xref: /openbmc/qemu/hw/intc/xive2.c (revision 7a40b50757b55c2d4233c304f32df5afdd1bd63a)
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          * Lower external interrupt line of requested ring and below except for
733          * USER, which doesn't exist.
734          */
735         if (xive_nsr_indicates_exception(cur_ring, nsr)) {
736             if (cur_ring == xive_nsr_exception_ring(cur_ring, nsr)) {
737                 xive_tctx_reset_signal(tctx, cur_ring);
738             }
739         }
740     }
741 
742     if (xive2_router_get_config(xrtr) & XIVE2_VP_SAVE_RESTORE && do_save) {
743         xive2_tctx_save_ctx(xrtr, tctx, ring, nvp_blk, nvp_idx);
744     }
745 
746     return target_ringw2;
747 }
748 
749 uint64_t xive2_tm_pull_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
750                               hwaddr offset, unsigned size)
751 {
752     return xive2_tm_pull_ctx(xptr, tctx, offset, size, TM_QW1_OS);
753 }
754 
755 uint64_t xive2_tm_pull_pool_ctx(XivePresenter *xptr, XiveTCTX *tctx,
756                                 hwaddr offset, unsigned size)
757 {
758     return xive2_tm_pull_ctx(xptr, tctx, offset, size, TM_QW2_HV_POOL);
759 }
760 
761 uint64_t xive2_tm_pull_phys_ctx(XivePresenter *xptr, XiveTCTX *tctx,
762                                 hwaddr offset, unsigned size)
763 {
764     return xive2_tm_pull_ctx(xptr, tctx, offset, size, TM_QW3_HV_PHYS);
765 }
766 
767 #define REPORT_LINE_GEN1_SIZE       16
768 
769 static void xive2_tm_report_line_gen1(XiveTCTX *tctx, uint8_t *data,
770                                       uint8_t size)
771 {
772     uint8_t *regs = tctx->regs;
773 
774     g_assert(size == REPORT_LINE_GEN1_SIZE);
775     memset(data, 0, size);
776     /*
777      * See xive architecture for description of what is saved. It is
778      * hand-picked information to fit in 16 bytes.
779      */
780     data[0x0] = regs[TM_QW3_HV_PHYS + TM_NSR];
781     data[0x1] = regs[TM_QW3_HV_PHYS + TM_CPPR];
782     data[0x2] = regs[TM_QW3_HV_PHYS + TM_IPB];
783     data[0x3] = regs[TM_QW2_HV_POOL + TM_IPB];
784     data[0x4] = regs[TM_QW1_OS + TM_ACK_CNT];
785     data[0x5] = regs[TM_QW3_HV_PHYS + TM_LGS];
786     data[0x6] = 0xFF;
787     data[0x7] = regs[TM_QW3_HV_PHYS + TM_WORD2] & 0x80;
788     data[0x7] |= (regs[TM_QW2_HV_POOL + TM_WORD2] & 0x80) >> 1;
789     data[0x7] |= (regs[TM_QW1_OS + TM_WORD2] & 0x80) >> 2;
790     data[0x7] |= (regs[TM_QW3_HV_PHYS + TM_WORD2] & 0x3);
791     data[0x8] = regs[TM_QW1_OS + TM_NSR];
792     data[0x9] = regs[TM_QW1_OS + TM_CPPR];
793     data[0xA] = regs[TM_QW1_OS + TM_IPB];
794     data[0xB] = regs[TM_QW1_OS + TM_LGS];
795     if (regs[TM_QW0_USER + TM_WORD2] & 0x80) {
796         /*
797          * Logical server extension, except VU bit replaced by EB bit
798          * from NSR
799          */
800         data[0xC] = regs[TM_QW0_USER + TM_WORD2];
801         data[0xC] &= ~0x80;
802         data[0xC] |= regs[TM_QW0_USER + TM_NSR] & 0x80;
803         data[0xD] = regs[TM_QW0_USER + TM_WORD2 + 1];
804         data[0xE] = regs[TM_QW0_USER + TM_WORD2 + 2];
805         data[0xF] = regs[TM_QW0_USER + TM_WORD2 + 3];
806     }
807 }
808 
809 static void xive2_tm_pull_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
810                                  hwaddr offset, uint64_t value,
811                                  unsigned size, uint8_t ring)
812 {
813     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
814     uint32_t hw_cam, nvp_idx, xive2_cfg, reserved;
815     uint8_t nvp_blk;
816     Xive2Nvp nvp;
817     uint64_t phys_addr;
818     MemTxResult result;
819 
820     hw_cam = xive2_tctx_hw_cam_line(xptr, tctx);
821     nvp_blk = xive2_nvp_blk(hw_cam);
822     nvp_idx = xive2_nvp_idx(hw_cam);
823 
824     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
825         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
826                       nvp_blk, nvp_idx);
827         return;
828     }
829 
830     if (!xive2_nvp_is_valid(&nvp)) {
831         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
832                       nvp_blk, nvp_idx);
833         return;
834     }
835 
836     xive2_cfg = xive2_router_get_config(xrtr);
837 
838     phys_addr = xive2_nvp_reporting_addr(&nvp) + 0x80; /* odd line */
839     if (xive2_cfg & XIVE2_GEN1_TIMA_OS) {
840         uint8_t pull_ctxt[REPORT_LINE_GEN1_SIZE];
841 
842         xive2_tm_report_line_gen1(tctx, pull_ctxt, REPORT_LINE_GEN1_SIZE);
843         result = dma_memory_write(&address_space_memory, phys_addr,
844                                   pull_ctxt, REPORT_LINE_GEN1_SIZE,
845                                   MEMTXATTRS_UNSPECIFIED);
846         assert(result == MEMTX_OK);
847     } else {
848         result = dma_memory_write(&address_space_memory, phys_addr,
849                                   &tctx->regs, sizeof(tctx->regs),
850                                   MEMTXATTRS_UNSPECIFIED);
851         assert(result == MEMTX_OK);
852         reserved = 0xFFFFFFFF;
853         result = dma_memory_write(&address_space_memory, phys_addr + 12,
854                                   &reserved, sizeof(reserved),
855                                   MEMTXATTRS_UNSPECIFIED);
856         assert(result == MEMTX_OK);
857     }
858 
859     /* the rest is similar to pull context to registers */
860     xive2_tm_pull_ctx(xptr, tctx, offset, size, ring);
861 }
862 
863 void xive2_tm_pull_os_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
864                              hwaddr offset, uint64_t value, unsigned size)
865 {
866     xive2_tm_pull_ctx_ol(xptr, tctx, offset, value, size, TM_QW1_OS);
867 }
868 
869 
870 void xive2_tm_pull_phys_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
871                                hwaddr offset, uint64_t value, unsigned size)
872 {
873     xive2_tm_pull_ctx_ol(xptr, tctx, offset, value, size, TM_QW3_HV_PHYS);
874 }
875 
876 static uint8_t xive2_tctx_restore_ctx(Xive2Router *xrtr, XiveTCTX *tctx,
877                                       uint8_t ring,
878                                       uint8_t nvp_blk, uint32_t nvp_idx,
879                                       Xive2Nvp *nvp)
880 {
881     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
882     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
883     uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
884     uint8_t *regs = &tctx->regs[ring];
885     uint8_t cppr;
886 
887     if (!xive2_nvp_is_hw(nvp)) {
888         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is not HW owned\n",
889                       nvp_blk, nvp_idx);
890         return 0;
891     }
892 
893     cppr = xive_get_field32(NVP2_W2_CPPR, nvp->w2);
894     nvp->w2 = xive_set_field32(NVP2_W2_CPPR, nvp->w2, 0);
895     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, nvp, 2);
896 
897     sig_regs[TM_CPPR] = cppr;
898     regs[TM_LSMFB] = xive_get_field32(NVP2_W2_LSMFB, nvp->w2);
899     regs[TM_LGS] = xive_get_field32(NVP2_W2_LGS, nvp->w2);
900     regs[TM_T] = xive_get_field32(NVP2_W2_T, nvp->w2);
901 
902     nvp->w1 = xive_set_field32(NVP2_W1_CO, nvp->w1, 1);
903     nvp->w1 = xive_set_field32(NVP2_W1_CO_THRID_VALID, nvp->w1, 1);
904     nvp->w1 = xive_set_field32(NVP2_W1_CO_THRID, nvp->w1, pir);
905 
906     /*
907      * Checkout privilege: 0:OS, 1:Pool, 2:Hard
908      *
909      * TODO: we don't support hard push/pull
910      */
911     switch (ring) {
912     case TM_QW1_OS:
913         nvp->w1 = xive_set_field32(NVP2_W1_CO_PRIV, nvp->w1, 0);
914         break;
915     case TM_QW2_HV_POOL:
916         nvp->w1 = xive_set_field32(NVP2_W1_CO_PRIV, nvp->w1, 1);
917         break;
918     default:
919         g_assert_not_reached();
920     }
921 
922     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, nvp, 1);
923 
924     /* return restored CPPR to generate a CPU exception if needed */
925     return cppr;
926 }
927 
928 static void xive2_tctx_process_pending(XiveTCTX *tctx, uint8_t sig_ring);
929 
930 static void xive2_tctx_need_resend(Xive2Router *xrtr, XiveTCTX *tctx,
931                                    uint8_t nvp_blk, uint32_t nvp_idx,
932                                    bool do_restore)
933 {
934     uint8_t *regs = &tctx->regs[TM_QW1_OS];
935     uint8_t ipb;
936     Xive2Nvp nvp;
937 
938     /*
939      * Grab the associated thread interrupt context registers in the
940      * associated NVP
941      */
942     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
943         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
944                       nvp_blk, nvp_idx);
945         return;
946     }
947 
948     if (!xive2_nvp_is_valid(&nvp)) {
949         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
950                       nvp_blk, nvp_idx);
951         return;
952     }
953 
954     /* Automatically restore thread context registers */
955     if (xive2_router_get_config(xrtr) & XIVE2_VP_SAVE_RESTORE && do_restore) {
956         xive2_tctx_restore_ctx(xrtr, tctx, TM_QW1_OS, nvp_blk, nvp_idx, &nvp);
957     }
958 
959     ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2);
960     if (ipb) {
961         nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, 0);
962         xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, &nvp, 2);
963     }
964     /* IPB bits in the backlog are merged with the TIMA IPB bits */
965     regs[TM_IPB] |= ipb;
966 
967     xive2_tctx_process_pending(tctx, TM_QW1_OS);
968 }
969 
970 /*
971  * Updating the OS CAM line can trigger a resend of interrupt
972  */
973 void xive2_tm_push_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
974                           hwaddr offset, uint64_t value, unsigned size)
975 {
976     uint32_t cam;
977     uint32_t qw1w2;
978     uint64_t qw1dw1;
979     uint8_t nvp_blk;
980     uint32_t nvp_idx;
981     bool vo;
982     bool do_restore;
983 
984     /* First update the thead context */
985     switch (size) {
986     case 4:
987         cam = value;
988         qw1w2 = cpu_to_be32(cam);
989         memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
990         break;
991     case 8:
992         cam = value >> 32;
993         qw1dw1 = cpu_to_be64(value);
994         memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1dw1, 8);
995         break;
996     default:
997         g_assert_not_reached();
998     }
999 
1000     xive2_cam_decode(cam, &nvp_blk, &nvp_idx, &vo, &do_restore);
1001 
1002     /* Check the interrupt pending bits */
1003     if (vo) {
1004         xive2_tctx_need_resend(XIVE2_ROUTER(xptr), tctx, nvp_blk, nvp_idx,
1005                                do_restore);
1006     }
1007 }
1008 
1009 /* returns -1 if ring is invalid, but still populates block and index */
1010 static int xive2_tctx_get_nvp_indexes(XiveTCTX *tctx, uint8_t ring,
1011                                       uint8_t *nvp_blk, uint32_t *nvp_idx)
1012 {
1013     uint32_t w2;
1014     uint32_t cam = 0;
1015     int rc = 0;
1016 
1017     w2 = xive_tctx_word2(&tctx->regs[ring]);
1018     switch (ring) {
1019     case TM_QW1_OS:
1020         if (!(be32_to_cpu(w2) & TM2_QW1W2_VO)) {
1021             rc = -1;
1022         }
1023         cam = xive_get_field32(TM2_QW1W2_OS_CAM, w2);
1024         break;
1025     case TM_QW2_HV_POOL:
1026         if (!(be32_to_cpu(w2) & TM2_QW2W2_VP)) {
1027             rc = -1;
1028         }
1029         cam = xive_get_field32(TM2_QW2W2_POOL_CAM, w2);
1030         break;
1031     case TM_QW3_HV_PHYS:
1032         if (!(be32_to_cpu(w2) & TM2_QW3W2_VT)) {
1033             rc = -1;
1034         }
1035         cam = xive2_tctx_hw_cam_line(tctx->xptr, tctx);
1036         break;
1037     default:
1038         rc = -1;
1039     }
1040     *nvp_blk = xive2_nvp_blk(cam);
1041     *nvp_idx = xive2_nvp_idx(cam);
1042     return rc;
1043 }
1044 
1045 static void xive2_tctx_accept_el(XivePresenter *xptr, XiveTCTX *tctx,
1046                                  uint8_t ring, uint8_t cl_ring)
1047 {
1048     uint64_t rd;
1049     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
1050     uint32_t nvp_idx, xive2_cfg;
1051     uint8_t nvp_blk;
1052     Xive2Nvp nvp;
1053     uint64_t phys_addr;
1054     uint8_t OGen = 0;
1055 
1056     xive2_tctx_get_nvp_indexes(tctx, cl_ring, &nvp_blk, &nvp_idx);
1057 
1058     if (xive2_router_get_nvp(xrtr, (uint8_t)nvp_blk, nvp_idx, &nvp)) {
1059         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
1060                       nvp_blk, nvp_idx);
1061         return;
1062     }
1063 
1064     if (!xive2_nvp_is_valid(&nvp)) {
1065         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1066                       nvp_blk, nvp_idx);
1067         return;
1068     }
1069 
1070 
1071     rd = xive_tctx_accept(tctx, ring);
1072 
1073     if (ring == TM_QW1_OS) {
1074         OGen = tctx->regs[ring + TM_OGEN];
1075     }
1076     xive2_cfg = xive2_router_get_config(xrtr);
1077     phys_addr = xive2_nvp_reporting_addr(&nvp);
1078     uint8_t report_data[REPORT_LINE_GEN1_SIZE];
1079     memset(report_data, 0xff, sizeof(report_data));
1080     if ((OGen == 1) || (xive2_cfg & XIVE2_GEN1_TIMA_OS)) {
1081         report_data[8] = (rd >> 8) & 0xff;
1082         report_data[9] = rd & 0xff;
1083     } else {
1084         report_data[0] = (rd >> 8) & 0xff;
1085         report_data[1] = rd & 0xff;
1086     }
1087     cpu_physical_memory_write(phys_addr, report_data, REPORT_LINE_GEN1_SIZE);
1088 }
1089 
1090 void xive2_tm_ack_os_el(XivePresenter *xptr, XiveTCTX *tctx,
1091                         hwaddr offset, uint64_t value, unsigned size)
1092 {
1093     xive2_tctx_accept_el(xptr, tctx, TM_QW1_OS, TM_QW1_OS);
1094 }
1095 
1096 /* Re-calculate and present pending interrupts */
1097 static void xive2_tctx_process_pending(XiveTCTX *tctx, uint8_t sig_ring)
1098 {
1099     uint8_t *sig_regs = &tctx->regs[sig_ring];
1100     Xive2Router *xrtr = XIVE2_ROUTER(tctx->xptr);
1101     uint8_t backlog_prio;
1102     uint8_t first_group;
1103     uint8_t group_level;
1104     uint8_t pipr_min;
1105     uint8_t lsmfb_min;
1106     uint8_t ring_min;
1107     uint8_t cppr = sig_regs[TM_CPPR];
1108     bool group_enabled;
1109     Xive2Nvp nvp;
1110     int rc;
1111 
1112     g_assert(sig_ring == TM_QW3_HV_PHYS || sig_ring == TM_QW1_OS);
1113     g_assert(!xive_nsr_indicates_group_exception(sig_ring, sig_regs[TM_NSR]));
1114 
1115     /*
1116      * Recompute the PIPR based on local pending interrupts. It will
1117      * be adjusted below if needed in case of pending group interrupts.
1118      */
1119 again:
1120     pipr_min = xive_ipb_to_pipr(sig_regs[TM_IPB]);
1121     group_enabled = !!sig_regs[TM_LGS];
1122     lsmfb_min = group_enabled ? sig_regs[TM_LSMFB] : 0xff;
1123     ring_min = sig_ring;
1124     group_level = 0;
1125 
1126     /* PHYS updates also depend on POOL values */
1127     if (sig_ring == TM_QW3_HV_PHYS) {
1128         uint8_t *pool_regs = &tctx->regs[TM_QW2_HV_POOL];
1129 
1130         /* POOL values only matter if POOL ctx is valid */
1131         if (pool_regs[TM_WORD2] & 0x80) {
1132             uint8_t pool_pipr = xive_ipb_to_pipr(pool_regs[TM_IPB]);
1133             uint8_t pool_lsmfb = pool_regs[TM_LSMFB];
1134 
1135             /*
1136              * Determine highest priority interrupt and
1137              * remember which ring has it.
1138              */
1139             if (pool_pipr < pipr_min) {
1140                 pipr_min = pool_pipr;
1141                 if (pool_pipr < lsmfb_min) {
1142                     ring_min = TM_QW2_HV_POOL;
1143                 }
1144             }
1145 
1146             /* Values needed for group priority calculation */
1147             if (pool_regs[TM_LGS] && (pool_lsmfb < lsmfb_min)) {
1148                 group_enabled = true;
1149                 lsmfb_min = pool_lsmfb;
1150                 if (lsmfb_min < pipr_min) {
1151                     ring_min = TM_QW2_HV_POOL;
1152                 }
1153             }
1154         }
1155     }
1156 
1157     if (group_enabled &&
1158         lsmfb_min < cppr &&
1159         lsmfb_min < pipr_min) {
1160 
1161         uint8_t nvp_blk;
1162         uint32_t nvp_idx;
1163 
1164         /*
1165          * Thread has seen a group interrupt with a higher priority
1166          * than the new cppr or pending local interrupt. Check the
1167          * backlog
1168          */
1169         rc = xive2_tctx_get_nvp_indexes(tctx, ring_min, &nvp_blk, &nvp_idx);
1170         if (rc) {
1171             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: set CPPR on invalid "
1172                                            "context\n");
1173             return;
1174         }
1175 
1176         if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
1177             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
1178                           nvp_blk, nvp_idx);
1179             return;
1180         }
1181 
1182         if (!xive2_nvp_is_valid(&nvp)) {
1183             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1184                           nvp_blk, nvp_idx);
1185             return;
1186         }
1187 
1188         first_group = xive_get_field32(NVP2_W0_PGOFIRST, nvp.w0);
1189         if (!first_group) {
1190             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1191                           nvp_blk, nvp_idx);
1192             return;
1193         }
1194 
1195         backlog_prio = xive2_presenter_backlog_scan(tctx->xptr,
1196                                                     nvp_blk, nvp_idx,
1197                                                     first_group, &group_level);
1198         tctx->regs[ring_min + TM_LSMFB] = backlog_prio;
1199         if (backlog_prio != lsmfb_min) {
1200             /*
1201              * If the group backlog scan finds a less favored or no interrupt,
1202              * then re-do the processing which may turn up a more favored
1203              * interrupt from IPB or the other pool. Backlog should not
1204              * find a priority < LSMFB.
1205              */
1206             g_assert(backlog_prio >= lsmfb_min);
1207             goto again;
1208         }
1209 
1210         xive2_presenter_backlog_decr(tctx->xptr, nvp_blk, nvp_idx,
1211                                      backlog_prio, group_level);
1212         pipr_min = backlog_prio;
1213     }
1214 
1215     if (pipr_min > cppr) {
1216         pipr_min = cppr;
1217     }
1218     xive_tctx_pipr_set(tctx, ring_min, pipr_min, group_level);
1219 }
1220 
1221 /* NOTE: CPPR only exists for TM_QW1_OS and TM_QW3_HV_PHYS */
1222 static void xive2_tctx_set_cppr(XiveTCTX *tctx, uint8_t sig_ring, uint8_t cppr)
1223 {
1224     uint8_t *sig_regs = &tctx->regs[sig_ring];
1225     Xive2Router *xrtr = XIVE2_ROUTER(tctx->xptr);
1226     uint8_t old_cppr;
1227     uint8_t nsr = sig_regs[TM_NSR];
1228 
1229     g_assert(sig_ring == TM_QW1_OS || sig_ring == TM_QW3_HV_PHYS);
1230 
1231     g_assert(tctx->regs[TM_QW2_HV_POOL + TM_NSR] == 0);
1232     g_assert(tctx->regs[TM_QW2_HV_POOL + TM_PIPR] == 0);
1233     g_assert(tctx->regs[TM_QW2_HV_POOL + TM_CPPR] == 0);
1234 
1235     /* XXX: should show pool IPB for PHYS ring */
1236     trace_xive_tctx_set_cppr(tctx->cs->cpu_index, sig_ring,
1237                              sig_regs[TM_IPB], sig_regs[TM_PIPR],
1238                              cppr, nsr);
1239 
1240     if (cppr > XIVE_PRIORITY_MAX) {
1241         cppr = 0xff;
1242     }
1243 
1244     old_cppr = sig_regs[TM_CPPR];
1245     sig_regs[TM_CPPR] = cppr;
1246 
1247     /* Handle increased CPPR priority (lower value) */
1248     if (cppr < old_cppr) {
1249         if (cppr <= sig_regs[TM_PIPR]) {
1250             /* CPPR lowered below PIPR, must un-present interrupt */
1251             if (xive_nsr_indicates_exception(sig_ring, nsr)) {
1252                 if (xive_nsr_indicates_group_exception(sig_ring, nsr)) {
1253                     /* redistribute precluded active grp interrupt */
1254                     xive2_redistribute(xrtr, tctx,
1255                                        xive_nsr_exception_ring(sig_ring, nsr));
1256                     return;
1257                 }
1258             }
1259 
1260             /* interrupt is VP directed, pending in IPB */
1261             xive_tctx_pipr_set(tctx, sig_ring, cppr, 0);
1262             return;
1263         } else {
1264             /* CPPR was lowered, but still above PIPR. No action needed. */
1265             return;
1266         }
1267     }
1268 
1269     /* CPPR didn't change, nothing needs to be done */
1270     if (cppr == old_cppr) {
1271         return;
1272     }
1273 
1274     /* CPPR priority decreased (higher value) */
1275     if (!xive_nsr_indicates_exception(sig_ring, nsr)) {
1276         xive2_tctx_process_pending(tctx, sig_ring);
1277     }
1278 }
1279 
1280 void xive2_tm_set_hv_cppr(XivePresenter *xptr, XiveTCTX *tctx,
1281                           hwaddr offset, uint64_t value, unsigned size)
1282 {
1283     xive2_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
1284 }
1285 
1286 void xive2_tm_set_os_cppr(XivePresenter *xptr, XiveTCTX *tctx,
1287                           hwaddr offset, uint64_t value, unsigned size)
1288 {
1289     xive2_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
1290 }
1291 
1292 static void xive2_tctx_set_target(XiveTCTX *tctx, uint8_t ring, uint8_t target)
1293 {
1294     uint8_t *regs = &tctx->regs[ring];
1295 
1296     regs[TM_T] = target;
1297 }
1298 
1299 void xive2_tm_set_hv_target(XivePresenter *xptr, XiveTCTX *tctx,
1300                             hwaddr offset, uint64_t value, unsigned size)
1301 {
1302     xive2_tctx_set_target(tctx, TM_QW3_HV_PHYS, value & 0xff);
1303 }
1304 
1305 /*
1306  * XIVE Router (aka. Virtualization Controller or IVRE)
1307  */
1308 
1309 int xive2_router_get_eas(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1310                          Xive2Eas *eas)
1311 {
1312     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1313 
1314     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1315 }
1316 
1317 static
1318 int xive2_router_get_pq(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1319                        uint8_t *pq)
1320 {
1321     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1322 
1323     return xrc->get_pq(xrtr, eas_blk, eas_idx, pq);
1324 }
1325 
1326 static
1327 int xive2_router_set_pq(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1328                        uint8_t *pq)
1329 {
1330     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1331 
1332     return xrc->set_pq(xrtr, eas_blk, eas_idx, pq);
1333 }
1334 
1335 int xive2_router_get_end(Xive2Router *xrtr, uint8_t end_blk, uint32_t end_idx,
1336                          Xive2End *end)
1337 {
1338    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1339 
1340    return xrc->get_end(xrtr, end_blk, end_idx, end);
1341 }
1342 
1343 int xive2_router_write_end(Xive2Router *xrtr, uint8_t end_blk, uint32_t end_idx,
1344                            Xive2End *end, uint8_t word_number)
1345 {
1346    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1347 
1348    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1349 }
1350 
1351 int xive2_router_get_nvp(Xive2Router *xrtr, uint8_t nvp_blk, uint32_t nvp_idx,
1352                          Xive2Nvp *nvp)
1353 {
1354    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1355 
1356    return xrc->get_nvp(xrtr, nvp_blk, nvp_idx, nvp);
1357 }
1358 
1359 int xive2_router_write_nvp(Xive2Router *xrtr, uint8_t nvp_blk, uint32_t nvp_idx,
1360                            Xive2Nvp *nvp, uint8_t word_number)
1361 {
1362    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1363 
1364    return xrc->write_nvp(xrtr, nvp_blk, nvp_idx, nvp, word_number);
1365 }
1366 
1367 int xive2_router_get_nvgc(Xive2Router *xrtr, bool crowd,
1368                           uint8_t nvgc_blk, uint32_t nvgc_idx,
1369                           Xive2Nvgc *nvgc)
1370 {
1371    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1372 
1373    return xrc->get_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
1374 }
1375 
1376 int xive2_router_write_nvgc(Xive2Router *xrtr, bool crowd,
1377                             uint8_t nvgc_blk, uint32_t nvgc_idx,
1378                             Xive2Nvgc *nvgc)
1379 {
1380    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1381 
1382    return xrc->write_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
1383 }
1384 
1385 static bool xive2_vp_match_mask(uint32_t cam1, uint32_t cam2,
1386                                 uint32_t vp_mask)
1387 {
1388     return (cam1 & vp_mask) == (cam2 & vp_mask);
1389 }
1390 
1391 static uint8_t xive2_get_vp_block_mask(uint32_t nvt_blk, bool crowd)
1392 {
1393     uint8_t block_mask = 0b1111;
1394 
1395     /* 3 supported crowd sizes: 2, 4, 16 */
1396     if (crowd) {
1397         uint32_t size = xive_get_vpgroup_size(nvt_blk);
1398 
1399         if (size != 2 && size != 4 && size != 16) {
1400             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid crowd size of %d",
1401                                            size);
1402             return block_mask;
1403         }
1404         block_mask &= ~(size - 1);
1405     }
1406     return block_mask;
1407 }
1408 
1409 static uint32_t xive2_get_vp_index_mask(uint32_t nvt_index, bool cam_ignore)
1410 {
1411     uint32_t index_mask = 0xFFFFFF; /* 24 bits */
1412 
1413     if (cam_ignore) {
1414         uint32_t size = xive_get_vpgroup_size(nvt_index);
1415 
1416         if (size < 2) {
1417             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid group size of %d",
1418                                            size);
1419             return index_mask;
1420         }
1421         index_mask &= ~(size - 1);
1422     }
1423     return index_mask;
1424 }
1425 
1426 /*
1427  * The thread context register words are in big-endian format.
1428  */
1429 int xive2_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1430                                uint8_t format,
1431                                uint8_t nvt_blk, uint32_t nvt_idx,
1432                                bool crowd, bool cam_ignore,
1433                                uint32_t logic_serv)
1434 {
1435     uint32_t cam =   xive2_nvp_cam_line(nvt_blk, nvt_idx);
1436     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1437     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1438     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1439     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1440 
1441     uint32_t index_mask, vp_mask;
1442     uint8_t block_mask;
1443 
1444     if (format == 0) {
1445         /*
1446          * i=0: Specific NVT notification
1447          * i=1: VP-group notification (bits ignored at the end of the
1448          *      NVT identifier)
1449          */
1450         block_mask = xive2_get_vp_block_mask(nvt_blk, crowd);
1451         index_mask = xive2_get_vp_index_mask(nvt_idx, cam_ignore);
1452         vp_mask = xive2_nvp_cam_line(block_mask, index_mask);
1453 
1454         /* For VP-group notifications, threads with LGS=0 are excluded */
1455 
1456         /* PHYS ring */
1457         if ((be32_to_cpu(qw3w2) & TM2_QW3W2_VT) &&
1458             !(cam_ignore && tctx->regs[TM_QW3_HV_PHYS + TM_LGS] == 0) &&
1459             xive2_vp_match_mask(cam,
1460                                 xive2_tctx_hw_cam_line(xptr, tctx),
1461                                 vp_mask)) {
1462             return TM_QW3_HV_PHYS;
1463         }
1464 
1465         /* HV POOL ring */
1466         if ((be32_to_cpu(qw2w2) & TM2_QW2W2_VP) &&
1467             !(cam_ignore && tctx->regs[TM_QW2_HV_POOL + TM_LGS] == 0) &&
1468             xive2_vp_match_mask(cam,
1469                                 xive_get_field32(TM2_QW2W2_POOL_CAM, qw2w2),
1470                                 vp_mask)) {
1471             return TM_QW2_HV_POOL;
1472         }
1473 
1474         /* OS ring */
1475         if ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
1476             !(cam_ignore && tctx->regs[TM_QW1_OS + TM_LGS] == 0) &&
1477             xive2_vp_match_mask(cam,
1478                                 xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2),
1479                                 vp_mask)) {
1480             return TM_QW1_OS;
1481         }
1482     } else {
1483         /* F=1 : User level Event-Based Branch (EBB) notification */
1484 
1485         /* FIXME: what if cam_ignore and LGS = 0 ? */
1486         /* USER ring */
1487         if  ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
1488              (cam == xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2)) &&
1489              (be32_to_cpu(qw0w2) & TM2_QW0W2_VU) &&
1490              (logic_serv == xive_get_field32(TM2_QW0W2_LOGIC_SERV, qw0w2))) {
1491             return TM_QW0_USER;
1492         }
1493     }
1494     return -1;
1495 }
1496 
1497 bool xive2_tm_irq_precluded(XiveTCTX *tctx, int ring, uint8_t priority)
1498 {
1499     uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
1500 
1501     /*
1502      * The xive2_presenter_tctx_match() above tells if there's a match
1503      * but for VP-group notification, we still need to look at the
1504      * priority to know if the thread can take the interrupt now or if
1505      * it is precluded.
1506      */
1507     if (priority < sig_regs[TM_PIPR]) {
1508         return false;
1509     }
1510     return true;
1511 }
1512 
1513 void xive2_tm_set_lsmfb(XiveTCTX *tctx, int ring, uint8_t priority)
1514 {
1515     uint8_t *regs = &tctx->regs[ring];
1516 
1517     /*
1518      * Called by the router during a VP-group notification when the
1519      * thread matches but can't take the interrupt because it's
1520      * already running at a more favored priority. It then stores the
1521      * new interrupt priority in the LSMFB field.
1522      */
1523     regs[TM_LSMFB] = priority;
1524 }
1525 
1526 static void xive2_router_realize(DeviceState *dev, Error **errp)
1527 {
1528     Xive2Router *xrtr = XIVE2_ROUTER(dev);
1529 
1530     assert(xrtr->xfb);
1531 }
1532 
1533 /*
1534  * Notification using the END ESe/ESn bit (Event State Buffer for
1535  * escalation and notification). Profide further coalescing in the
1536  * Router.
1537  */
1538 static bool xive2_router_end_es_notify(Xive2Router *xrtr, uint8_t end_blk,
1539                                        uint32_t end_idx, Xive2End *end,
1540                                        uint32_t end_esmask)
1541 {
1542     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1543     bool notify = xive_esb_trigger(&pq);
1544 
1545     if (pq != xive_get_field32(end_esmask, end->w1)) {
1546         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1547         xive2_router_write_end(xrtr, end_blk, end_idx, end, 1);
1548     }
1549 
1550     /* ESe/n[Q]=1 : end of notification */
1551     return notify;
1552 }
1553 
1554 /*
1555  * An END trigger can come from an event trigger (IPI or HW) or from
1556  * another chip. We don't model the PowerBus but the END trigger
1557  * message has the same parameters than in the function below.
1558  */
1559 static void xive2_router_end_notify(Xive2Router *xrtr, uint8_t end_blk,
1560                                     uint32_t end_idx, uint32_t end_data,
1561                                     bool redistribute)
1562 {
1563     Xive2End end;
1564     uint8_t priority;
1565     uint8_t format;
1566     XiveTCTXMatch match;
1567     bool crowd, cam_ignore;
1568     uint8_t nvx_blk;
1569     uint32_t nvx_idx;
1570 
1571     /* END cache lookup */
1572     if (xive2_router_get_end(xrtr, end_blk, end_idx, &end)) {
1573         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1574                       end_idx);
1575         return;
1576     }
1577 
1578     if (!xive2_end_is_valid(&end)) {
1579         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1580                       end_blk, end_idx);
1581         return;
1582     }
1583 
1584     if (xive2_end_is_crowd(&end) && !xive2_end_is_ignore(&end)) {
1585         qemu_log_mask(LOG_GUEST_ERROR,
1586                       "XIVE: invalid END, 'crowd' bit requires 'ignore' bit\n");
1587         return;
1588     }
1589 
1590     if (!redistribute && xive2_end_is_enqueue(&end)) {
1591         trace_xive_end_enqueue(end_blk, end_idx, end_data);
1592         xive2_end_enqueue(&end, end_data);
1593         /* Enqueuing event data modifies the EQ toggle and index */
1594         xive2_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1595     }
1596 
1597     /*
1598      * When the END is silent, we skip the notification part.
1599      */
1600     if (xive2_end_is_silent_escalation(&end)) {
1601         goto do_escalation;
1602     }
1603 
1604     /*
1605      * The W7 format depends on the F bit in W6. It defines the type
1606      * of the notification :
1607      *
1608      *   F=0 : single or multiple NVP notification
1609      *   F=1 : User level Event-Based Branch (EBB) notification, no
1610      *         priority
1611      */
1612     format = xive_get_field32(END2_W6_FORMAT_BIT, end.w6);
1613     priority = xive_get_field32(END2_W7_F0_PRIORITY, end.w7);
1614 
1615     /* The END is masked */
1616     if (format == 0 && priority == 0xff) {
1617         return;
1618     }
1619 
1620     /*
1621      * Check the END ESn (Event State Buffer for notification) for
1622      * even further coalescing in the Router
1623      */
1624     if (!xive2_end_is_notify(&end)) {
1625         /* ESn[Q]=1 : end of notification */
1626         if (!xive2_router_end_es_notify(xrtr, end_blk, end_idx,
1627                                        &end, END2_W1_ESn)) {
1628             return;
1629         }
1630     }
1631 
1632     /*
1633      * Follows IVPE notification
1634      */
1635     nvx_blk = xive_get_field32(END2_W6_VP_BLOCK, end.w6);
1636     nvx_idx = xive_get_field32(END2_W6_VP_OFFSET, end.w6);
1637     crowd = xive2_end_is_crowd(&end);
1638     cam_ignore = xive2_end_is_ignore(&end);
1639 
1640     /* TODO: Auto EOI. */
1641     if (xive_presenter_match(xrtr->xfb, format, nvx_blk, nvx_idx,
1642                              crowd, cam_ignore, priority,
1643                              xive_get_field32(END2_W7_F1_LOG_SERVER_ID, end.w7),
1644                              &match)) {
1645         XiveTCTX *tctx = match.tctx;
1646         uint8_t ring = match.ring;
1647         uint8_t *sig_regs = xive_tctx_signal_regs(tctx, ring);
1648         uint8_t nsr = sig_regs[TM_NSR];
1649         uint8_t group_level;
1650 
1651         if (priority < sig_regs[TM_PIPR] &&
1652             xive_nsr_indicates_group_exception(ring, nsr)) {
1653             xive2_redistribute(xrtr, tctx, xive_nsr_exception_ring(ring, nsr));
1654         }
1655 
1656         group_level = xive_get_group_level(crowd, cam_ignore, nvx_blk, nvx_idx);
1657         trace_xive_presenter_notify(nvx_blk, nvx_idx, ring, group_level);
1658         xive_tctx_pipr_present(tctx, ring, priority, group_level);
1659         return;
1660     }
1661 
1662     /*
1663      * If no matching NVP is dispatched on a HW thread :
1664      * - specific VP: update the NVP structure if backlog is activated
1665      * - VP-group: update the backlog counter for that priority in the NVG
1666      */
1667     if (xive2_end_is_backlog(&end)) {
1668 
1669         if (format == 1) {
1670             qemu_log_mask(LOG_GUEST_ERROR,
1671                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1672                           end_blk, end_idx);
1673             return;
1674         }
1675 
1676         if (!cam_ignore) {
1677             uint8_t ipb;
1678             Xive2Nvp nvp;
1679 
1680             /* NVP cache lookup */
1681             if (xive2_router_get_nvp(xrtr, nvx_blk, nvx_idx, &nvp)) {
1682                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVP %x/%x\n",
1683                               nvx_blk, nvx_idx);
1684                 return;
1685             }
1686 
1687             if (!xive2_nvp_is_valid(&nvp)) {
1688                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is invalid\n",
1689                               nvx_blk, nvx_idx);
1690                 return;
1691             }
1692 
1693             /*
1694              * Record the IPB in the associated NVP structure for later
1695              * use. The presenter will resend the interrupt when the vCPU
1696              * is dispatched again on a HW thread.
1697              */
1698             ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2) |
1699                 xive_priority_to_ipb(priority);
1700             nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, ipb);
1701             xive2_router_write_nvp(xrtr, nvx_blk, nvx_idx, &nvp, 2);
1702         } else {
1703             Xive2Nvgc nvgc;
1704             uint32_t backlog;
1705 
1706             /*
1707              * For groups and crowds, the per-priority backlog
1708              * counters are stored in the NVG/NVC structures
1709              */
1710             if (xive2_router_get_nvgc(xrtr, crowd,
1711                                       nvx_blk, nvx_idx, &nvgc)) {
1712                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no %s %x/%x\n",
1713                               crowd ? "NVC" : "NVG", nvx_blk, nvx_idx);
1714                 return;
1715             }
1716 
1717             if (!xive2_nvgc_is_valid(&nvgc)) {
1718                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVG %x/%x is invalid\n",
1719                               nvx_blk, nvx_idx);
1720                 return;
1721             }
1722 
1723             /*
1724              * Increment the backlog counter for that priority.
1725              * We only call broadcast the first time the counter is
1726              * incremented. broadcast will set the LSMFB field of the TIMA of
1727              * relevant threads so that they know an interrupt is pending.
1728              */
1729             backlog = xive2_nvgc_get_backlog(&nvgc, priority) + 1;
1730             xive2_nvgc_set_backlog(&nvgc, priority, backlog);
1731             xive2_router_write_nvgc(xrtr, crowd, nvx_blk, nvx_idx, &nvgc);
1732 
1733             if (backlog == 1) {
1734                 XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xrtr->xfb);
1735                 xfc->broadcast(xrtr->xfb, nvx_blk, nvx_idx,
1736                                crowd, cam_ignore, priority);
1737 
1738                 if (!xive2_end_is_precluded_escalation(&end)) {
1739                     /*
1740                      * The interrupt will be picked up when the
1741                      * matching thread lowers its priority level
1742                      */
1743                     return;
1744                 }
1745             }
1746         }
1747     }
1748 
1749 do_escalation:
1750     /*
1751      * If activated, escalate notification using the ESe PQ bits and
1752      * the EAS in w4-5
1753      */
1754     if (!xive2_end_is_escalate(&end)) {
1755         return;
1756     }
1757 
1758     /*
1759      * Check the END ESe (Event State Buffer for escalation) for even
1760      * further coalescing in the Router
1761      */
1762     if (!xive2_end_is_uncond_escalation(&end)) {
1763         /* ESe[Q]=1 : end of escalation notification */
1764         if (!xive2_router_end_es_notify(xrtr, end_blk, end_idx,
1765                                        &end, END2_W1_ESe)) {
1766             return;
1767         }
1768     }
1769 
1770     if (xive2_end_is_escalate_end(&end)) {
1771         /*
1772          * Perform END Adaptive escalation processing
1773          * The END trigger becomes an Escalation trigger
1774          */
1775         uint8_t esc_blk = xive_get_field32(END2_W4_END_BLOCK, end.w4);
1776         uint32_t esc_idx = xive_get_field32(END2_W4_ESC_END_INDEX, end.w4);
1777         uint32_t esc_data = xive_get_field32(END2_W5_ESC_END_DATA, end.w5);
1778         trace_xive_escalate_end(end_blk, end_idx, esc_blk, esc_idx, esc_data);
1779         xive2_router_end_notify(xrtr, esc_blk, esc_idx, esc_data, false);
1780     } /* end END adaptive escalation */
1781 
1782     else {
1783         uint32_t lisn;              /* Logical Interrupt Source Number */
1784 
1785         /*
1786          *  Perform ESB escalation processing
1787          *      E[N] == 1 --> N
1788          *      Req[Block] <- E[ESB_Block]
1789          *      Req[Index] <- E[ESB_Index]
1790          *      Req[Offset] <- 0x000
1791          *      Execute <ESB Store> Req command
1792          */
1793         lisn = XIVE_EAS(xive_get_field32(END2_W4_END_BLOCK,     end.w4),
1794                         xive_get_field32(END2_W4_ESC_END_INDEX, end.w4));
1795 
1796         trace_xive_escalate_esb(end_blk, end_idx, lisn);
1797         xive2_notify(xrtr, lisn, true /* pq_checked */);
1798     }
1799 
1800     return;
1801 }
1802 
1803 void xive2_notify(Xive2Router *xrtr , uint32_t lisn, bool pq_checked)
1804 {
1805     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1806     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1807     Xive2Eas eas;
1808 
1809     /* EAS cache lookup */
1810     if (xive2_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1811         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1812         return;
1813     }
1814 
1815     if (!pq_checked) {
1816         bool notify;
1817         uint8_t pq;
1818 
1819         /* PQ cache lookup */
1820         if (xive2_router_get_pq(xrtr, eas_blk, eas_idx, &pq)) {
1821             /* Set FIR */
1822             g_assert_not_reached();
1823         }
1824 
1825         notify = xive_esb_trigger(&pq);
1826 
1827         if (xive2_router_set_pq(xrtr, eas_blk, eas_idx, &pq)) {
1828             /* Set FIR */
1829             g_assert_not_reached();
1830         }
1831 
1832         if (!notify) {
1833             return;
1834         }
1835     }
1836 
1837     if (!xive2_eas_is_valid(&eas)) {
1838         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN %x\n", lisn);
1839         return;
1840     }
1841 
1842     if (xive2_eas_is_masked(&eas)) {
1843         /* Notification completed */
1844         return;
1845     }
1846 
1847     /* TODO: add support for EAS resume */
1848     if (xive2_eas_is_resume(&eas)) {
1849         qemu_log_mask(LOG_UNIMP,
1850                       "XIVE: EAS resume processing unimplemented - LISN %x\n",
1851                       lisn);
1852         return;
1853     }
1854 
1855     /*
1856      * The event trigger becomes an END trigger
1857      */
1858     xive2_router_end_notify(xrtr,
1859                             xive_get_field64(EAS2_END_BLOCK, eas.w),
1860                             xive_get_field64(EAS2_END_INDEX, eas.w),
1861                             xive_get_field64(EAS2_END_DATA,  eas.w),
1862                             false);
1863     return;
1864 }
1865 
1866 void xive2_router_notify(XiveNotifier *xn, uint32_t lisn, bool pq_checked)
1867 {
1868     Xive2Router *xrtr = XIVE2_ROUTER(xn);
1869 
1870     xive2_notify(xrtr, lisn, pq_checked);
1871     return;
1872 }
1873 
1874 static const Property xive2_router_properties[] = {
1875     DEFINE_PROP_LINK("xive-fabric", Xive2Router, xfb,
1876                      TYPE_XIVE_FABRIC, XiveFabric *),
1877 };
1878 
1879 static void xive2_router_class_init(ObjectClass *klass, const void *data)
1880 {
1881     DeviceClass *dc = DEVICE_CLASS(klass);
1882     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1883 
1884     dc->desc    = "XIVE2 Router Engine";
1885     device_class_set_props(dc, xive2_router_properties);
1886     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1887     dc->realize = xive2_router_realize;
1888     xnc->notify = xive2_router_notify;
1889 }
1890 
1891 static const TypeInfo xive2_router_info = {
1892     .name          = TYPE_XIVE2_ROUTER,
1893     .parent        = TYPE_SYS_BUS_DEVICE,
1894     .abstract      = true,
1895     .instance_size = sizeof(Xive2Router),
1896     .class_size    = sizeof(Xive2RouterClass),
1897     .class_init    = xive2_router_class_init,
1898     .interfaces    = (const InterfaceInfo[]) {
1899         { TYPE_XIVE_NOTIFIER },
1900         { TYPE_XIVE_PRESENTER },
1901         { }
1902     }
1903 };
1904 
1905 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
1906 {
1907     return !((addr >> shift) & 1);
1908 }
1909 
1910 static uint64_t xive2_end_source_read(void *opaque, hwaddr addr, unsigned size)
1911 {
1912     Xive2EndSource *xsrc = XIVE2_END_SOURCE(opaque);
1913     uint32_t offset = addr & 0xFFF;
1914     uint8_t end_blk;
1915     uint32_t end_idx;
1916     Xive2End end;
1917     uint32_t end_esmask;
1918     uint8_t pq;
1919     uint64_t ret;
1920 
1921     /*
1922      * The block id should be deduced from the load address on the END
1923      * ESB MMIO but our model only supports a single block per XIVE chip.
1924      */
1925     end_blk = xive2_router_get_block_id(xsrc->xrtr);
1926     end_idx = addr >> (xsrc->esb_shift + 1);
1927 
1928     if (xive2_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1929         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1930                       end_idx);
1931         return -1;
1932     }
1933 
1934     if (!xive2_end_is_valid(&end)) {
1935         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1936                       end_blk, end_idx);
1937         return -1;
1938     }
1939 
1940     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END2_W1_ESn :
1941         END2_W1_ESe;
1942     pq = xive_get_field32(end_esmask, end.w1);
1943 
1944     switch (offset) {
1945     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1946         ret = xive_esb_eoi(&pq);
1947 
1948         /* Forward the source event notification for routing ?? */
1949         break;
1950 
1951     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1952         ret = pq;
1953         break;
1954 
1955     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1956     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1957     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1958     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1959         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1960         break;
1961     default:
1962         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1963                       offset);
1964         return -1;
1965     }
1966 
1967     if (pq != xive_get_field32(end_esmask, end.w1)) {
1968         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1969         xive2_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1970     }
1971 
1972     return ret;
1973 }
1974 
1975 static void xive2_end_source_write(void *opaque, hwaddr addr,
1976                                    uint64_t value, unsigned size)
1977 {
1978     Xive2EndSource *xsrc = XIVE2_END_SOURCE(opaque);
1979     uint32_t offset = addr & 0xFFF;
1980     uint8_t end_blk;
1981     uint32_t end_idx;
1982     Xive2End end;
1983     uint32_t end_esmask;
1984     uint8_t pq;
1985     bool notify = false;
1986 
1987     /*
1988      * The block id should be deduced from the load address on the END
1989      * ESB MMIO but our model only supports a single block per XIVE chip.
1990      */
1991     end_blk = xive2_router_get_block_id(xsrc->xrtr);
1992     end_idx = addr >> (xsrc->esb_shift + 1);
1993 
1994     if (xive2_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1995         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1996                       end_idx);
1997         return;
1998     }
1999 
2000     if (!xive2_end_is_valid(&end)) {
2001         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
2002                       end_blk, end_idx);
2003         return;
2004     }
2005 
2006     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END2_W1_ESn :
2007         END2_W1_ESe;
2008     pq = xive_get_field32(end_esmask, end.w1);
2009 
2010     switch (offset) {
2011     case 0 ... 0x3FF:
2012         notify = xive_esb_trigger(&pq);
2013         break;
2014 
2015     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
2016         /* TODO: can we check StoreEOI availability from the router ? */
2017         notify = xive_esb_eoi(&pq);
2018         break;
2019 
2020     case XIVE_ESB_INJECT ... XIVE_ESB_INJECT + 0x3FF:
2021         if (end_esmask == END2_W1_ESe) {
2022             qemu_log_mask(LOG_GUEST_ERROR,
2023                           "XIVE: END %x/%x can not EQ inject on ESe\n",
2024                            end_blk, end_idx);
2025             return;
2026         }
2027         notify = true;
2028         break;
2029 
2030     default:
2031         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB write addr %d\n",
2032                       offset);
2033         return;
2034     }
2035 
2036     if (pq != xive_get_field32(end_esmask, end.w1)) {
2037         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
2038         xive2_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
2039     }
2040 
2041     /* TODO: Forward the source event notification for routing */
2042     if (notify) {
2043         ;
2044     }
2045 }
2046 
2047 static const MemoryRegionOps xive2_end_source_ops = {
2048     .read = xive2_end_source_read,
2049     .write = xive2_end_source_write,
2050     .endianness = DEVICE_BIG_ENDIAN,
2051     .valid = {
2052         .min_access_size = 1,
2053         .max_access_size = 8,
2054     },
2055     .impl = {
2056         .min_access_size = 1,
2057         .max_access_size = 8,
2058     },
2059 };
2060 
2061 static void xive2_end_source_realize(DeviceState *dev, Error **errp)
2062 {
2063     Xive2EndSource *xsrc = XIVE2_END_SOURCE(dev);
2064 
2065     assert(xsrc->xrtr);
2066 
2067     if (!xsrc->nr_ends) {
2068         error_setg(errp, "Number of interrupt needs to be greater than 0");
2069         return;
2070     }
2071 
2072     if (xsrc->esb_shift != XIVE_ESB_4K &&
2073         xsrc->esb_shift != XIVE_ESB_64K) {
2074         error_setg(errp, "Invalid ESB shift setting");
2075         return;
2076     }
2077 
2078     /*
2079      * Each END is assigned an even/odd pair of MMIO pages, the even page
2080      * manages the ESn field while the odd page manages the ESe field.
2081      */
2082     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
2083                           &xive2_end_source_ops, xsrc, "xive.end",
2084                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
2085 }
2086 
2087 static const Property xive2_end_source_properties[] = {
2088     DEFINE_PROP_UINT32("nr-ends", Xive2EndSource, nr_ends, 0),
2089     DEFINE_PROP_UINT32("shift", Xive2EndSource, esb_shift, XIVE_ESB_64K),
2090     DEFINE_PROP_LINK("xive", Xive2EndSource, xrtr, TYPE_XIVE2_ROUTER,
2091                      Xive2Router *),
2092 };
2093 
2094 static void xive2_end_source_class_init(ObjectClass *klass, const void *data)
2095 {
2096     DeviceClass *dc = DEVICE_CLASS(klass);
2097 
2098     dc->desc    = "XIVE END Source";
2099     device_class_set_props(dc, xive2_end_source_properties);
2100     dc->realize = xive2_end_source_realize;
2101     dc->user_creatable = false;
2102 }
2103 
2104 static const TypeInfo xive2_end_source_info = {
2105     .name          = TYPE_XIVE2_END_SOURCE,
2106     .parent        = TYPE_DEVICE,
2107     .instance_size = sizeof(Xive2EndSource),
2108     .class_init    = xive2_end_source_class_init,
2109 };
2110 
2111 static void xive2_register_types(void)
2112 {
2113     type_register_static(&xive2_router_info);
2114     type_register_static(&xive2_end_source_info);
2115 }
2116 
2117 type_init(xive2_register_types)
2118