xref: /openbmc/qemu/hw/intc/xive2.c (revision 365e322cfb86b2e7131c3290c3a61f8d2bb224d3)
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 static void xive2_cam_decode(uint32_t cam, uint8_t *nvp_blk,
587                              uint32_t *nvp_idx, bool *valid, bool *hw)
588 {
589     *nvp_blk = xive2_nvp_blk(cam);
590     *nvp_idx = xive2_nvp_idx(cam);
591     *valid = !!(cam & TM2_W2_VALID);
592     *hw = !!(cam & TM2_W2_HW);
593 }
594 
595 /*
596  * Encode the HW CAM line with 7bit or 8bit thread id. The thread id
597  * width and block id width is configurable at the IC level.
598  *
599  *    chipid << 24 | 0000 0000 0000 0000 1 threadid (7Bit)
600  *    chipid << 24 | 0000 0000 0000 0001 threadid   (8Bit)
601  */
602 static uint32_t xive2_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
603 {
604     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
605     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
606     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
607     uint8_t blk = xive2_router_get_block_id(xrtr);
608     uint8_t tid_shift =
609         xive2_router_get_config(xrtr) & XIVE2_THREADID_8BITS ? 8 : 7;
610     uint8_t tid_mask = (1 << tid_shift) - 1;
611 
612     return xive2_nvp_cam_line(blk, 1 << tid_shift | (pir & tid_mask));
613 }
614 
615 static void xive2_redistribute(Xive2Router *xrtr, XiveTCTX *tctx, uint8_t ring)
616 {
617     uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
618     uint8_t nsr = sig_regs[TM_NSR];
619     uint8_t pipr = sig_regs[TM_PIPR];
620     uint8_t crowd = NVx_CROWD_LVL(nsr);
621     uint8_t group = NVx_GROUP_LVL(nsr);
622     uint8_t nvgc_blk, end_blk, nvp_blk;
623     uint32_t nvgc_idx, end_idx, nvp_idx;
624     Xive2Nvgc nvgc;
625     uint8_t prio_limit;
626     uint32_t cfg;
627 
628     /* redistribution is only for group/crowd interrupts */
629     if (!xive_nsr_indicates_group_exception(ring, nsr)) {
630         return;
631     }
632 
633     /* Don't check return code since ring is expected to be invalidated */
634     xive2_tctx_get_nvp_indexes(tctx, ring, &nvp_blk, &nvp_idx);
635 
636     trace_xive_redistribute(tctx->cs->cpu_index, ring, nvp_blk, nvp_idx);
637 
638     trace_xive_redistribute(tctx->cs->cpu_index, ring, nvp_blk, nvp_idx);
639     /* convert crowd/group to blk/idx */
640     if (group > 0) {
641         nvgc_idx = (nvp_idx & (0xffffffff << group)) |
642                    ((1 << (group - 1)) - 1);
643     } else {
644         nvgc_idx = nvp_idx;
645     }
646 
647     if (crowd > 0) {
648         crowd = (crowd == 3) ? 4 : crowd;
649         nvgc_blk = (nvp_blk & (0xffffffff << crowd)) |
650                    ((1 << (crowd - 1)) - 1);
651     } else {
652         nvgc_blk = nvp_blk;
653     }
654 
655     /* Use blk/idx to retrieve the NVGC */
656     if (xive2_router_get_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, &nvgc)) {
657         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no %s %x/%x\n",
658                       crowd ? "NVC" : "NVG", nvgc_blk, nvgc_idx);
659         return;
660     }
661 
662     /* retrieve the END blk/idx from the NVGC */
663     end_blk = xive_get_field32(NVGC2_W1_END_BLK, nvgc.w1);
664     end_idx = xive_get_field32(NVGC2_W1_END_IDX, nvgc.w1);
665 
666     /* determine number of priorities being used */
667     cfg = xive2_router_get_config(xrtr);
668     if (cfg & XIVE2_EN_VP_GRP_PRIORITY) {
669         prio_limit = 1 << GETFIELD(NVGC2_W1_PSIZE, nvgc.w1);
670     } else {
671         prio_limit = 1 << GETFIELD(XIVE2_VP_INT_PRIO, cfg);
672     }
673 
674     /* add priority offset to end index */
675     end_idx += pipr % prio_limit;
676 
677     /* trigger the group END */
678     xive2_router_end_notify(xrtr, end_blk, end_idx, 0, true);
679 
680     /* clear interrupt indication for the context */
681     sig_regs[TM_NSR] = 0;
682     sig_regs[TM_PIPR] = sig_regs[TM_CPPR];
683     xive_tctx_reset_signal(tctx, ring);
684 }
685 
686 static uint64_t xive2_tm_pull_ctx(XivePresenter *xptr, XiveTCTX *tctx,
687                                   hwaddr offset, unsigned size, uint8_t ring)
688 {
689     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
690     uint32_t target_ringw2 = xive_tctx_word2(&tctx->regs[ring]);
691     uint32_t cam = be32_to_cpu(target_ringw2);
692     uint8_t nvp_blk;
693     uint32_t nvp_idx;
694     uint8_t cur_ring;
695     bool valid;
696     bool do_save;
697     uint8_t nsr;
698 
699     xive2_cam_decode(cam, &nvp_blk, &nvp_idx, &valid, &do_save);
700 
701     if (xive2_tctx_get_nvp_indexes(tctx, ring, &nvp_blk, &nvp_idx)) {
702         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: pulling invalid NVP %x/%x !?\n",
703                       nvp_blk, nvp_idx);
704     }
705 
706     /* Invalidate CAM line of requested ring and all lower rings */
707     for (cur_ring = TM_QW0_USER; cur_ring <= ring;
708          cur_ring += XIVE_TM_RING_SIZE) {
709         uint32_t ringw2 = xive_tctx_word2(&tctx->regs[cur_ring]);
710         uint32_t ringw2_new = xive_set_field32(TM2_QW1W2_VO, ringw2, 0);
711         bool is_valid = !!(xive_get_field32(TM2_QW1W2_VO, ringw2));
712         uint8_t *sig_regs;
713 
714         memcpy(&tctx->regs[cur_ring + TM_WORD2], &ringw2_new, 4);
715 
716         /* Skip the rest for USER or invalid contexts */
717         if ((cur_ring == TM_QW0_USER) || !is_valid) {
718             continue;
719         }
720 
721         /* Active group/crowd interrupts need to be redistributed */
722         sig_regs = xive_tctx_signal_regs(tctx, ring);
723         nsr = sig_regs[TM_NSR];
724         if (xive_nsr_indicates_group_exception(cur_ring, nsr)) {
725             /* Ensure ring matches NSR (for HV NSR POOL vs PHYS rings) */
726             if (cur_ring == xive_nsr_exception_ring(cur_ring, nsr)) {
727                 xive2_redistribute(xrtr, tctx, cur_ring);
728             }
729         }
730     }
731 
732     if (xive2_router_get_config(xrtr) & XIVE2_VP_SAVE_RESTORE && do_save) {
733         xive2_tctx_save_ctx(xrtr, tctx, ring, nvp_blk, nvp_idx);
734     }
735 
736     /*
737      * Lower external interrupt line of requested ring and below except for
738      * USER, which doesn't exist.
739      */
740     for (cur_ring = TM_QW1_OS; cur_ring <= ring;
741          cur_ring += XIVE_TM_RING_SIZE) {
742         xive_tctx_reset_signal(tctx, cur_ring);
743     }
744     return target_ringw2;
745 }
746 
747 uint64_t xive2_tm_pull_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
748                               hwaddr offset, unsigned size)
749 {
750     return xive2_tm_pull_ctx(xptr, tctx, offset, size, TM_QW1_OS);
751 }
752 
753 uint64_t xive2_tm_pull_pool_ctx(XivePresenter *xptr, XiveTCTX *tctx,
754                                 hwaddr offset, unsigned size)
755 {
756     return xive2_tm_pull_ctx(xptr, tctx, offset, size, TM_QW2_HV_POOL);
757 }
758 
759 uint64_t xive2_tm_pull_phys_ctx(XivePresenter *xptr, XiveTCTX *tctx,
760                                 hwaddr offset, unsigned size)
761 {
762     return xive2_tm_pull_ctx(xptr, tctx, offset, size, TM_QW3_HV_PHYS);
763 }
764 
765 #define REPORT_LINE_GEN1_SIZE       16
766 
767 static void xive2_tm_report_line_gen1(XiveTCTX *tctx, uint8_t *data,
768                                       uint8_t size)
769 {
770     uint8_t *regs = tctx->regs;
771 
772     g_assert(size == REPORT_LINE_GEN1_SIZE);
773     memset(data, 0, size);
774     /*
775      * See xive architecture for description of what is saved. It is
776      * hand-picked information to fit in 16 bytes.
777      */
778     data[0x0] = regs[TM_QW3_HV_PHYS + TM_NSR];
779     data[0x1] = regs[TM_QW3_HV_PHYS + TM_CPPR];
780     data[0x2] = regs[TM_QW3_HV_PHYS + TM_IPB];
781     data[0x3] = regs[TM_QW2_HV_POOL + TM_IPB];
782     data[0x4] = regs[TM_QW1_OS + TM_ACK_CNT];
783     data[0x5] = regs[TM_QW3_HV_PHYS + TM_LGS];
784     data[0x6] = 0xFF;
785     data[0x7] = regs[TM_QW3_HV_PHYS + TM_WORD2] & 0x80;
786     data[0x7] |= (regs[TM_QW2_HV_POOL + TM_WORD2] & 0x80) >> 1;
787     data[0x7] |= (regs[TM_QW1_OS + TM_WORD2] & 0x80) >> 2;
788     data[0x7] |= (regs[TM_QW3_HV_PHYS + TM_WORD2] & 0x3);
789     data[0x8] = regs[TM_QW1_OS + TM_NSR];
790     data[0x9] = regs[TM_QW1_OS + TM_CPPR];
791     data[0xA] = regs[TM_QW1_OS + TM_IPB];
792     data[0xB] = regs[TM_QW1_OS + TM_LGS];
793     if (regs[TM_QW0_USER + TM_WORD2] & 0x80) {
794         /*
795          * Logical server extension, except VU bit replaced by EB bit
796          * from NSR
797          */
798         data[0xC] = regs[TM_QW0_USER + TM_WORD2];
799         data[0xC] &= ~0x80;
800         data[0xC] |= regs[TM_QW0_USER + TM_NSR] & 0x80;
801         data[0xD] = regs[TM_QW0_USER + TM_WORD2 + 1];
802         data[0xE] = regs[TM_QW0_USER + TM_WORD2 + 2];
803         data[0xF] = regs[TM_QW0_USER + TM_WORD2 + 3];
804     }
805 }
806 
807 static void xive2_tm_pull_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
808                                  hwaddr offset, uint64_t value,
809                                  unsigned size, uint8_t ring)
810 {
811     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
812     uint32_t hw_cam, nvp_idx, xive2_cfg, reserved;
813     uint8_t nvp_blk;
814     Xive2Nvp nvp;
815     uint64_t phys_addr;
816     MemTxResult result;
817 
818     hw_cam = xive2_tctx_hw_cam_line(xptr, tctx);
819     nvp_blk = xive2_nvp_blk(hw_cam);
820     nvp_idx = xive2_nvp_idx(hw_cam);
821 
822     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
823         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
824                       nvp_blk, nvp_idx);
825         return;
826     }
827 
828     if (!xive2_nvp_is_valid(&nvp)) {
829         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
830                       nvp_blk, nvp_idx);
831         return;
832     }
833 
834     xive2_cfg = xive2_router_get_config(xrtr);
835 
836     phys_addr = xive2_nvp_reporting_addr(&nvp) + 0x80; /* odd line */
837     if (xive2_cfg & XIVE2_GEN1_TIMA_OS) {
838         uint8_t pull_ctxt[REPORT_LINE_GEN1_SIZE];
839 
840         xive2_tm_report_line_gen1(tctx, pull_ctxt, REPORT_LINE_GEN1_SIZE);
841         result = dma_memory_write(&address_space_memory, phys_addr,
842                                   pull_ctxt, REPORT_LINE_GEN1_SIZE,
843                                   MEMTXATTRS_UNSPECIFIED);
844         assert(result == MEMTX_OK);
845     } else {
846         result = dma_memory_write(&address_space_memory, phys_addr,
847                                   &tctx->regs, sizeof(tctx->regs),
848                                   MEMTXATTRS_UNSPECIFIED);
849         assert(result == MEMTX_OK);
850         reserved = 0xFFFFFFFF;
851         result = dma_memory_write(&address_space_memory, phys_addr + 12,
852                                   &reserved, sizeof(reserved),
853                                   MEMTXATTRS_UNSPECIFIED);
854         assert(result == MEMTX_OK);
855     }
856 
857     /* the rest is similar to pull context to registers */
858     xive2_tm_pull_ctx(xptr, tctx, offset, size, ring);
859 }
860 
861 void xive2_tm_pull_os_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
862                              hwaddr offset, uint64_t value, unsigned size)
863 {
864     xive2_tm_pull_ctx_ol(xptr, tctx, offset, value, size, TM_QW1_OS);
865 }
866 
867 
868 void xive2_tm_pull_phys_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
869                                hwaddr offset, uint64_t value, unsigned size)
870 {
871     xive2_tm_pull_ctx_ol(xptr, tctx, offset, value, size, TM_QW3_HV_PHYS);
872 }
873 
874 static uint8_t xive2_tctx_restore_ctx(Xive2Router *xrtr, XiveTCTX *tctx,
875                                       uint8_t ring,
876                                       uint8_t nvp_blk, uint32_t nvp_idx,
877                                       Xive2Nvp *nvp)
878 {
879     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
880     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
881     uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
882     uint8_t *regs = &tctx->regs[ring];
883     uint8_t cppr;
884 
885     if (!xive2_nvp_is_hw(nvp)) {
886         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is not HW owned\n",
887                       nvp_blk, nvp_idx);
888         return 0;
889     }
890 
891     cppr = xive_get_field32(NVP2_W2_CPPR, nvp->w2);
892     nvp->w2 = xive_set_field32(NVP2_W2_CPPR, nvp->w2, 0);
893     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, nvp, 2);
894 
895     sig_regs[TM_CPPR] = cppr;
896     regs[TM_LSMFB] = xive_get_field32(NVP2_W2_LSMFB, nvp->w2);
897     regs[TM_LGS] = xive_get_field32(NVP2_W2_LGS, nvp->w2);
898     regs[TM_T] = xive_get_field32(NVP2_W2_T, nvp->w2);
899 
900     nvp->w1 = xive_set_field32(NVP2_W1_CO, nvp->w1, 1);
901     nvp->w1 = xive_set_field32(NVP2_W1_CO_THRID_VALID, nvp->w1, 1);
902     nvp->w1 = xive_set_field32(NVP2_W1_CO_THRID, nvp->w1, pir);
903 
904     /*
905      * Checkout privilege: 0:OS, 1:Pool, 2:Hard
906      *
907      * TODO: we don't support hard push/pull
908      */
909     switch (ring) {
910     case TM_QW1_OS:
911         nvp->w1 = xive_set_field32(NVP2_W1_CO_PRIV, nvp->w1, 0);
912         break;
913     case TM_QW2_HV_POOL:
914         nvp->w1 = xive_set_field32(NVP2_W1_CO_PRIV, nvp->w1, 1);
915         break;
916     default:
917         g_assert_not_reached();
918     }
919 
920     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, nvp, 1);
921 
922     /* return restored CPPR to generate a CPU exception if needed */
923     return cppr;
924 }
925 
926 static void xive2_tctx_process_pending(XiveTCTX *tctx, uint8_t sig_ring);
927 
928 static void xive2_tctx_need_resend(Xive2Router *xrtr, XiveTCTX *tctx,
929                                    uint8_t nvp_blk, uint32_t nvp_idx,
930                                    bool do_restore)
931 {
932     uint8_t *regs = &tctx->regs[TM_QW1_OS];
933     uint8_t ipb;
934     Xive2Nvp nvp;
935 
936     /*
937      * Grab the associated thread interrupt context registers in the
938      * associated NVP
939      */
940     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
941         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
942                       nvp_blk, nvp_idx);
943         return;
944     }
945 
946     if (!xive2_nvp_is_valid(&nvp)) {
947         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
948                       nvp_blk, nvp_idx);
949         return;
950     }
951 
952     /* Automatically restore thread context registers */
953     if (xive2_router_get_config(xrtr) & XIVE2_VP_SAVE_RESTORE && do_restore) {
954         xive2_tctx_restore_ctx(xrtr, tctx, TM_QW1_OS, nvp_blk, nvp_idx, &nvp);
955     }
956 
957     ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2);
958     if (ipb) {
959         nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, 0);
960         xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, &nvp, 2);
961     }
962     /* IPB bits in the backlog are merged with the TIMA IPB bits */
963     regs[TM_IPB] |= ipb;
964 
965     xive2_tctx_process_pending(tctx, TM_QW1_OS);
966 }
967 
968 /*
969  * Updating the OS CAM line can trigger a resend of interrupt
970  */
971 void xive2_tm_push_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
972                           hwaddr offset, uint64_t value, unsigned size)
973 {
974     uint32_t cam;
975     uint32_t qw1w2;
976     uint64_t qw1dw1;
977     uint8_t nvp_blk;
978     uint32_t nvp_idx;
979     bool vo;
980     bool do_restore;
981 
982     /* First update the thead context */
983     switch (size) {
984     case 4:
985         cam = value;
986         qw1w2 = cpu_to_be32(cam);
987         memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
988         break;
989     case 8:
990         cam = value >> 32;
991         qw1dw1 = cpu_to_be64(value);
992         memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1dw1, 8);
993         break;
994     default:
995         g_assert_not_reached();
996     }
997 
998     xive2_cam_decode(cam, &nvp_blk, &nvp_idx, &vo, &do_restore);
999 
1000     /* Check the interrupt pending bits */
1001     if (vo) {
1002         xive2_tctx_need_resend(XIVE2_ROUTER(xptr), tctx, nvp_blk, nvp_idx,
1003                                do_restore);
1004     }
1005 }
1006 
1007 /* returns -1 if ring is invalid, but still populates block and index */
1008 static int xive2_tctx_get_nvp_indexes(XiveTCTX *tctx, uint8_t ring,
1009                                       uint8_t *nvp_blk, uint32_t *nvp_idx)
1010 {
1011     uint32_t w2;
1012     uint32_t cam = 0;
1013     int rc = 0;
1014 
1015     w2 = xive_tctx_word2(&tctx->regs[ring]);
1016     switch (ring) {
1017     case TM_QW1_OS:
1018         if (!(be32_to_cpu(w2) & TM2_QW1W2_VO)) {
1019             rc = -1;
1020         }
1021         cam = xive_get_field32(TM2_QW1W2_OS_CAM, w2);
1022         break;
1023     case TM_QW2_HV_POOL:
1024         if (!(be32_to_cpu(w2) & TM2_QW2W2_VP)) {
1025             rc = -1;
1026         }
1027         cam = xive_get_field32(TM2_QW2W2_POOL_CAM, w2);
1028         break;
1029     case TM_QW3_HV_PHYS:
1030         if (!(be32_to_cpu(w2) & TM2_QW3W2_VT)) {
1031             rc = -1;
1032         }
1033         cam = xive2_tctx_hw_cam_line(tctx->xptr, tctx);
1034         break;
1035     default:
1036         rc = -1;
1037     }
1038     *nvp_blk = xive2_nvp_blk(cam);
1039     *nvp_idx = xive2_nvp_idx(cam);
1040     return rc;
1041 }
1042 
1043 static void xive2_tctx_accept_el(XivePresenter *xptr, XiveTCTX *tctx,
1044                                  uint8_t ring, uint8_t cl_ring)
1045 {
1046     uint64_t rd;
1047     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
1048     uint32_t nvp_idx, xive2_cfg;
1049     uint8_t nvp_blk;
1050     Xive2Nvp nvp;
1051     uint64_t phys_addr;
1052     uint8_t OGen = 0;
1053 
1054     xive2_tctx_get_nvp_indexes(tctx, cl_ring, &nvp_blk, &nvp_idx);
1055 
1056     if (xive2_router_get_nvp(xrtr, (uint8_t)nvp_blk, nvp_idx, &nvp)) {
1057         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
1058                       nvp_blk, nvp_idx);
1059         return;
1060     }
1061 
1062     if (!xive2_nvp_is_valid(&nvp)) {
1063         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1064                       nvp_blk, nvp_idx);
1065         return;
1066     }
1067 
1068 
1069     rd = xive_tctx_accept(tctx, ring);
1070 
1071     if (ring == TM_QW1_OS) {
1072         OGen = tctx->regs[ring + TM_OGEN];
1073     }
1074     xive2_cfg = xive2_router_get_config(xrtr);
1075     phys_addr = xive2_nvp_reporting_addr(&nvp);
1076     uint8_t report_data[REPORT_LINE_GEN1_SIZE];
1077     memset(report_data, 0xff, sizeof(report_data));
1078     if ((OGen == 1) || (xive2_cfg & XIVE2_GEN1_TIMA_OS)) {
1079         report_data[8] = (rd >> 8) & 0xff;
1080         report_data[9] = rd & 0xff;
1081     } else {
1082         report_data[0] = (rd >> 8) & 0xff;
1083         report_data[1] = rd & 0xff;
1084     }
1085     cpu_physical_memory_write(phys_addr, report_data, REPORT_LINE_GEN1_SIZE);
1086 }
1087 
1088 void xive2_tm_ack_os_el(XivePresenter *xptr, XiveTCTX *tctx,
1089                         hwaddr offset, uint64_t value, unsigned size)
1090 {
1091     xive2_tctx_accept_el(xptr, tctx, TM_QW1_OS, TM_QW1_OS);
1092 }
1093 
1094 /* Re-calculate and present pending interrupts */
1095 static void xive2_tctx_process_pending(XiveTCTX *tctx, uint8_t sig_ring)
1096 {
1097     uint8_t *sig_regs = &tctx->regs[sig_ring];
1098     Xive2Router *xrtr = XIVE2_ROUTER(tctx->xptr);
1099     uint8_t backlog_prio;
1100     uint8_t first_group;
1101     uint8_t group_level;
1102     uint8_t pipr_min;
1103     uint8_t lsmfb_min;
1104     uint8_t ring_min;
1105     uint8_t cppr = sig_regs[TM_CPPR];
1106     bool group_enabled;
1107     Xive2Nvp nvp;
1108     int rc;
1109 
1110     g_assert(sig_ring == TM_QW3_HV_PHYS || sig_ring == TM_QW1_OS);
1111     g_assert(!xive_nsr_indicates_group_exception(sig_ring, sig_regs[TM_NSR]));
1112 
1113     /*
1114      * Recompute the PIPR based on local pending interrupts. It will
1115      * be adjusted below if needed in case of pending group interrupts.
1116      */
1117 again:
1118     pipr_min = xive_ipb_to_pipr(sig_regs[TM_IPB]);
1119     group_enabled = !!sig_regs[TM_LGS];
1120     lsmfb_min = group_enabled ? sig_regs[TM_LSMFB] : 0xff;
1121     ring_min = sig_ring;
1122     group_level = 0;
1123 
1124     /* PHYS updates also depend on POOL values */
1125     if (sig_ring == TM_QW3_HV_PHYS) {
1126         uint8_t *pool_regs = &tctx->regs[TM_QW2_HV_POOL];
1127 
1128         /* POOL values only matter if POOL ctx is valid */
1129         if (pool_regs[TM_WORD2] & 0x80) {
1130             uint8_t pool_pipr = xive_ipb_to_pipr(pool_regs[TM_IPB]);
1131             uint8_t pool_lsmfb = pool_regs[TM_LSMFB];
1132 
1133             /*
1134              * Determine highest priority interrupt and
1135              * remember which ring has it.
1136              */
1137             if (pool_pipr < pipr_min) {
1138                 pipr_min = pool_pipr;
1139                 if (pool_pipr < lsmfb_min) {
1140                     ring_min = TM_QW2_HV_POOL;
1141                 }
1142             }
1143 
1144             /* Values needed for group priority calculation */
1145             if (pool_regs[TM_LGS] && (pool_lsmfb < lsmfb_min)) {
1146                 group_enabled = true;
1147                 lsmfb_min = pool_lsmfb;
1148                 if (lsmfb_min < pipr_min) {
1149                     ring_min = TM_QW2_HV_POOL;
1150                 }
1151             }
1152         }
1153     }
1154 
1155     if (group_enabled &&
1156         lsmfb_min < cppr &&
1157         lsmfb_min < pipr_min) {
1158 
1159         uint8_t nvp_blk;
1160         uint32_t nvp_idx;
1161 
1162         /*
1163          * Thread has seen a group interrupt with a higher priority
1164          * than the new cppr or pending local interrupt. Check the
1165          * backlog
1166          */
1167         rc = xive2_tctx_get_nvp_indexes(tctx, ring_min, &nvp_blk, &nvp_idx);
1168         if (rc) {
1169             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: set CPPR on invalid "
1170                                            "context\n");
1171             return;
1172         }
1173 
1174         if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
1175             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
1176                           nvp_blk, nvp_idx);
1177             return;
1178         }
1179 
1180         if (!xive2_nvp_is_valid(&nvp)) {
1181             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1182                           nvp_blk, nvp_idx);
1183             return;
1184         }
1185 
1186         first_group = xive_get_field32(NVP2_W0_PGOFIRST, nvp.w0);
1187         if (!first_group) {
1188             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1189                           nvp_blk, nvp_idx);
1190             return;
1191         }
1192 
1193         backlog_prio = xive2_presenter_backlog_scan(tctx->xptr,
1194                                                     nvp_blk, nvp_idx,
1195                                                     first_group, &group_level);
1196         tctx->regs[ring_min + TM_LSMFB] = backlog_prio;
1197         if (backlog_prio != lsmfb_min) {
1198             /*
1199              * If the group backlog scan finds a less favored or no interrupt,
1200              * then re-do the processing which may turn up a more favored
1201              * interrupt from IPB or the other pool. Backlog should not
1202              * find a priority < LSMFB.
1203              */
1204             g_assert(backlog_prio >= lsmfb_min);
1205             goto again;
1206         }
1207 
1208         xive2_presenter_backlog_decr(tctx->xptr, nvp_blk, nvp_idx,
1209                                      backlog_prio, group_level);
1210         pipr_min = backlog_prio;
1211     }
1212 
1213     if (pipr_min > cppr) {
1214         pipr_min = cppr;
1215     }
1216     xive_tctx_pipr_set(tctx, ring_min, pipr_min, group_level);
1217 }
1218 
1219 /* NOTE: CPPR only exists for TM_QW1_OS and TM_QW3_HV_PHYS */
1220 static void xive2_tctx_set_cppr(XiveTCTX *tctx, uint8_t sig_ring, uint8_t cppr)
1221 {
1222     uint8_t *sig_regs = &tctx->regs[sig_ring];
1223     Xive2Router *xrtr = XIVE2_ROUTER(tctx->xptr);
1224     uint8_t old_cppr;
1225     uint8_t nsr = sig_regs[TM_NSR];
1226 
1227     g_assert(sig_ring == TM_QW1_OS || sig_ring == TM_QW3_HV_PHYS);
1228 
1229     g_assert(tctx->regs[TM_QW2_HV_POOL + TM_NSR] == 0);
1230     g_assert(tctx->regs[TM_QW2_HV_POOL + TM_PIPR] == 0);
1231     g_assert(tctx->regs[TM_QW2_HV_POOL + TM_CPPR] == 0);
1232 
1233     /* XXX: should show pool IPB for PHYS ring */
1234     trace_xive_tctx_set_cppr(tctx->cs->cpu_index, sig_ring,
1235                              sig_regs[TM_IPB], sig_regs[TM_PIPR],
1236                              cppr, nsr);
1237 
1238     if (cppr > XIVE_PRIORITY_MAX) {
1239         cppr = 0xff;
1240     }
1241 
1242     old_cppr = sig_regs[TM_CPPR];
1243     sig_regs[TM_CPPR] = cppr;
1244 
1245     /* Handle increased CPPR priority (lower value) */
1246     if (cppr < old_cppr) {
1247         if (cppr <= sig_regs[TM_PIPR]) {
1248             /* CPPR lowered below PIPR, must un-present interrupt */
1249             if (xive_nsr_indicates_exception(sig_ring, nsr)) {
1250                 if (xive_nsr_indicates_group_exception(sig_ring, nsr)) {
1251                     /* redistribute precluded active grp interrupt */
1252                     xive2_redistribute(xrtr, tctx,
1253                                        xive_nsr_exception_ring(sig_ring, nsr));
1254                     return;
1255                 }
1256             }
1257 
1258             /* interrupt is VP directed, pending in IPB */
1259             xive_tctx_pipr_set(tctx, sig_ring, cppr, 0);
1260             return;
1261         } else {
1262             /* CPPR was lowered, but still above PIPR. No action needed. */
1263             return;
1264         }
1265     }
1266 
1267     /* CPPR didn't change, nothing needs to be done */
1268     if (cppr == old_cppr) {
1269         return;
1270     }
1271 
1272     /* CPPR priority decreased (higher value) */
1273     if (!xive_nsr_indicates_exception(sig_ring, nsr)) {
1274         xive2_tctx_process_pending(tctx, sig_ring);
1275     }
1276 }
1277 
1278 void xive2_tm_set_hv_cppr(XivePresenter *xptr, XiveTCTX *tctx,
1279                           hwaddr offset, uint64_t value, unsigned size)
1280 {
1281     xive2_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
1282 }
1283 
1284 void xive2_tm_set_os_cppr(XivePresenter *xptr, XiveTCTX *tctx,
1285                           hwaddr offset, uint64_t value, unsigned size)
1286 {
1287     xive2_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
1288 }
1289 
1290 static void xive2_tctx_set_target(XiveTCTX *tctx, uint8_t ring, uint8_t target)
1291 {
1292     uint8_t *regs = &tctx->regs[ring];
1293 
1294     regs[TM_T] = target;
1295 }
1296 
1297 void xive2_tm_set_hv_target(XivePresenter *xptr, XiveTCTX *tctx,
1298                             hwaddr offset, uint64_t value, unsigned size)
1299 {
1300     xive2_tctx_set_target(tctx, TM_QW3_HV_PHYS, value & 0xff);
1301 }
1302 
1303 /*
1304  * XIVE Router (aka. Virtualization Controller or IVRE)
1305  */
1306 
1307 int xive2_router_get_eas(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1308                          Xive2Eas *eas)
1309 {
1310     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1311 
1312     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1313 }
1314 
1315 static
1316 int xive2_router_get_pq(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1317                        uint8_t *pq)
1318 {
1319     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1320 
1321     return xrc->get_pq(xrtr, eas_blk, eas_idx, pq);
1322 }
1323 
1324 static
1325 int xive2_router_set_pq(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1326                        uint8_t *pq)
1327 {
1328     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1329 
1330     return xrc->set_pq(xrtr, eas_blk, eas_idx, pq);
1331 }
1332 
1333 int xive2_router_get_end(Xive2Router *xrtr, uint8_t end_blk, uint32_t end_idx,
1334                          Xive2End *end)
1335 {
1336    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1337 
1338    return xrc->get_end(xrtr, end_blk, end_idx, end);
1339 }
1340 
1341 int xive2_router_write_end(Xive2Router *xrtr, uint8_t end_blk, uint32_t end_idx,
1342                            Xive2End *end, uint8_t word_number)
1343 {
1344    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1345 
1346    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1347 }
1348 
1349 int xive2_router_get_nvp(Xive2Router *xrtr, uint8_t nvp_blk, uint32_t nvp_idx,
1350                          Xive2Nvp *nvp)
1351 {
1352    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1353 
1354    return xrc->get_nvp(xrtr, nvp_blk, nvp_idx, nvp);
1355 }
1356 
1357 int xive2_router_write_nvp(Xive2Router *xrtr, uint8_t nvp_blk, uint32_t nvp_idx,
1358                            Xive2Nvp *nvp, uint8_t word_number)
1359 {
1360    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1361 
1362    return xrc->write_nvp(xrtr, nvp_blk, nvp_idx, nvp, word_number);
1363 }
1364 
1365 int xive2_router_get_nvgc(Xive2Router *xrtr, bool crowd,
1366                           uint8_t nvgc_blk, uint32_t nvgc_idx,
1367                           Xive2Nvgc *nvgc)
1368 {
1369    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1370 
1371    return xrc->get_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
1372 }
1373 
1374 int xive2_router_write_nvgc(Xive2Router *xrtr, bool crowd,
1375                             uint8_t nvgc_blk, uint32_t nvgc_idx,
1376                             Xive2Nvgc *nvgc)
1377 {
1378    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1379 
1380    return xrc->write_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
1381 }
1382 
1383 static bool xive2_vp_match_mask(uint32_t cam1, uint32_t cam2,
1384                                 uint32_t vp_mask)
1385 {
1386     return (cam1 & vp_mask) == (cam2 & vp_mask);
1387 }
1388 
1389 static uint8_t xive2_get_vp_block_mask(uint32_t nvt_blk, bool crowd)
1390 {
1391     uint8_t block_mask = 0b1111;
1392 
1393     /* 3 supported crowd sizes: 2, 4, 16 */
1394     if (crowd) {
1395         uint32_t size = xive_get_vpgroup_size(nvt_blk);
1396 
1397         if (size != 2 && size != 4 && size != 16) {
1398             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid crowd size of %d",
1399                                            size);
1400             return block_mask;
1401         }
1402         block_mask &= ~(size - 1);
1403     }
1404     return block_mask;
1405 }
1406 
1407 static uint32_t xive2_get_vp_index_mask(uint32_t nvt_index, bool cam_ignore)
1408 {
1409     uint32_t index_mask = 0xFFFFFF; /* 24 bits */
1410 
1411     if (cam_ignore) {
1412         uint32_t size = xive_get_vpgroup_size(nvt_index);
1413 
1414         if (size < 2) {
1415             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid group size of %d",
1416                                            size);
1417             return index_mask;
1418         }
1419         index_mask &= ~(size - 1);
1420     }
1421     return index_mask;
1422 }
1423 
1424 /*
1425  * The thread context register words are in big-endian format.
1426  */
1427 int xive2_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1428                                uint8_t format,
1429                                uint8_t nvt_blk, uint32_t nvt_idx,
1430                                bool crowd, bool cam_ignore,
1431                                uint32_t logic_serv)
1432 {
1433     uint32_t cam =   xive2_nvp_cam_line(nvt_blk, nvt_idx);
1434     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1435     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1436     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1437     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1438 
1439     uint32_t index_mask, vp_mask;
1440     uint8_t block_mask;
1441 
1442     if (format == 0) {
1443         /*
1444          * i=0: Specific NVT notification
1445          * i=1: VP-group notification (bits ignored at the end of the
1446          *      NVT identifier)
1447          */
1448         block_mask = xive2_get_vp_block_mask(nvt_blk, crowd);
1449         index_mask = xive2_get_vp_index_mask(nvt_idx, cam_ignore);
1450         vp_mask = xive2_nvp_cam_line(block_mask, index_mask);
1451 
1452         /* For VP-group notifications, threads with LGS=0 are excluded */
1453 
1454         /* PHYS ring */
1455         if ((be32_to_cpu(qw3w2) & TM2_QW3W2_VT) &&
1456             !(cam_ignore && tctx->regs[TM_QW3_HV_PHYS + TM_LGS] == 0) &&
1457             xive2_vp_match_mask(cam,
1458                                 xive2_tctx_hw_cam_line(xptr, tctx),
1459                                 vp_mask)) {
1460             return TM_QW3_HV_PHYS;
1461         }
1462 
1463         /* HV POOL ring */
1464         if ((be32_to_cpu(qw2w2) & TM2_QW2W2_VP) &&
1465             !(cam_ignore && tctx->regs[TM_QW2_HV_POOL + TM_LGS] == 0) &&
1466             xive2_vp_match_mask(cam,
1467                                 xive_get_field32(TM2_QW2W2_POOL_CAM, qw2w2),
1468                                 vp_mask)) {
1469             return TM_QW2_HV_POOL;
1470         }
1471 
1472         /* OS ring */
1473         if ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
1474             !(cam_ignore && tctx->regs[TM_QW1_OS + TM_LGS] == 0) &&
1475             xive2_vp_match_mask(cam,
1476                                 xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2),
1477                                 vp_mask)) {
1478             return TM_QW1_OS;
1479         }
1480     } else {
1481         /* F=1 : User level Event-Based Branch (EBB) notification */
1482 
1483         /* FIXME: what if cam_ignore and LGS = 0 ? */
1484         /* USER ring */
1485         if  ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
1486              (cam == xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2)) &&
1487              (be32_to_cpu(qw0w2) & TM2_QW0W2_VU) &&
1488              (logic_serv == xive_get_field32(TM2_QW0W2_LOGIC_SERV, qw0w2))) {
1489             return TM_QW0_USER;
1490         }
1491     }
1492     return -1;
1493 }
1494 
1495 bool xive2_tm_irq_precluded(XiveTCTX *tctx, int ring, uint8_t priority)
1496 {
1497     uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
1498 
1499     /*
1500      * The xive2_presenter_tctx_match() above tells if there's a match
1501      * but for VP-group notification, we still need to look at the
1502      * priority to know if the thread can take the interrupt now or if
1503      * it is precluded.
1504      */
1505     if (priority < sig_regs[TM_PIPR]) {
1506         return false;
1507     }
1508     return true;
1509 }
1510 
1511 void xive2_tm_set_lsmfb(XiveTCTX *tctx, int ring, uint8_t priority)
1512 {
1513     uint8_t *regs = &tctx->regs[ring];
1514 
1515     /*
1516      * Called by the router during a VP-group notification when the
1517      * thread matches but can't take the interrupt because it's
1518      * already running at a more favored priority. It then stores the
1519      * new interrupt priority in the LSMFB field.
1520      */
1521     regs[TM_LSMFB] = priority;
1522 }
1523 
1524 static void xive2_router_realize(DeviceState *dev, Error **errp)
1525 {
1526     Xive2Router *xrtr = XIVE2_ROUTER(dev);
1527 
1528     assert(xrtr->xfb);
1529 }
1530 
1531 /*
1532  * Notification using the END ESe/ESn bit (Event State Buffer for
1533  * escalation and notification). Profide further coalescing in the
1534  * Router.
1535  */
1536 static bool xive2_router_end_es_notify(Xive2Router *xrtr, uint8_t end_blk,
1537                                        uint32_t end_idx, Xive2End *end,
1538                                        uint32_t end_esmask)
1539 {
1540     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1541     bool notify = xive_esb_trigger(&pq);
1542 
1543     if (pq != xive_get_field32(end_esmask, end->w1)) {
1544         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1545         xive2_router_write_end(xrtr, end_blk, end_idx, end, 1);
1546     }
1547 
1548     /* ESe/n[Q]=1 : end of notification */
1549     return notify;
1550 }
1551 
1552 /*
1553  * An END trigger can come from an event trigger (IPI or HW) or from
1554  * another chip. We don't model the PowerBus but the END trigger
1555  * message has the same parameters than in the function below.
1556  */
1557 static void xive2_router_end_notify(Xive2Router *xrtr, uint8_t end_blk,
1558                                     uint32_t end_idx, uint32_t end_data,
1559                                     bool redistribute)
1560 {
1561     Xive2End end;
1562     uint8_t priority;
1563     uint8_t format;
1564     XiveTCTXMatch match;
1565     bool crowd, cam_ignore;
1566     uint8_t nvx_blk;
1567     uint32_t nvx_idx;
1568 
1569     /* END cache lookup */
1570     if (xive2_router_get_end(xrtr, end_blk, end_idx, &end)) {
1571         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1572                       end_idx);
1573         return;
1574     }
1575 
1576     if (!xive2_end_is_valid(&end)) {
1577         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1578                       end_blk, end_idx);
1579         return;
1580     }
1581 
1582     if (xive2_end_is_crowd(&end) && !xive2_end_is_ignore(&end)) {
1583         qemu_log_mask(LOG_GUEST_ERROR,
1584                       "XIVE: invalid END, 'crowd' bit requires 'ignore' bit\n");
1585         return;
1586     }
1587 
1588     if (!redistribute && xive2_end_is_enqueue(&end)) {
1589         trace_xive_end_enqueue(end_blk, end_idx, end_data);
1590         xive2_end_enqueue(&end, end_data);
1591         /* Enqueuing event data modifies the EQ toggle and index */
1592         xive2_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1593     }
1594 
1595     /*
1596      * When the END is silent, we skip the notification part.
1597      */
1598     if (xive2_end_is_silent_escalation(&end)) {
1599         goto do_escalation;
1600     }
1601 
1602     /*
1603      * The W7 format depends on the F bit in W6. It defines the type
1604      * of the notification :
1605      *
1606      *   F=0 : single or multiple NVP notification
1607      *   F=1 : User level Event-Based Branch (EBB) notification, no
1608      *         priority
1609      */
1610     format = xive_get_field32(END2_W6_FORMAT_BIT, end.w6);
1611     priority = xive_get_field32(END2_W7_F0_PRIORITY, end.w7);
1612 
1613     /* The END is masked */
1614     if (format == 0 && priority == 0xff) {
1615         return;
1616     }
1617 
1618     /*
1619      * Check the END ESn (Event State Buffer for notification) for
1620      * even further coalescing in the Router
1621      */
1622     if (!xive2_end_is_notify(&end)) {
1623         /* ESn[Q]=1 : end of notification */
1624         if (!xive2_router_end_es_notify(xrtr, end_blk, end_idx,
1625                                        &end, END2_W1_ESn)) {
1626             return;
1627         }
1628     }
1629 
1630     /*
1631      * Follows IVPE notification
1632      */
1633     nvx_blk = xive_get_field32(END2_W6_VP_BLOCK, end.w6);
1634     nvx_idx = xive_get_field32(END2_W6_VP_OFFSET, end.w6);
1635     crowd = xive2_end_is_crowd(&end);
1636     cam_ignore = xive2_end_is_ignore(&end);
1637 
1638     /* TODO: Auto EOI. */
1639     if (xive_presenter_match(xrtr->xfb, format, nvx_blk, nvx_idx,
1640                              crowd, cam_ignore, priority,
1641                              xive_get_field32(END2_W7_F1_LOG_SERVER_ID, end.w7),
1642                              &match)) {
1643         XiveTCTX *tctx = match.tctx;
1644         uint8_t ring = match.ring;
1645         uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
1646         uint8_t nsr = sig_regs[TM_NSR];
1647         uint8_t group_level;
1648 
1649         if (priority < sig_regs[TM_PIPR] &&
1650             xive_nsr_indicates_group_exception(ring, nsr)) {
1651             xive2_redistribute(xrtr, tctx, xive_nsr_exception_ring(ring, nsr));
1652         }
1653 
1654         group_level = xive_get_group_level(crowd, cam_ignore, nvx_blk, nvx_idx);
1655         trace_xive_presenter_notify(nvx_blk, nvx_idx, ring, group_level);
1656         xive_tctx_pipr_present(tctx, ring, priority, group_level);
1657         return;
1658     }
1659 
1660     /*
1661      * If no matching NVP is dispatched on a HW thread :
1662      * - specific VP: update the NVP structure if backlog is activated
1663      * - VP-group: update the backlog counter for that priority in the NVG
1664      */
1665     if (xive2_end_is_backlog(&end)) {
1666 
1667         if (format == 1) {
1668             qemu_log_mask(LOG_GUEST_ERROR,
1669                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1670                           end_blk, end_idx);
1671             return;
1672         }
1673 
1674         if (!cam_ignore) {
1675             uint8_t ipb;
1676             Xive2Nvp nvp;
1677 
1678             /* NVP cache lookup */
1679             if (xive2_router_get_nvp(xrtr, nvx_blk, nvx_idx, &nvp)) {
1680                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVP %x/%x\n",
1681                               nvx_blk, nvx_idx);
1682                 return;
1683             }
1684 
1685             if (!xive2_nvp_is_valid(&nvp)) {
1686                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is invalid\n",
1687                               nvx_blk, nvx_idx);
1688                 return;
1689             }
1690 
1691             /*
1692              * Record the IPB in the associated NVP structure for later
1693              * use. The presenter will resend the interrupt when the vCPU
1694              * is dispatched again on a HW thread.
1695              */
1696             ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2) |
1697                 xive_priority_to_ipb(priority);
1698             nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, ipb);
1699             xive2_router_write_nvp(xrtr, nvx_blk, nvx_idx, &nvp, 2);
1700         } else {
1701             Xive2Nvgc nvgc;
1702             uint32_t backlog;
1703 
1704             /*
1705              * For groups and crowds, the per-priority backlog
1706              * counters are stored in the NVG/NVC structures
1707              */
1708             if (xive2_router_get_nvgc(xrtr, crowd,
1709                                       nvx_blk, nvx_idx, &nvgc)) {
1710                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no %s %x/%x\n",
1711                               crowd ? "NVC" : "NVG", nvx_blk, nvx_idx);
1712                 return;
1713             }
1714 
1715             if (!xive2_nvgc_is_valid(&nvgc)) {
1716                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVG %x/%x is invalid\n",
1717                               nvx_blk, nvx_idx);
1718                 return;
1719             }
1720 
1721             /*
1722              * Increment the backlog counter for that priority.
1723              * We only call broadcast the first time the counter is
1724              * incremented. broadcast will set the LSMFB field of the TIMA of
1725              * relevant threads so that they know an interrupt is pending.
1726              */
1727             backlog = xive2_nvgc_get_backlog(&nvgc, priority) + 1;
1728             xive2_nvgc_set_backlog(&nvgc, priority, backlog);
1729             xive2_router_write_nvgc(xrtr, crowd, nvx_blk, nvx_idx, &nvgc);
1730 
1731             if (backlog == 1) {
1732                 XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xrtr->xfb);
1733                 xfc->broadcast(xrtr->xfb, nvx_blk, nvx_idx,
1734                                crowd, cam_ignore, priority);
1735 
1736                 if (!xive2_end_is_precluded_escalation(&end)) {
1737                     /*
1738                      * The interrupt will be picked up when the
1739                      * matching thread lowers its priority level
1740                      */
1741                     return;
1742                 }
1743             }
1744         }
1745     }
1746 
1747 do_escalation:
1748     /*
1749      * If activated, escalate notification using the ESe PQ bits and
1750      * the EAS in w4-5
1751      */
1752     if (!xive2_end_is_escalate(&end)) {
1753         return;
1754     }
1755 
1756     /*
1757      * Check the END ESe (Event State Buffer for escalation) for even
1758      * further coalescing in the Router
1759      */
1760     if (!xive2_end_is_uncond_escalation(&end)) {
1761         /* ESe[Q]=1 : end of escalation notification */
1762         if (!xive2_router_end_es_notify(xrtr, end_blk, end_idx,
1763                                        &end, END2_W1_ESe)) {
1764             return;
1765         }
1766     }
1767 
1768     if (xive2_end_is_escalate_end(&end)) {
1769         /*
1770          * Perform END Adaptive escalation processing
1771          * The END trigger becomes an Escalation trigger
1772          */
1773         uint8_t esc_blk = xive_get_field32(END2_W4_END_BLOCK, end.w4);
1774         uint32_t esc_idx = xive_get_field32(END2_W4_ESC_END_INDEX, end.w4);
1775         uint32_t esc_data = xive_get_field32(END2_W5_ESC_END_DATA, end.w5);
1776         trace_xive_escalate_end(end_blk, end_idx, esc_blk, esc_idx, esc_data);
1777         xive2_router_end_notify(xrtr, esc_blk, esc_idx, esc_data, false);
1778     } /* end END adaptive escalation */
1779 
1780     else {
1781         uint32_t lisn;              /* Logical Interrupt Source Number */
1782 
1783         /*
1784          *  Perform ESB escalation processing
1785          *      E[N] == 1 --> N
1786          *      Req[Block] <- E[ESB_Block]
1787          *      Req[Index] <- E[ESB_Index]
1788          *      Req[Offset] <- 0x000
1789          *      Execute <ESB Store> Req command
1790          */
1791         lisn = XIVE_EAS(xive_get_field32(END2_W4_END_BLOCK,     end.w4),
1792                         xive_get_field32(END2_W4_ESC_END_INDEX, end.w4));
1793 
1794         trace_xive_escalate_esb(end_blk, end_idx, lisn);
1795         xive2_notify(xrtr, lisn, true /* pq_checked */);
1796     }
1797 
1798     return;
1799 }
1800 
1801 void xive2_notify(Xive2Router *xrtr , uint32_t lisn, bool pq_checked)
1802 {
1803     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1804     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1805     Xive2Eas eas;
1806 
1807     /* EAS cache lookup */
1808     if (xive2_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1809         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1810         return;
1811     }
1812 
1813     if (!pq_checked) {
1814         bool notify;
1815         uint8_t pq;
1816 
1817         /* PQ cache lookup */
1818         if (xive2_router_get_pq(xrtr, eas_blk, eas_idx, &pq)) {
1819             /* Set FIR */
1820             g_assert_not_reached();
1821         }
1822 
1823         notify = xive_esb_trigger(&pq);
1824 
1825         if (xive2_router_set_pq(xrtr, eas_blk, eas_idx, &pq)) {
1826             /* Set FIR */
1827             g_assert_not_reached();
1828         }
1829 
1830         if (!notify) {
1831             return;
1832         }
1833     }
1834 
1835     if (!xive2_eas_is_valid(&eas)) {
1836         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN %x\n", lisn);
1837         return;
1838     }
1839 
1840     if (xive2_eas_is_masked(&eas)) {
1841         /* Notification completed */
1842         return;
1843     }
1844 
1845     /* TODO: add support for EAS resume */
1846     if (xive2_eas_is_resume(&eas)) {
1847         qemu_log_mask(LOG_UNIMP,
1848                       "XIVE: EAS resume processing unimplemented - LISN %x\n",
1849                       lisn);
1850         return;
1851     }
1852 
1853     /*
1854      * The event trigger becomes an END trigger
1855      */
1856     xive2_router_end_notify(xrtr,
1857                             xive_get_field64(EAS2_END_BLOCK, eas.w),
1858                             xive_get_field64(EAS2_END_INDEX, eas.w),
1859                             xive_get_field64(EAS2_END_DATA,  eas.w),
1860                             false);
1861     return;
1862 }
1863 
1864 void xive2_router_notify(XiveNotifier *xn, uint32_t lisn, bool pq_checked)
1865 {
1866     Xive2Router *xrtr = XIVE2_ROUTER(xn);
1867 
1868     xive2_notify(xrtr, lisn, pq_checked);
1869     return;
1870 }
1871 
1872 static const Property xive2_router_properties[] = {
1873     DEFINE_PROP_LINK("xive-fabric", Xive2Router, xfb,
1874                      TYPE_XIVE_FABRIC, XiveFabric *),
1875 };
1876 
1877 static void xive2_router_class_init(ObjectClass *klass, const void *data)
1878 {
1879     DeviceClass *dc = DEVICE_CLASS(klass);
1880     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1881 
1882     dc->desc    = "XIVE2 Router Engine";
1883     device_class_set_props(dc, xive2_router_properties);
1884     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1885     dc->realize = xive2_router_realize;
1886     xnc->notify = xive2_router_notify;
1887 }
1888 
1889 static const TypeInfo xive2_router_info = {
1890     .name          = TYPE_XIVE2_ROUTER,
1891     .parent        = TYPE_SYS_BUS_DEVICE,
1892     .abstract      = true,
1893     .instance_size = sizeof(Xive2Router),
1894     .class_size    = sizeof(Xive2RouterClass),
1895     .class_init    = xive2_router_class_init,
1896     .interfaces    = (const InterfaceInfo[]) {
1897         { TYPE_XIVE_NOTIFIER },
1898         { TYPE_XIVE_PRESENTER },
1899         { }
1900     }
1901 };
1902 
1903 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
1904 {
1905     return !((addr >> shift) & 1);
1906 }
1907 
1908 static uint64_t xive2_end_source_read(void *opaque, hwaddr addr, unsigned size)
1909 {
1910     Xive2EndSource *xsrc = XIVE2_END_SOURCE(opaque);
1911     uint32_t offset = addr & 0xFFF;
1912     uint8_t end_blk;
1913     uint32_t end_idx;
1914     Xive2End end;
1915     uint32_t end_esmask;
1916     uint8_t pq;
1917     uint64_t ret;
1918 
1919     /*
1920      * The block id should be deduced from the load address on the END
1921      * ESB MMIO but our model only supports a single block per XIVE chip.
1922      */
1923     end_blk = xive2_router_get_block_id(xsrc->xrtr);
1924     end_idx = addr >> (xsrc->esb_shift + 1);
1925 
1926     if (xive2_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1927         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1928                       end_idx);
1929         return -1;
1930     }
1931 
1932     if (!xive2_end_is_valid(&end)) {
1933         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1934                       end_blk, end_idx);
1935         return -1;
1936     }
1937 
1938     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END2_W1_ESn :
1939         END2_W1_ESe;
1940     pq = xive_get_field32(end_esmask, end.w1);
1941 
1942     switch (offset) {
1943     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1944         ret = xive_esb_eoi(&pq);
1945 
1946         /* Forward the source event notification for routing ?? */
1947         break;
1948 
1949     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1950         ret = pq;
1951         break;
1952 
1953     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1954     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1955     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1956     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1957         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1958         break;
1959     default:
1960         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1961                       offset);
1962         return -1;
1963     }
1964 
1965     if (pq != xive_get_field32(end_esmask, end.w1)) {
1966         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1967         xive2_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1968     }
1969 
1970     return ret;
1971 }
1972 
1973 static void xive2_end_source_write(void *opaque, hwaddr addr,
1974                                    uint64_t value, unsigned size)
1975 {
1976     Xive2EndSource *xsrc = XIVE2_END_SOURCE(opaque);
1977     uint32_t offset = addr & 0xFFF;
1978     uint8_t end_blk;
1979     uint32_t end_idx;
1980     Xive2End end;
1981     uint32_t end_esmask;
1982     uint8_t pq;
1983     bool notify = false;
1984 
1985     /*
1986      * The block id should be deduced from the load address on the END
1987      * ESB MMIO but our model only supports a single block per XIVE chip.
1988      */
1989     end_blk = xive2_router_get_block_id(xsrc->xrtr);
1990     end_idx = addr >> (xsrc->esb_shift + 1);
1991 
1992     if (xive2_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1993         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1994                       end_idx);
1995         return;
1996     }
1997 
1998     if (!xive2_end_is_valid(&end)) {
1999         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
2000                       end_blk, end_idx);
2001         return;
2002     }
2003 
2004     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END2_W1_ESn :
2005         END2_W1_ESe;
2006     pq = xive_get_field32(end_esmask, end.w1);
2007 
2008     switch (offset) {
2009     case 0 ... 0x3FF:
2010         notify = xive_esb_trigger(&pq);
2011         break;
2012 
2013     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
2014         /* TODO: can we check StoreEOI availability from the router ? */
2015         notify = xive_esb_eoi(&pq);
2016         break;
2017 
2018     case XIVE_ESB_INJECT ... XIVE_ESB_INJECT + 0x3FF:
2019         if (end_esmask == END2_W1_ESe) {
2020             qemu_log_mask(LOG_GUEST_ERROR,
2021                           "XIVE: END %x/%x can not EQ inject on ESe\n",
2022                            end_blk, end_idx);
2023             return;
2024         }
2025         notify = true;
2026         break;
2027 
2028     default:
2029         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB write addr %d\n",
2030                       offset);
2031         return;
2032     }
2033 
2034     if (pq != xive_get_field32(end_esmask, end.w1)) {
2035         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
2036         xive2_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
2037     }
2038 
2039     /* TODO: Forward the source event notification for routing */
2040     if (notify) {
2041         ;
2042     }
2043 }
2044 
2045 static const MemoryRegionOps xive2_end_source_ops = {
2046     .read = xive2_end_source_read,
2047     .write = xive2_end_source_write,
2048     .endianness = DEVICE_BIG_ENDIAN,
2049     .valid = {
2050         .min_access_size = 1,
2051         .max_access_size = 8,
2052     },
2053     .impl = {
2054         .min_access_size = 1,
2055         .max_access_size = 8,
2056     },
2057 };
2058 
2059 static void xive2_end_source_realize(DeviceState *dev, Error **errp)
2060 {
2061     Xive2EndSource *xsrc = XIVE2_END_SOURCE(dev);
2062 
2063     assert(xsrc->xrtr);
2064 
2065     if (!xsrc->nr_ends) {
2066         error_setg(errp, "Number of interrupt needs to be greater than 0");
2067         return;
2068     }
2069 
2070     if (xsrc->esb_shift != XIVE_ESB_4K &&
2071         xsrc->esb_shift != XIVE_ESB_64K) {
2072         error_setg(errp, "Invalid ESB shift setting");
2073         return;
2074     }
2075 
2076     /*
2077      * Each END is assigned an even/odd pair of MMIO pages, the even page
2078      * manages the ESn field while the odd page manages the ESe field.
2079      */
2080     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
2081                           &xive2_end_source_ops, xsrc, "xive.end",
2082                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
2083 }
2084 
2085 static const Property xive2_end_source_properties[] = {
2086     DEFINE_PROP_UINT32("nr-ends", Xive2EndSource, nr_ends, 0),
2087     DEFINE_PROP_UINT32("shift", Xive2EndSource, esb_shift, XIVE_ESB_64K),
2088     DEFINE_PROP_LINK("xive", Xive2EndSource, xrtr, TYPE_XIVE2_ROUTER,
2089                      Xive2Router *),
2090 };
2091 
2092 static void xive2_end_source_class_init(ObjectClass *klass, const void *data)
2093 {
2094     DeviceClass *dc = DEVICE_CLASS(klass);
2095 
2096     dc->desc    = "XIVE END Source";
2097     device_class_set_props(dc, xive2_end_source_properties);
2098     dc->realize = xive2_end_source_realize;
2099     dc->user_creatable = false;
2100 }
2101 
2102 static const TypeInfo xive2_end_source_info = {
2103     .name          = TYPE_XIVE2_END_SOURCE,
2104     .parent        = TYPE_DEVICE,
2105     .instance_size = sizeof(Xive2EndSource),
2106     .class_init    = xive2_end_source_class_init,
2107 };
2108 
2109 static void xive2_register_types(void)
2110 {
2111     type_register_static(&xive2_router_info);
2112     type_register_static(&xive2_end_source_info);
2113 }
2114 
2115 type_init(xive2_register_types)
2116