[deliverable/linux.git] / drivers / gpu / drm / amd / amdkfd / kfd_interrupt.c

/*
 * Copyright 2014 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 * OTHER DEALINGS IN THE SOFTWARE.
 */

/*
 * KFD Interrupts.
 *
 * AMD GPUs deliver interrupts by pushing an interrupt description onto the
 * interrupt ring and then sending an interrupt. KGD receives the interrupt
 * in ISR and sends us a pointer to each new entry on the interrupt ring.
 *
 * We generally can't process interrupt-signaled events from ISR, so we call
 * out to each interrupt client module (currently only the scheduler) to ask if
 * each interrupt is interesting. If they return true, then it requires further
 * processing so we copy it to an internal interrupt ring and call each
 * interrupt client again from a work-queue.
 *
 * There's no acknowledgment for the interrupts we use. The hardware simply
 * queues a new interrupt each time without waiting.
 *
 * The fixed-size internal queue means that it's possible for us to lose
 * interrupts because we have no back-pressure to the hardware.
 */

#include <linux/slab.h>
#include <linux/device.h>
#include "kfd_priv.h"

#define KFD_INTERRUPT_RING_SIZE 1024

static void interrupt_wq(struct work_struct *);

int kfd_interrupt_init(struct kfd_dev *kfd)
{
	void *interrupt_ring = kmalloc_array(KFD_INTERRUPT_RING_SIZE,
					kfd->device_info->ih_ring_entry_size,
					GFP_KERNEL);
	if (!interrupt_ring)
		return -ENOMEM;

	kfd->interrupt_ring = interrupt_ring;
	kfd->interrupt_ring_size =
		KFD_INTERRUPT_RING_SIZE * kfd->device_info->ih_ring_entry_size;
	atomic_set(&kfd->interrupt_ring_wptr, 0);
	atomic_set(&kfd->interrupt_ring_rptr, 0);

	spin_lock_init(&kfd->interrupt_lock);

	INIT_WORK(&kfd->interrupt_work, interrupt_wq);

	kfd->interrupts_active = true;

	/*
	 * After this function returns, the interrupt will be enabled. This
	 * barrier ensures that the interrupt running on a different processor
	 * sees all the above writes.
	 */
	smp_wmb();

	return 0;
}

void kfd_interrupt_exit(struct kfd_dev *kfd)
{
	/*
	 * Stop the interrupt handler from writing to the ring and scheduling
	 * workqueue items. The spinlock ensures that any interrupt running
	 * after we have unlocked sees interrupts_active = false.
	 */
	unsigned long flags;

	spin_lock_irqsave(&kfd->interrupt_lock, flags);
	kfd->interrupts_active = false;
	spin_unlock_irqrestore(&kfd->interrupt_lock, flags);

	/*
	 * Flush_scheduled_work ensures that there are no outstanding
	 * work-queue items that will access interrupt_ring. New work items
	 * can't be created because we stopped interrupt handling above.
	 */
	flush_scheduled_work();

	kfree(kfd->interrupt_ring);
}

/*
 * This assumes that it can't be called concurrently with itself
 * but only with dequeue_ih_ring_entry.
 */
bool enqueue_ih_ring_entry(struct kfd_dev *kfd,	const void *ih_ring_entry)
{
	unsigned int rptr = atomic_read(&kfd->interrupt_ring_rptr);
	unsigned int wptr = atomic_read(&kfd->interrupt_ring_wptr);

	if ((rptr - wptr) % kfd->interrupt_ring_size ==
					kfd->device_info->ih_ring_entry_size) {
		/* This is very bad, the system is likely to hang. */
		dev_err_ratelimited(kfd_chardev(),
			"Interrupt ring overflow, dropping interrupt.\n");
		return false;
	}

	memcpy(kfd->interrupt_ring + wptr, ih_ring_entry,
			kfd->device_info->ih_ring_entry_size);

	wptr = (wptr + kfd->device_info->ih_ring_entry_size) %
			kfd->interrupt_ring_size;
	smp_wmb(); /* Ensure memcpy'd data is visible before wptr update. */
	atomic_set(&kfd->interrupt_ring_wptr, wptr);

	return true;
}

/*
 * This assumes that it can't be called concurrently with itself
 * but only with enqueue_ih_ring_entry.
 */
static bool dequeue_ih_ring_entry(struct kfd_dev *kfd, void *ih_ring_entry)
{
	/*
	 * Assume that wait queues have an implicit barrier, i.e. anything that
	 * happened in the ISR before it queued work is visible.
	 */

	unsigned int wptr = atomic_read(&kfd->interrupt_ring_wptr);
	unsigned int rptr = atomic_read(&kfd->interrupt_ring_rptr);

	if (rptr == wptr)
		return false;

	memcpy(ih_ring_entry, kfd->interrupt_ring + rptr,
			kfd->device_info->ih_ring_entry_size);

	rptr = (rptr + kfd->device_info->ih_ring_entry_size) %
			kfd->interrupt_ring_size;

	/*
	 * Ensure the rptr write update is not visible until
	 * memcpy has finished reading.
	 */
	smp_mb();
	atomic_set(&kfd->interrupt_ring_rptr, rptr);

	return true;
}

static void interrupt_wq(struct work_struct *work)
{
	struct kfd_dev *dev = container_of(work, struct kfd_dev,
						interrupt_work);

	uint32_t ih_ring_entry[DIV_ROUND_UP(
				dev->device_info->ih_ring_entry_size,
				sizeof(uint32_t))];

	while (dequeue_ih_ring_entry(dev, ih_ring_entry))
		dev->device_info->event_interrupt_class->interrupt_wq(dev,
								ih_ring_entry);
}

bool interrupt_is_wanted(struct kfd_dev *dev, const uint32_t *ih_ring_entry)
{
	/* integer and bitwise OR so there is no boolean short-circuiting */
	unsigned wanted = 0;

	wanted |= dev->device_info->event_interrupt_class->interrupt_isr(dev,
								ih_ring_entry);

	return wanted != 0;
}
Commit	Line	Data
2249d558 AL	1	/*
	2	* Copyright 2014 Advanced Micro Devices, Inc.
	3	*
	4	* Permission is hereby granted, free of charge, to any person obtaining a
	5	* copy of this software and associated documentation files (the "Software"),
	6	* to deal in the Software without restriction, including without limitation
	7	* the rights to use, copy, modify, merge, publish, distribute, sublicense,
	8	* and/or sell copies of the Software, and to permit persons to whom the
	9	* Software is furnished to do so, subject to the following conditions:
	10	*
	11	* The above copyright notice and this permission notice shall be included in
	12	* all copies or substantial portions of the Software.
	13	*
	14	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
	15	* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
	16	* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
	17	* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
	18	* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
	19	* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
	20	* OTHER DEALINGS IN THE SOFTWARE.
	21	*/
	22
	23	/*
	24	* KFD Interrupts.
	25	*
	26	* AMD GPUs deliver interrupts by pushing an interrupt description onto the
	27	* interrupt ring and then sending an interrupt. KGD receives the interrupt
	28	* in ISR and sends us a pointer to each new entry on the interrupt ring.
	29	*
	30	* We generally can't process interrupt-signaled events from ISR, so we call
	31	* out to each interrupt client module (currently only the scheduler) to ask if
	32	* each interrupt is interesting. If they return true, then it requires further
	33	* processing so we copy it to an internal interrupt ring and call each
	34	* interrupt client again from a work-queue.
	35	*
	36	* There's no acknowledgment for the interrupts we use. The hardware simply
	37	* queues a new interrupt each time without waiting.
	38	*
	39	* The fixed-size internal queue means that it's possible for us to lose
	40	* interrupts because we have no back-pressure to the hardware.
	41	*/
	42
	43	#include <linux/slab.h>
	44	#include <linux/device.h>
	45	#include "kfd_priv.h"
	46
	47	#define KFD_INTERRUPT_RING_SIZE 1024
	48
	49	static void interrupt_wq(struct work_struct *);
	50
	51	int kfd_interrupt_init(struct kfd_dev *kfd)
	52	{
	53	void *interrupt_ring = kmalloc_array(KFD_INTERRUPT_RING_SIZE,
	54	kfd->device_info->ih_ring_entry_size,
	55	GFP_KERNEL);
	56	if (!interrupt_ring)
	57	return -ENOMEM;
	58
	59	kfd->interrupt_ring = interrupt_ring;
	60	kfd->interrupt_ring_size =
	61	KFD_INTERRUPT_RING_SIZE * kfd->device_info->ih_ring_entry_size;
	62	atomic_set(&kfd->interrupt_ring_wptr, 0);
	63	atomic_set(&kfd->interrupt_ring_rptr, 0);
	64
65	spin_lock_init(&kfd->interrupt_lock);
66
67	INIT_WORK(&kfd->interrupt_work, interrupt_wq);
68
69	kfd->interrupts_active = true;
70
71	/*
72	* After this function returns, the interrupt will be enabled. This
73	* barrier ensures that the interrupt running on a different processor
74	* sees all the above writes.
75	*/
76	smp_wmb();
77
78	return 0;
79	}
80
81	void kfd_interrupt_exit(struct kfd_dev *kfd)
82	{
83	/*
84	* Stop the interrupt handler from writing to the ring and scheduling
85	* workqueue items. The spinlock ensures that any interrupt running
86	* after we have unlocked sees interrupts_active = false.
87	*/
88	unsigned long flags;
89
90	spin_lock_irqsave(&kfd->interrupt_lock, flags);
91	kfd->interrupts_active = false;
92	spin_unlock_irqrestore(&kfd->interrupt_lock, flags);
93
94	/*
95	* Flush_scheduled_work ensures that there are no outstanding
96	* work-queue items that will access interrupt_ring. New work items
97	* can't be created because we stopped interrupt handling above.
98	*/
99	flush_scheduled_work();
100
101	kfree(kfd->interrupt_ring);
102	}
103
104	/*
105	* This assumes that it can't be called concurrently with itself
106	* but only with dequeue_ih_ring_entry.
107	*/
108	bool enqueue_ih_ring_entry(struct kfd_dev kfd, const void ih_ring_entry)
109	{
110	unsigned int rptr = atomic_read(&kfd->interrupt_ring_rptr);
111	unsigned int wptr = atomic_read(&kfd->interrupt_ring_wptr);
112
113	if ((rptr - wptr) % kfd->interrupt_ring_size ==
114	kfd->device_info->ih_ring_entry_size) {
115	/* This is very bad, the system is likely to hang. */
116	dev_err_ratelimited(kfd_chardev(),
117	"Interrupt ring overflow, dropping interrupt.\n");
118	return false;
119	}
120
121	memcpy(kfd->interrupt_ring + wptr, ih_ring_entry,
122	kfd->device_info->ih_ring_entry_size);
123
124	wptr = (wptr + kfd->device_info->ih_ring_entry_size) %
125	kfd->interrupt_ring_size;
126	smp_wmb(); /* Ensure memcpy'd data is visible before wptr update. */
127	atomic_set(&kfd->interrupt_ring_wptr, wptr);
128
129	return true;
130	}
131
132	/*
133	* This assumes that it can't be called concurrently with itself
134	* but only with enqueue_ih_ring_entry.
135	*/
136	static bool dequeue_ih_ring_entry(struct kfd_dev kfd, void ih_ring_entry)
137	{
138	/*
139	* Assume that wait queues have an implicit barrier, i.e. anything that
140	* happened in the ISR before it queued work is visible.
141	*/
142
143	unsigned int wptr = atomic_read(&kfd->interrupt_ring_wptr);
144	unsigned int rptr = atomic_read(&kfd->interrupt_ring_rptr);
145
146	if (rptr == wptr)
147	return false;
148
149	memcpy(ih_ring_entry, kfd->interrupt_ring + rptr,
150	kfd->device_info->ih_ring_entry_size);
151
152	rptr = (rptr + kfd->device_info->ih_ring_entry_size) %
153	kfd->interrupt_ring_size;
154
155	/*
156	* Ensure the rptr write update is not visible until
157	* memcpy has finished reading.
158	*/
159	smp_mb();
160	atomic_set(&kfd->interrupt_ring_rptr, rptr);
161
162	return true;
163	}
164
165	static void interrupt_wq(struct work_struct *work)
166	{
167	struct kfd_dev *dev = container_of(work, struct kfd_dev,
168	interrupt_work);
169
170	uint32_t ih_ring_entry[DIV_ROUND_UP(
171	dev->device_info->ih_ring_entry_size,
172	sizeof(uint32_t))];
173
174	while (dequeue_ih_ring_entry(dev, ih_ring_entry))
f3a39818 AL	175	dev->device_info->event_interrupt_class->interrupt_wq(dev,
f3a39818 AL	176	ih_ring_entry);
2249d558 AL	177	}
	178
	179	bool interrupt_is_wanted(struct kfd_dev dev, const uint32_t ih_ring_entry)
	180	{
f3a39818 AL	181	/* integer and bitwise OR so there is no boolean short-circuiting */
	182	unsigned wanted = 0;
	183
	184	wanted \|= dev->device_info->event_interrupt_class->interrupt_isr(dev,
	185	ih_ring_entry);
	186
	187	return wanted != 0;
2249d558	188	}