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
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
|
/*
* RISC-V IOMMU - Hardware Performance Monitor (HPM) helpers
*
* Copyright (C) 2022-2023 Rivos Inc.
*
* This program is free software; you can redistribute it and/or modify it
* under the terms and conditions of the GNU General Public License,
* version 2 or later, as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, see <http://www.gnu.org/licenses/>.
*/
#include "qemu/osdep.h"
#include "qemu/timer.h"
#include "cpu_bits.h"
#include "riscv-iommu-hpm.h"
#include "riscv-iommu.h"
#include "riscv-iommu-bits.h"
#include "trace.h"
/* For now we assume IOMMU HPM frequency to be 1GHz so 1-cycle is of 1-ns. */
static inline uint64_t get_cycles(void)
{
return qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
}
uint64_t riscv_iommu_hpmcycle_read(RISCVIOMMUState *s)
{
const uint64_t cycle = riscv_iommu_reg_get64(
s, RISCV_IOMMU_REG_IOHPMCYCLES);
const uint32_t inhibit = riscv_iommu_reg_get32(
s, RISCV_IOMMU_REG_IOCOUNTINH);
const uint64_t ctr_prev = s->hpmcycle_prev;
const uint64_t ctr_val = s->hpmcycle_val;
trace_riscv_iommu_hpm_read(cycle, inhibit, ctr_prev, ctr_val);
if (get_field(inhibit, RISCV_IOMMU_IOCOUNTINH_CY)) {
/*
* Counter should not increment if inhibit bit is set. We can't really
* stop the QEMU_CLOCK_VIRTUAL, so we just return the last updated
* counter value to indicate that counter was not incremented.
*/
return (ctr_val & RISCV_IOMMU_IOHPMCYCLES_COUNTER) |
(cycle & RISCV_IOMMU_IOHPMCYCLES_OVF);
}
return (ctr_val + get_cycles() - ctr_prev) |
(cycle & RISCV_IOMMU_IOHPMCYCLES_OVF);
}
static void hpm_incr_ctr(RISCVIOMMUState *s, uint32_t ctr_idx)
{
const uint32_t off = ctr_idx << 3;
uint64_t cntr_val;
cntr_val = ldq_le_p(&s->regs_rw[RISCV_IOMMU_REG_IOHPMCTR_BASE + off]);
stq_le_p(&s->regs_rw[RISCV_IOMMU_REG_IOHPMCTR_BASE + off], cntr_val + 1);
trace_riscv_iommu_hpm_incr_ctr(cntr_val);
/* Handle the overflow scenario. */
if (cntr_val == UINT64_MAX) {
/*
* Generate interrupt only if OF bit is clear. +1 to offset the cycle
* register OF bit.
*/
const uint32_t ovf =
riscv_iommu_reg_mod32(s, RISCV_IOMMU_REG_IOCOUNTOVF,
BIT(ctr_idx + 1), 0);
if (!get_field(ovf, BIT(ctr_idx + 1))) {
riscv_iommu_reg_mod64(s,
RISCV_IOMMU_REG_IOHPMEVT_BASE + off,
RISCV_IOMMU_IOHPMEVT_OF,
0);
riscv_iommu_notify(s, RISCV_IOMMU_INTR_PM);
}
}
}
void riscv_iommu_hpm_incr_ctr(RISCVIOMMUState *s, RISCVIOMMUContext *ctx,
unsigned event_id)
{
const uint32_t inhibit = riscv_iommu_reg_get32(
s, RISCV_IOMMU_REG_IOCOUNTINH);
uint32_t did_gscid;
uint32_t pid_pscid;
uint32_t ctr_idx;
gpointer value;
uint32_t ctrs;
uint64_t evt;
if (!(s->cap & RISCV_IOMMU_CAP_HPM)) {
return;
}
value = g_hash_table_lookup(s->hpm_event_ctr_map,
GUINT_TO_POINTER(event_id));
if (value == NULL) {
return;
}
for (ctrs = GPOINTER_TO_UINT(value); ctrs != 0; ctrs &= ctrs - 1) {
ctr_idx = ctz32(ctrs);
if (get_field(inhibit, BIT(ctr_idx + 1))) {
continue;
}
evt = riscv_iommu_reg_get64(s,
RISCV_IOMMU_REG_IOHPMEVT_BASE + (ctr_idx << 3));
/*
* It's quite possible that event ID has been changed in counter
* but hashtable hasn't been updated yet. We don't want to increment
* counter for the old event ID.
*/
if (event_id != get_field(evt, RISCV_IOMMU_IOHPMEVT_EVENT_ID)) {
continue;
}
if (get_field(evt, RISCV_IOMMU_IOHPMEVT_IDT)) {
did_gscid = get_field(ctx->gatp, RISCV_IOMMU_DC_IOHGATP_GSCID);
pid_pscid = get_field(ctx->ta, RISCV_IOMMU_DC_TA_PSCID);
} else {
did_gscid = ctx->devid;
pid_pscid = ctx->process_id;
}
if (get_field(evt, RISCV_IOMMU_IOHPMEVT_PV_PSCV)) {
/*
* If the transaction does not have a valid process_id, counter
* increments if device_id matches DID_GSCID. If the transaction
* has a valid process_id, counter increments if device_id
* matches DID_GSCID and process_id matches PID_PSCID. See
* IOMMU Specification, Chapter 5.23. Performance-monitoring
* event selector.
*/
if (ctx->process_id &&
get_field(evt, RISCV_IOMMU_IOHPMEVT_PID_PSCID) != pid_pscid) {
continue;
}
}
if (get_field(evt, RISCV_IOMMU_IOHPMEVT_DV_GSCV)) {
uint32_t mask = ~0;
if (get_field(evt, RISCV_IOMMU_IOHPMEVT_DMASK)) {
/*
* 1001 1011 mask = GSCID
* 0000 0111 mask = mask ^ (mask + 1)
* 1111 1000 mask = ~mask;
*/
mask = get_field(evt, RISCV_IOMMU_IOHPMEVT_DID_GSCID);
mask = mask ^ (mask + 1);
mask = ~mask;
}
if ((get_field(evt, RISCV_IOMMU_IOHPMEVT_DID_GSCID) & mask) !=
(did_gscid & mask)) {
continue;
}
}
hpm_incr_ctr(s, ctr_idx);
}
}
/* Timer callback for cycle counter overflow. */
void riscv_iommu_hpm_timer_cb(void *priv)
{
RISCVIOMMUState *s = priv;
const uint32_t inhibit = riscv_iommu_reg_get32(
s, RISCV_IOMMU_REG_IOCOUNTINH);
uint32_t ovf;
if (get_field(inhibit, RISCV_IOMMU_IOCOUNTINH_CY)) {
return;
}
if (s->irq_overflow_left > 0) {
uint64_t irq_trigger_at =
qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + s->irq_overflow_left;
timer_mod_anticipate_ns(s->hpm_timer, irq_trigger_at);
s->irq_overflow_left = 0;
return;
}
ovf = riscv_iommu_reg_get32(s, RISCV_IOMMU_REG_IOCOUNTOVF);
if (!get_field(ovf, RISCV_IOMMU_IOCOUNTOVF_CY)) {
/*
* We don't need to set hpmcycle_val to zero and update hpmcycle_prev to
* current clock value. The way we calculate iohpmcycs will overflow
* and return the correct value. This avoids the need to synchronize
* timer callback and write callback.
*/
riscv_iommu_reg_mod32(s, RISCV_IOMMU_REG_IOCOUNTOVF,
RISCV_IOMMU_IOCOUNTOVF_CY, 0);
riscv_iommu_reg_mod64(s, RISCV_IOMMU_REG_IOHPMCYCLES,
RISCV_IOMMU_IOHPMCYCLES_OVF, 0);
riscv_iommu_notify(s, RISCV_IOMMU_INTR_PM);
}
}
static void hpm_setup_timer(RISCVIOMMUState *s, uint64_t value)
{
const uint32_t inhibit = riscv_iommu_reg_get32(
s, RISCV_IOMMU_REG_IOCOUNTINH);
uint64_t overflow_at, overflow_ns;
if (get_field(inhibit, RISCV_IOMMU_IOCOUNTINH_CY)) {
return;
}
/*
* We are using INT64_MAX here instead to UINT64_MAX because cycle counter
* has 63-bit precision and INT64_MAX is the maximum it can store.
*/
if (value) {
overflow_ns = INT64_MAX - value + 1;
} else {
overflow_ns = INT64_MAX;
}
overflow_at = (uint64_t)qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + overflow_ns;
if (overflow_at > INT64_MAX) {
s->irq_overflow_left = overflow_at - INT64_MAX;
overflow_at = INT64_MAX;
}
timer_mod_anticipate_ns(s->hpm_timer, overflow_at);
}
/* Updates the internal cycle counter state when iocntinh:CY is changed. */
void riscv_iommu_process_iocntinh_cy(RISCVIOMMUState *s, bool prev_cy_inh)
{
const uint32_t inhibit = riscv_iommu_reg_get32(
s, RISCV_IOMMU_REG_IOCOUNTINH);
/* We only need to process CY bit toggle. */
if (!(inhibit ^ prev_cy_inh)) {
return;
}
trace_riscv_iommu_hpm_iocntinh_cy(prev_cy_inh);
if (!(inhibit & RISCV_IOMMU_IOCOUNTINH_CY)) {
/*
* Cycle counter is enabled. Just start the timer again and update
* the clock snapshot value to point to the current time to make
* sure iohpmcycles read is correct.
*/
s->hpmcycle_prev = get_cycles();
hpm_setup_timer(s, s->hpmcycle_val);
} else {
/*
* Cycle counter is disabled. Stop the timer and update the cycle
* counter to record the current value which is last programmed
* value + the cycles passed so far.
*/
s->hpmcycle_val = s->hpmcycle_val + (get_cycles() - s->hpmcycle_prev);
timer_del(s->hpm_timer);
}
}
void riscv_iommu_process_hpmcycle_write(RISCVIOMMUState *s)
{
const uint64_t val = riscv_iommu_reg_get64(s, RISCV_IOMMU_REG_IOHPMCYCLES);
const uint32_t ovf = riscv_iommu_reg_get32(s, RISCV_IOMMU_REG_IOCOUNTOVF);
trace_riscv_iommu_hpm_cycle_write(ovf, val);
/*
* Clear OF bit in IOCNTOVF if it's being cleared in IOHPMCYCLES register.
*/
if (get_field(ovf, RISCV_IOMMU_IOCOUNTOVF_CY) &&
!get_field(val, RISCV_IOMMU_IOHPMCYCLES_OVF)) {
riscv_iommu_reg_mod32(s, RISCV_IOMMU_REG_IOCOUNTOVF, 0,
RISCV_IOMMU_IOCOUNTOVF_CY);
}
s->hpmcycle_val = val & ~RISCV_IOMMU_IOHPMCYCLES_OVF;
s->hpmcycle_prev = get_cycles();
hpm_setup_timer(s, s->hpmcycle_val);
}
static inline bool check_valid_event_id(unsigned event_id)
{
return event_id > RISCV_IOMMU_HPMEVENT_INVALID &&
event_id < RISCV_IOMMU_HPMEVENT_MAX;
}
static gboolean hpm_event_equal(gpointer key, gpointer value, gpointer udata)
{
uint32_t *pair = udata;
if (GPOINTER_TO_UINT(value) & (1 << pair[0])) {
pair[1] = GPOINTER_TO_UINT(key);
return true;
}
return false;
}
/* Caller must check ctr_idx against hpm_ctrs to see if its supported or not. */
static void update_event_map(RISCVIOMMUState *s, uint64_t value,
uint32_t ctr_idx)
{
unsigned event_id = get_field(value, RISCV_IOMMU_IOHPMEVT_EVENT_ID);
uint32_t pair[2] = { ctr_idx, RISCV_IOMMU_HPMEVENT_INVALID };
uint32_t new_value = 1 << ctr_idx;
gpointer data;
/*
* If EventID field is RISCV_IOMMU_HPMEVENT_INVALID
* remove the current mapping.
*/
if (event_id == RISCV_IOMMU_HPMEVENT_INVALID) {
data = g_hash_table_find(s->hpm_event_ctr_map, hpm_event_equal, pair);
new_value = GPOINTER_TO_UINT(data) & ~(new_value);
if (new_value != 0) {
g_hash_table_replace(s->hpm_event_ctr_map,
GUINT_TO_POINTER(pair[1]),
GUINT_TO_POINTER(new_value));
} else {
g_hash_table_remove(s->hpm_event_ctr_map,
GUINT_TO_POINTER(pair[1]));
}
return;
}
/* Update the counter mask if the event is already enabled. */
if (g_hash_table_lookup_extended(s->hpm_event_ctr_map,
GUINT_TO_POINTER(event_id),
NULL,
&data)) {
new_value |= GPOINTER_TO_UINT(data);
}
g_hash_table_insert(s->hpm_event_ctr_map,
GUINT_TO_POINTER(event_id),
GUINT_TO_POINTER(new_value));
}
void riscv_iommu_process_hpmevt_write(RISCVIOMMUState *s, uint32_t evt_reg)
{
const uint32_t ctr_idx = (evt_reg - RISCV_IOMMU_REG_IOHPMEVT_BASE) >> 3;
const uint32_t ovf = riscv_iommu_reg_get32(s, RISCV_IOMMU_REG_IOCOUNTOVF);
uint64_t val = riscv_iommu_reg_get64(s, evt_reg);
if (ctr_idx >= s->hpm_cntrs) {
return;
}
trace_riscv_iommu_hpm_evt_write(ctr_idx, ovf, val);
/* Clear OF bit in IOCNTOVF if it's being cleared in IOHPMEVT register. */
if (get_field(ovf, BIT(ctr_idx + 1)) &&
!get_field(val, RISCV_IOMMU_IOHPMEVT_OF)) {
/* +1 to offset CYCLE register OF bit. */
riscv_iommu_reg_mod32(
s, RISCV_IOMMU_REG_IOCOUNTOVF, 0, BIT(ctr_idx + 1));
}
if (!check_valid_event_id(get_field(val, RISCV_IOMMU_IOHPMEVT_EVENT_ID))) {
/* Reset EventID (WARL) field to invalid. */
val = set_field(val, RISCV_IOMMU_IOHPMEVT_EVENT_ID,
RISCV_IOMMU_HPMEVENT_INVALID);
riscv_iommu_reg_set64(s, evt_reg, val);
}
update_event_map(s, val, ctr_idx);
}
|