1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * RackMac vu-meter driver
4  *
5  * (c) Copyright 2006 Benjamin Herrenschmidt, IBM Corp.
6  *                    <benh@kernel.crashing.org>
7  *
8  * Support the CPU-meter LEDs of the Xserve G5
9  *
10  * TODO: Implement PWM to do variable intensity and provide userland
11  * interface for fun. Also, the CPU-meter could be made nicer by being
12  * a bit less "immediate" but giving instead a more average load over
13  * time. Patches welcome :-)
14  */
15 #undef DEBUG
16 
17 #include <linux/types.h>
18 #include <linux/kernel.h>
19 #include <linux/slab.h>
20 #include <linux/device.h>
21 #include <linux/interrupt.h>
22 #include <linux/module.h>
23 #include <linux/pci.h>
24 #include <linux/dma-mapping.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/of_address.h>
27 #include <linux/of_irq.h>
28 
29 #include <asm/io.h>
30 #include <asm/prom.h>
31 #include <asm/machdep.h>
32 #include <asm/pmac_feature.h>
33 #include <asm/dbdma.h>
34 #include <asm/macio.h>
35 #include <asm/keylargo.h>
36 
37 /* Number of samples in a sample buffer */
38 #define SAMPLE_COUNT		256
39 
40 /* CPU meter sampling rate in ms */
41 #define CPU_SAMPLING_RATE	250
42 
43 struct rackmeter_dma {
44 	struct dbdma_cmd	cmd[4]			____cacheline_aligned;
45 	u32			mark			____cacheline_aligned;
46 	u32			buf1[SAMPLE_COUNT]	____cacheline_aligned;
47 	u32			buf2[SAMPLE_COUNT]	____cacheline_aligned;
48 } ____cacheline_aligned;
49 
50 struct rackmeter_cpu {
51 	struct delayed_work	sniffer;
52 	struct rackmeter	*rm;
53 	u64			prev_wall;
54 	u64			prev_idle;
55 	int			zero;
56 } ____cacheline_aligned;
57 
58 struct rackmeter {
59 	struct macio_dev		*mdev;
60 	unsigned int			irq;
61 	struct device_node		*i2s;
62 	u8				*ubuf;
63 	struct dbdma_regs __iomem	*dma_regs;
64 	void __iomem			*i2s_regs;
65 	dma_addr_t			dma_buf_p;
66 	struct rackmeter_dma		*dma_buf_v;
67 	int				stale_irq;
68 	struct rackmeter_cpu		cpu[2];
69 	int				paused;
70 	struct mutex			sem;
71 };
72 
73 /* To be set as a tunable */
74 static int rackmeter_ignore_nice;
75 
76 /* This GPIO is whacked by the OS X driver when initializing */
77 #define RACKMETER_MAGIC_GPIO	0x78
78 
79 /* This is copied from cpufreq_ondemand, maybe we should put it in
80  * a common header somewhere
81  */
82 static inline u64 get_cpu_idle_time(unsigned int cpu)
83 {
84 	struct kernel_cpustat *kcpustat = &kcpustat_cpu(cpu);
85 	u64 retval;
86 
87 	retval = kcpustat->cpustat[CPUTIME_IDLE] +
88 		 kcpustat->cpustat[CPUTIME_IOWAIT];
89 
90 	if (rackmeter_ignore_nice)
91 		retval += kcpustat_field(kcpustat, CPUTIME_NICE, cpu);
92 
93 	return retval;
94 }
95 
96 static void rackmeter_setup_i2s(struct rackmeter *rm)
97 {
98 	struct macio_chip *macio = rm->mdev->bus->chip;
99 
100 	/* First whack magic GPIO */
101 	pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, RACKMETER_MAGIC_GPIO, 5);
102 
103 
104 	/* Call feature code to enable the sound channel and the proper
105 	 * clock sources
106 	 */
107 	pmac_call_feature(PMAC_FTR_SOUND_CHIP_ENABLE, rm->i2s, 0, 1);
108 
109 	/* Power i2s and stop i2s clock. We whack MacIO FCRs directly for now.
110 	 * This is a bit racy, thus we should add new platform functions to
111 	 * handle that. snd-aoa needs that too
112 	 */
113 	MACIO_BIS(KEYLARGO_FCR1, KL1_I2S0_ENABLE);
114 	MACIO_BIC(KEYLARGO_FCR1, KL1_I2S0_CLK_ENABLE_BIT);
115 	(void)MACIO_IN32(KEYLARGO_FCR1);
116 	udelay(10);
117 
118 	/* Then setup i2s. For now, we use the same magic value that
119 	 * the OS X driver seems to use. We might want to play around
120 	 * with the clock divisors later
121 	 */
122 	out_le32(rm->i2s_regs + 0x10, 0x01fa0000);
123 	(void)in_le32(rm->i2s_regs + 0x10);
124 	udelay(10);
125 
126 	/* Fully restart i2s*/
127 	MACIO_BIS(KEYLARGO_FCR1, KL1_I2S0_CELL_ENABLE |
128 		  KL1_I2S0_CLK_ENABLE_BIT);
129 	(void)MACIO_IN32(KEYLARGO_FCR1);
130 	udelay(10);
131 }
132 
133 static void rackmeter_set_default_pattern(struct rackmeter *rm)
134 {
135 	int i;
136 
137 	for (i = 0; i < 16; i++) {
138 		if (i < 8)
139 			rm->ubuf[i] = (i & 1) * 255;
140 		else
141 			rm->ubuf[i] = ((~i) & 1) * 255;
142 	}
143 }
144 
145 static void rackmeter_do_pause(struct rackmeter *rm, int pause)
146 {
147 	struct rackmeter_dma *rdma = rm->dma_buf_v;
148 
149 	pr_debug("rackmeter: %s\n", pause ? "paused" : "started");
150 
151 	rm->paused = pause;
152 	if (pause) {
153 		DBDMA_DO_STOP(rm->dma_regs);
154 		return;
155 	}
156 	memset(rdma->buf1, 0, sizeof(rdma->buf1));
157 	memset(rdma->buf2, 0, sizeof(rdma->buf2));
158 
159 	rm->dma_buf_v->mark = 0;
160 
161 	mb();
162 	out_le32(&rm->dma_regs->cmdptr_hi, 0);
163 	out_le32(&rm->dma_regs->cmdptr, rm->dma_buf_p);
164 	out_le32(&rm->dma_regs->control, (RUN << 16) | RUN);
165 }
166 
167 static void rackmeter_setup_dbdma(struct rackmeter *rm)
168 {
169 	struct rackmeter_dma *db = rm->dma_buf_v;
170 	struct dbdma_cmd *cmd = db->cmd;
171 
172 	/* Make sure dbdma is reset */
173 	DBDMA_DO_RESET(rm->dma_regs);
174 
175 	pr_debug("rackmeter: mark offset=0x%zx\n",
176 		 offsetof(struct rackmeter_dma, mark));
177 	pr_debug("rackmeter: buf1 offset=0x%zx\n",
178 		 offsetof(struct rackmeter_dma, buf1));
179 	pr_debug("rackmeter: buf2 offset=0x%zx\n",
180 		 offsetof(struct rackmeter_dma, buf2));
181 
182 	/* Prepare 4 dbdma commands for the 2 buffers */
183 	memset(cmd, 0, 4 * sizeof(struct dbdma_cmd));
184 	cmd->req_count = cpu_to_le16(4);
185 	cmd->command = cpu_to_le16(STORE_WORD | INTR_ALWAYS | KEY_SYSTEM);
186 	cmd->phy_addr = cpu_to_le32(rm->dma_buf_p +
187 		offsetof(struct rackmeter_dma, mark));
188 	cmd->cmd_dep = cpu_to_le32(0x02000000);
189 	cmd++;
190 
191 	cmd->req_count = cpu_to_le16(SAMPLE_COUNT * 4);
192 	cmd->command = cpu_to_le16(OUTPUT_MORE);
193 	cmd->phy_addr = cpu_to_le32(rm->dma_buf_p +
194 		offsetof(struct rackmeter_dma, buf1));
195 	cmd++;
196 
197 	cmd->req_count = cpu_to_le16(4);
198 	cmd->command = cpu_to_le16(STORE_WORD | INTR_ALWAYS | KEY_SYSTEM);
199 	cmd->phy_addr = cpu_to_le32(rm->dma_buf_p +
200 		offsetof(struct rackmeter_dma, mark));
201 	cmd->cmd_dep = cpu_to_le32(0x01000000);
202 	cmd++;
203 
204 	cmd->req_count = cpu_to_le16(SAMPLE_COUNT * 4);
205 	cmd->command = cpu_to_le16(OUTPUT_MORE | BR_ALWAYS);
206 	cmd->phy_addr = cpu_to_le32(rm->dma_buf_p +
207 		offsetof(struct rackmeter_dma, buf2));
208 	cmd->cmd_dep = cpu_to_le32(rm->dma_buf_p);
209 
210 	rackmeter_do_pause(rm, 0);
211 }
212 
213 static void rackmeter_do_timer(struct work_struct *work)
214 {
215 	struct rackmeter_cpu *rcpu =
216 		container_of(work, struct rackmeter_cpu, sniffer.work);
217 	struct rackmeter *rm = rcpu->rm;
218 	unsigned int cpu = smp_processor_id();
219 	u64 cur_nsecs, total_idle_nsecs;
220 	u64 total_nsecs, idle_nsecs;
221 	int i, offset, load, cumm, pause;
222 
223 	cur_nsecs = jiffies64_to_nsecs(get_jiffies_64());
224 	total_nsecs = cur_nsecs - rcpu->prev_wall;
225 	rcpu->prev_wall = cur_nsecs;
226 
227 	total_idle_nsecs = get_cpu_idle_time(cpu);
228 	idle_nsecs = total_idle_nsecs - rcpu->prev_idle;
229 	idle_nsecs = min(idle_nsecs, total_nsecs);
230 	rcpu->prev_idle = total_idle_nsecs;
231 
232 	/* We do a very dumb calculation to update the LEDs for now,
233 	 * we'll do better once we have actual PWM implemented
234 	 */
235 	load = div64_u64(9 * (total_nsecs - idle_nsecs), total_nsecs);
236 
237 	offset = cpu << 3;
238 	cumm = 0;
239 	for (i = 0; i < 8; i++) {
240 		u8 ub = (load > i) ? 0xff : 0;
241 		rm->ubuf[i + offset] = ub;
242 		cumm |= ub;
243 	}
244 	rcpu->zero = (cumm == 0);
245 
246 	/* Now check if LEDs are all 0, we can stop DMA */
247 	pause = (rm->cpu[0].zero && rm->cpu[1].zero);
248 	if (pause != rm->paused) {
249 		mutex_lock(&rm->sem);
250 		pause = (rm->cpu[0].zero && rm->cpu[1].zero);
251 		rackmeter_do_pause(rm, pause);
252 		mutex_unlock(&rm->sem);
253 	}
254 	schedule_delayed_work_on(cpu, &rcpu->sniffer,
255 				 msecs_to_jiffies(CPU_SAMPLING_RATE));
256 }
257 
258 static void rackmeter_init_cpu_sniffer(struct rackmeter *rm)
259 {
260 	unsigned int cpu;
261 
262 	/* This driver works only with 1 or 2 CPUs numbered 0 and 1,
263 	 * but that's really all we have on Apple Xserve. It doesn't
264 	 * play very nice with CPU hotplug neither but we don't do that
265 	 * on those machines yet
266 	 */
267 
268 	rm->cpu[0].rm = rm;
269 	INIT_DELAYED_WORK(&rm->cpu[0].sniffer, rackmeter_do_timer);
270 	rm->cpu[1].rm = rm;
271 	INIT_DELAYED_WORK(&rm->cpu[1].sniffer, rackmeter_do_timer);
272 
273 	for_each_online_cpu(cpu) {
274 		struct rackmeter_cpu *rcpu;
275 
276 		if (cpu > 1)
277 			continue;
278 		rcpu = &rm->cpu[cpu];
279 		rcpu->prev_idle = get_cpu_idle_time(cpu);
280 		rcpu->prev_wall = jiffies64_to_nsecs(get_jiffies_64());
281 		schedule_delayed_work_on(cpu, &rm->cpu[cpu].sniffer,
282 					 msecs_to_jiffies(CPU_SAMPLING_RATE));
283 	}
284 }
285 
286 static void rackmeter_stop_cpu_sniffer(struct rackmeter *rm)
287 {
288 	cancel_delayed_work_sync(&rm->cpu[0].sniffer);
289 	cancel_delayed_work_sync(&rm->cpu[1].sniffer);
290 }
291 
292 static int rackmeter_setup(struct rackmeter *rm)
293 {
294 	pr_debug("rackmeter: setting up i2s..\n");
295 	rackmeter_setup_i2s(rm);
296 
297 	pr_debug("rackmeter: setting up default pattern..\n");
298 	rackmeter_set_default_pattern(rm);
299 
300 	pr_debug("rackmeter: setting up dbdma..\n");
301 	rackmeter_setup_dbdma(rm);
302 
303 	pr_debug("rackmeter: start CPU measurements..\n");
304 	rackmeter_init_cpu_sniffer(rm);
305 
306 	printk(KERN_INFO "RackMeter initialized\n");
307 
308 	return 0;
309 }
310 
311 /*  XXX FIXME: No PWM yet, this is 0/1 */
312 static u32 rackmeter_calc_sample(struct rackmeter *rm, unsigned int index)
313 {
314 	int led;
315 	u32 sample = 0;
316 
317 	for (led = 0; led < 16; led++) {
318 		sample >>= 1;
319 		sample |= ((rm->ubuf[led] >= 0x80) << 15);
320 	}
321 	return (sample << 17) | (sample >> 15);
322 }
323 
324 static irqreturn_t rackmeter_irq(int irq, void *arg)
325 {
326 	struct rackmeter *rm = arg;
327 	struct rackmeter_dma *db = rm->dma_buf_v;
328 	unsigned int mark, i;
329 	u32 *buf;
330 
331 	/* Flush PCI buffers with an MMIO read. Maybe we could actually
332 	 * check the status one day ... in case things go wrong, though
333 	 * this never happened to me
334 	 */
335 	(void)in_le32(&rm->dma_regs->status);
336 
337 	/* Make sure the CPU gets us in order */
338 	rmb();
339 
340 	/* Read mark */
341 	mark = db->mark;
342 	if (mark != 1 && mark != 2) {
343 		printk(KERN_WARNING "rackmeter: Incorrect DMA mark 0x%08x\n",
344 		       mark);
345 		/* We allow for 3 errors like that (stale DBDMA irqs) */
346 		if (++rm->stale_irq > 3) {
347 			printk(KERN_ERR "rackmeter: Too many errors,"
348 			       " stopping DMA\n");
349 			DBDMA_DO_RESET(rm->dma_regs);
350 		}
351 		return IRQ_HANDLED;
352 	}
353 
354 	/* Next buffer we need to fill is mark value */
355 	buf = mark == 1 ? db->buf1 : db->buf2;
356 
357 	/* Fill it now. This routine converts the 8 bits depth sample array
358 	 * into the PWM bitmap for each LED.
359 	 */
360 	for (i = 0; i < SAMPLE_COUNT; i++)
361 		buf[i] = rackmeter_calc_sample(rm, i);
362 
363 
364 	return IRQ_HANDLED;
365 }
366 
367 static int rackmeter_probe(struct macio_dev* mdev,
368 			   const struct of_device_id *match)
369 {
370 	struct device_node *i2s = NULL, *np = NULL;
371 	struct rackmeter *rm = NULL;
372 	struct resource ri2s, rdma;
373 	int rc = -ENODEV;
374 
375 	pr_debug("rackmeter_probe()\n");
376 
377 	/* Get i2s-a node */
378 	for_each_child_of_node(mdev->ofdev.dev.of_node, i2s)
379 		if (of_node_name_eq(i2s, "i2s-a"))
380 			break;
381 
382 	if (i2s == NULL) {
383 		pr_debug("  i2s-a child not found\n");
384 		goto bail;
385 	}
386 	/* Get lightshow or virtual sound */
387 	for_each_child_of_node(i2s, np) {
388 	       if (of_node_name_eq(np, "lightshow"))
389 		       break;
390 	       if (of_node_name_eq(np, "sound") &&
391 		   of_get_property(np, "virtual", NULL) != NULL)
392 		       break;
393 	}
394 	if (np == NULL) {
395 		pr_debug("  lightshow or sound+virtual child not found\n");
396 		goto bail;
397 	}
398 
399 	/* Create and initialize our instance data */
400 	rm = kzalloc(sizeof(*rm), GFP_KERNEL);
401 	if (rm == NULL) {
402 		printk(KERN_ERR "rackmeter: failed to allocate memory !\n");
403 		rc = -ENOMEM;
404 		goto bail_release;
405 	}
406 	rm->mdev = mdev;
407 	rm->i2s = i2s;
408 	mutex_init(&rm->sem);
409 	dev_set_drvdata(&mdev->ofdev.dev, rm);
410 	/* Check resources availability. We need at least resource 0 and 1 */
411 #if 0 /* Use that when i2s-a is finally an mdev per-se */
412 	if (macio_resource_count(mdev) < 2 || macio_irq_count(mdev) < 2) {
413 		printk(KERN_ERR
414 		       "rackmeter: found match but lacks resources: %pOF"
415 		       " (%d resources, %d interrupts)\n",
416 		       mdev->ofdev.dev.of_node);
417 		rc = -ENXIO;
418 		goto bail_free;
419 	}
420 	if (macio_request_resources(mdev, "rackmeter")) {
421 		printk(KERN_ERR
422 		       "rackmeter: failed to request resources: %pOF\n",
423 		       mdev->ofdev.dev.of_node);
424 		rc = -EBUSY;
425 		goto bail_free;
426 	}
427 	rm->irq = macio_irq(mdev, 1);
428 #else
429 	rm->irq = irq_of_parse_and_map(i2s, 1);
430 	if (!rm->irq ||
431 	    of_address_to_resource(i2s, 0, &ri2s) ||
432 	    of_address_to_resource(i2s, 1, &rdma)) {
433 		printk(KERN_ERR
434 		       "rackmeter: found match but lacks resources: %pOF",
435 		       mdev->ofdev.dev.of_node);
436 		rc = -ENXIO;
437 		goto bail_free;
438 	}
439 #endif
440 
441 	pr_debug("  i2s @0x%08x\n", (unsigned int)ri2s.start);
442 	pr_debug("  dma @0x%08x\n", (unsigned int)rdma.start);
443 	pr_debug("  irq %d\n", rm->irq);
444 
445 	rm->ubuf = (u8 *)__get_free_page(GFP_KERNEL);
446 	if (rm->ubuf == NULL) {
447 		printk(KERN_ERR
448 		       "rackmeter: failed to allocate samples page !\n");
449 		rc = -ENOMEM;
450 		goto bail_release;
451 	}
452 
453 	rm->dma_buf_v = dma_alloc_coherent(&macio_get_pci_dev(mdev)->dev,
454 					   sizeof(struct rackmeter_dma),
455 					   &rm->dma_buf_p, GFP_KERNEL);
456 	if (rm->dma_buf_v == NULL) {
457 		printk(KERN_ERR
458 		       "rackmeter: failed to allocate dma buffer !\n");
459 		rc = -ENOMEM;
460 		goto bail_free_samples;
461 	}
462 #if 0
463 	rm->i2s_regs = ioremap(macio_resource_start(mdev, 0), 0x1000);
464 #else
465 	rm->i2s_regs = ioremap(ri2s.start, 0x1000);
466 #endif
467 	if (rm->i2s_regs == NULL) {
468 		printk(KERN_ERR
469 		       "rackmeter: failed to map i2s registers !\n");
470 		rc = -ENXIO;
471 		goto bail_free_dma;
472 	}
473 #if 0
474 	rm->dma_regs = ioremap(macio_resource_start(mdev, 1), 0x100);
475 #else
476 	rm->dma_regs = ioremap(rdma.start, 0x100);
477 #endif
478 	if (rm->dma_regs == NULL) {
479 		printk(KERN_ERR
480 		       "rackmeter: failed to map dma registers !\n");
481 		rc = -ENXIO;
482 		goto bail_unmap_i2s;
483 	}
484 
485 	rc = rackmeter_setup(rm);
486 	if (rc) {
487 		printk(KERN_ERR
488 		       "rackmeter: failed to initialize !\n");
489 		rc = -ENXIO;
490 		goto bail_unmap_dma;
491 	}
492 
493 	rc = request_irq(rm->irq, rackmeter_irq, 0, "rackmeter", rm);
494 	if (rc != 0) {
495 		printk(KERN_ERR
496 		       "rackmeter: failed to request interrupt !\n");
497 		goto bail_stop_dma;
498 	}
499 	of_node_put(np);
500 	return 0;
501 
502  bail_stop_dma:
503 	DBDMA_DO_RESET(rm->dma_regs);
504  bail_unmap_dma:
505 	iounmap(rm->dma_regs);
506  bail_unmap_i2s:
507 	iounmap(rm->i2s_regs);
508  bail_free_dma:
509 	dma_free_coherent(&macio_get_pci_dev(mdev)->dev,
510 			  sizeof(struct rackmeter_dma),
511 			  rm->dma_buf_v, rm->dma_buf_p);
512  bail_free_samples:
513 	free_page((unsigned long)rm->ubuf);
514  bail_release:
515 #if 0
516 	macio_release_resources(mdev);
517 #endif
518  bail_free:
519 	kfree(rm);
520  bail:
521 	of_node_put(i2s);
522 	of_node_put(np);
523 	dev_set_drvdata(&mdev->ofdev.dev, NULL);
524 	return rc;
525 }
526 
527 static int rackmeter_remove(struct macio_dev* mdev)
528 {
529 	struct rackmeter *rm = dev_get_drvdata(&mdev->ofdev.dev);
530 
531 	/* Stop CPU sniffer timer & work queues */
532 	rackmeter_stop_cpu_sniffer(rm);
533 
534 	/* Clear reference to private data */
535 	dev_set_drvdata(&mdev->ofdev.dev, NULL);
536 
537 	/* Stop/reset dbdma */
538 	DBDMA_DO_RESET(rm->dma_regs);
539 
540 	/* Release the IRQ */
541 	free_irq(rm->irq, rm);
542 
543 	/* Unmap registers */
544 	iounmap(rm->dma_regs);
545 	iounmap(rm->i2s_regs);
546 
547 	/* Free DMA */
548 	dma_free_coherent(&macio_get_pci_dev(mdev)->dev,
549 			  sizeof(struct rackmeter_dma),
550 			  rm->dma_buf_v, rm->dma_buf_p);
551 
552 	/* Free samples */
553 	free_page((unsigned long)rm->ubuf);
554 
555 #if 0
556 	/* Release resources */
557 	macio_release_resources(mdev);
558 #endif
559 
560 	/* Get rid of me */
561 	kfree(rm);
562 
563 	return 0;
564 }
565 
566 static int rackmeter_shutdown(struct macio_dev* mdev)
567 {
568 	struct rackmeter *rm = dev_get_drvdata(&mdev->ofdev.dev);
569 
570 	if (rm == NULL)
571 		return -ENODEV;
572 
573 	/* Stop CPU sniffer timer & work queues */
574 	rackmeter_stop_cpu_sniffer(rm);
575 
576 	/* Stop/reset dbdma */
577 	DBDMA_DO_RESET(rm->dma_regs);
578 
579 	return 0;
580 }
581 
582 static const struct of_device_id rackmeter_match[] = {
583 	{ .name = "i2s" },
584 	{ }
585 };
586 MODULE_DEVICE_TABLE(of, rackmeter_match);
587 
588 static struct macio_driver rackmeter_driver = {
589 	.driver = {
590 		.name = "rackmeter",
591 		.owner = THIS_MODULE,
592 		.of_match_table = rackmeter_match,
593 	},
594 	.probe = rackmeter_probe,
595 	.remove = rackmeter_remove,
596 	.shutdown = rackmeter_shutdown,
597 };
598 
599 
600 static int __init rackmeter_init(void)
601 {
602 	pr_debug("rackmeter_init()\n");
603 
604 	return macio_register_driver(&rackmeter_driver);
605 }
606 
607 static void __exit rackmeter_exit(void)
608 {
609 	pr_debug("rackmeter_exit()\n");
610 
611 	macio_unregister_driver(&rackmeter_driver);
612 }
613 
614 module_init(rackmeter_init);
615 module_exit(rackmeter_exit);
616 
617 
618 MODULE_LICENSE("GPL");
619 MODULE_AUTHOR("Benjamin Herrenschmidt <benh@kernel.crashing.org>");
620 MODULE_DESCRIPTION("RackMeter: Support vu-meter on XServe front panel");
621