBACK(a3xx_fatal_err_callback),/* 24 - MISC_HANG_DETECT */

	ADRENO_IRQ_CALLBACK(a3xx_a4xx_err_callback),  /* 25 - UCHE_OOB_ACCESS */

	/* 26 to 31 - Unused */

};



static struct adreno_irq a3xx_irq = {

	.funcs = a3xx_irq_funcs,

	.funcs_count = ARRAY_SIZE(a3xx_irq_funcs),

	.mask = A3XX_INT_MASK,

};



/*

 * a3xx_irq_func_setup() - Sets up callback functions and mask for interrupts

 * @adreno_dev: Adreno device pointer

 *

 * Called during initialization

 */

void a3xx_irq_setup(struct adreno_device *adreno_dev)

{

	struct adreno_irq *irq_params = adreno_dev->gpudev->irq;

	int i;

	/* On a330v2 only the hang interrupt should be fatal */

	if (adreno_is_a330v2(adreno_dev)) {

		for (i = 0; i < irq_params->funcs_count; i++) {

			if ((irq_params->funcs[i].func ==

				a3xx_fatal_err_callback) &&

				A3XX_INT_MISC_HANG_DETECT != i)

				irq_params->funcs[i].func =

						a3xx_err_callback;

		}

	}

}



/*

 * a3xx_irq_handler() - Interrupt handler function

 * @adreno_dev: Pointer to adreno device
@@ -3758,16 +3828,18 @@ static struct { 
irqreturn_t a3xx_irq_handler(struct adreno_device *adreno_dev)

{

	struct kgsl_device *device = &adreno_dev->dev;

	struct adreno_irq *irq_params = adreno_dev->gpudev->irq;

	irqreturn_t ret = IRQ_NONE;

	unsigned int status, tmp;

	int i;



	adreno_readreg(adreno_dev, ADRENO_REG_RBBM_INT_0_STATUS, &status);



	for (tmp = status, i = 0; tmp && i < ARRAY_SIZE(a3xx_irq_funcs); i++) {

	for (tmp = status, i = 0; tmp &&

			i < irq_params->funcs_count; i++) {

		if (tmp & 1) {

			if (a3xx_irq_funcs[i].func != NULL) {

				a3xx_irq_funcs[i].func(adreno_dev, i);

			if (irq_params->funcs[i].func != NULL) {

				irq_params->funcs[i].func(adreno_dev, i);

				ret = IRQ_HANDLED;

			} else {

				KGSL_DRV_CRIT(device,
@@ -3797,7 +3869,9 @@ void a3xx_irq_control(struct adreno_device *adreno_dev, int state) 
{

	if (state)

		adreno_writereg(adreno_dev, ADRENO_REG_RBBM_INT_0_MASK,

				A3XX_INT_MASK);

			adreno_dev->gpudev->irq->mask |

			(test_bit(ADRENO_DEVICE_HANG_INTR, &adreno_dev->priv) ?

				(1 << A3XX_INT_MISC_HANG_DETECT) : 0));

	else

		adreno_writereg(adreno_dev, ADRENO_REG_RBBM_INT_0_MASK, 0);

}
@@ -4637,6 +4711,7 @@ const struct adreno_reg_offsets a3xx_reg_offsets = { 
struct adreno_gpudev adreno_a3xx_gpudev = {

	.reg_offsets = &a3xx_reg_offsets,

	.perfcounters = &a3xx_perfcounters,

	.irq = &a3xx_irq,



	.ctxt_create = a3xx_drawctxt_create,

	.rb_init = a3xx_rb_init,
@@ -4647,6 +4722,7 @@ struct adreno_gpudev adreno_a3xx_gpudev = { 
	.irq_control = a3xx_irq_control,

	.irq_handler = a3xx_irq_handler,

	.irq_pending = a3xx_irq_pending,

	.irq_setup = a3xx_irq_setup,

	.busy_cycles = a3xx_busy_cycles,

	.start = a3xx_start,

	.snapshot = a3xx_snapshot,

drivers/gpu/msm/adreno_a3xx.h

+5 −1
Original line number Diff line number Diff line
@@ -13,7 +13,6 @@ 
#ifndef __A3XX_H

#define __A3XX_H



void a3xx_err_callback(struct adreno_device *adreno_dev, int bit);

irqreturn_t a3xx_irq_handler(struct adreno_device *adreno_dev);

void a3xx_irq_control(struct adreno_device *adreno_dev, int state);

unsigned int a3xx_irq_pending(struct adreno_device *adreno_dev);
@@ -28,4 +27,9 @@ uint64_t a3xx_perfcounter_read(struct adreno_device *adreno_dev, 
void a3xx_perfcounter_disable(struct adreno_device *adreno_dev,

	unsigned int group, unsigned int counter);

void a3xx_soft_reset(struct adreno_device *adreno_dev);

void a3xx_irq_setup(struct adreno_device *adreno_dev);

void a3xx_a4xx_err_callback(struct adreno_device *adreno_dev, int bit);

void a3xx_fatal_err_callback(struct adreno_device *adreno_dev, int bit);

void a3xx_gpu_idle_callback(struct adreno_device *adreno_dev, int irq);

void a3xx_cp_callback(struct adreno_device *adreno_dev, int irq);

#endif /*__A3XX_H */