3 * athlon / K7 / K8 / Family 10h model-specific MSR operations
5 * @remark Copyright 2002-2009 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 (1UL << 4)
66 #define IBS_OP_BEGIN (1UL << 5)
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
;
109 static int has_ibs
; /* AMD Family10h and later */
111 struct op_ibs_config
{
112 unsigned long op_enabled
;
113 unsigned long fetch_enabled
;
114 unsigned long max_cnt_fetch
;
115 unsigned long max_cnt_op
;
116 unsigned long rand_en
;
117 unsigned long dispatched_ops
;
120 static struct op_ibs_config ibs_config
;
124 /* functions for op_amd_spec */
126 static void op_amd_fill_in_addresses(struct op_msrs
* const msrs
)
130 for (i
= 0; i
< NUM_COUNTERS
; i
++) {
131 if (reserve_perfctr_nmi(MSR_K7_PERFCTR0
+ i
))
132 msrs
->counters
[i
].addr
= MSR_K7_PERFCTR0
+ i
;
134 msrs
->counters
[i
].addr
= 0;
137 for (i
= 0; i
< NUM_CONTROLS
; i
++) {
138 if (reserve_evntsel_nmi(MSR_K7_EVNTSEL0
+ i
))
139 msrs
->controls
[i
].addr
= MSR_K7_EVNTSEL0
+ i
;
141 msrs
->controls
[i
].addr
= 0;
146 static void op_amd_setup_ctrs(struct op_msrs
const * const msrs
)
148 unsigned int low
, high
;
151 /* clear all counters */
152 for (i
= 0 ; i
< NUM_CONTROLS
; ++i
) {
153 if (unlikely(!CTRL_IS_RESERVED(msrs
, i
)))
155 CTRL_READ(low
, high
, msrs
, i
);
158 CTRL_WRITE(low
, high
, msrs
, i
);
161 /* avoid a false detection of ctr overflows in NMI handler */
162 for (i
= 0; i
< NUM_COUNTERS
; ++i
) {
163 if (unlikely(!CTR_IS_RESERVED(msrs
, i
)))
165 CTR_WRITE(1, msrs
, i
);
168 /* enable active counters */
169 for (i
= 0; i
< NUM_COUNTERS
; ++i
) {
170 if ((counter_config
[i
].enabled
) && (CTR_IS_RESERVED(msrs
, i
))) {
171 reset_value
[i
] = counter_config
[i
].count
;
173 CTR_WRITE(counter_config
[i
].count
, msrs
, i
);
175 CTRL_READ(low
, high
, msrs
, i
);
178 CTRL_SET_ENABLE(low
);
179 CTRL_SET_USR(low
, counter_config
[i
].user
);
180 CTRL_SET_KERN(low
, counter_config
[i
].kernel
);
181 CTRL_SET_UM(low
, counter_config
[i
].unit_mask
);
182 CTRL_SET_EVENT_LOW(low
, counter_config
[i
].event
);
183 CTRL_SET_EVENT_HIGH(high
, counter_config
[i
].event
);
184 CTRL_SET_HOST_ONLY(high
, 0);
185 CTRL_SET_GUEST_ONLY(high
, 0);
187 CTRL_WRITE(low
, high
, msrs
, i
);
194 #ifdef CONFIG_OPROFILE_IBS
197 op_amd_handle_ibs(struct pt_regs
* const regs
,
198 struct op_msrs
const * const msrs
)
200 unsigned int low
, high
;
201 struct ibs_fetch_sample ibs_fetch
;
202 struct ibs_op_sample ibs_op
;
207 if (ibs_config
.fetch_enabled
) {
208 rdmsr(MSR_AMD64_IBSFETCHCTL
, low
, high
);
209 if (high
& IBS_FETCH_HIGH_VALID_BIT
) {
210 ibs_fetch
.ibs_fetch_ctl_high
= high
;
211 ibs_fetch
.ibs_fetch_ctl_low
= low
;
212 rdmsr(MSR_AMD64_IBSFETCHLINAD
, low
, high
);
213 ibs_fetch
.ibs_fetch_lin_addr_high
= high
;
214 ibs_fetch
.ibs_fetch_lin_addr_low
= low
;
215 rdmsr(MSR_AMD64_IBSFETCHPHYSAD
, low
, high
);
216 ibs_fetch
.ibs_fetch_phys_addr_high
= high
;
217 ibs_fetch
.ibs_fetch_phys_addr_low
= low
;
219 oprofile_add_ibs_sample(regs
,
220 (unsigned int *)&ibs_fetch
,
223 /* reenable the IRQ */
224 rdmsr(MSR_AMD64_IBSFETCHCTL
, low
, high
);
225 high
&= ~IBS_FETCH_HIGH_VALID_BIT
;
226 high
|= IBS_FETCH_HIGH_ENABLE
;
227 low
&= IBS_FETCH_LOW_MAX_CNT_MASK
;
228 wrmsr(MSR_AMD64_IBSFETCHCTL
, low
, high
);
232 if (ibs_config
.op_enabled
) {
233 rdmsr(MSR_AMD64_IBSOPCTL
, low
, high
);
234 if (low
& IBS_OP_LOW_VALID_BIT
) {
235 rdmsr(MSR_AMD64_IBSOPRIP
, low
, high
);
236 ibs_op
.ibs_op_rip_low
= low
;
237 ibs_op
.ibs_op_rip_high
= high
;
238 rdmsr(MSR_AMD64_IBSOPDATA
, low
, high
);
239 ibs_op
.ibs_op_data1_low
= low
;
240 ibs_op
.ibs_op_data1_high
= high
;
241 rdmsr(MSR_AMD64_IBSOPDATA2
, low
, high
);
242 ibs_op
.ibs_op_data2_low
= low
;
243 ibs_op
.ibs_op_data2_high
= high
;
244 rdmsr(MSR_AMD64_IBSOPDATA3
, low
, high
);
245 ibs_op
.ibs_op_data3_low
= low
;
246 ibs_op
.ibs_op_data3_high
= high
;
247 rdmsr(MSR_AMD64_IBSDCLINAD
, low
, high
);
248 ibs_op
.ibs_dc_linear_low
= low
;
249 ibs_op
.ibs_dc_linear_high
= high
;
250 rdmsr(MSR_AMD64_IBSDCPHYSAD
, low
, high
);
251 ibs_op
.ibs_dc_phys_low
= low
;
252 ibs_op
.ibs_dc_phys_high
= high
;
254 /* reenable the IRQ */
255 oprofile_add_ibs_sample(regs
,
256 (unsigned int *)&ibs_op
,
258 rdmsr(MSR_AMD64_IBSOPCTL
, low
, high
);
260 low
&= ~IBS_OP_LOW_VALID_BIT
;
261 low
|= IBS_OP_LOW_ENABLE
;
262 wrmsr(MSR_AMD64_IBSOPCTL
, low
, high
);
271 static int op_amd_check_ctrs(struct pt_regs
* const regs
,
272 struct op_msrs
const * const msrs
)
274 unsigned int low
, high
;
277 for (i
= 0 ; i
< NUM_COUNTERS
; ++i
) {
280 CTR_READ(low
, high
, msrs
, i
);
281 if (CTR_OVERFLOWED(low
)) {
282 oprofile_add_sample(regs
, i
);
283 CTR_WRITE(reset_value
[i
], msrs
, i
);
287 #ifdef CONFIG_OPROFILE_IBS
288 op_amd_handle_ibs(regs
, msrs
);
291 /* See op_model_ppro.c */
295 static void op_amd_start(struct op_msrs
const * const msrs
)
297 unsigned int low
, high
;
299 for (i
= 0 ; i
< NUM_COUNTERS
; ++i
) {
300 if (reset_value
[i
]) {
301 CTRL_READ(low
, high
, msrs
, i
);
302 CTRL_SET_ACTIVE(low
);
303 CTRL_WRITE(low
, high
, msrs
, i
);
307 #ifdef CONFIG_OPROFILE_IBS
308 if (has_ibs
&& ibs_config
.fetch_enabled
) {
309 low
= (ibs_config
.max_cnt_fetch
>> 4) & 0xFFFF;
310 high
= ((ibs_config
.rand_en
& 0x1) << 25) /* bit 57 */
311 + IBS_FETCH_HIGH_ENABLE
;
312 wrmsr(MSR_AMD64_IBSFETCHCTL
, low
, high
);
315 if (has_ibs
&& ibs_config
.op_enabled
) {
316 low
= ((ibs_config
.max_cnt_op
>> 4) & 0xFFFF)
317 + ((ibs_config
.dispatched_ops
& 0x1) << 19) /* bit 19 */
320 wrmsr(MSR_AMD64_IBSOPCTL
, low
, high
);
326 static void op_amd_stop(struct op_msrs
const * const msrs
)
328 unsigned int low
, high
;
332 * Subtle: stop on all counters to avoid race with setting our
335 for (i
= 0 ; i
< NUM_COUNTERS
; ++i
) {
338 CTRL_READ(low
, high
, msrs
, i
);
339 CTRL_SET_INACTIVE(low
);
340 CTRL_WRITE(low
, high
, msrs
, i
);
343 #ifdef CONFIG_OPROFILE_IBS
344 if (has_ibs
&& ibs_config
.fetch_enabled
) {
345 /* clear max count and enable */
348 wrmsr(MSR_AMD64_IBSFETCHCTL
, low
, high
);
351 if (has_ibs
&& ibs_config
.op_enabled
) {
352 /* clear max count and enable */
355 wrmsr(MSR_AMD64_IBSOPCTL
, low
, high
);
360 static void op_amd_shutdown(struct op_msrs
const * const msrs
)
364 for (i
= 0 ; i
< NUM_COUNTERS
; ++i
) {
365 if (CTR_IS_RESERVED(msrs
, i
))
366 release_perfctr_nmi(MSR_K7_PERFCTR0
+ i
);
368 for (i
= 0 ; i
< NUM_CONTROLS
; ++i
) {
369 if (CTRL_IS_RESERVED(msrs
, i
))
370 release_evntsel_nmi(MSR_K7_EVNTSEL0
+ i
);
374 #ifdef CONFIG_OPROFILE_IBS
376 static u8 ibs_eilvt_off
;
378 static inline void apic_init_ibs_nmi_per_cpu(void *arg
)
380 ibs_eilvt_off
= setup_APIC_eilvt_ibs(0, APIC_EILVT_MSG_NMI
, 0);
383 static inline void apic_clear_ibs_nmi_per_cpu(void *arg
)
385 setup_APIC_eilvt_ibs(0, APIC_EILVT_MSG_FIX
, 1);
388 static int init_ibs_nmi(void)
390 #define IBSCTL_LVTOFFSETVAL (1 << 8)
392 struct pci_dev
*cpu_cfg
;
397 on_each_cpu(apic_init_ibs_nmi_per_cpu
, NULL
, 1);
402 cpu_cfg
= pci_get_device(PCI_VENDOR_ID_AMD
,
403 PCI_DEVICE_ID_AMD_10H_NB_MISC
,
408 pci_write_config_dword(cpu_cfg
, IBSCTL
, ibs_eilvt_off
409 | IBSCTL_LVTOFFSETVAL
);
410 pci_read_config_dword(cpu_cfg
, IBSCTL
, &value
);
411 if (value
!= (ibs_eilvt_off
| IBSCTL_LVTOFFSETVAL
)) {
412 pci_dev_put(cpu_cfg
);
413 printk(KERN_DEBUG
"Failed to setup IBS LVT offset, "
414 "IBSCTL = 0x%08x", value
);
420 printk(KERN_DEBUG
"No CPU node configured for IBS");
426 /* Works only for 64bit with proper numa implementation. */
427 if (nodes
!= num_possible_nodes()) {
428 printk(KERN_DEBUG
"Failed to setup CPU node(s) for IBS, "
429 "found: %d, expected %d",
430 nodes
, num_possible_nodes());
437 /* uninitialize the APIC for the IBS interrupts if needed */
438 static void clear_ibs_nmi(void)
441 on_each_cpu(apic_clear_ibs_nmi_per_cpu
, NULL
, 1);
444 /* initialize the APIC for the IBS interrupts if available */
445 static void ibs_init(void)
447 has_ibs
= boot_cpu_has(X86_FEATURE_IBS
);
452 if (init_ibs_nmi()) {
457 printk(KERN_INFO
"oprofile: AMD IBS detected\n");
460 static void ibs_exit(void)
468 static int (*create_arch_files
)(struct super_block
*sb
, struct dentry
*root
);
470 static int setup_ibs_files(struct super_block
*sb
, struct dentry
*root
)
475 /* architecture specific files */
476 if (create_arch_files
)
477 ret
= create_arch_files(sb
, root
);
485 /* model specific files */
487 /* setup some reasonable defaults */
488 ibs_config
.max_cnt_fetch
= 250000;
489 ibs_config
.fetch_enabled
= 0;
490 ibs_config
.max_cnt_op
= 250000;
491 ibs_config
.op_enabled
= 0;
492 ibs_config
.dispatched_ops
= 1;
494 dir
= oprofilefs_mkdir(sb
, root
, "ibs_fetch");
495 oprofilefs_create_ulong(sb
, dir
, "enable",
496 &ibs_config
.fetch_enabled
);
497 oprofilefs_create_ulong(sb
, dir
, "max_count",
498 &ibs_config
.max_cnt_fetch
);
499 oprofilefs_create_ulong(sb
, dir
, "rand_enable",
500 &ibs_config
.rand_en
);
502 dir
= oprofilefs_mkdir(sb
, root
, "ibs_op");
503 oprofilefs_create_ulong(sb
, dir
, "enable",
504 &ibs_config
.op_enabled
);
505 oprofilefs_create_ulong(sb
, dir
, "max_count",
506 &ibs_config
.max_cnt_op
);
507 oprofilefs_create_ulong(sb
, dir
, "dispatched_ops",
508 &ibs_config
.dispatched_ops
);
513 static int op_amd_init(struct oprofile_operations
*ops
)
516 create_arch_files
= ops
->create_files
;
517 ops
->create_files
= setup_ibs_files
;
521 static void op_amd_exit(void)
530 static int op_amd_init(struct oprofile_operations
*ops
)
535 static void op_amd_exit(void) {}
537 #endif /* CONFIG_OPROFILE_IBS */
539 struct op_x86_model_spec
const op_amd_spec
= {
542 .num_counters
= NUM_COUNTERS
,
543 .num_controls
= NUM_CONTROLS
,
544 .fill_in_addresses
= &op_amd_fill_in_addresses
,
545 .setup_ctrs
= &op_amd_setup_ctrs
,
546 .check_ctrs
= &op_amd_check_ctrs
,
547 .start
= &op_amd_start
,
548 .stop
= &op_amd_stop
,
549 .shutdown
= &op_amd_shutdown