3 * athlon / K7 / K8 / Family 10h model-specific MSR operations
5 * @remark Copyright 2002-2008 OProfile authors
6 * @remark Read the file COPYING
9 * @author Philippe Elie
10 * @author Graydon Hoare
11 * @author Robert Richter <robert.richter@amd.com>
12 * @author Barry Kasindorf
15 #include <linux/oprofile.h>
16 #include <linux/device.h>
17 #include <linux/pci.h>
19 #include <asm/ptrace.h>
23 #include "op_x86_model.h"
24 #include "op_counter.h"
26 #define NUM_COUNTERS 4
27 #define NUM_CONTROLS 4
29 #define CTR_IS_RESERVED(msrs, c) (msrs->counters[(c)].addr ? 1 : 0)
30 #define CTR_READ(l, h, msrs, c) do {rdmsr(msrs->counters[(c)].addr, (l), (h)); } while (0)
31 #define CTR_WRITE(l, msrs, c) do {wrmsr(msrs->counters[(c)].addr, -(unsigned int)(l), -1); } while (0)
32 #define CTR_OVERFLOWED(n) (!((n) & (1U<<31)))
34 #define CTRL_IS_RESERVED(msrs, c) (msrs->controls[(c)].addr ? 1 : 0)
35 #define CTRL_READ(l, h, msrs, c) do {rdmsr(msrs->controls[(c)].addr, (l), (h)); } while (0)
36 #define CTRL_WRITE(l, h, msrs, c) do {wrmsr(msrs->controls[(c)].addr, (l), (h)); } while (0)
37 #define CTRL_SET_ACTIVE(n) (n |= (1<<22))
38 #define CTRL_SET_INACTIVE(n) (n &= ~(1<<22))
39 #define CTRL_CLEAR_LO(x) (x &= (1<<21))
40 #define CTRL_CLEAR_HI(x) (x &= 0xfffffcf0)
41 #define CTRL_SET_ENABLE(val) (val |= 1<<20)
42 #define CTRL_SET_USR(val, u) (val |= ((u & 1) << 16))
43 #define CTRL_SET_KERN(val, k) (val |= ((k & 1) << 17))
44 #define CTRL_SET_UM(val, m) (val |= (m << 8))
45 #define CTRL_SET_EVENT_LOW(val, e) (val |= (e & 0xff))
46 #define CTRL_SET_EVENT_HIGH(val, e) (val |= ((e >> 8) & 0xf))
47 #define CTRL_SET_HOST_ONLY(val, h) (val |= ((h & 1) << 9))
48 #define CTRL_SET_GUEST_ONLY(val, h) (val |= ((h & 1) << 8))
50 static unsigned long reset_value
[NUM_COUNTERS
];
52 #ifdef CONFIG_OPROFILE_IBS
54 /* IbsFetchCtl bits/masks */
55 #define IBS_FETCH_HIGH_VALID_BIT (1UL << 17) /* bit 49 */
56 #define IBS_FETCH_HIGH_ENABLE (1UL << 16) /* bit 48 */
57 #define IBS_FETCH_LOW_MAX_CNT_MASK 0x0000FFFFUL /* MaxCnt mask */
60 #define IBS_OP_LOW_VALID_BIT (1ULL<<18) /* bit 18 */
61 #define IBS_OP_LOW_ENABLE (1ULL<<17) /* bit 17 */
63 /* Codes used in cpu_buffer.c */
64 /* This produces duplicate code, need to be fixed */
65 #define IBS_FETCH_BEGIN 3
66 #define IBS_OP_BEGIN 4
69 * The function interface needs to be fixed, something like add
70 * data. Should then be added to linux/oprofile.h.
73 oprofile_add_ibs_sample(struct pt_regs
* const regs
,
74 unsigned int * const ibs_sample
, int ibs_code
);
76 struct ibs_fetch_sample
{
77 /* MSRC001_1031 IBS Fetch Linear Address Register */
78 unsigned int ibs_fetch_lin_addr_low
;
79 unsigned int ibs_fetch_lin_addr_high
;
80 /* MSRC001_1030 IBS Fetch Control Register */
81 unsigned int ibs_fetch_ctl_low
;
82 unsigned int ibs_fetch_ctl_high
;
83 /* MSRC001_1032 IBS Fetch Physical Address Register */
84 unsigned int ibs_fetch_phys_addr_low
;
85 unsigned int ibs_fetch_phys_addr_high
;
88 struct ibs_op_sample
{
89 /* MSRC001_1034 IBS Op Logical Address Register (IbsRIP) */
90 unsigned int ibs_op_rip_low
;
91 unsigned int ibs_op_rip_high
;
92 /* MSRC001_1035 IBS Op Data Register */
93 unsigned int ibs_op_data1_low
;
94 unsigned int ibs_op_data1_high
;
95 /* MSRC001_1036 IBS Op Data 2 Register */
96 unsigned int ibs_op_data2_low
;
97 unsigned int ibs_op_data2_high
;
98 /* MSRC001_1037 IBS Op Data 3 Register */
99 unsigned int ibs_op_data3_low
;
100 unsigned int ibs_op_data3_high
;
101 /* MSRC001_1038 IBS DC Linear Address Register (IbsDcLinAd) */
102 unsigned int ibs_dc_linear_low
;
103 unsigned int ibs_dc_linear_high
;
104 /* MSRC001_1039 IBS DC Physical Address Register (IbsDcPhysAd) */
105 unsigned int ibs_dc_phys_low
;
106 unsigned int ibs_dc_phys_high
;
110 * unitialize the APIC for the IBS interrupts if needed on AMD Family10h+
112 static void clear_ibs_nmi(void);
114 static int ibs_allowed
; /* AMD Family10h and later */
116 struct op_ibs_config
{
117 unsigned long op_enabled
;
118 unsigned long fetch_enabled
;
119 unsigned long max_cnt_fetch
;
120 unsigned long max_cnt_op
;
121 unsigned long rand_en
;
122 unsigned long dispatched_ops
;
125 static struct op_ibs_config ibs_config
;
129 /* functions for op_amd_spec */
131 static void op_amd_fill_in_addresses(struct op_msrs
* const msrs
)
135 for (i
= 0; i
< NUM_COUNTERS
; i
++) {
136 if (reserve_perfctr_nmi(MSR_K7_PERFCTR0
+ i
))
137 msrs
->counters
[i
].addr
= MSR_K7_PERFCTR0
+ i
;
139 msrs
->counters
[i
].addr
= 0;
142 for (i
= 0; i
< NUM_CONTROLS
; i
++) {
143 if (reserve_evntsel_nmi(MSR_K7_EVNTSEL0
+ i
))
144 msrs
->controls
[i
].addr
= MSR_K7_EVNTSEL0
+ i
;
146 msrs
->controls
[i
].addr
= 0;
151 static void op_amd_setup_ctrs(struct op_msrs
const * const msrs
)
153 unsigned int low
, high
;
156 /* clear all counters */
157 for (i
= 0 ; i
< NUM_CONTROLS
; ++i
) {
158 if (unlikely(!CTRL_IS_RESERVED(msrs
, i
)))
160 CTRL_READ(low
, high
, msrs
, i
);
163 CTRL_WRITE(low
, high
, msrs
, i
);
166 /* avoid a false detection of ctr overflows in NMI handler */
167 for (i
= 0; i
< NUM_COUNTERS
; ++i
) {
168 if (unlikely(!CTR_IS_RESERVED(msrs
, i
)))
170 CTR_WRITE(1, msrs
, i
);
173 /* enable active counters */
174 for (i
= 0; i
< NUM_COUNTERS
; ++i
) {
175 if ((counter_config
[i
].enabled
) && (CTR_IS_RESERVED(msrs
, i
))) {
176 reset_value
[i
] = counter_config
[i
].count
;
178 CTR_WRITE(counter_config
[i
].count
, msrs
, i
);
180 CTRL_READ(low
, high
, msrs
, i
);
183 CTRL_SET_ENABLE(low
);
184 CTRL_SET_USR(low
, counter_config
[i
].user
);
185 CTRL_SET_KERN(low
, counter_config
[i
].kernel
);
186 CTRL_SET_UM(low
, counter_config
[i
].unit_mask
);
187 CTRL_SET_EVENT_LOW(low
, counter_config
[i
].event
);
188 CTRL_SET_EVENT_HIGH(high
, counter_config
[i
].event
);
189 CTRL_SET_HOST_ONLY(high
, 0);
190 CTRL_SET_GUEST_ONLY(high
, 0);
192 CTRL_WRITE(low
, high
, msrs
, i
);
199 #ifdef CONFIG_OPROFILE_IBS
202 op_amd_handle_ibs(struct pt_regs
* const regs
,
203 struct op_msrs
const * const msrs
)
205 unsigned int low
, high
;
206 struct ibs_fetch_sample ibs_fetch
;
207 struct ibs_op_sample ibs_op
;
212 if (ibs_config
.fetch_enabled
) {
213 rdmsr(MSR_AMD64_IBSFETCHCTL
, low
, high
);
214 if (high
& IBS_FETCH_HIGH_VALID_BIT
) {
215 ibs_fetch
.ibs_fetch_ctl_high
= high
;
216 ibs_fetch
.ibs_fetch_ctl_low
= low
;
217 rdmsr(MSR_AMD64_IBSFETCHLINAD
, low
, high
);
218 ibs_fetch
.ibs_fetch_lin_addr_high
= high
;
219 ibs_fetch
.ibs_fetch_lin_addr_low
= low
;
220 rdmsr(MSR_AMD64_IBSFETCHPHYSAD
, low
, high
);
221 ibs_fetch
.ibs_fetch_phys_addr_high
= high
;
222 ibs_fetch
.ibs_fetch_phys_addr_low
= low
;
224 oprofile_add_ibs_sample(regs
,
225 (unsigned int *)&ibs_fetch
,
228 /* reenable the IRQ */
229 rdmsr(MSR_AMD64_IBSFETCHCTL
, low
, high
);
230 high
&= ~IBS_FETCH_HIGH_VALID_BIT
;
231 high
|= IBS_FETCH_HIGH_ENABLE
;
232 low
&= IBS_FETCH_LOW_MAX_CNT_MASK
;
233 wrmsr(MSR_AMD64_IBSFETCHCTL
, low
, high
);
237 if (ibs_config
.op_enabled
) {
238 rdmsr(MSR_AMD64_IBSOPCTL
, low
, high
);
239 if (low
& IBS_OP_LOW_VALID_BIT
) {
240 rdmsr(MSR_AMD64_IBSOPRIP
, low
, high
);
241 ibs_op
.ibs_op_rip_low
= low
;
242 ibs_op
.ibs_op_rip_high
= high
;
243 rdmsr(MSR_AMD64_IBSOPDATA
, low
, high
);
244 ibs_op
.ibs_op_data1_low
= low
;
245 ibs_op
.ibs_op_data1_high
= high
;
246 rdmsr(MSR_AMD64_IBSOPDATA2
, low
, high
);
247 ibs_op
.ibs_op_data2_low
= low
;
248 ibs_op
.ibs_op_data2_high
= high
;
249 rdmsr(MSR_AMD64_IBSOPDATA3
, low
, high
);
250 ibs_op
.ibs_op_data3_low
= low
;
251 ibs_op
.ibs_op_data3_high
= high
;
252 rdmsr(MSR_AMD64_IBSDCLINAD
, low
, high
);
253 ibs_op
.ibs_dc_linear_low
= low
;
254 ibs_op
.ibs_dc_linear_high
= high
;
255 rdmsr(MSR_AMD64_IBSDCPHYSAD
, low
, high
);
256 ibs_op
.ibs_dc_phys_low
= low
;
257 ibs_op
.ibs_dc_phys_high
= high
;
259 /* reenable the IRQ */
260 oprofile_add_ibs_sample(regs
,
261 (unsigned int *)&ibs_op
,
263 rdmsr(MSR_AMD64_IBSOPCTL
, low
, high
);
265 low
&= ~IBS_OP_LOW_VALID_BIT
;
266 low
|= IBS_OP_LOW_ENABLE
;
267 wrmsr(MSR_AMD64_IBSOPCTL
, low
, high
);
276 static int op_amd_check_ctrs(struct pt_regs
* const regs
,
277 struct op_msrs
const * const msrs
)
279 unsigned int low
, high
;
282 for (i
= 0 ; i
< NUM_COUNTERS
; ++i
) {
285 CTR_READ(low
, high
, msrs
, i
);
286 if (CTR_OVERFLOWED(low
)) {
287 oprofile_add_sample(regs
, i
);
288 CTR_WRITE(reset_value
[i
], msrs
, i
);
292 #ifdef CONFIG_OPROFILE_IBS
293 op_amd_handle_ibs(regs
, msrs
);
296 /* See op_model_ppro.c */
300 static void op_amd_start(struct op_msrs
const * const msrs
)
302 unsigned int low
, high
;
304 for (i
= 0 ; i
< NUM_COUNTERS
; ++i
) {
305 if (reset_value
[i
]) {
306 CTRL_READ(low
, high
, msrs
, i
);
307 CTRL_SET_ACTIVE(low
);
308 CTRL_WRITE(low
, high
, msrs
, i
);
312 #ifdef CONFIG_OPROFILE_IBS
313 if (ibs_allowed
&& ibs_config
.fetch_enabled
) {
314 low
= (ibs_config
.max_cnt_fetch
>> 4) & 0xFFFF;
315 high
= ((ibs_config
.rand_en
& 0x1) << 25) /* bit 57 */
316 + IBS_FETCH_HIGH_ENABLE
;
317 wrmsr(MSR_AMD64_IBSFETCHCTL
, low
, high
);
320 if (ibs_allowed
&& ibs_config
.op_enabled
) {
321 low
= ((ibs_config
.max_cnt_op
>> 4) & 0xFFFF)
322 + ((ibs_config
.dispatched_ops
& 0x1) << 19) /* bit 19 */
325 wrmsr(MSR_AMD64_IBSOPCTL
, low
, high
);
331 static void op_amd_stop(struct op_msrs
const * const msrs
)
333 unsigned int low
, high
;
337 * Subtle: stop on all counters to avoid race with setting our
340 for (i
= 0 ; i
< NUM_COUNTERS
; ++i
) {
343 CTRL_READ(low
, high
, msrs
, i
);
344 CTRL_SET_INACTIVE(low
);
345 CTRL_WRITE(low
, high
, msrs
, i
);
348 #ifdef CONFIG_OPROFILE_IBS
349 if (ibs_allowed
&& ibs_config
.fetch_enabled
) {
350 /* clear max count and enable */
353 wrmsr(MSR_AMD64_IBSFETCHCTL
, low
, high
);
356 if (ibs_allowed
&& ibs_config
.op_enabled
) {
357 /* clear max count and enable */
360 wrmsr(MSR_AMD64_IBSOPCTL
, low
, high
);
365 static void op_amd_shutdown(struct op_msrs
const * const msrs
)
369 for (i
= 0 ; i
< NUM_COUNTERS
; ++i
) {
370 if (CTR_IS_RESERVED(msrs
, i
))
371 release_perfctr_nmi(MSR_K7_PERFCTR0
+ i
);
373 for (i
= 0 ; i
< NUM_CONTROLS
; ++i
) {
374 if (CTRL_IS_RESERVED(msrs
, i
))
375 release_evntsel_nmi(MSR_K7_EVNTSEL0
+ i
);
379 #ifndef CONFIG_OPROFILE_IBS
383 static int op_amd_init(struct oprofile_operations
*ops
)
388 static void op_amd_exit(void) {}
392 static u8 ibs_eilvt_off
;
394 static inline void apic_init_ibs_nmi_per_cpu(void *arg
)
396 ibs_eilvt_off
= setup_APIC_eilvt_ibs(0, APIC_EILVT_MSG_NMI
, 0);
399 static inline void apic_clear_ibs_nmi_per_cpu(void *arg
)
401 setup_APIC_eilvt_ibs(0, APIC_EILVT_MSG_FIX
, 1);
404 static int pfm_amd64_setup_eilvt(void)
406 #define IBSCTL_LVTOFFSETVAL (1 << 8)
408 struct pci_dev
*cpu_cfg
;
413 on_each_cpu(apic_init_ibs_nmi_per_cpu
, NULL
, 1);
418 cpu_cfg
= pci_get_device(PCI_VENDOR_ID_AMD
,
419 PCI_DEVICE_ID_AMD_10H_NB_MISC
,
424 pci_write_config_dword(cpu_cfg
, IBSCTL
, ibs_eilvt_off
425 | IBSCTL_LVTOFFSETVAL
);
426 pci_read_config_dword(cpu_cfg
, IBSCTL
, &value
);
427 if (value
!= (ibs_eilvt_off
| IBSCTL_LVTOFFSETVAL
)) {
428 printk(KERN_DEBUG
"Failed to setup IBS LVT offset, "
429 "IBSCTL = 0x%08x", value
);
435 printk(KERN_DEBUG
"No CPU node configured for IBS");
441 /* Works only for 64bit with proper numa implementation. */
442 if (nodes
!= num_possible_nodes()) {
443 printk(KERN_DEBUG
"Failed to setup CPU node(s) for IBS, "
444 "found: %d, expected %d",
445 nodes
, num_possible_nodes());
452 /* initialize the APIC for the IBS interrupts if available */
453 static void setup_ibs(void)
455 ibs_allowed
= boot_cpu_has(X86_FEATURE_IBS
);
460 if (pfm_amd64_setup_eilvt()) {
465 printk(KERN_INFO
"oprofile: AMD IBS detected\n");
469 /* uninitialize the APIC for the IBS interrupts if needed */
470 static void clear_ibs_nmi(void)
473 on_each_cpu(apic_clear_ibs_nmi_per_cpu
, NULL
, 1);
476 static int (*create_arch_files
)(struct super_block
*sb
, struct dentry
*root
);
478 static int setup_ibs_files(struct super_block
*sb
, struct dentry
*root
)
483 /* architecture specific files */
484 if (create_arch_files
)
485 ret
= create_arch_files(sb
, root
);
493 /* model specific files */
495 /* setup some reasonable defaults */
496 ibs_config
.max_cnt_fetch
= 250000;
497 ibs_config
.fetch_enabled
= 0;
498 ibs_config
.max_cnt_op
= 250000;
499 ibs_config
.op_enabled
= 0;
500 ibs_config
.dispatched_ops
= 1;
502 dir
= oprofilefs_mkdir(sb
, root
, "ibs_fetch");
503 oprofilefs_create_ulong(sb
, dir
, "enable",
504 &ibs_config
.fetch_enabled
);
505 oprofilefs_create_ulong(sb
, dir
, "max_count",
506 &ibs_config
.max_cnt_fetch
);
507 oprofilefs_create_ulong(sb
, dir
, "rand_enable",
508 &ibs_config
.rand_en
);
510 dir
= oprofilefs_mkdir(sb
, root
, "ibs_op");
511 oprofilefs_create_ulong(sb
, dir
, "enable",
512 &ibs_config
.op_enabled
);
513 oprofilefs_create_ulong(sb
, dir
, "max_count",
514 &ibs_config
.max_cnt_op
);
515 oprofilefs_create_ulong(sb
, dir
, "dispatched_ops",
516 &ibs_config
.dispatched_ops
);
521 static int op_amd_init(struct oprofile_operations
*ops
)
524 create_arch_files
= ops
->create_files
;
525 ops
->create_files
= setup_ibs_files
;
529 static void op_amd_exit(void)
536 struct op_x86_model_spec
const op_amd_spec
= {
539 .num_counters
= NUM_COUNTERS
,
540 .num_controls
= NUM_CONTROLS
,
541 .fill_in_addresses
= &op_amd_fill_in_addresses
,
542 .setup_ctrs
= &op_amd_setup_ctrs
,
543 .check_ctrs
= &op_amd_check_ctrs
,
544 .start
= &op_amd_start
,
545 .stop
= &op_amd_stop
,
546 .shutdown
= &op_amd_shutdown