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
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
|
From 6177b4c1237ec7719ddd93a59f9ac50dd8580d4d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig@amd.com>
Date: Wed, 7 Nov 2018 13:55:01 +0100
Subject: [PATCH 1617/2940] drm/amdgpu: use ring/hash for fault handling on
GMC9 v3
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Further testing showed that the idea with the chash doesn't work as expected.
Especially we can't predict when we can remove the entries from the hash again.
So replace the chash with a ring buffer/hash mix where entries in the container
age automatically based on their timestamp.
v2: use ring buffer / hash mix
v3: check the timeout to make sure all entries age
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> (v2)
Signed-off-by: Chaudhary Amit Kumar <Chaudharyamit.Kumar@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c | 55 +++++++++++++++++++++++
drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h | 34 ++++++++++++++
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 60 ++-----------------------
3 files changed, 92 insertions(+), 57 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
index 5a32a0d2ad31..250d9212cc38 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
@@ -240,3 +240,58 @@ void amdgpu_gmc_agp_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc)
dev_info(adev->dev, "AGP: %lluM 0x%016llX - 0x%016llX\n",
mc->agp_size >> 20, mc->agp_start, mc->agp_end);
}
+
+/**
+ * amdgpu_gmc_filter_faults - filter VM faults
+ *
+ * @adev: amdgpu device structure
+ * @addr: address of the VM fault
+ * @pasid: PASID of the process causing the fault
+ * @timestamp: timestamp of the fault
+ *
+ * Returns:
+ * True if the fault was filtered and should not be processed further.
+ * False if the fault is a new one and needs to be handled.
+ */
+bool amdgpu_gmc_filter_faults(struct amdgpu_device *adev, uint64_t addr,
+ uint16_t pasid, uint64_t timestamp)
+{
+ struct amdgpu_gmc *gmc = &adev->gmc;
+
+ uint64_t stamp, key = addr << 4 | pasid;
+ struct amdgpu_gmc_fault *fault;
+ uint32_t hash;
+
+ /* If we don't have space left in the ring buffer return immediately */
+ stamp = max(timestamp, AMDGPU_GMC_FAULT_TIMEOUT + 1) -
+ AMDGPU_GMC_FAULT_TIMEOUT;
+ if (gmc->fault_ring[gmc->last_fault].timestamp >= stamp)
+ return true;
+
+ /* Try to find the fault in the hash */
+ hash = hash_64(key, AMDGPU_GMC_FAULT_HASH_ORDER);
+ fault = &gmc->fault_ring[gmc->fault_hash[hash].idx];
+ while (fault->timestamp >= stamp) {
+ uint64_t tmp;
+
+ if (fault->key == key)
+ return true;
+
+ tmp = fault->timestamp;
+ fault = &gmc->fault_ring[fault->next];
+
+ /* Check if the entry was reused */
+ if (fault->timestamp >= tmp)
+ break;
+ }
+
+ /* Add the fault to the ring */
+ fault = &gmc->fault_ring[gmc->last_fault];
+ fault->key = key;
+ fault->timestamp = timestamp;
+
+ /* And update the hash */
+ fault->next = gmc->fault_hash[hash].idx;
+ gmc->fault_hash[hash].idx = gmc->last_fault++;
+ return false;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
index 2d63ec729bec..13479c3e6bfb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
@@ -43,8 +43,34 @@
*/
#define AMDGPU_GMC_HOLE_MASK 0x0000ffffffffffffULL
+/*
+ * Ring size as power of two for the log of recent faults.
+ */
+#define AMDGPU_GMC_FAULT_RING_ORDER 8
+#define AMDGPU_GMC_FAULT_RING_SIZE (1 << AMDGPU_GMC_FAULT_RING_ORDER)
+
+/*
+ * Hash size as power of two for the log of recent faults
+ */
+#define AMDGPU_GMC_FAULT_HASH_ORDER 8
+#define AMDGPU_GMC_FAULT_HASH_SIZE (1 << AMDGPU_GMC_FAULT_HASH_ORDER)
+
+/*
+ * Number of IH timestamp ticks until a fault is considered handled
+ */
+#define AMDGPU_GMC_FAULT_TIMEOUT 5000ULL
+
struct firmware;
+/*
+ * GMC page fault information
+ */
+struct amdgpu_gmc_fault {
+ uint64_t timestamp;
+ uint64_t next:AMDGPU_GMC_FAULT_RING_ORDER;
+ uint64_t key:52;
+};
+
/*
* VMHUB structures, functions & helpers
*/
@@ -141,6 +167,12 @@ struct amdgpu_gmc {
struct kfd_vm_fault_info *vm_fault_info;
atomic_t vm_fault_info_updated;
+ struct amdgpu_gmc_fault fault_ring[AMDGPU_GMC_FAULT_RING_SIZE];
+ struct {
+ uint64_t idx:AMDGPU_GMC_FAULT_RING_ORDER;
+ } fault_hash[AMDGPU_GMC_FAULT_HASH_SIZE];
+ uint64_t last_fault:AMDGPU_GMC_FAULT_RING_ORDER;
+
const struct amdgpu_gmc_funcs *gmc_funcs;
struct amdgpu_xgmi xgmi;
@@ -195,5 +227,7 @@ void amdgpu_gmc_gart_location(struct amdgpu_device *adev,
struct amdgpu_gmc *mc);
void amdgpu_gmc_agp_location(struct amdgpu_device *adev,
struct amdgpu_gmc *mc);
+bool amdgpu_gmc_filter_faults(struct amdgpu_device *adev, uint64_t addr,
+ uint16_t pasid, uint64_t timestamp);
#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index ad5de0e99794..593f93d5feb2 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -300,62 +300,6 @@ static int gmc_v9_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
return 0;
}
-/**
- * vega10_ih_prescreen_iv - prescreen an interrupt vector
- *
- * @adev: amdgpu_device pointer
- *
- * Returns true if the interrupt vector should be further processed.
- */
-static bool gmc_v9_0_prescreen_iv(struct amdgpu_device *adev,
- struct amdgpu_iv_entry *entry,
- uint64_t addr)
-{
- struct amdgpu_vm *vm;
- u64 key;
- int r;
-
- /* No PASID, can't identify faulting process */
- if (!entry->pasid)
- return true;
-
- /* Not a retry fault */
- if (!(entry->src_data[1] & 0x80))
- return true;
-
- /* Track retry faults in per-VM fault FIFO. */
- spin_lock(&adev->vm_manager.pasid_lock);
- vm = idr_find(&adev->vm_manager.pasid_idr, entry->pasid);
- if (!vm) {
- /* VM not found, process it normally */
- spin_unlock(&adev->vm_manager.pasid_lock);
- return true;
- }
-
- key = AMDGPU_VM_FAULT(entry->pasid, addr);
- r = amdgpu_vm_add_fault(vm->fault_hash, key);
-
- /* Hash table is full or the fault is already being processed,
- * ignore further page faults
- */
- if (r != 0) {
- spin_unlock(&adev->vm_manager.pasid_lock);
- return false;
- }
- /* No locking required with single writer and single reader */
- r = kfifo_put(&vm->faults, key);
- if (!r) {
- /* FIFO is full. Ignore it until there is space */
- amdgpu_vm_clear_fault(vm->fault_hash, key);
- spin_unlock(&adev->vm_manager.pasid_lock);
- return false;
- }
-
- spin_unlock(&adev->vm_manager.pasid_lock);
- /* It's the first fault for this address, process it normally */
- return true;
-}
-
static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
struct amdgpu_irq_src *source,
struct amdgpu_iv_entry *entry)
@@ -368,9 +312,11 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
addr = (u64)entry->src_data[0] << 12;
addr |= ((u64)entry->src_data[1] & 0xf) << 44;
- if (!gmc_v9_0_prescreen_iv(adev, entry, addr))
+ if (retry_fault && amdgpu_gmc_filter_faults(adev, addr, entry->pasid,
+ entry->timestamp))
return 1; /* This also prevents sending it to KFD */
+ /* If it's the first fault for this address, process it normally */
if (!amdgpu_sriov_vf(adev)) {
status = RREG32(hub->vm_l2_pro_fault_status);
WREG32_P(hub->vm_l2_pro_fault_cntl, 1, ~1);
--
2.17.1
|
