@@ -12,6 +12,6 @@ amdkfd-y := kfd_module.o kfd_device.o kfd_chardev.o kfd_topology.o \
kfd_kernel_queue_vi.o kfd_packet_manager.o \
kfd_process_queue_manager.o kfd_device_queue_manager.o \
kfd_device_queue_manager_cik.o kfd_device_queue_manager_vi.o \
- kfd_interrupt.o
+ kfd_interrupt.o kfd_events.o cik_event_interrupt.o
obj-$(CONFIG_HSA_AMD) += amdkfd.o
new file mode 100644
@@ -0,0 +1,57 @@
+/*
+ * 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.
+ */
+
+#include "kfd_priv.h"
+#include "kfd_events.h"
+#include "cik_int.h"
+
+static bool cik_event_interrupt_isr(struct kfd_dev *dev,
+ const uint32_t *ih_ring_entry)
+{
+ const struct cik_ih_ring_entry *ihre =
+ (const struct cik_ih_ring_entry *)ih_ring_entry;
+
+ /* Do not process in ISR, just request it to be forwarded to WQ. */
+ return (ihre->pasid != 0) &&
+ (ihre->source_id == CIK_INTSRC_CP_END_OF_PIPE ||
+ ihre->source_id == CIK_INTSRC_SQ_INTERRUPT_MSG);
+}
+
+static void cik_event_interrupt_wq(struct kfd_dev *dev,
+ const uint32_t *ih_ring_entry)
+{
+ const struct cik_ih_ring_entry *ihre =
+ (const struct cik_ih_ring_entry *)ih_ring_entry;
+
+ if (ihre->pasid == 0)
+ return;
+
+ if (ihre->source_id == CIK_INTSRC_CP_END_OF_PIPE)
+ kfd_signal_event_interrupt(ihre->pasid, 0, 0);
+ else if (ihre->source_id == CIK_INTSRC_SQ_INTERRUPT_MSG)
+ kfd_signal_event_interrupt(ihre->pasid, ihre->data & 0xFF, 8);
+}
+
+const struct kfd_event_interrupt_class event_interrupt_class_cik = {
+ .interrupt_isr = cik_event_interrupt_isr,
+ .interrupt_wq = cik_event_interrupt_wq,
+};
new file mode 100644
@@ -0,0 +1,54 @@
+/*
+ * 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.
+ */
+
+#ifndef HSA_RADEON_CIK_INT_H_INCLUDED
+#define HSA_RADEON_CIK_INT_H_INCLUDED
+
+#include <linux/types.h>
+
+struct cik_ih_ring_entry {
+ uint32_t source_id:8;
+ uint32_t reserved1:8;
+ uint32_t reserved2:16;
+
+ uint32_t data:28;
+ uint32_t reserved3:4;
+
+ /*
+ * pipeid, meid and unused3 are officially called RINGID,
+ * but for our purposes, they always decode into pipe and ME.
+ */
+ uint32_t pipeid:2;
+ uint32_t meid:2;
+ uint32_t reserved4:4;
+ uint32_t vmid:8;
+ uint32_t pasid:16;
+
+ uint32_t reserved5;
+};
+
+#define CIK_INTSRC_DEQUEUE_COMPLETE 0xC6
+#define CIK_INTSRC_CP_END_OF_PIPE 0xB5
+#define CIK_INTSRC_SQ_INTERRUPT_MSG 0xEF
+
+#endif
+
@@ -289,8 +289,10 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p,
args->queue_id = queue_id;
+
/* Return gpu_id as doorbell offset for mmap usage */
- args->doorbell_offset = args->gpu_id << PAGE_SHIFT;
+ args->doorbell_offset = (KFD_MMAP_DOORBELL_MASK | args->gpu_id);
+ args->doorbell_offset <<= PAGE_SHIFT;
mutex_unlock(&p->mutex);
@@ -684,5 +686,15 @@ static int kfd_mmap(struct file *filp, struct vm_area_struct *vma)
if (IS_ERR(process))
return PTR_ERR(process);
- return kfd_doorbell_mmap(process, vma);
+ if ((vma->vm_pgoff & KFD_MMAP_DOORBELL_MASK) ==
+ KFD_MMAP_DOORBELL_MASK) {
+ vma->vm_pgoff = vma->vm_pgoff ^ KFD_MMAP_DOORBELL_MASK;
+ return kfd_doorbell_mmap(process, vma);
+ } else if ((vma->vm_pgoff & KFD_MMAP_EVENTS_MASK) ==
+ KFD_MMAP_EVENTS_MASK) {
+ vma->vm_pgoff = vma->vm_pgoff ^ KFD_MMAP_EVENTS_MASK;
+ return radeon_kfd_event_mmap(process, vma);
+ }
+
+ return -EFAULT;
}
@@ -34,6 +34,7 @@ static const struct kfd_device_info kaveri_device_info = {
.asic_family = CHIP_KAVERI,
.max_pasid_bits = 16,
.ih_ring_entry_size = 4 * sizeof(uint32_t),
+ .event_interrupt_class = &event_interrupt_class_cik,
.mqd_size_aligned = MQD_SIZE_ALIGNED
};
new file mode 100644
@@ -0,0 +1,775 @@
+/*
+ * 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.
+ */
+
+#include <linux/mm_types.h>
+#include <linux/slab.h>
+#include <linux/types.h>
+#include <linux/sched.h>
+#include <linux/uaccess.h>
+#include <linux/mm.h>
+#include <linux/mman.h>
+#include <linux/memory.h>
+#include "kfd_priv.h"
+#include "kfd_events.h"
+
+#define SIGNAL_EVENT_LIMIT 256
+
+/*
+ * A task can only be on a single wait_queue at a time, but we need to support
+ * waiting on multiple events (any/all).
+ * Instead of each event simply having a wait_queue with sleeping tasks, it
+ * has a singly-linked list of tasks.
+ * A thread that wants to sleep creates an array of these, one for each event
+ * and adds one to each event's waiter chain.
+ */
+struct kfd_event_waiter {
+ struct list_head waiters;
+ struct task_struct *sleeping_task;
+
+ /* Transitions to true when the event this belongs to is signaled. */
+ bool activated;
+};
+
+/*
+ * Over-complicated pooled allocator for event notification slots.
+ *
+ * Each signal event needs a 64-bit signal slot where the signaler will write
+ * a 1 before sending an interrupt.l (This is needed because some interrupts
+ * do not contain enough spare data bits to identify an event.)
+ * We get whole pages from vmalloc and map them to the process VA.
+ * Individual signal events are then allocated a slot in a page.
+ */
+
+struct signal_page {
+ struct list_head event_pages; /* kfd_process.signal_event_pages */
+ uint64_t *kernel_address;
+ uint64_t __user *user_address;
+ uint32_t page_index; /* Index into the mmap aperture. */
+ unsigned int free_slots;
+ unsigned long used_slot_bitmap[0];
+};
+
+#define SLOTS_PER_PAGE (PAGE_SIZE / sizeof(uint64_t))
+#define SLOT_BITMAP_SIZE BITS_TO_LONGS(SLOTS_PER_PAGE)
+#define BITS_PER_PAGE (ilog2(SLOTS_PER_PAGE)+1)
+#define SIGNAL_PAGE_SIZE (sizeof(struct signal_page) + \
+ SLOT_BITMAP_SIZE * sizeof(long))
+
+/*
+ * For signal events, the event ID is used as the interrupt user data.
+ * For SQ s_sendmsg interrupts, this is limited to 8 bits. We will
+ * limit to 256 signal events and simply use all the available bits.
+ */
+
+#define INTERRUPT_DATA_BITS 8
+#define SIGNAL_EVENT_ID_SLOT_SHIFT 0
+
+static uint64_t *page_slots(struct signal_page *page)
+{
+ return page->kernel_address;
+}
+
+static bool allocate_free_slot(struct kfd_process *process,
+ struct signal_page **out_page,
+ unsigned int *out_slot_index)
+{
+ struct signal_page *page;
+
+ list_for_each_entry(page, &process->signal_event_pages, event_pages) {
+ if (page->free_slots > 0) {
+ unsigned int slot =
+ find_first_zero_bit(page->used_slot_bitmap,
+ SLOTS_PER_PAGE);
+
+ __set_bit(slot, page->used_slot_bitmap);
+ page->free_slots--;
+
+ page_slots(page)[slot] = UNSIGNALED_EVENT_SLOT;
+
+ *out_page = page;
+ *out_slot_index = slot;
+ return true;
+ }
+ }
+
+ return false;
+}
+
+#define list_tail_entry(head, type, member) \
+ list_entry((head)->prev, type, member)
+
+static bool allocate_signal_page(struct file *devkfd, struct kfd_process *p)
+{
+ void *backing_store;
+ struct signal_page *page;
+
+ page = kzalloc(SIGNAL_PAGE_SIZE, GFP_KERNEL);
+ if (!page)
+ goto fail_alloc_signal_page;
+
+ page->free_slots = SLOTS_PER_PAGE;
+
+ backing_store = (void *)get_zeroed_page(GFP_KERNEL);
+ if (!backing_store)
+ goto fail_alloc_signal_store;
+
+ /* prevent user-mode info leaks */
+ memset(backing_store, (uint8_t) UNSIGNALED_EVENT_SLOT, PAGE_SIZE);
+ page->kernel_address = backing_store;
+
+ if (list_empty(&p->signal_event_pages))
+ page->page_index = 0;
+ else
+ page->page_index = list_tail_entry(&p->signal_event_pages,
+ struct signal_page,
+ event_pages)->page_index + 1;
+
+ list_add(&page->event_pages, &p->signal_event_pages);
+
+ return true;
+
+fail_alloc_signal_store:
+ kfree(page);
+fail_alloc_signal_page:
+ return false;
+}
+
+static bool allocate_event_notification_slot(struct file *devkfd,
+ struct kfd_process *p,
+ struct signal_page **page,
+ unsigned int *signal_slot_index)
+{
+ if (!allocate_free_slot(p, page, signal_slot_index)) {
+ allocate_signal_page(devkfd, p);
+ return allocate_free_slot(p, page, signal_slot_index);
+ }
+
+ return true;
+}
+
+/* Assumes that the process's event_mutex is locked. */
+static void release_event_notification_slot(struct signal_page *page,
+ size_t slot_index)
+{
+ __clear_bit(slot_index, page->used_slot_bitmap);
+ page->free_slots++;
+
+ /* We don't free signal pages, they are retained by the process
+ * and reused until it exits. */
+}
+
+static struct signal_page *lookup_signal_page_by_index(struct kfd_process *p,
+ unsigned int page_index)
+{
+ struct signal_page *page;
+
+ /*
+ * This is safe because we don't delete signal pages until the
+ * process exits.
+ */
+ list_for_each_entry(page, &p->signal_event_pages, event_pages)
+ if (page->page_index == page_index)
+ return page;
+
+ return NULL;
+}
+
+/*
+ * Assumes that p->event_mutex is held and of course that p is not going
+ * away (current or locked).
+ */
+static struct kfd_event *lookup_event_by_id(struct kfd_process *p, uint32_t id)
+{
+ struct kfd_event *ev;
+
+ hash_for_each_possible(p->events, ev, events, id)
+ if (ev->event_id == id)
+ return ev;
+
+ return NULL;
+}
+
+static u32 make_signal_event_id(struct signal_page *page,
+ unsigned int signal_slot_index)
+{
+ return page->page_index |
+ (signal_slot_index << SIGNAL_EVENT_ID_SLOT_SHIFT);
+}
+
+/*
+ * Produce a kfd event id for a nonsignal event.
+ * These are arbitrary numbers, so we do a sequential search through
+ * the hash table for an unused number.
+ */
+static u32 make_nonsignal_event_id(struct kfd_process *p)
+{
+ u32 id;
+
+ for (id = p->next_nonsignal_event_id;
+ id < KFD_LAST_NONSIGNAL_EVENT_ID &&
+ lookup_event_by_id(p, id) != NULL;
+ id++)
+ ;
+
+ if (id < KFD_LAST_NONSIGNAL_EVENT_ID) {
+
+ /*
+ * What if id == LAST_NONSIGNAL_EVENT_ID - 1?
+ * Then next_nonsignal_event_id = LAST_NONSIGNAL_EVENT_ID so
+ * the first loop fails immediately and we proceed with the
+ * wraparound loop below.
+ */
+ p->next_nonsignal_event_id = id + 1;
+
+ return id;
+ }
+
+ for (id = KFD_FIRST_NONSIGNAL_EVENT_ID;
+ id < KFD_LAST_NONSIGNAL_EVENT_ID &&
+ lookup_event_by_id(p, id) != NULL;
+ id++)
+ ;
+
+
+ if (id < KFD_LAST_NONSIGNAL_EVENT_ID) {
+ p->next_nonsignal_event_id = id + 1;
+ return id;
+ }
+
+ p->next_nonsignal_event_id = KFD_FIRST_NONSIGNAL_EVENT_ID;
+ return 0;
+}
+
+static struct kfd_event *lookup_event_by_page_slot(struct kfd_process *p,
+ struct signal_page *page,
+ unsigned int signal_slot)
+{
+ return lookup_event_by_id(p, make_signal_event_id(page, signal_slot));
+}
+
+static int create_signal_event(struct file *devkfd,
+ struct kfd_process *p,
+ struct kfd_event *ev)
+{
+ if (p->signal_event_count == SIGNAL_EVENT_LIMIT)
+ return -ENOMEM;
+
+ if (!allocate_event_notification_slot(devkfd, p, &ev->signal_page,
+ &ev->signal_slot_index))
+ return -ENOMEM;
+
+ p->signal_event_count++;
+
+ ev->user_signal_address =
+ &ev->signal_page->user_address[ev->signal_slot_index];
+
+ ev->event_id = make_signal_event_id(ev->signal_page,
+ ev->signal_slot_index);
+
+ return 0;
+}
+
+/*
+ * No non-signal events are supported yet.
+ * We create them as events that never signal.
+ * Set event calls from user-mode are failed.
+ */
+static int create_other_event(struct kfd_process *p, struct kfd_event *ev)
+{
+ ev->event_id = make_nonsignal_event_id(p);
+ if (ev->event_id == 0)
+ return -ENOMEM;
+
+ return 0;
+}
+
+void kfd_event_init_process(struct kfd_process *p)
+{
+ mutex_init(&p->event_mutex);
+ hash_init(p->events);
+ INIT_LIST_HEAD(&p->signal_event_pages);
+ p->next_nonsignal_event_id = KFD_FIRST_NONSIGNAL_EVENT_ID;
+ p->signal_event_count = 0;
+}
+
+static void destroy_event(struct kfd_process *p, struct kfd_event *ev)
+{
+ if (ev->signal_page != NULL) {
+ release_event_notification_slot(ev->signal_page,
+ ev->signal_slot_index);
+ p->signal_event_count--;
+ }
+
+ /*
+ * Abandon the list of waiters. Individual waiting threads will
+ * clean up their own data.
+ */
+ list_del(&ev->waiters);
+
+ hash_del(&ev->events);
+ kfree(ev);
+}
+
+static void destroy_events(struct kfd_process *p)
+{
+ struct kfd_event *ev;
+ struct hlist_node *tmp;
+ unsigned int hash_bkt;
+
+ hash_for_each_safe(p->events, hash_bkt, tmp, ev, events)
+ destroy_event(p, ev);
+}
+
+/*
+ * We assume that the process is being destroyed and there is no need to
+ * unmap the pages or keep bookkeeping data in order.
+ */
+static void shutdown_signal_pages(struct kfd_process *p)
+{
+ struct signal_page *page, *tmp;
+
+ list_for_each_entry_safe(page, tmp, &p->signal_event_pages,
+ event_pages) {
+ free_page((unsigned long)page->kernel_address);
+ kfree(page);
+ }
+}
+
+void kfd_event_free_process(struct kfd_process *p)
+{
+ destroy_events(p);
+ shutdown_signal_pages(p);
+}
+
+static bool event_can_be_gpu_signaled(const struct kfd_event *ev)
+{
+ return ev->type == KFD_EVENT_TYPE_SIGNAL ||
+ ev->type == KFD_EVENT_TYPE_DEBUG;
+}
+
+static bool event_can_be_cpu_signaled(const struct kfd_event *ev)
+{
+ return ev->type == KFD_EVENT_TYPE_SIGNAL;
+}
+
+int kfd_event_create(struct file *devkfd, struct kfd_process *p,
+ uint32_t event_type, bool auto_reset, uint32_t node_id,
+ uint32_t *event_id, uint32_t *event_trigger_data,
+ uint64_t *event_page_offset, uint32_t *event_slot_index)
+{
+ int ret = 0;
+ struct kfd_event *ev = kzalloc(sizeof(*ev), GFP_KERNEL);
+
+ if (!ev)
+ return -ENOMEM;
+
+ ev->type = event_type;
+ ev->auto_reset = auto_reset;
+ ev->signaled = false;
+
+ INIT_LIST_HEAD(&ev->waiters);
+
+ *event_page_offset = 0;
+
+ mutex_lock(&p->event_mutex);
+
+ switch (event_type) {
+ case KFD_EVENT_TYPE_SIGNAL:
+ case KFD_EVENT_TYPE_DEBUG:
+ ret = create_signal_event(devkfd, p, ev);
+ if (ret == 0) {
+ *event_page_offset = (ev->signal_page->page_index |
+ KFD_MMAP_EVENTS_MASK);
+ *event_page_offset <<= PAGE_SHIFT;
+ *event_slot_index = ev->signal_slot_index;
+ } else
+ pr_err("amdkfd: error signal page\n");
+ break;
+ default:
+ ret = create_other_event(p, ev);
+ break;
+ }
+
+ if (!ret) {
+ hash_add(p->events, &ev->events, ev->event_id);
+
+ *event_id = ev->event_id;
+ *event_trigger_data = ev->event_id;
+ } else {
+ kfree(ev);
+ }
+
+ mutex_unlock(&p->event_mutex);
+
+ return ret;
+}
+
+/* Assumes that p is current. */
+int kfd_event_destroy(struct kfd_process *p, uint32_t event_id)
+{
+ struct kfd_event *ev;
+ int ret = 0;
+
+ mutex_lock(&p->event_mutex);
+
+ ev = lookup_event_by_id(p, event_id);
+
+ if (ev)
+ destroy_event(p, ev);
+ else
+ ret = -EINVAL;
+
+ mutex_unlock(&p->event_mutex);
+ return ret;
+}
+
+static void set_event(struct kfd_event *ev)
+{
+ struct kfd_event_waiter *waiter;
+ struct kfd_event_waiter *next;
+
+ /* Auto reset if the list is non-empty and we're waking someone. */
+ ev->signaled = !ev->auto_reset || list_empty(&ev->waiters);
+
+ list_for_each_entry_safe(waiter, next, &ev->waiters, waiters) {
+ waiter->activated = true;
+
+ /* _init because free_waiters will call list_del */
+ list_del_init(&waiter->waiters);
+
+ wake_up_process(waiter->sleeping_task);
+ }
+}
+
+/* Assumes that p is current. */
+int kfd_set_event(struct kfd_process *p, uint32_t event_id)
+{
+ int ret = 0;
+ struct kfd_event *ev;
+
+ mutex_lock(&p->event_mutex);
+
+ ev = lookup_event_by_id(p, event_id);
+
+ if (ev && event_can_be_cpu_signaled(ev))
+ set_event(ev);
+ else
+ ret = -EINVAL;
+
+ mutex_unlock(&p->event_mutex);
+ return ret;
+}
+
+static void reset_event(struct kfd_event *ev)
+{
+ ev->signaled = false;
+}
+
+/* Assumes that p is current. */
+int kfd_reset_event(struct kfd_process *p, uint32_t event_id)
+{
+ int ret = 0;
+ struct kfd_event *ev;
+
+ mutex_lock(&p->event_mutex);
+
+ ev = lookup_event_by_id(p, event_id);
+
+ if (ev && event_can_be_cpu_signaled(ev))
+ reset_event(ev);
+ else
+ ret = -EINVAL;
+
+ mutex_unlock(&p->event_mutex);
+ return ret;
+
+}
+
+static void acknowledge_signal(struct kfd_process *p, struct kfd_event *ev)
+{
+ page_slots(ev->signal_page)[ev->signal_slot_index] =
+ UNSIGNALED_EVENT_SLOT;
+}
+
+static bool is_slot_signaled(struct signal_page *page, unsigned int index)
+{
+ return page_slots(page)[index] != UNSIGNALED_EVENT_SLOT;
+}
+
+static void set_event_from_interrupt(struct kfd_process *p,
+ struct kfd_event *ev)
+{
+ if (ev && event_can_be_gpu_signaled(ev)) {
+ acknowledge_signal(p, ev);
+ set_event(ev);
+ }
+}
+
+void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id,
+ uint32_t valid_id_bits)
+{
+ struct kfd_event *ev;
+
+ /*
+ * Because we are called from arbitrary context (workqueue) as opposed
+ * to process context, kfd_process could attempt to exit while we are
+ * running so the lookup function returns a locked process.
+ */
+ struct kfd_process *p = kfd_lookup_process_by_pasid(pasid);
+
+ if (!p)
+ return; /* Presumably process exited. */
+
+ mutex_lock(&p->event_mutex);
+
+ if (valid_id_bits >= INTERRUPT_DATA_BITS) {
+ /* Partial ID is a full ID. */
+ ev = lookup_event_by_id(p, partial_id);
+ set_event_from_interrupt(p, ev);
+ } else {
+ /*
+ * Partial ID is in fact partial. For now we completely
+ * ignore it, but we could use any bits we did receive to
+ * search faster.
+ */
+ struct signal_page *page;
+ unsigned i;
+
+ list_for_each_entry(page, &p->signal_event_pages, event_pages)
+ for (i = 0; i < SLOTS_PER_PAGE; i++)
+ if (is_slot_signaled(page, i)) {
+ ev = lookup_event_by_page_slot(p,
+ page, i);
+ set_event_from_interrupt(p, ev);
+ }
+ }
+
+ mutex_unlock(&p->event_mutex);
+ mutex_unlock(&p->mutex);
+}
+
+static struct kfd_event_waiter *alloc_event_waiters(uint32_t num_events)
+{
+ struct kfd_event_waiter *event_waiters;
+ uint32_t i;
+
+ event_waiters = kmalloc_array(num_events,
+ sizeof(struct kfd_event_waiter),
+ GFP_KERNEL);
+
+ for (i = 0; (event_waiters) && (i < num_events) ; i++) {
+ INIT_LIST_HEAD(&event_waiters[i].waiters);
+ event_waiters[i].sleeping_task = current;
+ event_waiters[i].activated = false;
+ }
+
+ return event_waiters;
+}
+
+static int init_event_waiter(struct kfd_process *p,
+ struct kfd_event_waiter *waiter,
+ uint32_t event_id)
+{
+ struct kfd_event *ev = lookup_event_by_id(p, event_id);
+
+ if (!ev)
+ return -EINVAL;
+
+ waiter->activated = ev->signaled;
+ ev->signaled = ev->signaled && !ev->auto_reset;
+
+ list_add(&waiter->waiters, &ev->waiters);
+
+ return 0;
+}
+
+static bool test_event_condition(bool all, uint32_t num_events,
+ struct kfd_event_waiter *event_waiters)
+{
+ uint32_t i;
+ uint32_t activated_count = 0;
+
+ for (i = 0; i < num_events; i++) {
+ if (event_waiters[i].activated) {
+ if (!all)
+ return true;
+
+ activated_count++;
+ }
+ }
+
+ return activated_count == num_events;
+}
+
+static long user_timeout_to_jiffies(uint32_t user_timeout_ms)
+{
+ if (user_timeout_ms == KFD_EVENT_TIMEOUT_IMMEDIATE)
+ return 0;
+
+ if (user_timeout_ms == KFD_EVENT_TIMEOUT_INFINITE)
+ return MAX_SCHEDULE_TIMEOUT;
+
+ /*
+ * msecs_to_jiffies interprets all values above 2^31-1 as infinite,
+ * but we consider them finite.
+ * This hack is wrong, but nobody is likely to notice.
+ */
+ user_timeout_ms = min_t(uint32_t, user_timeout_ms, 0x7FFFFFFF);
+
+ return msecs_to_jiffies(user_timeout_ms) + 1;
+}
+
+static void free_waiters(uint32_t num_events, struct kfd_event_waiter *waiters)
+{
+ uint32_t i;
+
+ for (i = 0; i < num_events; i++)
+ list_del(&waiters[i].waiters);
+
+ kfree(waiters);
+}
+
+int kfd_wait_on_events(struct kfd_process *p,
+ uint32_t num_events, const uint32_t __user *event_ids,
+ bool all, uint32_t user_timeout_ms,
+ enum kfd_event_wait_result *wait_result)
+{
+ uint32_t i;
+ int ret = 0;
+ struct kfd_event_waiter *event_waiters = NULL;
+ long timeout = user_timeout_to_jiffies(user_timeout_ms);
+
+ mutex_lock(&p->event_mutex);
+
+ event_waiters = alloc_event_waiters(num_events);
+ if (!event_waiters) {
+ ret = -ENOMEM;
+ goto fail;
+ }
+
+ for (i = 0; i < num_events; i++) {
+ uint32_t event_id;
+
+ ret = get_user(event_id, &event_ids[i]);
+ if (ret)
+ goto fail;
+
+ ret = init_event_waiter(p, &event_waiters[i], event_id);
+ if (ret)
+ goto fail;
+ }
+
+ mutex_unlock(&p->event_mutex);
+
+ while (true) {
+ if (fatal_signal_pending(current)) {
+ ret = -EINTR;
+ break;
+ }
+
+ if (signal_pending(current)) {
+ /*
+ * This is wrong when a nonzero, non-infinite timeout
+ * is specified. We need to use
+ * ERESTARTSYS_RESTARTBLOCK, but struct restart_block
+ * contains a union with data for each user and it's
+ * in generic kernel code that I don't want to
+ * touch yet.
+ */
+ ret = -ERESTARTSYS;
+ break;
+ }
+
+ if (test_event_condition(all, num_events, event_waiters)) {
+ *wait_result = KFD_WAIT_COMPLETE;
+ break;
+ }
+
+ if (timeout <= 0) {
+ *wait_result = KFD_WAIT_TIMEOUT;
+ break;
+ }
+
+ timeout = schedule_timeout_interruptible(timeout);
+ }
+ __set_current_state(TASK_RUNNING);
+
+ mutex_lock(&p->event_mutex);
+ free_waiters(num_events, event_waiters);
+ mutex_unlock(&p->event_mutex);
+
+ return ret;
+
+fail:
+ if (event_waiters)
+ free_waiters(num_events, event_waiters);
+
+ mutex_unlock(&p->event_mutex);
+
+ *wait_result = KFD_WAIT_ERROR;
+
+ return ret;
+}
+
+int radeon_kfd_event_mmap(struct kfd_process *p,
+ struct vm_area_struct *vma)
+{
+
+ unsigned int page_index;
+ unsigned long pfn;
+ struct signal_page *page;
+
+ if (vma->vm_end - vma->vm_start != PAGE_SIZE) {
+ pr_debug("start address(0x%lx) - end address(0x%lx) != len(0x%lx)\n",
+ vma->vm_end, vma->vm_start, PAGE_SIZE);
+ return -EINVAL;
+ }
+
+ page_index = vma->vm_pgoff;
+
+ page = lookup_signal_page_by_index(p, page_index);
+ if (!page) {
+ /* Probably KFD bug, but mmap is user-accessible. */
+ pr_debug("signal page could not be found for page_index %u\n",
+ page_index);
+ return -EINVAL;
+ }
+
+ pfn = __pa(page->kernel_address);
+ pfn >>= PAGE_SHIFT;
+
+ vma->vm_flags |= VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_NORESERVE
+ | VM_DONTDUMP | VM_PFNMAP;
+
+ pr_debug("kfd: mapping signal page\n");
+ pr_debug(" target user address == 0x%016llX\n",
+ (unsigned long long int) vma->vm_start);
+ pr_debug(" pfn == 0x%016lX\n", (unsigned long)pfn);
+ pr_debug(" vm_flags == 0x%08lX\n", vma->vm_flags);
+ pr_debug(" size == 0x%08lX\n", PAGE_SIZE);
+
+ page->user_address = (uint64_t __user *)vma->vm_start;
+
+ /* mapping the page to user process */
+ return remap_pfn_range(vma, vma->vm_start, pfn, PAGE_SIZE,
+ vma->vm_page_prot);
+}
new file mode 100644
@@ -0,0 +1,76 @@
+/*
+ * 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.
+ */
+
+#ifndef KFD_EVENTS_H_INCLUDED
+#define KFD_EVENTS_H_INCLUDED
+
+#include <linux/kernel.h>
+#include <linux/hashtable.h>
+#include <linux/types.h>
+#include <linux/list.h>
+#include "kfd_priv.h"
+
+#define KFD_EVENT_ID_NONSIGNAL_MASK 0x80000000U
+#define KFD_FIRST_NONSIGNAL_EVENT_ID KFD_EVENT_ID_NONSIGNAL_MASK
+#define KFD_LAST_NONSIGNAL_EVENT_ID UINT_MAX
+
+/*
+ * Written into kfd_signal_slot_t to indicate that the event is not signaled.
+ * Since the event protocol may need to write the event ID into memory, this
+ * must not be a valid event ID.
+ * For the sake of easy memset-ing, this must be a byte pattern.
+ */
+#define UNSIGNALED_EVENT_SLOT ((uint64_t)-1)
+
+struct kfd_event_waiter;
+struct signal_page;
+
+struct kfd_event {
+ /* All events in process, rooted at kfd_process.events. */
+ struct hlist_node events;
+
+ u32 event_id;
+
+ bool signaled;
+ bool auto_reset;
+
+ int type;
+
+ struct list_head waiters; /* List of kfd_event_waiter by waiters. */
+
+ /* Only for signal events. */
+ struct signal_page *signal_page;
+ unsigned int signal_slot_index;
+ uint64_t __user *user_signal_address;
+};
+
+#define KFD_EVENT_TIMEOUT_IMMEDIATE 0
+#define KFD_EVENT_TIMEOUT_INFINITE 0xFFFFFFFFu
+
+/* Matching HSA_EVENTTYPE */
+#define KFD_EVENT_TYPE_SIGNAL 0
+#define KFD_EVENT_TYPE_DEBUG 5
+
+extern void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id,
+ uint32_t valid_id_bits);
+
+#endif
@@ -172,10 +172,17 @@ static void interrupt_wq(struct work_struct *work)
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)
{
- return false;
+ /* 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;
}
@@ -35,6 +35,9 @@
#define KFD_SYSFS_FILE_MODE 0444
+#define KFD_MMAP_DOORBELL_MASK 0x8000000000000
+#define KFD_MMAP_EVENTS_MASK 0x4000000000000
+
/*
* When working with cp scheduler we should assign the HIQ manually or via
* the radeon driver to a fixed hqd slot, here are the fixed HIQ hqd slot
@@ -108,8 +111,16 @@ enum asic_family_type {
CHIP_CARRIZO
};
+struct kfd_event_interrupt_class {
+ bool (*interrupt_isr)(struct kfd_dev *dev,
+ const uint32_t *ih_ring_entry);
+ void (*interrupt_wq)(struct kfd_dev *dev,
+ const uint32_t *ih_ring_entry);
+};
+
struct kfd_device_info {
unsigned int asic_family;
+ const struct kfd_event_interrupt_class *event_interrupt_class;
unsigned int max_pasid_bits;
size_t ih_ring_entry_size;
uint8_t num_of_watch_points;
@@ -490,6 +501,15 @@ struct kfd_process {
/*Is the user space process 32 bit?*/
bool is_32bit_user_mode;
+
+ /* Event-related data */
+ struct mutex event_mutex;
+ /* All events in process hashed by ID, linked on kfd_event.events. */
+ DECLARE_HASHTABLE(events, 4);
+ struct list_head signal_event_pages; /* struct slot_page_header.
+ event_pages */
+ u32 next_nonsignal_event_id;
+ size_t signal_event_count;
};
/**
@@ -514,6 +534,7 @@ void kfd_process_create_wq(void);
void kfd_process_destroy_wq(void);
struct kfd_process *kfd_create_process(const struct task_struct *);
struct kfd_process *kfd_get_process(const struct task_struct *);
+struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid);
struct kfd_process_device *kfd_bind_process_to_device(struct kfd_dev *dev,
struct kfd_process *p);
@@ -659,4 +680,31 @@ uint64_t kfd_get_number_elems(struct kfd_dev *kfd);
phys_addr_t kfd_get_process_doorbells(struct kfd_dev *dev,
struct kfd_process *process);
+/* Events */
+extern const struct kfd_event_interrupt_class event_interrupt_class_cik;
+
+enum kfd_event_wait_result {
+ KFD_WAIT_COMPLETE,
+ KFD_WAIT_TIMEOUT,
+ KFD_WAIT_ERROR
+};
+
+void kfd_event_init_process(struct kfd_process *p);
+void kfd_event_free_process(struct kfd_process *p);
+int radeon_kfd_event_mmap(struct kfd_process *process,
+ struct vm_area_struct *vma);
+int kfd_wait_on_events(struct kfd_process *p,
+ uint32_t num_events, const uint32_t __user *event_ids,
+ bool all, uint32_t user_timeout_ms,
+ enum kfd_event_wait_result *wait_result);
+void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id,
+ uint32_t valid_id_bits);
+int kfd_set_event(struct kfd_process *p, uint32_t event_id);
+int kfd_reset_event(struct kfd_process *p, uint32_t event_id);
+int kfd_event_create(struct file *devkfd, struct kfd_process *p,
+ uint32_t event_type, bool auto_reset, uint32_t node_id,
+ uint32_t *event_id, uint32_t *event_trigger_data,
+ uint64_t *event_page_offset, uint32_t *event_slot_index);
+int kfd_event_destroy(struct kfd_process *p, uint32_t event_id);
+
#endif
@@ -178,6 +178,8 @@ static void kfd_process_wq_release(struct work_struct *work)
kfree(pdd);
}
+ kfd_event_free_process(p);
+
kfd_pasid_free(p->pasid);
mutex_unlock(&p->mutex);
@@ -289,6 +291,8 @@ static struct kfd_process *create_process(const struct task_struct *thread)
INIT_LIST_HEAD(&process->per_device_data);
+ kfd_event_init_process(process);
+
err = pqm_init(&process->pqm, process);
if (err != 0)
goto err_process_pqm_init;
@@ -431,3 +435,23 @@ bool kfd_has_process_device_data(struct kfd_process *p)
{
return !(list_empty(&p->per_device_data));
}
+
+/* This returns with process->mutex locked. */
+struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid)
+{
+ struct kfd_process *p;
+ unsigned int temp;
+
+ int idx = srcu_read_lock(&kfd_processes_srcu);
+
+ hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
+ if (p->pasid == pasid) {
+ mutex_lock(&p->mutex);
+ break;
+ }
+ }
+
+ srcu_read_unlock(&kfd_processes_srcu, idx);
+
+ return p;
+}