xref: /openbmc/qemu/hw/intc/xive2.c (revision 714bae7351d9c85643d6b48109f59f20f05c8466)
1 /*
2  * QEMU PowerPC XIVE2 interrupt controller model (POWER10)
3  *
4  * Copyright (c) 2019-2024, IBM Corporation..
5  *
6  * SPDX-License-Identifier: GPL-2.0-or-later
7  */
8 
9 #include "qemu/osdep.h"
10 #include "qemu/log.h"
11 #include "qemu/module.h"
12 #include "qapi/error.h"
13 #include "target/ppc/cpu.h"
14 #include "system/cpus.h"
15 #include "system/dma.h"
16 #include "hw/qdev-properties.h"
17 #include "hw/ppc/xive.h"
18 #include "hw/ppc/xive2.h"
19 #include "hw/ppc/xive2_regs.h"
20 #include "trace.h"
21 
22 static void xive2_router_end_notify(Xive2Router *xrtr, uint8_t end_blk,
23                                     uint32_t end_idx, uint32_t end_data,
24                                     bool redistribute);
25 
26 static int xive2_tctx_get_nvp_indexes(XiveTCTX *tctx, uint8_t ring,
27                                       uint8_t *nvp_blk, uint32_t *nvp_idx);
28 
29 uint32_t xive2_router_get_config(Xive2Router *xrtr)
30 {
31     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
32 
33     return xrc->get_config(xrtr);
34 }
35 
36 static int xive2_router_get_block_id(Xive2Router *xrtr)
37 {
38    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
39 
40    return xrc->get_block_id(xrtr);
41 }
42 
43 static uint64_t xive2_nvp_reporting_addr(Xive2Nvp *nvp)
44 {
45     uint64_t cache_addr;
46 
47     cache_addr = xive_get_field32(NVP2_W6_REPORTING_LINE, nvp->w6) << 24 |
48         xive_get_field32(NVP2_W7_REPORTING_LINE, nvp->w7);
49     cache_addr <<= 8; /* aligned on a cache line pair */
50     return cache_addr;
51 }
52 
53 static uint32_t xive2_nvgc_get_backlog(Xive2Nvgc *nvgc, uint8_t priority)
54 {
55     uint32_t val = 0;
56     uint8_t *ptr, i;
57 
58     if (priority > 7) {
59         return 0;
60     }
61 
62     /*
63      * The per-priority backlog counters are 24-bit and the structure
64      * is stored in big endian. NVGC is 32-bytes long, so 24-bytes from
65      * w2, which fits 8 priorities * 24-bits per priority.
66      */
67     ptr = (uint8_t *)&nvgc->w2 + priority * 3;
68     for (i = 0; i < 3; i++, ptr++) {
69         val = (val << 8) + *ptr;
70     }
71     return val;
72 }
73 
74 static void xive2_nvgc_set_backlog(Xive2Nvgc *nvgc, uint8_t priority,
75                                    uint32_t val)
76 {
77     uint8_t *ptr, i;
78     uint32_t shift;
79 
80     if (priority > 7) {
81         return;
82     }
83 
84     if (val > 0xFFFFFF) {
85         val = 0xFFFFFF;
86     }
87     /*
88      * The per-priority backlog counters are 24-bit and the structure
89      * is stored in big endian
90      */
91     ptr = (uint8_t *)&nvgc->w2 + priority * 3;
92     for (i = 0; i < 3; i++, ptr++) {
93         shift = 8 * (2 - i);
94         *ptr = (val >> shift) & 0xFF;
95     }
96 }
97 
98 uint64_t xive2_presenter_nvgc_backlog_op(XivePresenter *xptr,
99                                          bool crowd,
100                                          uint8_t blk, uint32_t idx,
101                                          uint16_t offset, uint16_t val)
102 {
103     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
104     uint8_t priority = GETFIELD(NVx_BACKLOG_PRIO, offset);
105     uint8_t op = GETFIELD(NVx_BACKLOG_OP, offset);
106     Xive2Nvgc nvgc;
107     uint32_t count, old_count;
108 
109     if (xive2_router_get_nvgc(xrtr, crowd, blk, idx, &nvgc)) {
110         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No %s %x/%x\n",
111                       crowd ? "NVC" : "NVG", blk, idx);
112         return -1;
113     }
114     if (!xive2_nvgc_is_valid(&nvgc)) {
115         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid NVG %x/%x\n", blk, idx);
116         return -1;
117     }
118 
119     old_count = xive2_nvgc_get_backlog(&nvgc, priority);
120     count = old_count;
121     /*
122      * op:
123      * 0b00 => increment
124      * 0b01 => decrement
125      * 0b1- => read
126      */
127     if (op == 0b00 || op == 0b01) {
128         if (op == 0b00) {
129             count += val;
130         } else {
131             if (count > val) {
132                 count -= val;
133             } else {
134                 count = 0;
135             }
136         }
137         xive2_nvgc_set_backlog(&nvgc, priority, count);
138         xive2_router_write_nvgc(xrtr, crowd, blk, idx, &nvgc);
139     }
140     trace_xive_nvgc_backlog_op(crowd, blk, idx, op, priority, old_count);
141     return old_count;
142 }
143 
144 uint64_t xive2_presenter_nvp_backlog_op(XivePresenter *xptr,
145                                         uint8_t blk, uint32_t idx,
146                                         uint16_t offset)
147 {
148     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
149     uint8_t priority = GETFIELD(NVx_BACKLOG_PRIO, offset);
150     uint8_t op = GETFIELD(NVx_BACKLOG_OP, offset);
151     Xive2Nvp nvp;
152     uint8_t ipb, old_ipb, rc;
153 
154     if (xive2_router_get_nvp(xrtr, blk, idx, &nvp)) {
155         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n", blk, idx);
156         return -1;
157     }
158     if (!xive2_nvp_is_valid(&nvp)) {
159         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid NVP %x/%x\n", blk, idx);
160         return -1;
161     }
162 
163     old_ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2);
164     ipb = old_ipb;
165     /*
166      * op:
167      * 0b00 => set priority bit
168      * 0b01 => reset priority bit
169      * 0b1- => read
170      */
171     if (op == 0b00 || op == 0b01) {
172         if (op == 0b00) {
173             ipb |= xive_priority_to_ipb(priority);
174         } else {
175             ipb &= ~xive_priority_to_ipb(priority);
176         }
177         nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, ipb);
178         xive2_router_write_nvp(xrtr, blk, idx, &nvp, 2);
179     }
180     rc = !!(old_ipb & xive_priority_to_ipb(priority));
181     trace_xive_nvp_backlog_op(blk, idx, op, priority, rc);
182     return rc;
183 }
184 
185 void xive2_eas_pic_print_info(Xive2Eas *eas, uint32_t lisn, GString *buf)
186 {
187     if (!xive2_eas_is_valid(eas)) {
188         return;
189     }
190 
191     g_string_append_printf(buf, "  %08x %s end:%02x/%04x data:%08x\n",
192                            lisn, xive2_eas_is_masked(eas) ? "M" : " ",
193                            (uint8_t)  xive_get_field64(EAS2_END_BLOCK, eas->w),
194                            (uint32_t) xive_get_field64(EAS2_END_INDEX, eas->w),
195                            (uint32_t) xive_get_field64(EAS2_END_DATA, eas->w));
196 }
197 
198 #define XIVE2_QSIZE_CHUNK_CL    128
199 #define XIVE2_QSIZE_CHUNK_4k   4096
200 /* Calculate max number of queue entries for an END */
201 static uint32_t xive2_end_get_qentries(Xive2End *end)
202 {
203     uint32_t w3 = end->w3;
204     uint32_t qsize = xive_get_field32(END2_W3_QSIZE, w3);
205     if (xive_get_field32(END2_W3_CL, w3)) {
206         g_assert(qsize <= 4);
207         return (XIVE2_QSIZE_CHUNK_CL << qsize) / sizeof(uint32_t);
208     } else {
209         g_assert(qsize <= 12);
210         return (XIVE2_QSIZE_CHUNK_4k << qsize) / sizeof(uint32_t);
211     }
212 }
213 
214 void xive2_end_queue_pic_print_info(Xive2End *end, uint32_t width, GString *buf)
215 {
216     uint64_t qaddr_base = xive2_end_qaddr(end);
217     uint32_t qindex = xive_get_field32(END2_W1_PAGE_OFF, end->w1);
218     uint32_t qentries = xive2_end_get_qentries(end);
219     int i;
220 
221     /*
222      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
223      */
224     g_string_append_printf(buf, " [ ");
225     qindex = (qindex - (width - 1)) & (qentries - 1);
226     for (i = 0; i < width; i++) {
227         uint64_t qaddr = qaddr_base + (qindex << 2);
228         uint32_t qdata = -1;
229 
230         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
231                             sizeof(qdata), MEMTXATTRS_UNSPECIFIED)) {
232             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
233                           HWADDR_PRIx "\n", qaddr);
234             return;
235         }
236         g_string_append_printf(buf, "%s%08x ", i == width - 1 ? "^" : "",
237                                be32_to_cpu(qdata));
238         qindex = (qindex + 1) & (qentries - 1);
239     }
240     g_string_append_printf(buf, "]");
241 }
242 
243 void xive2_end_pic_print_info(Xive2End *end, uint32_t end_idx, GString *buf)
244 {
245     uint64_t qaddr_base = xive2_end_qaddr(end);
246     uint32_t qindex = xive_get_field32(END2_W1_PAGE_OFF, end->w1);
247     uint32_t qgen = xive_get_field32(END2_W1_GENERATION, end->w1);
248     uint32_t qentries = xive2_end_get_qentries(end);
249 
250     uint32_t nvx_blk = xive_get_field32(END2_W6_VP_BLOCK, end->w6);
251     uint32_t nvx_idx = xive_get_field32(END2_W6_VP_OFFSET, end->w6);
252     uint8_t priority = xive_get_field32(END2_W7_F0_PRIORITY, end->w7);
253     uint8_t pq;
254 
255     if (!xive2_end_is_valid(end)) {
256         return;
257     }
258 
259     pq = xive_get_field32(END2_W1_ESn, end->w1);
260 
261     g_string_append_printf(buf,
262                            "  %08x %c%c %c%c%c%c%c%c%c%c%c%c%c %c%c "
263                            "prio:%d nvp:%02x/%04x",
264                            end_idx,
265                            pq & XIVE_ESB_VAL_P ? 'P' : '-',
266                            pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
267                            xive2_end_is_valid(end)    ? 'v' : '-',
268                            xive2_end_is_enqueue(end)  ? 'q' : '-',
269                            xive2_end_is_notify(end)   ? 'n' : '-',
270                            xive2_end_is_backlog(end)  ? 'b' : '-',
271                            xive2_end_is_precluded_escalation(end) ? 'p' : '-',
272                            xive2_end_is_escalate(end) ? 'e' : '-',
273                            xive2_end_is_escalate_end(end) ? 'N' : '-',
274                            xive2_end_is_uncond_escalation(end)   ? 'u' : '-',
275                            xive2_end_is_silent_escalation(end)   ? 's' : '-',
276                            xive2_end_is_firmware1(end)   ? 'f' : '-',
277                            xive2_end_is_firmware2(end)   ? 'F' : '-',
278                            xive2_end_is_ignore(end) ? 'i' : '-',
279                            xive2_end_is_crowd(end)  ? 'c' : '-',
280                            priority, nvx_blk, nvx_idx);
281 
282     if (qaddr_base) {
283         g_string_append_printf(buf, " eq:@%08"PRIx64"% 6d/%5d ^%d",
284                                qaddr_base, qindex, qentries, qgen);
285         xive2_end_queue_pic_print_info(end, 6, buf);
286     }
287     g_string_append_c(buf, '\n');
288 }
289 
290 void xive2_end_eas_pic_print_info(Xive2End *end, uint32_t end_idx,
291                                   GString *buf)
292 {
293     Xive2Eas *eas = (Xive2Eas *) &end->w4;
294     uint8_t pq;
295 
296     if (!xive2_end_is_escalate(end)) {
297         return;
298     }
299 
300     pq = xive_get_field32(END2_W1_ESe, end->w1);
301 
302     g_string_append_printf(buf, "  %08x %c%c %c%c end:%02x/%04x data:%08x\n",
303                            end_idx,
304                            pq & XIVE_ESB_VAL_P ? 'P' : '-',
305                            pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
306                            xive2_eas_is_valid(eas) ? 'v' : ' ',
307                            xive2_eas_is_masked(eas) ? 'M' : ' ',
308                            (uint8_t)  xive_get_field64(EAS2_END_BLOCK, eas->w),
309                            (uint32_t) xive_get_field64(EAS2_END_INDEX, eas->w),
310                            (uint32_t) xive_get_field64(EAS2_END_DATA, eas->w));
311 }
312 
313 void xive2_nvp_pic_print_info(Xive2Nvp *nvp, uint32_t nvp_idx, GString *buf)
314 {
315     uint8_t  eq_blk = xive_get_field32(NVP2_W5_VP_END_BLOCK, nvp->w5);
316     uint32_t eq_idx = xive_get_field32(NVP2_W5_VP_END_INDEX, nvp->w5);
317     uint64_t cache_line = xive2_nvp_reporting_addr(nvp);
318 
319     if (!xive2_nvp_is_valid(nvp)) {
320         return;
321     }
322 
323     g_string_append_printf(buf, "  %08x end:%02x/%04x IPB:%02x PGoFirst:%02x",
324                            nvp_idx, eq_blk, eq_idx,
325                            xive_get_field32(NVP2_W2_IPB, nvp->w2),
326                            xive_get_field32(NVP2_W0_PGOFIRST, nvp->w0));
327     if (cache_line) {
328         g_string_append_printf(buf, "  reporting CL:%016"PRIx64, cache_line);
329     }
330 
331     /*
332      * When the NVP is HW controlled, more fields are updated
333      */
334     if (xive2_nvp_is_hw(nvp)) {
335         g_string_append_printf(buf, " CPPR:%02x",
336                                xive_get_field32(NVP2_W2_CPPR, nvp->w2));
337         if (xive2_nvp_is_co(nvp)) {
338             g_string_append_printf(buf, " CO:%04x",
339                                    xive_get_field32(NVP2_W1_CO_THRID, nvp->w1));
340         }
341     }
342     g_string_append_c(buf, '\n');
343 }
344 
345 void xive2_nvgc_pic_print_info(Xive2Nvgc *nvgc, uint32_t nvgc_idx, GString *buf)
346 {
347     uint8_t i;
348 
349     if (!xive2_nvgc_is_valid(nvgc)) {
350         return;
351     }
352 
353     g_string_append_printf(buf, "  %08x PGoNext:%02x bklog: ", nvgc_idx,
354                            xive_get_field32(NVGC2_W0_PGONEXT, nvgc->w0));
355     for (i = 0; i <= XIVE_PRIORITY_MAX; i++) {
356         g_string_append_printf(buf, "[%d]=0x%x ",
357                                i, xive2_nvgc_get_backlog(nvgc, i));
358     }
359     g_string_append_printf(buf, "\n");
360 }
361 
362 static void xive2_end_enqueue(Xive2End *end, uint32_t data)
363 {
364     uint64_t qaddr_base = xive2_end_qaddr(end);
365     uint32_t qindex = xive_get_field32(END2_W1_PAGE_OFF, end->w1);
366     uint32_t qgen = xive_get_field32(END2_W1_GENERATION, end->w1);
367 
368     uint64_t qaddr = qaddr_base + (qindex << 2);
369     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
370     uint32_t qentries = xive2_end_get_qentries(end);
371 
372     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata),
373                          MEMTXATTRS_UNSPECIFIED)) {
374         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
375                       HWADDR_PRIx "\n", qaddr);
376         return;
377     }
378 
379     qindex = (qindex + 1) & (qentries - 1);
380     if (qindex == 0) {
381         qgen ^= 1;
382         end->w1 = xive_set_field32(END2_W1_GENERATION, end->w1, qgen);
383 
384         /* Set gen flipped to 1, it gets reset on a cache watch operation */
385         end->w1 = xive_set_field32(END2_W1_GEN_FLIPPED, end->w1, 1);
386     }
387     end->w1 = xive_set_field32(END2_W1_PAGE_OFF, end->w1, qindex);
388 }
389 
390 static void xive2_pgofnext(uint8_t *nvgc_blk, uint32_t *nvgc_idx,
391                            uint8_t next_level)
392 {
393     uint32_t mask, next_idx;
394     uint8_t next_blk;
395 
396     /*
397      * Adjust the block and index of a VP for the next group/crowd
398      * size (PGofFirst/PGofNext field in the NVP and NVGC structures).
399      *
400      * The 6-bit group level is split into a 2-bit crowd and 4-bit
401      * group levels. Encoding is similar. However, we don't support
402      * crowd size of 8. So a crowd level of 0b11 is bumped to a crowd
403      * size of 16.
404      */
405     next_blk = NVx_CROWD_LVL(next_level);
406     if (next_blk == 3) {
407         next_blk = 4;
408     }
409     mask = (1 << next_blk) - 1;
410     *nvgc_blk &= ~mask;
411     *nvgc_blk |= mask >> 1;
412 
413     next_idx = NVx_GROUP_LVL(next_level);
414     mask = (1 << next_idx) - 1;
415     *nvgc_idx &= ~mask;
416     *nvgc_idx |= mask >> 1;
417 }
418 
419 /*
420  * Scan the group chain and return the highest priority and group
421  * level of pending group interrupts.
422  */
423 static uint8_t xive2_presenter_backlog_scan(XivePresenter *xptr,
424                                             uint8_t nvx_blk, uint32_t nvx_idx,
425                                             uint8_t first_group,
426                                             uint8_t *out_level)
427 {
428     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
429     uint32_t nvgc_idx;
430     uint32_t current_level, count;
431     uint8_t nvgc_blk, prio;
432     Xive2Nvgc nvgc;
433 
434     for (prio = 0; prio <= XIVE_PRIORITY_MAX; prio++) {
435         current_level = first_group & 0x3F;
436         nvgc_blk = nvx_blk;
437         nvgc_idx = nvx_idx;
438 
439         while (current_level) {
440             xive2_pgofnext(&nvgc_blk, &nvgc_idx, current_level);
441 
442             if (xive2_router_get_nvgc(xrtr, NVx_CROWD_LVL(current_level),
443                                       nvgc_blk, nvgc_idx, &nvgc)) {
444                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVGC %x/%x\n",
445                               nvgc_blk, nvgc_idx);
446                 return 0xFF;
447             }
448             if (!xive2_nvgc_is_valid(&nvgc)) {
449                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid NVGC %x/%x\n",
450                               nvgc_blk, nvgc_idx);
451                 return 0xFF;
452             }
453 
454             count = xive2_nvgc_get_backlog(&nvgc, prio);
455             if (count) {
456                 *out_level = current_level;
457                 return prio;
458             }
459             current_level = xive_get_field32(NVGC2_W0_PGONEXT, nvgc.w0) & 0x3F;
460         }
461     }
462     return 0xFF;
463 }
464 
465 static void xive2_presenter_backlog_decr(XivePresenter *xptr,
466                                          uint8_t nvx_blk, uint32_t nvx_idx,
467                                          uint8_t group_prio,
468                                          uint8_t group_level)
469 {
470     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
471     uint32_t nvgc_idx, count;
472     uint8_t nvgc_blk;
473     Xive2Nvgc nvgc;
474 
475     nvgc_blk = nvx_blk;
476     nvgc_idx = nvx_idx;
477     xive2_pgofnext(&nvgc_blk, &nvgc_idx, group_level);
478 
479     if (xive2_router_get_nvgc(xrtr, NVx_CROWD_LVL(group_level),
480                               nvgc_blk, nvgc_idx, &nvgc)) {
481         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVGC %x/%x\n",
482                       nvgc_blk, nvgc_idx);
483         return;
484     }
485     if (!xive2_nvgc_is_valid(&nvgc)) {
486         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid NVGC %x/%x\n",
487                       nvgc_blk, nvgc_idx);
488         return;
489     }
490     count = xive2_nvgc_get_backlog(&nvgc, group_prio);
491     if (!count) {
492         return;
493     }
494     xive2_nvgc_set_backlog(&nvgc, group_prio, count - 1);
495     xive2_router_write_nvgc(xrtr, NVx_CROWD_LVL(group_level),
496                             nvgc_blk, nvgc_idx, &nvgc);
497 }
498 
499 /*
500  * XIVE Thread Interrupt Management Area (TIMA) - Gen2 mode
501  *
502  * TIMA Gen2 VP “save & restore” (S&R) indicated by H bit next to V bit
503  *
504  *   - if a context is enabled with the H bit set, the VP context
505  *     information is retrieved from the NVP structure (“check out”)
506  *     and stored back on a context pull (“check in”), the SW receives
507  *     the same context pull information as on P9
508  *
509  *   - the H bit cannot be changed while the V bit is set, i.e. a
510  *     context cannot be set up in the TIMA and then be “pushed” into
511  *     the NVP by changing the H bit while the context is enabled
512  */
513 
514 static void xive2_tctx_save_ctx(Xive2Router *xrtr, XiveTCTX *tctx,
515                                 uint8_t ring,
516                                 uint8_t nvp_blk, uint32_t nvp_idx)
517 {
518     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
519     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
520     Xive2Nvp nvp;
521     uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
522     uint8_t *regs = &tctx->regs[ring];
523 
524     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
525         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
526                           nvp_blk, nvp_idx);
527         return;
528     }
529 
530     if (!xive2_nvp_is_valid(&nvp)) {
531         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
532                       nvp_blk, nvp_idx);
533         return;
534     }
535 
536     if (!xive2_nvp_is_hw(&nvp)) {
537         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is not HW owned\n",
538                       nvp_blk, nvp_idx);
539         return;
540     }
541 
542     if (!xive2_nvp_is_co(&nvp)) {
543         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is not checkout\n",
544                       nvp_blk, nvp_idx);
545         return;
546     }
547 
548     if (xive_get_field32(NVP2_W1_CO_THRID_VALID, nvp.w1) &&
549         xive_get_field32(NVP2_W1_CO_THRID, nvp.w1) != pir) {
550         qemu_log_mask(LOG_GUEST_ERROR,
551                       "XIVE: NVP %x/%x invalid checkout Thread %x\n",
552                       nvp_blk, nvp_idx, pir);
553         return;
554     }
555 
556     nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, regs[TM_IPB]);
557 
558     if ((nvp.w0 & NVP2_W0_P) || ring != TM_QW2_HV_POOL) {
559         /*
560          * Non-pool contexts always save CPPR (ignore p bit). XXX: Clarify
561          * whether that is the correct behaviour.
562          */
563         nvp.w2 = xive_set_field32(NVP2_W2_CPPR, nvp.w2, sig_regs[TM_CPPR]);
564     }
565     if (nvp.w0 & NVP2_W0_L) {
566         /*
567          * Typically not used. If LSMFB is restored with 0, it will
568          * force a backlog rescan
569          */
570         nvp.w2 = xive_set_field32(NVP2_W2_LSMFB, nvp.w2, regs[TM_LSMFB]);
571     }
572     if (nvp.w0 & NVP2_W0_G) {
573         nvp.w2 = xive_set_field32(NVP2_W2_LGS, nvp.w2, regs[TM_LGS]);
574     }
575     if (nvp.w0 & NVP2_W0_T) {
576         nvp.w2 = xive_set_field32(NVP2_W2_T, nvp.w2, regs[TM_T]);
577     }
578     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, &nvp, 2);
579 
580     nvp.w1 = xive_set_field32(NVP2_W1_CO, nvp.w1, 0);
581     /* NVP2_W1_CO_THRID_VALID only set once */
582     nvp.w1 = xive_set_field32(NVP2_W1_CO_THRID, nvp.w1, 0xFFFF);
583     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, &nvp, 1);
584 }
585 
586 /* POOL cam is the same as OS cam encoding */
587 static void xive2_cam_decode(uint32_t cam, uint8_t *nvp_blk,
588                              uint32_t *nvp_idx, bool *valid, bool *hw)
589 {
590     *nvp_blk = xive2_nvp_blk(cam);
591     *nvp_idx = xive2_nvp_idx(cam);
592     *valid = !!(cam & TM2_W2_VALID);
593     *hw = !!(cam & TM2_W2_HW);
594 }
595 
596 /*
597  * Encode the HW CAM line with 7bit or 8bit thread id. The thread id
598  * width and block id width is configurable at the IC level.
599  *
600  *    chipid << 24 | 0000 0000 0000 0000 1 threadid (7Bit)
601  *    chipid << 24 | 0000 0000 0000 0001 threadid   (8Bit)
602  */
603 static uint32_t xive2_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
604 {
605     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
606     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
607     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
608     uint8_t blk = xive2_router_get_block_id(xrtr);
609     uint8_t tid_shift =
610         xive2_router_get_config(xrtr) & XIVE2_THREADID_8BITS ? 8 : 7;
611     uint8_t tid_mask = (1 << tid_shift) - 1;
612 
613     return xive2_nvp_cam_line(blk, 1 << tid_shift | (pir & tid_mask));
614 }
615 
616 static void xive2_redistribute(Xive2Router *xrtr, XiveTCTX *tctx, uint8_t ring)
617 {
618     uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
619     uint8_t nsr = sig_regs[TM_NSR];
620     uint8_t pipr = sig_regs[TM_PIPR];
621     uint8_t crowd = NVx_CROWD_LVL(nsr);
622     uint8_t group = NVx_GROUP_LVL(nsr);
623     uint8_t nvgc_blk, end_blk, nvp_blk;
624     uint32_t nvgc_idx, end_idx, nvp_idx;
625     Xive2Nvgc nvgc;
626     uint8_t prio_limit;
627     uint32_t cfg;
628 
629     /* redistribution is only for group/crowd interrupts */
630     if (!xive_nsr_indicates_group_exception(ring, nsr)) {
631         return;
632     }
633 
634     /* Don't check return code since ring is expected to be invalidated */
635     xive2_tctx_get_nvp_indexes(tctx, ring, &nvp_blk, &nvp_idx);
636 
637     trace_xive_redistribute(tctx->cs->cpu_index, ring, nvp_blk, nvp_idx);
638 
639     trace_xive_redistribute(tctx->cs->cpu_index, ring, nvp_blk, nvp_idx);
640     /* convert crowd/group to blk/idx */
641     if (group > 0) {
642         nvgc_idx = (nvp_idx & (0xffffffff << group)) |
643                    ((1 << (group - 1)) - 1);
644     } else {
645         nvgc_idx = nvp_idx;
646     }
647 
648     if (crowd > 0) {
649         crowd = (crowd == 3) ? 4 : crowd;
650         nvgc_blk = (nvp_blk & (0xffffffff << crowd)) |
651                    ((1 << (crowd - 1)) - 1);
652     } else {
653         nvgc_blk = nvp_blk;
654     }
655 
656     /* Use blk/idx to retrieve the NVGC */
657     if (xive2_router_get_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, &nvgc)) {
658         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no %s %x/%x\n",
659                       crowd ? "NVC" : "NVG", nvgc_blk, nvgc_idx);
660         return;
661     }
662 
663     /* retrieve the END blk/idx from the NVGC */
664     end_blk = xive_get_field32(NVGC2_W1_END_BLK, nvgc.w1);
665     end_idx = xive_get_field32(NVGC2_W1_END_IDX, nvgc.w1);
666 
667     /* determine number of priorities being used */
668     cfg = xive2_router_get_config(xrtr);
669     if (cfg & XIVE2_EN_VP_GRP_PRIORITY) {
670         prio_limit = 1 << GETFIELD(NVGC2_W1_PSIZE, nvgc.w1);
671     } else {
672         prio_limit = 1 << GETFIELD(XIVE2_VP_INT_PRIO, cfg);
673     }
674 
675     /* add priority offset to end index */
676     end_idx += pipr % prio_limit;
677 
678     /* trigger the group END */
679     xive2_router_end_notify(xrtr, end_blk, end_idx, 0, true);
680 
681     /* clear interrupt indication for the context */
682     sig_regs[TM_NSR] = 0;
683     sig_regs[TM_PIPR] = sig_regs[TM_CPPR];
684     xive_tctx_reset_signal(tctx, ring);
685 }
686 
687 static void xive2_tctx_process_pending(XiveTCTX *tctx, uint8_t sig_ring);
688 
689 static uint64_t xive2_tm_pull_ctx(XivePresenter *xptr, XiveTCTX *tctx,
690                                   hwaddr offset, unsigned size, uint8_t ring)
691 {
692     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
693     uint32_t target_ringw2 = xive_tctx_word2(&tctx->regs[ring]);
694     uint32_t cam = be32_to_cpu(target_ringw2);
695     uint8_t nvp_blk;
696     uint32_t nvp_idx;
697     uint8_t cur_ring;
698     bool valid;
699     bool do_save;
700     uint8_t nsr;
701 
702     xive2_cam_decode(cam, &nvp_blk, &nvp_idx, &valid, &do_save);
703 
704     if (xive2_tctx_get_nvp_indexes(tctx, ring, &nvp_blk, &nvp_idx)) {
705         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: pulling invalid NVP %x/%x !?\n",
706                       nvp_blk, nvp_idx);
707     }
708 
709     /* Invalidate CAM line of requested ring and all lower rings */
710     for (cur_ring = TM_QW0_USER; cur_ring <= ring;
711          cur_ring += XIVE_TM_RING_SIZE) {
712         uint32_t ringw2 = xive_tctx_word2(&tctx->regs[cur_ring]);
713         uint32_t ringw2_new = xive_set_field32(TM2_QW1W2_VO, ringw2, 0);
714         bool is_valid = !!(xive_get_field32(TM2_QW1W2_VO, ringw2));
715         uint8_t *sig_regs;
716 
717         memcpy(&tctx->regs[cur_ring + TM_WORD2], &ringw2_new, 4);
718 
719         /* Skip the rest for USER or invalid contexts */
720         if ((cur_ring == TM_QW0_USER) || !is_valid) {
721             continue;
722         }
723 
724         /* Active group/crowd interrupts need to be redistributed */
725         sig_regs = xive_tctx_signal_regs(tctx, ring);
726         nsr = sig_regs[TM_NSR];
727         if (xive_nsr_indicates_group_exception(cur_ring, nsr)) {
728             /* Ensure ring matches NSR (for HV NSR POOL vs PHYS rings) */
729             if (cur_ring == xive_nsr_exception_ring(cur_ring, nsr)) {
730                 xive2_redistribute(xrtr, tctx, cur_ring);
731             }
732         }
733 
734         /*
735          * Lower external interrupt line of requested ring and below except for
736          * USER, which doesn't exist.
737          */
738         if (xive_nsr_indicates_exception(cur_ring, nsr)) {
739             if (cur_ring == xive_nsr_exception_ring(cur_ring, nsr)) {
740                 xive_tctx_reset_signal(tctx, cur_ring);
741             }
742         }
743     }
744 
745     if (ring == TM_QW2_HV_POOL) {
746         /* Re-check phys for interrupts if pool was disabled */
747         nsr = tctx->regs[TM_QW3_HV_PHYS + TM_NSR];
748         if (xive_nsr_indicates_exception(TM_QW3_HV_PHYS, nsr)) {
749             /* Ring must be PHYS because POOL would have been redistributed */
750             g_assert(xive_nsr_exception_ring(TM_QW3_HV_PHYS, nsr) ==
751                                                            TM_QW3_HV_PHYS);
752         } else {
753             xive2_tctx_process_pending(tctx, TM_QW3_HV_PHYS);
754         }
755     }
756 
757     if (xive2_router_get_config(xrtr) & XIVE2_VP_SAVE_RESTORE && do_save) {
758         xive2_tctx_save_ctx(xrtr, tctx, ring, nvp_blk, nvp_idx);
759     }
760 
761     return target_ringw2;
762 }
763 
764 uint64_t xive2_tm_pull_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
765                               hwaddr offset, unsigned size)
766 {
767     return xive2_tm_pull_ctx(xptr, tctx, offset, size, TM_QW1_OS);
768 }
769 
770 uint64_t xive2_tm_pull_pool_ctx(XivePresenter *xptr, XiveTCTX *tctx,
771                                 hwaddr offset, unsigned size)
772 {
773     return xive2_tm_pull_ctx(xptr, tctx, offset, size, TM_QW2_HV_POOL);
774 }
775 
776 uint64_t xive2_tm_pull_phys_ctx(XivePresenter *xptr, XiveTCTX *tctx,
777                                 hwaddr offset, unsigned size)
778 {
779     return xive2_tm_pull_ctx(xptr, tctx, offset, size, TM_QW3_HV_PHYS);
780 }
781 
782 #define REPORT_LINE_GEN1_SIZE       16
783 
784 static void xive2_tm_report_line_gen1(XiveTCTX *tctx, uint8_t *data,
785                                       uint8_t size)
786 {
787     uint8_t *regs = tctx->regs;
788 
789     g_assert(size == REPORT_LINE_GEN1_SIZE);
790     memset(data, 0, size);
791     /*
792      * See xive architecture for description of what is saved. It is
793      * hand-picked information to fit in 16 bytes.
794      */
795     data[0x0] = regs[TM_QW3_HV_PHYS + TM_NSR];
796     data[0x1] = regs[TM_QW3_HV_PHYS + TM_CPPR];
797     data[0x2] = regs[TM_QW3_HV_PHYS + TM_IPB];
798     data[0x3] = regs[TM_QW2_HV_POOL + TM_IPB];
799     data[0x4] = regs[TM_QW1_OS + TM_ACK_CNT];
800     data[0x5] = regs[TM_QW3_HV_PHYS + TM_LGS];
801     data[0x6] = 0xFF;
802     data[0x7] = regs[TM_QW3_HV_PHYS + TM_WORD2] & 0x80;
803     data[0x7] |= (regs[TM_QW2_HV_POOL + TM_WORD2] & 0x80) >> 1;
804     data[0x7] |= (regs[TM_QW1_OS + TM_WORD2] & 0x80) >> 2;
805     data[0x7] |= (regs[TM_QW3_HV_PHYS + TM_WORD2] & 0x3);
806     data[0x8] = regs[TM_QW1_OS + TM_NSR];
807     data[0x9] = regs[TM_QW1_OS + TM_CPPR];
808     data[0xA] = regs[TM_QW1_OS + TM_IPB];
809     data[0xB] = regs[TM_QW1_OS + TM_LGS];
810     if (regs[TM_QW0_USER + TM_WORD2] & 0x80) {
811         /*
812          * Logical server extension, except VU bit replaced by EB bit
813          * from NSR
814          */
815         data[0xC] = regs[TM_QW0_USER + TM_WORD2];
816         data[0xC] &= ~0x80;
817         data[0xC] |= regs[TM_QW0_USER + TM_NSR] & 0x80;
818         data[0xD] = regs[TM_QW0_USER + TM_WORD2 + 1];
819         data[0xE] = regs[TM_QW0_USER + TM_WORD2 + 2];
820         data[0xF] = regs[TM_QW0_USER + TM_WORD2 + 3];
821     }
822 }
823 
824 static void xive2_tm_pull_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
825                                  hwaddr offset, uint64_t value,
826                                  unsigned size, uint8_t ring)
827 {
828     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
829     uint32_t hw_cam, nvp_idx, xive2_cfg, reserved;
830     uint8_t nvp_blk;
831     Xive2Nvp nvp;
832     uint64_t phys_addr;
833     MemTxResult result;
834 
835     hw_cam = xive2_tctx_hw_cam_line(xptr, tctx);
836     nvp_blk = xive2_nvp_blk(hw_cam);
837     nvp_idx = xive2_nvp_idx(hw_cam);
838 
839     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
840         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
841                       nvp_blk, nvp_idx);
842         return;
843     }
844 
845     if (!xive2_nvp_is_valid(&nvp)) {
846         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
847                       nvp_blk, nvp_idx);
848         return;
849     }
850 
851     xive2_cfg = xive2_router_get_config(xrtr);
852 
853     phys_addr = xive2_nvp_reporting_addr(&nvp) + 0x80; /* odd line */
854     if (xive2_cfg & XIVE2_GEN1_TIMA_OS) {
855         uint8_t pull_ctxt[REPORT_LINE_GEN1_SIZE];
856 
857         xive2_tm_report_line_gen1(tctx, pull_ctxt, REPORT_LINE_GEN1_SIZE);
858         result = dma_memory_write(&address_space_memory, phys_addr,
859                                   pull_ctxt, REPORT_LINE_GEN1_SIZE,
860                                   MEMTXATTRS_UNSPECIFIED);
861         assert(result == MEMTX_OK);
862     } else {
863         result = dma_memory_write(&address_space_memory, phys_addr,
864                                   &tctx->regs, sizeof(tctx->regs),
865                                   MEMTXATTRS_UNSPECIFIED);
866         assert(result == MEMTX_OK);
867         reserved = 0xFFFFFFFF;
868         result = dma_memory_write(&address_space_memory, phys_addr + 12,
869                                   &reserved, sizeof(reserved),
870                                   MEMTXATTRS_UNSPECIFIED);
871         assert(result == MEMTX_OK);
872     }
873 
874     /* the rest is similar to pull context to registers */
875     xive2_tm_pull_ctx(xptr, tctx, offset, size, ring);
876 }
877 
878 void xive2_tm_pull_os_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
879                              hwaddr offset, uint64_t value, unsigned size)
880 {
881     xive2_tm_pull_ctx_ol(xptr, tctx, offset, value, size, TM_QW1_OS);
882 }
883 
884 
885 void xive2_tm_pull_phys_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
886                                hwaddr offset, uint64_t value, unsigned size)
887 {
888     xive2_tm_pull_ctx_ol(xptr, tctx, offset, value, size, TM_QW3_HV_PHYS);
889 }
890 
891 static uint8_t xive2_tctx_restore_ctx(Xive2Router *xrtr, XiveTCTX *tctx,
892                                       uint8_t ring,
893                                       uint8_t nvp_blk, uint32_t nvp_idx,
894                                       Xive2Nvp *nvp)
895 {
896     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
897     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
898     uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
899     uint8_t *regs = &tctx->regs[ring];
900     uint8_t cppr;
901 
902     if (!xive2_nvp_is_hw(nvp)) {
903         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is not HW owned\n",
904                       nvp_blk, nvp_idx);
905         return 0;
906     }
907 
908     cppr = xive_get_field32(NVP2_W2_CPPR, nvp->w2);
909     nvp->w2 = xive_set_field32(NVP2_W2_CPPR, nvp->w2, 0);
910     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, nvp, 2);
911 
912     sig_regs[TM_CPPR] = cppr;
913     regs[TM_LSMFB] = xive_get_field32(NVP2_W2_LSMFB, nvp->w2);
914     regs[TM_LGS] = xive_get_field32(NVP2_W2_LGS, nvp->w2);
915     regs[TM_T] = xive_get_field32(NVP2_W2_T, nvp->w2);
916 
917     nvp->w1 = xive_set_field32(NVP2_W1_CO, nvp->w1, 1);
918     nvp->w1 = xive_set_field32(NVP2_W1_CO_THRID_VALID, nvp->w1, 1);
919     nvp->w1 = xive_set_field32(NVP2_W1_CO_THRID, nvp->w1, pir);
920 
921     /*
922      * Checkout privilege: 0:OS, 1:Pool, 2:Hard
923      *
924      * TODO: we don't support hard push/pull
925      */
926     switch (ring) {
927     case TM_QW1_OS:
928         nvp->w1 = xive_set_field32(NVP2_W1_CO_PRIV, nvp->w1, 0);
929         break;
930     case TM_QW2_HV_POOL:
931         nvp->w1 = xive_set_field32(NVP2_W1_CO_PRIV, nvp->w1, 1);
932         break;
933     default:
934         g_assert_not_reached();
935     }
936 
937     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, nvp, 1);
938 
939     /* return restored CPPR to generate a CPU exception if needed */
940     return cppr;
941 }
942 
943 static void xive2_tctx_need_resend(Xive2Router *xrtr, XiveTCTX *tctx,
944                                    uint8_t ring,
945                                    uint8_t nvp_blk, uint32_t nvp_idx,
946                                    bool do_restore)
947 {
948     uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
949     uint8_t *regs = &tctx->regs[ring];
950     uint8_t ipb, nsr = sig_regs[TM_NSR];
951     Xive2Nvp nvp;
952 
953     /*
954      * Grab the associated thread interrupt context registers in the
955      * associated NVP
956      */
957     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
958         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
959                       nvp_blk, nvp_idx);
960         return;
961     }
962 
963     if (!xive2_nvp_is_valid(&nvp)) {
964         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
965                       nvp_blk, nvp_idx);
966         return;
967     }
968 
969     /* Automatically restore thread context registers */
970     if (xive2_router_get_config(xrtr) & XIVE2_VP_SAVE_RESTORE && do_restore) {
971         xive2_tctx_restore_ctx(xrtr, tctx, ring, nvp_blk, nvp_idx, &nvp);
972     }
973 
974     ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2);
975     if (ipb) {
976         nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, 0);
977         xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, &nvp, 2);
978     }
979     /* IPB bits in the backlog are merged with the TIMA IPB bits */
980     regs[TM_IPB] |= ipb;
981 
982     if (xive_nsr_indicates_group_exception(ring, nsr)) {
983         /* redistribute precluded active grp interrupt */
984         g_assert(ring == TM_QW2_HV_POOL); /* PHYS ring has the grp interrupt */
985         xive2_redistribute(xrtr, tctx, xive_nsr_exception_ring(ring, nsr));
986     }
987     xive2_tctx_process_pending(tctx, ring == TM_QW2_HV_POOL ?
988                                          TM_QW3_HV_PHYS : ring);
989 }
990 
991 /*
992  * Updating the ring CAM line can trigger a resend of interrupt
993  */
994 static void xive2_tm_push_ctx(XivePresenter *xptr, XiveTCTX *tctx,
995                               hwaddr offset, uint64_t value, unsigned size,
996                               uint8_t ring)
997 {
998     uint32_t cam;
999     uint32_t w2;
1000     uint64_t dw1;
1001     uint8_t nvp_blk;
1002     uint32_t nvp_idx;
1003     bool v;
1004     bool do_restore;
1005 
1006     /* First update the thead context */
1007     switch (size) {
1008     case 1:
1009         tctx->regs[ring + TM_WORD2] = value & 0xff;
1010         cam = xive2_tctx_hw_cam_line(xptr, tctx);
1011         cam |= ((value & 0xc0) << 24); /* V and H bits */
1012         break;
1013     case 4:
1014         cam = value;
1015         w2 = cpu_to_be32(cam);
1016         memcpy(&tctx->regs[ring + TM_WORD2], &w2, 4);
1017         break;
1018     case 8:
1019         cam = value >> 32;
1020         dw1 = cpu_to_be64(value);
1021         memcpy(&tctx->regs[ring + TM_WORD2], &dw1, 8);
1022         break;
1023     default:
1024         g_assert_not_reached();
1025     }
1026 
1027     xive2_cam_decode(cam, &nvp_blk, &nvp_idx, &v, &do_restore);
1028 
1029     /* Check the interrupt pending bits */
1030     if (v) {
1031         xive2_tctx_need_resend(XIVE2_ROUTER(xptr), tctx, ring,
1032                                nvp_blk, nvp_idx, do_restore);
1033     }
1034 }
1035 
1036 void xive2_tm_push_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
1037                           hwaddr offset, uint64_t value, unsigned size)
1038 {
1039     xive2_tm_push_ctx(xptr, tctx, offset, value, size, TM_QW1_OS);
1040 }
1041 
1042 void xive2_tm_push_pool_ctx(XivePresenter *xptr, XiveTCTX *tctx,
1043                             hwaddr offset, uint64_t value, unsigned size)
1044 {
1045     xive2_tm_push_ctx(xptr, tctx, offset, value, size, TM_QW2_HV_POOL);
1046 }
1047 
1048 void xive2_tm_push_phys_ctx(XivePresenter *xptr, XiveTCTX *tctx,
1049                             hwaddr offset, uint64_t value, unsigned size)
1050 {
1051     xive2_tm_push_ctx(xptr, tctx, offset, value, size, TM_QW3_HV_PHYS);
1052 }
1053 
1054 /* returns -1 if ring is invalid, but still populates block and index */
1055 static int xive2_tctx_get_nvp_indexes(XiveTCTX *tctx, uint8_t ring,
1056                                       uint8_t *nvp_blk, uint32_t *nvp_idx)
1057 {
1058     uint32_t w2;
1059     uint32_t cam = 0;
1060     int rc = 0;
1061 
1062     w2 = xive_tctx_word2(&tctx->regs[ring]);
1063     switch (ring) {
1064     case TM_QW1_OS:
1065         if (!(be32_to_cpu(w2) & TM2_QW1W2_VO)) {
1066             rc = -1;
1067         }
1068         cam = xive_get_field32(TM2_QW1W2_OS_CAM, w2);
1069         break;
1070     case TM_QW2_HV_POOL:
1071         if (!(be32_to_cpu(w2) & TM2_QW2W2_VP)) {
1072             rc = -1;
1073         }
1074         cam = xive_get_field32(TM2_QW2W2_POOL_CAM, w2);
1075         break;
1076     case TM_QW3_HV_PHYS:
1077         if (!(be32_to_cpu(w2) & TM2_QW3W2_VT)) {
1078             rc = -1;
1079         }
1080         cam = xive2_tctx_hw_cam_line(tctx->xptr, tctx);
1081         break;
1082     default:
1083         rc = -1;
1084     }
1085     *nvp_blk = xive2_nvp_blk(cam);
1086     *nvp_idx = xive2_nvp_idx(cam);
1087     return rc;
1088 }
1089 
1090 static void xive2_tctx_accept_el(XivePresenter *xptr, XiveTCTX *tctx,
1091                                  uint8_t ring, uint8_t cl_ring)
1092 {
1093     uint64_t rd;
1094     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
1095     uint32_t nvp_idx, xive2_cfg;
1096     uint8_t nvp_blk;
1097     Xive2Nvp nvp;
1098     uint64_t phys_addr;
1099     uint8_t OGen = 0;
1100 
1101     xive2_tctx_get_nvp_indexes(tctx, cl_ring, &nvp_blk, &nvp_idx);
1102 
1103     if (xive2_router_get_nvp(xrtr, (uint8_t)nvp_blk, nvp_idx, &nvp)) {
1104         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
1105                       nvp_blk, nvp_idx);
1106         return;
1107     }
1108 
1109     if (!xive2_nvp_is_valid(&nvp)) {
1110         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1111                       nvp_blk, nvp_idx);
1112         return;
1113     }
1114 
1115 
1116     rd = xive_tctx_accept(tctx, ring);
1117 
1118     if (ring == TM_QW1_OS) {
1119         OGen = tctx->regs[ring + TM_OGEN];
1120     }
1121     xive2_cfg = xive2_router_get_config(xrtr);
1122     phys_addr = xive2_nvp_reporting_addr(&nvp);
1123     uint8_t report_data[REPORT_LINE_GEN1_SIZE];
1124     memset(report_data, 0xff, sizeof(report_data));
1125     if ((OGen == 1) || (xive2_cfg & XIVE2_GEN1_TIMA_OS)) {
1126         report_data[8] = (rd >> 8) & 0xff;
1127         report_data[9] = rd & 0xff;
1128     } else {
1129         report_data[0] = (rd >> 8) & 0xff;
1130         report_data[1] = rd & 0xff;
1131     }
1132     cpu_physical_memory_write(phys_addr, report_data, REPORT_LINE_GEN1_SIZE);
1133 }
1134 
1135 void xive2_tm_ack_os_el(XivePresenter *xptr, XiveTCTX *tctx,
1136                         hwaddr offset, uint64_t value, unsigned size)
1137 {
1138     xive2_tctx_accept_el(xptr, tctx, TM_QW1_OS, TM_QW1_OS);
1139 }
1140 
1141 /* Re-calculate and present pending interrupts */
1142 static void xive2_tctx_process_pending(XiveTCTX *tctx, uint8_t sig_ring)
1143 {
1144     uint8_t *sig_regs = &tctx->regs[sig_ring];
1145     Xive2Router *xrtr = XIVE2_ROUTER(tctx->xptr);
1146     uint8_t backlog_prio;
1147     uint8_t first_group;
1148     uint8_t group_level;
1149     uint8_t pipr_min;
1150     uint8_t lsmfb_min;
1151     uint8_t ring_min;
1152     uint8_t cppr = sig_regs[TM_CPPR];
1153     bool group_enabled;
1154     Xive2Nvp nvp;
1155     int rc;
1156 
1157     g_assert(sig_ring == TM_QW3_HV_PHYS || sig_ring == TM_QW1_OS);
1158     g_assert(!xive_nsr_indicates_group_exception(sig_ring, sig_regs[TM_NSR]));
1159 
1160     /*
1161      * Recompute the PIPR based on local pending interrupts. It will
1162      * be adjusted below if needed in case of pending group interrupts.
1163      */
1164 again:
1165     pipr_min = xive_ipb_to_pipr(sig_regs[TM_IPB]);
1166     group_enabled = !!sig_regs[TM_LGS];
1167     lsmfb_min = group_enabled ? sig_regs[TM_LSMFB] : 0xff;
1168     ring_min = sig_ring;
1169     group_level = 0;
1170 
1171     /* PHYS updates also depend on POOL values */
1172     if (sig_ring == TM_QW3_HV_PHYS) {
1173         uint8_t *pool_regs = &tctx->regs[TM_QW2_HV_POOL];
1174 
1175         /* POOL values only matter if POOL ctx is valid */
1176         if (pool_regs[TM_WORD2] & 0x80) {
1177             uint8_t pool_pipr = xive_ipb_to_pipr(pool_regs[TM_IPB]);
1178             uint8_t pool_lsmfb = pool_regs[TM_LSMFB];
1179 
1180             /*
1181              * Determine highest priority interrupt and
1182              * remember which ring has it.
1183              */
1184             if (pool_pipr < pipr_min) {
1185                 pipr_min = pool_pipr;
1186                 if (pool_pipr < lsmfb_min) {
1187                     ring_min = TM_QW2_HV_POOL;
1188                 }
1189             }
1190 
1191             /* Values needed for group priority calculation */
1192             if (pool_regs[TM_LGS] && (pool_lsmfb < lsmfb_min)) {
1193                 group_enabled = true;
1194                 lsmfb_min = pool_lsmfb;
1195                 if (lsmfb_min < pipr_min) {
1196                     ring_min = TM_QW2_HV_POOL;
1197                 }
1198             }
1199         }
1200     }
1201 
1202     if (group_enabled &&
1203         lsmfb_min < cppr &&
1204         lsmfb_min < pipr_min) {
1205 
1206         uint8_t nvp_blk;
1207         uint32_t nvp_idx;
1208 
1209         /*
1210          * Thread has seen a group interrupt with a higher priority
1211          * than the new cppr or pending local interrupt. Check the
1212          * backlog
1213          */
1214         rc = xive2_tctx_get_nvp_indexes(tctx, ring_min, &nvp_blk, &nvp_idx);
1215         if (rc) {
1216             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: set CPPR on invalid "
1217                                            "context\n");
1218             return;
1219         }
1220 
1221         if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
1222             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
1223                           nvp_blk, nvp_idx);
1224             return;
1225         }
1226 
1227         if (!xive2_nvp_is_valid(&nvp)) {
1228             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1229                           nvp_blk, nvp_idx);
1230             return;
1231         }
1232 
1233         first_group = xive_get_field32(NVP2_W0_PGOFIRST, nvp.w0);
1234         if (!first_group) {
1235             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1236                           nvp_blk, nvp_idx);
1237             return;
1238         }
1239 
1240         backlog_prio = xive2_presenter_backlog_scan(tctx->xptr,
1241                                                     nvp_blk, nvp_idx,
1242                                                     first_group, &group_level);
1243         tctx->regs[ring_min + TM_LSMFB] = backlog_prio;
1244         if (backlog_prio != lsmfb_min) {
1245             /*
1246              * If the group backlog scan finds a less favored or no interrupt,
1247              * then re-do the processing which may turn up a more favored
1248              * interrupt from IPB or the other pool. Backlog should not
1249              * find a priority < LSMFB.
1250              */
1251             g_assert(backlog_prio >= lsmfb_min);
1252             goto again;
1253         }
1254 
1255         xive2_presenter_backlog_decr(tctx->xptr, nvp_blk, nvp_idx,
1256                                      backlog_prio, group_level);
1257         pipr_min = backlog_prio;
1258     }
1259 
1260     if (pipr_min > cppr) {
1261         pipr_min = cppr;
1262     }
1263     xive_tctx_pipr_set(tctx, ring_min, pipr_min, group_level);
1264 }
1265 
1266 /* NOTE: CPPR only exists for TM_QW1_OS and TM_QW3_HV_PHYS */
1267 static void xive2_tctx_set_cppr(XiveTCTX *tctx, uint8_t sig_ring, uint8_t cppr)
1268 {
1269     uint8_t *sig_regs = &tctx->regs[sig_ring];
1270     Xive2Router *xrtr = XIVE2_ROUTER(tctx->xptr);
1271     uint8_t old_cppr;
1272     uint8_t nsr = sig_regs[TM_NSR];
1273 
1274     g_assert(sig_ring == TM_QW1_OS || sig_ring == TM_QW3_HV_PHYS);
1275 
1276     g_assert(tctx->regs[TM_QW2_HV_POOL + TM_NSR] == 0);
1277     g_assert(tctx->regs[TM_QW2_HV_POOL + TM_PIPR] == 0);
1278     g_assert(tctx->regs[TM_QW2_HV_POOL + TM_CPPR] == 0);
1279 
1280     /* XXX: should show pool IPB for PHYS ring */
1281     trace_xive_tctx_set_cppr(tctx->cs->cpu_index, sig_ring,
1282                              sig_regs[TM_IPB], sig_regs[TM_PIPR],
1283                              cppr, nsr);
1284 
1285     if (cppr > XIVE_PRIORITY_MAX) {
1286         cppr = 0xff;
1287     }
1288 
1289     old_cppr = sig_regs[TM_CPPR];
1290     sig_regs[TM_CPPR] = cppr;
1291 
1292     /* Handle increased CPPR priority (lower value) */
1293     if (cppr < old_cppr) {
1294         if (cppr <= sig_regs[TM_PIPR]) {
1295             /* CPPR lowered below PIPR, must un-present interrupt */
1296             if (xive_nsr_indicates_exception(sig_ring, nsr)) {
1297                 if (xive_nsr_indicates_group_exception(sig_ring, nsr)) {
1298                     /* redistribute precluded active grp interrupt */
1299                     xive2_redistribute(xrtr, tctx,
1300                                        xive_nsr_exception_ring(sig_ring, nsr));
1301                     return;
1302                 }
1303             }
1304 
1305             /* interrupt is VP directed, pending in IPB */
1306             xive_tctx_pipr_set(tctx, sig_ring, cppr, 0);
1307             return;
1308         } else {
1309             /* CPPR was lowered, but still above PIPR. No action needed. */
1310             return;
1311         }
1312     }
1313 
1314     /* CPPR didn't change, nothing needs to be done */
1315     if (cppr == old_cppr) {
1316         return;
1317     }
1318 
1319     /* CPPR priority decreased (higher value) */
1320     if (!xive_nsr_indicates_exception(sig_ring, nsr)) {
1321         xive2_tctx_process_pending(tctx, sig_ring);
1322     }
1323 }
1324 
1325 void xive2_tm_set_hv_cppr(XivePresenter *xptr, XiveTCTX *tctx,
1326                           hwaddr offset, uint64_t value, unsigned size)
1327 {
1328     xive2_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
1329 }
1330 
1331 void xive2_tm_set_os_cppr(XivePresenter *xptr, XiveTCTX *tctx,
1332                           hwaddr offset, uint64_t value, unsigned size)
1333 {
1334     xive2_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
1335 }
1336 
1337 /*
1338  * Adjust the IPB to allow a CPU to process event queues of other
1339  * priorities during one physical interrupt cycle.
1340  */
1341 void xive2_tm_set_os_pending(XivePresenter *xptr, XiveTCTX *tctx,
1342                              hwaddr offset, uint64_t value, unsigned size)
1343 {
1344     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
1345     uint8_t ring = TM_QW1_OS;
1346     uint8_t *regs = &tctx->regs[ring];
1347     uint8_t priority = value & 0xff;
1348 
1349     /*
1350      * XXX: should this simply set a bit in IPB and wait for it to be picked
1351      * up next cycle, or is it supposed to present it now? We implement the
1352      * latter here.
1353      */
1354     regs[TM_IPB] |= xive_priority_to_ipb(priority);
1355     if (xive_ipb_to_pipr(regs[TM_IPB]) >= regs[TM_PIPR]) {
1356         return;
1357     }
1358     if (xive_nsr_indicates_group_exception(ring, regs[TM_NSR])) {
1359         xive2_redistribute(xrtr, tctx, ring);
1360     }
1361 
1362     xive_tctx_pipr_present(tctx, ring, priority, 0);
1363 }
1364 
1365 static void xive2_tctx_set_target(XiveTCTX *tctx, uint8_t ring, uint8_t target)
1366 {
1367     uint8_t *regs = &tctx->regs[ring];
1368 
1369     regs[TM_T] = target;
1370 }
1371 
1372 void xive2_tm_set_hv_target(XivePresenter *xptr, XiveTCTX *tctx,
1373                             hwaddr offset, uint64_t value, unsigned size)
1374 {
1375     xive2_tctx_set_target(tctx, TM_QW3_HV_PHYS, value & 0xff);
1376 }
1377 
1378 /*
1379  * XIVE Router (aka. Virtualization Controller or IVRE)
1380  */
1381 
1382 int xive2_router_get_eas(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1383                          Xive2Eas *eas)
1384 {
1385     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1386 
1387     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1388 }
1389 
1390 static
1391 int xive2_router_get_pq(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1392                        uint8_t *pq)
1393 {
1394     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1395 
1396     return xrc->get_pq(xrtr, eas_blk, eas_idx, pq);
1397 }
1398 
1399 static
1400 int xive2_router_set_pq(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1401                        uint8_t *pq)
1402 {
1403     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1404 
1405     return xrc->set_pq(xrtr, eas_blk, eas_idx, pq);
1406 }
1407 
1408 int xive2_router_get_end(Xive2Router *xrtr, uint8_t end_blk, uint32_t end_idx,
1409                          Xive2End *end)
1410 {
1411    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1412 
1413    return xrc->get_end(xrtr, end_blk, end_idx, end);
1414 }
1415 
1416 int xive2_router_write_end(Xive2Router *xrtr, uint8_t end_blk, uint32_t end_idx,
1417                            Xive2End *end, uint8_t word_number)
1418 {
1419    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1420 
1421    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1422 }
1423 
1424 int xive2_router_get_nvp(Xive2Router *xrtr, uint8_t nvp_blk, uint32_t nvp_idx,
1425                          Xive2Nvp *nvp)
1426 {
1427    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1428 
1429    return xrc->get_nvp(xrtr, nvp_blk, nvp_idx, nvp);
1430 }
1431 
1432 int xive2_router_write_nvp(Xive2Router *xrtr, uint8_t nvp_blk, uint32_t nvp_idx,
1433                            Xive2Nvp *nvp, uint8_t word_number)
1434 {
1435    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1436 
1437    return xrc->write_nvp(xrtr, nvp_blk, nvp_idx, nvp, word_number);
1438 }
1439 
1440 int xive2_router_get_nvgc(Xive2Router *xrtr, bool crowd,
1441                           uint8_t nvgc_blk, uint32_t nvgc_idx,
1442                           Xive2Nvgc *nvgc)
1443 {
1444    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1445 
1446    return xrc->get_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
1447 }
1448 
1449 int xive2_router_write_nvgc(Xive2Router *xrtr, bool crowd,
1450                             uint8_t nvgc_blk, uint32_t nvgc_idx,
1451                             Xive2Nvgc *nvgc)
1452 {
1453    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1454 
1455    return xrc->write_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
1456 }
1457 
1458 static bool xive2_vp_match_mask(uint32_t cam1, uint32_t cam2,
1459                                 uint32_t vp_mask)
1460 {
1461     return (cam1 & vp_mask) == (cam2 & vp_mask);
1462 }
1463 
1464 static uint8_t xive2_get_vp_block_mask(uint32_t nvt_blk, bool crowd)
1465 {
1466     uint8_t block_mask = 0b1111;
1467 
1468     /* 3 supported crowd sizes: 2, 4, 16 */
1469     if (crowd) {
1470         uint32_t size = xive_get_vpgroup_size(nvt_blk);
1471 
1472         if (size != 2 && size != 4 && size != 16) {
1473             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid crowd size of %d",
1474                                            size);
1475             return block_mask;
1476         }
1477         block_mask &= ~(size - 1);
1478     }
1479     return block_mask;
1480 }
1481 
1482 static uint32_t xive2_get_vp_index_mask(uint32_t nvt_index, bool cam_ignore)
1483 {
1484     uint32_t index_mask = 0xFFFFFF; /* 24 bits */
1485 
1486     if (cam_ignore) {
1487         uint32_t size = xive_get_vpgroup_size(nvt_index);
1488 
1489         if (size < 2) {
1490             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid group size of %d",
1491                                            size);
1492             return index_mask;
1493         }
1494         index_mask &= ~(size - 1);
1495     }
1496     return index_mask;
1497 }
1498 
1499 /*
1500  * The thread context register words are in big-endian format.
1501  */
1502 int xive2_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1503                                uint8_t format,
1504                                uint8_t nvt_blk, uint32_t nvt_idx,
1505                                bool crowd, bool cam_ignore,
1506                                uint32_t logic_serv)
1507 {
1508     uint32_t cam =   xive2_nvp_cam_line(nvt_blk, nvt_idx);
1509     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1510     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1511     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1512     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1513 
1514     uint32_t index_mask, vp_mask;
1515     uint8_t block_mask;
1516 
1517     if (format == 0) {
1518         /*
1519          * i=0: Specific NVT notification
1520          * i=1: VP-group notification (bits ignored at the end of the
1521          *      NVT identifier)
1522          */
1523         block_mask = xive2_get_vp_block_mask(nvt_blk, crowd);
1524         index_mask = xive2_get_vp_index_mask(nvt_idx, cam_ignore);
1525         vp_mask = xive2_nvp_cam_line(block_mask, index_mask);
1526 
1527         /* For VP-group notifications, threads with LGS=0 are excluded */
1528 
1529         /* PHYS ring */
1530         if ((be32_to_cpu(qw3w2) & TM2_QW3W2_VT) &&
1531             !(cam_ignore && tctx->regs[TM_QW3_HV_PHYS + TM_LGS] == 0) &&
1532             xive2_vp_match_mask(cam,
1533                                 xive2_tctx_hw_cam_line(xptr, tctx),
1534                                 vp_mask)) {
1535             return TM_QW3_HV_PHYS;
1536         }
1537 
1538         /* HV POOL ring */
1539         if ((be32_to_cpu(qw2w2) & TM2_QW2W2_VP) &&
1540             !(cam_ignore && tctx->regs[TM_QW2_HV_POOL + TM_LGS] == 0) &&
1541             xive2_vp_match_mask(cam,
1542                                 xive_get_field32(TM2_QW2W2_POOL_CAM, qw2w2),
1543                                 vp_mask)) {
1544             return TM_QW2_HV_POOL;
1545         }
1546 
1547         /* OS ring */
1548         if ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
1549             !(cam_ignore && tctx->regs[TM_QW1_OS + TM_LGS] == 0) &&
1550             xive2_vp_match_mask(cam,
1551                                 xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2),
1552                                 vp_mask)) {
1553             return TM_QW1_OS;
1554         }
1555     } else {
1556         /* F=1 : User level Event-Based Branch (EBB) notification */
1557 
1558         /* FIXME: what if cam_ignore and LGS = 0 ? */
1559         /* USER ring */
1560         if  ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
1561              (cam == xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2)) &&
1562              (be32_to_cpu(qw0w2) & TM2_QW0W2_VU) &&
1563              (logic_serv == xive_get_field32(TM2_QW0W2_LOGIC_SERV, qw0w2))) {
1564             return TM_QW0_USER;
1565         }
1566     }
1567     return -1;
1568 }
1569 
1570 bool xive2_tm_irq_precluded(XiveTCTX *tctx, int ring, uint8_t priority)
1571 {
1572     uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
1573 
1574     /*
1575      * The xive2_presenter_tctx_match() above tells if there's a match
1576      * but for VP-group notification, we still need to look at the
1577      * priority to know if the thread can take the interrupt now or if
1578      * it is precluded.
1579      */
1580     if (priority < sig_regs[TM_PIPR]) {
1581         return false;
1582     }
1583     return true;
1584 }
1585 
1586 void xive2_tm_set_lsmfb(XiveTCTX *tctx, int ring, uint8_t priority)
1587 {
1588     uint8_t *regs = &tctx->regs[ring];
1589 
1590     /*
1591      * Called by the router during a VP-group notification when the
1592      * thread matches but can't take the interrupt because it's
1593      * already running at a more favored priority. It then stores the
1594      * new interrupt priority in the LSMFB field.
1595      */
1596     regs[TM_LSMFB] = priority;
1597 }
1598 
1599 static void xive2_router_realize(DeviceState *dev, Error **errp)
1600 {
1601     Xive2Router *xrtr = XIVE2_ROUTER(dev);
1602 
1603     assert(xrtr->xfb);
1604 }
1605 
1606 /*
1607  * Notification using the END ESe/ESn bit (Event State Buffer for
1608  * escalation and notification). Profide further coalescing in the
1609  * Router.
1610  */
1611 static bool xive2_router_end_es_notify(Xive2Router *xrtr, uint8_t end_blk,
1612                                        uint32_t end_idx, Xive2End *end,
1613                                        uint32_t end_esmask)
1614 {
1615     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1616     bool notify = xive_esb_trigger(&pq);
1617 
1618     if (pq != xive_get_field32(end_esmask, end->w1)) {
1619         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1620         xive2_router_write_end(xrtr, end_blk, end_idx, end, 1);
1621     }
1622 
1623     /* ESe/n[Q]=1 : end of notification */
1624     return notify;
1625 }
1626 
1627 /*
1628  * An END trigger can come from an event trigger (IPI or HW) or from
1629  * another chip. We don't model the PowerBus but the END trigger
1630  * message has the same parameters than in the function below.
1631  */
1632 static void xive2_router_end_notify(Xive2Router *xrtr, uint8_t end_blk,
1633                                     uint32_t end_idx, uint32_t end_data,
1634                                     bool redistribute)
1635 {
1636     Xive2End end;
1637     uint8_t priority;
1638     uint8_t format;
1639     XiveTCTXMatch match;
1640     bool crowd, cam_ignore;
1641     uint8_t nvx_blk;
1642     uint32_t nvx_idx;
1643 
1644     /* END cache lookup */
1645     if (xive2_router_get_end(xrtr, end_blk, end_idx, &end)) {
1646         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1647                       end_idx);
1648         return;
1649     }
1650 
1651     if (!xive2_end_is_valid(&end)) {
1652         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1653                       end_blk, end_idx);
1654         return;
1655     }
1656 
1657     if (xive2_end_is_crowd(&end) && !xive2_end_is_ignore(&end)) {
1658         qemu_log_mask(LOG_GUEST_ERROR,
1659                       "XIVE: invalid END, 'crowd' bit requires 'ignore' bit\n");
1660         return;
1661     }
1662 
1663     if (!redistribute && xive2_end_is_enqueue(&end)) {
1664         trace_xive_end_enqueue(end_blk, end_idx, end_data);
1665         xive2_end_enqueue(&end, end_data);
1666         /* Enqueuing event data modifies the EQ toggle and index */
1667         xive2_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1668     }
1669 
1670     /*
1671      * When the END is silent, we skip the notification part.
1672      */
1673     if (xive2_end_is_silent_escalation(&end)) {
1674         goto do_escalation;
1675     }
1676 
1677     /*
1678      * The W7 format depends on the F bit in W6. It defines the type
1679      * of the notification :
1680      *
1681      *   F=0 : single or multiple NVP notification
1682      *   F=1 : User level Event-Based Branch (EBB) notification, no
1683      *         priority
1684      */
1685     format = xive_get_field32(END2_W6_FORMAT_BIT, end.w6);
1686     priority = xive_get_field32(END2_W7_F0_PRIORITY, end.w7);
1687 
1688     /* The END is masked */
1689     if (format == 0 && priority == 0xff) {
1690         return;
1691     }
1692 
1693     /*
1694      * Check the END ESn (Event State Buffer for notification) for
1695      * even further coalescing in the Router
1696      */
1697     if (!xive2_end_is_notify(&end)) {
1698         /* ESn[Q]=1 : end of notification */
1699         if (!xive2_router_end_es_notify(xrtr, end_blk, end_idx,
1700                                        &end, END2_W1_ESn)) {
1701             return;
1702         }
1703     }
1704 
1705     /*
1706      * Follows IVPE notification
1707      */
1708     nvx_blk = xive_get_field32(END2_W6_VP_BLOCK, end.w6);
1709     nvx_idx = xive_get_field32(END2_W6_VP_OFFSET, end.w6);
1710     crowd = xive2_end_is_crowd(&end);
1711     cam_ignore = xive2_end_is_ignore(&end);
1712 
1713     /* TODO: Auto EOI. */
1714     if (xive_presenter_match(xrtr->xfb, format, nvx_blk, nvx_idx,
1715                              crowd, cam_ignore, priority,
1716                              xive_get_field32(END2_W7_F1_LOG_SERVER_ID, end.w7),
1717                              &match)) {
1718         XiveTCTX *tctx = match.tctx;
1719         uint8_t ring = match.ring;
1720         uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
1721         uint8_t nsr = sig_regs[TM_NSR];
1722         uint8_t group_level;
1723 
1724         if (priority < sig_regs[TM_PIPR] &&
1725             xive_nsr_indicates_group_exception(ring, nsr)) {
1726             xive2_redistribute(xrtr, tctx, xive_nsr_exception_ring(ring, nsr));
1727         }
1728 
1729         group_level = xive_get_group_level(crowd, cam_ignore, nvx_blk, nvx_idx);
1730         trace_xive_presenter_notify(nvx_blk, nvx_idx, ring, group_level);
1731         xive_tctx_pipr_present(tctx, ring, priority, group_level);
1732         return;
1733     }
1734 
1735     /*
1736      * If no matching NVP is dispatched on a HW thread :
1737      * - specific VP: update the NVP structure if backlog is activated
1738      * - VP-group: update the backlog counter for that priority in the NVG
1739      */
1740     if (xive2_end_is_backlog(&end)) {
1741 
1742         if (format == 1) {
1743             qemu_log_mask(LOG_GUEST_ERROR,
1744                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1745                           end_blk, end_idx);
1746             return;
1747         }
1748 
1749         if (!cam_ignore) {
1750             uint8_t ipb;
1751             Xive2Nvp nvp;
1752 
1753             /* NVP cache lookup */
1754             if (xive2_router_get_nvp(xrtr, nvx_blk, nvx_idx, &nvp)) {
1755                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVP %x/%x\n",
1756                               nvx_blk, nvx_idx);
1757                 return;
1758             }
1759 
1760             if (!xive2_nvp_is_valid(&nvp)) {
1761                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is invalid\n",
1762                               nvx_blk, nvx_idx);
1763                 return;
1764             }
1765 
1766             /*
1767              * Record the IPB in the associated NVP structure for later
1768              * use. The presenter will resend the interrupt when the vCPU
1769              * is dispatched again on a HW thread.
1770              */
1771             ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2) |
1772                 xive_priority_to_ipb(priority);
1773             nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, ipb);
1774             xive2_router_write_nvp(xrtr, nvx_blk, nvx_idx, &nvp, 2);
1775         } else {
1776             Xive2Nvgc nvgc;
1777             uint32_t backlog;
1778 
1779             /*
1780              * For groups and crowds, the per-priority backlog
1781              * counters are stored in the NVG/NVC structures
1782              */
1783             if (xive2_router_get_nvgc(xrtr, crowd,
1784                                       nvx_blk, nvx_idx, &nvgc)) {
1785                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no %s %x/%x\n",
1786                               crowd ? "NVC" : "NVG", nvx_blk, nvx_idx);
1787                 return;
1788             }
1789 
1790             if (!xive2_nvgc_is_valid(&nvgc)) {
1791                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVG %x/%x is invalid\n",
1792                               nvx_blk, nvx_idx);
1793                 return;
1794             }
1795 
1796             /*
1797              * Increment the backlog counter for that priority.
1798              * We only call broadcast the first time the counter is
1799              * incremented. broadcast will set the LSMFB field of the TIMA of
1800              * relevant threads so that they know an interrupt is pending.
1801              */
1802             backlog = xive2_nvgc_get_backlog(&nvgc, priority) + 1;
1803             xive2_nvgc_set_backlog(&nvgc, priority, backlog);
1804             xive2_router_write_nvgc(xrtr, crowd, nvx_blk, nvx_idx, &nvgc);
1805 
1806             if (backlog == 1) {
1807                 XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xrtr->xfb);
1808                 xfc->broadcast(xrtr->xfb, nvx_blk, nvx_idx,
1809                                crowd, cam_ignore, priority);
1810 
1811                 if (!xive2_end_is_precluded_escalation(&end)) {
1812                     /*
1813                      * The interrupt will be picked up when the
1814                      * matching thread lowers its priority level
1815                      */
1816                     return;
1817                 }
1818             }
1819         }
1820     }
1821 
1822 do_escalation:
1823     /*
1824      * If activated, escalate notification using the ESe PQ bits and
1825      * the EAS in w4-5
1826      */
1827     if (!xive2_end_is_escalate(&end)) {
1828         return;
1829     }
1830 
1831     /*
1832      * Check the END ESe (Event State Buffer for escalation) for even
1833      * further coalescing in the Router
1834      */
1835     if (!xive2_end_is_uncond_escalation(&end)) {
1836         /* ESe[Q]=1 : end of escalation notification */
1837         if (!xive2_router_end_es_notify(xrtr, end_blk, end_idx,
1838                                        &end, END2_W1_ESe)) {
1839             return;
1840         }
1841     }
1842 
1843     if (xive2_end_is_escalate_end(&end)) {
1844         /*
1845          * Perform END Adaptive escalation processing
1846          * The END trigger becomes an Escalation trigger
1847          */
1848         uint8_t esc_blk = xive_get_field32(END2_W4_END_BLOCK, end.w4);
1849         uint32_t esc_idx = xive_get_field32(END2_W4_ESC_END_INDEX, end.w4);
1850         uint32_t esc_data = xive_get_field32(END2_W5_ESC_END_DATA, end.w5);
1851         trace_xive_escalate_end(end_blk, end_idx, esc_blk, esc_idx, esc_data);
1852         xive2_router_end_notify(xrtr, esc_blk, esc_idx, esc_data, false);
1853     } /* end END adaptive escalation */
1854 
1855     else {
1856         uint32_t lisn;              /* Logical Interrupt Source Number */
1857 
1858         /*
1859          *  Perform ESB escalation processing
1860          *      E[N] == 1 --> N
1861          *      Req[Block] <- E[ESB_Block]
1862          *      Req[Index] <- E[ESB_Index]
1863          *      Req[Offset] <- 0x000
1864          *      Execute <ESB Store> Req command
1865          */
1866         lisn = XIVE_EAS(xive_get_field32(END2_W4_END_BLOCK,     end.w4),
1867                         xive_get_field32(END2_W4_ESC_END_INDEX, end.w4));
1868 
1869         trace_xive_escalate_esb(end_blk, end_idx, lisn);
1870         xive2_notify(xrtr, lisn, true /* pq_checked */);
1871     }
1872 
1873     return;
1874 }
1875 
1876 void xive2_notify(Xive2Router *xrtr , uint32_t lisn, bool pq_checked)
1877 {
1878     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1879     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1880     Xive2Eas eas;
1881 
1882     /* EAS cache lookup */
1883     if (xive2_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1884         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1885         return;
1886     }
1887 
1888     if (!pq_checked) {
1889         bool notify;
1890         uint8_t pq;
1891 
1892         /* PQ cache lookup */
1893         if (xive2_router_get_pq(xrtr, eas_blk, eas_idx, &pq)) {
1894             /* Set FIR */
1895             g_assert_not_reached();
1896         }
1897 
1898         notify = xive_esb_trigger(&pq);
1899 
1900         if (xive2_router_set_pq(xrtr, eas_blk, eas_idx, &pq)) {
1901             /* Set FIR */
1902             g_assert_not_reached();
1903         }
1904 
1905         if (!notify) {
1906             return;
1907         }
1908     }
1909 
1910     if (!xive2_eas_is_valid(&eas)) {
1911         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN %x\n", lisn);
1912         return;
1913     }
1914 
1915     if (xive2_eas_is_masked(&eas)) {
1916         /* Notification completed */
1917         return;
1918     }
1919 
1920     /* TODO: add support for EAS resume */
1921     if (xive2_eas_is_resume(&eas)) {
1922         qemu_log_mask(LOG_UNIMP,
1923                       "XIVE: EAS resume processing unimplemented - LISN %x\n",
1924                       lisn);
1925         return;
1926     }
1927 
1928     /*
1929      * The event trigger becomes an END trigger
1930      */
1931     xive2_router_end_notify(xrtr,
1932                             xive_get_field64(EAS2_END_BLOCK, eas.w),
1933                             xive_get_field64(EAS2_END_INDEX, eas.w),
1934                             xive_get_field64(EAS2_END_DATA,  eas.w),
1935                             false);
1936     return;
1937 }
1938 
1939 void xive2_router_notify(XiveNotifier *xn, uint32_t lisn, bool pq_checked)
1940 {
1941     Xive2Router *xrtr = XIVE2_ROUTER(xn);
1942 
1943     xive2_notify(xrtr, lisn, pq_checked);
1944     return;
1945 }
1946 
1947 static const Property xive2_router_properties[] = {
1948     DEFINE_PROP_LINK("xive-fabric", Xive2Router, xfb,
1949                      TYPE_XIVE_FABRIC, XiveFabric *),
1950 };
1951 
1952 static void xive2_router_class_init(ObjectClass *klass, const void *data)
1953 {
1954     DeviceClass *dc = DEVICE_CLASS(klass);
1955     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1956 
1957     dc->desc    = "XIVE2 Router Engine";
1958     device_class_set_props(dc, xive2_router_properties);
1959     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1960     dc->realize = xive2_router_realize;
1961     xnc->notify = xive2_router_notify;
1962 }
1963 
1964 static const TypeInfo xive2_router_info = {
1965     .name          = TYPE_XIVE2_ROUTER,
1966     .parent        = TYPE_SYS_BUS_DEVICE,
1967     .abstract      = true,
1968     .instance_size = sizeof(Xive2Router),
1969     .class_size    = sizeof(Xive2RouterClass),
1970     .class_init    = xive2_router_class_init,
1971     .interfaces    = (const InterfaceInfo[]) {
1972         { TYPE_XIVE_NOTIFIER },
1973         { TYPE_XIVE_PRESENTER },
1974         { }
1975     }
1976 };
1977 
1978 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
1979 {
1980     return !((addr >> shift) & 1);
1981 }
1982 
1983 static uint64_t xive2_end_source_read(void *opaque, hwaddr addr, unsigned size)
1984 {
1985     Xive2EndSource *xsrc = XIVE2_END_SOURCE(opaque);
1986     uint32_t offset = addr & 0xFFF;
1987     uint8_t end_blk;
1988     uint32_t end_idx;
1989     Xive2End end;
1990     uint32_t end_esmask;
1991     uint8_t pq;
1992     uint64_t ret;
1993 
1994     /*
1995      * The block id should be deduced from the load address on the END
1996      * ESB MMIO but our model only supports a single block per XIVE chip.
1997      */
1998     end_blk = xive2_router_get_block_id(xsrc->xrtr);
1999     end_idx = addr >> (xsrc->esb_shift + 1);
2000 
2001     if (xive2_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
2002         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
2003                       end_idx);
2004         return -1;
2005     }
2006 
2007     if (!xive2_end_is_valid(&end)) {
2008         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
2009                       end_blk, end_idx);
2010         return -1;
2011     }
2012 
2013     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END2_W1_ESn :
2014         END2_W1_ESe;
2015     pq = xive_get_field32(end_esmask, end.w1);
2016 
2017     switch (offset) {
2018     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
2019         ret = xive_esb_eoi(&pq);
2020 
2021         /* Forward the source event notification for routing ?? */
2022         break;
2023 
2024     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
2025         ret = pq;
2026         break;
2027 
2028     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
2029     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
2030     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
2031     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
2032         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
2033         break;
2034     default:
2035         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
2036                       offset);
2037         return -1;
2038     }
2039 
2040     if (pq != xive_get_field32(end_esmask, end.w1)) {
2041         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
2042         xive2_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
2043     }
2044 
2045     return ret;
2046 }
2047 
2048 static void xive2_end_source_write(void *opaque, hwaddr addr,
2049                                    uint64_t value, unsigned size)
2050 {
2051     Xive2EndSource *xsrc = XIVE2_END_SOURCE(opaque);
2052     uint32_t offset = addr & 0xFFF;
2053     uint8_t end_blk;
2054     uint32_t end_idx;
2055     Xive2End end;
2056     uint32_t end_esmask;
2057     uint8_t pq;
2058     bool notify = false;
2059 
2060     /*
2061      * The block id should be deduced from the load address on the END
2062      * ESB MMIO but our model only supports a single block per XIVE chip.
2063      */
2064     end_blk = xive2_router_get_block_id(xsrc->xrtr);
2065     end_idx = addr >> (xsrc->esb_shift + 1);
2066 
2067     if (xive2_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
2068         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
2069                       end_idx);
2070         return;
2071     }
2072 
2073     if (!xive2_end_is_valid(&end)) {
2074         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
2075                       end_blk, end_idx);
2076         return;
2077     }
2078 
2079     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END2_W1_ESn :
2080         END2_W1_ESe;
2081     pq = xive_get_field32(end_esmask, end.w1);
2082 
2083     switch (offset) {
2084     case 0 ... 0x3FF:
2085         notify = xive_esb_trigger(&pq);
2086         break;
2087 
2088     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
2089         /* TODO: can we check StoreEOI availability from the router ? */
2090         notify = xive_esb_eoi(&pq);
2091         break;
2092 
2093     case XIVE_ESB_INJECT ... XIVE_ESB_INJECT + 0x3FF:
2094         if (end_esmask == END2_W1_ESe) {
2095             qemu_log_mask(LOG_GUEST_ERROR,
2096                           "XIVE: END %x/%x can not EQ inject on ESe\n",
2097                            end_blk, end_idx);
2098             return;
2099         }
2100         notify = true;
2101         break;
2102 
2103     default:
2104         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB write addr %d\n",
2105                       offset);
2106         return;
2107     }
2108 
2109     if (pq != xive_get_field32(end_esmask, end.w1)) {
2110         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
2111         xive2_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
2112     }
2113 
2114     /* TODO: Forward the source event notification for routing */
2115     if (notify) {
2116         ;
2117     }
2118 }
2119 
2120 static const MemoryRegionOps xive2_end_source_ops = {
2121     .read = xive2_end_source_read,
2122     .write = xive2_end_source_write,
2123     .endianness = DEVICE_BIG_ENDIAN,
2124     .valid = {
2125         .min_access_size = 1,
2126         .max_access_size = 8,
2127     },
2128     .impl = {
2129         .min_access_size = 1,
2130         .max_access_size = 8,
2131     },
2132 };
2133 
2134 static void xive2_end_source_realize(DeviceState *dev, Error **errp)
2135 {
2136     Xive2EndSource *xsrc = XIVE2_END_SOURCE(dev);
2137 
2138     assert(xsrc->xrtr);
2139 
2140     if (!xsrc->nr_ends) {
2141         error_setg(errp, "Number of interrupt needs to be greater than 0");
2142         return;
2143     }
2144 
2145     if (xsrc->esb_shift != XIVE_ESB_4K &&
2146         xsrc->esb_shift != XIVE_ESB_64K) {
2147         error_setg(errp, "Invalid ESB shift setting");
2148         return;
2149     }
2150 
2151     /*
2152      * Each END is assigned an even/odd pair of MMIO pages, the even page
2153      * manages the ESn field while the odd page manages the ESe field.
2154      */
2155     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
2156                           &xive2_end_source_ops, xsrc, "xive.end",
2157                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
2158 }
2159 
2160 static const Property xive2_end_source_properties[] = {
2161     DEFINE_PROP_UINT32("nr-ends", Xive2EndSource, nr_ends, 0),
2162     DEFINE_PROP_UINT32("shift", Xive2EndSource, esb_shift, XIVE_ESB_64K),
2163     DEFINE_PROP_LINK("xive", Xive2EndSource, xrtr, TYPE_XIVE2_ROUTER,
2164                      Xive2Router *),
2165 };
2166 
2167 static void xive2_end_source_class_init(ObjectClass *klass, const void *data)
2168 {
2169     DeviceClass *dc = DEVICE_CLASS(klass);
2170 
2171     dc->desc    = "XIVE END Source";
2172     device_class_set_props(dc, xive2_end_source_properties);
2173     dc->realize = xive2_end_source_realize;
2174     dc->user_creatable = false;
2175 }
2176 
2177 static const TypeInfo xive2_end_source_info = {
2178     .name          = TYPE_XIVE2_END_SOURCE,
2179     .parent        = TYPE_DEVICE,
2180     .instance_size = sizeof(Xive2EndSource),
2181     .class_init    = xive2_end_source_class_init,
2182 };
2183 
2184 static void xive2_register_types(void)
2185 {
2186     type_register_static(&xive2_router_info);
2187     type_register_static(&xive2_end_source_info);
2188 }
2189 
2190 type_init(xive2_register_types)
2191