xref: /openbmc/linux/drivers/misc/cxl/guest.c (revision aa74c44b)
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 	struct pci_driver *afu_drv;
24 	const struct pci_error_handlers *err_handler;
25 
26 	if (afu->phb == NULL)
27 		return;
28 
29 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
30 		afu_drv = to_pci_driver(afu_dev->dev.driver);
31 		if (!afu_drv)
32 			continue;
33 
34 		err_handler = afu_drv->err_handler;
35 		switch (bus_error_event) {
36 		case CXL_ERROR_DETECTED_EVENT:
37 			afu_dev->error_state = state;
38 
39 			if (err_handler &&
40 			    err_handler->error_detected)
41 				err_handler->error_detected(afu_dev, state);
42 			break;
43 		case CXL_SLOT_RESET_EVENT:
44 			afu_dev->error_state = state;
45 
46 			if (err_handler &&
47 			    err_handler->slot_reset)
48 				err_handler->slot_reset(afu_dev);
49 			break;
50 		case CXL_RESUME_EVENT:
51 			if (err_handler &&
52 			    err_handler->resume)
53 				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 = kcalloc(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 	spin_lock(&adapter->afu_list_lock);
271 	for (i = 0; i < adapter->slices; i++) {
272 		if ((afu = adapter->afu[i])) {
273 			pci_error_handlers(afu, CXL_ERROR_DETECTED_EVENT,
274 					pci_channel_io_frozen);
275 			cxl_context_detach_all(afu);
276 		}
277 	}
278 
279 	rc = cxl_h_reset_adapter(adapter->guest->handle);
280 	for (i = 0; i < adapter->slices; i++) {
281 		if (!rc && (afu = adapter->afu[i])) {
282 			pci_error_handlers(afu, CXL_SLOT_RESET_EVENT,
283 					pci_channel_io_normal);
284 			pci_error_handlers(afu, CXL_RESUME_EVENT, 0);
285 		}
286 	}
287 	spin_unlock(&adapter->afu_list_lock);
288 	return rc;
289 }
290 
291 static int guest_alloc_one_irq(struct cxl *adapter)
292 {
293 	int irq;
294 
295 	spin_lock(&adapter->guest->irq_alloc_lock);
296 	if (irq_alloc_range(adapter, 1, &irq))
297 		irq = -ENOSPC;
298 	spin_unlock(&adapter->guest->irq_alloc_lock);
299 	return irq;
300 }
301 
302 static void guest_release_one_irq(struct cxl *adapter, int irq)
303 {
304 	spin_lock(&adapter->guest->irq_alloc_lock);
305 	irq_free_range(adapter, irq, 1);
306 	spin_unlock(&adapter->guest->irq_alloc_lock);
307 }
308 
309 static int guest_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
310 				struct cxl *adapter, unsigned int num)
311 {
312 	int i, try, irq;
313 
314 	memset(irqs, 0, sizeof(struct cxl_irq_ranges));
315 
316 	spin_lock(&adapter->guest->irq_alloc_lock);
317 	for (i = 0; i < CXL_IRQ_RANGES && num; i++) {
318 		try = num;
319 		while (try) {
320 			if (irq_alloc_range(adapter, try, &irq) == 0)
321 				break;
322 			try /= 2;
323 		}
324 		if (!try)
325 			goto error;
326 		irqs->offset[i] = irq;
327 		irqs->range[i] = try;
328 		num -= try;
329 	}
330 	if (num)
331 		goto error;
332 	spin_unlock(&adapter->guest->irq_alloc_lock);
333 	return 0;
334 
335 error:
336 	for (i = 0; i < CXL_IRQ_RANGES; i++)
337 		irq_free_range(adapter, irqs->offset[i], irqs->range[i]);
338 	spin_unlock(&adapter->guest->irq_alloc_lock);
339 	return -ENOSPC;
340 }
341 
342 static void guest_release_irq_ranges(struct cxl_irq_ranges *irqs,
343 				struct cxl *adapter)
344 {
345 	int i;
346 
347 	spin_lock(&adapter->guest->irq_alloc_lock);
348 	for (i = 0; i < CXL_IRQ_RANGES; i++)
349 		irq_free_range(adapter, irqs->offset[i], irqs->range[i]);
350 	spin_unlock(&adapter->guest->irq_alloc_lock);
351 }
352 
353 static int guest_register_serr_irq(struct cxl_afu *afu)
354 {
355 	afu->err_irq_name = kasprintf(GFP_KERNEL, "cxl-%s-err",
356 				      dev_name(&afu->dev));
357 	if (!afu->err_irq_name)
358 		return -ENOMEM;
359 
360 	if (!(afu->serr_virq = cxl_map_irq(afu->adapter, afu->serr_hwirq,
361 				 guest_slice_irq_err, afu, afu->err_irq_name))) {
362 		kfree(afu->err_irq_name);
363 		afu->err_irq_name = NULL;
364 		return -ENOMEM;
365 	}
366 
367 	return 0;
368 }
369 
370 static void guest_release_serr_irq(struct cxl_afu *afu)
371 {
372 	cxl_unmap_irq(afu->serr_virq, afu);
373 	cxl_ops->release_one_irq(afu->adapter, afu->serr_hwirq);
374 	kfree(afu->err_irq_name);
375 }
376 
377 static int guest_ack_irq(struct cxl_context *ctx, u64 tfc, u64 psl_reset_mask)
378 {
379 	return cxl_h_control_faults(ctx->afu->guest->handle, ctx->process_token,
380 				tfc >> 32, (psl_reset_mask != 0));
381 }
382 
383 static void disable_afu_irqs(struct cxl_context *ctx)
384 {
385 	irq_hw_number_t hwirq;
386 	unsigned int virq;
387 	int r, i;
388 
389 	pr_devel("Disabling AFU(%d) interrupts\n", ctx->afu->slice);
390 	for (r = 0; r < CXL_IRQ_RANGES; r++) {
391 		hwirq = ctx->irqs.offset[r];
392 		for (i = 0; i < ctx->irqs.range[r]; hwirq++, i++) {
393 			virq = irq_find_mapping(NULL, hwirq);
394 			disable_irq(virq);
395 		}
396 	}
397 }
398 
399 static void enable_afu_irqs(struct cxl_context *ctx)
400 {
401 	irq_hw_number_t hwirq;
402 	unsigned int virq;
403 	int r, i;
404 
405 	pr_devel("Enabling AFU(%d) interrupts\n", ctx->afu->slice);
406 	for (r = 0; r < CXL_IRQ_RANGES; r++) {
407 		hwirq = ctx->irqs.offset[r];
408 		for (i = 0; i < ctx->irqs.range[r]; hwirq++, i++) {
409 			virq = irq_find_mapping(NULL, hwirq);
410 			enable_irq(virq);
411 		}
412 	}
413 }
414 
415 static int _guest_afu_cr_readXX(int sz, struct cxl_afu *afu, int cr_idx,
416 			u64 offset, u64 *val)
417 {
418 	unsigned long cr;
419 	char c;
420 	int rc = 0;
421 
422 	if (afu->crs_len < sz)
423 		return -ENOENT;
424 
425 	if (unlikely(offset >= afu->crs_len))
426 		return -ERANGE;
427 
428 	cr = get_zeroed_page(GFP_KERNEL);
429 	if (!cr)
430 		return -ENOMEM;
431 
432 	rc = cxl_h_get_config(afu->guest->handle, cr_idx, offset,
433 			virt_to_phys((void *)cr), sz);
434 	if (rc)
435 		goto err;
436 
437 	switch (sz) {
438 	case 1:
439 		c = *((char *) cr);
440 		*val = c;
441 		break;
442 	case 2:
443 		*val = in_le16((u16 *)cr);
444 		break;
445 	case 4:
446 		*val = in_le32((unsigned *)cr);
447 		break;
448 	case 8:
449 		*val = in_le64((u64 *)cr);
450 		break;
451 	default:
452 		WARN_ON(1);
453 	}
454 err:
455 	free_page(cr);
456 	return rc;
457 }
458 
459 static int guest_afu_cr_read32(struct cxl_afu *afu, int cr_idx, u64 offset,
460 			u32 *out)
461 {
462 	int rc;
463 	u64 val;
464 
465 	rc = _guest_afu_cr_readXX(4, afu, cr_idx, offset, &val);
466 	if (!rc)
467 		*out = (u32) val;
468 	return rc;
469 }
470 
471 static int guest_afu_cr_read16(struct cxl_afu *afu, int cr_idx, u64 offset,
472 			u16 *out)
473 {
474 	int rc;
475 	u64 val;
476 
477 	rc = _guest_afu_cr_readXX(2, afu, cr_idx, offset, &val);
478 	if (!rc)
479 		*out = (u16) val;
480 	return rc;
481 }
482 
483 static int guest_afu_cr_read8(struct cxl_afu *afu, int cr_idx, u64 offset,
484 			u8 *out)
485 {
486 	int rc;
487 	u64 val;
488 
489 	rc = _guest_afu_cr_readXX(1, afu, cr_idx, offset, &val);
490 	if (!rc)
491 		*out = (u8) val;
492 	return rc;
493 }
494 
495 static int guest_afu_cr_read64(struct cxl_afu *afu, int cr_idx, u64 offset,
496 			u64 *out)
497 {
498 	return _guest_afu_cr_readXX(8, afu, cr_idx, offset, out);
499 }
500 
501 static int guest_afu_cr_write32(struct cxl_afu *afu, int cr, u64 off, u32 in)
502 {
503 	/* config record is not writable from guest */
504 	return -EPERM;
505 }
506 
507 static int guest_afu_cr_write16(struct cxl_afu *afu, int cr, u64 off, u16 in)
508 {
509 	/* config record is not writable from guest */
510 	return -EPERM;
511 }
512 
513 static int guest_afu_cr_write8(struct cxl_afu *afu, int cr, u64 off, u8 in)
514 {
515 	/* config record is not writable from guest */
516 	return -EPERM;
517 }
518 
519 static int attach_afu_directed(struct cxl_context *ctx, u64 wed, u64 amr)
520 {
521 	struct cxl_process_element_hcall *elem;
522 	struct cxl *adapter = ctx->afu->adapter;
523 	const struct cred *cred;
524 	u32 pid, idx;
525 	int rc, r, i;
526 	u64 mmio_addr, mmio_size;
527 	__be64 flags = 0;
528 
529 	/* Must be 8 byte aligned and cannot cross a 4096 byte boundary */
530 	if (!(elem = (struct cxl_process_element_hcall *)
531 			get_zeroed_page(GFP_KERNEL)))
532 		return -ENOMEM;
533 
534 	elem->version = cpu_to_be64(CXL_PROCESS_ELEMENT_VERSION);
535 	if (ctx->kernel) {
536 		pid = 0;
537 		flags |= CXL_PE_TRANSLATION_ENABLED;
538 		flags |= CXL_PE_PRIVILEGED_PROCESS;
539 		if (mfmsr() & MSR_SF)
540 			flags |= CXL_PE_64_BIT;
541 	} else {
542 		pid = current->pid;
543 		flags |= CXL_PE_PROBLEM_STATE;
544 		flags |= CXL_PE_TRANSLATION_ENABLED;
545 		if (!test_tsk_thread_flag(current, TIF_32BIT))
546 			flags |= CXL_PE_64_BIT;
547 		cred = get_current_cred();
548 		if (uid_eq(cred->euid, GLOBAL_ROOT_UID))
549 			flags |= CXL_PE_PRIVILEGED_PROCESS;
550 		put_cred(cred);
551 	}
552 	elem->flags         = cpu_to_be64(flags);
553 	elem->common.tid    = cpu_to_be32(0); /* Unused */
554 	elem->common.pid    = cpu_to_be32(pid);
555 	elem->common.csrp   = cpu_to_be64(0); /* disable */
556 	elem->common.u.psl8.aurp0  = cpu_to_be64(0); /* disable */
557 	elem->common.u.psl8.aurp1  = cpu_to_be64(0); /* disable */
558 
559 	cxl_prefault(ctx, wed);
560 
561 	elem->common.u.psl8.sstp0  = cpu_to_be64(ctx->sstp0);
562 	elem->common.u.psl8.sstp1  = cpu_to_be64(ctx->sstp1);
563 
564 	/*
565 	 * Ensure we have at least one interrupt allocated to take faults for
566 	 * kernel contexts that may not have allocated any AFU IRQs at all:
567 	 */
568 	if (ctx->irqs.range[0] == 0) {
569 		rc = afu_register_irqs(ctx, 0);
570 		if (rc)
571 			goto out_free;
572 	}
573 
574 	for (r = 0; r < CXL_IRQ_RANGES; r++) {
575 		for (i = 0; i < ctx->irqs.range[r]; i++) {
576 			if (r == 0 && i == 0) {
577 				elem->pslVirtualIsn = cpu_to_be32(ctx->irqs.offset[0]);
578 			} else {
579 				idx = ctx->irqs.offset[r] + i - adapter->guest->irq_base_offset;
580 				elem->applicationVirtualIsnBitmap[idx / 8] |= 0x80 >> (idx % 8);
581 			}
582 		}
583 	}
584 	elem->common.amr = cpu_to_be64(amr);
585 	elem->common.wed = cpu_to_be64(wed);
586 
587 	disable_afu_irqs(ctx);
588 
589 	rc = cxl_h_attach_process(ctx->afu->guest->handle, elem,
590 				&ctx->process_token, &mmio_addr, &mmio_size);
591 	if (rc == H_SUCCESS) {
592 		if (ctx->master || !ctx->afu->pp_psa) {
593 			ctx->psn_phys = ctx->afu->psn_phys;
594 			ctx->psn_size = ctx->afu->adapter->ps_size;
595 		} else {
596 			ctx->psn_phys = mmio_addr;
597 			ctx->psn_size = mmio_size;
598 		}
599 		if (ctx->afu->pp_psa && mmio_size &&
600 			ctx->afu->pp_size == 0) {
601 			/*
602 			 * There's no property in the device tree to read the
603 			 * pp_size. We only find out at the 1st attach.
604 			 * Compared to bare-metal, it is too late and we
605 			 * should really lock here. However, on powerVM,
606 			 * pp_size is really only used to display in /sys.
607 			 * Being discussed with pHyp for their next release.
608 			 */
609 			ctx->afu->pp_size = mmio_size;
610 		}
611 		/* from PAPR: process element is bytes 4-7 of process token */
612 		ctx->external_pe = ctx->process_token & 0xFFFFFFFF;
613 		pr_devel("CXL pe=%i is known as %i for pHyp, mmio_size=%#llx",
614 			ctx->pe, ctx->external_pe, ctx->psn_size);
615 		ctx->pe_inserted = true;
616 		enable_afu_irqs(ctx);
617 	}
618 
619 out_free:
620 	free_page((u64)elem);
621 	return rc;
622 }
623 
624 static int guest_attach_process(struct cxl_context *ctx, bool kernel, u64 wed, u64 amr)
625 {
626 	pr_devel("in %s\n", __func__);
627 
628 	ctx->kernel = kernel;
629 	if (ctx->afu->current_mode == CXL_MODE_DIRECTED)
630 		return attach_afu_directed(ctx, wed, amr);
631 
632 	/* dedicated mode not supported on FW840 */
633 
634 	return -EINVAL;
635 }
636 
637 static int detach_afu_directed(struct cxl_context *ctx)
638 {
639 	if (!ctx->pe_inserted)
640 		return 0;
641 	if (cxl_h_detach_process(ctx->afu->guest->handle, ctx->process_token))
642 		return -1;
643 	return 0;
644 }
645 
646 static int guest_detach_process(struct cxl_context *ctx)
647 {
648 	pr_devel("in %s\n", __func__);
649 	trace_cxl_detach(ctx);
650 
651 	if (!cxl_ops->link_ok(ctx->afu->adapter, ctx->afu))
652 		return -EIO;
653 
654 	if (ctx->afu->current_mode == CXL_MODE_DIRECTED)
655 		return detach_afu_directed(ctx);
656 
657 	return -EINVAL;
658 }
659 
660 static void guest_release_afu(struct device *dev)
661 {
662 	struct cxl_afu *afu = to_cxl_afu(dev);
663 
664 	pr_devel("%s\n", __func__);
665 
666 	idr_destroy(&afu->contexts_idr);
667 
668 	kfree(afu->guest);
669 	kfree(afu);
670 }
671 
672 ssize_t cxl_guest_read_afu_vpd(struct cxl_afu *afu, void *buf, size_t len)
673 {
674 	return guest_collect_vpd(NULL, afu, buf, len);
675 }
676 
677 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
678 static ssize_t guest_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
679 					loff_t off, size_t count)
680 {
681 	void *tbuf = NULL;
682 	int rc = 0;
683 
684 	tbuf = (void *) get_zeroed_page(GFP_KERNEL);
685 	if (!tbuf)
686 		return -ENOMEM;
687 
688 	rc = cxl_h_get_afu_err(afu->guest->handle,
689 			       off & 0x7,
690 			       virt_to_phys(tbuf),
691 			       count);
692 	if (rc)
693 		goto err;
694 
695 	if (count > ERR_BUFF_MAX_COPY_SIZE)
696 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
697 	memcpy(buf, tbuf, count);
698 err:
699 	free_page((u64)tbuf);
700 
701 	return rc;
702 }
703 
704 static int guest_afu_check_and_enable(struct cxl_afu *afu)
705 {
706 	return 0;
707 }
708 
709 static bool guest_support_attributes(const char *attr_name,
710 				     enum cxl_attrs type)
711 {
712 	switch (type) {
713 	case CXL_ADAPTER_ATTRS:
714 		if ((strcmp(attr_name, "base_image") == 0) ||
715 			(strcmp(attr_name, "load_image_on_perst") == 0) ||
716 			(strcmp(attr_name, "perst_reloads_same_image") == 0) ||
717 			(strcmp(attr_name, "image_loaded") == 0))
718 			return false;
719 		break;
720 	case CXL_AFU_MASTER_ATTRS:
721 		if ((strcmp(attr_name, "pp_mmio_off") == 0))
722 			return false;
723 		break;
724 	case CXL_AFU_ATTRS:
725 		break;
726 	default:
727 		break;
728 	}
729 
730 	return true;
731 }
732 
733 static int activate_afu_directed(struct cxl_afu *afu)
734 {
735 	int rc;
736 
737 	dev_info(&afu->dev, "Activating AFU(%d) directed mode\n", afu->slice);
738 
739 	afu->current_mode = CXL_MODE_DIRECTED;
740 
741 	afu->num_procs = afu->max_procs_virtualised;
742 
743 	if ((rc = cxl_chardev_m_afu_add(afu)))
744 		return rc;
745 
746 	if ((rc = cxl_sysfs_afu_m_add(afu)))
747 		goto err;
748 
749 	if ((rc = cxl_chardev_s_afu_add(afu)))
750 		goto err1;
751 
752 	return 0;
753 err1:
754 	cxl_sysfs_afu_m_remove(afu);
755 err:
756 	cxl_chardev_afu_remove(afu);
757 	return rc;
758 }
759 
760 static int guest_afu_activate_mode(struct cxl_afu *afu, int mode)
761 {
762 	if (!mode)
763 		return 0;
764 	if (!(mode & afu->modes_supported))
765 		return -EINVAL;
766 
767 	if (mode == CXL_MODE_DIRECTED)
768 		return activate_afu_directed(afu);
769 
770 	if (mode == CXL_MODE_DEDICATED)
771 		dev_err(&afu->dev, "Dedicated mode not supported\n");
772 
773 	return -EINVAL;
774 }
775 
776 static int deactivate_afu_directed(struct cxl_afu *afu)
777 {
778 	dev_info(&afu->dev, "Deactivating AFU(%d) directed mode\n", afu->slice);
779 
780 	afu->current_mode = 0;
781 	afu->num_procs = 0;
782 
783 	cxl_sysfs_afu_m_remove(afu);
784 	cxl_chardev_afu_remove(afu);
785 
786 	cxl_ops->afu_reset(afu);
787 
788 	return 0;
789 }
790 
791 static int guest_afu_deactivate_mode(struct cxl_afu *afu, int mode)
792 {
793 	if (!mode)
794 		return 0;
795 	if (!(mode & afu->modes_supported))
796 		return -EINVAL;
797 
798 	if (mode == CXL_MODE_DIRECTED)
799 		return deactivate_afu_directed(afu);
800 	return 0;
801 }
802 
803 static int guest_afu_reset(struct cxl_afu *afu)
804 {
805 	pr_devel("AFU(%d) reset request\n", afu->slice);
806 	return cxl_h_reset_afu(afu->guest->handle);
807 }
808 
809 static int guest_map_slice_regs(struct cxl_afu *afu)
810 {
811 	if (!(afu->p2n_mmio = ioremap(afu->guest->p2n_phys, afu->guest->p2n_size))) {
812 		dev_err(&afu->dev, "Error mapping AFU(%d) MMIO regions\n",
813 			afu->slice);
814 		return -ENOMEM;
815 	}
816 	return 0;
817 }
818 
819 static void guest_unmap_slice_regs(struct cxl_afu *afu)
820 {
821 	if (afu->p2n_mmio)
822 		iounmap(afu->p2n_mmio);
823 }
824 
825 static int afu_update_state(struct cxl_afu *afu)
826 {
827 	int rc, cur_state;
828 
829 	rc = afu_read_error_state(afu, &cur_state);
830 	if (rc)
831 		return rc;
832 
833 	if (afu->guest->previous_state == cur_state)
834 		return 0;
835 
836 	pr_devel("AFU(%d) update state to %#x\n", afu->slice, cur_state);
837 
838 	switch (cur_state) {
839 	case H_STATE_NORMAL:
840 		afu->guest->previous_state = cur_state;
841 		break;
842 
843 	case H_STATE_DISABLE:
844 		pci_error_handlers(afu, CXL_ERROR_DETECTED_EVENT,
845 				pci_channel_io_frozen);
846 
847 		cxl_context_detach_all(afu);
848 		if ((rc = cxl_ops->afu_reset(afu)))
849 			pr_devel("reset hcall failed %d\n", rc);
850 
851 		rc = afu_read_error_state(afu, &cur_state);
852 		if (!rc && cur_state == H_STATE_NORMAL) {
853 			pci_error_handlers(afu, CXL_SLOT_RESET_EVENT,
854 					pci_channel_io_normal);
855 			pci_error_handlers(afu, CXL_RESUME_EVENT, 0);
856 		}
857 		afu->guest->previous_state = 0;
858 		break;
859 
860 	case H_STATE_TEMP_UNAVAILABLE:
861 		afu->guest->previous_state = cur_state;
862 		break;
863 
864 	case H_STATE_PERM_UNAVAILABLE:
865 		dev_err(&afu->dev, "AFU is in permanent error state\n");
866 		pci_error_handlers(afu, CXL_ERROR_DETECTED_EVENT,
867 				pci_channel_io_perm_failure);
868 		afu->guest->previous_state = cur_state;
869 		break;
870 
871 	default:
872 		pr_err("Unexpected AFU(%d) error state: %#x\n",
873 		       afu->slice, cur_state);
874 		return -EINVAL;
875 	}
876 
877 	return rc;
878 }
879 
880 static void afu_handle_errstate(struct work_struct *work)
881 {
882 	struct cxl_afu_guest *afu_guest =
883 		container_of(to_delayed_work(work), struct cxl_afu_guest, work_err);
884 
885 	if (!afu_update_state(afu_guest->parent) &&
886 	    afu_guest->previous_state == H_STATE_PERM_UNAVAILABLE)
887 		return;
888 
889 	if (afu_guest->handle_err)
890 		schedule_delayed_work(&afu_guest->work_err,
891 				      msecs_to_jiffies(3000));
892 }
893 
894 static bool guest_link_ok(struct cxl *cxl, struct cxl_afu *afu)
895 {
896 	int state;
897 
898 	if (afu && (!afu_read_error_state(afu, &state))) {
899 		if (state == H_STATE_NORMAL)
900 			return true;
901 	}
902 
903 	return false;
904 }
905 
906 static int afu_properties_look_ok(struct cxl_afu *afu)
907 {
908 	if (afu->pp_irqs < 0) {
909 		dev_err(&afu->dev, "Unexpected per-process minimum interrupt value\n");
910 		return -EINVAL;
911 	}
912 
913 	if (afu->max_procs_virtualised < 1) {
914 		dev_err(&afu->dev, "Unexpected max number of processes virtualised value\n");
915 		return -EINVAL;
916 	}
917 
918 	return 0;
919 }
920 
921 int cxl_guest_init_afu(struct cxl *adapter, int slice, struct device_node *afu_np)
922 {
923 	struct cxl_afu *afu;
924 	bool free = true;
925 	int rc;
926 
927 	pr_devel("in %s - AFU(%d)\n", __func__, slice);
928 	if (!(afu = cxl_alloc_afu(adapter, slice)))
929 		return -ENOMEM;
930 
931 	if (!(afu->guest = kzalloc(sizeof(struct cxl_afu_guest), GFP_KERNEL))) {
932 		kfree(afu);
933 		return -ENOMEM;
934 	}
935 
936 	if ((rc = dev_set_name(&afu->dev, "afu%i.%i",
937 					  adapter->adapter_num,
938 					  slice)))
939 		goto err1;
940 
941 	adapter->slices++;
942 
943 	if ((rc = cxl_of_read_afu_handle(afu, afu_np)))
944 		goto err1;
945 
946 	if ((rc = cxl_ops->afu_reset(afu)))
947 		goto err1;
948 
949 	if ((rc = cxl_of_read_afu_properties(afu, afu_np)))
950 		goto err1;
951 
952 	if ((rc = afu_properties_look_ok(afu)))
953 		goto err1;
954 
955 	if ((rc = guest_map_slice_regs(afu)))
956 		goto err1;
957 
958 	if ((rc = guest_register_serr_irq(afu)))
959 		goto err2;
960 
961 	/*
962 	 * After we call this function we must not free the afu directly, even
963 	 * if it returns an error!
964 	 */
965 	if ((rc = cxl_register_afu(afu)))
966 		goto err_put1;
967 
968 	if ((rc = cxl_sysfs_afu_add(afu)))
969 		goto err_put1;
970 
971 	/*
972 	 * pHyp doesn't expose the programming models supported by the
973 	 * AFU. pHyp currently only supports directed mode. If it adds
974 	 * dedicated mode later, this version of cxl has no way to
975 	 * detect it. So we'll initialize the driver, but the first
976 	 * attach will fail.
977 	 * Being discussed with pHyp to do better (likely new property)
978 	 */
979 	if (afu->max_procs_virtualised == 1)
980 		afu->modes_supported = CXL_MODE_DEDICATED;
981 	else
982 		afu->modes_supported = CXL_MODE_DIRECTED;
983 
984 	if ((rc = cxl_afu_select_best_mode(afu)))
985 		goto err_put2;
986 
987 	adapter->afu[afu->slice] = afu;
988 
989 	afu->enabled = true;
990 
991 	/*
992 	 * wake up the cpu periodically to check the state
993 	 * of the AFU using "afu" stored in the guest structure.
994 	 */
995 	afu->guest->parent = afu;
996 	afu->guest->handle_err = true;
997 	INIT_DELAYED_WORK(&afu->guest->work_err, afu_handle_errstate);
998 	schedule_delayed_work(&afu->guest->work_err, msecs_to_jiffies(1000));
999 
1000 	if ((rc = cxl_pci_vphb_add(afu)))
1001 		dev_info(&afu->dev, "Can't register vPHB\n");
1002 
1003 	return 0;
1004 
1005 err_put2:
1006 	cxl_sysfs_afu_remove(afu);
1007 err_put1:
1008 	device_unregister(&afu->dev);
1009 	free = false;
1010 	guest_release_serr_irq(afu);
1011 err2:
1012 	guest_unmap_slice_regs(afu);
1013 err1:
1014 	if (free) {
1015 		kfree(afu->guest);
1016 		kfree(afu);
1017 	}
1018 	return rc;
1019 }
1020 
1021 void cxl_guest_remove_afu(struct cxl_afu *afu)
1022 {
1023 	if (!afu)
1024 		return;
1025 
1026 	/* flush and stop pending job */
1027 	afu->guest->handle_err = false;
1028 	flush_delayed_work(&afu->guest->work_err);
1029 
1030 	cxl_pci_vphb_remove(afu);
1031 	cxl_sysfs_afu_remove(afu);
1032 
1033 	spin_lock(&afu->adapter->afu_list_lock);
1034 	afu->adapter->afu[afu->slice] = NULL;
1035 	spin_unlock(&afu->adapter->afu_list_lock);
1036 
1037 	cxl_context_detach_all(afu);
1038 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1039 	guest_release_serr_irq(afu);
1040 	guest_unmap_slice_regs(afu);
1041 
1042 	device_unregister(&afu->dev);
1043 }
1044 
1045 static void free_adapter(struct cxl *adapter)
1046 {
1047 	struct irq_avail *cur;
1048 	int i;
1049 
1050 	if (adapter->guest) {
1051 		if (adapter->guest->irq_avail) {
1052 			for (i = 0; i < adapter->guest->irq_nranges; i++) {
1053 				cur = &adapter->guest->irq_avail[i];
1054 				kfree(cur->bitmap);
1055 			}
1056 			kfree(adapter->guest->irq_avail);
1057 		}
1058 		kfree(adapter->guest->status);
1059 		kfree(adapter->guest);
1060 	}
1061 	cxl_remove_adapter_nr(adapter);
1062 	kfree(adapter);
1063 }
1064 
1065 static int properties_look_ok(struct cxl *adapter)
1066 {
1067 	/* The absence of this property means that the operational
1068 	 * status is unknown or okay
1069 	 */
1070 	if (strlen(adapter->guest->status) &&
1071 	    strcmp(adapter->guest->status, "okay")) {
1072 		pr_err("ABORTING:Bad operational status of the device\n");
1073 		return -EINVAL;
1074 	}
1075 
1076 	return 0;
1077 }
1078 
1079 ssize_t cxl_guest_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1080 {
1081 	return guest_collect_vpd(adapter, NULL, buf, len);
1082 }
1083 
1084 void cxl_guest_remove_adapter(struct cxl *adapter)
1085 {
1086 	pr_devel("in %s\n", __func__);
1087 
1088 	cxl_sysfs_adapter_remove(adapter);
1089 
1090 	cxl_guest_remove_chardev(adapter);
1091 	device_unregister(&adapter->dev);
1092 }
1093 
1094 static void release_adapter(struct device *dev)
1095 {
1096 	free_adapter(to_cxl_adapter(dev));
1097 }
1098 
1099 struct cxl *cxl_guest_init_adapter(struct device_node *np, struct platform_device *pdev)
1100 {
1101 	struct cxl *adapter;
1102 	bool free = true;
1103 	int rc;
1104 
1105 	if (!(adapter = cxl_alloc_adapter()))
1106 		return ERR_PTR(-ENOMEM);
1107 
1108 	if (!(adapter->guest = kzalloc(sizeof(struct cxl_guest), GFP_KERNEL))) {
1109 		free_adapter(adapter);
1110 		return ERR_PTR(-ENOMEM);
1111 	}
1112 
1113 	adapter->slices = 0;
1114 	adapter->guest->pdev = pdev;
1115 	adapter->dev.parent = &pdev->dev;
1116 	adapter->dev.release = release_adapter;
1117 	dev_set_drvdata(&pdev->dev, adapter);
1118 
1119 	/*
1120 	 * Hypervisor controls PSL timebase initialization (p1 register).
1121 	 * On FW840, PSL is initialized.
1122 	 */
1123 	adapter->psl_timebase_synced = true;
1124 
1125 	if ((rc = cxl_of_read_adapter_handle(adapter, np)))
1126 		goto err1;
1127 
1128 	if ((rc = cxl_of_read_adapter_properties(adapter, np)))
1129 		goto err1;
1130 
1131 	if ((rc = properties_look_ok(adapter)))
1132 		goto err1;
1133 
1134 	if ((rc = cxl_guest_add_chardev(adapter)))
1135 		goto err1;
1136 
1137 	/*
1138 	 * After we call this function we must not free the adapter directly,
1139 	 * even if it returns an error!
1140 	 */
1141 	if ((rc = cxl_register_adapter(adapter)))
1142 		goto err_put1;
1143 
1144 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1145 		goto err_put1;
1146 
1147 	/* release the context lock as the adapter is configured */
1148 	cxl_adapter_context_unlock(adapter);
1149 
1150 	return adapter;
1151 
1152 err_put1:
1153 	device_unregister(&adapter->dev);
1154 	free = false;
1155 	cxl_guest_remove_chardev(adapter);
1156 err1:
1157 	if (free)
1158 		free_adapter(adapter);
1159 	return ERR_PTR(rc);
1160 }
1161 
1162 void cxl_guest_reload_module(struct cxl *adapter)
1163 {
1164 	struct platform_device *pdev;
1165 
1166 	pdev = adapter->guest->pdev;
1167 	cxl_guest_remove_adapter(adapter);
1168 
1169 	cxl_of_probe(pdev);
1170 }
1171 
1172 const struct cxl_backend_ops cxl_guest_ops = {
1173 	.module = THIS_MODULE,
1174 	.adapter_reset = guest_reset,
1175 	.alloc_one_irq = guest_alloc_one_irq,
1176 	.release_one_irq = guest_release_one_irq,
1177 	.alloc_irq_ranges = guest_alloc_irq_ranges,
1178 	.release_irq_ranges = guest_release_irq_ranges,
1179 	.setup_irq = NULL,
1180 	.handle_psl_slice_error = guest_handle_psl_slice_error,
1181 	.psl_interrupt = guest_psl_irq,
1182 	.ack_irq = guest_ack_irq,
1183 	.attach_process = guest_attach_process,
1184 	.detach_process = guest_detach_process,
1185 	.update_ivtes = NULL,
1186 	.support_attributes = guest_support_attributes,
1187 	.link_ok = guest_link_ok,
1188 	.release_afu = guest_release_afu,
1189 	.afu_read_err_buffer = guest_afu_read_err_buffer,
1190 	.afu_check_and_enable = guest_afu_check_and_enable,
1191 	.afu_activate_mode = guest_afu_activate_mode,
1192 	.afu_deactivate_mode = guest_afu_deactivate_mode,
1193 	.afu_reset = guest_afu_reset,
1194 	.afu_cr_read8 = guest_afu_cr_read8,
1195 	.afu_cr_read16 = guest_afu_cr_read16,
1196 	.afu_cr_read32 = guest_afu_cr_read32,
1197 	.afu_cr_read64 = guest_afu_cr_read64,
1198 	.afu_cr_write8 = guest_afu_cr_write8,
1199 	.afu_cr_write16 = guest_afu_cr_write16,
1200 	.afu_cr_write32 = guest_afu_cr_write32,
1201 	.read_adapter_vpd = cxl_guest_read_adapter_vpd,
1202 };
1203