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