aboutsummaryrefslogtreecommitdiffstats
path: root/arch/parisc/kernel/perf.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
commit1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch)
tree0bba044c4ce775e45a88a51686b5d9f90697ea9d /arch/parisc/kernel/perf.c
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
Diffstat (limited to 'arch/parisc/kernel/perf.c')
-rw-r--r--arch/parisc/kernel/perf.c841
1 files changed, 841 insertions, 0 deletions
diff --git a/arch/parisc/kernel/perf.c b/arch/parisc/kernel/perf.c
new file mode 100644
index 000000000000..b3ad0a505b87
--- /dev/null
+++ b/arch/parisc/kernel/perf.c
@@ -0,0 +1,841 @@
1/*
2 * Parisc performance counters
3 * Copyright (C) 2001 Randolph Chung <tausq@debian.org>
4 *
5 * This code is derived, with permission, from HP/UX sources.
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2, or (at your option)
10 * any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 */
21
22/*
23 * Edited comment from original sources:
24 *
25 * This driver programs the PCX-U/PCX-W performance counters
26 * on the PA-RISC 2.0 chips. The driver keeps all images now
27 * internally to the kernel to hopefully eliminate the possiblity
28 * of a bad image halting the CPU. Also, there are different
29 * images for the PCX-W and later chips vs the PCX-U chips.
30 *
31 * Only 1 process is allowed to access the driver at any time,
32 * so the only protection that is needed is at open and close.
33 * A variable "perf_enabled" is used to hold the state of the
34 * driver. The spinlock "perf_lock" is used to protect the
35 * modification of the state during open/close operations so
36 * multiple processes don't get into the driver simultaneously.
37 *
38 * This driver accesses the processor directly vs going through
39 * the PDC INTRIGUE calls. This is done to eliminate bugs introduced
40 * in various PDC revisions. The code is much more maintainable
41 * and reliable this way vs having to debug on every version of PDC
42 * on every box.
43 */
44
45#include <linux/init.h>
46#include <linux/proc_fs.h>
47#include <linux/miscdevice.h>
48#include <linux/spinlock.h>
49
50#include <asm/uaccess.h>
51#include <asm/perf.h>
52#include <asm/parisc-device.h>
53#include <asm/processor.h>
54#include <asm/runway.h>
55#include <asm/io.h> /* for __raw_read() */
56
57#include "perf_images.h"
58
59#define MAX_RDR_WORDS 24
60#define PERF_VERSION 2 /* derived from hpux's PI v2 interface */
61
62/* definition of RDR regs */
63struct rdr_tbl_ent {
64 uint16_t width;
65 uint8_t num_words;
66 uint8_t write_control;
67};
68
69static int perf_processor_interface = UNKNOWN_INTF;
70static int perf_enabled = 0;
71static spinlock_t perf_lock;
72struct parisc_device *cpu_device = NULL;
73
74/* RDRs to write for PCX-W */
75static int perf_rdrs_W[] =
76 { 0, 1, 4, 5, 6, 15, 16, 17, 18, 20, 21, 22, 23, 24, 25, -1 };
77
78/* RDRs to write for PCX-U */
79static int perf_rdrs_U[] =
80 { 0, 1, 4, 5, 6, 7, 16, 17, 18, 20, 21, 22, 23, 24, 25, -1 };
81
82/* RDR register descriptions for PCX-W */
83static struct rdr_tbl_ent perf_rdr_tbl_W[] = {
84 { 19, 1, 8 }, /* RDR 0 */
85 { 16, 1, 16 }, /* RDR 1 */
86 { 72, 2, 0 }, /* RDR 2 */
87 { 81, 2, 0 }, /* RDR 3 */
88 { 328, 6, 0 }, /* RDR 4 */
89 { 160, 3, 0 }, /* RDR 5 */
90 { 336, 6, 0 }, /* RDR 6 */
91 { 164, 3, 0 }, /* RDR 7 */
92 { 0, 0, 0 }, /* RDR 8 */
93 { 35, 1, 0 }, /* RDR 9 */
94 { 6, 1, 0 }, /* RDR 10 */
95 { 18, 1, 0 }, /* RDR 11 */
96 { 13, 1, 0 }, /* RDR 12 */
97 { 8, 1, 0 }, /* RDR 13 */
98 { 8, 1, 0 }, /* RDR 14 */
99 { 8, 1, 0 }, /* RDR 15 */
100 { 1530, 24, 0 }, /* RDR 16 */
101 { 16, 1, 0 }, /* RDR 17 */
102 { 4, 1, 0 }, /* RDR 18 */
103 { 0, 0, 0 }, /* RDR 19 */
104 { 152, 3, 24 }, /* RDR 20 */
105 { 152, 3, 24 }, /* RDR 21 */
106 { 233, 4, 48 }, /* RDR 22 */
107 { 233, 4, 48 }, /* RDR 23 */
108 { 71, 2, 0 }, /* RDR 24 */
109 { 71, 2, 0 }, /* RDR 25 */
110 { 11, 1, 0 }, /* RDR 26 */
111 { 18, 1, 0 }, /* RDR 27 */
112 { 128, 2, 0 }, /* RDR 28 */
113 { 0, 0, 0 }, /* RDR 29 */
114 { 16, 1, 0 }, /* RDR 30 */
115 { 16, 1, 0 }, /* RDR 31 */
116};
117
118/* RDR register descriptions for PCX-U */
119static struct rdr_tbl_ent perf_rdr_tbl_U[] = {
120 { 19, 1, 8 }, /* RDR 0 */
121 { 32, 1, 16 }, /* RDR 1 */
122 { 20, 1, 0 }, /* RDR 2 */
123 { 0, 0, 0 }, /* RDR 3 */
124 { 344, 6, 0 }, /* RDR 4 */
125 { 176, 3, 0 }, /* RDR 5 */
126 { 336, 6, 0 }, /* RDR 6 */
127 { 0, 0, 0 }, /* RDR 7 */
128 { 0, 0, 0 }, /* RDR 8 */
129 { 0, 0, 0 }, /* RDR 9 */
130 { 28, 1, 0 }, /* RDR 10 */
131 { 33, 1, 0 }, /* RDR 11 */
132 { 0, 0, 0 }, /* RDR 12 */
133 { 230, 4, 0 }, /* RDR 13 */
134 { 32, 1, 0 }, /* RDR 14 */
135 { 128, 2, 0 }, /* RDR 15 */
136 { 1494, 24, 0 }, /* RDR 16 */
137 { 18, 1, 0 }, /* RDR 17 */
138 { 4, 1, 0 }, /* RDR 18 */
139 { 0, 0, 0 }, /* RDR 19 */
140 { 158, 3, 24 }, /* RDR 20 */
141 { 158, 3, 24 }, /* RDR 21 */
142 { 194, 4, 48 }, /* RDR 22 */
143 { 194, 4, 48 }, /* RDR 23 */
144 { 71, 2, 0 }, /* RDR 24 */
145 { 71, 2, 0 }, /* RDR 25 */
146 { 28, 1, 0 }, /* RDR 26 */
147 { 33, 1, 0 }, /* RDR 27 */
148 { 88, 2, 0 }, /* RDR 28 */
149 { 32, 1, 0 }, /* RDR 29 */
150 { 24, 1, 0 }, /* RDR 30 */
151 { 16, 1, 0 }, /* RDR 31 */
152};
153
154/*
155 * A non-zero write_control in the above tables is a byte offset into
156 * this array.
157 */
158static uint64_t perf_bitmasks[] = {
159 0x0000000000000000ul, /* first dbl word must be zero */
160 0xfdffe00000000000ul, /* RDR0 bitmask */
161 0x003f000000000000ul, /* RDR1 bitmask */
162 0x00fffffffffffffful, /* RDR20-RDR21 bitmask (152 bits) */
163 0xfffffffffffffffful,
164 0xfffffffc00000000ul,
165 0xfffffffffffffffful, /* RDR22-RDR23 bitmask (233 bits) */
166 0xfffffffffffffffful,
167 0xfffffffffffffffcul,
168 0xff00000000000000ul
169};
170
171/*
172 * Write control bitmasks for Pa-8700 processor given
173 * somethings have changed slightly.
174 */
175static uint64_t perf_bitmasks_piranha[] = {
176 0x0000000000000000ul, /* first dbl word must be zero */
177 0xfdffe00000000000ul, /* RDR0 bitmask */
178 0x003f000000000000ul, /* RDR1 bitmask */
179 0x00fffffffffffffful, /* RDR20-RDR21 bitmask (158 bits) */
180 0xfffffffffffffffful,
181 0xfffffffc00000000ul,
182 0xfffffffffffffffful, /* RDR22-RDR23 bitmask (210 bits) */
183 0xfffffffffffffffful,
184 0xfffffffffffffffful,
185 0xfffc000000000000ul
186};
187
188static uint64_t *bitmask_array; /* array of bitmasks to use */
189
190/******************************************************************************
191 * Function Prototypes
192 *****************************************************************************/
193static int perf_config(uint32_t *image_ptr);
194static int perf_release(struct inode *inode, struct file *file);
195static int perf_open(struct inode *inode, struct file *file);
196static ssize_t perf_read(struct file *file, char __user *buf, size_t cnt, loff_t *ppos);
197static ssize_t perf_write(struct file *file, const char __user *buf, size_t count,
198 loff_t *ppos);
199static int perf_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
200 unsigned long arg);
201static void perf_start_counters(void);
202static int perf_stop_counters(uint32_t *raddr);
203static struct rdr_tbl_ent * perf_rdr_get_entry(uint32_t rdr_num);
204static int perf_rdr_read_ubuf(uint32_t rdr_num, uint64_t *buffer);
205static int perf_rdr_clear(uint32_t rdr_num);
206static int perf_write_image(uint64_t *memaddr);
207static void perf_rdr_write(uint32_t rdr_num, uint64_t *buffer);
208
209/* External Assembly Routines */
210extern uint64_t perf_rdr_shift_in_W (uint32_t rdr_num, uint16_t width);
211extern uint64_t perf_rdr_shift_in_U (uint32_t rdr_num, uint16_t width);
212extern void perf_rdr_shift_out_W (uint32_t rdr_num, uint64_t buffer);
213extern void perf_rdr_shift_out_U (uint32_t rdr_num, uint64_t buffer);
214extern void perf_intrigue_enable_perf_counters (void);
215extern void perf_intrigue_disable_perf_counters (void);
216
217/******************************************************************************
218 * Function Definitions
219 *****************************************************************************/
220
221
222/*
223 * configure:
224 *
225 * Configure the cpu with a given data image. First turn off the counters,
226 * then download the image, then turn the counters back on.
227 */
228static int perf_config(uint32_t *image_ptr)
229{
230 long error;
231 uint32_t raddr[4];
232
233 /* Stop the counters*/
234 error = perf_stop_counters(raddr);
235 if (error != 0) {
236 printk("perf_config: perf_stop_counters = %ld\n", error);
237 return -EINVAL;
238 }
239
240printk("Preparing to write image\n");
241 /* Write the image to the chip */
242 error = perf_write_image((uint64_t *)image_ptr);
243 if (error != 0) {
244 printk("perf_config: DOWNLOAD = %ld\n", error);
245 return -EINVAL;
246 }
247
248printk("Preparing to start counters\n");
249
250 /* Start the counters */
251 perf_start_counters();
252
253 return sizeof(uint32_t);
254}
255
256/*
257 * Open the device and initialize all of its memory. The device is only
258 * opened once, but can be "queried" by multiple processes that know its
259 * file descriptor.
260 */
261static int perf_open(struct inode *inode, struct file *file)
262{
263 spin_lock(&perf_lock);
264 if (perf_enabled) {
265 spin_unlock(&perf_lock);
266 return -EBUSY;
267 }
268 perf_enabled = 1;
269 spin_unlock(&perf_lock);
270
271 return 0;
272}
273
274/*
275 * Close the device.
276 */
277static int perf_release(struct inode *inode, struct file *file)
278{
279 spin_lock(&perf_lock);
280 perf_enabled = 0;
281 spin_unlock(&perf_lock);
282
283 return 0;
284}
285
286/*
287 * Read does nothing for this driver
288 */
289static ssize_t perf_read(struct file *file, char __user *buf, size_t cnt, loff_t *ppos)
290{
291 return 0;
292}
293
294/*
295 * write:
296 *
297 * This routine downloads the image to the chip. It must be
298 * called on the processor that the download should happen
299 * on.
300 */
301static ssize_t perf_write(struct file *file, const char __user *buf, size_t count,
302 loff_t *ppos)
303{
304 int err;
305 size_t image_size;
306 uint32_t image_type;
307 uint32_t interface_type;
308 uint32_t test;
309
310 if (perf_processor_interface == ONYX_INTF)
311 image_size = PCXU_IMAGE_SIZE;
312 else if (perf_processor_interface == CUDA_INTF)
313 image_size = PCXW_IMAGE_SIZE;
314 else
315 return -EFAULT;
316
317 if (!capable(CAP_SYS_ADMIN))
318 return -EACCES;
319
320 if (count != sizeof(uint32_t))
321 return -EIO;
322
323 if ((err = copy_from_user(&image_type, buf, sizeof(uint32_t))) != 0)
324 return err;
325
326 /* Get the interface type and test type */
327 interface_type = (image_type >> 16) & 0xffff;
328 test = (image_type & 0xffff);
329
330 /* Make sure everything makes sense */
331
332 /* First check the machine type is correct for
333 the requested image */
334 if (((perf_processor_interface == CUDA_INTF) &&
335 (interface_type != CUDA_INTF)) ||
336 ((perf_processor_interface == ONYX_INTF) &&
337 (interface_type != ONYX_INTF)))
338 return -EINVAL;
339
340 /* Next check to make sure the requested image
341 is valid */
342 if (((interface_type == CUDA_INTF) &&
343 (test >= MAX_CUDA_IMAGES)) ||
344 ((interface_type == ONYX_INTF) &&
345 (test >= MAX_ONYX_IMAGES)))
346 return -EINVAL;
347
348 /* Copy the image into the processor */
349 if (interface_type == CUDA_INTF)
350 return perf_config(cuda_images[test]);
351 else
352 return perf_config(onyx_images[test]);
353
354 return count;
355}
356
357/*
358 * Patch the images that need to know the IVA addresses.
359 */
360static void perf_patch_images(void)
361{
362#if 0 /* FIXME!! */
363/*
364 * NOTE: this routine is VERY specific to the current TLB image.
365 * If the image is changed, this routine might also need to be changed.
366 */
367 extern void $i_itlb_miss_2_0();
368 extern void $i_dtlb_miss_2_0();
369 extern void PA2_0_iva();
370
371 /*
372 * We can only use the lower 32-bits, the upper 32-bits should be 0
373 * anyway given this is in the kernel
374 */
375 uint32_t itlb_addr = (uint32_t)&($i_itlb_miss_2_0);
376 uint32_t dtlb_addr = (uint32_t)&($i_dtlb_miss_2_0);
377 uint32_t IVAaddress = (uint32_t)&PA2_0_iva;
378
379 if (perf_processor_interface == ONYX_INTF) {
380 /* clear last 2 bytes */
381 onyx_images[TLBMISS][15] &= 0xffffff00;
382 /* set 2 bytes */
383 onyx_images[TLBMISS][15] |= (0x000000ff&((dtlb_addr) >> 24));
384 onyx_images[TLBMISS][16] = (dtlb_addr << 8)&0xffffff00;
385 onyx_images[TLBMISS][17] = itlb_addr;
386
387 /* clear last 2 bytes */
388 onyx_images[TLBHANDMISS][15] &= 0xffffff00;
389 /* set 2 bytes */
390 onyx_images[TLBHANDMISS][15] |= (0x000000ff&((dtlb_addr) >> 24));
391 onyx_images[TLBHANDMISS][16] = (dtlb_addr << 8)&0xffffff00;
392 onyx_images[TLBHANDMISS][17] = itlb_addr;
393
394 /* clear last 2 bytes */
395 onyx_images[BIG_CPI][15] &= 0xffffff00;
396 /* set 2 bytes */
397 onyx_images[BIG_CPI][15] |= (0x000000ff&((dtlb_addr) >> 24));
398 onyx_images[BIG_CPI][16] = (dtlb_addr << 8)&0xffffff00;
399 onyx_images[BIG_CPI][17] = itlb_addr;
400
401 onyx_images[PANIC][15] &= 0xffffff00; /* clear last 2 bytes */
402 onyx_images[PANIC][15] |= (0x000000ff&((IVAaddress) >> 24)); /* set 2 bytes */
403 onyx_images[PANIC][16] = (IVAaddress << 8)&0xffffff00;
404
405
406 } else if (perf_processor_interface == CUDA_INTF) {
407 /* Cuda interface */
408 cuda_images[TLBMISS][16] =
409 (cuda_images[TLBMISS][16]&0xffff0000) |
410 ((dtlb_addr >> 8)&0x0000ffff);
411 cuda_images[TLBMISS][17] =
412 ((dtlb_addr << 24)&0xff000000) | ((itlb_addr >> 16)&0x000000ff);
413 cuda_images[TLBMISS][18] = (itlb_addr << 16)&0xffff0000;
414
415 cuda_images[TLBHANDMISS][16] =
416 (cuda_images[TLBHANDMISS][16]&0xffff0000) |
417 ((dtlb_addr >> 8)&0x0000ffff);
418 cuda_images[TLBHANDMISS][17] =
419 ((dtlb_addr << 24)&0xff000000) | ((itlb_addr >> 16)&0x000000ff);
420 cuda_images[TLBHANDMISS][18] = (itlb_addr << 16)&0xffff0000;
421
422 cuda_images[BIG_CPI][16] =
423 (cuda_images[BIG_CPI][16]&0xffff0000) |
424 ((dtlb_addr >> 8)&0x0000ffff);
425 cuda_images[BIG_CPI][17] =
426 ((dtlb_addr << 24)&0xff000000) | ((itlb_addr >> 16)&0x000000ff);
427 cuda_images[BIG_CPI][18] = (itlb_addr << 16)&0xffff0000;
428 } else {
429 /* Unknown type */
430 }
431#endif
432}
433
434
435/*
436 * ioctl routine
437 * All routines effect the processor that they are executed on. Thus you
438 * must be running on the processor that you wish to change.
439 */
440
441static int perf_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
442 unsigned long arg)
443{
444 long error_start;
445 uint32_t raddr[4];
446
447 switch (cmd) {
448
449 case PA_PERF_ON:
450 /* Start the counters */
451 perf_start_counters();
452 return 0;
453
454 case PA_PERF_OFF:
455 error_start = perf_stop_counters(raddr);
456 if (error_start != 0) {
457 printk(KERN_ERR "perf_off: perf_stop_counters = %ld\n", error_start);
458 return -EFAULT;
459 }
460
461 /* copy out the Counters */
462 if (copy_to_user((void __user *)arg, raddr,
463 sizeof (raddr)) != 0) {
464 return -EFAULT;
465 }
466 return 0;
467
468 case PA_PERF_VERSION:
469 /* Return the version # */
470 return put_user(PERF_VERSION, (int *)arg);
471
472 default:
473 break;
474 }
475 return -ENOTTY;
476}
477
478static struct file_operations perf_fops = {
479 .llseek = no_llseek,
480 .read = perf_read,
481 .write = perf_write,
482 .ioctl = perf_ioctl,
483 .open = perf_open,
484 .release = perf_release
485};
486
487static struct miscdevice perf_dev = {
488 MISC_DYNAMIC_MINOR,
489 PA_PERF_DEV,
490 &perf_fops
491};
492
493/*
494 * Initialize the module
495 */
496static int __init perf_init(void)
497{
498 int ret;
499
500 /* Determine correct processor interface to use */
501 bitmask_array = perf_bitmasks;
502
503 if (boot_cpu_data.cpu_type == pcxu ||
504 boot_cpu_data.cpu_type == pcxu_) {
505 perf_processor_interface = ONYX_INTF;
506 } else if (boot_cpu_data.cpu_type == pcxw ||
507 boot_cpu_data.cpu_type == pcxw_ ||
508 boot_cpu_data.cpu_type == pcxw2 ||
509 boot_cpu_data.cpu_type == mako) {
510 perf_processor_interface = CUDA_INTF;
511 if (boot_cpu_data.cpu_type == pcxw2 ||
512 boot_cpu_data.cpu_type == mako)
513 bitmask_array = perf_bitmasks_piranha;
514 } else {
515 perf_processor_interface = UNKNOWN_INTF;
516 printk("Performance monitoring counters not supported on this processor\n");
517 return -ENODEV;
518 }
519
520 ret = misc_register(&perf_dev);
521 if (ret) {
522 printk(KERN_ERR "Performance monitoring counters: "
523 "cannot register misc device.\n");
524 return ret;
525 }
526
527 /* Patch the images to match the system */
528 perf_patch_images();
529
530 spin_lock_init(&perf_lock);
531
532 /* TODO: this only lets us access the first cpu.. what to do for SMP? */
533 cpu_device = cpu_data[0].dev;
534 printk("Performance monitoring counters enabled for %s\n",
535 cpu_data[0].dev->name);
536
537 return 0;
538}
539
540/*
541 * perf_start_counters(void)
542 *
543 * Start the counters.
544 */
545static void perf_start_counters(void)
546{
547 /* Enable performance monitor counters */
548 perf_intrigue_enable_perf_counters();
549}
550
551/*
552 * perf_stop_counters
553 *
554 * Stop the performance counters and save counts
555 * in a per_processor array.
556 */
557static int perf_stop_counters(uint32_t *raddr)
558{
559 uint64_t userbuf[MAX_RDR_WORDS];
560
561 /* Disable performance counters */
562 perf_intrigue_disable_perf_counters();
563
564 if (perf_processor_interface == ONYX_INTF) {
565 uint64_t tmp64;
566 /*
567 * Read the counters
568 */
569 if (!perf_rdr_read_ubuf(16, userbuf))
570 return -13;
571
572 /* Counter0 is bits 1398 thru 1429 */
573 tmp64 = (userbuf[21] << 22) & 0x00000000ffc00000;
574 tmp64 |= (userbuf[22] >> 42) & 0x00000000003fffff;
575 /* OR sticky0 (bit 1430) to counter0 bit 32 */
576 tmp64 |= (userbuf[22] >> 10) & 0x0000000080000000;
577 raddr[0] = (uint32_t)tmp64;
578
579 /* Counter1 is bits 1431 thru 1462 */
580 tmp64 = (userbuf[22] >> 9) & 0x00000000ffffffff;
581 /* OR sticky1 (bit 1463) to counter1 bit 32 */
582 tmp64 |= (userbuf[22] << 23) & 0x0000000080000000;
583 raddr[1] = (uint32_t)tmp64;
584
585 /* Counter2 is bits 1464 thru 1495 */
586 tmp64 = (userbuf[22] << 24) & 0x00000000ff000000;
587 tmp64 |= (userbuf[23] >> 40) & 0x0000000000ffffff;
588 /* OR sticky2 (bit 1496) to counter2 bit 32 */
589 tmp64 |= (userbuf[23] >> 8) & 0x0000000080000000;
590 raddr[2] = (uint32_t)tmp64;
591
592 /* Counter3 is bits 1497 thru 1528 */
593 tmp64 = (userbuf[23] >> 7) & 0x00000000ffffffff;
594 /* OR sticky3 (bit 1529) to counter3 bit 32 */
595 tmp64 |= (userbuf[23] << 25) & 0x0000000080000000;
596 raddr[3] = (uint32_t)tmp64;
597
598 /*
599 * Zero out the counters
600 */
601
602 /*
603 * The counters and sticky-bits comprise the last 132 bits
604 * (1398 - 1529) of RDR16 on a U chip. We'll zero these
605 * out the easy way: zero out last 10 bits of dword 21,
606 * all of dword 22 and 58 bits (plus 6 don't care bits) of
607 * dword 23.
608 */
609 userbuf[21] &= 0xfffffffffffffc00ul; /* 0 to last 10 bits */
610 userbuf[22] = 0;
611 userbuf[23] = 0;
612
613 /*
614 * Write back the zero'ed bytes + the image given
615 * the read was destructive.
616 */
617 perf_rdr_write(16, userbuf);
618 } else {
619
620 /*
621 * Read RDR-15 which contains the counters and sticky bits
622 */
623 if (!perf_rdr_read_ubuf(15, userbuf)) {
624 return -13;
625 }
626
627 /*
628 * Clear out the counters
629 */
630 perf_rdr_clear(15);
631
632 /*
633 * Copy the counters
634 */
635 raddr[0] = (uint32_t)((userbuf[0] >> 32) & 0x00000000ffffffffUL);
636 raddr[1] = (uint32_t)(userbuf[0] & 0x00000000ffffffffUL);
637 raddr[2] = (uint32_t)((userbuf[1] >> 32) & 0x00000000ffffffffUL);
638 raddr[3] = (uint32_t)(userbuf[1] & 0x00000000ffffffffUL);
639 }
640
641 return 0;
642}
643
644/*
645 * perf_rdr_get_entry
646 *
647 * Retrieve a pointer to the description of what this
648 * RDR contains.
649 */
650static struct rdr_tbl_ent * perf_rdr_get_entry(uint32_t rdr_num)
651{
652 if (perf_processor_interface == ONYX_INTF) {
653 return &perf_rdr_tbl_U[rdr_num];
654 } else {
655 return &perf_rdr_tbl_W[rdr_num];
656 }
657}
658
659/*
660 * perf_rdr_read_ubuf
661 *
662 * Read the RDR value into the buffer specified.
663 */
664static int perf_rdr_read_ubuf(uint32_t rdr_num, uint64_t *buffer)
665{
666 uint64_t data, data_mask = 0;
667 uint32_t width, xbits, i;
668 struct rdr_tbl_ent *tentry;
669
670 tentry = perf_rdr_get_entry(rdr_num);
671 if ((width = tentry->width) == 0)
672 return 0;
673
674 /* Clear out buffer */
675 i = tentry->num_words;
676 while (i--) {
677 buffer[i] = 0;
678 }
679
680 /* Check for bits an even number of 64 */
681 if ((xbits = width & 0x03f) != 0) {
682 data_mask = 1;
683 data_mask <<= (64 - xbits);
684 data_mask--;
685 }
686
687 /* Grab all of the data */
688 i = tentry->num_words;
689 while (i--) {
690
691 if (perf_processor_interface == ONYX_INTF) {
692 data = perf_rdr_shift_in_U(rdr_num, width);
693 } else {
694 data = perf_rdr_shift_in_W(rdr_num, width);
695 }
696 if (xbits) {
697 buffer[i] |= (data << (64 - xbits));
698 if (i) {
699 buffer[i-1] |= ((data >> xbits) & data_mask);
700 }
701 } else {
702 buffer[i] = data;
703 }
704 }
705
706 return 1;
707}
708
709/*
710 * perf_rdr_clear
711 *
712 * Zero out the given RDR register
713 */
714static int perf_rdr_clear(uint32_t rdr_num)
715{
716 struct rdr_tbl_ent *tentry;
717 int32_t i;
718
719 tentry = perf_rdr_get_entry(rdr_num);
720
721 if (tentry->width == 0) {
722 return -1;
723 }
724
725 i = tentry->num_words;
726 while (i--) {
727 if (perf_processor_interface == ONYX_INTF) {
728 perf_rdr_shift_out_U(rdr_num, 0UL);
729 } else {
730 perf_rdr_shift_out_W(rdr_num, 0UL);
731 }
732 }
733
734 return 0;
735}
736
737
738/*
739 * perf_write_image
740 *
741 * Write the given image out to the processor
742 */
743static int perf_write_image(uint64_t *memaddr)
744{
745 uint64_t buffer[MAX_RDR_WORDS];
746 uint64_t *bptr;
747 uint32_t dwords;
748 uint32_t *intrigue_rdr;
749 uint64_t *intrigue_bitmask, tmp64, proc_hpa;
750 struct rdr_tbl_ent *tentry;
751 int i;
752
753 /* Clear out counters */
754 if (perf_processor_interface == ONYX_INTF) {
755
756 perf_rdr_clear(16);
757
758 /* Toggle performance monitor */
759 perf_intrigue_enable_perf_counters();
760 perf_intrigue_disable_perf_counters();
761
762 intrigue_rdr = perf_rdrs_U;
763 } else {
764 perf_rdr_clear(15);
765 intrigue_rdr = perf_rdrs_W;
766 }
767
768 /* Write all RDRs */
769 while (*intrigue_rdr != -1) {
770 tentry = perf_rdr_get_entry(*intrigue_rdr);
771 perf_rdr_read_ubuf(*intrigue_rdr, buffer);
772 bptr = &buffer[0];
773 dwords = tentry->num_words;
774 if (tentry->write_control) {
775 intrigue_bitmask = &bitmask_array[tentry->write_control >> 3];
776 while (dwords--) {
777 tmp64 = *intrigue_bitmask & *memaddr++;
778 tmp64 |= (~(*intrigue_bitmask++)) & *bptr;
779 *bptr++ = tmp64;
780 }
781 } else {
782 while (dwords--) {
783 *bptr++ = *memaddr++;
784 }
785 }
786
787 perf_rdr_write(*intrigue_rdr, buffer);
788 intrigue_rdr++;
789 }
790
791 /*
792 * Now copy out the Runway stuff which is not in RDRs
793 */
794
795 if (cpu_device == NULL)
796 {
797 printk(KERN_ERR "write_image: cpu_device not yet initialized!\n");
798 return -1;
799 }
800
801 proc_hpa = cpu_device->hpa;
802
803 /* Merge intrigue bits into Runway STATUS 0 */
804 tmp64 = __raw_readq(proc_hpa + RUNWAY_STATUS) & 0xffecfffffffffffful;
805 __raw_writeq(tmp64 | (*memaddr++ & 0x0013000000000000ul), proc_hpa + RUNWAY_STATUS);
806
807 /* Write RUNWAY DEBUG registers */
808 for (i = 0; i < 8; i++) {
809 __raw_writeq(*memaddr++, proc_hpa + RUNWAY_DEBUG + i);
810 }
811
812 return 0;
813}
814
815/*
816 * perf_rdr_write
817 *
818 * Write the given RDR register with the contents
819 * of the given buffer.
820 */
821static void perf_rdr_write(uint32_t rdr_num, uint64_t *buffer)
822{
823 struct rdr_tbl_ent *tentry;
824 int32_t i;
825
826printk("perf_rdr_write\n");
827 tentry = perf_rdr_get_entry(rdr_num);
828 if (tentry->width == 0) { return; }
829
830 i = tentry->num_words;
831 while (i--) {
832 if (perf_processor_interface == ONYX_INTF) {
833 perf_rdr_shift_out_U(rdr_num, buffer[i]);
834 } else {
835 perf_rdr_shift_out_W(rdr_num, buffer[i]);
836 }
837 }
838printk("perf_rdr_write done\n");
839}
840
841module_init(perf_init);
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-01 12:42:15 -0400