2 * Copyright 2014 IBM Corp.
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.
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>
16 #include <linux/uaccess.h>
17 #include <linux/delay.h>
18 #include <asm/synch.h>
19 #include <misc/cxl-base.h>
24 static int afu_control(struct cxl_afu *afu, u64 command, u64 clear,
25 u64 result, u64 mask, bool enabled)
28 unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
31 spin_lock(&afu->afu_cntl_lock);
32 pr_devel("AFU command starting: %llx\n", command);
34 trace_cxl_afu_ctrl(afu, command);
36 AFU_Cntl = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
37 cxl_p2n_write(afu, CXL_AFU_Cntl_An, (AFU_Cntl & ~clear) | command);
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");
47 if (!cxl_ops->link_ok(afu->adapter, afu)) {
48 afu->enabled = enabled;
53 pr_devel_ratelimited("AFU control... (0x%016llx)\n",
56 AFU_Cntl = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
59 if (AFU_Cntl & CXL_AFU_Cntl_An_RA) {
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.
65 cxl_p2n_write(afu, CXL_AFU_Cntl_An, AFU_Cntl & ~CXL_AFU_Cntl_An_RA);
68 pr_devel("AFU command complete: %llx\n", command);
69 afu->enabled = enabled;
71 trace_cxl_afu_ctrl_done(afu, command, rc);
72 spin_unlock(&afu->afu_cntl_lock);
77 static int afu_enable(struct cxl_afu *afu)
79 pr_devel("AFU enable request\n");
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);
86 int cxl_afu_disable(struct cxl_afu *afu)
88 pr_devel("AFU disable request\n");
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);
95 /* This will disable as well as reset */
96 static int native_afu_reset(struct cxl_afu *afu)
101 pr_devel("AFU reset request\n");
103 rc = afu_control(afu, CXL_AFU_Cntl_An_RA, 0,
104 CXL_AFU_Cntl_An_RS_Complete | CXL_AFU_Cntl_An_ES_Disabled,
105 CXL_AFU_Cntl_An_RS_MASK | CXL_AFU_Cntl_An_ES_MASK,
109 * Re-enable any masked interrupts when the AFU is not
110 * activated to avoid side effects after attaching a process
113 if (afu->current_mode == 0) {
114 serr = cxl_p1n_read(afu, CXL_PSL_SERR_An);
115 serr &= ~CXL_PSL_SERR_An_IRQ_MASKS;
116 cxl_p1n_write(afu, CXL_PSL_SERR_An, serr);
122 static int native_afu_check_and_enable(struct cxl_afu *afu)
124 if (!cxl_ops->link_ok(afu->adapter, afu)) {
125 WARN(1, "Refusing to enable afu while link down!\n");
130 return afu_enable(afu);
133 int cxl_psl_purge(struct cxl_afu *afu)
135 u64 PSL_CNTL = cxl_p1n_read(afu, CXL_PSL_SCNTL_An);
136 u64 AFU_Cntl = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
139 u64 trans_fault = 0x0ULL;
140 unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
143 trace_cxl_psl_ctrl(afu, CXL_PSL_SCNTL_An_Pc);
145 pr_devel("PSL purge request\n");
148 trans_fault = CXL_PSL_DSISR_TRANS;
150 trans_fault = CXL_PSL9_DSISR_An_TF;
152 if (!cxl_ops->link_ok(afu->adapter, afu)) {
153 dev_warn(&afu->dev, "PSL Purge called with link down, ignoring\n");
158 if ((AFU_Cntl & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
159 WARN(1, "psl_purge request while AFU not disabled!\n");
160 cxl_afu_disable(afu);
163 cxl_p1n_write(afu, CXL_PSL_SCNTL_An,
164 PSL_CNTL | CXL_PSL_SCNTL_An_Pc);
165 start = local_clock();
166 PSL_CNTL = cxl_p1n_read(afu, CXL_PSL_SCNTL_An);
167 while ((PSL_CNTL & CXL_PSL_SCNTL_An_Ps_MASK)
168 == CXL_PSL_SCNTL_An_Ps_Pending) {
169 if (time_after_eq(jiffies, timeout)) {
170 dev_warn(&afu->dev, "WARNING: PSL Purge timed out!\n");
174 if (!cxl_ops->link_ok(afu->adapter, afu)) {
179 dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
180 pr_devel_ratelimited("PSL purging... PSL_CNTL: 0x%016llx PSL_DSISR: 0x%016llx\n",
183 if (dsisr & trans_fault) {
184 dar = cxl_p2n_read(afu, CXL_PSL_DAR_An);
185 dev_notice(&afu->dev, "PSL purge terminating pending translation, DSISR: 0x%016llx, DAR: 0x%016llx\n",
187 cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
189 dev_notice(&afu->dev, "PSL purge acknowledging pending non-translation fault, DSISR: 0x%016llx\n",
191 cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
195 PSL_CNTL = cxl_p1n_read(afu, CXL_PSL_SCNTL_An);
198 pr_devel("PSL purged in %lld ns\n", end - start);
200 cxl_p1n_write(afu, CXL_PSL_SCNTL_An,
201 PSL_CNTL & ~CXL_PSL_SCNTL_An_Pc);
203 trace_cxl_psl_ctrl_done(afu, CXL_PSL_SCNTL_An_Pc, rc);
207 static int spa_max_procs(int spa_size)
211 * end_of_SPA_area = SPA_Base + ((n+4) * 128) + (( ((n*8) + 127) >> 7) * 128) + 255
212 * Most of that junk is really just an overly-complicated way of saying
213 * the last 256 bytes are __aligned(128), so it's really:
214 * end_of_SPA_area = end_of_PSL_queue_area + __aligned(128) 255
216 * end_of_PSL_queue_area = SPA_Base + ((n+4) * 128) + (n*8) - 1
218 * sizeof(SPA) = ((n+4) * 128) + (n*8) + __aligned(128) 256
219 * Ignore the alignment (which is safe in this case as long as we are
220 * careful with our rounding) and solve for n:
222 return ((spa_size / 8) - 96) / 17;
225 static int cxl_alloc_spa(struct cxl_afu *afu, int mode)
229 /* Work out how many pages to allocate */
230 afu->native->spa_order = -1;
232 afu->native->spa_order++;
233 spa_size = (1 << afu->native->spa_order) * PAGE_SIZE;
235 if (spa_size > 0x100000) {
236 dev_warn(&afu->dev, "num_of_processes too large for the SPA, limiting to %i (0x%x)\n",
237 afu->native->spa_max_procs, afu->native->spa_size);
238 if (mode != CXL_MODE_DEDICATED)
239 afu->num_procs = afu->native->spa_max_procs;
243 afu->native->spa_size = spa_size;
244 afu->native->spa_max_procs = spa_max_procs(afu->native->spa_size);
245 } while (afu->native->spa_max_procs < afu->num_procs);
247 if (!(afu->native->spa = (struct cxl_process_element *)
248 __get_free_pages(GFP_KERNEL | __GFP_ZERO, afu->native->spa_order))) {
249 pr_err("cxl_alloc_spa: Unable to allocate scheduled process area\n");
252 pr_devel("spa pages: %i afu->spa_max_procs: %i afu->num_procs: %i\n",
253 1<<afu->native->spa_order, afu->native->spa_max_procs, afu->num_procs);
258 static void attach_spa(struct cxl_afu *afu)
262 afu->native->sw_command_status = (__be64 *)((char *)afu->native->spa +
263 ((afu->native->spa_max_procs + 3) * 128));
265 spap = virt_to_phys(afu->native->spa) & CXL_PSL_SPAP_Addr;
266 spap |= ((afu->native->spa_size >> (12 - CXL_PSL_SPAP_Size_Shift)) - 1) & CXL_PSL_SPAP_Size;
267 spap |= CXL_PSL_SPAP_V;
268 pr_devel("cxl: SPA allocated at 0x%p. Max processes: %i, sw_command_status: 0x%p CXL_PSL_SPAP_An=0x%016llx\n",
269 afu->native->spa, afu->native->spa_max_procs,
270 afu->native->sw_command_status, spap);
271 cxl_p1n_write(afu, CXL_PSL_SPAP_An, spap);
274 static inline void detach_spa(struct cxl_afu *afu)
276 cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0);
279 void cxl_release_spa(struct cxl_afu *afu)
281 if (afu->native->spa) {
282 free_pages((unsigned long) afu->native->spa,
283 afu->native->spa_order);
284 afu->native->spa = NULL;
289 * Invalidation of all ERAT entries is no longer required by CAIA2. Use
292 int cxl_invalidate_all_psl9(struct cxl *adapter)
294 unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
297 pr_devel("CXL adapter - invalidation of all ERAT entries\n");
299 /* Invalidates all ERAT entries for Radix or HPT */
300 ierat = CXL_XSL9_IERAT_IALL;
302 ierat |= CXL_XSL9_IERAT_INVR;
303 cxl_p1_write(adapter, CXL_XSL9_IERAT, ierat);
305 while (cxl_p1_read(adapter, CXL_XSL9_IERAT) & CXL_XSL9_IERAT_IINPROG) {
306 if (time_after_eq(jiffies, timeout)) {
307 dev_warn(&adapter->dev,
308 "WARNING: CXL adapter invalidation of all ERAT entries timed out!\n");
311 if (!cxl_ops->link_ok(adapter, NULL))
318 int cxl_invalidate_all_psl8(struct cxl *adapter)
320 unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
322 pr_devel("CXL adapter wide TLBIA & SLBIA\n");
324 cxl_p1_write(adapter, CXL_PSL_AFUSEL, CXL_PSL_AFUSEL_A);
326 cxl_p1_write(adapter, CXL_PSL_TLBIA, CXL_TLB_SLB_IQ_ALL);
327 while (cxl_p1_read(adapter, CXL_PSL_TLBIA) & CXL_TLB_SLB_P) {
328 if (time_after_eq(jiffies, timeout)) {
329 dev_warn(&adapter->dev, "WARNING: CXL adapter wide TLBIA timed out!\n");
332 if (!cxl_ops->link_ok(adapter, NULL))
337 cxl_p1_write(adapter, CXL_PSL_SLBIA, CXL_TLB_SLB_IQ_ALL);
338 while (cxl_p1_read(adapter, CXL_PSL_SLBIA) & CXL_TLB_SLB_P) {
339 if (time_after_eq(jiffies, timeout)) {
340 dev_warn(&adapter->dev, "WARNING: CXL adapter wide SLBIA timed out!\n");
343 if (!cxl_ops->link_ok(adapter, NULL))
350 int cxl_data_cache_flush(struct cxl *adapter)
353 unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
356 * Do a datacache flush only if datacache is available.
357 * In case of PSL9D datacache absent hence flush operation.
360 if (adapter->native->no_data_cache) {
361 pr_devel("No PSL data cache. Ignoring cache flush req.\n");
365 pr_devel("Flushing data cache\n");
366 reg = cxl_p1_read(adapter, CXL_PSL_Control);
367 reg |= CXL_PSL_Control_Fr;
368 cxl_p1_write(adapter, CXL_PSL_Control, reg);
370 reg = cxl_p1_read(adapter, CXL_PSL_Control);
371 while ((reg & CXL_PSL_Control_Fs_MASK) != CXL_PSL_Control_Fs_Complete) {
372 if (time_after_eq(jiffies, timeout)) {
373 dev_warn(&adapter->dev, "WARNING: cache flush timed out!\n");
377 if (!cxl_ops->link_ok(adapter, NULL)) {
378 dev_warn(&adapter->dev, "WARNING: link down when flushing cache\n");
382 reg = cxl_p1_read(adapter, CXL_PSL_Control);
385 reg &= ~CXL_PSL_Control_Fr;
386 cxl_p1_write(adapter, CXL_PSL_Control, reg);
390 static int cxl_write_sstp(struct cxl_afu *afu, u64 sstp0, u64 sstp1)
394 /* 1. Disable SSTP by writing 0 to SSTP1[V] */
395 cxl_p2n_write(afu, CXL_SSTP1_An, 0);
397 /* 2. Invalidate all SLB entries */
398 if ((rc = cxl_afu_slbia(afu)))
401 /* 3. Set SSTP0_An */
402 cxl_p2n_write(afu, CXL_SSTP0_An, sstp0);
404 /* 4. Set SSTP1_An */
405 cxl_p2n_write(afu, CXL_SSTP1_An, sstp1);
410 /* Using per slice version may improve performance here. (ie. SLBIA_An) */
411 static void slb_invalid(struct cxl_context *ctx)
413 struct cxl *adapter = ctx->afu->adapter;
416 WARN_ON(!mutex_is_locked(&ctx->afu->native->spa_mutex));
418 cxl_p1_write(adapter, CXL_PSL_LBISEL,
419 ((u64)be32_to_cpu(ctx->elem->common.pid) << 32) |
420 be32_to_cpu(ctx->elem->lpid));
421 cxl_p1_write(adapter, CXL_PSL_SLBIA, CXL_TLB_SLB_IQ_LPIDPID);
424 if (!cxl_ops->link_ok(adapter, NULL))
426 slbia = cxl_p1_read(adapter, CXL_PSL_SLBIA);
427 if (!(slbia & CXL_TLB_SLB_P))
433 static int do_process_element_cmd(struct cxl_context *ctx,
434 u64 cmd, u64 pe_state)
437 unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
440 trace_cxl_llcmd(ctx, cmd);
442 WARN_ON(!ctx->afu->enabled);
444 ctx->elem->software_state = cpu_to_be32(pe_state);
446 *(ctx->afu->native->sw_command_status) = cpu_to_be64(cmd | 0 | ctx->pe);
448 cxl_p1n_write(ctx->afu, CXL_PSL_LLCMD_An, cmd | ctx->pe);
450 if (time_after_eq(jiffies, timeout)) {
451 dev_warn(&ctx->afu->dev, "WARNING: Process Element Command timed out!\n");
455 if (!cxl_ops->link_ok(ctx->afu->adapter, ctx->afu)) {
456 dev_warn(&ctx->afu->dev, "WARNING: Device link down, aborting Process Element Command!\n");
460 state = be64_to_cpup(ctx->afu->native->sw_command_status);
461 if (state == ~0ULL) {
462 pr_err("cxl: Error adding process element to AFU\n");
466 if ((state & (CXL_SPA_SW_CMD_MASK | CXL_SPA_SW_STATE_MASK | CXL_SPA_SW_LINK_MASK)) ==
467 (cmd | (cmd >> 16) | ctx->pe))
470 * The command won't finish in the PSL if there are
471 * outstanding DSIs. Hence we need to yield here in
472 * case there are outstanding DSIs that we need to
473 * service. Tuning possiblity: we could wait for a
480 trace_cxl_llcmd_done(ctx, cmd, rc);
484 static int add_process_element(struct cxl_context *ctx)
488 mutex_lock(&ctx->afu->native->spa_mutex);
489 pr_devel("%s Adding pe: %i started\n", __func__, ctx->pe);
490 if (!(rc = do_process_element_cmd(ctx, CXL_SPA_SW_CMD_ADD, CXL_PE_SOFTWARE_STATE_V)))
491 ctx->pe_inserted = true;
492 pr_devel("%s Adding pe: %i finished\n", __func__, ctx->pe);
493 mutex_unlock(&ctx->afu->native->spa_mutex);
497 static int terminate_process_element(struct cxl_context *ctx)
501 /* fast path terminate if it's already invalid */
502 if (!(ctx->elem->software_state & cpu_to_be32(CXL_PE_SOFTWARE_STATE_V)))
505 mutex_lock(&ctx->afu->native->spa_mutex);
506 pr_devel("%s Terminate pe: %i started\n", __func__, ctx->pe);
507 /* We could be asked to terminate when the hw is down. That
508 * should always succeed: it's not running if the hw has gone
509 * away and is being reset.
511 if (cxl_ops->link_ok(ctx->afu->adapter, ctx->afu))
512 rc = do_process_element_cmd(ctx, CXL_SPA_SW_CMD_TERMINATE,
513 CXL_PE_SOFTWARE_STATE_V | CXL_PE_SOFTWARE_STATE_T);
514 ctx->elem->software_state = 0; /* Remove Valid bit */
515 pr_devel("%s Terminate pe: %i finished\n", __func__, ctx->pe);
516 mutex_unlock(&ctx->afu->native->spa_mutex);
520 static int remove_process_element(struct cxl_context *ctx)
524 mutex_lock(&ctx->afu->native->spa_mutex);
525 pr_devel("%s Remove pe: %i started\n", __func__, ctx->pe);
527 /* We could be asked to remove when the hw is down. Again, if
528 * the hw is down, the PE is gone, so we succeed.
530 if (cxl_ops->link_ok(ctx->afu->adapter, ctx->afu))
531 rc = do_process_element_cmd(ctx, CXL_SPA_SW_CMD_REMOVE, 0);
534 ctx->pe_inserted = false;
537 pr_devel("%s Remove pe: %i finished\n", __func__, ctx->pe);
538 mutex_unlock(&ctx->afu->native->spa_mutex);
543 void cxl_assign_psn_space(struct cxl_context *ctx)
545 if (!ctx->afu->pp_size || ctx->master) {
546 ctx->psn_phys = ctx->afu->psn_phys;
547 ctx->psn_size = ctx->afu->adapter->ps_size;
549 ctx->psn_phys = ctx->afu->psn_phys +
550 (ctx->afu->native->pp_offset + ctx->afu->pp_size * ctx->pe);
551 ctx->psn_size = ctx->afu->pp_size;
555 static int activate_afu_directed(struct cxl_afu *afu)
559 dev_info(&afu->dev, "Activating AFU directed mode\n");
561 afu->num_procs = afu->max_procs_virtualised;
562 if (afu->native->spa == NULL) {
563 if (cxl_alloc_spa(afu, CXL_MODE_DIRECTED))
568 cxl_p1n_write(afu, CXL_PSL_SCNTL_An, CXL_PSL_SCNTL_An_PM_AFU);
570 cxl_p1n_write(afu, CXL_PSL_AMOR_An, 0xFFFFFFFFFFFFFFFFULL);
571 cxl_p1n_write(afu, CXL_PSL_ID_An, CXL_PSL_ID_An_F | CXL_PSL_ID_An_L);
573 afu->current_mode = CXL_MODE_DIRECTED;
575 if ((rc = cxl_chardev_m_afu_add(afu)))
578 if ((rc = cxl_sysfs_afu_m_add(afu)))
581 if ((rc = cxl_chardev_s_afu_add(afu)))
586 cxl_sysfs_afu_m_remove(afu);
588 cxl_chardev_afu_remove(afu);
592 #ifdef CONFIG_CPU_LITTLE_ENDIAN
593 #define set_endian(sr) ((sr) |= CXL_PSL_SR_An_LE)
595 #define set_endian(sr) ((sr) &= ~(CXL_PSL_SR_An_LE))
598 u64 cxl_calculate_sr(bool master, bool kernel, bool real_mode, bool p9)
604 sr |= CXL_PSL_SR_An_MP;
605 if (mfspr(SPRN_LPCR) & LPCR_TC)
606 sr |= CXL_PSL_SR_An_TC;
609 sr |= CXL_PSL_SR_An_R;
610 sr |= (mfmsr() & MSR_SF) | CXL_PSL_SR_An_HV;
612 sr |= CXL_PSL_SR_An_PR | CXL_PSL_SR_An_R;
614 sr |= CXL_PSL_SR_An_HV;
616 sr &= ~(CXL_PSL_SR_An_HV);
617 if (!test_tsk_thread_flag(current, TIF_32BIT))
618 sr |= CXL_PSL_SR_An_SF;
622 sr |= CXL_PSL_SR_An_XLAT_ror;
624 sr |= CXL_PSL_SR_An_XLAT_hpt;
629 static u64 calculate_sr(struct cxl_context *ctx)
631 return cxl_calculate_sr(ctx->master, ctx->kernel, ctx->real_mode,
635 static void update_ivtes_directed(struct cxl_context *ctx)
637 bool need_update = (ctx->status == STARTED);
641 WARN_ON(terminate_process_element(ctx));
642 WARN_ON(remove_process_element(ctx));
645 for (r = 0; r < CXL_IRQ_RANGES; r++) {
646 ctx->elem->ivte_offsets[r] = cpu_to_be16(ctx->irqs.offset[r]);
647 ctx->elem->ivte_ranges[r] = cpu_to_be16(ctx->irqs.range[r]);
651 * Theoretically we could use the update llcmd, instead of a
652 * terminate/remove/add (or if an atomic update was required we could
653 * do a suspend/update/resume), however it seems there might be issues
654 * with the update llcmd on some cards (including those using an XSL on
655 * an ASIC) so for now it's safest to go with the commands that are
656 * known to work. In the future if we come across a situation where the
657 * card may be performing transactions using the same PE while we are
658 * doing this update we might need to revisit this.
661 WARN_ON(add_process_element(ctx));
664 static int process_element_entry_psl9(struct cxl_context *ctx, u64 wed, u64 amr)
668 cxl_assign_psn_space(ctx);
670 ctx->elem->ctxtime = 0; /* disable */
671 ctx->elem->lpid = cpu_to_be32(mfspr(SPRN_LPID));
672 ctx->elem->haurp = 0; /* disable */
677 if (ctx->mm == NULL) {
678 pr_devel("%s: unable to get mm for pe=%d pid=%i\n",
679 __func__, ctx->pe, pid_nr(ctx->pid));
682 pid = ctx->mm->context.id;
685 ctx->elem->common.tid = 0;
686 ctx->elem->common.pid = cpu_to_be32(pid);
688 ctx->elem->sr = cpu_to_be64(calculate_sr(ctx));
690 ctx->elem->common.csrp = 0; /* disable */
692 cxl_prefault(ctx, wed);
695 * Ensure we have the multiplexed PSL interrupt set up to take faults
696 * for kernel contexts that may not have allocated any AFU IRQs at all:
698 if (ctx->irqs.range[0] == 0) {
699 ctx->irqs.offset[0] = ctx->afu->native->psl_hwirq;
700 ctx->irqs.range[0] = 1;
703 ctx->elem->common.amr = cpu_to_be64(amr);
704 ctx->elem->common.wed = cpu_to_be64(wed);
709 int cxl_attach_afu_directed_psl9(struct cxl_context *ctx, u64 wed, u64 amr)
713 /* fill the process element entry */
714 result = process_element_entry_psl9(ctx, wed, amr);
718 update_ivtes_directed(ctx);
720 /* first guy needs to enable */
721 result = cxl_ops->afu_check_and_enable(ctx->afu);
725 return add_process_element(ctx);
728 int cxl_attach_afu_directed_psl8(struct cxl_context *ctx, u64 wed, u64 amr)
733 cxl_assign_psn_space(ctx);
735 ctx->elem->ctxtime = 0; /* disable */
736 ctx->elem->lpid = cpu_to_be32(mfspr(SPRN_LPID));
737 ctx->elem->haurp = 0; /* disable */
738 ctx->elem->u.sdr = cpu_to_be64(mfspr(SPRN_SDR1));
743 ctx->elem->common.tid = 0;
744 ctx->elem->common.pid = cpu_to_be32(pid);
746 ctx->elem->sr = cpu_to_be64(calculate_sr(ctx));
748 ctx->elem->common.csrp = 0; /* disable */
749 ctx->elem->common.u.psl8.aurp0 = 0; /* disable */
750 ctx->elem->common.u.psl8.aurp1 = 0; /* disable */
752 cxl_prefault(ctx, wed);
754 ctx->elem->common.u.psl8.sstp0 = cpu_to_be64(ctx->sstp0);
755 ctx->elem->common.u.psl8.sstp1 = cpu_to_be64(ctx->sstp1);
758 * Ensure we have the multiplexed PSL interrupt set up to take faults
759 * for kernel contexts that may not have allocated any AFU IRQs at all:
761 if (ctx->irqs.range[0] == 0) {
762 ctx->irqs.offset[0] = ctx->afu->native->psl_hwirq;
763 ctx->irqs.range[0] = 1;
766 update_ivtes_directed(ctx);
768 ctx->elem->common.amr = cpu_to_be64(amr);
769 ctx->elem->common.wed = cpu_to_be64(wed);
771 /* first guy needs to enable */
772 if ((result = cxl_ops->afu_check_and_enable(ctx->afu)))
775 return add_process_element(ctx);
778 static int deactivate_afu_directed(struct cxl_afu *afu)
780 dev_info(&afu->dev, "Deactivating AFU directed mode\n");
782 afu->current_mode = 0;
785 cxl_sysfs_afu_m_remove(afu);
786 cxl_chardev_afu_remove(afu);
789 * The CAIA section 2.2.1 indicates that the procedure for starting and
790 * stopping an AFU in AFU directed mode is AFU specific, which is not
791 * ideal since this code is generic and with one exception has no
792 * knowledge of the AFU. This is in contrast to the procedure for
793 * disabling a dedicated process AFU, which is documented to just
794 * require a reset. The architecture does indicate that both an AFU
795 * reset and an AFU disable should result in the AFU being disabled and
796 * we do both followed by a PSL purge for safety.
798 * Notably we used to have some issues with the disable sequence on PSL
799 * cards, which is why we ended up using this heavy weight procedure in
800 * the first place, however a bug was discovered that had rendered the
801 * disable operation ineffective, so it is conceivable that was the
802 * sole explanation for those difficulties. Careful regression testing
803 * is recommended if anyone attempts to remove or reorder these
806 * The XSL on the Mellanox CX4 behaves a little differently from the
807 * PSL based cards and will time out an AFU reset if the AFU is still
808 * enabled. That card is special in that we do have a means to identify
809 * it from this code, so in that case we skip the reset and just use a
810 * disable/purge to avoid the timeout and corresponding noise in the
813 if (afu->adapter->native->sl_ops->needs_reset_before_disable)
814 cxl_ops->afu_reset(afu);
815 cxl_afu_disable(afu);
821 int cxl_activate_dedicated_process_psl9(struct cxl_afu *afu)
823 dev_info(&afu->dev, "Activating dedicated process mode\n");
826 * If XSL is set to dedicated mode (Set in PSL_SCNTL reg), the
827 * XSL and AFU are programmed to work with a single context.
828 * The context information should be configured in the SPA area
829 * index 0 (so PSL_SPAP must be configured before enabling the
833 if (afu->native->spa == NULL) {
834 if (cxl_alloc_spa(afu, CXL_MODE_DEDICATED))
839 cxl_p1n_write(afu, CXL_PSL_SCNTL_An, CXL_PSL_SCNTL_An_PM_Process);
840 cxl_p1n_write(afu, CXL_PSL_ID_An, CXL_PSL_ID_An_F | CXL_PSL_ID_An_L);
842 afu->current_mode = CXL_MODE_DEDICATED;
844 return cxl_chardev_d_afu_add(afu);
847 int cxl_activate_dedicated_process_psl8(struct cxl_afu *afu)
849 dev_info(&afu->dev, "Activating dedicated process mode\n");
851 cxl_p1n_write(afu, CXL_PSL_SCNTL_An, CXL_PSL_SCNTL_An_PM_Process);
853 cxl_p1n_write(afu, CXL_PSL_CtxTime_An, 0); /* disable */
854 cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0); /* disable */
855 cxl_p1n_write(afu, CXL_PSL_AMOR_An, 0xFFFFFFFFFFFFFFFFULL);
856 cxl_p1n_write(afu, CXL_PSL_LPID_An, mfspr(SPRN_LPID));
857 cxl_p1n_write(afu, CXL_HAURP_An, 0); /* disable */
858 cxl_p1n_write(afu, CXL_PSL_SDR_An, mfspr(SPRN_SDR1));
860 cxl_p2n_write(afu, CXL_CSRP_An, 0); /* disable */
861 cxl_p2n_write(afu, CXL_AURP0_An, 0); /* disable */
862 cxl_p2n_write(afu, CXL_AURP1_An, 0); /* disable */
864 afu->current_mode = CXL_MODE_DEDICATED;
867 return cxl_chardev_d_afu_add(afu);
870 void cxl_update_dedicated_ivtes_psl9(struct cxl_context *ctx)
874 for (r = 0; r < CXL_IRQ_RANGES; r++) {
875 ctx->elem->ivte_offsets[r] = cpu_to_be16(ctx->irqs.offset[r]);
876 ctx->elem->ivte_ranges[r] = cpu_to_be16(ctx->irqs.range[r]);
880 void cxl_update_dedicated_ivtes_psl8(struct cxl_context *ctx)
882 struct cxl_afu *afu = ctx->afu;
884 cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An,
885 (((u64)ctx->irqs.offset[0] & 0xffff) << 48) |
886 (((u64)ctx->irqs.offset[1] & 0xffff) << 32) |
887 (((u64)ctx->irqs.offset[2] & 0xffff) << 16) |
888 ((u64)ctx->irqs.offset[3] & 0xffff));
889 cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, (u64)
890 (((u64)ctx->irqs.range[0] & 0xffff) << 48) |
891 (((u64)ctx->irqs.range[1] & 0xffff) << 32) |
892 (((u64)ctx->irqs.range[2] & 0xffff) << 16) |
893 ((u64)ctx->irqs.range[3] & 0xffff));
896 int cxl_attach_dedicated_process_psl9(struct cxl_context *ctx, u64 wed, u64 amr)
898 struct cxl_afu *afu = ctx->afu;
901 /* fill the process element entry */
902 result = process_element_entry_psl9(ctx, wed, amr);
906 if (ctx->afu->adapter->native->sl_ops->update_dedicated_ivtes)
907 afu->adapter->native->sl_ops->update_dedicated_ivtes(ctx);
909 result = cxl_ops->afu_reset(afu);
913 return afu_enable(afu);
916 int cxl_attach_dedicated_process_psl8(struct cxl_context *ctx, u64 wed, u64 amr)
918 struct cxl_afu *afu = ctx->afu;
922 pid = (u64)current->pid << 32;
925 cxl_p2n_write(afu, CXL_PSL_PID_TID_An, pid);
927 cxl_p1n_write(afu, CXL_PSL_SR_An, calculate_sr(ctx));
929 if ((rc = cxl_write_sstp(afu, ctx->sstp0, ctx->sstp1)))
932 cxl_prefault(ctx, wed);
934 if (ctx->afu->adapter->native->sl_ops->update_dedicated_ivtes)
935 afu->adapter->native->sl_ops->update_dedicated_ivtes(ctx);
937 cxl_p2n_write(afu, CXL_PSL_AMR_An, amr);
939 /* master only context for dedicated */
940 cxl_assign_psn_space(ctx);
942 if ((rc = cxl_ops->afu_reset(afu)))
945 cxl_p2n_write(afu, CXL_PSL_WED_An, wed);
947 return afu_enable(afu);
950 static int deactivate_dedicated_process(struct cxl_afu *afu)
952 dev_info(&afu->dev, "Deactivating dedicated process mode\n");
954 afu->current_mode = 0;
957 cxl_chardev_afu_remove(afu);
962 static int native_afu_deactivate_mode(struct cxl_afu *afu, int mode)
964 if (mode == CXL_MODE_DIRECTED)
965 return deactivate_afu_directed(afu);
966 if (mode == CXL_MODE_DEDICATED)
967 return deactivate_dedicated_process(afu);
971 static int native_afu_activate_mode(struct cxl_afu *afu, int mode)
975 if (!(mode & afu->modes_supported))
978 if (!cxl_ops->link_ok(afu->adapter, afu)) {
979 WARN(1, "Device link is down, refusing to activate!\n");
983 if (mode == CXL_MODE_DIRECTED)
984 return activate_afu_directed(afu);
985 if ((mode == CXL_MODE_DEDICATED) &&
986 (afu->adapter->native->sl_ops->activate_dedicated_process))
987 return afu->adapter->native->sl_ops->activate_dedicated_process(afu);
992 static int native_attach_process(struct cxl_context *ctx, bool kernel,
995 if (!cxl_ops->link_ok(ctx->afu->adapter, ctx->afu)) {
996 WARN(1, "Device link is down, refusing to attach process!\n");
1000 ctx->kernel = kernel;
1001 if ((ctx->afu->current_mode == CXL_MODE_DIRECTED) &&
1002 (ctx->afu->adapter->native->sl_ops->attach_afu_directed))
1003 return ctx->afu->adapter->native->sl_ops->attach_afu_directed(ctx, wed, amr);
1005 if ((ctx->afu->current_mode == CXL_MODE_DEDICATED) &&
1006 (ctx->afu->adapter->native->sl_ops->attach_dedicated_process))
1007 return ctx->afu->adapter->native->sl_ops->attach_dedicated_process(ctx, wed, amr);
1012 static inline int detach_process_native_dedicated(struct cxl_context *ctx)
1015 * The CAIA section 2.1.1 indicates that we need to do an AFU reset to
1016 * stop the AFU in dedicated mode (we therefore do not make that
1017 * optional like we do in the afu directed path). It does not indicate
1018 * that we need to do an explicit disable (which should occur
1019 * implicitly as part of the reset) or purge, but we do these as well
1020 * to be on the safe side.
1022 * Notably we used to have some issues with the disable sequence
1023 * (before the sequence was spelled out in the architecture) which is
1024 * why we were so heavy weight in the first place, however a bug was
1025 * discovered that had rendered the disable operation ineffective, so
1026 * it is conceivable that was the sole explanation for those
1027 * difficulties. Point is, we should be careful and do some regression
1028 * testing if we ever attempt to remove any part of this procedure.
1030 cxl_ops->afu_reset(ctx->afu);
1031 cxl_afu_disable(ctx->afu);
1032 cxl_psl_purge(ctx->afu);
1036 static void native_update_ivtes(struct cxl_context *ctx)
1038 if (ctx->afu->current_mode == CXL_MODE_DIRECTED)
1039 return update_ivtes_directed(ctx);
1040 if ((ctx->afu->current_mode == CXL_MODE_DEDICATED) &&
1041 (ctx->afu->adapter->native->sl_ops->update_dedicated_ivtes))
1042 return ctx->afu->adapter->native->sl_ops->update_dedicated_ivtes(ctx);
1043 WARN(1, "native_update_ivtes: Bad mode\n");
1046 static inline int detach_process_native_afu_directed(struct cxl_context *ctx)
1048 if (!ctx->pe_inserted)
1050 if (terminate_process_element(ctx))
1052 if (remove_process_element(ctx))
1058 static int native_detach_process(struct cxl_context *ctx)
1060 trace_cxl_detach(ctx);
1062 if (ctx->afu->current_mode == CXL_MODE_DEDICATED)
1063 return detach_process_native_dedicated(ctx);
1065 return detach_process_native_afu_directed(ctx);
1068 static int native_get_irq_info(struct cxl_afu *afu, struct cxl_irq_info *info)
1070 /* If the adapter has gone away, we can't get any meaningful
1073 if (!cxl_ops->link_ok(afu->adapter, afu))
1076 info->dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1077 info->dar = cxl_p2n_read(afu, CXL_PSL_DAR_An);
1078 if (cxl_is_power8())
1079 info->dsr = cxl_p2n_read(afu, CXL_PSL_DSR_An);
1080 info->afu_err = cxl_p2n_read(afu, CXL_AFU_ERR_An);
1081 info->errstat = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1082 info->proc_handle = 0;
1087 void cxl_native_irq_dump_regs_psl9(struct cxl_context *ctx)
1089 u64 fir1, fir2, serr;
1091 fir1 = cxl_p1_read(ctx->afu->adapter, CXL_PSL9_FIR1);
1092 fir2 = cxl_p1_read(ctx->afu->adapter, CXL_PSL9_FIR2);
1094 dev_crit(&ctx->afu->dev, "PSL_FIR1: 0x%016llx\n", fir1);
1095 dev_crit(&ctx->afu->dev, "PSL_FIR2: 0x%016llx\n", fir2);
1096 if (ctx->afu->adapter->native->sl_ops->register_serr_irq) {
1097 serr = cxl_p1n_read(ctx->afu, CXL_PSL_SERR_An);
1098 cxl_afu_decode_psl_serr(ctx->afu, serr);
1102 void cxl_native_irq_dump_regs_psl8(struct cxl_context *ctx)
1104 u64 fir1, fir2, fir_slice, serr, afu_debug;
1106 fir1 = cxl_p1_read(ctx->afu->adapter, CXL_PSL_FIR1);
1107 fir2 = cxl_p1_read(ctx->afu->adapter, CXL_PSL_FIR2);
1108 fir_slice = cxl_p1n_read(ctx->afu, CXL_PSL_FIR_SLICE_An);
1109 afu_debug = cxl_p1n_read(ctx->afu, CXL_AFU_DEBUG_An);
1111 dev_crit(&ctx->afu->dev, "PSL_FIR1: 0x%016llx\n", fir1);
1112 dev_crit(&ctx->afu->dev, "PSL_FIR2: 0x%016llx\n", fir2);
1113 if (ctx->afu->adapter->native->sl_ops->register_serr_irq) {
1114 serr = cxl_p1n_read(ctx->afu, CXL_PSL_SERR_An);
1115 cxl_afu_decode_psl_serr(ctx->afu, serr);
1117 dev_crit(&ctx->afu->dev, "PSL_FIR_SLICE_An: 0x%016llx\n", fir_slice);
1118 dev_crit(&ctx->afu->dev, "CXL_PSL_AFU_DEBUG_An: 0x%016llx\n", afu_debug);
1121 static irqreturn_t native_handle_psl_slice_error(struct cxl_context *ctx,
1122 u64 dsisr, u64 errstat)
1125 dev_crit(&ctx->afu->dev, "PSL ERROR STATUS: 0x%016llx\n", errstat);
1127 if (ctx->afu->adapter->native->sl_ops->psl_irq_dump_registers)
1128 ctx->afu->adapter->native->sl_ops->psl_irq_dump_registers(ctx);
1130 if (ctx->afu->adapter->native->sl_ops->debugfs_stop_trace) {
1131 dev_crit(&ctx->afu->dev, "STOPPING CXL TRACE\n");
1132 ctx->afu->adapter->native->sl_ops->debugfs_stop_trace(ctx->afu->adapter);
1135 return cxl_ops->ack_irq(ctx, 0, errstat);
1138 static bool cxl_is_translation_fault(struct cxl_afu *afu, u64 dsisr)
1140 if ((cxl_is_power8()) && (dsisr & CXL_PSL_DSISR_TRANS))
1143 if ((cxl_is_power9()) && (dsisr & CXL_PSL9_DSISR_An_TF))
1149 irqreturn_t cxl_fail_irq_psl(struct cxl_afu *afu, struct cxl_irq_info *irq_info)
1151 if (cxl_is_translation_fault(afu, irq_info->dsisr))
1152 cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1154 cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1159 static irqreturn_t native_irq_multiplexed(int irq, void *data)
1161 struct cxl_afu *afu = data;
1162 struct cxl_context *ctx;
1163 struct cxl_irq_info irq_info;
1164 u64 phreg = cxl_p2n_read(afu, CXL_PSL_PEHandle_An);
1165 int ph, ret = IRQ_HANDLED, res;
1167 /* check if eeh kicked in while the interrupt was in flight */
1168 if (unlikely(phreg == ~0ULL)) {
1170 "Ignoring slice interrupt(%d) due to fenced card",
1174 /* Mask the pe-handle from register value */
1175 ph = phreg & 0xffff;
1176 if ((res = native_get_irq_info(afu, &irq_info))) {
1177 WARN(1, "Unable to get CXL IRQ Info: %i\n", res);
1178 if (afu->adapter->native->sl_ops->fail_irq)
1179 return afu->adapter->native->sl_ops->fail_irq(afu, &irq_info);
1184 ctx = idr_find(&afu->contexts_idr, ph);
1186 if (afu->adapter->native->sl_ops->handle_interrupt)
1187 ret = afu->adapter->native->sl_ops->handle_interrupt(irq, ctx, &irq_info);
1193 WARN(1, "Unable to demultiplex CXL PSL IRQ for PE %i DSISR %016llx DAR"
1194 " %016llx\n(Possible AFU HW issue - was a term/remove acked"
1195 " with outstanding transactions?)\n", ph, irq_info.dsisr,
1197 if (afu->adapter->native->sl_ops->fail_irq)
1198 ret = afu->adapter->native->sl_ops->fail_irq(afu, &irq_info);
1202 static void native_irq_wait(struct cxl_context *ctx)
1209 * Wait until no further interrupts are presented by the PSL
1213 ph = cxl_p2n_read(ctx->afu, CXL_PSL_PEHandle_An) & 0xffff;
1216 dsisr = cxl_p2n_read(ctx->afu, CXL_PSL_DSISR_An);
1217 if (cxl_is_power8() &&
1218 ((dsisr & CXL_PSL_DSISR_PENDING) == 0))
1220 if (cxl_is_power9() &&
1221 ((dsisr & CXL_PSL9_DSISR_PENDING) == 0))
1224 * We are waiting for the workqueue to process our
1225 * irq, so need to let that run here.
1230 dev_warn(&ctx->afu->dev, "WARNING: waiting on DSI for PE %i"
1231 " DSISR %016llx!\n", ph, dsisr);
1235 static irqreturn_t native_slice_irq_err(int irq, void *data)
1237 struct cxl_afu *afu = data;
1238 u64 errstat, serr, afu_error, dsisr;
1239 u64 fir_slice, afu_debug, irq_mask;
1242 * slice err interrupt is only used with full PSL (no XSL)
1244 serr = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1245 errstat = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1246 afu_error = cxl_p2n_read(afu, CXL_AFU_ERR_An);
1247 dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1248 cxl_afu_decode_psl_serr(afu, serr);
1250 if (cxl_is_power8()) {
1251 fir_slice = cxl_p1n_read(afu, CXL_PSL_FIR_SLICE_An);
1252 afu_debug = cxl_p1n_read(afu, CXL_AFU_DEBUG_An);
1253 dev_crit(&afu->dev, "PSL_FIR_SLICE_An: 0x%016llx\n", fir_slice);
1254 dev_crit(&afu->dev, "CXL_PSL_AFU_DEBUG_An: 0x%016llx\n", afu_debug);
1256 dev_crit(&afu->dev, "CXL_PSL_ErrStat_An: 0x%016llx\n", errstat);
1257 dev_crit(&afu->dev, "AFU_ERR_An: 0x%.16llx\n", afu_error);
1258 dev_crit(&afu->dev, "PSL_DSISR_An: 0x%.16llx\n", dsisr);
1260 /* mask off the IRQ so it won't retrigger until the AFU is reset */
1261 irq_mask = (serr & CXL_PSL_SERR_An_IRQS) >> 32;
1263 cxl_p1n_write(afu, CXL_PSL_SERR_An, serr);
1264 dev_info(&afu->dev, "Further such interrupts will be masked until the AFU is reset\n");
1269 void cxl_native_err_irq_dump_regs(struct cxl *adapter)
1273 fir1 = cxl_p1_read(adapter, CXL_PSL_FIR1);
1274 fir2 = cxl_p1_read(adapter, CXL_PSL_FIR2);
1276 dev_crit(&adapter->dev, "PSL_FIR1: 0x%016llx\nPSL_FIR2: 0x%016llx\n", fir1, fir2);
1279 static irqreturn_t native_irq_err(int irq, void *data)
1281 struct cxl *adapter = data;
1284 WARN(1, "CXL ERROR interrupt %i\n", irq);
1286 err_ivte = cxl_p1_read(adapter, CXL_PSL_ErrIVTE);
1287 dev_crit(&adapter->dev, "PSL_ErrIVTE: 0x%016llx\n", err_ivte);
1289 if (adapter->native->sl_ops->debugfs_stop_trace) {
1290 dev_crit(&adapter->dev, "STOPPING CXL TRACE\n");
1291 adapter->native->sl_ops->debugfs_stop_trace(adapter);
1294 if (adapter->native->sl_ops->err_irq_dump_registers)
1295 adapter->native->sl_ops->err_irq_dump_registers(adapter);
1300 int cxl_native_register_psl_err_irq(struct cxl *adapter)
1304 adapter->irq_name = kasprintf(GFP_KERNEL, "cxl-%s-err",
1305 dev_name(&adapter->dev));
1306 if (!adapter->irq_name)
1309 if ((rc = cxl_register_one_irq(adapter, native_irq_err, adapter,
1310 &adapter->native->err_hwirq,
1311 &adapter->native->err_virq,
1312 adapter->irq_name))) {
1313 kfree(adapter->irq_name);
1314 adapter->irq_name = NULL;
1318 cxl_p1_write(adapter, CXL_PSL_ErrIVTE, adapter->native->err_hwirq & 0xffff);
1323 void cxl_native_release_psl_err_irq(struct cxl *adapter)
1325 if (adapter->native->err_virq == 0 ||
1326 adapter->native->err_virq !=
1327 irq_find_mapping(NULL, adapter->native->err_hwirq))
1330 cxl_p1_write(adapter, CXL_PSL_ErrIVTE, 0x0000000000000000);
1331 cxl_unmap_irq(adapter->native->err_virq, adapter);
1332 cxl_ops->release_one_irq(adapter, adapter->native->err_hwirq);
1333 kfree(adapter->irq_name);
1334 adapter->native->err_virq = 0;
1337 int cxl_native_register_serr_irq(struct cxl_afu *afu)
1342 afu->err_irq_name = kasprintf(GFP_KERNEL, "cxl-%s-err",
1343 dev_name(&afu->dev));
1344 if (!afu->err_irq_name)
1347 if ((rc = cxl_register_one_irq(afu->adapter, native_slice_irq_err, afu,
1349 &afu->serr_virq, afu->err_irq_name))) {
1350 kfree(afu->err_irq_name);
1351 afu->err_irq_name = NULL;
1355 serr = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1356 if (cxl_is_power8())
1357 serr = (serr & 0x00ffffffffff0000ULL) | (afu->serr_hwirq & 0xffff);
1358 if (cxl_is_power9()) {
1360 * By default, all errors are masked. So don't set all masks.
1361 * Slice errors will be transfered.
1363 serr = (serr & ~0xff0000007fffffffULL) | (afu->serr_hwirq & 0xffff);
1365 cxl_p1n_write(afu, CXL_PSL_SERR_An, serr);
1370 void cxl_native_release_serr_irq(struct cxl_afu *afu)
1372 if (afu->serr_virq == 0 ||
1373 afu->serr_virq != irq_find_mapping(NULL, afu->serr_hwirq))
1376 cxl_p1n_write(afu, CXL_PSL_SERR_An, 0x0000000000000000);
1377 cxl_unmap_irq(afu->serr_virq, afu);
1378 cxl_ops->release_one_irq(afu->adapter, afu->serr_hwirq);
1379 kfree(afu->err_irq_name);
1383 int cxl_native_register_psl_irq(struct cxl_afu *afu)
1387 afu->psl_irq_name = kasprintf(GFP_KERNEL, "cxl-%s",
1388 dev_name(&afu->dev));
1389 if (!afu->psl_irq_name)
1392 if ((rc = cxl_register_one_irq(afu->adapter, native_irq_multiplexed,
1393 afu, &afu->native->psl_hwirq, &afu->native->psl_virq,
1394 afu->psl_irq_name))) {
1395 kfree(afu->psl_irq_name);
1396 afu->psl_irq_name = NULL;
1401 void cxl_native_release_psl_irq(struct cxl_afu *afu)
1403 if (afu->native->psl_virq == 0 ||
1404 afu->native->psl_virq !=
1405 irq_find_mapping(NULL, afu->native->psl_hwirq))
1408 cxl_unmap_irq(afu->native->psl_virq, afu);
1409 cxl_ops->release_one_irq(afu->adapter, afu->native->psl_hwirq);
1410 kfree(afu->psl_irq_name);
1411 afu->native->psl_virq = 0;
1414 static void recover_psl_err(struct cxl_afu *afu, u64 errstat)
1418 pr_devel("RECOVERING FROM PSL ERROR... (0x%016llx)\n", errstat);
1420 /* Clear PSL_DSISR[PE] */
1421 dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1422 cxl_p2n_write(afu, CXL_PSL_DSISR_An, dsisr & ~CXL_PSL_DSISR_An_PE);
1424 /* Write 1s to clear error status bits */
1425 cxl_p2n_write(afu, CXL_PSL_ErrStat_An, errstat);
1428 static int native_ack_irq(struct cxl_context *ctx, u64 tfc, u64 psl_reset_mask)
1430 trace_cxl_psl_irq_ack(ctx, tfc);
1432 cxl_p2n_write(ctx->afu, CXL_PSL_TFC_An, tfc);
1434 recover_psl_err(ctx->afu, psl_reset_mask);
1439 int cxl_check_error(struct cxl_afu *afu)
1441 return (cxl_p1n_read(afu, CXL_PSL_SCNTL_An) == ~0ULL);
1444 static bool native_support_attributes(const char *attr_name,
1445 enum cxl_attrs type)
1450 static int native_afu_cr_read64(struct cxl_afu *afu, int cr, u64 off, u64 *out)
1452 if (unlikely(!cxl_ops->link_ok(afu->adapter, afu)))
1454 if (unlikely(off >= afu->crs_len))
1456 *out = in_le64(afu->native->afu_desc_mmio + afu->crs_offset +
1457 (cr * afu->crs_len) + off);
1461 static int native_afu_cr_read32(struct cxl_afu *afu, int cr, u64 off, u32 *out)
1463 if (unlikely(!cxl_ops->link_ok(afu->adapter, afu)))
1465 if (unlikely(off >= afu->crs_len))
1467 *out = in_le32(afu->native->afu_desc_mmio + afu->crs_offset +
1468 (cr * afu->crs_len) + off);
1472 static int native_afu_cr_read16(struct cxl_afu *afu, int cr, u64 off, u16 *out)
1474 u64 aligned_off = off & ~0x3L;
1478 rc = native_afu_cr_read32(afu, cr, aligned_off, &val);
1480 *out = (val >> ((off & 0x3) * 8)) & 0xffff;
1484 static int native_afu_cr_read8(struct cxl_afu *afu, int cr, u64 off, u8 *out)
1486 u64 aligned_off = off & ~0x3L;
1490 rc = native_afu_cr_read32(afu, cr, aligned_off, &val);
1492 *out = (val >> ((off & 0x3) * 8)) & 0xff;
1496 static int native_afu_cr_write32(struct cxl_afu *afu, int cr, u64 off, u32 in)
1498 if (unlikely(!cxl_ops->link_ok(afu->adapter, afu)))
1500 if (unlikely(off >= afu->crs_len))
1502 out_le32(afu->native->afu_desc_mmio + afu->crs_offset +
1503 (cr * afu->crs_len) + off, in);
1507 static int native_afu_cr_write16(struct cxl_afu *afu, int cr, u64 off, u16 in)
1509 u64 aligned_off = off & ~0x3L;
1510 u32 val32, mask, shift;
1513 rc = native_afu_cr_read32(afu, cr, aligned_off, &val32);
1516 shift = (off & 0x3) * 8;
1517 WARN_ON(shift == 24);
1518 mask = 0xffff << shift;
1519 val32 = (val32 & ~mask) | (in << shift);
1521 rc = native_afu_cr_write32(afu, cr, aligned_off, val32);
1525 static int native_afu_cr_write8(struct cxl_afu *afu, int cr, u64 off, u8 in)
1527 u64 aligned_off = off & ~0x3L;
1528 u32 val32, mask, shift;
1531 rc = native_afu_cr_read32(afu, cr, aligned_off, &val32);
1534 shift = (off & 0x3) * 8;
1535 mask = 0xff << shift;
1536 val32 = (val32 & ~mask) | (in << shift);
1538 rc = native_afu_cr_write32(afu, cr, aligned_off, val32);
1542 const struct cxl_backend_ops cxl_native_ops = {
1543 .module = THIS_MODULE,
1544 .adapter_reset = cxl_pci_reset,
1545 .alloc_one_irq = cxl_pci_alloc_one_irq,
1546 .release_one_irq = cxl_pci_release_one_irq,
1547 .alloc_irq_ranges = cxl_pci_alloc_irq_ranges,
1548 .release_irq_ranges = cxl_pci_release_irq_ranges,
1549 .setup_irq = cxl_pci_setup_irq,
1550 .handle_psl_slice_error = native_handle_psl_slice_error,
1551 .psl_interrupt = NULL,
1552 .ack_irq = native_ack_irq,
1553 .irq_wait = native_irq_wait,
1554 .attach_process = native_attach_process,
1555 .detach_process = native_detach_process,
1556 .update_ivtes = native_update_ivtes,
1557 .support_attributes = native_support_attributes,
1558 .link_ok = cxl_adapter_link_ok,
1559 .release_afu = cxl_pci_release_afu,
1560 .afu_read_err_buffer = cxl_pci_afu_read_err_buffer,
1561 .afu_check_and_enable = native_afu_check_and_enable,
1562 .afu_activate_mode = native_afu_activate_mode,
1563 .afu_deactivate_mode = native_afu_deactivate_mode,
1564 .afu_reset = native_afu_reset,
1565 .afu_cr_read8 = native_afu_cr_read8,
1566 .afu_cr_read16 = native_afu_cr_read16,
1567 .afu_cr_read32 = native_afu_cr_read32,
1568 .afu_cr_read64 = native_afu_cr_read64,
1569 .afu_cr_write8 = native_afu_cr_write8,
1570 .afu_cr_write16 = native_afu_cr_write16,
1571 .afu_cr_write32 = native_afu_cr_write32,
1572 .read_adapter_vpd = cxl_pci_read_adapter_vpd,