1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
|
From 3cd54dfc7075e64fd4ea5349775d90e429aa4a5b Mon Sep 17 00:00:00 2001
From: Felix Kuehling <Felix.Kuehling@amd.com>
Date: Mon, 25 Jul 2016 16:02:03 -0400
Subject: [PATCH 1161/4131] drm/amdkfd: Make kfd_process reference counted
This will be used to elliminate the use of the processe lock for
preventing process destruction. This will simplify lock
dependencies between KFD and KGD.
This also simplifies the process destruction in two ways:
* Don't allocate work struct dynamically
* Remove unnecessary hack that increments mm reference counter
Change-Id: I498e55856c5ab7dcc5a619a74a99db895fdaa7e3
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Conflicts:
drivers/gpu/drm/amd/amdkfd/kfd_process.c
---
drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 4 +++
drivers/gpu/drm/amd/amdkfd/kfd_process.c | 55 +++++++++++---------------------
2 files changed, 23 insertions(+), 36 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 2465685..1a9aab7 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -35,6 +35,7 @@
#include <linux/pid.h>
#include <linux/interval_tree.h>
#include <linux/seq_file.h>
+#include <linux/kref.h>
#include <kgd_kfd_interface.h>
#include "amd_rdma.h"
@@ -608,6 +609,9 @@ struct kfd_process {
struct mm_struct *mm;
+ struct kref ref;
+ struct work_struct release_work;
+
struct rw_semaphore lock;
/*
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index b0fb998..ec066d8 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -51,11 +51,6 @@ DEFINE_STATIC_SRCU(kfd_processes_srcu);
static struct workqueue_struct *kfd_process_wq;
-struct kfd_process_release_work {
- struct work_struct kfd_work;
- struct kfd_process *p;
-};
-
#define MIN_IDR_ID 1
#define MAX_IDR_ID 0 /*0 - for unlimited*/
@@ -415,23 +410,18 @@ int restore(struct kfd_dev *kfd)
* is not findable any more. We must assume that no other thread is
* using it any more, otherwise we couldn't safely free the process
* stucture in the end. */
-static void kfd_process_wq_release(struct work_struct *work)
+static void kfd_process_ref_release(struct kref *ref)
{
- struct kfd_process_release_work *my_work;
+ struct kfd_process *p = container_of(ref, struct kfd_process, ref);
struct kfd_process_device *pdd, *temp, *peer_pdd;
- struct kfd_process *p;
struct kfd_bo *buf_obj;
int id;
- my_work = (struct kfd_process_release_work *) work;
-
- p = my_work->p;
-
- pr_debug("Releasing process (pasid %d) in workqueue\n",
+ pr_debug("Releasing process (pasid %d)\n",
p->pasid);
list_for_each_entry(pdd, &p->per_device_data, per_device_list) {
- pr_debug("Releasing pdd (topology id %d) for process (pasid %d) in workqueue\n",
+ pr_debug("Releasing pdd (topology id %d) for process (pasid %d)\n",
pdd->dev->id, p->pasid);
if (pdd->dev->device_info->is_need_iommu_device) {
@@ -477,29 +467,24 @@ static void kfd_process_wq_release(struct work_struct *work)
kfd_pasid_free(p->pasid);
kfree(p);
+}
+
+static void kfd_process_wq_release(struct work_struct *work)
+{
+ struct kfd_process *p = container_of(work, struct kfd_process,
+ release_work);
- kfree((void *)work);
+ kref_put(&p->ref, kfd_process_ref_release);
}
static void kfd_process_destroy_delayed(struct rcu_head *rcu)
{
- struct kfd_process_release_work *work;
- struct kfd_process *p;
+ struct kfd_process *p = container_of(rcu, struct kfd_process, rcu);
BUG_ON(!kfd_process_wq);
- p = container_of(rcu, struct kfd_process, rcu);
- BUG_ON(atomic_read(&p->mm->mm_count) <= 0);
-
- mmdrop(p->mm);
-
- work = kmalloc(sizeof(struct kfd_process_release_work), GFP_ATOMIC);
-
- if (work) {
- INIT_WORK((struct work_struct *) work, kfd_process_wq_release);
- work->p = p;
- queue_work(kfd_process_wq, (struct work_struct *) work);
- }
+ INIT_WORK(&p->release_work, kfd_process_wq_release);
+ queue_work(kfd_process_wq, &p->release_work);
}
static void kfd_process_notifier_release(struct mmu_notifier *mn,
@@ -564,15 +549,12 @@ static void kfd_process_notifier_release(struct mmu_notifier *mn,
}
}
+ /* Indicate to other users that MM is no longer valid */
+ p->mm = NULL;
+
up_write(&p->lock);
- /*
- * Because we drop mm_count inside kfd_process_destroy_delayed
- * and because the mmu_notifier_unregister function also drop
- * mm_count we need to take an extra count here.
- */
- mmgrab(p->mm);
- mmu_notifier_unregister_no_release(&p->mmu_notifier, p->mm);
+ mmu_notifier_unregister_no_release(&p->mmu_notifier, mm);
mmu_notifier_call_srcu(&p->rcu, &kfd_process_destroy_delayed);
}
@@ -650,6 +632,7 @@ static struct kfd_process *create_process(const struct task_struct *thread)
if (process->pasid == 0)
goto err_alloc_pasid;
+ kref_init(&process->ref);
init_rwsem(&process->lock);
process->mm = thread->mm;
--
2.7.4
|
