xref: /openbmc/qemu/hw/intc/xive2.c (revision 6ef77843603b89b1e48a06ca0644e74e45297839)
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 4:
1009         cam = value;
1010         w2 = cpu_to_be32(cam);
1011         memcpy(&tctx->regs[ring + TM_WORD2], &w2, 4);
1012         break;
1013     case 8:
1014         cam = value >> 32;
1015         dw1 = cpu_to_be64(value);
1016         memcpy(&tctx->regs[ring + TM_WORD2], &dw1, 8);
1017         break;
1018     default:
1019         g_assert_not_reached();
1020     }
1021 
1022     xive2_cam_decode(cam, &nvp_blk, &nvp_idx, &v, &do_restore);
1023 
1024     /* Check the interrupt pending bits */
1025     if (v) {
1026         xive2_tctx_need_resend(XIVE2_ROUTER(xptr), tctx, ring,
1027                                nvp_blk, nvp_idx, do_restore);
1028     }
1029 }
1030 
1031 void xive2_tm_push_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
1032                           hwaddr offset, uint64_t value, unsigned size)
1033 {
1034     xive2_tm_push_ctx(xptr, tctx, offset, value, size, TM_QW1_OS);
1035 }
1036 
1037 void xive2_tm_push_pool_ctx(XivePresenter *xptr, XiveTCTX *tctx,
1038                             hwaddr offset, uint64_t value, unsigned size)
1039 {
1040     xive2_tm_push_ctx(xptr, tctx, offset, value, size, TM_QW2_HV_POOL);
1041 }
1042 
1043 /* returns -1 if ring is invalid, but still populates block and index */
1044 static int xive2_tctx_get_nvp_indexes(XiveTCTX *tctx, uint8_t ring,
1045                                       uint8_t *nvp_blk, uint32_t *nvp_idx)
1046 {
1047     uint32_t w2;
1048     uint32_t cam = 0;
1049     int rc = 0;
1050 
1051     w2 = xive_tctx_word2(&tctx->regs[ring]);
1052     switch (ring) {
1053     case TM_QW1_OS:
1054         if (!(be32_to_cpu(w2) & TM2_QW1W2_VO)) {
1055             rc = -1;
1056         }
1057         cam = xive_get_field32(TM2_QW1W2_OS_CAM, w2);
1058         break;
1059     case TM_QW2_HV_POOL:
1060         if (!(be32_to_cpu(w2) & TM2_QW2W2_VP)) {
1061             rc = -1;
1062         }
1063         cam = xive_get_field32(TM2_QW2W2_POOL_CAM, w2);
1064         break;
1065     case TM_QW3_HV_PHYS:
1066         if (!(be32_to_cpu(w2) & TM2_QW3W2_VT)) {
1067             rc = -1;
1068         }
1069         cam = xive2_tctx_hw_cam_line(tctx->xptr, tctx);
1070         break;
1071     default:
1072         rc = -1;
1073     }
1074     *nvp_blk = xive2_nvp_blk(cam);
1075     *nvp_idx = xive2_nvp_idx(cam);
1076     return rc;
1077 }
1078 
1079 static void xive2_tctx_accept_el(XivePresenter *xptr, XiveTCTX *tctx,
1080                                  uint8_t ring, uint8_t cl_ring)
1081 {
1082     uint64_t rd;
1083     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
1084     uint32_t nvp_idx, xive2_cfg;
1085     uint8_t nvp_blk;
1086     Xive2Nvp nvp;
1087     uint64_t phys_addr;
1088     uint8_t OGen = 0;
1089 
1090     xive2_tctx_get_nvp_indexes(tctx, cl_ring, &nvp_blk, &nvp_idx);
1091 
1092     if (xive2_router_get_nvp(xrtr, (uint8_t)nvp_blk, nvp_idx, &nvp)) {
1093         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
1094                       nvp_blk, nvp_idx);
1095         return;
1096     }
1097 
1098     if (!xive2_nvp_is_valid(&nvp)) {
1099         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1100                       nvp_blk, nvp_idx);
1101         return;
1102     }
1103 
1104 
1105     rd = xive_tctx_accept(tctx, ring);
1106 
1107     if (ring == TM_QW1_OS) {
1108         OGen = tctx->regs[ring + TM_OGEN];
1109     }
1110     xive2_cfg = xive2_router_get_config(xrtr);
1111     phys_addr = xive2_nvp_reporting_addr(&nvp);
1112     uint8_t report_data[REPORT_LINE_GEN1_SIZE];
1113     memset(report_data, 0xff, sizeof(report_data));
1114     if ((OGen == 1) || (xive2_cfg & XIVE2_GEN1_TIMA_OS)) {
1115         report_data[8] = (rd >> 8) & 0xff;
1116         report_data[9] = rd & 0xff;
1117     } else {
1118         report_data[0] = (rd >> 8) & 0xff;
1119         report_data[1] = rd & 0xff;
1120     }
1121     cpu_physical_memory_write(phys_addr, report_data, REPORT_LINE_GEN1_SIZE);
1122 }
1123 
1124 void xive2_tm_ack_os_el(XivePresenter *xptr, XiveTCTX *tctx,
1125                         hwaddr offset, uint64_t value, unsigned size)
1126 {
1127     xive2_tctx_accept_el(xptr, tctx, TM_QW1_OS, TM_QW1_OS);
1128 }
1129 
1130 /* Re-calculate and present pending interrupts */
1131 static void xive2_tctx_process_pending(XiveTCTX *tctx, uint8_t sig_ring)
1132 {
1133     uint8_t *sig_regs = &tctx->regs[sig_ring];
1134     Xive2Router *xrtr = XIVE2_ROUTER(tctx->xptr);
1135     uint8_t backlog_prio;
1136     uint8_t first_group;
1137     uint8_t group_level;
1138     uint8_t pipr_min;
1139     uint8_t lsmfb_min;
1140     uint8_t ring_min;
1141     uint8_t cppr = sig_regs[TM_CPPR];
1142     bool group_enabled;
1143     Xive2Nvp nvp;
1144     int rc;
1145 
1146     g_assert(sig_ring == TM_QW3_HV_PHYS || sig_ring == TM_QW1_OS);
1147     g_assert(!xive_nsr_indicates_group_exception(sig_ring, sig_regs[TM_NSR]));
1148 
1149     /*
1150      * Recompute the PIPR based on local pending interrupts. It will
1151      * be adjusted below if needed in case of pending group interrupts.
1152      */
1153 again:
1154     pipr_min = xive_ipb_to_pipr(sig_regs[TM_IPB]);
1155     group_enabled = !!sig_regs[TM_LGS];
1156     lsmfb_min = group_enabled ? sig_regs[TM_LSMFB] : 0xff;
1157     ring_min = sig_ring;
1158     group_level = 0;
1159 
1160     /* PHYS updates also depend on POOL values */
1161     if (sig_ring == TM_QW3_HV_PHYS) {
1162         uint8_t *pool_regs = &tctx->regs[TM_QW2_HV_POOL];
1163 
1164         /* POOL values only matter if POOL ctx is valid */
1165         if (pool_regs[TM_WORD2] & 0x80) {
1166             uint8_t pool_pipr = xive_ipb_to_pipr(pool_regs[TM_IPB]);
1167             uint8_t pool_lsmfb = pool_regs[TM_LSMFB];
1168 
1169             /*
1170              * Determine highest priority interrupt and
1171              * remember which ring has it.
1172              */
1173             if (pool_pipr < pipr_min) {
1174                 pipr_min = pool_pipr;
1175                 if (pool_pipr < lsmfb_min) {
1176                     ring_min = TM_QW2_HV_POOL;
1177                 }
1178             }
1179 
1180             /* Values needed for group priority calculation */
1181             if (pool_regs[TM_LGS] && (pool_lsmfb < lsmfb_min)) {
1182                 group_enabled = true;
1183                 lsmfb_min = pool_lsmfb;
1184                 if (lsmfb_min < pipr_min) {
1185                     ring_min = TM_QW2_HV_POOL;
1186                 }
1187             }
1188         }
1189     }
1190 
1191     if (group_enabled &&
1192         lsmfb_min < cppr &&
1193         lsmfb_min < pipr_min) {
1194 
1195         uint8_t nvp_blk;
1196         uint32_t nvp_idx;
1197 
1198         /*
1199          * Thread has seen a group interrupt with a higher priority
1200          * than the new cppr or pending local interrupt. Check the
1201          * backlog
1202          */
1203         rc = xive2_tctx_get_nvp_indexes(tctx, ring_min, &nvp_blk, &nvp_idx);
1204         if (rc) {
1205             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: set CPPR on invalid "
1206                                            "context\n");
1207             return;
1208         }
1209 
1210         if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
1211             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
1212                           nvp_blk, nvp_idx);
1213             return;
1214         }
1215 
1216         if (!xive2_nvp_is_valid(&nvp)) {
1217             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1218                           nvp_blk, nvp_idx);
1219             return;
1220         }
1221 
1222         first_group = xive_get_field32(NVP2_W0_PGOFIRST, nvp.w0);
1223         if (!first_group) {
1224             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1225                           nvp_blk, nvp_idx);
1226             return;
1227         }
1228 
1229         backlog_prio = xive2_presenter_backlog_scan(tctx->xptr,
1230                                                     nvp_blk, nvp_idx,
1231                                                     first_group, &group_level);
1232         tctx->regs[ring_min + TM_LSMFB] = backlog_prio;
1233         if (backlog_prio != lsmfb_min) {
1234             /*
1235              * If the group backlog scan finds a less favored or no interrupt,
1236              * then re-do the processing which may turn up a more favored
1237              * interrupt from IPB or the other pool. Backlog should not
1238              * find a priority < LSMFB.
1239              */
1240             g_assert(backlog_prio >= lsmfb_min);
1241             goto again;
1242         }
1243 
1244         xive2_presenter_backlog_decr(tctx->xptr, nvp_blk, nvp_idx,
1245                                      backlog_prio, group_level);
1246         pipr_min = backlog_prio;
1247     }
1248 
1249     if (pipr_min > cppr) {
1250         pipr_min = cppr;
1251     }
1252     xive_tctx_pipr_set(tctx, ring_min, pipr_min, group_level);
1253 }
1254 
1255 /* NOTE: CPPR only exists for TM_QW1_OS and TM_QW3_HV_PHYS */
1256 static void xive2_tctx_set_cppr(XiveTCTX *tctx, uint8_t sig_ring, uint8_t cppr)
1257 {
1258     uint8_t *sig_regs = &tctx->regs[sig_ring];
1259     Xive2Router *xrtr = XIVE2_ROUTER(tctx->xptr);
1260     uint8_t old_cppr;
1261     uint8_t nsr = sig_regs[TM_NSR];
1262 
1263     g_assert(sig_ring == TM_QW1_OS || sig_ring == TM_QW3_HV_PHYS);
1264 
1265     g_assert(tctx->regs[TM_QW2_HV_POOL + TM_NSR] == 0);
1266     g_assert(tctx->regs[TM_QW2_HV_POOL + TM_PIPR] == 0);
1267     g_assert(tctx->regs[TM_QW2_HV_POOL + TM_CPPR] == 0);
1268 
1269     /* XXX: should show pool IPB for PHYS ring */
1270     trace_xive_tctx_set_cppr(tctx->cs->cpu_index, sig_ring,
1271                              sig_regs[TM_IPB], sig_regs[TM_PIPR],
1272                              cppr, nsr);
1273 
1274     if (cppr > XIVE_PRIORITY_MAX) {
1275         cppr = 0xff;
1276     }
1277 
1278     old_cppr = sig_regs[TM_CPPR];
1279     sig_regs[TM_CPPR] = cppr;
1280 
1281     /* Handle increased CPPR priority (lower value) */
1282     if (cppr < old_cppr) {
1283         if (cppr <= sig_regs[TM_PIPR]) {
1284             /* CPPR lowered below PIPR, must un-present interrupt */
1285             if (xive_nsr_indicates_exception(sig_ring, nsr)) {
1286                 if (xive_nsr_indicates_group_exception(sig_ring, nsr)) {
1287                     /* redistribute precluded active grp interrupt */
1288                     xive2_redistribute(xrtr, tctx,
1289                                        xive_nsr_exception_ring(sig_ring, nsr));
1290                     return;
1291                 }
1292             }
1293 
1294             /* interrupt is VP directed, pending in IPB */
1295             xive_tctx_pipr_set(tctx, sig_ring, cppr, 0);
1296             return;
1297         } else {
1298             /* CPPR was lowered, but still above PIPR. No action needed. */
1299             return;
1300         }
1301     }
1302 
1303     /* CPPR didn't change, nothing needs to be done */
1304     if (cppr == old_cppr) {
1305         return;
1306     }
1307 
1308     /* CPPR priority decreased (higher value) */
1309     if (!xive_nsr_indicates_exception(sig_ring, nsr)) {
1310         xive2_tctx_process_pending(tctx, sig_ring);
1311     }
1312 }
1313 
1314 void xive2_tm_set_hv_cppr(XivePresenter *xptr, XiveTCTX *tctx,
1315                           hwaddr offset, uint64_t value, unsigned size)
1316 {
1317     xive2_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
1318 }
1319 
1320 void xive2_tm_set_os_cppr(XivePresenter *xptr, XiveTCTX *tctx,
1321                           hwaddr offset, uint64_t value, unsigned size)
1322 {
1323     xive2_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
1324 }
1325 
1326 /*
1327  * Adjust the IPB to allow a CPU to process event queues of other
1328  * priorities during one physical interrupt cycle.
1329  */
1330 void xive2_tm_set_os_pending(XivePresenter *xptr, XiveTCTX *tctx,
1331                              hwaddr offset, uint64_t value, unsigned size)
1332 {
1333     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
1334     uint8_t ring = TM_QW1_OS;
1335     uint8_t *regs = &tctx->regs[ring];
1336     uint8_t priority = value & 0xff;
1337 
1338     /*
1339      * XXX: should this simply set a bit in IPB and wait for it to be picked
1340      * up next cycle, or is it supposed to present it now? We implement the
1341      * latter here.
1342      */
1343     regs[TM_IPB] |= xive_priority_to_ipb(priority);
1344     if (xive_ipb_to_pipr(regs[TM_IPB]) >= regs[TM_PIPR]) {
1345         return;
1346     }
1347     if (xive_nsr_indicates_group_exception(ring, regs[TM_NSR])) {
1348         xive2_redistribute(xrtr, tctx, ring);
1349     }
1350 
1351     xive_tctx_pipr_present(tctx, ring, priority, 0);
1352 }
1353 
1354 static void xive2_tctx_set_target(XiveTCTX *tctx, uint8_t ring, uint8_t target)
1355 {
1356     uint8_t *regs = &tctx->regs[ring];
1357 
1358     regs[TM_T] = target;
1359 }
1360 
1361 void xive2_tm_set_hv_target(XivePresenter *xptr, XiveTCTX *tctx,
1362                             hwaddr offset, uint64_t value, unsigned size)
1363 {
1364     xive2_tctx_set_target(tctx, TM_QW3_HV_PHYS, value & 0xff);
1365 }
1366 
1367 /*
1368  * XIVE Router (aka. Virtualization Controller or IVRE)
1369  */
1370 
1371 int xive2_router_get_eas(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1372                          Xive2Eas *eas)
1373 {
1374     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1375 
1376     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1377 }
1378 
1379 static
1380 int xive2_router_get_pq(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1381                        uint8_t *pq)
1382 {
1383     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1384 
1385     return xrc->get_pq(xrtr, eas_blk, eas_idx, pq);
1386 }
1387 
1388 static
1389 int xive2_router_set_pq(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1390                        uint8_t *pq)
1391 {
1392     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1393 
1394     return xrc->set_pq(xrtr, eas_blk, eas_idx, pq);
1395 }
1396 
1397 int xive2_router_get_end(Xive2Router *xrtr, uint8_t end_blk, uint32_t end_idx,
1398                          Xive2End *end)
1399 {
1400    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1401 
1402    return xrc->get_end(xrtr, end_blk, end_idx, end);
1403 }
1404 
1405 int xive2_router_write_end(Xive2Router *xrtr, uint8_t end_blk, uint32_t end_idx,
1406                            Xive2End *end, uint8_t word_number)
1407 {
1408    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1409 
1410    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1411 }
1412 
1413 int xive2_router_get_nvp(Xive2Router *xrtr, uint8_t nvp_blk, uint32_t nvp_idx,
1414                          Xive2Nvp *nvp)
1415 {
1416    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1417 
1418    return xrc->get_nvp(xrtr, nvp_blk, nvp_idx, nvp);
1419 }
1420 
1421 int xive2_router_write_nvp(Xive2Router *xrtr, uint8_t nvp_blk, uint32_t nvp_idx,
1422                            Xive2Nvp *nvp, uint8_t word_number)
1423 {
1424    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1425 
1426    return xrc->write_nvp(xrtr, nvp_blk, nvp_idx, nvp, word_number);
1427 }
1428 
1429 int xive2_router_get_nvgc(Xive2Router *xrtr, bool crowd,
1430                           uint8_t nvgc_blk, uint32_t nvgc_idx,
1431                           Xive2Nvgc *nvgc)
1432 {
1433    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1434 
1435    return xrc->get_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
1436 }
1437 
1438 int xive2_router_write_nvgc(Xive2Router *xrtr, bool crowd,
1439                             uint8_t nvgc_blk, uint32_t nvgc_idx,
1440                             Xive2Nvgc *nvgc)
1441 {
1442    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1443 
1444    return xrc->write_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
1445 }
1446 
1447 static bool xive2_vp_match_mask(uint32_t cam1, uint32_t cam2,
1448                                 uint32_t vp_mask)
1449 {
1450     return (cam1 & vp_mask) == (cam2 & vp_mask);
1451 }
1452 
1453 static uint8_t xive2_get_vp_block_mask(uint32_t nvt_blk, bool crowd)
1454 {
1455     uint8_t block_mask = 0b1111;
1456 
1457     /* 3 supported crowd sizes: 2, 4, 16 */
1458     if (crowd) {
1459         uint32_t size = xive_get_vpgroup_size(nvt_blk);
1460 
1461         if (size != 2 && size != 4 && size != 16) {
1462             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid crowd size of %d",
1463                                            size);
1464             return block_mask;
1465         }
1466         block_mask &= ~(size - 1);
1467     }
1468     return block_mask;
1469 }
1470 
1471 static uint32_t xive2_get_vp_index_mask(uint32_t nvt_index, bool cam_ignore)
1472 {
1473     uint32_t index_mask = 0xFFFFFF; /* 24 bits */
1474 
1475     if (cam_ignore) {
1476         uint32_t size = xive_get_vpgroup_size(nvt_index);
1477 
1478         if (size < 2) {
1479             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid group size of %d",
1480                                            size);
1481             return index_mask;
1482         }
1483         index_mask &= ~(size - 1);
1484     }
1485     return index_mask;
1486 }
1487 
1488 /*
1489  * The thread context register words are in big-endian format.
1490  */
1491 int xive2_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1492                                uint8_t format,
1493                                uint8_t nvt_blk, uint32_t nvt_idx,
1494                                bool crowd, bool cam_ignore,
1495                                uint32_t logic_serv)
1496 {
1497     uint32_t cam =   xive2_nvp_cam_line(nvt_blk, nvt_idx);
1498     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1499     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1500     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1501     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1502 
1503     uint32_t index_mask, vp_mask;
1504     uint8_t block_mask;
1505 
1506     if (format == 0) {
1507         /*
1508          * i=0: Specific NVT notification
1509          * i=1: VP-group notification (bits ignored at the end of the
1510          *      NVT identifier)
1511          */
1512         block_mask = xive2_get_vp_block_mask(nvt_blk, crowd);
1513         index_mask = xive2_get_vp_index_mask(nvt_idx, cam_ignore);
1514         vp_mask = xive2_nvp_cam_line(block_mask, index_mask);
1515 
1516         /* For VP-group notifications, threads with LGS=0 are excluded */
1517 
1518         /* PHYS ring */
1519         if ((be32_to_cpu(qw3w2) & TM2_QW3W2_VT) &&
1520             !(cam_ignore && tctx->regs[TM_QW3_HV_PHYS + TM_LGS] == 0) &&
1521             xive2_vp_match_mask(cam,
1522                                 xive2_tctx_hw_cam_line(xptr, tctx),
1523                                 vp_mask)) {
1524             return TM_QW3_HV_PHYS;
1525         }
1526 
1527         /* HV POOL ring */
1528         if ((be32_to_cpu(qw2w2) & TM2_QW2W2_VP) &&
1529             !(cam_ignore && tctx->regs[TM_QW2_HV_POOL + TM_LGS] == 0) &&
1530             xive2_vp_match_mask(cam,
1531                                 xive_get_field32(TM2_QW2W2_POOL_CAM, qw2w2),
1532                                 vp_mask)) {
1533             return TM_QW2_HV_POOL;
1534         }
1535 
1536         /* OS ring */
1537         if ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
1538             !(cam_ignore && tctx->regs[TM_QW1_OS + TM_LGS] == 0) &&
1539             xive2_vp_match_mask(cam,
1540                                 xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2),
1541                                 vp_mask)) {
1542             return TM_QW1_OS;
1543         }
1544     } else {
1545         /* F=1 : User level Event-Based Branch (EBB) notification */
1546 
1547         /* FIXME: what if cam_ignore and LGS = 0 ? */
1548         /* USER ring */
1549         if  ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
1550              (cam == xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2)) &&
1551              (be32_to_cpu(qw0w2) & TM2_QW0W2_VU) &&
1552              (logic_serv == xive_get_field32(TM2_QW0W2_LOGIC_SERV, qw0w2))) {
1553             return TM_QW0_USER;
1554         }
1555     }
1556     return -1;
1557 }
1558 
1559 bool xive2_tm_irq_precluded(XiveTCTX *tctx, int ring, uint8_t priority)
1560 {
1561     uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
1562 
1563     /*
1564      * The xive2_presenter_tctx_match() above tells if there's a match
1565      * but for VP-group notification, we still need to look at the
1566      * priority to know if the thread can take the interrupt now or if
1567      * it is precluded.
1568      */
1569     if (priority < sig_regs[TM_PIPR]) {
1570         return false;
1571     }
1572     return true;
1573 }
1574 
1575 void xive2_tm_set_lsmfb(XiveTCTX *tctx, int ring, uint8_t priority)
1576 {
1577     uint8_t *regs = &tctx->regs[ring];
1578 
1579     /*
1580      * Called by the router during a VP-group notification when the
1581      * thread matches but can't take the interrupt because it's
1582      * already running at a more favored priority. It then stores the
1583      * new interrupt priority in the LSMFB field.
1584      */
1585     regs[TM_LSMFB] = priority;
1586 }
1587 
1588 static void xive2_router_realize(DeviceState *dev, Error **errp)
1589 {
1590     Xive2Router *xrtr = XIVE2_ROUTER(dev);
1591 
1592     assert(xrtr->xfb);
1593 }
1594 
1595 /*
1596  * Notification using the END ESe/ESn bit (Event State Buffer for
1597  * escalation and notification). Profide further coalescing in the
1598  * Router.
1599  */
1600 static bool xive2_router_end_es_notify(Xive2Router *xrtr, uint8_t end_blk,
1601                                        uint32_t end_idx, Xive2End *end,
1602                                        uint32_t end_esmask)
1603 {
1604     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1605     bool notify = xive_esb_trigger(&pq);
1606 
1607     if (pq != xive_get_field32(end_esmask, end->w1)) {
1608         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1609         xive2_router_write_end(xrtr, end_blk, end_idx, end, 1);
1610     }
1611 
1612     /* ESe/n[Q]=1 : end of notification */
1613     return notify;
1614 }
1615 
1616 /*
1617  * An END trigger can come from an event trigger (IPI or HW) or from
1618  * another chip. We don't model the PowerBus but the END trigger
1619  * message has the same parameters than in the function below.
1620  */
1621 static void xive2_router_end_notify(Xive2Router *xrtr, uint8_t end_blk,
1622                                     uint32_t end_idx, uint32_t end_data,
1623                                     bool redistribute)
1624 {
1625     Xive2End end;
1626     uint8_t priority;
1627     uint8_t format;
1628     XiveTCTXMatch match;
1629     bool crowd, cam_ignore;
1630     uint8_t nvx_blk;
1631     uint32_t nvx_idx;
1632 
1633     /* END cache lookup */
1634     if (xive2_router_get_end(xrtr, end_blk, end_idx, &end)) {
1635         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1636                       end_idx);
1637         return;
1638     }
1639 
1640     if (!xive2_end_is_valid(&end)) {
1641         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1642                       end_blk, end_idx);
1643         return;
1644     }
1645 
1646     if (xive2_end_is_crowd(&end) && !xive2_end_is_ignore(&end)) {
1647         qemu_log_mask(LOG_GUEST_ERROR,
1648                       "XIVE: invalid END, 'crowd' bit requires 'ignore' bit\n");
1649         return;
1650     }
1651 
1652     if (!redistribute && xive2_end_is_enqueue(&end)) {
1653         trace_xive_end_enqueue(end_blk, end_idx, end_data);
1654         xive2_end_enqueue(&end, end_data);
1655         /* Enqueuing event data modifies the EQ toggle and index */
1656         xive2_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1657     }
1658 
1659     /*
1660      * When the END is silent, we skip the notification part.
1661      */
1662     if (xive2_end_is_silent_escalation(&end)) {
1663         goto do_escalation;
1664     }
1665 
1666     /*
1667      * The W7 format depends on the F bit in W6. It defines the type
1668      * of the notification :
1669      *
1670      *   F=0 : single or multiple NVP notification
1671      *   F=1 : User level Event-Based Branch (EBB) notification, no
1672      *         priority
1673      */
1674     format = xive_get_field32(END2_W6_FORMAT_BIT, end.w6);
1675     priority = xive_get_field32(END2_W7_F0_PRIORITY, end.w7);
1676 
1677     /* The END is masked */
1678     if (format == 0 && priority == 0xff) {
1679         return;
1680     }
1681 
1682     /*
1683      * Check the END ESn (Event State Buffer for notification) for
1684      * even further coalescing in the Router
1685      */
1686     if (!xive2_end_is_notify(&end)) {
1687         /* ESn[Q]=1 : end of notification */
1688         if (!xive2_router_end_es_notify(xrtr, end_blk, end_idx,
1689                                        &end, END2_W1_ESn)) {
1690             return;
1691         }
1692     }
1693 
1694     /*
1695      * Follows IVPE notification
1696      */
1697     nvx_blk = xive_get_field32(END2_W6_VP_BLOCK, end.w6);
1698     nvx_idx = xive_get_field32(END2_W6_VP_OFFSET, end.w6);
1699     crowd = xive2_end_is_crowd(&end);
1700     cam_ignore = xive2_end_is_ignore(&end);
1701 
1702     /* TODO: Auto EOI. */
1703     if (xive_presenter_match(xrtr->xfb, format, nvx_blk, nvx_idx,
1704                              crowd, cam_ignore, priority,
1705                              xive_get_field32(END2_W7_F1_LOG_SERVER_ID, end.w7),
1706                              &match)) {
1707         XiveTCTX *tctx = match.tctx;
1708         uint8_t ring = match.ring;
1709         uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
1710         uint8_t nsr = sig_regs[TM_NSR];
1711         uint8_t group_level;
1712 
1713         if (priority < sig_regs[TM_PIPR] &&
1714             xive_nsr_indicates_group_exception(ring, nsr)) {
1715             xive2_redistribute(xrtr, tctx, xive_nsr_exception_ring(ring, nsr));
1716         }
1717 
1718         group_level = xive_get_group_level(crowd, cam_ignore, nvx_blk, nvx_idx);
1719         trace_xive_presenter_notify(nvx_blk, nvx_idx, ring, group_level);
1720         xive_tctx_pipr_present(tctx, ring, priority, group_level);
1721         return;
1722     }
1723 
1724     /*
1725      * If no matching NVP is dispatched on a HW thread :
1726      * - specific VP: update the NVP structure if backlog is activated
1727      * - VP-group: update the backlog counter for that priority in the NVG
1728      */
1729     if (xive2_end_is_backlog(&end)) {
1730 
1731         if (format == 1) {
1732             qemu_log_mask(LOG_GUEST_ERROR,
1733                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1734                           end_blk, end_idx);
1735             return;
1736         }
1737 
1738         if (!cam_ignore) {
1739             uint8_t ipb;
1740             Xive2Nvp nvp;
1741 
1742             /* NVP cache lookup */
1743             if (xive2_router_get_nvp(xrtr, nvx_blk, nvx_idx, &nvp)) {
1744                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVP %x/%x\n",
1745                               nvx_blk, nvx_idx);
1746                 return;
1747             }
1748 
1749             if (!xive2_nvp_is_valid(&nvp)) {
1750                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is invalid\n",
1751                               nvx_blk, nvx_idx);
1752                 return;
1753             }
1754 
1755             /*
1756              * Record the IPB in the associated NVP structure for later
1757              * use. The presenter will resend the interrupt when the vCPU
1758              * is dispatched again on a HW thread.
1759              */
1760             ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2) |
1761                 xive_priority_to_ipb(priority);
1762             nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, ipb);
1763             xive2_router_write_nvp(xrtr, nvx_blk, nvx_idx, &nvp, 2);
1764         } else {
1765             Xive2Nvgc nvgc;
1766             uint32_t backlog;
1767 
1768             /*
1769              * For groups and crowds, the per-priority backlog
1770              * counters are stored in the NVG/NVC structures
1771              */
1772             if (xive2_router_get_nvgc(xrtr, crowd,
1773                                       nvx_blk, nvx_idx, &nvgc)) {
1774                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no %s %x/%x\n",
1775                               crowd ? "NVC" : "NVG", nvx_blk, nvx_idx);
1776                 return;
1777             }
1778 
1779             if (!xive2_nvgc_is_valid(&nvgc)) {
1780                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVG %x/%x is invalid\n",
1781                               nvx_blk, nvx_idx);
1782                 return;
1783             }
1784 
1785             /*
1786              * Increment the backlog counter for that priority.
1787              * We only call broadcast the first time the counter is
1788              * incremented. broadcast will set the LSMFB field of the TIMA of
1789              * relevant threads so that they know an interrupt is pending.
1790              */
1791             backlog = xive2_nvgc_get_backlog(&nvgc, priority) + 1;
1792             xive2_nvgc_set_backlog(&nvgc, priority, backlog);
1793             xive2_router_write_nvgc(xrtr, crowd, nvx_blk, nvx_idx, &nvgc);
1794 
1795             if (backlog == 1) {
1796                 XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xrtr->xfb);
1797                 xfc->broadcast(xrtr->xfb, nvx_blk, nvx_idx,
1798                                crowd, cam_ignore, priority);
1799 
1800                 if (!xive2_end_is_precluded_escalation(&end)) {
1801                     /*
1802                      * The interrupt will be picked up when the
1803                      * matching thread lowers its priority level
1804                      */
1805                     return;
1806                 }
1807             }
1808         }
1809     }
1810 
1811 do_escalation:
1812     /*
1813      * If activated, escalate notification using the ESe PQ bits and
1814      * the EAS in w4-5
1815      */
1816     if (!xive2_end_is_escalate(&end)) {
1817         return;
1818     }
1819 
1820     /*
1821      * Check the END ESe (Event State Buffer for escalation) for even
1822      * further coalescing in the Router
1823      */
1824     if (!xive2_end_is_uncond_escalation(&end)) {
1825         /* ESe[Q]=1 : end of escalation notification */
1826         if (!xive2_router_end_es_notify(xrtr, end_blk, end_idx,
1827                                        &end, END2_W1_ESe)) {
1828             return;
1829         }
1830     }
1831 
1832     if (xive2_end_is_escalate_end(&end)) {
1833         /*
1834          * Perform END Adaptive escalation processing
1835          * The END trigger becomes an Escalation trigger
1836          */
1837         uint8_t esc_blk = xive_get_field32(END2_W4_END_BLOCK, end.w4);
1838         uint32_t esc_idx = xive_get_field32(END2_W4_ESC_END_INDEX, end.w4);
1839         uint32_t esc_data = xive_get_field32(END2_W5_ESC_END_DATA, end.w5);
1840         trace_xive_escalate_end(end_blk, end_idx, esc_blk, esc_idx, esc_data);
1841         xive2_router_end_notify(xrtr, esc_blk, esc_idx, esc_data, false);
1842     } /* end END adaptive escalation */
1843 
1844     else {
1845         uint32_t lisn;              /* Logical Interrupt Source Number */
1846 
1847         /*
1848          *  Perform ESB escalation processing
1849          *      E[N] == 1 --> N
1850          *      Req[Block] <- E[ESB_Block]
1851          *      Req[Index] <- E[ESB_Index]
1852          *      Req[Offset] <- 0x000
1853          *      Execute <ESB Store> Req command
1854          */
1855         lisn = XIVE_EAS(xive_get_field32(END2_W4_END_BLOCK,     end.w4),
1856                         xive_get_field32(END2_W4_ESC_END_INDEX, end.w4));
1857 
1858         trace_xive_escalate_esb(end_blk, end_idx, lisn);
1859         xive2_notify(xrtr, lisn, true /* pq_checked */);
1860     }
1861 
1862     return;
1863 }
1864 
1865 void xive2_notify(Xive2Router *xrtr , uint32_t lisn, bool pq_checked)
1866 {
1867     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1868     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1869     Xive2Eas eas;
1870 
1871     /* EAS cache lookup */
1872     if (xive2_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1873         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1874         return;
1875     }
1876 
1877     if (!pq_checked) {
1878         bool notify;
1879         uint8_t pq;
1880 
1881         /* PQ cache lookup */
1882         if (xive2_router_get_pq(xrtr, eas_blk, eas_idx, &pq)) {
1883             /* Set FIR */
1884             g_assert_not_reached();
1885         }
1886 
1887         notify = xive_esb_trigger(&pq);
1888 
1889         if (xive2_router_set_pq(xrtr, eas_blk, eas_idx, &pq)) {
1890             /* Set FIR */
1891             g_assert_not_reached();
1892         }
1893 
1894         if (!notify) {
1895             return;
1896         }
1897     }
1898 
1899     if (!xive2_eas_is_valid(&eas)) {
1900         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN %x\n", lisn);
1901         return;
1902     }
1903 
1904     if (xive2_eas_is_masked(&eas)) {
1905         /* Notification completed */
1906         return;
1907     }
1908 
1909     /* TODO: add support for EAS resume */
1910     if (xive2_eas_is_resume(&eas)) {
1911         qemu_log_mask(LOG_UNIMP,
1912                       "XIVE: EAS resume processing unimplemented - LISN %x\n",
1913                       lisn);
1914         return;
1915     }
1916 
1917     /*
1918      * The event trigger becomes an END trigger
1919      */
1920     xive2_router_end_notify(xrtr,
1921                             xive_get_field64(EAS2_END_BLOCK, eas.w),
1922                             xive_get_field64(EAS2_END_INDEX, eas.w),
1923                             xive_get_field64(EAS2_END_DATA,  eas.w),
1924                             false);
1925     return;
1926 }
1927 
1928 void xive2_router_notify(XiveNotifier *xn, uint32_t lisn, bool pq_checked)
1929 {
1930     Xive2Router *xrtr = XIVE2_ROUTER(xn);
1931 
1932     xive2_notify(xrtr, lisn, pq_checked);
1933     return;
1934 }
1935 
1936 static const Property xive2_router_properties[] = {
1937     DEFINE_PROP_LINK("xive-fabric", Xive2Router, xfb,
1938                      TYPE_XIVE_FABRIC, XiveFabric *),
1939 };
1940 
1941 static void xive2_router_class_init(ObjectClass *klass, const void *data)
1942 {
1943     DeviceClass *dc = DEVICE_CLASS(klass);
1944     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1945 
1946     dc->desc    = "XIVE2 Router Engine";
1947     device_class_set_props(dc, xive2_router_properties);
1948     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1949     dc->realize = xive2_router_realize;
1950     xnc->notify = xive2_router_notify;
1951 }
1952 
1953 static const TypeInfo xive2_router_info = {
1954     .name          = TYPE_XIVE2_ROUTER,
1955     .parent        = TYPE_SYS_BUS_DEVICE,
1956     .abstract      = true,
1957     .instance_size = sizeof(Xive2Router),
1958     .class_size    = sizeof(Xive2RouterClass),
1959     .class_init    = xive2_router_class_init,
1960     .interfaces    = (const InterfaceInfo[]) {
1961         { TYPE_XIVE_NOTIFIER },
1962         { TYPE_XIVE_PRESENTER },
1963         { }
1964     }
1965 };
1966 
1967 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
1968 {
1969     return !((addr >> shift) & 1);
1970 }
1971 
1972 static uint64_t xive2_end_source_read(void *opaque, hwaddr addr, unsigned size)
1973 {
1974     Xive2EndSource *xsrc = XIVE2_END_SOURCE(opaque);
1975     uint32_t offset = addr & 0xFFF;
1976     uint8_t end_blk;
1977     uint32_t end_idx;
1978     Xive2End end;
1979     uint32_t end_esmask;
1980     uint8_t pq;
1981     uint64_t ret;
1982 
1983     /*
1984      * The block id should be deduced from the load address on the END
1985      * ESB MMIO but our model only supports a single block per XIVE chip.
1986      */
1987     end_blk = xive2_router_get_block_id(xsrc->xrtr);
1988     end_idx = addr >> (xsrc->esb_shift + 1);
1989 
1990     if (xive2_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1991         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1992                       end_idx);
1993         return -1;
1994     }
1995 
1996     if (!xive2_end_is_valid(&end)) {
1997         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1998                       end_blk, end_idx);
1999         return -1;
2000     }
2001 
2002     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END2_W1_ESn :
2003         END2_W1_ESe;
2004     pq = xive_get_field32(end_esmask, end.w1);
2005 
2006     switch (offset) {
2007     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
2008         ret = xive_esb_eoi(&pq);
2009 
2010         /* Forward the source event notification for routing ?? */
2011         break;
2012 
2013     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
2014         ret = pq;
2015         break;
2016 
2017     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
2018     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
2019     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
2020     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
2021         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
2022         break;
2023     default:
2024         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
2025                       offset);
2026         return -1;
2027     }
2028 
2029     if (pq != xive_get_field32(end_esmask, end.w1)) {
2030         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
2031         xive2_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
2032     }
2033 
2034     return ret;
2035 }
2036 
2037 static void xive2_end_source_write(void *opaque, hwaddr addr,
2038                                    uint64_t value, unsigned size)
2039 {
2040     Xive2EndSource *xsrc = XIVE2_END_SOURCE(opaque);
2041     uint32_t offset = addr & 0xFFF;
2042     uint8_t end_blk;
2043     uint32_t end_idx;
2044     Xive2End end;
2045     uint32_t end_esmask;
2046     uint8_t pq;
2047     bool notify = false;
2048 
2049     /*
2050      * The block id should be deduced from the load address on the END
2051      * ESB MMIO but our model only supports a single block per XIVE chip.
2052      */
2053     end_blk = xive2_router_get_block_id(xsrc->xrtr);
2054     end_idx = addr >> (xsrc->esb_shift + 1);
2055 
2056     if (xive2_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
2057         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
2058                       end_idx);
2059         return;
2060     }
2061 
2062     if (!xive2_end_is_valid(&end)) {
2063         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
2064                       end_blk, end_idx);
2065         return;
2066     }
2067 
2068     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END2_W1_ESn :
2069         END2_W1_ESe;
2070     pq = xive_get_field32(end_esmask, end.w1);
2071 
2072     switch (offset) {
2073     case 0 ... 0x3FF:
2074         notify = xive_esb_trigger(&pq);
2075         break;
2076 
2077     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
2078         /* TODO: can we check StoreEOI availability from the router ? */
2079         notify = xive_esb_eoi(&pq);
2080         break;
2081 
2082     case XIVE_ESB_INJECT ... XIVE_ESB_INJECT + 0x3FF:
2083         if (end_esmask == END2_W1_ESe) {
2084             qemu_log_mask(LOG_GUEST_ERROR,
2085                           "XIVE: END %x/%x can not EQ inject on ESe\n",
2086                            end_blk, end_idx);
2087             return;
2088         }
2089         notify = true;
2090         break;
2091 
2092     default:
2093         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB write addr %d\n",
2094                       offset);
2095         return;
2096     }
2097 
2098     if (pq != xive_get_field32(end_esmask, end.w1)) {
2099         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
2100         xive2_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
2101     }
2102 
2103     /* TODO: Forward the source event notification for routing */
2104     if (notify) {
2105         ;
2106     }
2107 }
2108 
2109 static const MemoryRegionOps xive2_end_source_ops = {
2110     .read = xive2_end_source_read,
2111     .write = xive2_end_source_write,
2112     .endianness = DEVICE_BIG_ENDIAN,
2113     .valid = {
2114         .min_access_size = 1,
2115         .max_access_size = 8,
2116     },
2117     .impl = {
2118         .min_access_size = 1,
2119         .max_access_size = 8,
2120     },
2121 };
2122 
2123 static void xive2_end_source_realize(DeviceState *dev, Error **errp)
2124 {
2125     Xive2EndSource *xsrc = XIVE2_END_SOURCE(dev);
2126 
2127     assert(xsrc->xrtr);
2128 
2129     if (!xsrc->nr_ends) {
2130         error_setg(errp, "Number of interrupt needs to be greater than 0");
2131         return;
2132     }
2133 
2134     if (xsrc->esb_shift != XIVE_ESB_4K &&
2135         xsrc->esb_shift != XIVE_ESB_64K) {
2136         error_setg(errp, "Invalid ESB shift setting");
2137         return;
2138     }
2139 
2140     /*
2141      * Each END is assigned an even/odd pair of MMIO pages, the even page
2142      * manages the ESn field while the odd page manages the ESe field.
2143      */
2144     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
2145                           &xive2_end_source_ops, xsrc, "xive.end",
2146                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
2147 }
2148 
2149 static const Property xive2_end_source_properties[] = {
2150     DEFINE_PROP_UINT32("nr-ends", Xive2EndSource, nr_ends, 0),
2151     DEFINE_PROP_UINT32("shift", Xive2EndSource, esb_shift, XIVE_ESB_64K),
2152     DEFINE_PROP_LINK("xive", Xive2EndSource, xrtr, TYPE_XIVE2_ROUTER,
2153                      Xive2Router *),
2154 };
2155 
2156 static void xive2_end_source_class_init(ObjectClass *klass, const void *data)
2157 {
2158     DeviceClass *dc = DEVICE_CLASS(klass);
2159 
2160     dc->desc    = "XIVE END Source";
2161     device_class_set_props(dc, xive2_end_source_properties);
2162     dc->realize = xive2_end_source_realize;
2163     dc->user_creatable = false;
2164 }
2165 
2166 static const TypeInfo xive2_end_source_info = {
2167     .name          = TYPE_XIVE2_END_SOURCE,
2168     .parent        = TYPE_DEVICE,
2169     .instance_size = sizeof(Xive2EndSource),
2170     .class_init    = xive2_end_source_class_init,
2171 };
2172 
2173 static void xive2_register_types(void)
2174 {
2175     type_register_static(&xive2_router_info);
2176     type_register_static(&xive2_end_source_info);
2177 }
2178 
2179 type_init(xive2_register_types)
2180