xref: /openbmc/linux/drivers/misc/cxl/irq.c (revision c819e2cf)
1 /*
2  * Copyright 2014 IBM Corp.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public License
6  * as published by the Free Software Foundation; either version
7  * 2 of the License, or (at your option) any later version.
8  */
9 
10 #include <linux/interrupt.h>
11 #include <linux/workqueue.h>
12 #include <linux/sched.h>
13 #include <linux/wait.h>
14 #include <linux/slab.h>
15 #include <linux/pid.h>
16 #include <asm/cputable.h>
17 #include <misc/cxl.h>
18 
19 #include "cxl.h"
20 
21 /* XXX: This is implementation specific */
22 static irqreturn_t handle_psl_slice_error(struct cxl_context *ctx, u64 dsisr, u64 errstat)
23 {
24 	u64 fir1, fir2, fir_slice, serr, afu_debug;
25 
26 	fir1 = cxl_p1_read(ctx->afu->adapter, CXL_PSL_FIR1);
27 	fir2 = cxl_p1_read(ctx->afu->adapter, CXL_PSL_FIR2);
28 	fir_slice = cxl_p1n_read(ctx->afu, CXL_PSL_FIR_SLICE_An);
29 	serr = cxl_p1n_read(ctx->afu, CXL_PSL_SERR_An);
30 	afu_debug = cxl_p1n_read(ctx->afu, CXL_AFU_DEBUG_An);
31 
32 	dev_crit(&ctx->afu->dev, "PSL ERROR STATUS: 0x%.16llx\n", errstat);
33 	dev_crit(&ctx->afu->dev, "PSL_FIR1: 0x%.16llx\n", fir1);
34 	dev_crit(&ctx->afu->dev, "PSL_FIR2: 0x%.16llx\n", fir2);
35 	dev_crit(&ctx->afu->dev, "PSL_SERR_An: 0x%.16llx\n", serr);
36 	dev_crit(&ctx->afu->dev, "PSL_FIR_SLICE_An: 0x%.16llx\n", fir_slice);
37 	dev_crit(&ctx->afu->dev, "CXL_PSL_AFU_DEBUG_An: 0x%.16llx\n", afu_debug);
38 
39 	dev_crit(&ctx->afu->dev, "STOPPING CXL TRACE\n");
40 	cxl_stop_trace(ctx->afu->adapter);
41 
42 	return cxl_ack_irq(ctx, 0, errstat);
43 }
44 
45 irqreturn_t cxl_slice_irq_err(int irq, void *data)
46 {
47 	struct cxl_afu *afu = data;
48 	u64 fir_slice, errstat, serr, afu_debug;
49 
50 	WARN(irq, "CXL SLICE ERROR interrupt %i\n", irq);
51 
52 	serr = cxl_p1n_read(afu, CXL_PSL_SERR_An);
53 	fir_slice = cxl_p1n_read(afu, CXL_PSL_FIR_SLICE_An);
54 	errstat = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
55 	afu_debug = cxl_p1n_read(afu, CXL_AFU_DEBUG_An);
56 	dev_crit(&afu->dev, "PSL_SERR_An: 0x%.16llx\n", serr);
57 	dev_crit(&afu->dev, "PSL_FIR_SLICE_An: 0x%.16llx\n", fir_slice);
58 	dev_crit(&afu->dev, "CXL_PSL_ErrStat_An: 0x%.16llx\n", errstat);
59 	dev_crit(&afu->dev, "CXL_PSL_AFU_DEBUG_An: 0x%.16llx\n", afu_debug);
60 
61 	cxl_p1n_write(afu, CXL_PSL_SERR_An, serr);
62 
63 	return IRQ_HANDLED;
64 }
65 
66 static irqreturn_t cxl_irq_err(int irq, void *data)
67 {
68 	struct cxl *adapter = data;
69 	u64 fir1, fir2, err_ivte;
70 
71 	WARN(1, "CXL ERROR interrupt %i\n", irq);
72 
73 	err_ivte = cxl_p1_read(adapter, CXL_PSL_ErrIVTE);
74 	dev_crit(&adapter->dev, "PSL_ErrIVTE: 0x%.16llx\n", err_ivte);
75 
76 	dev_crit(&adapter->dev, "STOPPING CXL TRACE\n");
77 	cxl_stop_trace(adapter);
78 
79 	fir1 = cxl_p1_read(adapter, CXL_PSL_FIR1);
80 	fir2 = cxl_p1_read(adapter, CXL_PSL_FIR2);
81 
82 	dev_crit(&adapter->dev, "PSL_FIR1: 0x%.16llx\nPSL_FIR2: 0x%.16llx\n", fir1, fir2);
83 
84 	return IRQ_HANDLED;
85 }
86 
87 static irqreturn_t schedule_cxl_fault(struct cxl_context *ctx, u64 dsisr, u64 dar)
88 {
89 	ctx->dsisr = dsisr;
90 	ctx->dar = dar;
91 	schedule_work(&ctx->fault_work);
92 	return IRQ_HANDLED;
93 }
94 
95 static irqreturn_t cxl_irq(int irq, void *data, struct cxl_irq_info *irq_info)
96 {
97 	struct cxl_context *ctx = data;
98 	u64 dsisr, dar;
99 
100 	dsisr = irq_info->dsisr;
101 	dar = irq_info->dar;
102 
103 	pr_devel("CXL interrupt %i for afu pe: %i DSISR: %#llx DAR: %#llx\n", irq, ctx->pe, dsisr, dar);
104 
105 	if (dsisr & CXL_PSL_DSISR_An_DS) {
106 		/*
107 		 * We don't inherently need to sleep to handle this, but we do
108 		 * need to get a ref to the task's mm, which we can't do from
109 		 * irq context without the potential for a deadlock since it
110 		 * takes the task_lock. An alternate option would be to keep a
111 		 * reference to the task's mm the entire time it has cxl open,
112 		 * but to do that we need to solve the issue where we hold a
113 		 * ref to the mm, but the mm can hold a ref to the fd after an
114 		 * mmap preventing anything from being cleaned up.
115 		 */
116 		pr_devel("Scheduling segment miss handling for later pe: %i\n", ctx->pe);
117 		return schedule_cxl_fault(ctx, dsisr, dar);
118 	}
119 
120 	if (dsisr & CXL_PSL_DSISR_An_M)
121 		pr_devel("CXL interrupt: PTE not found\n");
122 	if (dsisr & CXL_PSL_DSISR_An_P)
123 		pr_devel("CXL interrupt: Storage protection violation\n");
124 	if (dsisr & CXL_PSL_DSISR_An_A)
125 		pr_devel("CXL interrupt: AFU lock access to write through or cache inhibited storage\n");
126 	if (dsisr & CXL_PSL_DSISR_An_S)
127 		pr_devel("CXL interrupt: Access was afu_wr or afu_zero\n");
128 	if (dsisr & CXL_PSL_DSISR_An_K)
129 		pr_devel("CXL interrupt: Access not permitted by virtual page class key protection\n");
130 
131 	if (dsisr & CXL_PSL_DSISR_An_DM) {
132 		/*
133 		 * In some cases we might be able to handle the fault
134 		 * immediately if hash_page would succeed, but we still need
135 		 * the task's mm, which as above we can't get without a lock
136 		 */
137 		pr_devel("Scheduling page fault handling for later pe: %i\n", ctx->pe);
138 		return schedule_cxl_fault(ctx, dsisr, dar);
139 	}
140 	if (dsisr & CXL_PSL_DSISR_An_ST)
141 		WARN(1, "CXL interrupt: Segment Table PTE not found\n");
142 	if (dsisr & CXL_PSL_DSISR_An_UR)
143 		pr_devel("CXL interrupt: AURP PTE not found\n");
144 	if (dsisr & CXL_PSL_DSISR_An_PE)
145 		return handle_psl_slice_error(ctx, dsisr, irq_info->errstat);
146 	if (dsisr & CXL_PSL_DSISR_An_AE) {
147 		pr_devel("CXL interrupt: AFU Error %.llx\n", irq_info->afu_err);
148 
149 		if (ctx->pending_afu_err) {
150 			/*
151 			 * This shouldn't happen - the PSL treats these errors
152 			 * as fatal and will have reset the AFU, so there's not
153 			 * much point buffering multiple AFU errors.
154 			 * OTOH if we DO ever see a storm of these come in it's
155 			 * probably best that we log them somewhere:
156 			 */
157 			dev_err_ratelimited(&ctx->afu->dev, "CXL AFU Error "
158 					    "undelivered to pe %i: %.llx\n",
159 					    ctx->pe, irq_info->afu_err);
160 		} else {
161 			spin_lock(&ctx->lock);
162 			ctx->afu_err = irq_info->afu_err;
163 			ctx->pending_afu_err = 1;
164 			spin_unlock(&ctx->lock);
165 
166 			wake_up_all(&ctx->wq);
167 		}
168 
169 		cxl_ack_irq(ctx, CXL_PSL_TFC_An_A, 0);
170 	}
171 	if (dsisr & CXL_PSL_DSISR_An_OC)
172 		pr_devel("CXL interrupt: OS Context Warning\n");
173 
174 	WARN(1, "Unhandled CXL PSL IRQ\n");
175 	return IRQ_HANDLED;
176 }
177 
178 static irqreturn_t fail_psl_irq(struct cxl_afu *afu, struct cxl_irq_info *irq_info)
179 {
180 	if (irq_info->dsisr & CXL_PSL_DSISR_TRANS)
181 		cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
182 	else
183 		cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
184 
185 	return IRQ_HANDLED;
186 }
187 
188 static irqreturn_t cxl_irq_multiplexed(int irq, void *data)
189 {
190 	struct cxl_afu *afu = data;
191 	struct cxl_context *ctx;
192 	struct cxl_irq_info irq_info;
193 	int ph = cxl_p2n_read(afu, CXL_PSL_PEHandle_An) & 0xffff;
194 	int ret;
195 
196 	if ((ret = cxl_get_irq(afu, &irq_info))) {
197 		WARN(1, "Unable to get CXL IRQ Info: %i\n", ret);
198 		return fail_psl_irq(afu, &irq_info);
199 	}
200 
201 	rcu_read_lock();
202 	ctx = idr_find(&afu->contexts_idr, ph);
203 	if (ctx) {
204 		ret = cxl_irq(irq, ctx, &irq_info);
205 		rcu_read_unlock();
206 		return ret;
207 	}
208 	rcu_read_unlock();
209 
210 	WARN(1, "Unable to demultiplex CXL PSL IRQ for PE %i DSISR %.16llx DAR"
211 		" %.16llx\n(Possible AFU HW issue - was a term/remove acked"
212 		" with outstanding transactions?)\n", ph, irq_info.dsisr,
213 		irq_info.dar);
214 	return fail_psl_irq(afu, &irq_info);
215 }
216 
217 static irqreturn_t cxl_irq_afu(int irq, void *data)
218 {
219 	struct cxl_context *ctx = data;
220 	irq_hw_number_t hwirq = irqd_to_hwirq(irq_get_irq_data(irq));
221 	int irq_off, afu_irq = 1;
222 	__u16 range;
223 	int r;
224 
225 	for (r = 1; r < CXL_IRQ_RANGES; r++) {
226 		irq_off = hwirq - ctx->irqs.offset[r];
227 		range = ctx->irqs.range[r];
228 		if (irq_off >= 0 && irq_off < range) {
229 			afu_irq += irq_off;
230 			break;
231 		}
232 		afu_irq += range;
233 	}
234 	if (unlikely(r >= CXL_IRQ_RANGES)) {
235 		WARN(1, "Recieved AFU IRQ out of range for pe %i (virq %i hwirq %lx)\n",
236 		     ctx->pe, irq, hwirq);
237 		return IRQ_HANDLED;
238 	}
239 
240 	pr_devel("Received AFU interrupt %i for pe: %i (virq %i hwirq %lx)\n",
241 	       afu_irq, ctx->pe, irq, hwirq);
242 
243 	if (unlikely(!ctx->irq_bitmap)) {
244 		WARN(1, "Recieved AFU IRQ for context with no IRQ bitmap\n");
245 		return IRQ_HANDLED;
246 	}
247 	spin_lock(&ctx->lock);
248 	set_bit(afu_irq - 1, ctx->irq_bitmap);
249 	ctx->pending_irq = true;
250 	spin_unlock(&ctx->lock);
251 
252 	wake_up_all(&ctx->wq);
253 
254 	return IRQ_HANDLED;
255 }
256 
257 unsigned int cxl_map_irq(struct cxl *adapter, irq_hw_number_t hwirq,
258 			 irq_handler_t handler, void *cookie, const char *name)
259 {
260 	unsigned int virq;
261 	int result;
262 
263 	/* IRQ Domain? */
264 	virq = irq_create_mapping(NULL, hwirq);
265 	if (!virq) {
266 		dev_warn(&adapter->dev, "cxl_map_irq: irq_create_mapping failed\n");
267 		return 0;
268 	}
269 
270 	cxl_setup_irq(adapter, hwirq, virq);
271 
272 	pr_devel("hwirq %#lx mapped to virq %u\n", hwirq, virq);
273 
274 	result = request_irq(virq, handler, 0, name, cookie);
275 	if (result) {
276 		dev_warn(&adapter->dev, "cxl_map_irq: request_irq failed: %i\n", result);
277 		return 0;
278 	}
279 
280 	return virq;
281 }
282 
283 void cxl_unmap_irq(unsigned int virq, void *cookie)
284 {
285 	free_irq(virq, cookie);
286 	irq_dispose_mapping(virq);
287 }
288 
289 static int cxl_register_one_irq(struct cxl *adapter,
290 				irq_handler_t handler,
291 				void *cookie,
292 				irq_hw_number_t *dest_hwirq,
293 				unsigned int *dest_virq,
294 				const char *name)
295 {
296 	int hwirq, virq;
297 
298 	if ((hwirq = cxl_alloc_one_irq(adapter)) < 0)
299 		return hwirq;
300 
301 	if (!(virq = cxl_map_irq(adapter, hwirq, handler, cookie, name)))
302 		goto err;
303 
304 	*dest_hwirq = hwirq;
305 	*dest_virq = virq;
306 
307 	return 0;
308 
309 err:
310 	cxl_release_one_irq(adapter, hwirq);
311 	return -ENOMEM;
312 }
313 
314 int cxl_register_psl_err_irq(struct cxl *adapter)
315 {
316 	int rc;
317 
318 	adapter->irq_name = kasprintf(GFP_KERNEL, "cxl-%s-err",
319 				      dev_name(&adapter->dev));
320 	if (!adapter->irq_name)
321 		return -ENOMEM;
322 
323 	if ((rc = cxl_register_one_irq(adapter, cxl_irq_err, adapter,
324 				       &adapter->err_hwirq,
325 				       &adapter->err_virq,
326 				       adapter->irq_name))) {
327 		kfree(adapter->irq_name);
328 		adapter->irq_name = NULL;
329 		return rc;
330 	}
331 
332 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, adapter->err_hwirq & 0xffff);
333 
334 	return 0;
335 }
336 
337 void cxl_release_psl_err_irq(struct cxl *adapter)
338 {
339 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, 0x0000000000000000);
340 	cxl_unmap_irq(adapter->err_virq, adapter);
341 	cxl_release_one_irq(adapter, adapter->err_hwirq);
342 	kfree(adapter->irq_name);
343 }
344 
345 int cxl_register_serr_irq(struct cxl_afu *afu)
346 {
347 	u64 serr;
348 	int rc;
349 
350 	afu->err_irq_name = kasprintf(GFP_KERNEL, "cxl-%s-err",
351 				      dev_name(&afu->dev));
352 	if (!afu->err_irq_name)
353 		return -ENOMEM;
354 
355 	if ((rc = cxl_register_one_irq(afu->adapter, cxl_slice_irq_err, afu,
356 				       &afu->serr_hwirq,
357 				       &afu->serr_virq, afu->err_irq_name))) {
358 		kfree(afu->err_irq_name);
359 		afu->err_irq_name = NULL;
360 		return rc;
361 	}
362 
363 	serr = cxl_p1n_read(afu, CXL_PSL_SERR_An);
364 	serr = (serr & 0x00ffffffffff0000ULL) | (afu->serr_hwirq & 0xffff);
365 	cxl_p1n_write(afu, CXL_PSL_SERR_An, serr);
366 
367 	return 0;
368 }
369 
370 void cxl_release_serr_irq(struct cxl_afu *afu)
371 {
372 	cxl_p1n_write(afu, CXL_PSL_SERR_An, 0x0000000000000000);
373 	cxl_unmap_irq(afu->serr_virq, afu);
374 	cxl_release_one_irq(afu->adapter, afu->serr_hwirq);
375 	kfree(afu->err_irq_name);
376 }
377 
378 int cxl_register_psl_irq(struct cxl_afu *afu)
379 {
380 	int rc;
381 
382 	afu->psl_irq_name = kasprintf(GFP_KERNEL, "cxl-%s",
383 				      dev_name(&afu->dev));
384 	if (!afu->psl_irq_name)
385 		return -ENOMEM;
386 
387 	if ((rc = cxl_register_one_irq(afu->adapter, cxl_irq_multiplexed, afu,
388 				    &afu->psl_hwirq, &afu->psl_virq,
389 				    afu->psl_irq_name))) {
390 		kfree(afu->psl_irq_name);
391 		afu->psl_irq_name = NULL;
392 	}
393 	return rc;
394 }
395 
396 void cxl_release_psl_irq(struct cxl_afu *afu)
397 {
398 	cxl_unmap_irq(afu->psl_virq, afu);
399 	cxl_release_one_irq(afu->adapter, afu->psl_hwirq);
400 	kfree(afu->psl_irq_name);
401 }
402 
403 void afu_irq_name_free(struct cxl_context *ctx)
404 {
405 	struct cxl_irq_name *irq_name, *tmp;
406 
407 	list_for_each_entry_safe(irq_name, tmp, &ctx->irq_names, list) {
408 		kfree(irq_name->name);
409 		list_del(&irq_name->list);
410 		kfree(irq_name);
411 	}
412 }
413 
414 int afu_register_irqs(struct cxl_context *ctx, u32 count)
415 {
416 	irq_hw_number_t hwirq;
417 	int rc, r, i, j = 1;
418 	struct cxl_irq_name *irq_name;
419 
420 	if ((rc = cxl_alloc_irq_ranges(&ctx->irqs, ctx->afu->adapter, count)))
421 		return rc;
422 
423 	/* Multiplexed PSL Interrupt */
424 	ctx->irqs.offset[0] = ctx->afu->psl_hwirq;
425 	ctx->irqs.range[0] = 1;
426 
427 	ctx->irq_count = count;
428 	ctx->irq_bitmap = kcalloc(BITS_TO_LONGS(count),
429 				  sizeof(*ctx->irq_bitmap), GFP_KERNEL);
430 	if (!ctx->irq_bitmap)
431 		return -ENOMEM;
432 
433 	/*
434 	 * Allocate names first.  If any fail, bail out before allocating
435 	 * actual hardware IRQs.
436 	 */
437 	INIT_LIST_HEAD(&ctx->irq_names);
438 	for (r = 1; r < CXL_IRQ_RANGES; r++) {
439 		for (i = 0; i < ctx->irqs.range[r]; hwirq++, i++) {
440 			irq_name = kmalloc(sizeof(struct cxl_irq_name),
441 					   GFP_KERNEL);
442 			if (!irq_name)
443 				goto out;
444 			irq_name->name = kasprintf(GFP_KERNEL, "cxl-%s-pe%i-%i",
445 						   dev_name(&ctx->afu->dev),
446 						   ctx->pe, j);
447 			if (!irq_name->name) {
448 				kfree(irq_name);
449 				goto out;
450 			}
451 			/* Add to tail so next look get the correct order */
452 			list_add_tail(&irq_name->list, &ctx->irq_names);
453 			j++;
454 		}
455 	}
456 
457 	/* We've allocated all memory now, so let's do the irq allocations */
458 	irq_name = list_first_entry(&ctx->irq_names, struct cxl_irq_name, list);
459 	for (r = 1; r < CXL_IRQ_RANGES; r++) {
460 		hwirq = ctx->irqs.offset[r];
461 		for (i = 0; i < ctx->irqs.range[r]; hwirq++, i++) {
462 			cxl_map_irq(ctx->afu->adapter, hwirq,
463 				    cxl_irq_afu, ctx, irq_name->name);
464 			irq_name = list_next_entry(irq_name, list);
465 		}
466 	}
467 
468 	return 0;
469 
470 out:
471 	afu_irq_name_free(ctx);
472 	return -ENOMEM;
473 }
474 
475 void afu_release_irqs(struct cxl_context *ctx)
476 {
477 	irq_hw_number_t hwirq;
478 	unsigned int virq;
479 	int r, i;
480 
481 	for (r = 1; r < CXL_IRQ_RANGES; r++) {
482 		hwirq = ctx->irqs.offset[r];
483 		for (i = 0; i < ctx->irqs.range[r]; hwirq++, i++) {
484 			virq = irq_find_mapping(NULL, hwirq);
485 			if (virq)
486 				cxl_unmap_irq(virq, ctx);
487 		}
488 	}
489 
490 	afu_irq_name_free(ctx);
491 	cxl_release_irq_ranges(&ctx->irqs, ctx->afu->adapter);
492 }
493