xref: /linux/drivers/misc/cxl/guest.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
1 /*
2  * Copyright 2015 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/spinlock.h>
11 #include <linux/uaccess.h>
12 #include <linux/delay.h>
13 
14 #include "cxl.h"
15 #include "hcalls.h"
16 #include "trace.h"
17 
18 #define CXL_ERROR_DETECTED_EVENT	1
19 #define CXL_SLOT_RESET_EVENT		2
20 #define CXL_RESUME_EVENT		3
21 
22 static void pci_error_handlers(struct cxl_afu *afu,
23 				int bus_error_event,
24 				pci_channel_state_t state)
25 {
26 	struct pci_dev *afu_dev;
27 
28 	if (afu->phb == NULL)
29 		return;
30 
31 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
32 		if (!afu_dev->driver)
33 			continue;
34 
35 		switch (bus_error_event) {
36 		case CXL_ERROR_DETECTED_EVENT:
37 			afu_dev->error_state = state;
38 
39 			if (afu_dev->driver->err_handler &&
40 			    afu_dev->driver->err_handler->error_detected)
41 				afu_dev->driver->err_handler->error_detected(afu_dev, state);
42 		break;
43 		case CXL_SLOT_RESET_EVENT:
44 			afu_dev->error_state = state;
45 
46 			if (afu_dev->driver->err_handler &&
47 			    afu_dev->driver->err_handler->slot_reset)
48 				afu_dev->driver->err_handler->slot_reset(afu_dev);
49 		break;
50 		case CXL_RESUME_EVENT:
51 			if (afu_dev->driver->err_handler &&
52 			    afu_dev->driver->err_handler->resume)
53 				afu_dev->driver->err_handler->resume(afu_dev);
54 		break;
55 		}
56 	}
57 }
58 
59 static irqreturn_t guest_handle_psl_slice_error(struct cxl_context *ctx, u64 dsisr,
60 					u64 errstat)
61 {
62 	pr_devel("in %s\n", __func__);
63 	dev_crit(&ctx->afu->dev, "PSL ERROR STATUS: 0x%.16llx\n", errstat);
64 
65 	return cxl_ops->ack_irq(ctx, 0, errstat);
66 }
67 
68 static ssize_t guest_collect_vpd(struct cxl *adapter, struct cxl_afu *afu,
69 			void *buf, size_t len)
70 {
71 	unsigned int entries, mod;
72 	unsigned long **vpd_buf = NULL;
73 	struct sg_list *le;
74 	int rc = 0, i, tocopy;
75 	u64 out = 0;
76 
77 	if (buf == NULL)
78 		return -EINVAL;
79 
80 	/* number of entries in the list */
81 	entries = len / SG_BUFFER_SIZE;
82 	mod = len % SG_BUFFER_SIZE;
83 	if (mod)
84 		entries++;
85 
86 	if (entries > SG_MAX_ENTRIES) {
87 		entries = SG_MAX_ENTRIES;
88 		len = SG_MAX_ENTRIES * SG_BUFFER_SIZE;
89 		mod = 0;
90 	}
91 
92 	vpd_buf = kzalloc(entries * sizeof(unsigned long *), GFP_KERNEL);
93 	if (!vpd_buf)
94 		return -ENOMEM;
95 
96 	le = (struct sg_list *)get_zeroed_page(GFP_KERNEL);
97 	if (!le) {
98 		rc = -ENOMEM;
99 		goto err1;
100 	}
101 
102 	for (i = 0; i < entries; i++) {
103 		vpd_buf[i] = (unsigned long *)get_zeroed_page(GFP_KERNEL);
104 		if (!vpd_buf[i]) {
105 			rc = -ENOMEM;
106 			goto err2;
107 		}
108 		le[i].phys_addr = cpu_to_be64(virt_to_phys(vpd_buf[i]));
109 		le[i].len = cpu_to_be64(SG_BUFFER_SIZE);
110 		if ((i == (entries - 1)) && mod)
111 			le[i].len = cpu_to_be64(mod);
112 	}
113 
114 	if (adapter)
115 		rc = cxl_h_collect_vpd_adapter(adapter->guest->handle,
116 					virt_to_phys(le), entries, &out);
117 	else
118 		rc = cxl_h_collect_vpd(afu->guest->handle, 0,
119 				virt_to_phys(le), entries, &out);
120 	pr_devel("length of available (entries: %i), vpd: %#llx\n",
121 		entries, out);
122 
123 	if (!rc) {
124 		/*
125 		 * hcall returns in 'out' the size of available VPDs.
126 		 * It fills the buffer with as much data as possible.
127 		 */
128 		if (out < len)
129 			len = out;
130 		rc = len;
131 		if (out) {
132 			for (i = 0; i < entries; i++) {
133 				if (len < SG_BUFFER_SIZE)
134 					tocopy = len;
135 				else
136 					tocopy = SG_BUFFER_SIZE;
137 				memcpy(buf, vpd_buf[i], tocopy);
138 				buf += tocopy;
139 				len -= tocopy;
140 			}
141 		}
142 	}
143 err2:
144 	for (i = 0; i < entries; i++) {
145 		if (vpd_buf[i])
146 			free_page((unsigned long) vpd_buf[i]);
147 	}
148 	free_page((unsigned long) le);
149 err1:
150 	kfree(vpd_buf);
151 	return rc;
152 }
153 
154 static int guest_get_irq_info(struct cxl_context *ctx, struct cxl_irq_info *info)
155 {
156 	return cxl_h_collect_int_info(ctx->afu->guest->handle, ctx->process_token, info);
157 }
158 
159 static irqreturn_t guest_psl_irq(int irq, void *data)
160 {
161 	struct cxl_context *ctx = data;
162 	struct cxl_irq_info irq_info;
163 	int rc;
164 
165 	pr_devel("%d: received PSL interrupt %i\n", ctx->pe, irq);
166 	rc = guest_get_irq_info(ctx, &irq_info);
167 	if (rc) {
168 		WARN(1, "Unable to get IRQ info: %i\n", rc);
169 		return IRQ_HANDLED;
170 	}
171 
172 	rc = cxl_irq_psl8(irq, ctx, &irq_info);
173 	return rc;
174 }
175 
176 static int afu_read_error_state(struct cxl_afu *afu, int *state_out)
177 {
178 	u64 state;
179 	int rc = 0;
180 
181 	if (!afu)
182 		return -EIO;
183 
184 	rc = cxl_h_read_error_state(afu->guest->handle, &state);
185 	if (!rc) {
186 		WARN_ON(state != H_STATE_NORMAL &&
187 			state != H_STATE_DISABLE &&
188 			state != H_STATE_TEMP_UNAVAILABLE &&
189 			state != H_STATE_PERM_UNAVAILABLE);
190 		*state_out = state & 0xffffffff;
191 	}
192 	return rc;
193 }
194 
195 static irqreturn_t guest_slice_irq_err(int irq, void *data)
196 {
197 	struct cxl_afu *afu = data;
198 	int rc;
199 	u64 serr, afu_error, dsisr;
200 
201 	rc = cxl_h_get_fn_error_interrupt(afu->guest->handle, &serr);
202 	if (rc) {
203 		dev_crit(&afu->dev, "Couldn't read PSL_SERR_An: %d\n", rc);
204 		return IRQ_HANDLED;
205 	}
206 	afu_error = cxl_p2n_read(afu, CXL_AFU_ERR_An);
207 	dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
208 	cxl_afu_decode_psl_serr(afu, serr);
209 	dev_crit(&afu->dev, "AFU_ERR_An: 0x%.16llx\n", afu_error);
210 	dev_crit(&afu->dev, "PSL_DSISR_An: 0x%.16llx\n", dsisr);
211 
212 	rc = cxl_h_ack_fn_error_interrupt(afu->guest->handle, serr);
213 	if (rc)
214 		dev_crit(&afu->dev, "Couldn't ack slice error interrupt: %d\n",
215 			rc);
216 
217 	return IRQ_HANDLED;
218 }
219 
220 
221 static int irq_alloc_range(struct cxl *adapter, int len, int *irq)
222 {
223 	int i, n;
224 	struct irq_avail *cur;
225 
226 	for (i = 0; i < adapter->guest->irq_nranges; i++) {
227 		cur = &adapter->guest->irq_avail[i];
228 		n = bitmap_find_next_zero_area(cur->bitmap, cur->range,
229 					0, len, 0);
230 		if (n < cur->range) {
231 			bitmap_set(cur->bitmap, n, len);
232 			*irq = cur->offset + n;
233 			pr_devel("guest: allocate IRQs %#x->%#x\n",
234 				*irq, *irq + len - 1);
235 
236 			return 0;
237 		}
238 	}
239 	return -ENOSPC;
240 }
241 
242 static int irq_free_range(struct cxl *adapter, int irq, int len)
243 {
244 	int i, n;
245 	struct irq_avail *cur;
246 
247 	if (len == 0)
248 		return -ENOENT;
249 
250 	for (i = 0; i < adapter->guest->irq_nranges; i++) {
251 		cur = &adapter->guest->irq_avail[i];
252 		if (irq >= cur->offset &&
253 			(irq + len) <= (cur->offset + cur->range)) {
254 			n = irq - cur->offset;
255 			bitmap_clear(cur->bitmap, n, len);
256 			pr_devel("guest: release IRQs %#x->%#x\n",
257 				irq, irq + len - 1);
258 			return 0;
259 		}
260 	}
261 	return -ENOENT;
262 }
263 
264 static int guest_reset(struct cxl *adapter)
265 {
266 	struct cxl_afu *afu = NULL;
267 	int i, rc;
268 
269 	pr_devel("Adapter reset request\n");
270 	for (i = 0; i < adapter->slices; i++) {
271 		if ((afu = adapter->afu[i])) {
272 			pci_error_handlers(afu, CXL_ERROR_DETECTED_EVENT,
273 					pci_channel_io_frozen);
274 			cxl_context_detach_all(afu);
275 		}
276 	}
277 
278 	rc = cxl_h_reset_adapter(adapter->guest->handle);
279 	for (i = 0; i < adapter->slices; i++) {
280 		if (!rc && (afu = adapter->afu[i])) {
281 			pci_error_handlers(afu, CXL_SLOT_RESET_EVENT,
282 					pci_channel_io_normal);
283 			pci_error_handlers(afu, CXL_RESUME_EVENT, 0);
284 		}
285 	}
286 	return rc;
287 }
288 
289 static int guest_alloc_one_irq(struct cxl *adapter)
290 {
291 	int irq;
292 
293 	spin_lock(&adapter->guest->irq_alloc_lock);
294 	if (irq_alloc_range(adapter, 1, &irq))
295 		irq = -ENOSPC;
296 	spin_unlock(&adapter->guest->irq_alloc_lock);
297 	return irq;
298 }
299 
300 static void guest_release_one_irq(struct cxl *adapter, int irq)
301 {
302 	spin_lock(&adapter->guest->irq_alloc_lock);
303 	irq_free_range(adapter, irq, 1);
304 	spin_unlock(&adapter->guest->irq_alloc_lock);
305 }
306 
307 static int guest_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
308 				struct cxl *adapter, unsigned int num)
309 {
310 	int i, try, irq;
311 
312 	memset(irqs, 0, sizeof(struct cxl_irq_ranges));
313 
314 	spin_lock(&adapter->guest->irq_alloc_lock);
315 	for (i = 0; i < CXL_IRQ_RANGES && num; i++) {
316 		try = num;
317 		while (try) {
318 			if (irq_alloc_range(adapter, try, &irq) == 0)
319 				break;
320 			try /= 2;
321 		}
322 		if (!try)
323 			goto error;
324 		irqs->offset[i] = irq;
325 		irqs->range[i] = try;
326 		num -= try;
327 	}
328 	if (num)
329 		goto error;
330 	spin_unlock(&adapter->guest->irq_alloc_lock);
331 	return 0;
332 
333 error:
334 	for (i = 0; i < CXL_IRQ_RANGES; i++)
335 		irq_free_range(adapter, irqs->offset[i], irqs->range[i]);
336 	spin_unlock(&adapter->guest->irq_alloc_lock);
337 	return -ENOSPC;
338 }
339 
340 static void guest_release_irq_ranges(struct cxl_irq_ranges *irqs,
341 				struct cxl *adapter)
342 {
343 	int i;
344 
345 	spin_lock(&adapter->guest->irq_alloc_lock);
346 	for (i = 0; i < CXL_IRQ_RANGES; i++)
347 		irq_free_range(adapter, irqs->offset[i], irqs->range[i]);
348 	spin_unlock(&adapter->guest->irq_alloc_lock);
349 }
350 
351 static int guest_register_serr_irq(struct cxl_afu *afu)
352 {
353 	afu->err_irq_name = kasprintf(GFP_KERNEL, "cxl-%s-err",
354 				      dev_name(&afu->dev));
355 	if (!afu->err_irq_name)
356 		return -ENOMEM;
357 
358 	if (!(afu->serr_virq = cxl_map_irq(afu->adapter, afu->serr_hwirq,
359 				 guest_slice_irq_err, afu, afu->err_irq_name))) {
360 		kfree(afu->err_irq_name);
361 		afu->err_irq_name = NULL;
362 		return -ENOMEM;
363 	}
364 
365 	return 0;
366 }
367 
368 static void guest_release_serr_irq(struct cxl_afu *afu)
369 {
370 	cxl_unmap_irq(afu->serr_virq, afu);
371 	cxl_ops->release_one_irq(afu->adapter, afu->serr_hwirq);
372 	kfree(afu->err_irq_name);
373 }
374 
375 static int guest_ack_irq(struct cxl_context *ctx, u64 tfc, u64 psl_reset_mask)
376 {
377 	return cxl_h_control_faults(ctx->afu->guest->handle, ctx->process_token,
378 				tfc >> 32, (psl_reset_mask != 0));
379 }
380 
381 static void disable_afu_irqs(struct cxl_context *ctx)
382 {
383 	irq_hw_number_t hwirq;
384 	unsigned int virq;
385 	int r, i;
386 
387 	pr_devel("Disabling AFU(%d) interrupts\n", ctx->afu->slice);
388 	for (r = 0; r < CXL_IRQ_RANGES; r++) {
389 		hwirq = ctx->irqs.offset[r];
390 		for (i = 0; i < ctx->irqs.range[r]; hwirq++, i++) {
391 			virq = irq_find_mapping(NULL, hwirq);
392 			disable_irq(virq);
393 		}
394 	}
395 }
396 
397 static void enable_afu_irqs(struct cxl_context *ctx)
398 {
399 	irq_hw_number_t hwirq;
400 	unsigned int virq;
401 	int r, i;
402 
403 	pr_devel("Enabling AFU(%d) interrupts\n", ctx->afu->slice);
404 	for (r = 0; r < CXL_IRQ_RANGES; r++) {
405 		hwirq = ctx->irqs.offset[r];
406 		for (i = 0; i < ctx->irqs.range[r]; hwirq++, i++) {
407 			virq = irq_find_mapping(NULL, hwirq);
408 			enable_irq(virq);
409 		}
410 	}
411 }
412 
413 static int _guest_afu_cr_readXX(int sz, struct cxl_afu *afu, int cr_idx,
414 			u64 offset, u64 *val)
415 {
416 	unsigned long cr;
417 	char c;
418 	int rc = 0;
419 
420 	if (afu->crs_len < sz)
421 		return -ENOENT;
422 
423 	if (unlikely(offset >= afu->crs_len))
424 		return -ERANGE;
425 
426 	cr = get_zeroed_page(GFP_KERNEL);
427 	if (!cr)
428 		return -ENOMEM;
429 
430 	rc = cxl_h_get_config(afu->guest->handle, cr_idx, offset,
431 			virt_to_phys((void *)cr), sz);
432 	if (rc)
433 		goto err;
434 
435 	switch (sz) {
436 	case 1:
437 		c = *((char *) cr);
438 		*val = c;
439 		break;
440 	case 2:
441 		*val = in_le16((u16 *)cr);
442 		break;
443 	case 4:
444 		*val = in_le32((unsigned *)cr);
445 		break;
446 	case 8:
447 		*val = in_le64((u64 *)cr);
448 		break;
449 	default:
450 		WARN_ON(1);
451 	}
452 err:
453 	free_page(cr);
454 	return rc;
455 }
456 
457 static int guest_afu_cr_read32(struct cxl_afu *afu, int cr_idx, u64 offset,
458 			u32 *out)
459 {
460 	int rc;
461 	u64 val;
462 
463 	rc = _guest_afu_cr_readXX(4, afu, cr_idx, offset, &val);
464 	if (!rc)
465 		*out = (u32) val;
466 	return rc;
467 }
468 
469 static int guest_afu_cr_read16(struct cxl_afu *afu, int cr_idx, u64 offset,
470 			u16 *out)
471 {
472 	int rc;
473 	u64 val;
474 
475 	rc = _guest_afu_cr_readXX(2, afu, cr_idx, offset, &val);
476 	if (!rc)
477 		*out = (u16) val;
478 	return rc;
479 }
480 
481 static int guest_afu_cr_read8(struct cxl_afu *afu, int cr_idx, u64 offset,
482 			u8 *out)
483 {
484 	int rc;
485 	u64 val;
486 
487 	rc = _guest_afu_cr_readXX(1, afu, cr_idx, offset, &val);
488 	if (!rc)
489 		*out = (u8) val;
490 	return rc;
491 }
492 
493 static int guest_afu_cr_read64(struct cxl_afu *afu, int cr_idx, u64 offset,
494 			u64 *out)
495 {
496 	return _guest_afu_cr_readXX(8, afu, cr_idx, offset, out);
497 }
498 
499 static int guest_afu_cr_write32(struct cxl_afu *afu, int cr, u64 off, u32 in)
500 {
501 	/* config record is not writable from guest */
502 	return -EPERM;
503 }
504 
505 static int guest_afu_cr_write16(struct cxl_afu *afu, int cr, u64 off, u16 in)
506 {
507 	/* config record is not writable from guest */
508 	return -EPERM;
509 }
510 
511 static int guest_afu_cr_write8(struct cxl_afu *afu, int cr, u64 off, u8 in)
512 {
513 	/* config record is not writable from guest */
514 	return -EPERM;
515 }
516 
517 static int attach_afu_directed(struct cxl_context *ctx, u64 wed, u64 amr)
518 {
519 	struct cxl_process_element_hcall *elem;
520 	struct cxl *adapter = ctx->afu->adapter;
521 	const struct cred *cred;
522 	u32 pid, idx;
523 	int rc, r, i;
524 	u64 mmio_addr, mmio_size;
525 	__be64 flags = 0;
526 
527 	/* Must be 8 byte aligned and cannot cross a 4096 byte boundary */
528 	if (!(elem = (struct cxl_process_element_hcall *)
529 			get_zeroed_page(GFP_KERNEL)))
530 		return -ENOMEM;
531 
532 	elem->version = cpu_to_be64(CXL_PROCESS_ELEMENT_VERSION);
533 	if (ctx->kernel) {
534 		pid = 0;
535 		flags |= CXL_PE_TRANSLATION_ENABLED;
536 		flags |= CXL_PE_PRIVILEGED_PROCESS;
537 		if (mfmsr() & MSR_SF)
538 			flags |= CXL_PE_64_BIT;
539 	} else {
540 		pid = current->pid;
541 		flags |= CXL_PE_PROBLEM_STATE;
542 		flags |= CXL_PE_TRANSLATION_ENABLED;
543 		if (!test_tsk_thread_flag(current, TIF_32BIT))
544 			flags |= CXL_PE_64_BIT;
545 		cred = get_current_cred();
546 		if (uid_eq(cred->euid, GLOBAL_ROOT_UID))
547 			flags |= CXL_PE_PRIVILEGED_PROCESS;
548 		put_cred(cred);
549 	}
550 	elem->flags         = cpu_to_be64(flags);
551 	elem->common.tid    = cpu_to_be32(0); /* Unused */
552 	elem->common.pid    = cpu_to_be32(pid);
553 	elem->common.csrp   = cpu_to_be64(0); /* disable */
554 	elem->common.u.psl8.aurp0  = cpu_to_be64(0); /* disable */
555 	elem->common.u.psl8.aurp1  = cpu_to_be64(0); /* disable */
556 
557 	cxl_prefault(ctx, wed);
558 
559 	elem->common.u.psl8.sstp0  = cpu_to_be64(ctx->sstp0);
560 	elem->common.u.psl8.sstp1  = cpu_to_be64(ctx->sstp1);
561 
562 	/*
563 	 * Ensure we have at least one interrupt allocated to take faults for
564 	 * kernel contexts that may not have allocated any AFU IRQs at all:
565 	 */
566 	if (ctx->irqs.range[0] == 0) {
567 		rc = afu_register_irqs(ctx, 0);
568 		if (rc)
569 			goto out_free;
570 	}
571 
572 	for (r = 0; r < CXL_IRQ_RANGES; r++) {
573 		for (i = 0; i < ctx->irqs.range[r]; i++) {
574 			if (r == 0 && i == 0) {
575 				elem->pslVirtualIsn = cpu_to_be32(ctx->irqs.offset[0]);
576 			} else {
577 				idx = ctx->irqs.offset[r] + i - adapter->guest->irq_base_offset;
578 				elem->applicationVirtualIsnBitmap[idx / 8] |= 0x80 >> (idx % 8);
579 			}
580 		}
581 	}
582 	elem->common.amr = cpu_to_be64(amr);
583 	elem->common.wed = cpu_to_be64(wed);
584 
585 	disable_afu_irqs(ctx);
586 
587 	rc = cxl_h_attach_process(ctx->afu->guest->handle, elem,
588 				&ctx->process_token, &mmio_addr, &mmio_size);
589 	if (rc == H_SUCCESS) {
590 		if (ctx->master || !ctx->afu->pp_psa) {
591 			ctx->psn_phys = ctx->afu->psn_phys;
592 			ctx->psn_size = ctx->afu->adapter->ps_size;
593 		} else {
594 			ctx->psn_phys = mmio_addr;
595 			ctx->psn_size = mmio_size;
596 		}
597 		if (ctx->afu->pp_psa && mmio_size &&
598 			ctx->afu->pp_size == 0) {
599 			/*
600 			 * There's no property in the device tree to read the
601 			 * pp_size. We only find out at the 1st attach.
602 			 * Compared to bare-metal, it is too late and we
603 			 * should really lock here. However, on powerVM,
604 			 * pp_size is really only used to display in /sys.
605 			 * Being discussed with pHyp for their next release.
606 			 */
607 			ctx->afu->pp_size = mmio_size;
608 		}
609 		/* from PAPR: process element is bytes 4-7 of process token */
610 		ctx->external_pe = ctx->process_token & 0xFFFFFFFF;
611 		pr_devel("CXL pe=%i is known as %i for pHyp, mmio_size=%#llx",
612 			ctx->pe, ctx->external_pe, ctx->psn_size);
613 		ctx->pe_inserted = true;
614 		enable_afu_irqs(ctx);
615 	}
616 
617 out_free:
618 	free_page((u64)elem);
619 	return rc;
620 }
621 
622 static int guest_attach_process(struct cxl_context *ctx, bool kernel, u64 wed, u64 amr)
623 {
624 	pr_devel("in %s\n", __func__);
625 
626 	if (ctx->real_mode)
627 		return -EPERM;
628 
629 	ctx->kernel = kernel;
630 	if (ctx->afu->current_mode == CXL_MODE_DIRECTED)
631 		return attach_afu_directed(ctx, wed, amr);
632 
633 	/* dedicated mode not supported on FW840 */
634 
635 	return -EINVAL;
636 }
637 
638 static int detach_afu_directed(struct cxl_context *ctx)
639 {
640 	if (!ctx->pe_inserted)
641 		return 0;
642 	if (cxl_h_detach_process(ctx->afu->guest->handle, ctx->process_token))
643 		return -1;
644 	return 0;
645 }
646 
647 static int guest_detach_process(struct cxl_context *ctx)
648 {
649 	pr_devel("in %s\n", __func__);
650 	trace_cxl_detach(ctx);
651 
652 	if (!cxl_ops->link_ok(ctx->afu->adapter, ctx->afu))
653 		return -EIO;
654 
655 	if (ctx->afu->current_mode == CXL_MODE_DIRECTED)
656 		return detach_afu_directed(ctx);
657 
658 	return -EINVAL;
659 }
660 
661 static void guest_release_afu(struct device *dev)
662 {
663 	struct cxl_afu *afu = to_cxl_afu(dev);
664 
665 	pr_devel("%s\n", __func__);
666 
667 	idr_destroy(&afu->contexts_idr);
668 
669 	kfree(afu->guest);
670 	kfree(afu);
671 }
672 
673 ssize_t cxl_guest_read_afu_vpd(struct cxl_afu *afu, void *buf, size_t len)
674 {
675 	return guest_collect_vpd(NULL, afu, buf, len);
676 }
677 
678 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
679 static ssize_t guest_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
680 					loff_t off, size_t count)
681 {
682 	void *tbuf = NULL;
683 	int rc = 0;
684 
685 	tbuf = (void *) get_zeroed_page(GFP_KERNEL);
686 	if (!tbuf)
687 		return -ENOMEM;
688 
689 	rc = cxl_h_get_afu_err(afu->guest->handle,
690 			       off & 0x7,
691 			       virt_to_phys(tbuf),
692 			       count);
693 	if (rc)
694 		goto err;
695 
696 	if (count > ERR_BUFF_MAX_COPY_SIZE)
697 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
698 	memcpy(buf, tbuf, count);
699 err:
700 	free_page((u64)tbuf);
701 
702 	return rc;
703 }
704 
705 static int guest_afu_check_and_enable(struct cxl_afu *afu)
706 {
707 	return 0;
708 }
709 
710 static bool guest_support_attributes(const char *attr_name,
711 				     enum cxl_attrs type)
712 {
713 	switch (type) {
714 	case CXL_ADAPTER_ATTRS:
715 		if ((strcmp(attr_name, "base_image") == 0) ||
716 			(strcmp(attr_name, "load_image_on_perst") == 0) ||
717 			(strcmp(attr_name, "perst_reloads_same_image") == 0) ||
718 			(strcmp(attr_name, "image_loaded") == 0))
719 			return false;
720 		break;
721 	case CXL_AFU_MASTER_ATTRS:
722 		if ((strcmp(attr_name, "pp_mmio_off") == 0))
723 			return false;
724 		break;
725 	case CXL_AFU_ATTRS:
726 		break;
727 	default:
728 		break;
729 	}
730 
731 	return true;
732 }
733 
734 static int activate_afu_directed(struct cxl_afu *afu)
735 {
736 	int rc;
737 
738 	dev_info(&afu->dev, "Activating AFU(%d) directed mode\n", afu->slice);
739 
740 	afu->current_mode = CXL_MODE_DIRECTED;
741 
742 	afu->num_procs = afu->max_procs_virtualised;
743 
744 	if ((rc = cxl_chardev_m_afu_add(afu)))
745 		return rc;
746 
747 	if ((rc = cxl_sysfs_afu_m_add(afu)))
748 		goto err;
749 
750 	if ((rc = cxl_chardev_s_afu_add(afu)))
751 		goto err1;
752 
753 	return 0;
754 err1:
755 	cxl_sysfs_afu_m_remove(afu);
756 err:
757 	cxl_chardev_afu_remove(afu);
758 	return rc;
759 }
760 
761 static int guest_afu_activate_mode(struct cxl_afu *afu, int mode)
762 {
763 	if (!mode)
764 		return 0;
765 	if (!(mode & afu->modes_supported))
766 		return -EINVAL;
767 
768 	if (mode == CXL_MODE_DIRECTED)
769 		return activate_afu_directed(afu);
770 
771 	if (mode == CXL_MODE_DEDICATED)
772 		dev_err(&afu->dev, "Dedicated mode not supported\n");
773 
774 	return -EINVAL;
775 }
776 
777 static int deactivate_afu_directed(struct cxl_afu *afu)
778 {
779 	dev_info(&afu->dev, "Deactivating AFU(%d) directed mode\n", afu->slice);
780 
781 	afu->current_mode = 0;
782 	afu->num_procs = 0;
783 
784 	cxl_sysfs_afu_m_remove(afu);
785 	cxl_chardev_afu_remove(afu);
786 
787 	cxl_ops->afu_reset(afu);
788 
789 	return 0;
790 }
791 
792 static int guest_afu_deactivate_mode(struct cxl_afu *afu, int mode)
793 {
794 	if (!mode)
795 		return 0;
796 	if (!(mode & afu->modes_supported))
797 		return -EINVAL;
798 
799 	if (mode == CXL_MODE_DIRECTED)
800 		return deactivate_afu_directed(afu);
801 	return 0;
802 }
803 
804 static int guest_afu_reset(struct cxl_afu *afu)
805 {
806 	pr_devel("AFU(%d) reset request\n", afu->slice);
807 	return cxl_h_reset_afu(afu->guest->handle);
808 }
809 
810 static int guest_map_slice_regs(struct cxl_afu *afu)
811 {
812 	if (!(afu->p2n_mmio = ioremap(afu->guest->p2n_phys, afu->guest->p2n_size))) {
813 		dev_err(&afu->dev, "Error mapping AFU(%d) MMIO regions\n",
814 			afu->slice);
815 		return -ENOMEM;
816 	}
817 	return 0;
818 }
819 
820 static void guest_unmap_slice_regs(struct cxl_afu *afu)
821 {
822 	if (afu->p2n_mmio)
823 		iounmap(afu->p2n_mmio);
824 }
825 
826 static int afu_update_state(struct cxl_afu *afu)
827 {
828 	int rc, cur_state;
829 
830 	rc = afu_read_error_state(afu, &cur_state);
831 	if (rc)
832 		return rc;
833 
834 	if (afu->guest->previous_state == cur_state)
835 		return 0;
836 
837 	pr_devel("AFU(%d) update state to %#x\n", afu->slice, cur_state);
838 
839 	switch (cur_state) {
840 	case H_STATE_NORMAL:
841 		afu->guest->previous_state = cur_state;
842 		break;
843 
844 	case H_STATE_DISABLE:
845 		pci_error_handlers(afu, CXL_ERROR_DETECTED_EVENT,
846 				pci_channel_io_frozen);
847 
848 		cxl_context_detach_all(afu);
849 		if ((rc = cxl_ops->afu_reset(afu)))
850 			pr_devel("reset hcall failed %d\n", rc);
851 
852 		rc = afu_read_error_state(afu, &cur_state);
853 		if (!rc && cur_state == H_STATE_NORMAL) {
854 			pci_error_handlers(afu, CXL_SLOT_RESET_EVENT,
855 					pci_channel_io_normal);
856 			pci_error_handlers(afu, CXL_RESUME_EVENT, 0);
857 		}
858 		afu->guest->previous_state = 0;
859 		break;
860 
861 	case H_STATE_TEMP_UNAVAILABLE:
862 		afu->guest->previous_state = cur_state;
863 		break;
864 
865 	case H_STATE_PERM_UNAVAILABLE:
866 		dev_err(&afu->dev, "AFU is in permanent error state\n");
867 		pci_error_handlers(afu, CXL_ERROR_DETECTED_EVENT,
868 				pci_channel_io_perm_failure);
869 		afu->guest->previous_state = cur_state;
870 		break;
871 
872 	default:
873 		pr_err("Unexpected AFU(%d) error state: %#x\n",
874 		       afu->slice, cur_state);
875 		return -EINVAL;
876 	}
877 
878 	return rc;
879 }
880 
881 static void afu_handle_errstate(struct work_struct *work)
882 {
883 	struct cxl_afu_guest *afu_guest =
884 		container_of(to_delayed_work(work), struct cxl_afu_guest, work_err);
885 
886 	if (!afu_update_state(afu_guest->parent) &&
887 	    afu_guest->previous_state == H_STATE_PERM_UNAVAILABLE)
888 		return;
889 
890 	if (afu_guest->handle_err)
891 		schedule_delayed_work(&afu_guest->work_err,
892 				      msecs_to_jiffies(3000));
893 }
894 
895 static bool guest_link_ok(struct cxl *cxl, struct cxl_afu *afu)
896 {
897 	int state;
898 
899 	if (afu && (!afu_read_error_state(afu, &state))) {
900 		if (state == H_STATE_NORMAL)
901 			return true;
902 	}
903 
904 	return false;
905 }
906 
907 static int afu_properties_look_ok(struct cxl_afu *afu)
908 {
909 	if (afu->pp_irqs < 0) {
910 		dev_err(&afu->dev, "Unexpected per-process minimum interrupt value\n");
911 		return -EINVAL;
912 	}
913 
914 	if (afu->max_procs_virtualised < 1) {
915 		dev_err(&afu->dev, "Unexpected max number of processes virtualised value\n");
916 		return -EINVAL;
917 	}
918 
919 	if (afu->crs_len < 0) {
920 		dev_err(&afu->dev, "Unexpected configuration record size value\n");
921 		return -EINVAL;
922 	}
923 
924 	return 0;
925 }
926 
927 int cxl_guest_init_afu(struct cxl *adapter, int slice, struct device_node *afu_np)
928 {
929 	struct cxl_afu *afu;
930 	bool free = true;
931 	int rc;
932 
933 	pr_devel("in %s - AFU(%d)\n", __func__, slice);
934 	if (!(afu = cxl_alloc_afu(adapter, slice)))
935 		return -ENOMEM;
936 
937 	if (!(afu->guest = kzalloc(sizeof(struct cxl_afu_guest), GFP_KERNEL))) {
938 		kfree(afu);
939 		return -ENOMEM;
940 	}
941 
942 	if ((rc = dev_set_name(&afu->dev, "afu%i.%i",
943 					  adapter->adapter_num,
944 					  slice)))
945 		goto err1;
946 
947 	adapter->slices++;
948 
949 	if ((rc = cxl_of_read_afu_handle(afu, afu_np)))
950 		goto err1;
951 
952 	if ((rc = cxl_ops->afu_reset(afu)))
953 		goto err1;
954 
955 	if ((rc = cxl_of_read_afu_properties(afu, afu_np)))
956 		goto err1;
957 
958 	if ((rc = afu_properties_look_ok(afu)))
959 		goto err1;
960 
961 	if ((rc = guest_map_slice_regs(afu)))
962 		goto err1;
963 
964 	if ((rc = guest_register_serr_irq(afu)))
965 		goto err2;
966 
967 	/*
968 	 * After we call this function we must not free the afu directly, even
969 	 * if it returns an error!
970 	 */
971 	if ((rc = cxl_register_afu(afu)))
972 		goto err_put1;
973 
974 	if ((rc = cxl_sysfs_afu_add(afu)))
975 		goto err_put1;
976 
977 	/*
978 	 * pHyp doesn't expose the programming models supported by the
979 	 * AFU. pHyp currently only supports directed mode. If it adds
980 	 * dedicated mode later, this version of cxl has no way to
981 	 * detect it. So we'll initialize the driver, but the first
982 	 * attach will fail.
983 	 * Being discussed with pHyp to do better (likely new property)
984 	 */
985 	if (afu->max_procs_virtualised == 1)
986 		afu->modes_supported = CXL_MODE_DEDICATED;
987 	else
988 		afu->modes_supported = CXL_MODE_DIRECTED;
989 
990 	if ((rc = cxl_afu_select_best_mode(afu)))
991 		goto err_put2;
992 
993 	adapter->afu[afu->slice] = afu;
994 
995 	afu->enabled = true;
996 
997 	/*
998 	 * wake up the cpu periodically to check the state
999 	 * of the AFU using "afu" stored in the guest structure.
1000 	 */
1001 	afu->guest->parent = afu;
1002 	afu->guest->handle_err = true;
1003 	INIT_DELAYED_WORK(&afu->guest->work_err, afu_handle_errstate);
1004 	schedule_delayed_work(&afu->guest->work_err, msecs_to_jiffies(1000));
1005 
1006 	if ((rc = cxl_pci_vphb_add(afu)))
1007 		dev_info(&afu->dev, "Can't register vPHB\n");
1008 
1009 	return 0;
1010 
1011 err_put2:
1012 	cxl_sysfs_afu_remove(afu);
1013 err_put1:
1014 	device_unregister(&afu->dev);
1015 	free = false;
1016 	guest_release_serr_irq(afu);
1017 err2:
1018 	guest_unmap_slice_regs(afu);
1019 err1:
1020 	if (free) {
1021 		kfree(afu->guest);
1022 		kfree(afu);
1023 	}
1024 	return rc;
1025 }
1026 
1027 void cxl_guest_remove_afu(struct cxl_afu *afu)
1028 {
1029 	pr_devel("in %s - AFU(%d)\n", __func__, afu->slice);
1030 
1031 	if (!afu)
1032 		return;
1033 
1034 	/* flush and stop pending job */
1035 	afu->guest->handle_err = false;
1036 	flush_delayed_work(&afu->guest->work_err);
1037 
1038 	cxl_pci_vphb_remove(afu);
1039 	cxl_sysfs_afu_remove(afu);
1040 
1041 	spin_lock(&afu->adapter->afu_list_lock);
1042 	afu->adapter->afu[afu->slice] = NULL;
1043 	spin_unlock(&afu->adapter->afu_list_lock);
1044 
1045 	cxl_context_detach_all(afu);
1046 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1047 	guest_release_serr_irq(afu);
1048 	guest_unmap_slice_regs(afu);
1049 
1050 	device_unregister(&afu->dev);
1051 }
1052 
1053 static void free_adapter(struct cxl *adapter)
1054 {
1055 	struct irq_avail *cur;
1056 	int i;
1057 
1058 	if (adapter->guest) {
1059 		if (adapter->guest->irq_avail) {
1060 			for (i = 0; i < adapter->guest->irq_nranges; i++) {
1061 				cur = &adapter->guest->irq_avail[i];
1062 				kfree(cur->bitmap);
1063 			}
1064 			kfree(adapter->guest->irq_avail);
1065 		}
1066 		kfree(adapter->guest->status);
1067 		kfree(adapter->guest);
1068 	}
1069 	cxl_remove_adapter_nr(adapter);
1070 	kfree(adapter);
1071 }
1072 
1073 static int properties_look_ok(struct cxl *adapter)
1074 {
1075 	/* The absence of this property means that the operational
1076 	 * status is unknown or okay
1077 	 */
1078 	if (strlen(adapter->guest->status) &&
1079 	    strcmp(adapter->guest->status, "okay")) {
1080 		pr_err("ABORTING:Bad operational status of the device\n");
1081 		return -EINVAL;
1082 	}
1083 
1084 	return 0;
1085 }
1086 
1087 ssize_t cxl_guest_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1088 {
1089 	return guest_collect_vpd(adapter, NULL, buf, len);
1090 }
1091 
1092 void cxl_guest_remove_adapter(struct cxl *adapter)
1093 {
1094 	pr_devel("in %s\n", __func__);
1095 
1096 	cxl_sysfs_adapter_remove(adapter);
1097 
1098 	cxl_guest_remove_chardev(adapter);
1099 	device_unregister(&adapter->dev);
1100 }
1101 
1102 static void release_adapter(struct device *dev)
1103 {
1104 	free_adapter(to_cxl_adapter(dev));
1105 }
1106 
1107 struct cxl *cxl_guest_init_adapter(struct device_node *np, struct platform_device *pdev)
1108 {
1109 	struct cxl *adapter;
1110 	bool free = true;
1111 	int rc;
1112 
1113 	if (!(adapter = cxl_alloc_adapter()))
1114 		return ERR_PTR(-ENOMEM);
1115 
1116 	if (!(adapter->guest = kzalloc(sizeof(struct cxl_guest), GFP_KERNEL))) {
1117 		free_adapter(adapter);
1118 		return ERR_PTR(-ENOMEM);
1119 	}
1120 
1121 	adapter->slices = 0;
1122 	adapter->guest->pdev = pdev;
1123 	adapter->dev.parent = &pdev->dev;
1124 	adapter->dev.release = release_adapter;
1125 	dev_set_drvdata(&pdev->dev, adapter);
1126 
1127 	/*
1128 	 * Hypervisor controls PSL timebase initialization (p1 register).
1129 	 * On FW840, PSL is initialized.
1130 	 */
1131 	adapter->psl_timebase_synced = true;
1132 
1133 	if ((rc = cxl_of_read_adapter_handle(adapter, np)))
1134 		goto err1;
1135 
1136 	if ((rc = cxl_of_read_adapter_properties(adapter, np)))
1137 		goto err1;
1138 
1139 	if ((rc = properties_look_ok(adapter)))
1140 		goto err1;
1141 
1142 	if ((rc = cxl_guest_add_chardev(adapter)))
1143 		goto err1;
1144 
1145 	/*
1146 	 * After we call this function we must not free the adapter directly,
1147 	 * even if it returns an error!
1148 	 */
1149 	if ((rc = cxl_register_adapter(adapter)))
1150 		goto err_put1;
1151 
1152 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1153 		goto err_put1;
1154 
1155 	/* release the context lock as the adapter is configured */
1156 	cxl_adapter_context_unlock(adapter);
1157 
1158 	return adapter;
1159 
1160 err_put1:
1161 	device_unregister(&adapter->dev);
1162 	free = false;
1163 	cxl_guest_remove_chardev(adapter);
1164 err1:
1165 	if (free)
1166 		free_adapter(adapter);
1167 	return ERR_PTR(rc);
1168 }
1169 
1170 void cxl_guest_reload_module(struct cxl *adapter)
1171 {
1172 	struct platform_device *pdev;
1173 
1174 	pdev = adapter->guest->pdev;
1175 	cxl_guest_remove_adapter(adapter);
1176 
1177 	cxl_of_probe(pdev);
1178 }
1179 
1180 const struct cxl_backend_ops cxl_guest_ops = {
1181 	.module = THIS_MODULE,
1182 	.adapter_reset = guest_reset,
1183 	.alloc_one_irq = guest_alloc_one_irq,
1184 	.release_one_irq = guest_release_one_irq,
1185 	.alloc_irq_ranges = guest_alloc_irq_ranges,
1186 	.release_irq_ranges = guest_release_irq_ranges,
1187 	.setup_irq = NULL,
1188 	.handle_psl_slice_error = guest_handle_psl_slice_error,
1189 	.psl_interrupt = guest_psl_irq,
1190 	.ack_irq = guest_ack_irq,
1191 	.attach_process = guest_attach_process,
1192 	.detach_process = guest_detach_process,
1193 	.update_ivtes = NULL,
1194 	.support_attributes = guest_support_attributes,
1195 	.link_ok = guest_link_ok,
1196 	.release_afu = guest_release_afu,
1197 	.afu_read_err_buffer = guest_afu_read_err_buffer,
1198 	.afu_check_and_enable = guest_afu_check_and_enable,
1199 	.afu_activate_mode = guest_afu_activate_mode,
1200 	.afu_deactivate_mode = guest_afu_deactivate_mode,
1201 	.afu_reset = guest_afu_reset,
1202 	.afu_cr_read8 = guest_afu_cr_read8,
1203 	.afu_cr_read16 = guest_afu_cr_read16,
1204 	.afu_cr_read32 = guest_afu_cr_read32,
1205 	.afu_cr_read64 = guest_afu_cr_read64,
1206 	.afu_cr_write8 = guest_afu_cr_write8,
1207 	.afu_cr_write16 = guest_afu_cr_write16,
1208 	.afu_cr_write32 = guest_afu_cr_write32,
1209 	.read_adapter_vpd = cxl_guest_read_adapter_vpd,
1210 };
1211