xref: /openbmc/qemu/hw/intc/xive2.c (revision 9ad30401ce9aefa319364cb8efdc6893a5bc20ad)
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 uint32_t xive2_router_get_config(Xive2Router *xrtr)
27 {
28     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
29 
30     return xrc->get_config(xrtr);
31 }
32 
33 static int xive2_router_get_block_id(Xive2Router *xrtr)
34 {
35    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
36 
37    return xrc->get_block_id(xrtr);
38 }
39 
40 static uint64_t xive2_nvp_reporting_addr(Xive2Nvp *nvp)
41 {
42     uint64_t cache_addr;
43 
44     cache_addr = xive_get_field32(NVP2_W6_REPORTING_LINE, nvp->w6) << 24 |
45         xive_get_field32(NVP2_W7_REPORTING_LINE, nvp->w7);
46     cache_addr <<= 8; /* aligned on a cache line pair */
47     return cache_addr;
48 }
49 
50 static uint32_t xive2_nvgc_get_backlog(Xive2Nvgc *nvgc, uint8_t priority)
51 {
52     uint32_t val = 0;
53     uint8_t *ptr, i;
54 
55     if (priority > 7) {
56         return 0;
57     }
58 
59     /*
60      * The per-priority backlog counters are 24-bit and the structure
61      * is stored in big endian. NVGC is 32-bytes long, so 24-bytes from
62      * w2, which fits 8 priorities * 24-bits per priority.
63      */
64     ptr = (uint8_t *)&nvgc->w2 + priority * 3;
65     for (i = 0; i < 3; i++, ptr++) {
66         val = (val << 8) + *ptr;
67     }
68     return val;
69 }
70 
71 static void xive2_nvgc_set_backlog(Xive2Nvgc *nvgc, uint8_t priority,
72                                    uint32_t val)
73 {
74     uint8_t *ptr, i;
75     uint32_t shift;
76 
77     if (priority > 7) {
78         return;
79     }
80 
81     if (val > 0xFFFFFF) {
82         val = 0xFFFFFF;
83     }
84     /*
85      * The per-priority backlog counters are 24-bit and the structure
86      * is stored in big endian
87      */
88     ptr = (uint8_t *)&nvgc->w2 + priority * 3;
89     for (i = 0; i < 3; i++, ptr++) {
90         shift = 8 * (2 - i);
91         *ptr = (val >> shift) & 0xFF;
92     }
93 }
94 
95 uint64_t xive2_presenter_nvgc_backlog_op(XivePresenter *xptr,
96                                          bool crowd,
97                                          uint8_t blk, uint32_t idx,
98                                          uint16_t offset, uint16_t val)
99 {
100     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
101     uint8_t priority = GETFIELD(NVx_BACKLOG_PRIO, offset);
102     uint8_t op = GETFIELD(NVx_BACKLOG_OP, offset);
103     Xive2Nvgc nvgc;
104     uint32_t count, old_count;
105 
106     if (xive2_router_get_nvgc(xrtr, crowd, blk, idx, &nvgc)) {
107         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No %s %x/%x\n",
108                       crowd ? "NVC" : "NVG", blk, idx);
109         return -1;
110     }
111     if (!xive2_nvgc_is_valid(&nvgc)) {
112         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid NVG %x/%x\n", blk, idx);
113         return -1;
114     }
115 
116     old_count = xive2_nvgc_get_backlog(&nvgc, priority);
117     count = old_count;
118     /*
119      * op:
120      * 0b00 => increment
121      * 0b01 => decrement
122      * 0b1- => read
123      */
124     if (op == 0b00 || op == 0b01) {
125         if (op == 0b00) {
126             count += val;
127         } else {
128             if (count > val) {
129                 count -= val;
130             } else {
131                 count = 0;
132             }
133         }
134         xive2_nvgc_set_backlog(&nvgc, priority, count);
135         xive2_router_write_nvgc(xrtr, crowd, blk, idx, &nvgc);
136     }
137     trace_xive_nvgc_backlog_op(crowd, blk, idx, op, priority, old_count);
138     return old_count;
139 }
140 
141 uint64_t xive2_presenter_nvp_backlog_op(XivePresenter *xptr,
142                                         uint8_t blk, uint32_t idx,
143                                         uint16_t offset)
144 {
145     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
146     uint8_t priority = GETFIELD(NVx_BACKLOG_PRIO, offset);
147     uint8_t op = GETFIELD(NVx_BACKLOG_OP, offset);
148     Xive2Nvp nvp;
149     uint8_t ipb, old_ipb, rc;
150 
151     if (xive2_router_get_nvp(xrtr, blk, idx, &nvp)) {
152         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n", blk, idx);
153         return -1;
154     }
155     if (!xive2_nvp_is_valid(&nvp)) {
156         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid NVP %x/%x\n", blk, idx);
157         return -1;
158     }
159 
160     old_ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2);
161     ipb = old_ipb;
162     /*
163      * op:
164      * 0b00 => set priority bit
165      * 0b01 => reset priority bit
166      * 0b1- => read
167      */
168     if (op == 0b00 || op == 0b01) {
169         if (op == 0b00) {
170             ipb |= xive_priority_to_ipb(priority);
171         } else {
172             ipb &= ~xive_priority_to_ipb(priority);
173         }
174         nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, ipb);
175         xive2_router_write_nvp(xrtr, blk, idx, &nvp, 2);
176     }
177     rc = !!(old_ipb & xive_priority_to_ipb(priority));
178     trace_xive_nvp_backlog_op(blk, idx, op, priority, rc);
179     return rc;
180 }
181 
182 void xive2_eas_pic_print_info(Xive2Eas *eas, uint32_t lisn, GString *buf)
183 {
184     if (!xive2_eas_is_valid(eas)) {
185         return;
186     }
187 
188     g_string_append_printf(buf, "  %08x %s end:%02x/%04x data:%08x\n",
189                            lisn, xive2_eas_is_masked(eas) ? "M" : " ",
190                            (uint8_t)  xive_get_field64(EAS2_END_BLOCK, eas->w),
191                            (uint32_t) xive_get_field64(EAS2_END_INDEX, eas->w),
192                            (uint32_t) xive_get_field64(EAS2_END_DATA, eas->w));
193 }
194 
195 #define XIVE2_QSIZE_CHUNK_CL    128
196 #define XIVE2_QSIZE_CHUNK_4k   4096
197 /* Calculate max number of queue entries for an END */
198 static uint32_t xive2_end_get_qentries(Xive2End *end)
199 {
200     uint32_t w3 = end->w3;
201     uint32_t qsize = xive_get_field32(END2_W3_QSIZE, w3);
202     if (xive_get_field32(END2_W3_CL, w3)) {
203         g_assert(qsize <= 4);
204         return (XIVE2_QSIZE_CHUNK_CL << qsize) / sizeof(uint32_t);
205     } else {
206         g_assert(qsize <= 12);
207         return (XIVE2_QSIZE_CHUNK_4k << qsize) / sizeof(uint32_t);
208     }
209 }
210 
211 void xive2_end_queue_pic_print_info(Xive2End *end, uint32_t width, GString *buf)
212 {
213     uint64_t qaddr_base = xive2_end_qaddr(end);
214     uint32_t qindex = xive_get_field32(END2_W1_PAGE_OFF, end->w1);
215     uint32_t qentries = xive2_end_get_qentries(end);
216     int i;
217 
218     /*
219      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
220      */
221     g_string_append_printf(buf, " [ ");
222     qindex = (qindex - (width - 1)) & (qentries - 1);
223     for (i = 0; i < width; i++) {
224         uint64_t qaddr = qaddr_base + (qindex << 2);
225         uint32_t qdata = -1;
226 
227         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
228                             sizeof(qdata), MEMTXATTRS_UNSPECIFIED)) {
229             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
230                           HWADDR_PRIx "\n", qaddr);
231             return;
232         }
233         g_string_append_printf(buf, "%s%08x ", i == width - 1 ? "^" : "",
234                                be32_to_cpu(qdata));
235         qindex = (qindex + 1) & (qentries - 1);
236     }
237     g_string_append_printf(buf, "]");
238 }
239 
240 void xive2_end_pic_print_info(Xive2End *end, uint32_t end_idx, GString *buf)
241 {
242     uint64_t qaddr_base = xive2_end_qaddr(end);
243     uint32_t qindex = xive_get_field32(END2_W1_PAGE_OFF, end->w1);
244     uint32_t qgen = xive_get_field32(END2_W1_GENERATION, end->w1);
245     uint32_t qentries = xive2_end_get_qentries(end);
246 
247     uint32_t nvx_blk = xive_get_field32(END2_W6_VP_BLOCK, end->w6);
248     uint32_t nvx_idx = xive_get_field32(END2_W6_VP_OFFSET, end->w6);
249     uint8_t priority = xive_get_field32(END2_W7_F0_PRIORITY, end->w7);
250     uint8_t pq;
251 
252     if (!xive2_end_is_valid(end)) {
253         return;
254     }
255 
256     pq = xive_get_field32(END2_W1_ESn, end->w1);
257 
258     g_string_append_printf(buf,
259                            "  %08x %c%c %c%c%c%c%c%c%c%c%c%c%c %c%c "
260                            "prio:%d nvp:%02x/%04x",
261                            end_idx,
262                            pq & XIVE_ESB_VAL_P ? 'P' : '-',
263                            pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
264                            xive2_end_is_valid(end)    ? 'v' : '-',
265                            xive2_end_is_enqueue(end)  ? 'q' : '-',
266                            xive2_end_is_notify(end)   ? 'n' : '-',
267                            xive2_end_is_backlog(end)  ? 'b' : '-',
268                            xive2_end_is_precluded_escalation(end) ? 'p' : '-',
269                            xive2_end_is_escalate(end) ? 'e' : '-',
270                            xive2_end_is_escalate_end(end) ? 'N' : '-',
271                            xive2_end_is_uncond_escalation(end)   ? 'u' : '-',
272                            xive2_end_is_silent_escalation(end)   ? 's' : '-',
273                            xive2_end_is_firmware1(end)   ? 'f' : '-',
274                            xive2_end_is_firmware2(end)   ? 'F' : '-',
275                            xive2_end_is_ignore(end) ? 'i' : '-',
276                            xive2_end_is_crowd(end)  ? 'c' : '-',
277                            priority, nvx_blk, nvx_idx);
278 
279     if (qaddr_base) {
280         g_string_append_printf(buf, " eq:@%08"PRIx64"% 6d/%5d ^%d",
281                                qaddr_base, qindex, qentries, qgen);
282         xive2_end_queue_pic_print_info(end, 6, buf);
283     }
284     g_string_append_c(buf, '\n');
285 }
286 
287 void xive2_end_eas_pic_print_info(Xive2End *end, uint32_t end_idx,
288                                   GString *buf)
289 {
290     Xive2Eas *eas = (Xive2Eas *) &end->w4;
291     uint8_t pq;
292 
293     if (!xive2_end_is_escalate(end)) {
294         return;
295     }
296 
297     pq = xive_get_field32(END2_W1_ESe, end->w1);
298 
299     g_string_append_printf(buf, "  %08x %c%c %c%c end:%02x/%04x data:%08x\n",
300                            end_idx,
301                            pq & XIVE_ESB_VAL_P ? 'P' : '-',
302                            pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
303                            xive2_eas_is_valid(eas) ? 'v' : ' ',
304                            xive2_eas_is_masked(eas) ? 'M' : ' ',
305                            (uint8_t)  xive_get_field64(EAS2_END_BLOCK, eas->w),
306                            (uint32_t) xive_get_field64(EAS2_END_INDEX, eas->w),
307                            (uint32_t) xive_get_field64(EAS2_END_DATA, eas->w));
308 }
309 
310 void xive2_nvp_pic_print_info(Xive2Nvp *nvp, uint32_t nvp_idx, GString *buf)
311 {
312     uint8_t  eq_blk = xive_get_field32(NVP2_W5_VP_END_BLOCK, nvp->w5);
313     uint32_t eq_idx = xive_get_field32(NVP2_W5_VP_END_INDEX, nvp->w5);
314     uint64_t cache_line = xive2_nvp_reporting_addr(nvp);
315 
316     if (!xive2_nvp_is_valid(nvp)) {
317         return;
318     }
319 
320     g_string_append_printf(buf, "  %08x end:%02x/%04x IPB:%02x PGoFirst:%02x",
321                            nvp_idx, eq_blk, eq_idx,
322                            xive_get_field32(NVP2_W2_IPB, nvp->w2),
323                            xive_get_field32(NVP2_W0_PGOFIRST, nvp->w0));
324     if (cache_line) {
325         g_string_append_printf(buf, "  reporting CL:%016"PRIx64, cache_line);
326     }
327 
328     /*
329      * When the NVP is HW controlled, more fields are updated
330      */
331     if (xive2_nvp_is_hw(nvp)) {
332         g_string_append_printf(buf, " CPPR:%02x",
333                                xive_get_field32(NVP2_W2_CPPR, nvp->w2));
334         if (xive2_nvp_is_co(nvp)) {
335             g_string_append_printf(buf, " CO:%04x",
336                                    xive_get_field32(NVP2_W1_CO_THRID, nvp->w1));
337         }
338     }
339     g_string_append_c(buf, '\n');
340 }
341 
342 void xive2_nvgc_pic_print_info(Xive2Nvgc *nvgc, uint32_t nvgc_idx, GString *buf)
343 {
344     uint8_t i;
345 
346     if (!xive2_nvgc_is_valid(nvgc)) {
347         return;
348     }
349 
350     g_string_append_printf(buf, "  %08x PGoNext:%02x bklog: ", nvgc_idx,
351                            xive_get_field32(NVGC2_W0_PGONEXT, nvgc->w0));
352     for (i = 0; i <= XIVE_PRIORITY_MAX; i++) {
353         g_string_append_printf(buf, "[%d]=0x%x ",
354                                i, xive2_nvgc_get_backlog(nvgc, i));
355     }
356     g_string_append_printf(buf, "\n");
357 }
358 
359 static void xive2_end_enqueue(Xive2End *end, uint32_t data)
360 {
361     uint64_t qaddr_base = xive2_end_qaddr(end);
362     uint32_t qindex = xive_get_field32(END2_W1_PAGE_OFF, end->w1);
363     uint32_t qgen = xive_get_field32(END2_W1_GENERATION, end->w1);
364 
365     uint64_t qaddr = qaddr_base + (qindex << 2);
366     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
367     uint32_t qentries = xive2_end_get_qentries(end);
368 
369     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata),
370                          MEMTXATTRS_UNSPECIFIED)) {
371         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
372                       HWADDR_PRIx "\n", qaddr);
373         return;
374     }
375 
376     qindex = (qindex + 1) & (qentries - 1);
377     if (qindex == 0) {
378         qgen ^= 1;
379         end->w1 = xive_set_field32(END2_W1_GENERATION, end->w1, qgen);
380 
381         /* Set gen flipped to 1, it gets reset on a cache watch operation */
382         end->w1 = xive_set_field32(END2_W1_GEN_FLIPPED, end->w1, 1);
383     }
384     end->w1 = xive_set_field32(END2_W1_PAGE_OFF, end->w1, qindex);
385 }
386 
387 static void xive2_pgofnext(uint8_t *nvgc_blk, uint32_t *nvgc_idx,
388                            uint8_t next_level)
389 {
390     uint32_t mask, next_idx;
391     uint8_t next_blk;
392 
393     /*
394      * Adjust the block and index of a VP for the next group/crowd
395      * size (PGofFirst/PGofNext field in the NVP and NVGC structures).
396      *
397      * The 6-bit group level is split into a 2-bit crowd and 4-bit
398      * group levels. Encoding is similar. However, we don't support
399      * crowd size of 8. So a crowd level of 0b11 is bumped to a crowd
400      * size of 16.
401      */
402     next_blk = NVx_CROWD_LVL(next_level);
403     if (next_blk == 3) {
404         next_blk = 4;
405     }
406     mask = (1 << next_blk) - 1;
407     *nvgc_blk &= ~mask;
408     *nvgc_blk |= mask >> 1;
409 
410     next_idx = NVx_GROUP_LVL(next_level);
411     mask = (1 << next_idx) - 1;
412     *nvgc_idx &= ~mask;
413     *nvgc_idx |= mask >> 1;
414 }
415 
416 /*
417  * Scan the group chain and return the highest priority and group
418  * level of pending group interrupts.
419  */
420 static uint8_t xive2_presenter_backlog_scan(XivePresenter *xptr,
421                                             uint8_t nvx_blk, uint32_t nvx_idx,
422                                             uint8_t first_group,
423                                             uint8_t *out_level)
424 {
425     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
426     uint32_t nvgc_idx;
427     uint32_t current_level, count;
428     uint8_t nvgc_blk, prio;
429     Xive2Nvgc nvgc;
430 
431     for (prio = 0; prio <= XIVE_PRIORITY_MAX; prio++) {
432         current_level = first_group & 0x3F;
433         nvgc_blk = nvx_blk;
434         nvgc_idx = nvx_idx;
435 
436         while (current_level) {
437             xive2_pgofnext(&nvgc_blk, &nvgc_idx, current_level);
438 
439             if (xive2_router_get_nvgc(xrtr, NVx_CROWD_LVL(current_level),
440                                       nvgc_blk, nvgc_idx, &nvgc)) {
441                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVGC %x/%x\n",
442                               nvgc_blk, nvgc_idx);
443                 return 0xFF;
444             }
445             if (!xive2_nvgc_is_valid(&nvgc)) {
446                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid NVGC %x/%x\n",
447                               nvgc_blk, nvgc_idx);
448                 return 0xFF;
449             }
450 
451             count = xive2_nvgc_get_backlog(&nvgc, prio);
452             if (count) {
453                 *out_level = current_level;
454                 return prio;
455             }
456             current_level = xive_get_field32(NVGC2_W0_PGONEXT, nvgc.w0) & 0x3F;
457         }
458     }
459     return 0xFF;
460 }
461 
462 static void xive2_presenter_backlog_decr(XivePresenter *xptr,
463                                          uint8_t nvx_blk, uint32_t nvx_idx,
464                                          uint8_t group_prio,
465                                          uint8_t group_level)
466 {
467     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
468     uint32_t nvgc_idx, count;
469     uint8_t nvgc_blk;
470     Xive2Nvgc nvgc;
471 
472     nvgc_blk = nvx_blk;
473     nvgc_idx = nvx_idx;
474     xive2_pgofnext(&nvgc_blk, &nvgc_idx, group_level);
475 
476     if (xive2_router_get_nvgc(xrtr, NVx_CROWD_LVL(group_level),
477                               nvgc_blk, nvgc_idx, &nvgc)) {
478         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVGC %x/%x\n",
479                       nvgc_blk, nvgc_idx);
480         return;
481     }
482     if (!xive2_nvgc_is_valid(&nvgc)) {
483         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid NVGC %x/%x\n",
484                       nvgc_blk, nvgc_idx);
485         return;
486     }
487     count = xive2_nvgc_get_backlog(&nvgc, group_prio);
488     if (!count) {
489         return;
490     }
491     xive2_nvgc_set_backlog(&nvgc, group_prio, count - 1);
492     xive2_router_write_nvgc(xrtr, NVx_CROWD_LVL(group_level),
493                             nvgc_blk, nvgc_idx, &nvgc);
494 }
495 
496 /*
497  * XIVE Thread Interrupt Management Area (TIMA) - Gen2 mode
498  *
499  * TIMA Gen2 VP “save & restore” (S&R) indicated by H bit next to V bit
500  *
501  *   - if a context is enabled with the H bit set, the VP context
502  *     information is retrieved from the NVP structure (“check out”)
503  *     and stored back on a context pull (“check in”), the SW receives
504  *     the same context pull information as on P9
505  *
506  *   - the H bit cannot be changed while the V bit is set, i.e. a
507  *     context cannot be set up in the TIMA and then be “pushed” into
508  *     the NVP by changing the H bit while the context is enabled
509  */
510 
511 static void xive2_tctx_save_ctx(Xive2Router *xrtr, XiveTCTX *tctx,
512                                 uint8_t nvp_blk, uint32_t nvp_idx,
513                                 uint8_t ring)
514 {
515     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
516     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
517     Xive2Nvp nvp;
518     uint8_t *regs = &tctx->regs[ring];
519 
520     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
521         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
522                           nvp_blk, nvp_idx);
523         return;
524     }
525 
526     if (!xive2_nvp_is_valid(&nvp)) {
527         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
528                       nvp_blk, nvp_idx);
529         return;
530     }
531 
532     if (!xive2_nvp_is_hw(&nvp)) {
533         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is not HW owned\n",
534                       nvp_blk, nvp_idx);
535         return;
536     }
537 
538     if (!xive2_nvp_is_co(&nvp)) {
539         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is not checkout\n",
540                       nvp_blk, nvp_idx);
541         return;
542     }
543 
544     if (xive_get_field32(NVP2_W1_CO_THRID_VALID, nvp.w1) &&
545         xive_get_field32(NVP2_W1_CO_THRID, nvp.w1) != pir) {
546         qemu_log_mask(LOG_GUEST_ERROR,
547                       "XIVE: NVP %x/%x invalid checkout Thread %x\n",
548                       nvp_blk, nvp_idx, pir);
549         return;
550     }
551 
552     nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, regs[TM_IPB]);
553     nvp.w2 = xive_set_field32(NVP2_W2_CPPR, nvp.w2, regs[TM_CPPR]);
554     if (nvp.w0 & NVP2_W0_L) {
555         /*
556          * Typically not used. If LSMFB is restored with 0, it will
557          * force a backlog rescan
558          */
559         nvp.w2 = xive_set_field32(NVP2_W2_LSMFB, nvp.w2, regs[TM_LSMFB]);
560     }
561     if (nvp.w0 & NVP2_W0_G) {
562         nvp.w2 = xive_set_field32(NVP2_W2_LGS, nvp.w2, regs[TM_LGS]);
563     }
564     if (nvp.w0 & NVP2_W0_T) {
565         nvp.w2 = xive_set_field32(NVP2_W2_T, nvp.w2, regs[TM_T]);
566     }
567     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, &nvp, 2);
568 
569     nvp.w1 = xive_set_field32(NVP2_W1_CO, nvp.w1, 0);
570     /* NVP2_W1_CO_THRID_VALID only set once */
571     nvp.w1 = xive_set_field32(NVP2_W1_CO_THRID, nvp.w1, 0xFFFF);
572     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, &nvp, 1);
573 }
574 
575 static void xive2_cam_decode(uint32_t cam, uint8_t *nvp_blk,
576                              uint32_t *nvp_idx, bool *valid, bool *hw)
577 {
578     *nvp_blk = xive2_nvp_blk(cam);
579     *nvp_idx = xive2_nvp_idx(cam);
580     *valid = !!(cam & TM2_W2_VALID);
581     *hw = !!(cam & TM2_W2_HW);
582 }
583 
584 /*
585  * Encode the HW CAM line with 7bit or 8bit thread id. The thread id
586  * width and block id width is configurable at the IC level.
587  *
588  *    chipid << 24 | 0000 0000 0000 0000 1 threadid (7Bit)
589  *    chipid << 24 | 0000 0000 0000 0001 threadid   (8Bit)
590  */
591 static uint32_t xive2_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
592 {
593     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
594     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
595     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
596     uint8_t blk = xive2_router_get_block_id(xrtr);
597     uint8_t tid_shift =
598         xive2_router_get_config(xrtr) & XIVE2_THREADID_8BITS ? 8 : 7;
599     uint8_t tid_mask = (1 << tid_shift) - 1;
600 
601     return xive2_nvp_cam_line(blk, 1 << tid_shift | (pir & tid_mask));
602 }
603 
604 static void xive2_redistribute(Xive2Router *xrtr, XiveTCTX *tctx, uint8_t ring)
605 {
606     uint8_t *regs = &tctx->regs[ring];
607     uint8_t nsr = regs[TM_NSR];
608     uint8_t pipr = regs[TM_PIPR];
609     uint8_t crowd = NVx_CROWD_LVL(nsr);
610     uint8_t group = NVx_GROUP_LVL(nsr);
611     uint8_t nvgc_blk, end_blk, nvp_blk;
612     uint32_t nvgc_idx, end_idx, nvp_idx;
613     Xive2Nvgc nvgc;
614     uint8_t prio_limit;
615     uint32_t cfg;
616     uint8_t alt_ring;
617     uint32_t target_ringw2;
618     uint32_t cam;
619     bool valid;
620     bool hw;
621 
622     /* redistribution is only for group/crowd interrupts */
623     if (!xive_nsr_indicates_group_exception(ring, nsr)) {
624         return;
625     }
626 
627     alt_ring = xive_nsr_exception_ring(ring, nsr);
628     target_ringw2 = xive_tctx_word2(&tctx->regs[alt_ring]);
629     cam = be32_to_cpu(target_ringw2);
630 
631     /* extract nvp block and index from targeted ring's cam */
632     xive2_cam_decode(cam, &nvp_blk, &nvp_idx, &valid, &hw);
633 
634     trace_xive_redistribute(tctx->cs->cpu_index, alt_ring, nvp_blk, nvp_idx);
635 
636     trace_xive_redistribute(tctx->cs->cpu_index, ring, nvp_blk, nvp_idx);
637     /* convert crowd/group to blk/idx */
638     if (group > 0) {
639         nvgc_idx = (nvp_idx & (0xffffffff << group)) |
640                    ((1 << (group - 1)) - 1);
641     } else {
642         nvgc_idx = nvp_idx;
643     }
644 
645     if (crowd > 0) {
646         crowd = (crowd == 3) ? 4 : crowd;
647         nvgc_blk = (nvp_blk & (0xffffffff << crowd)) |
648                    ((1 << (crowd - 1)) - 1);
649     } else {
650         nvgc_blk = nvp_blk;
651     }
652 
653     /* Use blk/idx to retrieve the NVGC */
654     if (xive2_router_get_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, &nvgc)) {
655         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no %s %x/%x\n",
656                       crowd ? "NVC" : "NVG", nvgc_blk, nvgc_idx);
657         return;
658     }
659 
660     /* retrieve the END blk/idx from the NVGC */
661     end_blk = xive_get_field32(NVGC2_W1_END_BLK, nvgc.w1);
662     end_idx = xive_get_field32(NVGC2_W1_END_IDX, nvgc.w1);
663 
664     /* determine number of priorities being used */
665     cfg = xive2_router_get_config(xrtr);
666     if (cfg & XIVE2_EN_VP_GRP_PRIORITY) {
667         prio_limit = 1 << GETFIELD(NVGC2_W1_PSIZE, nvgc.w1);
668     } else {
669         prio_limit = 1 << GETFIELD(XIVE2_VP_INT_PRIO, cfg);
670     }
671 
672     /* add priority offset to end index */
673     end_idx += pipr % prio_limit;
674 
675     /* trigger the group END */
676     xive2_router_end_notify(xrtr, end_blk, end_idx, 0, true);
677 
678     /* clear interrupt indication for the context */
679     regs[TM_NSR] = 0;
680     regs[TM_PIPR] = regs[TM_CPPR];
681     xive_tctx_reset_signal(tctx, ring);
682 }
683 
684 static uint64_t xive2_tm_pull_ctx(XivePresenter *xptr, XiveTCTX *tctx,
685                                   hwaddr offset, unsigned size, uint8_t ring)
686 {
687     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
688     uint32_t target_ringw2 = xive_tctx_word2(&tctx->regs[ring]);
689     uint32_t cam = be32_to_cpu(target_ringw2);
690     uint8_t nvp_blk;
691     uint32_t nvp_idx;
692     uint8_t cur_ring;
693     bool valid;
694     bool do_save;
695     uint8_t nsr;
696 
697     xive2_cam_decode(cam, &nvp_blk, &nvp_idx, &valid, &do_save);
698 
699     if (!valid) {
700         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: pulling invalid NVP %x/%x !?\n",
701                       nvp_blk, nvp_idx);
702     }
703 
704     /* Invalidate CAM line of requested ring and all lower rings */
705     for (cur_ring = TM_QW0_USER; cur_ring <= ring;
706          cur_ring += XIVE_TM_RING_SIZE) {
707         uint32_t ringw2 = xive_tctx_word2(&tctx->regs[cur_ring]);
708         uint32_t ringw2_new = xive_set_field32(TM2_QW1W2_VO, ringw2, 0);
709         memcpy(&tctx->regs[cur_ring + TM_WORD2], &ringw2_new, 4);
710     }
711 
712     /* Active group/crowd interrupts need to be redistributed */
713     nsr = tctx->regs[ring + TM_NSR];
714     if (xive_nsr_indicates_group_exception(ring, nsr)) {
715         xive2_redistribute(xrtr, tctx, ring);
716     }
717 
718     if (xive2_router_get_config(xrtr) & XIVE2_VP_SAVE_RESTORE && do_save) {
719         xive2_tctx_save_ctx(xrtr, tctx, nvp_blk, nvp_idx, ring);
720     }
721 
722     /*
723      * Lower external interrupt line of requested ring and below except for
724      * USER, which doesn't exist.
725      */
726     for (cur_ring = TM_QW1_OS; cur_ring <= ring;
727          cur_ring += XIVE_TM_RING_SIZE) {
728         xive_tctx_reset_signal(tctx, cur_ring);
729     }
730     return target_ringw2;
731 }
732 
733 uint64_t xive2_tm_pull_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
734                               hwaddr offset, unsigned size)
735 {
736     return xive2_tm_pull_ctx(xptr, tctx, offset, size, TM_QW1_OS);
737 }
738 
739 #define REPORT_LINE_GEN1_SIZE       16
740 
741 static void xive2_tm_report_line_gen1(XiveTCTX *tctx, uint8_t *data,
742                                       uint8_t size)
743 {
744     uint8_t *regs = tctx->regs;
745 
746     g_assert(size == REPORT_LINE_GEN1_SIZE);
747     memset(data, 0, size);
748     /*
749      * See xive architecture for description of what is saved. It is
750      * hand-picked information to fit in 16 bytes.
751      */
752     data[0x0] = regs[TM_QW3_HV_PHYS + TM_NSR];
753     data[0x1] = regs[TM_QW3_HV_PHYS + TM_CPPR];
754     data[0x2] = regs[TM_QW3_HV_PHYS + TM_IPB];
755     data[0x3] = regs[TM_QW2_HV_POOL + TM_IPB];
756     data[0x4] = regs[TM_QW1_OS + TM_ACK_CNT];
757     data[0x5] = regs[TM_QW3_HV_PHYS + TM_LGS];
758     data[0x6] = 0xFF;
759     data[0x7] = regs[TM_QW3_HV_PHYS + TM_WORD2] & 0x80;
760     data[0x7] |= (regs[TM_QW2_HV_POOL + TM_WORD2] & 0x80) >> 1;
761     data[0x7] |= (regs[TM_QW1_OS + TM_WORD2] & 0x80) >> 2;
762     data[0x7] |= (regs[TM_QW3_HV_PHYS + TM_WORD2] & 0x3);
763     data[0x8] = regs[TM_QW1_OS + TM_NSR];
764     data[0x9] = regs[TM_QW1_OS + TM_CPPR];
765     data[0xA] = regs[TM_QW1_OS + TM_IPB];
766     data[0xB] = regs[TM_QW1_OS + TM_LGS];
767     if (regs[TM_QW0_USER + TM_WORD2] & 0x80) {
768         /*
769          * Logical server extension, except VU bit replaced by EB bit
770          * from NSR
771          */
772         data[0xC] = regs[TM_QW0_USER + TM_WORD2];
773         data[0xC] &= ~0x80;
774         data[0xC] |= regs[TM_QW0_USER + TM_NSR] & 0x80;
775         data[0xD] = regs[TM_QW0_USER + TM_WORD2 + 1];
776         data[0xE] = regs[TM_QW0_USER + TM_WORD2 + 2];
777         data[0xF] = regs[TM_QW0_USER + TM_WORD2 + 3];
778     }
779 }
780 
781 static void xive2_tm_pull_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
782                                  hwaddr offset, uint64_t value,
783                                  unsigned size, uint8_t ring)
784 {
785     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
786     uint32_t hw_cam, nvp_idx, xive2_cfg, reserved;
787     uint8_t nvp_blk;
788     Xive2Nvp nvp;
789     uint64_t phys_addr;
790     MemTxResult result;
791 
792     hw_cam = xive2_tctx_hw_cam_line(xptr, tctx);
793     nvp_blk = xive2_nvp_blk(hw_cam);
794     nvp_idx = xive2_nvp_idx(hw_cam);
795 
796     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
797         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
798                       nvp_blk, nvp_idx);
799         return;
800     }
801 
802     if (!xive2_nvp_is_valid(&nvp)) {
803         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
804                       nvp_blk, nvp_idx);
805         return;
806     }
807 
808     xive2_cfg = xive2_router_get_config(xrtr);
809 
810     phys_addr = xive2_nvp_reporting_addr(&nvp) + 0x80; /* odd line */
811     if (xive2_cfg & XIVE2_GEN1_TIMA_OS) {
812         uint8_t pull_ctxt[REPORT_LINE_GEN1_SIZE];
813 
814         xive2_tm_report_line_gen1(tctx, pull_ctxt, REPORT_LINE_GEN1_SIZE);
815         result = dma_memory_write(&address_space_memory, phys_addr,
816                                   pull_ctxt, REPORT_LINE_GEN1_SIZE,
817                                   MEMTXATTRS_UNSPECIFIED);
818         assert(result == MEMTX_OK);
819     } else {
820         result = dma_memory_write(&address_space_memory, phys_addr,
821                                   &tctx->regs, sizeof(tctx->regs),
822                                   MEMTXATTRS_UNSPECIFIED);
823         assert(result == MEMTX_OK);
824         reserved = 0xFFFFFFFF;
825         result = dma_memory_write(&address_space_memory, phys_addr + 12,
826                                   &reserved, sizeof(reserved),
827                                   MEMTXATTRS_UNSPECIFIED);
828         assert(result == MEMTX_OK);
829     }
830 
831     /* the rest is similar to pull context to registers */
832     xive2_tm_pull_ctx(xptr, tctx, offset, size, ring);
833 }
834 
835 void xive2_tm_pull_os_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
836                              hwaddr offset, uint64_t value, unsigned size)
837 {
838     xive2_tm_pull_ctx_ol(xptr, tctx, offset, value, size, TM_QW1_OS);
839 }
840 
841 
842 void xive2_tm_pull_phys_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
843                                hwaddr offset, uint64_t value, unsigned size)
844 {
845     xive2_tm_pull_ctx_ol(xptr, tctx, offset, value, size, TM_QW3_HV_PHYS);
846 }
847 
848 static uint8_t xive2_tctx_restore_os_ctx(Xive2Router *xrtr, XiveTCTX *tctx,
849                                         uint8_t nvp_blk, uint32_t nvp_idx,
850                                         Xive2Nvp *nvp)
851 {
852     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
853     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
854     uint8_t cppr;
855 
856     if (!xive2_nvp_is_hw(nvp)) {
857         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is not HW owned\n",
858                       nvp_blk, nvp_idx);
859         return 0;
860     }
861 
862     cppr = xive_get_field32(NVP2_W2_CPPR, nvp->w2);
863     nvp->w2 = xive_set_field32(NVP2_W2_CPPR, nvp->w2, 0);
864     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, nvp, 2);
865 
866     tctx->regs[TM_QW1_OS + TM_CPPR] = cppr;
867     tctx->regs[TM_QW1_OS + TM_LSMFB] = xive_get_field32(NVP2_W2_LSMFB, nvp->w2);
868     tctx->regs[TM_QW1_OS + TM_LGS] = xive_get_field32(NVP2_W2_LGS, nvp->w2);
869     tctx->regs[TM_QW1_OS + TM_T] = xive_get_field32(NVP2_W2_T, nvp->w2);
870 
871     nvp->w1 = xive_set_field32(NVP2_W1_CO, nvp->w1, 1);
872     nvp->w1 = xive_set_field32(NVP2_W1_CO_THRID_VALID, nvp->w1, 1);
873     nvp->w1 = xive_set_field32(NVP2_W1_CO_THRID, nvp->w1, pir);
874 
875     /*
876      * Checkout privilege: 0:OS, 1:Pool, 2:Hard
877      *
878      * TODO: we only support OS push/pull
879      */
880     nvp->w1 = xive_set_field32(NVP2_W1_CO_PRIV, nvp->w1, 0);
881 
882     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, nvp, 1);
883 
884     /* return restored CPPR to generate a CPU exception if needed */
885     return cppr;
886 }
887 
888 static void xive2_tctx_need_resend(Xive2Router *xrtr, XiveTCTX *tctx,
889                                    uint8_t nvp_blk, uint32_t nvp_idx,
890                                    bool do_restore)
891 {
892     XivePresenter *xptr = XIVE_PRESENTER(xrtr);
893     uint8_t ipb;
894     uint8_t backlog_level;
895     uint8_t group_level;
896     uint8_t first_group;
897     uint8_t backlog_prio;
898     uint8_t group_prio;
899     uint8_t *regs = &tctx->regs[TM_QW1_OS];
900     Xive2Nvp nvp;
901 
902     /*
903      * Grab the associated thread interrupt context registers in the
904      * associated NVP
905      */
906     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
907         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
908                       nvp_blk, nvp_idx);
909         return;
910     }
911 
912     if (!xive2_nvp_is_valid(&nvp)) {
913         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
914                       nvp_blk, nvp_idx);
915         return;
916     }
917 
918     /* Automatically restore thread context registers */
919     if (xive2_router_get_config(xrtr) & XIVE2_VP_SAVE_RESTORE &&
920         do_restore) {
921         xive2_tctx_restore_os_ctx(xrtr, tctx, nvp_blk, nvp_idx, &nvp);
922     }
923 
924     ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2);
925     if (ipb) {
926         nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, 0);
927         xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, &nvp, 2);
928     }
929     /* IPB bits in the backlog are merged with the TIMA IPB bits */
930     regs[TM_IPB] |= ipb;
931     backlog_prio = xive_ipb_to_pipr(regs[TM_IPB]);
932     backlog_level = 0;
933 
934     first_group = xive_get_field32(NVP2_W0_PGOFIRST, nvp.w0);
935     if (first_group && regs[TM_LSMFB] < backlog_prio) {
936         group_prio = xive2_presenter_backlog_scan(xptr, nvp_blk, nvp_idx,
937                                                   first_group, &group_level);
938         regs[TM_LSMFB] = group_prio;
939         if (regs[TM_LGS] && group_prio < backlog_prio &&
940             group_prio < regs[TM_CPPR]) {
941 
942             /* VP can take a group interrupt */
943             xive2_presenter_backlog_decr(xptr, nvp_blk, nvp_idx,
944                                          group_prio, group_level);
945             backlog_prio = group_prio;
946             backlog_level = group_level;
947         }
948     }
949 
950     /*
951      * Compute the PIPR based on the restored state.
952      * It will raise the External interrupt signal if needed.
953      */
954     xive_tctx_pipr_update(tctx, TM_QW1_OS, backlog_prio, backlog_level);
955 }
956 
957 /*
958  * Updating the OS CAM line can trigger a resend of interrupt
959  */
960 void xive2_tm_push_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
961                           hwaddr offset, uint64_t value, unsigned size)
962 {
963     uint32_t cam;
964     uint32_t qw1w2;
965     uint64_t qw1dw1;
966     uint8_t nvp_blk;
967     uint32_t nvp_idx;
968     bool vo;
969     bool do_restore;
970 
971     /* First update the thead context */
972     switch (size) {
973     case 4:
974         cam = value;
975         qw1w2 = cpu_to_be32(cam);
976         memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
977         break;
978     case 8:
979         cam = value >> 32;
980         qw1dw1 = cpu_to_be64(value);
981         memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1dw1, 8);
982         break;
983     default:
984         g_assert_not_reached();
985     }
986 
987     xive2_cam_decode(cam, &nvp_blk, &nvp_idx, &vo, &do_restore);
988 
989     /* Check the interrupt pending bits */
990     if (vo) {
991         xive2_tctx_need_resend(XIVE2_ROUTER(xptr), tctx, nvp_blk, nvp_idx,
992                                do_restore);
993     }
994 }
995 
996 static int xive2_tctx_get_nvp_indexes(XiveTCTX *tctx, uint8_t ring,
997                                       uint32_t *nvp_blk, uint32_t *nvp_idx)
998 {
999     uint32_t w2, cam;
1000 
1001     w2 = xive_tctx_word2(&tctx->regs[ring]);
1002     switch (ring) {
1003     case TM_QW1_OS:
1004         if (!(be32_to_cpu(w2) & TM2_QW1W2_VO)) {
1005             return -1;
1006         }
1007         cam = xive_get_field32(TM2_QW1W2_OS_CAM, w2);
1008         break;
1009     case TM_QW2_HV_POOL:
1010         if (!(be32_to_cpu(w2) & TM2_QW2W2_VP)) {
1011             return -1;
1012         }
1013         cam = xive_get_field32(TM2_QW2W2_POOL_CAM, w2);
1014         break;
1015     case TM_QW3_HV_PHYS:
1016         if (!(be32_to_cpu(w2) & TM2_QW3W2_VT)) {
1017             return -1;
1018         }
1019         cam = xive2_tctx_hw_cam_line(tctx->xptr, tctx);
1020         break;
1021     default:
1022         return -1;
1023     }
1024     *nvp_blk = xive2_nvp_blk(cam);
1025     *nvp_idx = xive2_nvp_idx(cam);
1026     return 0;
1027 }
1028 
1029 static void xive2_tctx_accept_el(XivePresenter *xptr, XiveTCTX *tctx,
1030                                  uint8_t ring, uint8_t cl_ring)
1031 {
1032     uint64_t rd;
1033     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
1034     uint32_t nvp_blk, nvp_idx, xive2_cfg;
1035     Xive2Nvp nvp;
1036     uint64_t phys_addr;
1037     uint8_t OGen = 0;
1038 
1039     xive2_tctx_get_nvp_indexes(tctx, cl_ring, &nvp_blk, &nvp_idx);
1040 
1041     if (xive2_router_get_nvp(xrtr, (uint8_t)nvp_blk, nvp_idx, &nvp)) {
1042         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
1043                       nvp_blk, nvp_idx);
1044         return;
1045     }
1046 
1047     if (!xive2_nvp_is_valid(&nvp)) {
1048         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1049                       nvp_blk, nvp_idx);
1050         return;
1051     }
1052 
1053 
1054     rd = xive_tctx_accept(tctx, ring);
1055 
1056     if (ring == TM_QW1_OS) {
1057         OGen = tctx->regs[ring + TM_OGEN];
1058     }
1059     xive2_cfg = xive2_router_get_config(xrtr);
1060     phys_addr = xive2_nvp_reporting_addr(&nvp);
1061     uint8_t report_data[REPORT_LINE_GEN1_SIZE];
1062     memset(report_data, 0xff, sizeof(report_data));
1063     if ((OGen == 1) || (xive2_cfg & XIVE2_GEN1_TIMA_OS)) {
1064         report_data[8] = (rd >> 8) & 0xff;
1065         report_data[9] = rd & 0xff;
1066     } else {
1067         report_data[0] = (rd >> 8) & 0xff;
1068         report_data[1] = rd & 0xff;
1069     }
1070     cpu_physical_memory_write(phys_addr, report_data, REPORT_LINE_GEN1_SIZE);
1071 }
1072 
1073 void xive2_tm_ack_os_el(XivePresenter *xptr, XiveTCTX *tctx,
1074                         hwaddr offset, uint64_t value, unsigned size)
1075 {
1076     xive2_tctx_accept_el(xptr, tctx, TM_QW1_OS, TM_QW1_OS);
1077 }
1078 
1079 /* NOTE: CPPR only exists for TM_QW1_OS and TM_QW3_HV_PHYS */
1080 static void xive2_tctx_set_cppr(XiveTCTX *tctx, uint8_t ring, uint8_t cppr)
1081 {
1082     uint8_t *regs = &tctx->regs[ring];
1083     Xive2Router *xrtr = XIVE2_ROUTER(tctx->xptr);
1084     uint8_t old_cppr, backlog_prio, first_group, group_level;
1085     uint8_t pipr_min, lsmfb_min, ring_min;
1086     bool group_enabled;
1087     uint32_t nvp_blk, nvp_idx;
1088     Xive2Nvp nvp;
1089     int rc;
1090     uint8_t nsr = regs[TM_NSR];
1091 
1092     trace_xive_tctx_set_cppr(tctx->cs->cpu_index, ring,
1093                              regs[TM_IPB], regs[TM_PIPR],
1094                              cppr, nsr);
1095 
1096     if (cppr > XIVE_PRIORITY_MAX) {
1097         cppr = 0xff;
1098     }
1099 
1100     old_cppr = regs[TM_CPPR];
1101     regs[TM_CPPR] = cppr;
1102 
1103     /* Handle increased CPPR priority (lower value) */
1104     if (cppr < old_cppr) {
1105         if (cppr <= regs[TM_PIPR]) {
1106             /* CPPR lowered below PIPR, must un-present interrupt */
1107             if (xive_nsr_indicates_exception(ring, nsr)) {
1108                 if (xive_nsr_indicates_group_exception(ring, nsr)) {
1109                     /* redistribute precluded active grp interrupt */
1110                     xive2_redistribute(xrtr, tctx, ring);
1111                     return;
1112                 }
1113             }
1114 
1115             /* interrupt is VP directed, pending in IPB */
1116             regs[TM_PIPR] = cppr;
1117             xive_tctx_notify(tctx, ring, 0); /* Ensure interrupt is cleared */
1118             return;
1119         } else {
1120             /* CPPR was lowered, but still above PIPR. No action needed. */
1121             return;
1122         }
1123     }
1124 
1125     /* CPPR didn't change, nothing needs to be done */
1126     if (cppr == old_cppr) {
1127         return;
1128     }
1129 
1130     /* CPPR priority decreased (higher value) */
1131 
1132     /*
1133      * Recompute the PIPR based on local pending interrupts. It will
1134      * be adjusted below if needed in case of pending group interrupts.
1135      */
1136 again:
1137     pipr_min = xive_ipb_to_pipr(regs[TM_IPB]);
1138     group_enabled = !!regs[TM_LGS];
1139     lsmfb_min = group_enabled ? regs[TM_LSMFB] : 0xff;
1140     ring_min = ring;
1141     group_level = 0;
1142 
1143     /* PHYS updates also depend on POOL values */
1144     if (ring == TM_QW3_HV_PHYS) {
1145         uint8_t *pool_regs = &tctx->regs[TM_QW2_HV_POOL];
1146 
1147         /* POOL values only matter if POOL ctx is valid */
1148         if (pool_regs[TM_WORD2] & 0x80) {
1149             uint8_t pool_pipr = xive_ipb_to_pipr(pool_regs[TM_IPB]);
1150             uint8_t pool_lsmfb = pool_regs[TM_LSMFB];
1151 
1152             /*
1153              * Determine highest priority interrupt and
1154              * remember which ring has it.
1155              */
1156             if (pool_pipr < pipr_min) {
1157                 pipr_min = pool_pipr;
1158                 if (pool_pipr < lsmfb_min) {
1159                     ring_min = TM_QW2_HV_POOL;
1160                 }
1161             }
1162 
1163             /* Values needed for group priority calculation */
1164             if (pool_regs[TM_LGS] && (pool_lsmfb < lsmfb_min)) {
1165                 group_enabled = true;
1166                 lsmfb_min = pool_lsmfb;
1167                 if (lsmfb_min < pipr_min) {
1168                     ring_min = TM_QW2_HV_POOL;
1169                 }
1170             }
1171         }
1172     }
1173 
1174     rc = xive2_tctx_get_nvp_indexes(tctx, ring_min, &nvp_blk, &nvp_idx);
1175     if (rc) {
1176         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: set CPPR on invalid context\n");
1177         return;
1178     }
1179 
1180     if (group_enabled &&
1181         lsmfb_min < cppr &&
1182         lsmfb_min < pipr_min) {
1183         /*
1184          * Thread has seen a group interrupt with a higher priority
1185          * than the new cppr or pending local interrupt. Check the
1186          * backlog
1187          */
1188         if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
1189             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
1190                           nvp_blk, nvp_idx);
1191             return;
1192         }
1193 
1194         if (!xive2_nvp_is_valid(&nvp)) {
1195             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1196                           nvp_blk, nvp_idx);
1197             return;
1198         }
1199 
1200         first_group = xive_get_field32(NVP2_W0_PGOFIRST, nvp.w0);
1201         if (!first_group) {
1202             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
1203                           nvp_blk, nvp_idx);
1204             return;
1205         }
1206 
1207         backlog_prio = xive2_presenter_backlog_scan(tctx->xptr,
1208                                                     nvp_blk, nvp_idx,
1209                                                     first_group, &group_level);
1210         tctx->regs[ring_min + TM_LSMFB] = backlog_prio;
1211         if (backlog_prio != lsmfb_min) {
1212             /*
1213              * If the group backlog scan finds a less favored or no interrupt,
1214              * then re-do the processing which may turn up a more favored
1215              * interrupt from IPB or the other pool. Backlog should not
1216              * find a priority < LSMFB.
1217              */
1218             g_assert(backlog_prio >= lsmfb_min);
1219             goto again;
1220         }
1221 
1222         xive2_presenter_backlog_decr(tctx->xptr, nvp_blk, nvp_idx,
1223                                      backlog_prio, group_level);
1224         pipr_min = backlog_prio;
1225     }
1226 
1227     /* PIPR should not be set to a value greater than CPPR */
1228     regs[TM_PIPR] = (pipr_min > cppr) ? cppr : pipr_min;
1229 
1230     /* CPPR has changed, check if we need to raise a pending exception */
1231     xive_tctx_notify(tctx, ring_min, group_level);
1232 }
1233 
1234 void xive2_tm_set_hv_cppr(XivePresenter *xptr, XiveTCTX *tctx,
1235                           hwaddr offset, uint64_t value, unsigned size)
1236 {
1237     xive2_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
1238 }
1239 
1240 void xive2_tm_set_os_cppr(XivePresenter *xptr, XiveTCTX *tctx,
1241                           hwaddr offset, uint64_t value, unsigned size)
1242 {
1243     xive2_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
1244 }
1245 
1246 static void xive2_tctx_set_target(XiveTCTX *tctx, uint8_t ring, uint8_t target)
1247 {
1248     uint8_t *regs = &tctx->regs[ring];
1249 
1250     regs[TM_T] = target;
1251 }
1252 
1253 void xive2_tm_set_hv_target(XivePresenter *xptr, XiveTCTX *tctx,
1254                             hwaddr offset, uint64_t value, unsigned size)
1255 {
1256     xive2_tctx_set_target(tctx, TM_QW3_HV_PHYS, value & 0xff);
1257 }
1258 
1259 /*
1260  * XIVE Router (aka. Virtualization Controller or IVRE)
1261  */
1262 
1263 int xive2_router_get_eas(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1264                          Xive2Eas *eas)
1265 {
1266     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1267 
1268     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1269 }
1270 
1271 static
1272 int xive2_router_get_pq(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1273                        uint8_t *pq)
1274 {
1275     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1276 
1277     return xrc->get_pq(xrtr, eas_blk, eas_idx, pq);
1278 }
1279 
1280 static
1281 int xive2_router_set_pq(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1282                        uint8_t *pq)
1283 {
1284     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1285 
1286     return xrc->set_pq(xrtr, eas_blk, eas_idx, pq);
1287 }
1288 
1289 int xive2_router_get_end(Xive2Router *xrtr, uint8_t end_blk, uint32_t end_idx,
1290                          Xive2End *end)
1291 {
1292    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1293 
1294    return xrc->get_end(xrtr, end_blk, end_idx, end);
1295 }
1296 
1297 int xive2_router_write_end(Xive2Router *xrtr, uint8_t end_blk, uint32_t end_idx,
1298                            Xive2End *end, uint8_t word_number)
1299 {
1300    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1301 
1302    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1303 }
1304 
1305 int xive2_router_get_nvp(Xive2Router *xrtr, uint8_t nvp_blk, uint32_t nvp_idx,
1306                          Xive2Nvp *nvp)
1307 {
1308    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1309 
1310    return xrc->get_nvp(xrtr, nvp_blk, nvp_idx, nvp);
1311 }
1312 
1313 int xive2_router_write_nvp(Xive2Router *xrtr, uint8_t nvp_blk, uint32_t nvp_idx,
1314                            Xive2Nvp *nvp, uint8_t word_number)
1315 {
1316    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1317 
1318    return xrc->write_nvp(xrtr, nvp_blk, nvp_idx, nvp, word_number);
1319 }
1320 
1321 int xive2_router_get_nvgc(Xive2Router *xrtr, bool crowd,
1322                           uint8_t nvgc_blk, uint32_t nvgc_idx,
1323                           Xive2Nvgc *nvgc)
1324 {
1325    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1326 
1327    return xrc->get_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
1328 }
1329 
1330 int xive2_router_write_nvgc(Xive2Router *xrtr, bool crowd,
1331                             uint8_t nvgc_blk, uint32_t nvgc_idx,
1332                             Xive2Nvgc *nvgc)
1333 {
1334    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1335 
1336    return xrc->write_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
1337 }
1338 
1339 static bool xive2_vp_match_mask(uint32_t cam1, uint32_t cam2,
1340                                 uint32_t vp_mask)
1341 {
1342     return (cam1 & vp_mask) == (cam2 & vp_mask);
1343 }
1344 
1345 static uint8_t xive2_get_vp_block_mask(uint32_t nvt_blk, bool crowd)
1346 {
1347     uint8_t block_mask = 0b1111;
1348 
1349     /* 3 supported crowd sizes: 2, 4, 16 */
1350     if (crowd) {
1351         uint32_t size = xive_get_vpgroup_size(nvt_blk);
1352 
1353         if (size != 2 && size != 4 && size != 16) {
1354             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid crowd size of %d",
1355                                            size);
1356             return block_mask;
1357         }
1358         block_mask &= ~(size - 1);
1359     }
1360     return block_mask;
1361 }
1362 
1363 static uint32_t xive2_get_vp_index_mask(uint32_t nvt_index, bool cam_ignore)
1364 {
1365     uint32_t index_mask = 0xFFFFFF; /* 24 bits */
1366 
1367     if (cam_ignore) {
1368         uint32_t size = xive_get_vpgroup_size(nvt_index);
1369 
1370         if (size < 2) {
1371             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid group size of %d",
1372                                            size);
1373             return index_mask;
1374         }
1375         index_mask &= ~(size - 1);
1376     }
1377     return index_mask;
1378 }
1379 
1380 /*
1381  * The thread context register words are in big-endian format.
1382  */
1383 int xive2_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1384                                uint8_t format,
1385                                uint8_t nvt_blk, uint32_t nvt_idx,
1386                                bool crowd, bool cam_ignore,
1387                                uint32_t logic_serv)
1388 {
1389     uint32_t cam =   xive2_nvp_cam_line(nvt_blk, nvt_idx);
1390     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1391     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1392     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1393     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1394 
1395     uint32_t index_mask, vp_mask;
1396     uint8_t block_mask;
1397 
1398     if (format == 0) {
1399         /*
1400          * i=0: Specific NVT notification
1401          * i=1: VP-group notification (bits ignored at the end of the
1402          *      NVT identifier)
1403          */
1404         block_mask = xive2_get_vp_block_mask(nvt_blk, crowd);
1405         index_mask = xive2_get_vp_index_mask(nvt_idx, cam_ignore);
1406         vp_mask = xive2_nvp_cam_line(block_mask, index_mask);
1407 
1408         /* For VP-group notifications, threads with LGS=0 are excluded */
1409 
1410         /* PHYS ring */
1411         if ((be32_to_cpu(qw3w2) & TM2_QW3W2_VT) &&
1412             !(cam_ignore && tctx->regs[TM_QW3_HV_PHYS + TM_LGS] == 0) &&
1413             xive2_vp_match_mask(cam,
1414                                 xive2_tctx_hw_cam_line(xptr, tctx),
1415                                 vp_mask)) {
1416             return TM_QW3_HV_PHYS;
1417         }
1418 
1419         /* HV POOL ring */
1420         if ((be32_to_cpu(qw2w2) & TM2_QW2W2_VP) &&
1421             !(cam_ignore && tctx->regs[TM_QW2_HV_POOL + TM_LGS] == 0) &&
1422             xive2_vp_match_mask(cam,
1423                                 xive_get_field32(TM2_QW2W2_POOL_CAM, qw2w2),
1424                                 vp_mask)) {
1425             return TM_QW2_HV_POOL;
1426         }
1427 
1428         /* OS ring */
1429         if ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
1430             !(cam_ignore && tctx->regs[TM_QW1_OS + TM_LGS] == 0) &&
1431             xive2_vp_match_mask(cam,
1432                                 xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2),
1433                                 vp_mask)) {
1434             return TM_QW1_OS;
1435         }
1436     } else {
1437         /* F=1 : User level Event-Based Branch (EBB) notification */
1438 
1439         /* FIXME: what if cam_ignore and LGS = 0 ? */
1440         /* USER ring */
1441         if  ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
1442              (cam == xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2)) &&
1443              (be32_to_cpu(qw0w2) & TM2_QW0W2_VU) &&
1444              (logic_serv == xive_get_field32(TM2_QW0W2_LOGIC_SERV, qw0w2))) {
1445             return TM_QW0_USER;
1446         }
1447     }
1448     return -1;
1449 }
1450 
1451 bool xive2_tm_irq_precluded(XiveTCTX *tctx, int ring, uint8_t priority)
1452 {
1453     /* HV_POOL ring uses HV_PHYS NSR, CPPR and PIPR registers */
1454     uint8_t alt_ring = (ring == TM_QW2_HV_POOL) ? TM_QW3_HV_PHYS : ring;
1455     uint8_t *alt_regs = &tctx->regs[alt_ring];
1456 
1457     /*
1458      * The xive2_presenter_tctx_match() above tells if there's a match
1459      * but for VP-group notification, we still need to look at the
1460      * priority to know if the thread can take the interrupt now or if
1461      * it is precluded.
1462      */
1463     if (priority < alt_regs[TM_PIPR]) {
1464         return false;
1465     }
1466     return true;
1467 }
1468 
1469 void xive2_tm_set_lsmfb(XiveTCTX *tctx, int ring, uint8_t priority)
1470 {
1471     uint8_t *regs = &tctx->regs[ring];
1472 
1473     /*
1474      * Called by the router during a VP-group notification when the
1475      * thread matches but can't take the interrupt because it's
1476      * already running at a more favored priority. It then stores the
1477      * new interrupt priority in the LSMFB field.
1478      */
1479     regs[TM_LSMFB] = priority;
1480 }
1481 
1482 static void xive2_router_realize(DeviceState *dev, Error **errp)
1483 {
1484     Xive2Router *xrtr = XIVE2_ROUTER(dev);
1485 
1486     assert(xrtr->xfb);
1487 }
1488 
1489 /*
1490  * Notification using the END ESe/ESn bit (Event State Buffer for
1491  * escalation and notification). Profide further coalescing in the
1492  * Router.
1493  */
1494 static bool xive2_router_end_es_notify(Xive2Router *xrtr, uint8_t end_blk,
1495                                        uint32_t end_idx, Xive2End *end,
1496                                        uint32_t end_esmask)
1497 {
1498     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1499     bool notify = xive_esb_trigger(&pq);
1500 
1501     if (pq != xive_get_field32(end_esmask, end->w1)) {
1502         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1503         xive2_router_write_end(xrtr, end_blk, end_idx, end, 1);
1504     }
1505 
1506     /* ESe/n[Q]=1 : end of notification */
1507     return notify;
1508 }
1509 
1510 /*
1511  * An END trigger can come from an event trigger (IPI or HW) or from
1512  * another chip. We don't model the PowerBus but the END trigger
1513  * message has the same parameters than in the function below.
1514  */
1515 static void xive2_router_end_notify(Xive2Router *xrtr, uint8_t end_blk,
1516                                     uint32_t end_idx, uint32_t end_data,
1517                                     bool redistribute)
1518 {
1519     Xive2End end;
1520     uint8_t priority;
1521     uint8_t format;
1522     bool found, precluded;
1523     uint8_t nvx_blk;
1524     uint32_t nvx_idx;
1525 
1526     /* END cache lookup */
1527     if (xive2_router_get_end(xrtr, end_blk, end_idx, &end)) {
1528         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1529                       end_idx);
1530         return;
1531     }
1532 
1533     if (!xive2_end_is_valid(&end)) {
1534         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1535                       end_blk, end_idx);
1536         return;
1537     }
1538 
1539     if (xive2_end_is_crowd(&end) && !xive2_end_is_ignore(&end)) {
1540         qemu_log_mask(LOG_GUEST_ERROR,
1541                       "XIVE: invalid END, 'crowd' bit requires 'ignore' bit\n");
1542         return;
1543     }
1544 
1545     if (!redistribute && xive2_end_is_enqueue(&end)) {
1546         trace_xive_end_enqueue(end_blk, end_idx, end_data);
1547         xive2_end_enqueue(&end, end_data);
1548         /* Enqueuing event data modifies the EQ toggle and index */
1549         xive2_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1550     }
1551 
1552     /*
1553      * When the END is silent, we skip the notification part.
1554      */
1555     if (xive2_end_is_silent_escalation(&end)) {
1556         goto do_escalation;
1557     }
1558 
1559     /*
1560      * The W7 format depends on the F bit in W6. It defines the type
1561      * of the notification :
1562      *
1563      *   F=0 : single or multiple NVP notification
1564      *   F=1 : User level Event-Based Branch (EBB) notification, no
1565      *         priority
1566      */
1567     format = xive_get_field32(END2_W6_FORMAT_BIT, end.w6);
1568     priority = xive_get_field32(END2_W7_F0_PRIORITY, end.w7);
1569 
1570     /* The END is masked */
1571     if (format == 0 && priority == 0xff) {
1572         return;
1573     }
1574 
1575     /*
1576      * Check the END ESn (Event State Buffer for notification) for
1577      * even further coalescing in the Router
1578      */
1579     if (!xive2_end_is_notify(&end)) {
1580         /* ESn[Q]=1 : end of notification */
1581         if (!xive2_router_end_es_notify(xrtr, end_blk, end_idx,
1582                                        &end, END2_W1_ESn)) {
1583             return;
1584         }
1585     }
1586 
1587     /*
1588      * Follows IVPE notification
1589      */
1590     nvx_blk = xive_get_field32(END2_W6_VP_BLOCK, end.w6);
1591     nvx_idx = xive_get_field32(END2_W6_VP_OFFSET, end.w6);
1592 
1593     found = xive_presenter_notify(xrtr->xfb, format, nvx_blk, nvx_idx,
1594                           xive2_end_is_crowd(&end), xive2_end_is_ignore(&end),
1595                           priority,
1596                           xive_get_field32(END2_W7_F1_LOG_SERVER_ID, end.w7),
1597                           &precluded);
1598 
1599     /* TODO: Auto EOI. */
1600 
1601     if (found) {
1602         return;
1603     }
1604 
1605     /*
1606      * If no matching NVP is dispatched on a HW thread :
1607      * - specific VP: update the NVP structure if backlog is activated
1608      * - VP-group: update the backlog counter for that priority in the NVG
1609      */
1610     if (xive2_end_is_backlog(&end)) {
1611 
1612         if (format == 1) {
1613             qemu_log_mask(LOG_GUEST_ERROR,
1614                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1615                           end_blk, end_idx);
1616             return;
1617         }
1618 
1619         if (!xive2_end_is_ignore(&end)) {
1620             uint8_t ipb;
1621             Xive2Nvp nvp;
1622 
1623             /* NVP cache lookup */
1624             if (xive2_router_get_nvp(xrtr, nvx_blk, nvx_idx, &nvp)) {
1625                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVP %x/%x\n",
1626                               nvx_blk, nvx_idx);
1627                 return;
1628             }
1629 
1630             if (!xive2_nvp_is_valid(&nvp)) {
1631                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is invalid\n",
1632                               nvx_blk, nvx_idx);
1633                 return;
1634             }
1635 
1636             /*
1637              * Record the IPB in the associated NVP structure for later
1638              * use. The presenter will resend the interrupt when the vCPU
1639              * is dispatched again on a HW thread.
1640              */
1641             ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2) |
1642                 xive_priority_to_ipb(priority);
1643             nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, ipb);
1644             xive2_router_write_nvp(xrtr, nvx_blk, nvx_idx, &nvp, 2);
1645         } else {
1646             Xive2Nvgc nvgc;
1647             uint32_t backlog;
1648             bool crowd;
1649 
1650             crowd = xive2_end_is_crowd(&end);
1651 
1652             /*
1653              * For groups and crowds, the per-priority backlog
1654              * counters are stored in the NVG/NVC structures
1655              */
1656             if (xive2_router_get_nvgc(xrtr, crowd,
1657                                       nvx_blk, nvx_idx, &nvgc)) {
1658                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no %s %x/%x\n",
1659                               crowd ? "NVC" : "NVG", nvx_blk, nvx_idx);
1660                 return;
1661             }
1662 
1663             if (!xive2_nvgc_is_valid(&nvgc)) {
1664                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVG %x/%x is invalid\n",
1665                               nvx_blk, nvx_idx);
1666                 return;
1667             }
1668 
1669             /*
1670              * Increment the backlog counter for that priority.
1671              * We only call broadcast the first time the counter is
1672              * incremented. broadcast will set the LSMFB field of the TIMA of
1673              * relevant threads so that they know an interrupt is pending.
1674              */
1675             backlog = xive2_nvgc_get_backlog(&nvgc, priority) + 1;
1676             xive2_nvgc_set_backlog(&nvgc, priority, backlog);
1677             xive2_router_write_nvgc(xrtr, crowd, nvx_blk, nvx_idx, &nvgc);
1678 
1679             if (backlog == 1) {
1680                 XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xrtr->xfb);
1681                 xfc->broadcast(xrtr->xfb, nvx_blk, nvx_idx,
1682                                xive2_end_is_crowd(&end),
1683                                xive2_end_is_ignore(&end),
1684                                priority);
1685 
1686                 if (!xive2_end_is_precluded_escalation(&end)) {
1687                     /*
1688                      * The interrupt will be picked up when the
1689                      * matching thread lowers its priority level
1690                      */
1691                     return;
1692                 }
1693             }
1694         }
1695     }
1696 
1697 do_escalation:
1698     /*
1699      * If activated, escalate notification using the ESe PQ bits and
1700      * the EAS in w4-5
1701      */
1702     if (!xive2_end_is_escalate(&end)) {
1703         return;
1704     }
1705 
1706     /*
1707      * Check the END ESe (Event State Buffer for escalation) for even
1708      * further coalescing in the Router
1709      */
1710     if (!xive2_end_is_uncond_escalation(&end)) {
1711         /* ESe[Q]=1 : end of escalation notification */
1712         if (!xive2_router_end_es_notify(xrtr, end_blk, end_idx,
1713                                        &end, END2_W1_ESe)) {
1714             return;
1715         }
1716     }
1717 
1718     if (xive2_end_is_escalate_end(&end)) {
1719         /*
1720          * Perform END Adaptive escalation processing
1721          * The END trigger becomes an Escalation trigger
1722          */
1723         uint8_t esc_blk = xive_get_field32(END2_W4_END_BLOCK, end.w4);
1724         uint32_t esc_idx = xive_get_field32(END2_W4_ESC_END_INDEX, end.w4);
1725         uint32_t esc_data = xive_get_field32(END2_W5_ESC_END_DATA, end.w5);
1726         trace_xive_escalate_end(end_blk, end_idx, esc_blk, esc_idx, esc_data);
1727         xive2_router_end_notify(xrtr, esc_blk, esc_idx, esc_data, false);
1728     } /* end END adaptive escalation */
1729 
1730     else {
1731         uint32_t lisn;              /* Logical Interrupt Source Number */
1732 
1733         /*
1734          *  Perform ESB escalation processing
1735          *      E[N] == 1 --> N
1736          *      Req[Block] <- E[ESB_Block]
1737          *      Req[Index] <- E[ESB_Index]
1738          *      Req[Offset] <- 0x000
1739          *      Execute <ESB Store> Req command
1740          */
1741         lisn = XIVE_EAS(xive_get_field32(END2_W4_END_BLOCK,     end.w4),
1742                         xive_get_field32(END2_W4_ESC_END_INDEX, end.w4));
1743 
1744         trace_xive_escalate_esb(end_blk, end_idx, lisn);
1745         xive2_notify(xrtr, lisn, true /* pq_checked */);
1746     }
1747 
1748     return;
1749 }
1750 
1751 void xive2_notify(Xive2Router *xrtr , uint32_t lisn, bool pq_checked)
1752 {
1753     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1754     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1755     Xive2Eas eas;
1756 
1757     /* EAS cache lookup */
1758     if (xive2_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1759         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1760         return;
1761     }
1762 
1763     if (!pq_checked) {
1764         bool notify;
1765         uint8_t pq;
1766 
1767         /* PQ cache lookup */
1768         if (xive2_router_get_pq(xrtr, eas_blk, eas_idx, &pq)) {
1769             /* Set FIR */
1770             g_assert_not_reached();
1771         }
1772 
1773         notify = xive_esb_trigger(&pq);
1774 
1775         if (xive2_router_set_pq(xrtr, eas_blk, eas_idx, &pq)) {
1776             /* Set FIR */
1777             g_assert_not_reached();
1778         }
1779 
1780         if (!notify) {
1781             return;
1782         }
1783     }
1784 
1785     if (!xive2_eas_is_valid(&eas)) {
1786         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN %x\n", lisn);
1787         return;
1788     }
1789 
1790     if (xive2_eas_is_masked(&eas)) {
1791         /* Notification completed */
1792         return;
1793     }
1794 
1795     /* TODO: add support for EAS resume */
1796     if (xive2_eas_is_resume(&eas)) {
1797         qemu_log_mask(LOG_UNIMP,
1798                       "XIVE: EAS resume processing unimplemented - LISN %x\n",
1799                       lisn);
1800         return;
1801     }
1802 
1803     /*
1804      * The event trigger becomes an END trigger
1805      */
1806     xive2_router_end_notify(xrtr,
1807                             xive_get_field64(EAS2_END_BLOCK, eas.w),
1808                             xive_get_field64(EAS2_END_INDEX, eas.w),
1809                             xive_get_field64(EAS2_END_DATA,  eas.w),
1810                             false);
1811     return;
1812 }
1813 
1814 void xive2_router_notify(XiveNotifier *xn, uint32_t lisn, bool pq_checked)
1815 {
1816     Xive2Router *xrtr = XIVE2_ROUTER(xn);
1817 
1818     xive2_notify(xrtr, lisn, pq_checked);
1819     return;
1820 }
1821 
1822 static const Property xive2_router_properties[] = {
1823     DEFINE_PROP_LINK("xive-fabric", Xive2Router, xfb,
1824                      TYPE_XIVE_FABRIC, XiveFabric *),
1825 };
1826 
1827 static void xive2_router_class_init(ObjectClass *klass, const void *data)
1828 {
1829     DeviceClass *dc = DEVICE_CLASS(klass);
1830     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1831 
1832     dc->desc    = "XIVE2 Router Engine";
1833     device_class_set_props(dc, xive2_router_properties);
1834     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1835     dc->realize = xive2_router_realize;
1836     xnc->notify = xive2_router_notify;
1837 }
1838 
1839 static const TypeInfo xive2_router_info = {
1840     .name          = TYPE_XIVE2_ROUTER,
1841     .parent        = TYPE_SYS_BUS_DEVICE,
1842     .abstract      = true,
1843     .instance_size = sizeof(Xive2Router),
1844     .class_size    = sizeof(Xive2RouterClass),
1845     .class_init    = xive2_router_class_init,
1846     .interfaces    = (const InterfaceInfo[]) {
1847         { TYPE_XIVE_NOTIFIER },
1848         { TYPE_XIVE_PRESENTER },
1849         { }
1850     }
1851 };
1852 
1853 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
1854 {
1855     return !((addr >> shift) & 1);
1856 }
1857 
1858 static uint64_t xive2_end_source_read(void *opaque, hwaddr addr, unsigned size)
1859 {
1860     Xive2EndSource *xsrc = XIVE2_END_SOURCE(opaque);
1861     uint32_t offset = addr & 0xFFF;
1862     uint8_t end_blk;
1863     uint32_t end_idx;
1864     Xive2End end;
1865     uint32_t end_esmask;
1866     uint8_t pq;
1867     uint64_t ret;
1868 
1869     /*
1870      * The block id should be deduced from the load address on the END
1871      * ESB MMIO but our model only supports a single block per XIVE chip.
1872      */
1873     end_blk = xive2_router_get_block_id(xsrc->xrtr);
1874     end_idx = addr >> (xsrc->esb_shift + 1);
1875 
1876     if (xive2_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1877         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1878                       end_idx);
1879         return -1;
1880     }
1881 
1882     if (!xive2_end_is_valid(&end)) {
1883         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1884                       end_blk, end_idx);
1885         return -1;
1886     }
1887 
1888     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END2_W1_ESn :
1889         END2_W1_ESe;
1890     pq = xive_get_field32(end_esmask, end.w1);
1891 
1892     switch (offset) {
1893     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1894         ret = xive_esb_eoi(&pq);
1895 
1896         /* Forward the source event notification for routing ?? */
1897         break;
1898 
1899     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1900         ret = pq;
1901         break;
1902 
1903     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1904     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1905     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1906     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1907         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1908         break;
1909     default:
1910         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1911                       offset);
1912         return -1;
1913     }
1914 
1915     if (pq != xive_get_field32(end_esmask, end.w1)) {
1916         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1917         xive2_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1918     }
1919 
1920     return ret;
1921 }
1922 
1923 static void xive2_end_source_write(void *opaque, hwaddr addr,
1924                                    uint64_t value, unsigned size)
1925 {
1926     Xive2EndSource *xsrc = XIVE2_END_SOURCE(opaque);
1927     uint32_t offset = addr & 0xFFF;
1928     uint8_t end_blk;
1929     uint32_t end_idx;
1930     Xive2End end;
1931     uint32_t end_esmask;
1932     uint8_t pq;
1933     bool notify = false;
1934 
1935     /*
1936      * The block id should be deduced from the load address on the END
1937      * ESB MMIO but our model only supports a single block per XIVE chip.
1938      */
1939     end_blk = xive2_router_get_block_id(xsrc->xrtr);
1940     end_idx = addr >> (xsrc->esb_shift + 1);
1941 
1942     if (xive2_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1943         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1944                       end_idx);
1945         return;
1946     }
1947 
1948     if (!xive2_end_is_valid(&end)) {
1949         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1950                       end_blk, end_idx);
1951         return;
1952     }
1953 
1954     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END2_W1_ESn :
1955         END2_W1_ESe;
1956     pq = xive_get_field32(end_esmask, end.w1);
1957 
1958     switch (offset) {
1959     case 0 ... 0x3FF:
1960         notify = xive_esb_trigger(&pq);
1961         break;
1962 
1963     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
1964         /* TODO: can we check StoreEOI availability from the router ? */
1965         notify = xive_esb_eoi(&pq);
1966         break;
1967 
1968     case XIVE_ESB_INJECT ... XIVE_ESB_INJECT + 0x3FF:
1969         if (end_esmask == END2_W1_ESe) {
1970             qemu_log_mask(LOG_GUEST_ERROR,
1971                           "XIVE: END %x/%x can not EQ inject on ESe\n",
1972                            end_blk, end_idx);
1973             return;
1974         }
1975         notify = true;
1976         break;
1977 
1978     default:
1979         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB write addr %d\n",
1980                       offset);
1981         return;
1982     }
1983 
1984     if (pq != xive_get_field32(end_esmask, end.w1)) {
1985         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1986         xive2_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1987     }
1988 
1989     /* TODO: Forward the source event notification for routing */
1990     if (notify) {
1991         ;
1992     }
1993 }
1994 
1995 static const MemoryRegionOps xive2_end_source_ops = {
1996     .read = xive2_end_source_read,
1997     .write = xive2_end_source_write,
1998     .endianness = DEVICE_BIG_ENDIAN,
1999     .valid = {
2000         .min_access_size = 1,
2001         .max_access_size = 8,
2002     },
2003     .impl = {
2004         .min_access_size = 1,
2005         .max_access_size = 8,
2006     },
2007 };
2008 
2009 static void xive2_end_source_realize(DeviceState *dev, Error **errp)
2010 {
2011     Xive2EndSource *xsrc = XIVE2_END_SOURCE(dev);
2012 
2013     assert(xsrc->xrtr);
2014 
2015     if (!xsrc->nr_ends) {
2016         error_setg(errp, "Number of interrupt needs to be greater than 0");
2017         return;
2018     }
2019 
2020     if (xsrc->esb_shift != XIVE_ESB_4K &&
2021         xsrc->esb_shift != XIVE_ESB_64K) {
2022         error_setg(errp, "Invalid ESB shift setting");
2023         return;
2024     }
2025 
2026     /*
2027      * Each END is assigned an even/odd pair of MMIO pages, the even page
2028      * manages the ESn field while the odd page manages the ESe field.
2029      */
2030     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
2031                           &xive2_end_source_ops, xsrc, "xive.end",
2032                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
2033 }
2034 
2035 static const Property xive2_end_source_properties[] = {
2036     DEFINE_PROP_UINT32("nr-ends", Xive2EndSource, nr_ends, 0),
2037     DEFINE_PROP_UINT32("shift", Xive2EndSource, esb_shift, XIVE_ESB_64K),
2038     DEFINE_PROP_LINK("xive", Xive2EndSource, xrtr, TYPE_XIVE2_ROUTER,
2039                      Xive2Router *),
2040 };
2041 
2042 static void xive2_end_source_class_init(ObjectClass *klass, const void *data)
2043 {
2044     DeviceClass *dc = DEVICE_CLASS(klass);
2045 
2046     dc->desc    = "XIVE END Source";
2047     device_class_set_props(dc, xive2_end_source_properties);
2048     dc->realize = xive2_end_source_realize;
2049     dc->user_creatable = false;
2050 }
2051 
2052 static const TypeInfo xive2_end_source_info = {
2053     .name          = TYPE_XIVE2_END_SOURCE,
2054     .parent        = TYPE_DEVICE,
2055     .instance_size = sizeof(Xive2EndSource),
2056     .class_init    = xive2_end_source_class_init,
2057 };
2058 
2059 static void xive2_register_types(void)
2060 {
2061     type_register_static(&xive2_router_info);
2062     type_register_static(&xive2_end_source_info);
2063 }
2064 
2065 type_init(xive2_register_types)
2066