xref: /linux/drivers/misc/cxl/guest.c (revision 87c9c16317882dd6dbbc07e349bc3223e14f3244)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright 2015 IBM Corp.
4  */
5 
6 #include <linux/spinlock.h>
7 #include <linux/uaccess.h>
8 #include <linux/delay.h>
9 
10 #include "cxl.h"
11 #include "hcalls.h"
12 #include "trace.h"
13 
14 #define CXL_ERROR_DETECTED_EVENT	1
15 #define CXL_SLOT_RESET_EVENT		2
16 #define CXL_RESUME_EVENT		3
17 
18 static void pci_error_handlers(struct cxl_afu *afu,
19 				int bus_error_event,
20 				pci_channel_state_t state)
21 {
22 	struct pci_dev *afu_dev;
23 
24 	if (afu->phb == NULL)
25 		return;
26 
27 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
28 		if (!afu_dev->driver)
29 			continue;
30 
31 		switch (bus_error_event) {
32 		case CXL_ERROR_DETECTED_EVENT:
33 			afu_dev->error_state = state;
34 
35 			if (afu_dev->driver->err_handler &&
36 			    afu_dev->driver->err_handler->error_detected)
37 				afu_dev->driver->err_handler->error_detected(afu_dev, state);
38 		break;
39 		case CXL_SLOT_RESET_EVENT:
40 			afu_dev->error_state = state;
41 
42 			if (afu_dev->driver->err_handler &&
43 			    afu_dev->driver->err_handler->slot_reset)
44 				afu_dev->driver->err_handler->slot_reset(afu_dev);
45 		break;
46 		case CXL_RESUME_EVENT:
47 			if (afu_dev->driver->err_handler &&
48 			    afu_dev->driver->err_handler->resume)
49 				afu_dev->driver->err_handler->resume(afu_dev);
50 		break;
51 		}
52 	}
53 }
54 
55 static irqreturn_t guest_handle_psl_slice_error(struct cxl_context *ctx, u64 dsisr,
56 					u64 errstat)
57 {
58 	pr_devel("in %s\n", __func__);
59 	dev_crit(&ctx->afu->dev, "PSL ERROR STATUS: 0x%.16llx\n", errstat);
60 
61 	return cxl_ops->ack_irq(ctx, 0, errstat);
62 }
63 
64 static ssize_t guest_collect_vpd(struct cxl *adapter, struct cxl_afu *afu,
65 			void *buf, size_t len)
66 {
67 	unsigned int entries, mod;
68 	unsigned long **vpd_buf = NULL;
69 	struct sg_list *le;
70 	int rc = 0, i, tocopy;
71 	u64 out = 0;
72 
73 	if (buf == NULL)
74 		return -EINVAL;
75 
76 	/* number of entries in the list */
77 	entries = len / SG_BUFFER_SIZE;
78 	mod = len % SG_BUFFER_SIZE;
79 	if (mod)
80 		entries++;
81 
82 	if (entries > SG_MAX_ENTRIES) {
83 		entries = SG_MAX_ENTRIES;
84 		len = SG_MAX_ENTRIES * SG_BUFFER_SIZE;
85 		mod = 0;
86 	}
87 
88 	vpd_buf = kcalloc(entries, sizeof(unsigned long *), GFP_KERNEL);
89 	if (!vpd_buf)
90 		return -ENOMEM;
91 
92 	le = (struct sg_list *)get_zeroed_page(GFP_KERNEL);
93 	if (!le) {
94 		rc = -ENOMEM;
95 		goto err1;
96 	}
97 
98 	for (i = 0; i < entries; i++) {
99 		vpd_buf[i] = (unsigned long *)get_zeroed_page(GFP_KERNEL);
100 		if (!vpd_buf[i]) {
101 			rc = -ENOMEM;
102 			goto err2;
103 		}
104 		le[i].phys_addr = cpu_to_be64(virt_to_phys(vpd_buf[i]));
105 		le[i].len = cpu_to_be64(SG_BUFFER_SIZE);
106 		if ((i == (entries - 1)) && mod)
107 			le[i].len = cpu_to_be64(mod);
108 	}
109 
110 	if (adapter)
111 		rc = cxl_h_collect_vpd_adapter(adapter->guest->handle,
112 					virt_to_phys(le), entries, &out);
113 	else
114 		rc = cxl_h_collect_vpd(afu->guest->handle, 0,
115 				virt_to_phys(le), entries, &out);
116 	pr_devel("length of available (entries: %i), vpd: %#llx\n",
117 		entries, out);
118 
119 	if (!rc) {
120 		/*
121 		 * hcall returns in 'out' the size of available VPDs.
122 		 * It fills the buffer with as much data as possible.
123 		 */
124 		if (out < len)
125 			len = out;
126 		rc = len;
127 		if (out) {
128 			for (i = 0; i < entries; i++) {
129 				if (len < SG_BUFFER_SIZE)
130 					tocopy = len;
131 				else
132 					tocopy = SG_BUFFER_SIZE;
133 				memcpy(buf, vpd_buf[i], tocopy);
134 				buf += tocopy;
135 				len -= tocopy;
136 			}
137 		}
138 	}
139 err2:
140 	for (i = 0; i < entries; i++) {
141 		if (vpd_buf[i])
142 			free_page((unsigned long) vpd_buf[i]);
143 	}
144 	free_page((unsigned long) le);
145 err1:
146 	kfree(vpd_buf);
147 	return rc;
148 }
149 
150 static int guest_get_irq_info(struct cxl_context *ctx, struct cxl_irq_info *info)
151 {
152 	return cxl_h_collect_int_info(ctx->afu->guest->handle, ctx->process_token, info);
153 }
154 
155 static irqreturn_t guest_psl_irq(int irq, void *data)
156 {
157 	struct cxl_context *ctx = data;
158 	struct cxl_irq_info irq_info;
159 	int rc;
160 
161 	pr_devel("%d: received PSL interrupt %i\n", ctx->pe, irq);
162 	rc = guest_get_irq_info(ctx, &irq_info);
163 	if (rc) {
164 		WARN(1, "Unable to get IRQ info: %i\n", rc);
165 		return IRQ_HANDLED;
166 	}
167 
168 	rc = cxl_irq_psl8(irq, ctx, &irq_info);
169 	return rc;
170 }
171 
172 static int afu_read_error_state(struct cxl_afu *afu, int *state_out)
173 {
174 	u64 state;
175 	int rc = 0;
176 
177 	if (!afu)
178 		return -EIO;
179 
180 	rc = cxl_h_read_error_state(afu->guest->handle, &state);
181 	if (!rc) {
182 		WARN_ON(state != H_STATE_NORMAL &&
183 			state != H_STATE_DISABLE &&
184 			state != H_STATE_TEMP_UNAVAILABLE &&
185 			state != H_STATE_PERM_UNAVAILABLE);
186 		*state_out = state & 0xffffffff;
187 	}
188 	return rc;
189 }
190 
191 static irqreturn_t guest_slice_irq_err(int irq, void *data)
192 {
193 	struct cxl_afu *afu = data;
194 	int rc;
195 	u64 serr, afu_error, dsisr;
196 
197 	rc = cxl_h_get_fn_error_interrupt(afu->guest->handle, &serr);
198 	if (rc) {
199 		dev_crit(&afu->dev, "Couldn't read PSL_SERR_An: %d\n", rc);
200 		return IRQ_HANDLED;
201 	}
202 	afu_error = cxl_p2n_read(afu, CXL_AFU_ERR_An);
203 	dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
204 	cxl_afu_decode_psl_serr(afu, serr);
205 	dev_crit(&afu->dev, "AFU_ERR_An: 0x%.16llx\n", afu_error);
206 	dev_crit(&afu->dev, "PSL_DSISR_An: 0x%.16llx\n", dsisr);
207 
208 	rc = cxl_h_ack_fn_error_interrupt(afu->guest->handle, serr);
209 	if (rc)
210 		dev_crit(&afu->dev, "Couldn't ack slice error interrupt: %d\n",
211 			rc);
212 
213 	return IRQ_HANDLED;
214 }
215 
216 
217 static int irq_alloc_range(struct cxl *adapter, int len, int *irq)
218 {
219 	int i, n;
220 	struct irq_avail *cur;
221 
222 	for (i = 0; i < adapter->guest->irq_nranges; i++) {
223 		cur = &adapter->guest->irq_avail[i];
224 		n = bitmap_find_next_zero_area(cur->bitmap, cur->range,
225 					0, len, 0);
226 		if (n < cur->range) {
227 			bitmap_set(cur->bitmap, n, len);
228 			*irq = cur->offset + n;
229 			pr_devel("guest: allocate IRQs %#x->%#x\n",
230 				*irq, *irq + len - 1);
231 
232 			return 0;
233 		}
234 	}
235 	return -ENOSPC;
236 }
237 
238 static int irq_free_range(struct cxl *adapter, int irq, int len)
239 {
240 	int i, n;
241 	struct irq_avail *cur;
242 
243 	if (len == 0)
244 		return -ENOENT;
245 
246 	for (i = 0; i < adapter->guest->irq_nranges; i++) {
247 		cur = &adapter->guest->irq_avail[i];
248 		if (irq >= cur->offset &&
249 			(irq + len) <= (cur->offset + cur->range)) {
250 			n = irq - cur->offset;
251 			bitmap_clear(cur->bitmap, n, len);
252 			pr_devel("guest: release IRQs %#x->%#x\n",
253 				irq, irq + len - 1);
254 			return 0;
255 		}
256 	}
257 	return -ENOENT;
258 }
259 
260 static int guest_reset(struct cxl *adapter)
261 {
262 	struct cxl_afu *afu = NULL;
263 	int i, rc;
264 
265 	pr_devel("Adapter reset request\n");
266 	spin_lock(&adapter->afu_list_lock);
267 	for (i = 0; i < adapter->slices; i++) {
268 		if ((afu = adapter->afu[i])) {
269 			pci_error_handlers(afu, CXL_ERROR_DETECTED_EVENT,
270 					pci_channel_io_frozen);
271 			cxl_context_detach_all(afu);
272 		}
273 	}
274 
275 	rc = cxl_h_reset_adapter(adapter->guest->handle);
276 	for (i = 0; i < adapter->slices; i++) {
277 		if (!rc && (afu = adapter->afu[i])) {
278 			pci_error_handlers(afu, CXL_SLOT_RESET_EVENT,
279 					pci_channel_io_normal);
280 			pci_error_handlers(afu, CXL_RESUME_EVENT, 0);
281 		}
282 	}
283 	spin_unlock(&adapter->afu_list_lock);
284 	return rc;
285 }
286 
287 static int guest_alloc_one_irq(struct cxl *adapter)
288 {
289 	int irq;
290 
291 	spin_lock(&adapter->guest->irq_alloc_lock);
292 	if (irq_alloc_range(adapter, 1, &irq))
293 		irq = -ENOSPC;
294 	spin_unlock(&adapter->guest->irq_alloc_lock);
295 	return irq;
296 }
297 
298 static void guest_release_one_irq(struct cxl *adapter, int irq)
299 {
300 	spin_lock(&adapter->guest->irq_alloc_lock);
301 	irq_free_range(adapter, irq, 1);
302 	spin_unlock(&adapter->guest->irq_alloc_lock);
303 }
304 
305 static int guest_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
306 				struct cxl *adapter, unsigned int num)
307 {
308 	int i, try, irq;
309 
310 	memset(irqs, 0, sizeof(struct cxl_irq_ranges));
311 
312 	spin_lock(&adapter->guest->irq_alloc_lock);
313 	for (i = 0; i < CXL_IRQ_RANGES && num; i++) {
314 		try = num;
315 		while (try) {
316 			if (irq_alloc_range(adapter, try, &irq) == 0)
317 				break;
318 			try /= 2;
319 		}
320 		if (!try)
321 			goto error;
322 		irqs->offset[i] = irq;
323 		irqs->range[i] = try;
324 		num -= try;
325 	}
326 	if (num)
327 		goto error;
328 	spin_unlock(&adapter->guest->irq_alloc_lock);
329 	return 0;
330 
331 error:
332 	for (i = 0; i < CXL_IRQ_RANGES; i++)
333 		irq_free_range(adapter, irqs->offset[i], irqs->range[i]);
334 	spin_unlock(&adapter->guest->irq_alloc_lock);
335 	return -ENOSPC;
336 }
337 
338 static void guest_release_irq_ranges(struct cxl_irq_ranges *irqs,
339 				struct cxl *adapter)
340 {
341 	int i;
342 
343 	spin_lock(&adapter->guest->irq_alloc_lock);
344 	for (i = 0; i < CXL_IRQ_RANGES; i++)
345 		irq_free_range(adapter, irqs->offset[i], irqs->range[i]);
346 	spin_unlock(&adapter->guest->irq_alloc_lock);
347 }
348 
349 static int guest_register_serr_irq(struct cxl_afu *afu)
350 {
351 	afu->err_irq_name = kasprintf(GFP_KERNEL, "cxl-%s-err",
352 				      dev_name(&afu->dev));
353 	if (!afu->err_irq_name)
354 		return -ENOMEM;
355 
356 	if (!(afu->serr_virq = cxl_map_irq(afu->adapter, afu->serr_hwirq,
357 				 guest_slice_irq_err, afu, afu->err_irq_name))) {
358 		kfree(afu->err_irq_name);
359 		afu->err_irq_name = NULL;
360 		return -ENOMEM;
361 	}
362 
363 	return 0;
364 }
365 
366 static void guest_release_serr_irq(struct cxl_afu *afu)
367 {
368 	cxl_unmap_irq(afu->serr_virq, afu);
369 	cxl_ops->release_one_irq(afu->adapter, afu->serr_hwirq);
370 	kfree(afu->err_irq_name);
371 }
372 
373 static int guest_ack_irq(struct cxl_context *ctx, u64 tfc, u64 psl_reset_mask)
374 {
375 	return cxl_h_control_faults(ctx->afu->guest->handle, ctx->process_token,
376 				tfc >> 32, (psl_reset_mask != 0));
377 }
378 
379 static void disable_afu_irqs(struct cxl_context *ctx)
380 {
381 	irq_hw_number_t hwirq;
382 	unsigned int virq;
383 	int r, i;
384 
385 	pr_devel("Disabling AFU(%d) interrupts\n", ctx->afu->slice);
386 	for (r = 0; r < CXL_IRQ_RANGES; r++) {
387 		hwirq = ctx->irqs.offset[r];
388 		for (i = 0; i < ctx->irqs.range[r]; hwirq++, i++) {
389 			virq = irq_find_mapping(NULL, hwirq);
390 			disable_irq(virq);
391 		}
392 	}
393 }
394 
395 static void enable_afu_irqs(struct cxl_context *ctx)
396 {
397 	irq_hw_number_t hwirq;
398 	unsigned int virq;
399 	int r, i;
400 
401 	pr_devel("Enabling AFU(%d) interrupts\n", ctx->afu->slice);
402 	for (r = 0; r < CXL_IRQ_RANGES; r++) {
403 		hwirq = ctx->irqs.offset[r];
404 		for (i = 0; i < ctx->irqs.range[r]; hwirq++, i++) {
405 			virq = irq_find_mapping(NULL, hwirq);
406 			enable_irq(virq);
407 		}
408 	}
409 }
410 
411 static int _guest_afu_cr_readXX(int sz, struct cxl_afu *afu, int cr_idx,
412 			u64 offset, u64 *val)
413 {
414 	unsigned long cr;
415 	char c;
416 	int rc = 0;
417 
418 	if (afu->crs_len < sz)
419 		return -ENOENT;
420 
421 	if (unlikely(offset >= afu->crs_len))
422 		return -ERANGE;
423 
424 	cr = get_zeroed_page(GFP_KERNEL);
425 	if (!cr)
426 		return -ENOMEM;
427 
428 	rc = cxl_h_get_config(afu->guest->handle, cr_idx, offset,
429 			virt_to_phys((void *)cr), sz);
430 	if (rc)
431 		goto err;
432 
433 	switch (sz) {
434 	case 1:
435 		c = *((char *) cr);
436 		*val = c;
437 		break;
438 	case 2:
439 		*val = in_le16((u16 *)cr);
440 		break;
441 	case 4:
442 		*val = in_le32((unsigned *)cr);
443 		break;
444 	case 8:
445 		*val = in_le64((u64 *)cr);
446 		break;
447 	default:
448 		WARN_ON(1);
449 	}
450 err:
451 	free_page(cr);
452 	return rc;
453 }
454 
455 static int guest_afu_cr_read32(struct cxl_afu *afu, int cr_idx, u64 offset,
456 			u32 *out)
457 {
458 	int rc;
459 	u64 val;
460 
461 	rc = _guest_afu_cr_readXX(4, afu, cr_idx, offset, &val);
462 	if (!rc)
463 		*out = (u32) val;
464 	return rc;
465 }
466 
467 static int guest_afu_cr_read16(struct cxl_afu *afu, int cr_idx, u64 offset,
468 			u16 *out)
469 {
470 	int rc;
471 	u64 val;
472 
473 	rc = _guest_afu_cr_readXX(2, afu, cr_idx, offset, &val);
474 	if (!rc)
475 		*out = (u16) val;
476 	return rc;
477 }
478 
479 static int guest_afu_cr_read8(struct cxl_afu *afu, int cr_idx, u64 offset,
480 			u8 *out)
481 {
482 	int rc;
483 	u64 val;
484 
485 	rc = _guest_afu_cr_readXX(1, afu, cr_idx, offset, &val);
486 	if (!rc)
487 		*out = (u8) val;
488 	return rc;
489 }
490 
491 static int guest_afu_cr_read64(struct cxl_afu *afu, int cr_idx, u64 offset,
492 			u64 *out)
493 {
494 	return _guest_afu_cr_readXX(8, afu, cr_idx, offset, out);
495 }
496 
497 static int guest_afu_cr_write32(struct cxl_afu *afu, int cr, u64 off, u32 in)
498 {
499 	/* config record is not writable from guest */
500 	return -EPERM;
501 }
502 
503 static int guest_afu_cr_write16(struct cxl_afu *afu, int cr, u64 off, u16 in)
504 {
505 	/* config record is not writable from guest */
506 	return -EPERM;
507 }
508 
509 static int guest_afu_cr_write8(struct cxl_afu *afu, int cr, u64 off, u8 in)
510 {
511 	/* config record is not writable from guest */
512 	return -EPERM;
513 }
514 
515 static int attach_afu_directed(struct cxl_context *ctx, u64 wed, u64 amr)
516 {
517 	struct cxl_process_element_hcall *elem;
518 	struct cxl *adapter = ctx->afu->adapter;
519 	const struct cred *cred;
520 	u32 pid, idx;
521 	int rc, r, i;
522 	u64 mmio_addr, mmio_size;
523 	__be64 flags = 0;
524 
525 	/* Must be 8 byte aligned and cannot cross a 4096 byte boundary */
526 	if (!(elem = (struct cxl_process_element_hcall *)
527 			get_zeroed_page(GFP_KERNEL)))
528 		return -ENOMEM;
529 
530 	elem->version = cpu_to_be64(CXL_PROCESS_ELEMENT_VERSION);
531 	if (ctx->kernel) {
532 		pid = 0;
533 		flags |= CXL_PE_TRANSLATION_ENABLED;
534 		flags |= CXL_PE_PRIVILEGED_PROCESS;
535 		if (mfmsr() & MSR_SF)
536 			flags |= CXL_PE_64_BIT;
537 	} else {
538 		pid = current->pid;
539 		flags |= CXL_PE_PROBLEM_STATE;
540 		flags |= CXL_PE_TRANSLATION_ENABLED;
541 		if (!test_tsk_thread_flag(current, TIF_32BIT))
542 			flags |= CXL_PE_64_BIT;
543 		cred = get_current_cred();
544 		if (uid_eq(cred->euid, GLOBAL_ROOT_UID))
545 			flags |= CXL_PE_PRIVILEGED_PROCESS;
546 		put_cred(cred);
547 	}
548 	elem->flags         = cpu_to_be64(flags);
549 	elem->common.tid    = cpu_to_be32(0); /* Unused */
550 	elem->common.pid    = cpu_to_be32(pid);
551 	elem->common.csrp   = cpu_to_be64(0); /* disable */
552 	elem->common.u.psl8.aurp0  = cpu_to_be64(0); /* disable */
553 	elem->common.u.psl8.aurp1  = cpu_to_be64(0); /* disable */
554 
555 	cxl_prefault(ctx, wed);
556 
557 	elem->common.u.psl8.sstp0  = cpu_to_be64(ctx->sstp0);
558 	elem->common.u.psl8.sstp1  = cpu_to_be64(ctx->sstp1);
559 
560 	/*
561 	 * Ensure we have at least one interrupt allocated to take faults for
562 	 * kernel contexts that may not have allocated any AFU IRQs at all:
563 	 */
564 	if (ctx->irqs.range[0] == 0) {
565 		rc = afu_register_irqs(ctx, 0);
566 		if (rc)
567 			goto out_free;
568 	}
569 
570 	for (r = 0; r < CXL_IRQ_RANGES; r++) {
571 		for (i = 0; i < ctx->irqs.range[r]; i++) {
572 			if (r == 0 && i == 0) {
573 				elem->pslVirtualIsn = cpu_to_be32(ctx->irqs.offset[0]);
574 			} else {
575 				idx = ctx->irqs.offset[r] + i - adapter->guest->irq_base_offset;
576 				elem->applicationVirtualIsnBitmap[idx / 8] |= 0x80 >> (idx % 8);
577 			}
578 		}
579 	}
580 	elem->common.amr = cpu_to_be64(amr);
581 	elem->common.wed = cpu_to_be64(wed);
582 
583 	disable_afu_irqs(ctx);
584 
585 	rc = cxl_h_attach_process(ctx->afu->guest->handle, elem,
586 				&ctx->process_token, &mmio_addr, &mmio_size);
587 	if (rc == H_SUCCESS) {
588 		if (ctx->master || !ctx->afu->pp_psa) {
589 			ctx->psn_phys = ctx->afu->psn_phys;
590 			ctx->psn_size = ctx->afu->adapter->ps_size;
591 		} else {
592 			ctx->psn_phys = mmio_addr;
593 			ctx->psn_size = mmio_size;
594 		}
595 		if (ctx->afu->pp_psa && mmio_size &&
596 			ctx->afu->pp_size == 0) {
597 			/*
598 			 * There's no property in the device tree to read the
599 			 * pp_size. We only find out at the 1st attach.
600 			 * Compared to bare-metal, it is too late and we
601 			 * should really lock here. However, on powerVM,
602 			 * pp_size is really only used to display in /sys.
603 			 * Being discussed with pHyp for their next release.
604 			 */
605 			ctx->afu->pp_size = mmio_size;
606 		}
607 		/* from PAPR: process element is bytes 4-7 of process token */
608 		ctx->external_pe = ctx->process_token & 0xFFFFFFFF;
609 		pr_devel("CXL pe=%i is known as %i for pHyp, mmio_size=%#llx",
610 			ctx->pe, ctx->external_pe, ctx->psn_size);
611 		ctx->pe_inserted = true;
612 		enable_afu_irqs(ctx);
613 	}
614 
615 out_free:
616 	free_page((u64)elem);
617 	return rc;
618 }
619 
620 static int guest_attach_process(struct cxl_context *ctx, bool kernel, u64 wed, u64 amr)
621 {
622 	pr_devel("in %s\n", __func__);
623 
624 	ctx->kernel = kernel;
625 	if (ctx->afu->current_mode == CXL_MODE_DIRECTED)
626 		return attach_afu_directed(ctx, wed, amr);
627 
628 	/* dedicated mode not supported on FW840 */
629 
630 	return -EINVAL;
631 }
632 
633 static int detach_afu_directed(struct cxl_context *ctx)
634 {
635 	if (!ctx->pe_inserted)
636 		return 0;
637 	if (cxl_h_detach_process(ctx->afu->guest->handle, ctx->process_token))
638 		return -1;
639 	return 0;
640 }
641 
642 static int guest_detach_process(struct cxl_context *ctx)
643 {
644 	pr_devel("in %s\n", __func__);
645 	trace_cxl_detach(ctx);
646 
647 	if (!cxl_ops->link_ok(ctx->afu->adapter, ctx->afu))
648 		return -EIO;
649 
650 	if (ctx->afu->current_mode == CXL_MODE_DIRECTED)
651 		return detach_afu_directed(ctx);
652 
653 	return -EINVAL;
654 }
655 
656 static void guest_release_afu(struct device *dev)
657 {
658 	struct cxl_afu *afu = to_cxl_afu(dev);
659 
660 	pr_devel("%s\n", __func__);
661 
662 	idr_destroy(&afu->contexts_idr);
663 
664 	kfree(afu->guest);
665 	kfree(afu);
666 }
667 
668 ssize_t cxl_guest_read_afu_vpd(struct cxl_afu *afu, void *buf, size_t len)
669 {
670 	return guest_collect_vpd(NULL, afu, buf, len);
671 }
672 
673 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
674 static ssize_t guest_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
675 					loff_t off, size_t count)
676 {
677 	void *tbuf = NULL;
678 	int rc = 0;
679 
680 	tbuf = (void *) get_zeroed_page(GFP_KERNEL);
681 	if (!tbuf)
682 		return -ENOMEM;
683 
684 	rc = cxl_h_get_afu_err(afu->guest->handle,
685 			       off & 0x7,
686 			       virt_to_phys(tbuf),
687 			       count);
688 	if (rc)
689 		goto err;
690 
691 	if (count > ERR_BUFF_MAX_COPY_SIZE)
692 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
693 	memcpy(buf, tbuf, count);
694 err:
695 	free_page((u64)tbuf);
696 
697 	return rc;
698 }
699 
700 static int guest_afu_check_and_enable(struct cxl_afu *afu)
701 {
702 	return 0;
703 }
704 
705 static bool guest_support_attributes(const char *attr_name,
706 				     enum cxl_attrs type)
707 {
708 	switch (type) {
709 	case CXL_ADAPTER_ATTRS:
710 		if ((strcmp(attr_name, "base_image") == 0) ||
711 			(strcmp(attr_name, "load_image_on_perst") == 0) ||
712 			(strcmp(attr_name, "perst_reloads_same_image") == 0) ||
713 			(strcmp(attr_name, "image_loaded") == 0))
714 			return false;
715 		break;
716 	case CXL_AFU_MASTER_ATTRS:
717 		if ((strcmp(attr_name, "pp_mmio_off") == 0))
718 			return false;
719 		break;
720 	case CXL_AFU_ATTRS:
721 		break;
722 	default:
723 		break;
724 	}
725 
726 	return true;
727 }
728 
729 static int activate_afu_directed(struct cxl_afu *afu)
730 {
731 	int rc;
732 
733 	dev_info(&afu->dev, "Activating AFU(%d) directed mode\n", afu->slice);
734 
735 	afu->current_mode = CXL_MODE_DIRECTED;
736 
737 	afu->num_procs = afu->max_procs_virtualised;
738 
739 	if ((rc = cxl_chardev_m_afu_add(afu)))
740 		return rc;
741 
742 	if ((rc = cxl_sysfs_afu_m_add(afu)))
743 		goto err;
744 
745 	if ((rc = cxl_chardev_s_afu_add(afu)))
746 		goto err1;
747 
748 	return 0;
749 err1:
750 	cxl_sysfs_afu_m_remove(afu);
751 err:
752 	cxl_chardev_afu_remove(afu);
753 	return rc;
754 }
755 
756 static int guest_afu_activate_mode(struct cxl_afu *afu, int mode)
757 {
758 	if (!mode)
759 		return 0;
760 	if (!(mode & afu->modes_supported))
761 		return -EINVAL;
762 
763 	if (mode == CXL_MODE_DIRECTED)
764 		return activate_afu_directed(afu);
765 
766 	if (mode == CXL_MODE_DEDICATED)
767 		dev_err(&afu->dev, "Dedicated mode not supported\n");
768 
769 	return -EINVAL;
770 }
771 
772 static int deactivate_afu_directed(struct cxl_afu *afu)
773 {
774 	dev_info(&afu->dev, "Deactivating AFU(%d) directed mode\n", afu->slice);
775 
776 	afu->current_mode = 0;
777 	afu->num_procs = 0;
778 
779 	cxl_sysfs_afu_m_remove(afu);
780 	cxl_chardev_afu_remove(afu);
781 
782 	cxl_ops->afu_reset(afu);
783 
784 	return 0;
785 }
786 
787 static int guest_afu_deactivate_mode(struct cxl_afu *afu, int mode)
788 {
789 	if (!mode)
790 		return 0;
791 	if (!(mode & afu->modes_supported))
792 		return -EINVAL;
793 
794 	if (mode == CXL_MODE_DIRECTED)
795 		return deactivate_afu_directed(afu);
796 	return 0;
797 }
798 
799 static int guest_afu_reset(struct cxl_afu *afu)
800 {
801 	pr_devel("AFU(%d) reset request\n", afu->slice);
802 	return cxl_h_reset_afu(afu->guest->handle);
803 }
804 
805 static int guest_map_slice_regs(struct cxl_afu *afu)
806 {
807 	if (!(afu->p2n_mmio = ioremap(afu->guest->p2n_phys, afu->guest->p2n_size))) {
808 		dev_err(&afu->dev, "Error mapping AFU(%d) MMIO regions\n",
809 			afu->slice);
810 		return -ENOMEM;
811 	}
812 	return 0;
813 }
814 
815 static void guest_unmap_slice_regs(struct cxl_afu *afu)
816 {
817 	if (afu->p2n_mmio)
818 		iounmap(afu->p2n_mmio);
819 }
820 
821 static int afu_update_state(struct cxl_afu *afu)
822 {
823 	int rc, cur_state;
824 
825 	rc = afu_read_error_state(afu, &cur_state);
826 	if (rc)
827 		return rc;
828 
829 	if (afu->guest->previous_state == cur_state)
830 		return 0;
831 
832 	pr_devel("AFU(%d) update state to %#x\n", afu->slice, cur_state);
833 
834 	switch (cur_state) {
835 	case H_STATE_NORMAL:
836 		afu->guest->previous_state = cur_state;
837 		break;
838 
839 	case H_STATE_DISABLE:
840 		pci_error_handlers(afu, CXL_ERROR_DETECTED_EVENT,
841 				pci_channel_io_frozen);
842 
843 		cxl_context_detach_all(afu);
844 		if ((rc = cxl_ops->afu_reset(afu)))
845 			pr_devel("reset hcall failed %d\n", rc);
846 
847 		rc = afu_read_error_state(afu, &cur_state);
848 		if (!rc && cur_state == H_STATE_NORMAL) {
849 			pci_error_handlers(afu, CXL_SLOT_RESET_EVENT,
850 					pci_channel_io_normal);
851 			pci_error_handlers(afu, CXL_RESUME_EVENT, 0);
852 		}
853 		afu->guest->previous_state = 0;
854 		break;
855 
856 	case H_STATE_TEMP_UNAVAILABLE:
857 		afu->guest->previous_state = cur_state;
858 		break;
859 
860 	case H_STATE_PERM_UNAVAILABLE:
861 		dev_err(&afu->dev, "AFU is in permanent error state\n");
862 		pci_error_handlers(afu, CXL_ERROR_DETECTED_EVENT,
863 				pci_channel_io_perm_failure);
864 		afu->guest->previous_state = cur_state;
865 		break;
866 
867 	default:
868 		pr_err("Unexpected AFU(%d) error state: %#x\n",
869 		       afu->slice, cur_state);
870 		return -EINVAL;
871 	}
872 
873 	return rc;
874 }
875 
876 static void afu_handle_errstate(struct work_struct *work)
877 {
878 	struct cxl_afu_guest *afu_guest =
879 		container_of(to_delayed_work(work), struct cxl_afu_guest, work_err);
880 
881 	if (!afu_update_state(afu_guest->parent) &&
882 	    afu_guest->previous_state == H_STATE_PERM_UNAVAILABLE)
883 		return;
884 
885 	if (afu_guest->handle_err)
886 		schedule_delayed_work(&afu_guest->work_err,
887 				      msecs_to_jiffies(3000));
888 }
889 
890 static bool guest_link_ok(struct cxl *cxl, struct cxl_afu *afu)
891 {
892 	int state;
893 
894 	if (afu && (!afu_read_error_state(afu, &state))) {
895 		if (state == H_STATE_NORMAL)
896 			return true;
897 	}
898 
899 	return false;
900 }
901 
902 static int afu_properties_look_ok(struct cxl_afu *afu)
903 {
904 	if (afu->pp_irqs < 0) {
905 		dev_err(&afu->dev, "Unexpected per-process minimum interrupt value\n");
906 		return -EINVAL;
907 	}
908 
909 	if (afu->max_procs_virtualised < 1) {
910 		dev_err(&afu->dev, "Unexpected max number of processes virtualised value\n");
911 		return -EINVAL;
912 	}
913 
914 	return 0;
915 }
916 
917 int cxl_guest_init_afu(struct cxl *adapter, int slice, struct device_node *afu_np)
918 {
919 	struct cxl_afu *afu;
920 	bool free = true;
921 	int rc;
922 
923 	pr_devel("in %s - AFU(%d)\n", __func__, slice);
924 	if (!(afu = cxl_alloc_afu(adapter, slice)))
925 		return -ENOMEM;
926 
927 	if (!(afu->guest = kzalloc(sizeof(struct cxl_afu_guest), GFP_KERNEL))) {
928 		kfree(afu);
929 		return -ENOMEM;
930 	}
931 
932 	if ((rc = dev_set_name(&afu->dev, "afu%i.%i",
933 					  adapter->adapter_num,
934 					  slice)))
935 		goto err1;
936 
937 	adapter->slices++;
938 
939 	if ((rc = cxl_of_read_afu_handle(afu, afu_np)))
940 		goto err1;
941 
942 	if ((rc = cxl_ops->afu_reset(afu)))
943 		goto err1;
944 
945 	if ((rc = cxl_of_read_afu_properties(afu, afu_np)))
946 		goto err1;
947 
948 	if ((rc = afu_properties_look_ok(afu)))
949 		goto err1;
950 
951 	if ((rc = guest_map_slice_regs(afu)))
952 		goto err1;
953 
954 	if ((rc = guest_register_serr_irq(afu)))
955 		goto err2;
956 
957 	/*
958 	 * After we call this function we must not free the afu directly, even
959 	 * if it returns an error!
960 	 */
961 	if ((rc = cxl_register_afu(afu)))
962 		goto err_put1;
963 
964 	if ((rc = cxl_sysfs_afu_add(afu)))
965 		goto err_put1;
966 
967 	/*
968 	 * pHyp doesn't expose the programming models supported by the
969 	 * AFU. pHyp currently only supports directed mode. If it adds
970 	 * dedicated mode later, this version of cxl has no way to
971 	 * detect it. So we'll initialize the driver, but the first
972 	 * attach will fail.
973 	 * Being discussed with pHyp to do better (likely new property)
974 	 */
975 	if (afu->max_procs_virtualised == 1)
976 		afu->modes_supported = CXL_MODE_DEDICATED;
977 	else
978 		afu->modes_supported = CXL_MODE_DIRECTED;
979 
980 	if ((rc = cxl_afu_select_best_mode(afu)))
981 		goto err_put2;
982 
983 	adapter->afu[afu->slice] = afu;
984 
985 	afu->enabled = true;
986 
987 	/*
988 	 * wake up the cpu periodically to check the state
989 	 * of the AFU using "afu" stored in the guest structure.
990 	 */
991 	afu->guest->parent = afu;
992 	afu->guest->handle_err = true;
993 	INIT_DELAYED_WORK(&afu->guest->work_err, afu_handle_errstate);
994 	schedule_delayed_work(&afu->guest->work_err, msecs_to_jiffies(1000));
995 
996 	if ((rc = cxl_pci_vphb_add(afu)))
997 		dev_info(&afu->dev, "Can't register vPHB\n");
998 
999 	return 0;
1000 
1001 err_put2:
1002 	cxl_sysfs_afu_remove(afu);
1003 err_put1:
1004 	device_unregister(&afu->dev);
1005 	free = false;
1006 	guest_release_serr_irq(afu);
1007 err2:
1008 	guest_unmap_slice_regs(afu);
1009 err1:
1010 	if (free) {
1011 		kfree(afu->guest);
1012 		kfree(afu);
1013 	}
1014 	return rc;
1015 }
1016 
1017 void cxl_guest_remove_afu(struct cxl_afu *afu)
1018 {
1019 	if (!afu)
1020 		return;
1021 
1022 	/* flush and stop pending job */
1023 	afu->guest->handle_err = false;
1024 	flush_delayed_work(&afu->guest->work_err);
1025 
1026 	cxl_pci_vphb_remove(afu);
1027 	cxl_sysfs_afu_remove(afu);
1028 
1029 	spin_lock(&afu->adapter->afu_list_lock);
1030 	afu->adapter->afu[afu->slice] = NULL;
1031 	spin_unlock(&afu->adapter->afu_list_lock);
1032 
1033 	cxl_context_detach_all(afu);
1034 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1035 	guest_release_serr_irq(afu);
1036 	guest_unmap_slice_regs(afu);
1037 
1038 	device_unregister(&afu->dev);
1039 }
1040 
1041 static void free_adapter(struct cxl *adapter)
1042 {
1043 	struct irq_avail *cur;
1044 	int i;
1045 
1046 	if (adapter->guest) {
1047 		if (adapter->guest->irq_avail) {
1048 			for (i = 0; i < adapter->guest->irq_nranges; i++) {
1049 				cur = &adapter->guest->irq_avail[i];
1050 				kfree(cur->bitmap);
1051 			}
1052 			kfree(adapter->guest->irq_avail);
1053 		}
1054 		kfree(adapter->guest->status);
1055 		kfree(adapter->guest);
1056 	}
1057 	cxl_remove_adapter_nr(adapter);
1058 	kfree(adapter);
1059 }
1060 
1061 static int properties_look_ok(struct cxl *adapter)
1062 {
1063 	/* The absence of this property means that the operational
1064 	 * status is unknown or okay
1065 	 */
1066 	if (strlen(adapter->guest->status) &&
1067 	    strcmp(adapter->guest->status, "okay")) {
1068 		pr_err("ABORTING:Bad operational status of the device\n");
1069 		return -EINVAL;
1070 	}
1071 
1072 	return 0;
1073 }
1074 
1075 ssize_t cxl_guest_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1076 {
1077 	return guest_collect_vpd(adapter, NULL, buf, len);
1078 }
1079 
1080 void cxl_guest_remove_adapter(struct cxl *adapter)
1081 {
1082 	pr_devel("in %s\n", __func__);
1083 
1084 	cxl_sysfs_adapter_remove(adapter);
1085 
1086 	cxl_guest_remove_chardev(adapter);
1087 	device_unregister(&adapter->dev);
1088 }
1089 
1090 static void release_adapter(struct device *dev)
1091 {
1092 	free_adapter(to_cxl_adapter(dev));
1093 }
1094 
1095 struct cxl *cxl_guest_init_adapter(struct device_node *np, struct platform_device *pdev)
1096 {
1097 	struct cxl *adapter;
1098 	bool free = true;
1099 	int rc;
1100 
1101 	if (!(adapter = cxl_alloc_adapter()))
1102 		return ERR_PTR(-ENOMEM);
1103 
1104 	if (!(adapter->guest = kzalloc(sizeof(struct cxl_guest), GFP_KERNEL))) {
1105 		free_adapter(adapter);
1106 		return ERR_PTR(-ENOMEM);
1107 	}
1108 
1109 	adapter->slices = 0;
1110 	adapter->guest->pdev = pdev;
1111 	adapter->dev.parent = &pdev->dev;
1112 	adapter->dev.release = release_adapter;
1113 	dev_set_drvdata(&pdev->dev, adapter);
1114 
1115 	/*
1116 	 * Hypervisor controls PSL timebase initialization (p1 register).
1117 	 * On FW840, PSL is initialized.
1118 	 */
1119 	adapter->psl_timebase_synced = true;
1120 
1121 	if ((rc = cxl_of_read_adapter_handle(adapter, np)))
1122 		goto err1;
1123 
1124 	if ((rc = cxl_of_read_adapter_properties(adapter, np)))
1125 		goto err1;
1126 
1127 	if ((rc = properties_look_ok(adapter)))
1128 		goto err1;
1129 
1130 	if ((rc = cxl_guest_add_chardev(adapter)))
1131 		goto err1;
1132 
1133 	/*
1134 	 * After we call this function we must not free the adapter directly,
1135 	 * even if it returns an error!
1136 	 */
1137 	if ((rc = cxl_register_adapter(adapter)))
1138 		goto err_put1;
1139 
1140 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1141 		goto err_put1;
1142 
1143 	/* release the context lock as the adapter is configured */
1144 	cxl_adapter_context_unlock(adapter);
1145 
1146 	return adapter;
1147 
1148 err_put1:
1149 	device_unregister(&adapter->dev);
1150 	free = false;
1151 	cxl_guest_remove_chardev(adapter);
1152 err1:
1153 	if (free)
1154 		free_adapter(adapter);
1155 	return ERR_PTR(rc);
1156 }
1157 
1158 void cxl_guest_reload_module(struct cxl *adapter)
1159 {
1160 	struct platform_device *pdev;
1161 
1162 	pdev = adapter->guest->pdev;
1163 	cxl_guest_remove_adapter(adapter);
1164 
1165 	cxl_of_probe(pdev);
1166 }
1167 
1168 const struct cxl_backend_ops cxl_guest_ops = {
1169 	.module = THIS_MODULE,
1170 	.adapter_reset = guest_reset,
1171 	.alloc_one_irq = guest_alloc_one_irq,
1172 	.release_one_irq = guest_release_one_irq,
1173 	.alloc_irq_ranges = guest_alloc_irq_ranges,
1174 	.release_irq_ranges = guest_release_irq_ranges,
1175 	.setup_irq = NULL,
1176 	.handle_psl_slice_error = guest_handle_psl_slice_error,
1177 	.psl_interrupt = guest_psl_irq,
1178 	.ack_irq = guest_ack_irq,
1179 	.attach_process = guest_attach_process,
1180 	.detach_process = guest_detach_process,
1181 	.update_ivtes = NULL,
1182 	.support_attributes = guest_support_attributes,
1183 	.link_ok = guest_link_ok,
1184 	.release_afu = guest_release_afu,
1185 	.afu_read_err_buffer = guest_afu_read_err_buffer,
1186 	.afu_check_and_enable = guest_afu_check_and_enable,
1187 	.afu_activate_mode = guest_afu_activate_mode,
1188 	.afu_deactivate_mode = guest_afu_deactivate_mode,
1189 	.afu_reset = guest_afu_reset,
1190 	.afu_cr_read8 = guest_afu_cr_read8,
1191 	.afu_cr_read16 = guest_afu_cr_read16,
1192 	.afu_cr_read32 = guest_afu_cr_read32,
1193 	.afu_cr_read64 = guest_afu_cr_read64,
1194 	.afu_cr_write8 = guest_afu_cr_write8,
1195 	.afu_cr_write16 = guest_afu_cr_write16,
1196 	.afu_cr_write32 = guest_afu_cr_write32,
1197 	.read_adapter_vpd = cxl_guest_read_adapter_vpd,
1198 };
1199