xref: /openbmc/linux/drivers/misc/cxl/native.c (revision 60772e48)
1 /*
2  * Copyright 2014 IBM Corp.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public License
6  * as published by the Free Software Foundation; either version
7  * 2 of the License, or (at your option) any later version.
8  */
9 
10 #include <linux/spinlock.h>
11 #include <linux/sched.h>
12 #include <linux/sched/clock.h>
13 #include <linux/slab.h>
14 #include <linux/mutex.h>
15 #include <linux/mm.h>
16 #include <linux/uaccess.h>
17 #include <linux/delay.h>
18 #include <asm/synch.h>
19 #include <asm/switch_to.h>
20 #include <misc/cxl-base.h>
21 
22 #include "cxl.h"
23 #include "trace.h"
24 
25 static int afu_control(struct cxl_afu *afu, u64 command, u64 clear,
26 		       u64 result, u64 mask, bool enabled)
27 {
28 	u64 AFU_Cntl;
29 	unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
30 	int rc = 0;
31 
32 	spin_lock(&afu->afu_cntl_lock);
33 	pr_devel("AFU command starting: %llx\n", command);
34 
35 	trace_cxl_afu_ctrl(afu, command);
36 
37 	AFU_Cntl = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
38 	cxl_p2n_write(afu, CXL_AFU_Cntl_An, (AFU_Cntl & ~clear) | command);
39 
40 	AFU_Cntl = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
41 	while ((AFU_Cntl & mask) != result) {
42 		if (time_after_eq(jiffies, timeout)) {
43 			dev_warn(&afu->dev, "WARNING: AFU control timed out!\n");
44 			rc = -EBUSY;
45 			goto out;
46 		}
47 
48 		if (!cxl_ops->link_ok(afu->adapter, afu)) {
49 			afu->enabled = enabled;
50 			rc = -EIO;
51 			goto out;
52 		}
53 
54 		pr_devel_ratelimited("AFU control... (0x%016llx)\n",
55 				     AFU_Cntl | command);
56 		cpu_relax();
57 		AFU_Cntl = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
58 	}
59 
60 	if (AFU_Cntl & CXL_AFU_Cntl_An_RA) {
61 		/*
62 		 * Workaround for a bug in the XSL used in the Mellanox CX4
63 		 * that fails to clear the RA bit after an AFU reset,
64 		 * preventing subsequent AFU resets from working.
65 		 */
66 		cxl_p2n_write(afu, CXL_AFU_Cntl_An, AFU_Cntl & ~CXL_AFU_Cntl_An_RA);
67 	}
68 
69 	pr_devel("AFU command complete: %llx\n", command);
70 	afu->enabled = enabled;
71 out:
72 	trace_cxl_afu_ctrl_done(afu, command, rc);
73 	spin_unlock(&afu->afu_cntl_lock);
74 
75 	return rc;
76 }
77 
78 static int afu_enable(struct cxl_afu *afu)
79 {
80 	pr_devel("AFU enable request\n");
81 
82 	return afu_control(afu, CXL_AFU_Cntl_An_E, 0,
83 			   CXL_AFU_Cntl_An_ES_Enabled,
84 			   CXL_AFU_Cntl_An_ES_MASK, true);
85 }
86 
87 int cxl_afu_disable(struct cxl_afu *afu)
88 {
89 	pr_devel("AFU disable request\n");
90 
91 	return afu_control(afu, 0, CXL_AFU_Cntl_An_E,
92 			   CXL_AFU_Cntl_An_ES_Disabled,
93 			   CXL_AFU_Cntl_An_ES_MASK, false);
94 }
95 
96 /* This will disable as well as reset */
97 static int native_afu_reset(struct cxl_afu *afu)
98 {
99 	int rc;
100 	u64 serr;
101 
102 	pr_devel("AFU reset request\n");
103 
104 	rc = afu_control(afu, CXL_AFU_Cntl_An_RA, 0,
105 			   CXL_AFU_Cntl_An_RS_Complete | CXL_AFU_Cntl_An_ES_Disabled,
106 			   CXL_AFU_Cntl_An_RS_MASK | CXL_AFU_Cntl_An_ES_MASK,
107 			   false);
108 
109 	/*
110 	 * Re-enable any masked interrupts when the AFU is not
111 	 * activated to avoid side effects after attaching a process
112 	 * in dedicated mode.
113 	 */
114 	if (afu->current_mode == 0) {
115 		serr = cxl_p1n_read(afu, CXL_PSL_SERR_An);
116 		serr &= ~CXL_PSL_SERR_An_IRQ_MASKS;
117 		cxl_p1n_write(afu, CXL_PSL_SERR_An, serr);
118 	}
119 
120 	return rc;
121 }
122 
123 static int native_afu_check_and_enable(struct cxl_afu *afu)
124 {
125 	if (!cxl_ops->link_ok(afu->adapter, afu)) {
126 		WARN(1, "Refusing to enable afu while link down!\n");
127 		return -EIO;
128 	}
129 	if (afu->enabled)
130 		return 0;
131 	return afu_enable(afu);
132 }
133 
134 int cxl_psl_purge(struct cxl_afu *afu)
135 {
136 	u64 PSL_CNTL = cxl_p1n_read(afu, CXL_PSL_SCNTL_An);
137 	u64 AFU_Cntl = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
138 	u64 dsisr, dar;
139 	u64 start, end;
140 	u64 trans_fault = 0x0ULL;
141 	unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
142 	int rc = 0;
143 
144 	trace_cxl_psl_ctrl(afu, CXL_PSL_SCNTL_An_Pc);
145 
146 	pr_devel("PSL purge request\n");
147 
148 	if (cxl_is_power8())
149 		trans_fault = CXL_PSL_DSISR_TRANS;
150 	if (cxl_is_power9())
151 		trans_fault = CXL_PSL9_DSISR_An_TF;
152 
153 	if (!cxl_ops->link_ok(afu->adapter, afu)) {
154 		dev_warn(&afu->dev, "PSL Purge called with link down, ignoring\n");
155 		rc = -EIO;
156 		goto out;
157 	}
158 
159 	if ((AFU_Cntl & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
160 		WARN(1, "psl_purge request while AFU not disabled!\n");
161 		cxl_afu_disable(afu);
162 	}
163 
164 	cxl_p1n_write(afu, CXL_PSL_SCNTL_An,
165 		       PSL_CNTL | CXL_PSL_SCNTL_An_Pc);
166 	start = local_clock();
167 	PSL_CNTL = cxl_p1n_read(afu, CXL_PSL_SCNTL_An);
168 	while ((PSL_CNTL &  CXL_PSL_SCNTL_An_Ps_MASK)
169 			== CXL_PSL_SCNTL_An_Ps_Pending) {
170 		if (time_after_eq(jiffies, timeout)) {
171 			dev_warn(&afu->dev, "WARNING: PSL Purge timed out!\n");
172 			rc = -EBUSY;
173 			goto out;
174 		}
175 		if (!cxl_ops->link_ok(afu->adapter, afu)) {
176 			rc = -EIO;
177 			goto out;
178 		}
179 
180 		dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
181 		pr_devel_ratelimited("PSL purging... PSL_CNTL: 0x%016llx  PSL_DSISR: 0x%016llx\n",
182 				     PSL_CNTL, dsisr);
183 
184 		if (dsisr & trans_fault) {
185 			dar = cxl_p2n_read(afu, CXL_PSL_DAR_An);
186 			dev_notice(&afu->dev, "PSL purge terminating pending translation, DSISR: 0x%016llx, DAR: 0x%016llx\n",
187 				   dsisr, dar);
188 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
189 		} else if (dsisr) {
190 			dev_notice(&afu->dev, "PSL purge acknowledging pending non-translation fault, DSISR: 0x%016llx\n",
191 				   dsisr);
192 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
193 		} else {
194 			cpu_relax();
195 		}
196 		PSL_CNTL = cxl_p1n_read(afu, CXL_PSL_SCNTL_An);
197 	}
198 	end = local_clock();
199 	pr_devel("PSL purged in %lld ns\n", end - start);
200 
201 	cxl_p1n_write(afu, CXL_PSL_SCNTL_An,
202 		       PSL_CNTL & ~CXL_PSL_SCNTL_An_Pc);
203 out:
204 	trace_cxl_psl_ctrl_done(afu, CXL_PSL_SCNTL_An_Pc, rc);
205 	return rc;
206 }
207 
208 static int spa_max_procs(int spa_size)
209 {
210 	/*
211 	 * From the CAIA:
212 	 *    end_of_SPA_area = SPA_Base + ((n+4) * 128) + (( ((n*8) + 127) >> 7) * 128) + 255
213 	 * Most of that junk is really just an overly-complicated way of saying
214 	 * the last 256 bytes are __aligned(128), so it's really:
215 	 *    end_of_SPA_area = end_of_PSL_queue_area + __aligned(128) 255
216 	 * and
217 	 *    end_of_PSL_queue_area = SPA_Base + ((n+4) * 128) + (n*8) - 1
218 	 * so
219 	 *    sizeof(SPA) = ((n+4) * 128) + (n*8) + __aligned(128) 256
220 	 * Ignore the alignment (which is safe in this case as long as we are
221 	 * careful with our rounding) and solve for n:
222 	 */
223 	return ((spa_size / 8) - 96) / 17;
224 }
225 
226 static int cxl_alloc_spa(struct cxl_afu *afu, int mode)
227 {
228 	unsigned spa_size;
229 
230 	/* Work out how many pages to allocate */
231 	afu->native->spa_order = -1;
232 	do {
233 		afu->native->spa_order++;
234 		spa_size = (1 << afu->native->spa_order) * PAGE_SIZE;
235 
236 		if (spa_size > 0x100000) {
237 			dev_warn(&afu->dev, "num_of_processes too large for the SPA, limiting to %i (0x%x)\n",
238 					afu->native->spa_max_procs, afu->native->spa_size);
239 			if (mode != CXL_MODE_DEDICATED)
240 				afu->num_procs = afu->native->spa_max_procs;
241 			break;
242 		}
243 
244 		afu->native->spa_size = spa_size;
245 		afu->native->spa_max_procs = spa_max_procs(afu->native->spa_size);
246 	} while (afu->native->spa_max_procs < afu->num_procs);
247 
248 	if (!(afu->native->spa = (struct cxl_process_element *)
249 	      __get_free_pages(GFP_KERNEL | __GFP_ZERO, afu->native->spa_order))) {
250 		pr_err("cxl_alloc_spa: Unable to allocate scheduled process area\n");
251 		return -ENOMEM;
252 	}
253 	pr_devel("spa pages: %i afu->spa_max_procs: %i   afu->num_procs: %i\n",
254 		 1<<afu->native->spa_order, afu->native->spa_max_procs, afu->num_procs);
255 
256 	return 0;
257 }
258 
259 static void attach_spa(struct cxl_afu *afu)
260 {
261 	u64 spap;
262 
263 	afu->native->sw_command_status = (__be64 *)((char *)afu->native->spa +
264 					    ((afu->native->spa_max_procs + 3) * 128));
265 
266 	spap = virt_to_phys(afu->native->spa) & CXL_PSL_SPAP_Addr;
267 	spap |= ((afu->native->spa_size >> (12 - CXL_PSL_SPAP_Size_Shift)) - 1) & CXL_PSL_SPAP_Size;
268 	spap |= CXL_PSL_SPAP_V;
269 	pr_devel("cxl: SPA allocated at 0x%p. Max processes: %i, sw_command_status: 0x%p CXL_PSL_SPAP_An=0x%016llx\n",
270 		afu->native->spa, afu->native->spa_max_procs,
271 		afu->native->sw_command_status, spap);
272 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, spap);
273 }
274 
275 static inline void detach_spa(struct cxl_afu *afu)
276 {
277 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0);
278 }
279 
280 void cxl_release_spa(struct cxl_afu *afu)
281 {
282 	if (afu->native->spa) {
283 		free_pages((unsigned long) afu->native->spa,
284 			afu->native->spa_order);
285 		afu->native->spa = NULL;
286 	}
287 }
288 
289 /*
290  * Invalidation of all ERAT entries is no longer required by CAIA2. Use
291  * only for debug.
292  */
293 int cxl_invalidate_all_psl9(struct cxl *adapter)
294 {
295 	unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
296 	u64 ierat;
297 
298 	pr_devel("CXL adapter - invalidation of all ERAT entries\n");
299 
300 	/* Invalidates all ERAT entries for Radix or HPT */
301 	ierat = CXL_XSL9_IERAT_IALL;
302 	if (radix_enabled())
303 		ierat |= CXL_XSL9_IERAT_INVR;
304 	cxl_p1_write(adapter, CXL_XSL9_IERAT, ierat);
305 
306 	while (cxl_p1_read(adapter, CXL_XSL9_IERAT) & CXL_XSL9_IERAT_IINPROG) {
307 		if (time_after_eq(jiffies, timeout)) {
308 			dev_warn(&adapter->dev,
309 			"WARNING: CXL adapter invalidation of all ERAT entries timed out!\n");
310 			return -EBUSY;
311 		}
312 		if (!cxl_ops->link_ok(adapter, NULL))
313 			return -EIO;
314 		cpu_relax();
315 	}
316 	return 0;
317 }
318 
319 int cxl_invalidate_all_psl8(struct cxl *adapter)
320 {
321 	unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
322 
323 	pr_devel("CXL adapter wide TLBIA & SLBIA\n");
324 
325 	cxl_p1_write(adapter, CXL_PSL_AFUSEL, CXL_PSL_AFUSEL_A);
326 
327 	cxl_p1_write(adapter, CXL_PSL_TLBIA, CXL_TLB_SLB_IQ_ALL);
328 	while (cxl_p1_read(adapter, CXL_PSL_TLBIA) & CXL_TLB_SLB_P) {
329 		if (time_after_eq(jiffies, timeout)) {
330 			dev_warn(&adapter->dev, "WARNING: CXL adapter wide TLBIA timed out!\n");
331 			return -EBUSY;
332 		}
333 		if (!cxl_ops->link_ok(adapter, NULL))
334 			return -EIO;
335 		cpu_relax();
336 	}
337 
338 	cxl_p1_write(adapter, CXL_PSL_SLBIA, CXL_TLB_SLB_IQ_ALL);
339 	while (cxl_p1_read(adapter, CXL_PSL_SLBIA) & CXL_TLB_SLB_P) {
340 		if (time_after_eq(jiffies, timeout)) {
341 			dev_warn(&adapter->dev, "WARNING: CXL adapter wide SLBIA timed out!\n");
342 			return -EBUSY;
343 		}
344 		if (!cxl_ops->link_ok(adapter, NULL))
345 			return -EIO;
346 		cpu_relax();
347 	}
348 	return 0;
349 }
350 
351 int cxl_data_cache_flush(struct cxl *adapter)
352 {
353 	u64 reg;
354 	unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
355 
356 	pr_devel("Flushing data cache\n");
357 
358 	reg = cxl_p1_read(adapter, CXL_PSL_Control);
359 	reg |= CXL_PSL_Control_Fr;
360 	cxl_p1_write(adapter, CXL_PSL_Control, reg);
361 
362 	reg = cxl_p1_read(adapter, CXL_PSL_Control);
363 	while ((reg & CXL_PSL_Control_Fs_MASK) != CXL_PSL_Control_Fs_Complete) {
364 		if (time_after_eq(jiffies, timeout)) {
365 			dev_warn(&adapter->dev, "WARNING: cache flush timed out!\n");
366 			return -EBUSY;
367 		}
368 
369 		if (!cxl_ops->link_ok(adapter, NULL)) {
370 			dev_warn(&adapter->dev, "WARNING: link down when flushing cache\n");
371 			return -EIO;
372 		}
373 		cpu_relax();
374 		reg = cxl_p1_read(adapter, CXL_PSL_Control);
375 	}
376 
377 	reg &= ~CXL_PSL_Control_Fr;
378 	cxl_p1_write(adapter, CXL_PSL_Control, reg);
379 	return 0;
380 }
381 
382 static int cxl_write_sstp(struct cxl_afu *afu, u64 sstp0, u64 sstp1)
383 {
384 	int rc;
385 
386 	/* 1. Disable SSTP by writing 0 to SSTP1[V] */
387 	cxl_p2n_write(afu, CXL_SSTP1_An, 0);
388 
389 	/* 2. Invalidate all SLB entries */
390 	if ((rc = cxl_afu_slbia(afu)))
391 		return rc;
392 
393 	/* 3. Set SSTP0_An */
394 	cxl_p2n_write(afu, CXL_SSTP0_An, sstp0);
395 
396 	/* 4. Set SSTP1_An */
397 	cxl_p2n_write(afu, CXL_SSTP1_An, sstp1);
398 
399 	return 0;
400 }
401 
402 /* Using per slice version may improve performance here. (ie. SLBIA_An) */
403 static void slb_invalid(struct cxl_context *ctx)
404 {
405 	struct cxl *adapter = ctx->afu->adapter;
406 	u64 slbia;
407 
408 	WARN_ON(!mutex_is_locked(&ctx->afu->native->spa_mutex));
409 
410 	cxl_p1_write(adapter, CXL_PSL_LBISEL,
411 			((u64)be32_to_cpu(ctx->elem->common.pid) << 32) |
412 			be32_to_cpu(ctx->elem->lpid));
413 	cxl_p1_write(adapter, CXL_PSL_SLBIA, CXL_TLB_SLB_IQ_LPIDPID);
414 
415 	while (1) {
416 		if (!cxl_ops->link_ok(adapter, NULL))
417 			break;
418 		slbia = cxl_p1_read(adapter, CXL_PSL_SLBIA);
419 		if (!(slbia & CXL_TLB_SLB_P))
420 			break;
421 		cpu_relax();
422 	}
423 }
424 
425 static int do_process_element_cmd(struct cxl_context *ctx,
426 				  u64 cmd, u64 pe_state)
427 {
428 	u64 state;
429 	unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
430 	int rc = 0;
431 
432 	trace_cxl_llcmd(ctx, cmd);
433 
434 	WARN_ON(!ctx->afu->enabled);
435 
436 	ctx->elem->software_state = cpu_to_be32(pe_state);
437 	smp_wmb();
438 	*(ctx->afu->native->sw_command_status) = cpu_to_be64(cmd | 0 | ctx->pe);
439 	smp_mb();
440 	cxl_p1n_write(ctx->afu, CXL_PSL_LLCMD_An, cmd | ctx->pe);
441 	while (1) {
442 		if (time_after_eq(jiffies, timeout)) {
443 			dev_warn(&ctx->afu->dev, "WARNING: Process Element Command timed out!\n");
444 			rc = -EBUSY;
445 			goto out;
446 		}
447 		if (!cxl_ops->link_ok(ctx->afu->adapter, ctx->afu)) {
448 			dev_warn(&ctx->afu->dev, "WARNING: Device link down, aborting Process Element Command!\n");
449 			rc = -EIO;
450 			goto out;
451 		}
452 		state = be64_to_cpup(ctx->afu->native->sw_command_status);
453 		if (state == ~0ULL) {
454 			pr_err("cxl: Error adding process element to AFU\n");
455 			rc = -1;
456 			goto out;
457 		}
458 		if ((state & (CXL_SPA_SW_CMD_MASK | CXL_SPA_SW_STATE_MASK  | CXL_SPA_SW_LINK_MASK)) ==
459 		    (cmd | (cmd >> 16) | ctx->pe))
460 			break;
461 		/*
462 		 * The command won't finish in the PSL if there are
463 		 * outstanding DSIs.  Hence we need to yield here in
464 		 * case there are outstanding DSIs that we need to
465 		 * service.  Tuning possiblity: we could wait for a
466 		 * while before sched
467 		 */
468 		schedule();
469 
470 	}
471 out:
472 	trace_cxl_llcmd_done(ctx, cmd, rc);
473 	return rc;
474 }
475 
476 static int add_process_element(struct cxl_context *ctx)
477 {
478 	int rc = 0;
479 
480 	mutex_lock(&ctx->afu->native->spa_mutex);
481 	pr_devel("%s Adding pe: %i started\n", __func__, ctx->pe);
482 	if (!(rc = do_process_element_cmd(ctx, CXL_SPA_SW_CMD_ADD, CXL_PE_SOFTWARE_STATE_V)))
483 		ctx->pe_inserted = true;
484 	pr_devel("%s Adding pe: %i finished\n", __func__, ctx->pe);
485 	mutex_unlock(&ctx->afu->native->spa_mutex);
486 	return rc;
487 }
488 
489 static int terminate_process_element(struct cxl_context *ctx)
490 {
491 	int rc = 0;
492 
493 	/* fast path terminate if it's already invalid */
494 	if (!(ctx->elem->software_state & cpu_to_be32(CXL_PE_SOFTWARE_STATE_V)))
495 		return rc;
496 
497 	mutex_lock(&ctx->afu->native->spa_mutex);
498 	pr_devel("%s Terminate pe: %i started\n", __func__, ctx->pe);
499 	/* We could be asked to terminate when the hw is down. That
500 	 * should always succeed: it's not running if the hw has gone
501 	 * away and is being reset.
502 	 */
503 	if (cxl_ops->link_ok(ctx->afu->adapter, ctx->afu))
504 		rc = do_process_element_cmd(ctx, CXL_SPA_SW_CMD_TERMINATE,
505 					    CXL_PE_SOFTWARE_STATE_V | CXL_PE_SOFTWARE_STATE_T);
506 	ctx->elem->software_state = 0;	/* Remove Valid bit */
507 	pr_devel("%s Terminate pe: %i finished\n", __func__, ctx->pe);
508 	mutex_unlock(&ctx->afu->native->spa_mutex);
509 	return rc;
510 }
511 
512 static int remove_process_element(struct cxl_context *ctx)
513 {
514 	int rc = 0;
515 
516 	mutex_lock(&ctx->afu->native->spa_mutex);
517 	pr_devel("%s Remove pe: %i started\n", __func__, ctx->pe);
518 
519 	/* We could be asked to remove when the hw is down. Again, if
520 	 * the hw is down, the PE is gone, so we succeed.
521 	 */
522 	if (cxl_ops->link_ok(ctx->afu->adapter, ctx->afu))
523 		rc = do_process_element_cmd(ctx, CXL_SPA_SW_CMD_REMOVE, 0);
524 
525 	if (!rc)
526 		ctx->pe_inserted = false;
527 	if (cxl_is_power8())
528 		slb_invalid(ctx);
529 	pr_devel("%s Remove pe: %i finished\n", __func__, ctx->pe);
530 	mutex_unlock(&ctx->afu->native->spa_mutex);
531 
532 	return rc;
533 }
534 
535 void cxl_assign_psn_space(struct cxl_context *ctx)
536 {
537 	if (!ctx->afu->pp_size || ctx->master) {
538 		ctx->psn_phys = ctx->afu->psn_phys;
539 		ctx->psn_size = ctx->afu->adapter->ps_size;
540 	} else {
541 		ctx->psn_phys = ctx->afu->psn_phys +
542 			(ctx->afu->native->pp_offset + ctx->afu->pp_size * ctx->pe);
543 		ctx->psn_size = ctx->afu->pp_size;
544 	}
545 }
546 
547 static int activate_afu_directed(struct cxl_afu *afu)
548 {
549 	int rc;
550 
551 	dev_info(&afu->dev, "Activating AFU directed mode\n");
552 
553 	afu->num_procs = afu->max_procs_virtualised;
554 	if (afu->native->spa == NULL) {
555 		if (cxl_alloc_spa(afu, CXL_MODE_DIRECTED))
556 			return -ENOMEM;
557 	}
558 	attach_spa(afu);
559 
560 	cxl_p1n_write(afu, CXL_PSL_SCNTL_An, CXL_PSL_SCNTL_An_PM_AFU);
561 	if (cxl_is_power8())
562 		cxl_p1n_write(afu, CXL_PSL_AMOR_An, 0xFFFFFFFFFFFFFFFFULL);
563 	cxl_p1n_write(afu, CXL_PSL_ID_An, CXL_PSL_ID_An_F | CXL_PSL_ID_An_L);
564 
565 	afu->current_mode = CXL_MODE_DIRECTED;
566 
567 	if ((rc = cxl_chardev_m_afu_add(afu)))
568 		return rc;
569 
570 	if ((rc = cxl_sysfs_afu_m_add(afu)))
571 		goto err;
572 
573 	if ((rc = cxl_chardev_s_afu_add(afu)))
574 		goto err1;
575 
576 	return 0;
577 err1:
578 	cxl_sysfs_afu_m_remove(afu);
579 err:
580 	cxl_chardev_afu_remove(afu);
581 	return rc;
582 }
583 
584 #ifdef CONFIG_CPU_LITTLE_ENDIAN
585 #define set_endian(sr) ((sr) |= CXL_PSL_SR_An_LE)
586 #else
587 #define set_endian(sr) ((sr) &= ~(CXL_PSL_SR_An_LE))
588 #endif
589 
590 u64 cxl_calculate_sr(bool master, bool kernel, bool real_mode, bool p9)
591 {
592 	u64 sr = 0;
593 
594 	set_endian(sr);
595 	if (master)
596 		sr |= CXL_PSL_SR_An_MP;
597 	if (mfspr(SPRN_LPCR) & LPCR_TC)
598 		sr |= CXL_PSL_SR_An_TC;
599 	if (kernel) {
600 		if (!real_mode)
601 			sr |= CXL_PSL_SR_An_R;
602 		sr |= (mfmsr() & MSR_SF) | CXL_PSL_SR_An_HV;
603 	} else {
604 		sr |= CXL_PSL_SR_An_PR | CXL_PSL_SR_An_R;
605 		if (radix_enabled())
606 			sr |= CXL_PSL_SR_An_HV;
607 		else
608 			sr &= ~(CXL_PSL_SR_An_HV);
609 		if (!test_tsk_thread_flag(current, TIF_32BIT))
610 			sr |= CXL_PSL_SR_An_SF;
611 	}
612 	if (p9) {
613 		if (radix_enabled())
614 			sr |= CXL_PSL_SR_An_XLAT_ror;
615 		else
616 			sr |= CXL_PSL_SR_An_XLAT_hpt;
617 	}
618 	return sr;
619 }
620 
621 static u64 calculate_sr(struct cxl_context *ctx)
622 {
623 	return cxl_calculate_sr(ctx->master, ctx->kernel, ctx->real_mode,
624 				cxl_is_power9());
625 }
626 
627 static void update_ivtes_directed(struct cxl_context *ctx)
628 {
629 	bool need_update = (ctx->status == STARTED);
630 	int r;
631 
632 	if (need_update) {
633 		WARN_ON(terminate_process_element(ctx));
634 		WARN_ON(remove_process_element(ctx));
635 	}
636 
637 	for (r = 0; r < CXL_IRQ_RANGES; r++) {
638 		ctx->elem->ivte_offsets[r] = cpu_to_be16(ctx->irqs.offset[r]);
639 		ctx->elem->ivte_ranges[r] = cpu_to_be16(ctx->irqs.range[r]);
640 	}
641 
642 	/*
643 	 * Theoretically we could use the update llcmd, instead of a
644 	 * terminate/remove/add (or if an atomic update was required we could
645 	 * do a suspend/update/resume), however it seems there might be issues
646 	 * with the update llcmd on some cards (including those using an XSL on
647 	 * an ASIC) so for now it's safest to go with the commands that are
648 	 * known to work. In the future if we come across a situation where the
649 	 * card may be performing transactions using the same PE while we are
650 	 * doing this update we might need to revisit this.
651 	 */
652 	if (need_update)
653 		WARN_ON(add_process_element(ctx));
654 }
655 
656 static int process_element_entry_psl9(struct cxl_context *ctx, u64 wed, u64 amr)
657 {
658 	u32 pid;
659 	int rc;
660 
661 	cxl_assign_psn_space(ctx);
662 
663 	ctx->elem->ctxtime = 0; /* disable */
664 	ctx->elem->lpid = cpu_to_be32(mfspr(SPRN_LPID));
665 	ctx->elem->haurp = 0; /* disable */
666 
667 	if (ctx->kernel)
668 		pid = 0;
669 	else {
670 		if (ctx->mm == NULL) {
671 			pr_devel("%s: unable to get mm for pe=%d pid=%i\n",
672 				__func__, ctx->pe, pid_nr(ctx->pid));
673 			return -EINVAL;
674 		}
675 		pid = ctx->mm->context.id;
676 	}
677 
678 	/* Assign a unique TIDR (thread id) for the current thread */
679 	if (!(ctx->tidr) && (ctx->assign_tidr)) {
680 		rc = set_thread_tidr(current);
681 		if (rc)
682 			return -ENODEV;
683 		ctx->tidr = current->thread.tidr;
684 		pr_devel("%s: current tidr: %d\n", __func__, ctx->tidr);
685 	}
686 
687 	ctx->elem->common.tid = cpu_to_be32(ctx->tidr);
688 	ctx->elem->common.pid = cpu_to_be32(pid);
689 
690 	ctx->elem->sr = cpu_to_be64(calculate_sr(ctx));
691 
692 	ctx->elem->common.csrp = 0; /* disable */
693 
694 	cxl_prefault(ctx, wed);
695 
696 	/*
697 	 * Ensure we have the multiplexed PSL interrupt set up to take faults
698 	 * for kernel contexts that may not have allocated any AFU IRQs at all:
699 	 */
700 	if (ctx->irqs.range[0] == 0) {
701 		ctx->irqs.offset[0] = ctx->afu->native->psl_hwirq;
702 		ctx->irqs.range[0] = 1;
703 	}
704 
705 	ctx->elem->common.amr = cpu_to_be64(amr);
706 	ctx->elem->common.wed = cpu_to_be64(wed);
707 
708 	return 0;
709 }
710 
711 int cxl_attach_afu_directed_psl9(struct cxl_context *ctx, u64 wed, u64 amr)
712 {
713 	int result;
714 
715 	/* fill the process element entry */
716 	result = process_element_entry_psl9(ctx, wed, amr);
717 	if (result)
718 		return result;
719 
720 	update_ivtes_directed(ctx);
721 
722 	/* first guy needs to enable */
723 	result = cxl_ops->afu_check_and_enable(ctx->afu);
724 	if (result)
725 		return result;
726 
727 	return add_process_element(ctx);
728 }
729 
730 int cxl_attach_afu_directed_psl8(struct cxl_context *ctx, u64 wed, u64 amr)
731 {
732 	u32 pid;
733 	int result;
734 
735 	cxl_assign_psn_space(ctx);
736 
737 	ctx->elem->ctxtime = 0; /* disable */
738 	ctx->elem->lpid = cpu_to_be32(mfspr(SPRN_LPID));
739 	ctx->elem->haurp = 0; /* disable */
740 	ctx->elem->u.sdr = cpu_to_be64(mfspr(SPRN_SDR1));
741 
742 	pid = current->pid;
743 	if (ctx->kernel)
744 		pid = 0;
745 	ctx->elem->common.tid = 0;
746 	ctx->elem->common.pid = cpu_to_be32(pid);
747 
748 	ctx->elem->sr = cpu_to_be64(calculate_sr(ctx));
749 
750 	ctx->elem->common.csrp = 0; /* disable */
751 	ctx->elem->common.u.psl8.aurp0 = 0; /* disable */
752 	ctx->elem->common.u.psl8.aurp1 = 0; /* disable */
753 
754 	cxl_prefault(ctx, wed);
755 
756 	ctx->elem->common.u.psl8.sstp0 = cpu_to_be64(ctx->sstp0);
757 	ctx->elem->common.u.psl8.sstp1 = cpu_to_be64(ctx->sstp1);
758 
759 	/*
760 	 * Ensure we have the multiplexed PSL interrupt set up to take faults
761 	 * for kernel contexts that may not have allocated any AFU IRQs at all:
762 	 */
763 	if (ctx->irqs.range[0] == 0) {
764 		ctx->irqs.offset[0] = ctx->afu->native->psl_hwirq;
765 		ctx->irqs.range[0] = 1;
766 	}
767 
768 	update_ivtes_directed(ctx);
769 
770 	ctx->elem->common.amr = cpu_to_be64(amr);
771 	ctx->elem->common.wed = cpu_to_be64(wed);
772 
773 	/* first guy needs to enable */
774 	if ((result = cxl_ops->afu_check_and_enable(ctx->afu)))
775 		return result;
776 
777 	return add_process_element(ctx);
778 }
779 
780 static int deactivate_afu_directed(struct cxl_afu *afu)
781 {
782 	dev_info(&afu->dev, "Deactivating AFU directed mode\n");
783 
784 	afu->current_mode = 0;
785 	afu->num_procs = 0;
786 
787 	cxl_sysfs_afu_m_remove(afu);
788 	cxl_chardev_afu_remove(afu);
789 
790 	/*
791 	 * The CAIA section 2.2.1 indicates that the procedure for starting and
792 	 * stopping an AFU in AFU directed mode is AFU specific, which is not
793 	 * ideal since this code is generic and with one exception has no
794 	 * knowledge of the AFU. This is in contrast to the procedure for
795 	 * disabling a dedicated process AFU, which is documented to just
796 	 * require a reset. The architecture does indicate that both an AFU
797 	 * reset and an AFU disable should result in the AFU being disabled and
798 	 * we do both followed by a PSL purge for safety.
799 	 *
800 	 * Notably we used to have some issues with the disable sequence on PSL
801 	 * cards, which is why we ended up using this heavy weight procedure in
802 	 * the first place, however a bug was discovered that had rendered the
803 	 * disable operation ineffective, so it is conceivable that was the
804 	 * sole explanation for those difficulties. Careful regression testing
805 	 * is recommended if anyone attempts to remove or reorder these
806 	 * operations.
807 	 *
808 	 * The XSL on the Mellanox CX4 behaves a little differently from the
809 	 * PSL based cards and will time out an AFU reset if the AFU is still
810 	 * enabled. That card is special in that we do have a means to identify
811 	 * it from this code, so in that case we skip the reset and just use a
812 	 * disable/purge to avoid the timeout and corresponding noise in the
813 	 * kernel log.
814 	 */
815 	if (afu->adapter->native->sl_ops->needs_reset_before_disable)
816 		cxl_ops->afu_reset(afu);
817 	cxl_afu_disable(afu);
818 	cxl_psl_purge(afu);
819 
820 	return 0;
821 }
822 
823 int cxl_activate_dedicated_process_psl9(struct cxl_afu *afu)
824 {
825 	dev_info(&afu->dev, "Activating dedicated process mode\n");
826 
827 	/*
828 	 * If XSL is set to dedicated mode (Set in PSL_SCNTL reg), the
829 	 * XSL and AFU are programmed to work with a single context.
830 	 * The context information should be configured in the SPA area
831 	 * index 0 (so PSL_SPAP must be configured before enabling the
832 	 * AFU).
833 	 */
834 	afu->num_procs = 1;
835 	if (afu->native->spa == NULL) {
836 		if (cxl_alloc_spa(afu, CXL_MODE_DEDICATED))
837 			return -ENOMEM;
838 	}
839 	attach_spa(afu);
840 
841 	cxl_p1n_write(afu, CXL_PSL_SCNTL_An, CXL_PSL_SCNTL_An_PM_Process);
842 	cxl_p1n_write(afu, CXL_PSL_ID_An, CXL_PSL_ID_An_F | CXL_PSL_ID_An_L);
843 
844 	afu->current_mode = CXL_MODE_DEDICATED;
845 
846 	return cxl_chardev_d_afu_add(afu);
847 }
848 
849 int cxl_activate_dedicated_process_psl8(struct cxl_afu *afu)
850 {
851 	dev_info(&afu->dev, "Activating dedicated process mode\n");
852 
853 	cxl_p1n_write(afu, CXL_PSL_SCNTL_An, CXL_PSL_SCNTL_An_PM_Process);
854 
855 	cxl_p1n_write(afu, CXL_PSL_CtxTime_An, 0); /* disable */
856 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0);    /* disable */
857 	cxl_p1n_write(afu, CXL_PSL_AMOR_An, 0xFFFFFFFFFFFFFFFFULL);
858 	cxl_p1n_write(afu, CXL_PSL_LPID_An, mfspr(SPRN_LPID));
859 	cxl_p1n_write(afu, CXL_HAURP_An, 0);       /* disable */
860 	cxl_p1n_write(afu, CXL_PSL_SDR_An, mfspr(SPRN_SDR1));
861 
862 	cxl_p2n_write(afu, CXL_CSRP_An, 0);        /* disable */
863 	cxl_p2n_write(afu, CXL_AURP0_An, 0);       /* disable */
864 	cxl_p2n_write(afu, CXL_AURP1_An, 0);       /* disable */
865 
866 	afu->current_mode = CXL_MODE_DEDICATED;
867 	afu->num_procs = 1;
868 
869 	return cxl_chardev_d_afu_add(afu);
870 }
871 
872 void cxl_update_dedicated_ivtes_psl9(struct cxl_context *ctx)
873 {
874 	int r;
875 
876 	for (r = 0; r < CXL_IRQ_RANGES; r++) {
877 		ctx->elem->ivte_offsets[r] = cpu_to_be16(ctx->irqs.offset[r]);
878 		ctx->elem->ivte_ranges[r] = cpu_to_be16(ctx->irqs.range[r]);
879 	}
880 }
881 
882 void cxl_update_dedicated_ivtes_psl8(struct cxl_context *ctx)
883 {
884 	struct cxl_afu *afu = ctx->afu;
885 
886 	cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An,
887 		       (((u64)ctx->irqs.offset[0] & 0xffff) << 48) |
888 		       (((u64)ctx->irqs.offset[1] & 0xffff) << 32) |
889 		       (((u64)ctx->irqs.offset[2] & 0xffff) << 16) |
890 			((u64)ctx->irqs.offset[3] & 0xffff));
891 	cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, (u64)
892 		       (((u64)ctx->irqs.range[0] & 0xffff) << 48) |
893 		       (((u64)ctx->irqs.range[1] & 0xffff) << 32) |
894 		       (((u64)ctx->irqs.range[2] & 0xffff) << 16) |
895 			((u64)ctx->irqs.range[3] & 0xffff));
896 }
897 
898 int cxl_attach_dedicated_process_psl9(struct cxl_context *ctx, u64 wed, u64 amr)
899 {
900 	struct cxl_afu *afu = ctx->afu;
901 	int result;
902 
903 	/* fill the process element entry */
904 	result = process_element_entry_psl9(ctx, wed, amr);
905 	if (result)
906 		return result;
907 
908 	if (ctx->afu->adapter->native->sl_ops->update_dedicated_ivtes)
909 		afu->adapter->native->sl_ops->update_dedicated_ivtes(ctx);
910 
911 	ctx->elem->software_state = cpu_to_be32(CXL_PE_SOFTWARE_STATE_V);
912 	/*
913 	 * Ideally we should do a wmb() here to make sure the changes to the
914 	 * PE are visible to the card before we call afu_enable.
915 	 * On ppc64 though all mmios are preceded by a 'sync' instruction hence
916 	 * we dont dont need one here.
917 	 */
918 
919 	result = cxl_ops->afu_reset(afu);
920 	if (result)
921 		return result;
922 
923 	return afu_enable(afu);
924 }
925 
926 int cxl_attach_dedicated_process_psl8(struct cxl_context *ctx, u64 wed, u64 amr)
927 {
928 	struct cxl_afu *afu = ctx->afu;
929 	u64 pid;
930 	int rc;
931 
932 	pid = (u64)current->pid << 32;
933 	if (ctx->kernel)
934 		pid = 0;
935 	cxl_p2n_write(afu, CXL_PSL_PID_TID_An, pid);
936 
937 	cxl_p1n_write(afu, CXL_PSL_SR_An, calculate_sr(ctx));
938 
939 	if ((rc = cxl_write_sstp(afu, ctx->sstp0, ctx->sstp1)))
940 		return rc;
941 
942 	cxl_prefault(ctx, wed);
943 
944 	if (ctx->afu->adapter->native->sl_ops->update_dedicated_ivtes)
945 		afu->adapter->native->sl_ops->update_dedicated_ivtes(ctx);
946 
947 	cxl_p2n_write(afu, CXL_PSL_AMR_An, amr);
948 
949 	/* master only context for dedicated */
950 	cxl_assign_psn_space(ctx);
951 
952 	if ((rc = cxl_ops->afu_reset(afu)))
953 		return rc;
954 
955 	cxl_p2n_write(afu, CXL_PSL_WED_An, wed);
956 
957 	return afu_enable(afu);
958 }
959 
960 static int deactivate_dedicated_process(struct cxl_afu *afu)
961 {
962 	dev_info(&afu->dev, "Deactivating dedicated process mode\n");
963 
964 	afu->current_mode = 0;
965 	afu->num_procs = 0;
966 
967 	cxl_chardev_afu_remove(afu);
968 
969 	return 0;
970 }
971 
972 static int native_afu_deactivate_mode(struct cxl_afu *afu, int mode)
973 {
974 	if (mode == CXL_MODE_DIRECTED)
975 		return deactivate_afu_directed(afu);
976 	if (mode == CXL_MODE_DEDICATED)
977 		return deactivate_dedicated_process(afu);
978 	return 0;
979 }
980 
981 static int native_afu_activate_mode(struct cxl_afu *afu, int mode)
982 {
983 	if (!mode)
984 		return 0;
985 	if (!(mode & afu->modes_supported))
986 		return -EINVAL;
987 
988 	if (!cxl_ops->link_ok(afu->adapter, afu)) {
989 		WARN(1, "Device link is down, refusing to activate!\n");
990 		return -EIO;
991 	}
992 
993 	if (mode == CXL_MODE_DIRECTED)
994 		return activate_afu_directed(afu);
995 	if ((mode == CXL_MODE_DEDICATED) &&
996 	    (afu->adapter->native->sl_ops->activate_dedicated_process))
997 		return afu->adapter->native->sl_ops->activate_dedicated_process(afu);
998 
999 	return -EINVAL;
1000 }
1001 
1002 static int native_attach_process(struct cxl_context *ctx, bool kernel,
1003 				u64 wed, u64 amr)
1004 {
1005 	if (!cxl_ops->link_ok(ctx->afu->adapter, ctx->afu)) {
1006 		WARN(1, "Device link is down, refusing to attach process!\n");
1007 		return -EIO;
1008 	}
1009 
1010 	ctx->kernel = kernel;
1011 	if ((ctx->afu->current_mode == CXL_MODE_DIRECTED) &&
1012 	    (ctx->afu->adapter->native->sl_ops->attach_afu_directed))
1013 		return ctx->afu->adapter->native->sl_ops->attach_afu_directed(ctx, wed, amr);
1014 
1015 	if ((ctx->afu->current_mode == CXL_MODE_DEDICATED) &&
1016 	    (ctx->afu->adapter->native->sl_ops->attach_dedicated_process))
1017 		return ctx->afu->adapter->native->sl_ops->attach_dedicated_process(ctx, wed, amr);
1018 
1019 	return -EINVAL;
1020 }
1021 
1022 static inline int detach_process_native_dedicated(struct cxl_context *ctx)
1023 {
1024 	/*
1025 	 * The CAIA section 2.1.1 indicates that we need to do an AFU reset to
1026 	 * stop the AFU in dedicated mode (we therefore do not make that
1027 	 * optional like we do in the afu directed path). It does not indicate
1028 	 * that we need to do an explicit disable (which should occur
1029 	 * implicitly as part of the reset) or purge, but we do these as well
1030 	 * to be on the safe side.
1031 	 *
1032 	 * Notably we used to have some issues with the disable sequence
1033 	 * (before the sequence was spelled out in the architecture) which is
1034 	 * why we were so heavy weight in the first place, however a bug was
1035 	 * discovered that had rendered the disable operation ineffective, so
1036 	 * it is conceivable that was the sole explanation for those
1037 	 * difficulties. Point is, we should be careful and do some regression
1038 	 * testing if we ever attempt to remove any part of this procedure.
1039 	 */
1040 	cxl_ops->afu_reset(ctx->afu);
1041 	cxl_afu_disable(ctx->afu);
1042 	cxl_psl_purge(ctx->afu);
1043 	return 0;
1044 }
1045 
1046 static void native_update_ivtes(struct cxl_context *ctx)
1047 {
1048 	if (ctx->afu->current_mode == CXL_MODE_DIRECTED)
1049 		return update_ivtes_directed(ctx);
1050 	if ((ctx->afu->current_mode == CXL_MODE_DEDICATED) &&
1051 	    (ctx->afu->adapter->native->sl_ops->update_dedicated_ivtes))
1052 		return ctx->afu->adapter->native->sl_ops->update_dedicated_ivtes(ctx);
1053 	WARN(1, "native_update_ivtes: Bad mode\n");
1054 }
1055 
1056 static inline int detach_process_native_afu_directed(struct cxl_context *ctx)
1057 {
1058 	if (!ctx->pe_inserted)
1059 		return 0;
1060 	if (terminate_process_element(ctx))
1061 		return -1;
1062 	if (remove_process_element(ctx))
1063 		return -1;
1064 
1065 	return 0;
1066 }
1067 
1068 static int native_detach_process(struct cxl_context *ctx)
1069 {
1070 	trace_cxl_detach(ctx);
1071 
1072 	if (ctx->afu->current_mode == CXL_MODE_DEDICATED)
1073 		return detach_process_native_dedicated(ctx);
1074 
1075 	return detach_process_native_afu_directed(ctx);
1076 }
1077 
1078 static int native_get_irq_info(struct cxl_afu *afu, struct cxl_irq_info *info)
1079 {
1080 	/* If the adapter has gone away, we can't get any meaningful
1081 	 * information.
1082 	 */
1083 	if (!cxl_ops->link_ok(afu->adapter, afu))
1084 		return -EIO;
1085 
1086 	info->dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1087 	info->dar = cxl_p2n_read(afu, CXL_PSL_DAR_An);
1088 	if (cxl_is_power8())
1089 		info->dsr = cxl_p2n_read(afu, CXL_PSL_DSR_An);
1090 	info->afu_err = cxl_p2n_read(afu, CXL_AFU_ERR_An);
1091 	info->errstat = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1092 	info->proc_handle = 0;
1093 
1094 	return 0;
1095 }
1096 
1097 void cxl_native_irq_dump_regs_psl9(struct cxl_context *ctx)
1098 {
1099 	u64 fir1, serr;
1100 
1101 	fir1 = cxl_p1_read(ctx->afu->adapter, CXL_PSL9_FIR1);
1102 
1103 	dev_crit(&ctx->afu->dev, "PSL_FIR1: 0x%016llx\n", fir1);
1104 	if (ctx->afu->adapter->native->sl_ops->register_serr_irq) {
1105 		serr = cxl_p1n_read(ctx->afu, CXL_PSL_SERR_An);
1106 		cxl_afu_decode_psl_serr(ctx->afu, serr);
1107 	}
1108 }
1109 
1110 void cxl_native_irq_dump_regs_psl8(struct cxl_context *ctx)
1111 {
1112 	u64 fir1, fir2, fir_slice, serr, afu_debug;
1113 
1114 	fir1 = cxl_p1_read(ctx->afu->adapter, CXL_PSL_FIR1);
1115 	fir2 = cxl_p1_read(ctx->afu->adapter, CXL_PSL_FIR2);
1116 	fir_slice = cxl_p1n_read(ctx->afu, CXL_PSL_FIR_SLICE_An);
1117 	afu_debug = cxl_p1n_read(ctx->afu, CXL_AFU_DEBUG_An);
1118 
1119 	dev_crit(&ctx->afu->dev, "PSL_FIR1: 0x%016llx\n", fir1);
1120 	dev_crit(&ctx->afu->dev, "PSL_FIR2: 0x%016llx\n", fir2);
1121 	if (ctx->afu->adapter->native->sl_ops->register_serr_irq) {
1122 		serr = cxl_p1n_read(ctx->afu, CXL_PSL_SERR_An);
1123 		cxl_afu_decode_psl_serr(ctx->afu, serr);
1124 	}
1125 	dev_crit(&ctx->afu->dev, "PSL_FIR_SLICE_An: 0x%016llx\n", fir_slice);
1126 	dev_crit(&ctx->afu->dev, "CXL_PSL_AFU_DEBUG_An: 0x%016llx\n", afu_debug);
1127 }
1128 
1129 static irqreturn_t native_handle_psl_slice_error(struct cxl_context *ctx,
1130 						u64 dsisr, u64 errstat)
1131 {
1132 
1133 	dev_crit(&ctx->afu->dev, "PSL ERROR STATUS: 0x%016llx\n", errstat);
1134 
1135 	if (ctx->afu->adapter->native->sl_ops->psl_irq_dump_registers)
1136 		ctx->afu->adapter->native->sl_ops->psl_irq_dump_registers(ctx);
1137 
1138 	if (ctx->afu->adapter->native->sl_ops->debugfs_stop_trace) {
1139 		dev_crit(&ctx->afu->dev, "STOPPING CXL TRACE\n");
1140 		ctx->afu->adapter->native->sl_ops->debugfs_stop_trace(ctx->afu->adapter);
1141 	}
1142 
1143 	return cxl_ops->ack_irq(ctx, 0, errstat);
1144 }
1145 
1146 static bool cxl_is_translation_fault(struct cxl_afu *afu, u64 dsisr)
1147 {
1148 	if ((cxl_is_power8()) && (dsisr & CXL_PSL_DSISR_TRANS))
1149 		return true;
1150 
1151 	if ((cxl_is_power9()) && (dsisr & CXL_PSL9_DSISR_An_TF))
1152 		return true;
1153 
1154 	return false;
1155 }
1156 
1157 irqreturn_t cxl_fail_irq_psl(struct cxl_afu *afu, struct cxl_irq_info *irq_info)
1158 {
1159 	if (cxl_is_translation_fault(afu, irq_info->dsisr))
1160 		cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1161 	else
1162 		cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1163 
1164 	return IRQ_HANDLED;
1165 }
1166 
1167 static irqreturn_t native_irq_multiplexed(int irq, void *data)
1168 {
1169 	struct cxl_afu *afu = data;
1170 	struct cxl_context *ctx;
1171 	struct cxl_irq_info irq_info;
1172 	u64 phreg = cxl_p2n_read(afu, CXL_PSL_PEHandle_An);
1173 	int ph, ret = IRQ_HANDLED, res;
1174 
1175 	/* check if eeh kicked in while the interrupt was in flight */
1176 	if (unlikely(phreg == ~0ULL)) {
1177 		dev_warn(&afu->dev,
1178 			 "Ignoring slice interrupt(%d) due to fenced card",
1179 			 irq);
1180 		return IRQ_HANDLED;
1181 	}
1182 	/* Mask the pe-handle from register value */
1183 	ph = phreg & 0xffff;
1184 	if ((res = native_get_irq_info(afu, &irq_info))) {
1185 		WARN(1, "Unable to get CXL IRQ Info: %i\n", res);
1186 		if (afu->adapter->native->sl_ops->fail_irq)
1187 			return afu->adapter->native->sl_ops->fail_irq(afu, &irq_info);
1188 		return ret;
1189 	}
1190 
1191 	rcu_read_lock();
1192 	ctx = idr_find(&afu->contexts_idr, ph);
1193 	if (ctx) {
1194 		if (afu->adapter->native->sl_ops->handle_interrupt)
1195 			ret = afu->adapter->native->sl_ops->handle_interrupt(irq, ctx, &irq_info);
1196 		rcu_read_unlock();
1197 		return ret;
1198 	}
1199 	rcu_read_unlock();
1200 
1201 	WARN(1, "Unable to demultiplex CXL PSL IRQ for PE %i DSISR %016llx DAR"
1202 		" %016llx\n(Possible AFU HW issue - was a term/remove acked"
1203 		" with outstanding transactions?)\n", ph, irq_info.dsisr,
1204 		irq_info.dar);
1205 	if (afu->adapter->native->sl_ops->fail_irq)
1206 		ret = afu->adapter->native->sl_ops->fail_irq(afu, &irq_info);
1207 	return ret;
1208 }
1209 
1210 static void native_irq_wait(struct cxl_context *ctx)
1211 {
1212 	u64 dsisr;
1213 	int timeout = 1000;
1214 	int ph;
1215 
1216 	/*
1217 	 * Wait until no further interrupts are presented by the PSL
1218 	 * for this context.
1219 	 */
1220 	while (timeout--) {
1221 		ph = cxl_p2n_read(ctx->afu, CXL_PSL_PEHandle_An) & 0xffff;
1222 		if (ph != ctx->pe)
1223 			return;
1224 		dsisr = cxl_p2n_read(ctx->afu, CXL_PSL_DSISR_An);
1225 		if (cxl_is_power8() &&
1226 		   ((dsisr & CXL_PSL_DSISR_PENDING) == 0))
1227 			return;
1228 		if (cxl_is_power9() &&
1229 		   ((dsisr & CXL_PSL9_DSISR_PENDING) == 0))
1230 			return;
1231 		/*
1232 		 * We are waiting for the workqueue to process our
1233 		 * irq, so need to let that run here.
1234 		 */
1235 		msleep(1);
1236 	}
1237 
1238 	dev_warn(&ctx->afu->dev, "WARNING: waiting on DSI for PE %i"
1239 		 " DSISR %016llx!\n", ph, dsisr);
1240 	return;
1241 }
1242 
1243 static irqreturn_t native_slice_irq_err(int irq, void *data)
1244 {
1245 	struct cxl_afu *afu = data;
1246 	u64 errstat, serr, afu_error, dsisr;
1247 	u64 fir_slice, afu_debug, irq_mask;
1248 
1249 	/*
1250 	 * slice err interrupt is only used with full PSL (no XSL)
1251 	 */
1252 	serr = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1253 	errstat = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1254 	afu_error = cxl_p2n_read(afu, CXL_AFU_ERR_An);
1255 	dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1256 	cxl_afu_decode_psl_serr(afu, serr);
1257 
1258 	if (cxl_is_power8()) {
1259 		fir_slice = cxl_p1n_read(afu, CXL_PSL_FIR_SLICE_An);
1260 		afu_debug = cxl_p1n_read(afu, CXL_AFU_DEBUG_An);
1261 		dev_crit(&afu->dev, "PSL_FIR_SLICE_An: 0x%016llx\n", fir_slice);
1262 		dev_crit(&afu->dev, "CXL_PSL_AFU_DEBUG_An: 0x%016llx\n", afu_debug);
1263 	}
1264 	dev_crit(&afu->dev, "CXL_PSL_ErrStat_An: 0x%016llx\n", errstat);
1265 	dev_crit(&afu->dev, "AFU_ERR_An: 0x%.16llx\n", afu_error);
1266 	dev_crit(&afu->dev, "PSL_DSISR_An: 0x%.16llx\n", dsisr);
1267 
1268 	/* mask off the IRQ so it won't retrigger until the AFU is reset */
1269 	irq_mask = (serr & CXL_PSL_SERR_An_IRQS) >> 32;
1270 	serr |= irq_mask;
1271 	cxl_p1n_write(afu, CXL_PSL_SERR_An, serr);
1272 	dev_info(&afu->dev, "Further such interrupts will be masked until the AFU is reset\n");
1273 
1274 	return IRQ_HANDLED;
1275 }
1276 
1277 void cxl_native_err_irq_dump_regs_psl9(struct cxl *adapter)
1278 {
1279 	u64 fir1;
1280 
1281 	fir1 = cxl_p1_read(adapter, CXL_PSL9_FIR1);
1282 	dev_crit(&adapter->dev, "PSL_FIR: 0x%016llx\n", fir1);
1283 }
1284 
1285 void cxl_native_err_irq_dump_regs_psl8(struct cxl *adapter)
1286 {
1287 	u64 fir1, fir2;
1288 
1289 	fir1 = cxl_p1_read(adapter, CXL_PSL_FIR1);
1290 	fir2 = cxl_p1_read(adapter, CXL_PSL_FIR2);
1291 	dev_crit(&adapter->dev,
1292 		 "PSL_FIR1: 0x%016llx\nPSL_FIR2: 0x%016llx\n",
1293 		 fir1, fir2);
1294 }
1295 
1296 static irqreturn_t native_irq_err(int irq, void *data)
1297 {
1298 	struct cxl *adapter = data;
1299 	u64 err_ivte;
1300 
1301 	WARN(1, "CXL ERROR interrupt %i\n", irq);
1302 
1303 	err_ivte = cxl_p1_read(adapter, CXL_PSL_ErrIVTE);
1304 	dev_crit(&adapter->dev, "PSL_ErrIVTE: 0x%016llx\n", err_ivte);
1305 
1306 	if (adapter->native->sl_ops->debugfs_stop_trace) {
1307 		dev_crit(&adapter->dev, "STOPPING CXL TRACE\n");
1308 		adapter->native->sl_ops->debugfs_stop_trace(adapter);
1309 	}
1310 
1311 	if (adapter->native->sl_ops->err_irq_dump_registers)
1312 		adapter->native->sl_ops->err_irq_dump_registers(adapter);
1313 
1314 	return IRQ_HANDLED;
1315 }
1316 
1317 int cxl_native_register_psl_err_irq(struct cxl *adapter)
1318 {
1319 	int rc;
1320 
1321 	adapter->irq_name = kasprintf(GFP_KERNEL, "cxl-%s-err",
1322 				      dev_name(&adapter->dev));
1323 	if (!adapter->irq_name)
1324 		return -ENOMEM;
1325 
1326 	if ((rc = cxl_register_one_irq(adapter, native_irq_err, adapter,
1327 				       &adapter->native->err_hwirq,
1328 				       &adapter->native->err_virq,
1329 				       adapter->irq_name))) {
1330 		kfree(adapter->irq_name);
1331 		adapter->irq_name = NULL;
1332 		return rc;
1333 	}
1334 
1335 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, adapter->native->err_hwirq & 0xffff);
1336 
1337 	return 0;
1338 }
1339 
1340 void cxl_native_release_psl_err_irq(struct cxl *adapter)
1341 {
1342 	if (adapter->native->err_virq == 0 ||
1343 	    adapter->native->err_virq !=
1344 	    irq_find_mapping(NULL, adapter->native->err_hwirq))
1345 		return;
1346 
1347 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, 0x0000000000000000);
1348 	cxl_unmap_irq(adapter->native->err_virq, adapter);
1349 	cxl_ops->release_one_irq(adapter, adapter->native->err_hwirq);
1350 	kfree(adapter->irq_name);
1351 	adapter->native->err_virq = 0;
1352 }
1353 
1354 int cxl_native_register_serr_irq(struct cxl_afu *afu)
1355 {
1356 	u64 serr;
1357 	int rc;
1358 
1359 	afu->err_irq_name = kasprintf(GFP_KERNEL, "cxl-%s-err",
1360 				      dev_name(&afu->dev));
1361 	if (!afu->err_irq_name)
1362 		return -ENOMEM;
1363 
1364 	if ((rc = cxl_register_one_irq(afu->adapter, native_slice_irq_err, afu,
1365 				       &afu->serr_hwirq,
1366 				       &afu->serr_virq, afu->err_irq_name))) {
1367 		kfree(afu->err_irq_name);
1368 		afu->err_irq_name = NULL;
1369 		return rc;
1370 	}
1371 
1372 	serr = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1373 	if (cxl_is_power8())
1374 		serr = (serr & 0x00ffffffffff0000ULL) | (afu->serr_hwirq & 0xffff);
1375 	if (cxl_is_power9()) {
1376 		/*
1377 		 * By default, all errors are masked. So don't set all masks.
1378 		 * Slice errors will be transfered.
1379 		 */
1380 		serr = (serr & ~0xff0000007fffffffULL) | (afu->serr_hwirq & 0xffff);
1381 	}
1382 	cxl_p1n_write(afu, CXL_PSL_SERR_An, serr);
1383 
1384 	return 0;
1385 }
1386 
1387 void cxl_native_release_serr_irq(struct cxl_afu *afu)
1388 {
1389 	if (afu->serr_virq == 0 ||
1390 	    afu->serr_virq != irq_find_mapping(NULL, afu->serr_hwirq))
1391 		return;
1392 
1393 	cxl_p1n_write(afu, CXL_PSL_SERR_An, 0x0000000000000000);
1394 	cxl_unmap_irq(afu->serr_virq, afu);
1395 	cxl_ops->release_one_irq(afu->adapter, afu->serr_hwirq);
1396 	kfree(afu->err_irq_name);
1397 	afu->serr_virq = 0;
1398 }
1399 
1400 int cxl_native_register_psl_irq(struct cxl_afu *afu)
1401 {
1402 	int rc;
1403 
1404 	afu->psl_irq_name = kasprintf(GFP_KERNEL, "cxl-%s",
1405 				      dev_name(&afu->dev));
1406 	if (!afu->psl_irq_name)
1407 		return -ENOMEM;
1408 
1409 	if ((rc = cxl_register_one_irq(afu->adapter, native_irq_multiplexed,
1410 				    afu, &afu->native->psl_hwirq, &afu->native->psl_virq,
1411 				    afu->psl_irq_name))) {
1412 		kfree(afu->psl_irq_name);
1413 		afu->psl_irq_name = NULL;
1414 	}
1415 	return rc;
1416 }
1417 
1418 void cxl_native_release_psl_irq(struct cxl_afu *afu)
1419 {
1420 	if (afu->native->psl_virq == 0 ||
1421 	    afu->native->psl_virq !=
1422 	    irq_find_mapping(NULL, afu->native->psl_hwirq))
1423 		return;
1424 
1425 	cxl_unmap_irq(afu->native->psl_virq, afu);
1426 	cxl_ops->release_one_irq(afu->adapter, afu->native->psl_hwirq);
1427 	kfree(afu->psl_irq_name);
1428 	afu->native->psl_virq = 0;
1429 }
1430 
1431 static void recover_psl_err(struct cxl_afu *afu, u64 errstat)
1432 {
1433 	u64 dsisr;
1434 
1435 	pr_devel("RECOVERING FROM PSL ERROR... (0x%016llx)\n", errstat);
1436 
1437 	/* Clear PSL_DSISR[PE] */
1438 	dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1439 	cxl_p2n_write(afu, CXL_PSL_DSISR_An, dsisr & ~CXL_PSL_DSISR_An_PE);
1440 
1441 	/* Write 1s to clear error status bits */
1442 	cxl_p2n_write(afu, CXL_PSL_ErrStat_An, errstat);
1443 }
1444 
1445 static int native_ack_irq(struct cxl_context *ctx, u64 tfc, u64 psl_reset_mask)
1446 {
1447 	trace_cxl_psl_irq_ack(ctx, tfc);
1448 	if (tfc)
1449 		cxl_p2n_write(ctx->afu, CXL_PSL_TFC_An, tfc);
1450 	if (psl_reset_mask)
1451 		recover_psl_err(ctx->afu, psl_reset_mask);
1452 
1453 	return 0;
1454 }
1455 
1456 int cxl_check_error(struct cxl_afu *afu)
1457 {
1458 	return (cxl_p1n_read(afu, CXL_PSL_SCNTL_An) == ~0ULL);
1459 }
1460 
1461 static bool native_support_attributes(const char *attr_name,
1462 				      enum cxl_attrs type)
1463 {
1464 	return true;
1465 }
1466 
1467 static int native_afu_cr_read64(struct cxl_afu *afu, int cr, u64 off, u64 *out)
1468 {
1469 	if (unlikely(!cxl_ops->link_ok(afu->adapter, afu)))
1470 		return -EIO;
1471 	if (unlikely(off >= afu->crs_len))
1472 		return -ERANGE;
1473 	*out = in_le64(afu->native->afu_desc_mmio + afu->crs_offset +
1474 		(cr * afu->crs_len) + off);
1475 	return 0;
1476 }
1477 
1478 static int native_afu_cr_read32(struct cxl_afu *afu, int cr, u64 off, u32 *out)
1479 {
1480 	if (unlikely(!cxl_ops->link_ok(afu->adapter, afu)))
1481 		return -EIO;
1482 	if (unlikely(off >= afu->crs_len))
1483 		return -ERANGE;
1484 	*out = in_le32(afu->native->afu_desc_mmio + afu->crs_offset +
1485 		(cr * afu->crs_len) + off);
1486 	return 0;
1487 }
1488 
1489 static int native_afu_cr_read16(struct cxl_afu *afu, int cr, u64 off, u16 *out)
1490 {
1491 	u64 aligned_off = off & ~0x3L;
1492 	u32 val;
1493 	int rc;
1494 
1495 	rc = native_afu_cr_read32(afu, cr, aligned_off, &val);
1496 	if (!rc)
1497 		*out = (val >> ((off & 0x3) * 8)) & 0xffff;
1498 	return rc;
1499 }
1500 
1501 static int native_afu_cr_read8(struct cxl_afu *afu, int cr, u64 off, u8 *out)
1502 {
1503 	u64 aligned_off = off & ~0x3L;
1504 	u32 val;
1505 	int rc;
1506 
1507 	rc = native_afu_cr_read32(afu, cr, aligned_off, &val);
1508 	if (!rc)
1509 		*out = (val >> ((off & 0x3) * 8)) & 0xff;
1510 	return rc;
1511 }
1512 
1513 static int native_afu_cr_write32(struct cxl_afu *afu, int cr, u64 off, u32 in)
1514 {
1515 	if (unlikely(!cxl_ops->link_ok(afu->adapter, afu)))
1516 		return -EIO;
1517 	if (unlikely(off >= afu->crs_len))
1518 		return -ERANGE;
1519 	out_le32(afu->native->afu_desc_mmio + afu->crs_offset +
1520 		(cr * afu->crs_len) + off, in);
1521 	return 0;
1522 }
1523 
1524 static int native_afu_cr_write16(struct cxl_afu *afu, int cr, u64 off, u16 in)
1525 {
1526 	u64 aligned_off = off & ~0x3L;
1527 	u32 val32, mask, shift;
1528 	int rc;
1529 
1530 	rc = native_afu_cr_read32(afu, cr, aligned_off, &val32);
1531 	if (rc)
1532 		return rc;
1533 	shift = (off & 0x3) * 8;
1534 	WARN_ON(shift == 24);
1535 	mask = 0xffff << shift;
1536 	val32 = (val32 & ~mask) | (in << shift);
1537 
1538 	rc = native_afu_cr_write32(afu, cr, aligned_off, val32);
1539 	return rc;
1540 }
1541 
1542 static int native_afu_cr_write8(struct cxl_afu *afu, int cr, u64 off, u8 in)
1543 {
1544 	u64 aligned_off = off & ~0x3L;
1545 	u32 val32, mask, shift;
1546 	int rc;
1547 
1548 	rc = native_afu_cr_read32(afu, cr, aligned_off, &val32);
1549 	if (rc)
1550 		return rc;
1551 	shift = (off & 0x3) * 8;
1552 	mask = 0xff << shift;
1553 	val32 = (val32 & ~mask) | (in << shift);
1554 
1555 	rc = native_afu_cr_write32(afu, cr, aligned_off, val32);
1556 	return rc;
1557 }
1558 
1559 const struct cxl_backend_ops cxl_native_ops = {
1560 	.module = THIS_MODULE,
1561 	.adapter_reset = cxl_pci_reset,
1562 	.alloc_one_irq = cxl_pci_alloc_one_irq,
1563 	.release_one_irq = cxl_pci_release_one_irq,
1564 	.alloc_irq_ranges = cxl_pci_alloc_irq_ranges,
1565 	.release_irq_ranges = cxl_pci_release_irq_ranges,
1566 	.setup_irq = cxl_pci_setup_irq,
1567 	.handle_psl_slice_error = native_handle_psl_slice_error,
1568 	.psl_interrupt = NULL,
1569 	.ack_irq = native_ack_irq,
1570 	.irq_wait = native_irq_wait,
1571 	.attach_process = native_attach_process,
1572 	.detach_process = native_detach_process,
1573 	.update_ivtes = native_update_ivtes,
1574 	.support_attributes = native_support_attributes,
1575 	.link_ok = cxl_adapter_link_ok,
1576 	.release_afu = cxl_pci_release_afu,
1577 	.afu_read_err_buffer = cxl_pci_afu_read_err_buffer,
1578 	.afu_check_and_enable = native_afu_check_and_enable,
1579 	.afu_activate_mode = native_afu_activate_mode,
1580 	.afu_deactivate_mode = native_afu_deactivate_mode,
1581 	.afu_reset = native_afu_reset,
1582 	.afu_cr_read8 = native_afu_cr_read8,
1583 	.afu_cr_read16 = native_afu_cr_read16,
1584 	.afu_cr_read32 = native_afu_cr_read32,
1585 	.afu_cr_read64 = native_afu_cr_read64,
1586 	.afu_cr_write8 = native_afu_cr_write8,
1587 	.afu_cr_write16 = native_afu_cr_write16,
1588 	.afu_cr_write32 = native_afu_cr_write32,
1589 	.read_adapter_vpd = cxl_pci_read_adapter_vpd,
1590 };
1591