xref: /openbmc/qemu/hw/intc/xive2.c (revision 26c55b99418173107897b28ffdb8171e913339e9)
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 uint32_t xive2_router_get_config(Xive2Router *xrtr)
23 {
24     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
25 
26     return xrc->get_config(xrtr);
27 }
28 
29 static int xive2_router_get_block_id(Xive2Router *xrtr)
30 {
31    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
32 
33    return xrc->get_block_id(xrtr);
34 }
35 
36 static uint64_t xive2_nvp_reporting_addr(Xive2Nvp *nvp)
37 {
38     uint64_t cache_addr;
39 
40     cache_addr = xive_get_field32(NVP2_W6_REPORTING_LINE, nvp->w6) << 24 |
41         xive_get_field32(NVP2_W7_REPORTING_LINE, nvp->w7);
42     cache_addr <<= 8; /* aligned on a cache line pair */
43     return cache_addr;
44 }
45 
46 static uint32_t xive2_nvgc_get_backlog(Xive2Nvgc *nvgc, uint8_t priority)
47 {
48     uint32_t val = 0;
49     uint8_t *ptr, i;
50 
51     if (priority > 7) {
52         return 0;
53     }
54 
55     /*
56      * The per-priority backlog counters are 24-bit and the structure
57      * is stored in big endian. NVGC is 32-bytes long, so 24-bytes from
58      * w2, which fits 8 priorities * 24-bits per priority.
59      */
60     ptr = (uint8_t *)&nvgc->w2 + priority * 3;
61     for (i = 0; i < 3; i++, ptr++) {
62         val = (val << 8) + *ptr;
63     }
64     return val;
65 }
66 
67 static void xive2_nvgc_set_backlog(Xive2Nvgc *nvgc, uint8_t priority,
68                                    uint32_t val)
69 {
70     uint8_t *ptr, i;
71     uint32_t shift;
72 
73     if (priority > 7) {
74         return;
75     }
76 
77     if (val > 0xFFFFFF) {
78         val = 0xFFFFFF;
79     }
80     /*
81      * The per-priority backlog counters are 24-bit and the structure
82      * is stored in big endian
83      */
84     ptr = (uint8_t *)&nvgc->w2 + priority * 3;
85     for (i = 0; i < 3; i++, ptr++) {
86         shift = 8 * (2 - i);
87         *ptr = (val >> shift) & 0xFF;
88     }
89 }
90 
91 void xive2_eas_pic_print_info(Xive2Eas *eas, uint32_t lisn, GString *buf)
92 {
93     if (!xive2_eas_is_valid(eas)) {
94         return;
95     }
96 
97     g_string_append_printf(buf, "  %08x %s end:%02x/%04x data:%08x\n",
98                            lisn, xive2_eas_is_masked(eas) ? "M" : " ",
99                            (uint8_t)  xive_get_field64(EAS2_END_BLOCK, eas->w),
100                            (uint32_t) xive_get_field64(EAS2_END_INDEX, eas->w),
101                            (uint32_t) xive_get_field64(EAS2_END_DATA, eas->w));
102 }
103 
104 void xive2_end_queue_pic_print_info(Xive2End *end, uint32_t width, GString *buf)
105 {
106     uint64_t qaddr_base = xive2_end_qaddr(end);
107     uint32_t qsize = xive_get_field32(END2_W3_QSIZE, end->w3);
108     uint32_t qindex = xive_get_field32(END2_W1_PAGE_OFF, end->w1);
109     uint32_t qentries = 1 << (qsize + 10);
110     int i;
111 
112     /*
113      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
114      */
115     g_string_append_printf(buf, " [ ");
116     qindex = (qindex - (width - 1)) & (qentries - 1);
117     for (i = 0; i < width; i++) {
118         uint64_t qaddr = qaddr_base + (qindex << 2);
119         uint32_t qdata = -1;
120 
121         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
122                             sizeof(qdata), MEMTXATTRS_UNSPECIFIED)) {
123             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
124                           HWADDR_PRIx "\n", qaddr);
125             return;
126         }
127         g_string_append_printf(buf, "%s%08x ", i == width - 1 ? "^" : "",
128                                be32_to_cpu(qdata));
129         qindex = (qindex + 1) & (qentries - 1);
130     }
131     g_string_append_printf(buf, "]");
132 }
133 
134 void xive2_end_pic_print_info(Xive2End *end, uint32_t end_idx, GString *buf)
135 {
136     uint64_t qaddr_base = xive2_end_qaddr(end);
137     uint32_t qindex = xive_get_field32(END2_W1_PAGE_OFF, end->w1);
138     uint32_t qgen = xive_get_field32(END2_W1_GENERATION, end->w1);
139     uint32_t qsize = xive_get_field32(END2_W3_QSIZE, end->w3);
140     uint32_t qentries = 1 << (qsize + 10);
141 
142     uint32_t nvp_blk = xive_get_field32(END2_W6_VP_BLOCK, end->w6);
143     uint32_t nvp_idx = xive_get_field32(END2_W6_VP_OFFSET, end->w6);
144     uint8_t priority = xive_get_field32(END2_W7_F0_PRIORITY, end->w7);
145     uint8_t pq;
146 
147     if (!xive2_end_is_valid(end)) {
148         return;
149     }
150 
151     pq = xive_get_field32(END2_W1_ESn, end->w1);
152 
153     g_string_append_printf(buf,
154                            "  %08x %c%c %c%c%c%c%c%c%c%c%c%c%c %c%c "
155                            "prio:%d nvp:%02x/%04x",
156                            end_idx,
157                            pq & XIVE_ESB_VAL_P ? 'P' : '-',
158                            pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
159                            xive2_end_is_valid(end)    ? 'v' : '-',
160                            xive2_end_is_enqueue(end)  ? 'q' : '-',
161                            xive2_end_is_notify(end)   ? 'n' : '-',
162                            xive2_end_is_backlog(end)  ? 'b' : '-',
163                            xive2_end_is_precluded_escalation(end) ? 'p' : '-',
164                            xive2_end_is_escalate(end) ? 'e' : '-',
165                            xive2_end_is_escalate_end(end) ? 'N' : '-',
166                            xive2_end_is_uncond_escalation(end)   ? 'u' : '-',
167                            xive2_end_is_silent_escalation(end)   ? 's' : '-',
168                            xive2_end_is_firmware1(end)   ? 'f' : '-',
169                            xive2_end_is_firmware2(end)   ? 'F' : '-',
170                            xive2_end_is_ignore(end) ? 'i' : '-',
171                            xive2_end_is_crowd(end)  ? 'c' : '-',
172                            priority, nvp_blk, nvp_idx);
173 
174     if (qaddr_base) {
175         g_string_append_printf(buf, " eq:@%08"PRIx64"% 6d/%5d ^%d",
176                                qaddr_base, qindex, qentries, qgen);
177         xive2_end_queue_pic_print_info(end, 6, buf);
178     }
179     g_string_append_c(buf, '\n');
180 }
181 
182 void xive2_end_eas_pic_print_info(Xive2End *end, uint32_t end_idx,
183                                   GString *buf)
184 {
185     Xive2Eas *eas = (Xive2Eas *) &end->w4;
186     uint8_t pq;
187 
188     if (!xive2_end_is_escalate(end)) {
189         return;
190     }
191 
192     pq = xive_get_field32(END2_W1_ESe, end->w1);
193 
194     g_string_append_printf(buf, "  %08x %c%c %c%c end:%02x/%04x data:%08x\n",
195                            end_idx,
196                            pq & XIVE_ESB_VAL_P ? 'P' : '-',
197                            pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
198                            xive2_eas_is_valid(eas) ? 'v' : ' ',
199                            xive2_eas_is_masked(eas) ? 'M' : ' ',
200                            (uint8_t)  xive_get_field64(EAS2_END_BLOCK, eas->w),
201                            (uint32_t) xive_get_field64(EAS2_END_INDEX, eas->w),
202                            (uint32_t) xive_get_field64(EAS2_END_DATA, eas->w));
203 }
204 
205 void xive2_nvp_pic_print_info(Xive2Nvp *nvp, uint32_t nvp_idx, GString *buf)
206 {
207     uint8_t  eq_blk = xive_get_field32(NVP2_W5_VP_END_BLOCK, nvp->w5);
208     uint32_t eq_idx = xive_get_field32(NVP2_W5_VP_END_INDEX, nvp->w5);
209     uint64_t cache_line = xive2_nvp_reporting_addr(nvp);
210 
211     if (!xive2_nvp_is_valid(nvp)) {
212         return;
213     }
214 
215     g_string_append_printf(buf, "  %08x end:%02x/%04x IPB:%02x PGoFirst:%02x",
216                            nvp_idx, eq_blk, eq_idx,
217                            xive_get_field32(NVP2_W2_IPB, nvp->w2),
218                            xive_get_field32(NVP2_W0_PGOFIRST, nvp->w0));
219     if (cache_line) {
220         g_string_append_printf(buf, "  reporting CL:%016"PRIx64, cache_line);
221     }
222 
223     /*
224      * When the NVP is HW controlled, more fields are updated
225      */
226     if (xive2_nvp_is_hw(nvp)) {
227         g_string_append_printf(buf, " CPPR:%02x",
228                                xive_get_field32(NVP2_W2_CPPR, nvp->w2));
229         if (xive2_nvp_is_co(nvp)) {
230             g_string_append_printf(buf, " CO:%04x",
231                                    xive_get_field32(NVP2_W1_CO_THRID, nvp->w1));
232         }
233     }
234     g_string_append_c(buf, '\n');
235 }
236 
237 void xive2_nvgc_pic_print_info(Xive2Nvgc *nvgc, uint32_t nvgc_idx, GString *buf)
238 {
239     uint8_t i;
240 
241     if (!xive2_nvgc_is_valid(nvgc)) {
242         return;
243     }
244 
245     g_string_append_printf(buf, "  %08x PGoNext:%02x bklog: ", nvgc_idx,
246                            xive_get_field32(NVGC2_W0_PGONEXT, nvgc->w0));
247     for (i = 0; i <= XIVE_PRIORITY_MAX; i++) {
248         g_string_append_printf(buf, "[%d]=0x%x ",
249                                i, xive2_nvgc_get_backlog(nvgc, i));
250     }
251     g_string_append_printf(buf, "\n");
252 }
253 
254 static void xive2_end_enqueue(Xive2End *end, uint32_t data)
255 {
256     uint64_t qaddr_base = xive2_end_qaddr(end);
257     uint32_t qsize = xive_get_field32(END2_W3_QSIZE, end->w3);
258     uint32_t qindex = xive_get_field32(END2_W1_PAGE_OFF, end->w1);
259     uint32_t qgen = xive_get_field32(END2_W1_GENERATION, end->w1);
260 
261     uint64_t qaddr = qaddr_base + (qindex << 2);
262     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
263     uint32_t qentries = 1 << (qsize + 10);
264 
265     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata),
266                          MEMTXATTRS_UNSPECIFIED)) {
267         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
268                       HWADDR_PRIx "\n", qaddr);
269         return;
270     }
271 
272     qindex = (qindex + 1) & (qentries - 1);
273     if (qindex == 0) {
274         qgen ^= 1;
275         end->w1 = xive_set_field32(END2_W1_GENERATION, end->w1, qgen);
276 
277         /* TODO(PowerNV): reset GF bit on a cache watch operation */
278         end->w1 = xive_set_field32(END2_W1_GEN_FLIPPED, end->w1, qgen);
279     }
280     end->w1 = xive_set_field32(END2_W1_PAGE_OFF, end->w1, qindex);
281 }
282 
283 /*
284  * Scan the group chain and return the highest priority and group
285  * level of pending group interrupts.
286  */
287 static uint8_t xive2_presenter_backlog_scan(XivePresenter *xptr,
288                                             uint8_t nvp_blk, uint32_t nvp_idx,
289                                             uint8_t first_group,
290                                             uint8_t *out_level)
291 {
292     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
293     uint32_t nvgc_idx, mask;
294     uint32_t current_level, count;
295     uint8_t prio;
296     Xive2Nvgc nvgc;
297 
298     for (prio = 0; prio <= XIVE_PRIORITY_MAX; prio++) {
299         current_level = first_group & 0xF;
300 
301         while (current_level) {
302             mask = (1 << current_level) - 1;
303             nvgc_idx = nvp_idx & ~mask;
304             nvgc_idx |= mask >> 1;
305             qemu_log("fxb %s checking backlog for prio %d group idx %x\n",
306                      __func__, prio, nvgc_idx);
307 
308             if (xive2_router_get_nvgc(xrtr, false, nvp_blk, nvgc_idx, &nvgc)) {
309                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVG %x/%x\n",
310                               nvp_blk, nvgc_idx);
311                 return 0xFF;
312             }
313             if (!xive2_nvgc_is_valid(&nvgc)) {
314                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid NVG %x/%x\n",
315                               nvp_blk, nvgc_idx);
316                 return 0xFF;
317             }
318 
319             count = xive2_nvgc_get_backlog(&nvgc, prio);
320             if (count) {
321                 *out_level = current_level;
322                 return prio;
323             }
324             current_level = xive_get_field32(NVGC2_W0_PGONEXT, nvgc.w0) & 0xF;
325         }
326     }
327     return 0xFF;
328 }
329 
330 static void xive2_presenter_backlog_decr(XivePresenter *xptr,
331                                          uint8_t nvp_blk, uint32_t nvp_idx,
332                                          uint8_t group_prio,
333                                          uint8_t group_level)
334 {
335     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
336     uint32_t nvgc_idx, mask, count;
337     Xive2Nvgc nvgc;
338 
339     group_level &= 0xF;
340     mask = (1 << group_level) - 1;
341     nvgc_idx = nvp_idx & ~mask;
342     nvgc_idx |= mask >> 1;
343 
344     if (xive2_router_get_nvgc(xrtr, false, nvp_blk, nvgc_idx, &nvgc)) {
345         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVG %x/%x\n",
346                       nvp_blk, nvgc_idx);
347         return;
348     }
349     if (!xive2_nvgc_is_valid(&nvgc)) {
350         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid NVG %x/%x\n",
351                       nvp_blk, nvgc_idx);
352         return;
353     }
354     count = xive2_nvgc_get_backlog(&nvgc, group_prio);
355     if (!count) {
356         return;
357     }
358     xive2_nvgc_set_backlog(&nvgc, group_prio, count - 1);
359     xive2_router_write_nvgc(xrtr, false, nvp_blk, nvgc_idx, &nvgc);
360 }
361 
362 /*
363  * XIVE Thread Interrupt Management Area (TIMA) - Gen2 mode
364  *
365  * TIMA Gen2 VP “save & restore” (S&R) indicated by H bit next to V bit
366  *
367  *   - if a context is enabled with the H bit set, the VP context
368  *     information is retrieved from the NVP structure (“check out”)
369  *     and stored back on a context pull (“check in”), the SW receives
370  *     the same context pull information as on P9
371  *
372  *   - the H bit cannot be changed while the V bit is set, i.e. a
373  *     context cannot be set up in the TIMA and then be “pushed” into
374  *     the NVP by changing the H bit while the context is enabled
375  */
376 
377 static void xive2_tctx_save_ctx(Xive2Router *xrtr, XiveTCTX *tctx,
378                                 uint8_t nvp_blk, uint32_t nvp_idx,
379                                 uint8_t ring)
380 {
381     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
382     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
383     Xive2Nvp nvp;
384     uint8_t *regs = &tctx->regs[ring];
385 
386     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
387         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
388                           nvp_blk, nvp_idx);
389         return;
390     }
391 
392     if (!xive2_nvp_is_valid(&nvp)) {
393         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
394                       nvp_blk, nvp_idx);
395         return;
396     }
397 
398     if (!xive2_nvp_is_hw(&nvp)) {
399         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is not HW owned\n",
400                       nvp_blk, nvp_idx);
401         return;
402     }
403 
404     if (!xive2_nvp_is_co(&nvp)) {
405         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is not checkout\n",
406                       nvp_blk, nvp_idx);
407         return;
408     }
409 
410     if (xive_get_field32(NVP2_W1_CO_THRID_VALID, nvp.w1) &&
411         xive_get_field32(NVP2_W1_CO_THRID, nvp.w1) != pir) {
412         qemu_log_mask(LOG_GUEST_ERROR,
413                       "XIVE: NVP %x/%x invalid checkout Thread %x\n",
414                       nvp_blk, nvp_idx, pir);
415         return;
416     }
417 
418     nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, regs[TM_IPB]);
419     nvp.w2 = xive_set_field32(NVP2_W2_CPPR, nvp.w2, regs[TM_CPPR]);
420     if (nvp.w0 & NVP2_W0_L) {
421         /*
422          * Typically not used. If LSMFB is restored with 0, it will
423          * force a backlog rescan
424          */
425         nvp.w2 = xive_set_field32(NVP2_W2_LSMFB, nvp.w2, regs[TM_LSMFB]);
426     }
427     if (nvp.w0 & NVP2_W0_G) {
428         nvp.w2 = xive_set_field32(NVP2_W2_LGS, nvp.w2, regs[TM_LGS]);
429     }
430     if (nvp.w0 & NVP2_W0_T) {
431         nvp.w2 = xive_set_field32(NVP2_W2_T, nvp.w2, regs[TM_T]);
432     }
433     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, &nvp, 2);
434 
435     nvp.w1 = xive_set_field32(NVP2_W1_CO, nvp.w1, 0);
436     /* NVP2_W1_CO_THRID_VALID only set once */
437     nvp.w1 = xive_set_field32(NVP2_W1_CO_THRID, nvp.w1, 0xFFFF);
438     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, &nvp, 1);
439 }
440 
441 static void xive2_cam_decode(uint32_t cam, uint8_t *nvp_blk,
442                              uint32_t *nvp_idx, bool *valid, bool *hw)
443 {
444     *nvp_blk = xive2_nvp_blk(cam);
445     *nvp_idx = xive2_nvp_idx(cam);
446     *valid = !!(cam & TM2_W2_VALID);
447     *hw = !!(cam & TM2_W2_HW);
448 }
449 
450 /*
451  * Encode the HW CAM line with 7bit or 8bit thread id. The thread id
452  * width and block id width is configurable at the IC level.
453  *
454  *    chipid << 24 | 0000 0000 0000 0000 1 threadid (7Bit)
455  *    chipid << 24 | 0000 0000 0000 0001 threadid   (8Bit)
456  */
457 static uint32_t xive2_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
458 {
459     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
460     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
461     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
462     uint8_t blk = xive2_router_get_block_id(xrtr);
463     uint8_t tid_shift =
464         xive2_router_get_config(xrtr) & XIVE2_THREADID_8BITS ? 8 : 7;
465     uint8_t tid_mask = (1 << tid_shift) - 1;
466 
467     return xive2_nvp_cam_line(blk, 1 << tid_shift | (pir & tid_mask));
468 }
469 
470 static uint64_t xive2_tm_pull_ctx(XivePresenter *xptr, XiveTCTX *tctx,
471                                   hwaddr offset, unsigned size, uint8_t ring)
472 {
473     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
474     uint32_t target_ringw2 = xive_tctx_word2(&tctx->regs[ring]);
475     uint32_t cam = be32_to_cpu(target_ringw2);
476     uint8_t nvp_blk;
477     uint32_t nvp_idx;
478     uint8_t cur_ring;
479     bool valid;
480     bool do_save;
481 
482     xive2_cam_decode(cam, &nvp_blk, &nvp_idx, &valid, &do_save);
483 
484     if (!valid) {
485         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: pulling invalid NVP %x/%x !?\n",
486                       nvp_blk, nvp_idx);
487     }
488 
489     /* Invalidate CAM line of requested ring and all lower rings */
490     for (cur_ring = TM_QW0_USER; cur_ring <= ring;
491          cur_ring += XIVE_TM_RING_SIZE) {
492         uint32_t ringw2 = xive_tctx_word2(&tctx->regs[cur_ring]);
493         uint32_t ringw2_new = xive_set_field32(TM2_QW1W2_VO, ringw2, 0);
494         memcpy(&tctx->regs[cur_ring + TM_WORD2], &ringw2_new, 4);
495     }
496 
497     if (xive2_router_get_config(xrtr) & XIVE2_VP_SAVE_RESTORE && do_save) {
498         xive2_tctx_save_ctx(xrtr, tctx, nvp_blk, nvp_idx, ring);
499     }
500 
501     /*
502      * Lower external interrupt line of requested ring and below except for
503      * USER, which doesn't exist.
504      */
505     for (cur_ring = TM_QW1_OS; cur_ring <= ring;
506          cur_ring += XIVE_TM_RING_SIZE) {
507         xive_tctx_reset_signal(tctx, cur_ring);
508     }
509     return target_ringw2;
510 }
511 
512 uint64_t xive2_tm_pull_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
513                               hwaddr offset, unsigned size)
514 {
515     return xive2_tm_pull_ctx(xptr, tctx, offset, size, TM_QW1_OS);
516 }
517 
518 #define REPORT_LINE_GEN1_SIZE       16
519 
520 static void xive2_tm_report_line_gen1(XiveTCTX *tctx, uint8_t *data,
521                                       uint8_t size)
522 {
523     uint8_t *regs = tctx->regs;
524 
525     g_assert(size == REPORT_LINE_GEN1_SIZE);
526     memset(data, 0, size);
527     /*
528      * See xive architecture for description of what is saved. It is
529      * hand-picked information to fit in 16 bytes.
530      */
531     data[0x0] = regs[TM_QW3_HV_PHYS + TM_NSR];
532     data[0x1] = regs[TM_QW3_HV_PHYS + TM_CPPR];
533     data[0x2] = regs[TM_QW3_HV_PHYS + TM_IPB];
534     data[0x3] = regs[TM_QW2_HV_POOL + TM_IPB];
535     data[0x4] = regs[TM_QW1_OS + TM_ACK_CNT];
536     data[0x5] = regs[TM_QW3_HV_PHYS + TM_LGS];
537     data[0x6] = 0xFF;
538     data[0x7] = regs[TM_QW3_HV_PHYS + TM_WORD2] & 0x80;
539     data[0x7] |= (regs[TM_QW2_HV_POOL + TM_WORD2] & 0x80) >> 1;
540     data[0x7] |= (regs[TM_QW1_OS + TM_WORD2] & 0x80) >> 2;
541     data[0x7] |= (regs[TM_QW3_HV_PHYS + TM_WORD2] & 0x3);
542     data[0x8] = regs[TM_QW1_OS + TM_NSR];
543     data[0x9] = regs[TM_QW1_OS + TM_CPPR];
544     data[0xA] = regs[TM_QW1_OS + TM_IPB];
545     data[0xB] = regs[TM_QW1_OS + TM_LGS];
546     if (regs[TM_QW0_USER + TM_WORD2] & 0x80) {
547         /*
548          * Logical server extension, except VU bit replaced by EB bit
549          * from NSR
550          */
551         data[0xC] = regs[TM_QW0_USER + TM_WORD2];
552         data[0xC] &= ~0x80;
553         data[0xC] |= regs[TM_QW0_USER + TM_NSR] & 0x80;
554         data[0xD] = regs[TM_QW0_USER + TM_WORD2 + 1];
555         data[0xE] = regs[TM_QW0_USER + TM_WORD2 + 2];
556         data[0xF] = regs[TM_QW0_USER + TM_WORD2 + 3];
557     }
558 }
559 
560 static void xive2_tm_pull_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
561                                  hwaddr offset, uint64_t value,
562                                  unsigned size, uint8_t ring)
563 {
564     Xive2Router *xrtr = XIVE2_ROUTER(xptr);
565     uint32_t hw_cam, nvp_idx, xive2_cfg, reserved;
566     uint8_t nvp_blk;
567     Xive2Nvp nvp;
568     uint64_t phys_addr;
569     MemTxResult result;
570 
571     hw_cam = xive2_tctx_hw_cam_line(xptr, tctx);
572     nvp_blk = xive2_nvp_blk(hw_cam);
573     nvp_idx = xive2_nvp_idx(hw_cam);
574 
575     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
576         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
577                       nvp_blk, nvp_idx);
578         return;
579     }
580 
581     if (!xive2_nvp_is_valid(&nvp)) {
582         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
583                       nvp_blk, nvp_idx);
584         return;
585     }
586 
587     xive2_cfg = xive2_router_get_config(xrtr);
588 
589     phys_addr = xive2_nvp_reporting_addr(&nvp) + 0x80; /* odd line */
590     if (xive2_cfg & XIVE2_GEN1_TIMA_OS) {
591         uint8_t pull_ctxt[REPORT_LINE_GEN1_SIZE];
592 
593         xive2_tm_report_line_gen1(tctx, pull_ctxt, REPORT_LINE_GEN1_SIZE);
594         result = dma_memory_write(&address_space_memory, phys_addr,
595                                   pull_ctxt, REPORT_LINE_GEN1_SIZE,
596                                   MEMTXATTRS_UNSPECIFIED);
597         assert(result == MEMTX_OK);
598     } else {
599         result = dma_memory_write(&address_space_memory, phys_addr,
600                                   &tctx->regs, sizeof(tctx->regs),
601                                   MEMTXATTRS_UNSPECIFIED);
602         assert(result == MEMTX_OK);
603         reserved = 0xFFFFFFFF;
604         result = dma_memory_write(&address_space_memory, phys_addr + 12,
605                                   &reserved, sizeof(reserved),
606                                   MEMTXATTRS_UNSPECIFIED);
607         assert(result == MEMTX_OK);
608     }
609 
610     /* the rest is similar to pull context to registers */
611     xive2_tm_pull_ctx(xptr, tctx, offset, size, ring);
612 }
613 
614 void xive2_tm_pull_os_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
615                              hwaddr offset, uint64_t value, unsigned size)
616 {
617     xive2_tm_pull_ctx_ol(xptr, tctx, offset, value, size, TM_QW1_OS);
618 }
619 
620 
621 void xive2_tm_pull_phys_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
622                                hwaddr offset, uint64_t value, unsigned size)
623 {
624     xive2_tm_pull_ctx_ol(xptr, tctx, offset, value, size, TM_QW3_HV_PHYS);
625 }
626 
627 static uint8_t xive2_tctx_restore_os_ctx(Xive2Router *xrtr, XiveTCTX *tctx,
628                                         uint8_t nvp_blk, uint32_t nvp_idx,
629                                         Xive2Nvp *nvp)
630 {
631     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
632     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
633     uint8_t cppr;
634 
635     if (!xive2_nvp_is_hw(nvp)) {
636         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is not HW owned\n",
637                       nvp_blk, nvp_idx);
638         return 0;
639     }
640 
641     cppr = xive_get_field32(NVP2_W2_CPPR, nvp->w2);
642     nvp->w2 = xive_set_field32(NVP2_W2_CPPR, nvp->w2, 0);
643     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, nvp, 2);
644 
645     tctx->regs[TM_QW1_OS + TM_CPPR] = cppr;
646     tctx->regs[TM_QW1_OS + TM_LSMFB] = xive_get_field32(NVP2_W2_LSMFB, nvp->w2);
647     tctx->regs[TM_QW1_OS + TM_LGS] = xive_get_field32(NVP2_W2_LGS, nvp->w2);
648     tctx->regs[TM_QW1_OS + TM_T] = xive_get_field32(NVP2_W2_T, nvp->w2);
649 
650     nvp->w1 = xive_set_field32(NVP2_W1_CO, nvp->w1, 1);
651     nvp->w1 = xive_set_field32(NVP2_W1_CO_THRID_VALID, nvp->w1, 1);
652     nvp->w1 = xive_set_field32(NVP2_W1_CO_THRID, nvp->w1, pir);
653 
654     /*
655      * Checkout privilege: 0:OS, 1:Pool, 2:Hard
656      *
657      * TODO: we only support OS push/pull
658      */
659     nvp->w1 = xive_set_field32(NVP2_W1_CO_PRIV, nvp->w1, 0);
660 
661     xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, nvp, 1);
662 
663     /* return restored CPPR to generate a CPU exception if needed */
664     return cppr;
665 }
666 
667 static void xive2_tctx_need_resend(Xive2Router *xrtr, XiveTCTX *tctx,
668                                    uint8_t nvp_blk, uint32_t nvp_idx,
669                                    bool do_restore)
670 {
671     XivePresenter *xptr = XIVE_PRESENTER(xrtr);
672     uint8_t ipb;
673     uint8_t backlog_level;
674     uint8_t group_level;
675     uint8_t first_group;
676     uint8_t backlog_prio;
677     uint8_t group_prio;
678     uint8_t *regs = &tctx->regs[TM_QW1_OS];
679     Xive2Nvp nvp;
680 
681     /*
682      * Grab the associated thread interrupt context registers in the
683      * associated NVP
684      */
685     if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
686         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
687                       nvp_blk, nvp_idx);
688         return;
689     }
690 
691     if (!xive2_nvp_is_valid(&nvp)) {
692         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
693                       nvp_blk, nvp_idx);
694         return;
695     }
696 
697     /* Automatically restore thread context registers */
698     if (xive2_router_get_config(xrtr) & XIVE2_VP_SAVE_RESTORE &&
699         do_restore) {
700         xive2_tctx_restore_os_ctx(xrtr, tctx, nvp_blk, nvp_idx, &nvp);
701     }
702 
703     ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2);
704     if (ipb) {
705         nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, 0);
706         xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, &nvp, 2);
707     }
708     regs[TM_IPB] |= ipb;
709     backlog_prio = xive_ipb_to_pipr(ipb);
710     backlog_level = 0;
711 
712     first_group = xive_get_field32(NVP2_W0_PGOFIRST, nvp.w0);
713     if (first_group && regs[TM_LSMFB] < backlog_prio) {
714         group_prio = xive2_presenter_backlog_scan(xptr, nvp_blk, nvp_idx,
715                                                   first_group, &group_level);
716         regs[TM_LSMFB] = group_prio;
717         if (regs[TM_LGS] && group_prio < backlog_prio) {
718             /* VP can take a group interrupt */
719             xive2_presenter_backlog_decr(xptr, nvp_blk, nvp_idx,
720                                          group_prio, group_level);
721             backlog_prio = group_prio;
722             backlog_level = group_level;
723         }
724     }
725 
726     /*
727      * Compute the PIPR based on the restored state.
728      * It will raise the External interrupt signal if needed.
729      */
730     xive_tctx_pipr_update(tctx, TM_QW1_OS, backlog_prio, backlog_level);
731 }
732 
733 /*
734  * Updating the OS CAM line can trigger a resend of interrupt
735  */
736 void xive2_tm_push_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
737                           hwaddr offset, uint64_t value, unsigned size)
738 {
739     uint32_t cam;
740     uint32_t qw1w2;
741     uint64_t qw1dw1;
742     uint8_t nvp_blk;
743     uint32_t nvp_idx;
744     bool vo;
745     bool do_restore;
746 
747     /* First update the thead context */
748     switch (size) {
749     case 4:
750         cam = value;
751         qw1w2 = cpu_to_be32(cam);
752         memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
753         break;
754     case 8:
755         cam = value >> 32;
756         qw1dw1 = cpu_to_be64(value);
757         memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1dw1, 8);
758         break;
759     default:
760         g_assert_not_reached();
761     }
762 
763     xive2_cam_decode(cam, &nvp_blk, &nvp_idx, &vo, &do_restore);
764 
765     /* Check the interrupt pending bits */
766     if (vo) {
767         xive2_tctx_need_resend(XIVE2_ROUTER(xptr), tctx, nvp_blk, nvp_idx,
768                                do_restore);
769     }
770 }
771 
772 static int xive2_tctx_get_nvp_indexes(XiveTCTX *tctx, uint8_t ring,
773                                       uint32_t *nvp_blk, uint32_t *nvp_idx)
774 {
775     uint32_t w2, cam;
776 
777     w2 = xive_tctx_word2(&tctx->regs[ring]);
778     switch (ring) {
779     case TM_QW1_OS:
780         if (!(be32_to_cpu(w2) & TM2_QW1W2_VO)) {
781             return -1;
782         }
783         cam = xive_get_field32(TM2_QW1W2_OS_CAM, w2);
784         break;
785     case TM_QW2_HV_POOL:
786         if (!(be32_to_cpu(w2) & TM2_QW2W2_VP)) {
787             return -1;
788         }
789         cam = xive_get_field32(TM2_QW2W2_POOL_CAM, w2);
790         break;
791     case TM_QW3_HV_PHYS:
792         if (!(be32_to_cpu(w2) & TM2_QW3W2_VT)) {
793             return -1;
794         }
795         cam = xive2_tctx_hw_cam_line(tctx->xptr, tctx);
796         break;
797     default:
798         return -1;
799     }
800     *nvp_blk = xive2_nvp_blk(cam);
801     *nvp_idx = xive2_nvp_idx(cam);
802     return 0;
803 }
804 
805 static void xive2_tctx_set_cppr(XiveTCTX *tctx, uint8_t ring, uint8_t cppr)
806 {
807     uint8_t *regs = &tctx->regs[ring];
808     Xive2Router *xrtr = XIVE2_ROUTER(tctx->xptr);
809     uint8_t old_cppr, backlog_prio, first_group, group_level = 0;
810     uint8_t pipr_min, lsmfb_min, ring_min;
811     bool group_enabled;
812     uint32_t nvp_blk, nvp_idx;
813     Xive2Nvp nvp;
814     int rc;
815 
816     trace_xive_tctx_set_cppr(tctx->cs->cpu_index, ring,
817                              regs[TM_IPB], regs[TM_PIPR],
818                              cppr, regs[TM_NSR]);
819 
820     if (cppr > XIVE_PRIORITY_MAX) {
821         cppr = 0xff;
822     }
823 
824     old_cppr = regs[TM_CPPR];
825     regs[TM_CPPR] = cppr;
826 
827     /*
828      * Recompute the PIPR based on local pending interrupts. It will
829      * be adjusted below if needed in case of pending group interrupts.
830      */
831     pipr_min = xive_ipb_to_pipr(regs[TM_IPB]);
832     group_enabled = !!regs[TM_LGS];
833     lsmfb_min = (group_enabled) ? regs[TM_LSMFB] : 0xff;
834     ring_min = ring;
835 
836     /* PHYS updates also depend on POOL values */
837     if (ring == TM_QW3_HV_PHYS) {
838         uint8_t *pregs = &tctx->regs[TM_QW2_HV_POOL];
839 
840         /* POOL values only matter if POOL ctx is valid */
841         if (pregs[TM_WORD2] & 0x80) {
842 
843             uint8_t pool_pipr = xive_ipb_to_pipr(pregs[TM_IPB]);
844             uint8_t pool_lsmfb = pregs[TM_LSMFB];
845 
846             /*
847              * Determine highest priority interrupt and
848              * remember which ring has it.
849              */
850             if (pool_pipr < pipr_min) {
851                 pipr_min = pool_pipr;
852                 if (pool_pipr < lsmfb_min) {
853                     ring_min = TM_QW2_HV_POOL;
854                 }
855             }
856 
857             /* Values needed for group priority calculation */
858             if (pregs[TM_LGS] && (pool_lsmfb < lsmfb_min)) {
859                 group_enabled = true;
860                 lsmfb_min = pool_lsmfb;
861                 if (lsmfb_min < pipr_min) {
862                     ring_min = TM_QW2_HV_POOL;
863                 }
864             }
865         }
866     }
867     regs[TM_PIPR] = pipr_min;
868 
869     rc = xive2_tctx_get_nvp_indexes(tctx, ring_min, &nvp_blk, &nvp_idx);
870     if (rc) {
871         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: set CPPR on invalid context\n");
872         return;
873     }
874 
875     if (cppr < old_cppr) {
876         /*
877          * FIXME: check if there's a group interrupt being presented
878          * and if the new cppr prevents it. If so, then the group
879          * interrupt needs to be re-added to the backlog and
880          * re-triggered (see re-trigger END info in the NVGC
881          * structure)
882          */
883     }
884 
885     if (group_enabled &&
886         lsmfb_min < cppr &&
887         lsmfb_min < regs[TM_PIPR]) {
888         /*
889          * Thread has seen a group interrupt with a higher priority
890          * than the new cppr or pending local interrupt. Check the
891          * backlog
892          */
893         if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
894             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No NVP %x/%x\n",
895                           nvp_blk, nvp_idx);
896             return;
897         }
898 
899         if (!xive2_nvp_is_valid(&nvp)) {
900             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
901                           nvp_blk, nvp_idx);
902             return;
903         }
904 
905         first_group = xive_get_field32(NVP2_W0_PGOFIRST, nvp.w0);
906         if (!first_group) {
907             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVP %x/%x\n",
908                           nvp_blk, nvp_idx);
909             return;
910         }
911 
912         backlog_prio = xive2_presenter_backlog_scan(tctx->xptr,
913                                                     nvp_blk, nvp_idx,
914                                                     first_group, &group_level);
915         tctx->regs[ring_min + TM_LSMFB] = backlog_prio;
916         if (backlog_prio != 0xFF) {
917             xive2_presenter_backlog_decr(tctx->xptr, nvp_blk, nvp_idx,
918                                          backlog_prio, group_level);
919             regs[TM_PIPR] = backlog_prio;
920         }
921     }
922     /* CPPR has changed, check if we need to raise a pending exception */
923     xive_tctx_notify(tctx, ring_min, group_level);
924 }
925 
926 void xive2_tm_set_hv_cppr(XivePresenter *xptr, XiveTCTX *tctx,
927                           hwaddr offset, uint64_t value, unsigned size)
928 {
929     xive2_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
930 }
931 
932 void xive2_tm_set_os_cppr(XivePresenter *xptr, XiveTCTX *tctx,
933                           hwaddr offset, uint64_t value, unsigned size)
934 {
935     xive2_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
936 }
937 
938 static void xive2_tctx_set_target(XiveTCTX *tctx, uint8_t ring, uint8_t target)
939 {
940     uint8_t *regs = &tctx->regs[ring];
941 
942     regs[TM_T] = target;
943 }
944 
945 void xive2_tm_set_hv_target(XivePresenter *xptr, XiveTCTX *tctx,
946                             hwaddr offset, uint64_t value, unsigned size)
947 {
948     xive2_tctx_set_target(tctx, TM_QW3_HV_PHYS, value & 0xff);
949 }
950 
951 /*
952  * XIVE Router (aka. Virtualization Controller or IVRE)
953  */
954 
955 int xive2_router_get_eas(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
956                          Xive2Eas *eas)
957 {
958     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
959 
960     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
961 }
962 
963 static
964 int xive2_router_get_pq(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
965                        uint8_t *pq)
966 {
967     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
968 
969     return xrc->get_pq(xrtr, eas_blk, eas_idx, pq);
970 }
971 
972 static
973 int xive2_router_set_pq(Xive2Router *xrtr, uint8_t eas_blk, uint32_t eas_idx,
974                        uint8_t *pq)
975 {
976     Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
977 
978     return xrc->set_pq(xrtr, eas_blk, eas_idx, pq);
979 }
980 
981 int xive2_router_get_end(Xive2Router *xrtr, uint8_t end_blk, uint32_t end_idx,
982                          Xive2End *end)
983 {
984    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
985 
986    return xrc->get_end(xrtr, end_blk, end_idx, end);
987 }
988 
989 int xive2_router_write_end(Xive2Router *xrtr, uint8_t end_blk, uint32_t end_idx,
990                            Xive2End *end, uint8_t word_number)
991 {
992    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
993 
994    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
995 }
996 
997 int xive2_router_get_nvp(Xive2Router *xrtr, uint8_t nvp_blk, uint32_t nvp_idx,
998                          Xive2Nvp *nvp)
999 {
1000    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1001 
1002    return xrc->get_nvp(xrtr, nvp_blk, nvp_idx, nvp);
1003 }
1004 
1005 int xive2_router_write_nvp(Xive2Router *xrtr, uint8_t nvp_blk, uint32_t nvp_idx,
1006                            Xive2Nvp *nvp, uint8_t word_number)
1007 {
1008    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1009 
1010    return xrc->write_nvp(xrtr, nvp_blk, nvp_idx, nvp, word_number);
1011 }
1012 
1013 int xive2_router_get_nvgc(Xive2Router *xrtr, bool crowd,
1014                           uint8_t nvgc_blk, uint32_t nvgc_idx,
1015                           Xive2Nvgc *nvgc)
1016 {
1017    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1018 
1019    return xrc->get_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
1020 }
1021 
1022 int xive2_router_write_nvgc(Xive2Router *xrtr, bool crowd,
1023                             uint8_t nvgc_blk, uint32_t nvgc_idx,
1024                             Xive2Nvgc *nvgc)
1025 {
1026    Xive2RouterClass *xrc = XIVE2_ROUTER_GET_CLASS(xrtr);
1027 
1028    return xrc->write_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
1029 }
1030 
1031 static bool xive2_vp_match_mask(uint32_t cam1, uint32_t cam2,
1032                                 uint32_t vp_mask)
1033 {
1034     return (cam1 & vp_mask) == (cam2 & vp_mask);
1035 }
1036 
1037 /*
1038  * The thread context register words are in big-endian format.
1039  */
1040 int xive2_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1041                                uint8_t format,
1042                                uint8_t nvt_blk, uint32_t nvt_idx,
1043                                bool cam_ignore, uint32_t logic_serv)
1044 {
1045     uint32_t cam =   xive2_nvp_cam_line(nvt_blk, nvt_idx);
1046     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1047     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1048     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1049     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1050 
1051     uint32_t vp_mask = 0xFFFFFFFF;
1052 
1053     if (format == 0) {
1054         /*
1055          * i=0: Specific NVT notification
1056          * i=1: VP-group notification (bits ignored at the end of the
1057          *      NVT identifier)
1058          */
1059         if (cam_ignore) {
1060             vp_mask = ~(xive_get_vpgroup_size(nvt_idx) - 1);
1061         }
1062 
1063         /* For VP-group notifications, threads with LGS=0 are excluded */
1064 
1065         /* PHYS ring */
1066         if ((be32_to_cpu(qw3w2) & TM2_QW3W2_VT) &&
1067             !(cam_ignore && tctx->regs[TM_QW3_HV_PHYS + TM_LGS] == 0) &&
1068             xive2_vp_match_mask(cam,
1069                                 xive2_tctx_hw_cam_line(xptr, tctx),
1070                                 vp_mask)) {
1071             return TM_QW3_HV_PHYS;
1072         }
1073 
1074         /* HV POOL ring */
1075         if ((be32_to_cpu(qw2w2) & TM2_QW2W2_VP) &&
1076             !(cam_ignore && tctx->regs[TM_QW2_HV_POOL + TM_LGS] == 0) &&
1077             xive2_vp_match_mask(cam,
1078                                 xive_get_field32(TM2_QW2W2_POOL_CAM, qw2w2),
1079                                 vp_mask)) {
1080             return TM_QW2_HV_POOL;
1081         }
1082 
1083         /* OS ring */
1084         if ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
1085             !(cam_ignore && tctx->regs[TM_QW1_OS + TM_LGS] == 0) &&
1086             xive2_vp_match_mask(cam,
1087                                 xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2),
1088                                 vp_mask)) {
1089             return TM_QW1_OS;
1090         }
1091     } else {
1092         /* F=1 : User level Event-Based Branch (EBB) notification */
1093 
1094         /* FIXME: what if cam_ignore and LGS = 0 ? */
1095         /* USER ring */
1096         if  ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
1097              (cam == xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2)) &&
1098              (be32_to_cpu(qw0w2) & TM2_QW0W2_VU) &&
1099              (logic_serv == xive_get_field32(TM2_QW0W2_LOGIC_SERV, qw0w2))) {
1100             return TM_QW0_USER;
1101         }
1102     }
1103     return -1;
1104 }
1105 
1106 bool xive2_tm_irq_precluded(XiveTCTX *tctx, int ring, uint8_t priority)
1107 {
1108     /* HV_POOL ring uses HV_PHYS NSR, CPPR and PIPR registers */
1109     uint8_t alt_ring = (ring == TM_QW2_HV_POOL) ? TM_QW3_HV_PHYS : ring;
1110     uint8_t *alt_regs = &tctx->regs[alt_ring];
1111 
1112     /*
1113      * The xive2_presenter_tctx_match() above tells if there's a match
1114      * but for VP-group notification, we still need to look at the
1115      * priority to know if the thread can take the interrupt now or if
1116      * it is precluded.
1117      */
1118     if (priority < alt_regs[TM_CPPR]) {
1119         return false;
1120     }
1121     return true;
1122 }
1123 
1124 void xive2_tm_set_lsmfb(XiveTCTX *tctx, int ring, uint8_t priority)
1125 {
1126     uint8_t *regs = &tctx->regs[ring];
1127 
1128     /*
1129      * Called by the router during a VP-group notification when the
1130      * thread matches but can't take the interrupt because it's
1131      * already running at a more favored priority. It then stores the
1132      * new interrupt priority in the LSMFB field.
1133      */
1134     regs[TM_LSMFB] = priority;
1135 }
1136 
1137 static void xive2_router_realize(DeviceState *dev, Error **errp)
1138 {
1139     Xive2Router *xrtr = XIVE2_ROUTER(dev);
1140 
1141     assert(xrtr->xfb);
1142 }
1143 
1144 /*
1145  * Notification using the END ESe/ESn bit (Event State Buffer for
1146  * escalation and notification). Profide further coalescing in the
1147  * Router.
1148  */
1149 static bool xive2_router_end_es_notify(Xive2Router *xrtr, uint8_t end_blk,
1150                                        uint32_t end_idx, Xive2End *end,
1151                                        uint32_t end_esmask)
1152 {
1153     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1154     bool notify = xive_esb_trigger(&pq);
1155 
1156     if (pq != xive_get_field32(end_esmask, end->w1)) {
1157         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1158         xive2_router_write_end(xrtr, end_blk, end_idx, end, 1);
1159     }
1160 
1161     /* ESe/n[Q]=1 : end of notification */
1162     return notify;
1163 }
1164 
1165 /*
1166  * An END trigger can come from an event trigger (IPI or HW) or from
1167  * another chip. We don't model the PowerBus but the END trigger
1168  * message has the same parameters than in the function below.
1169  */
1170 static void xive2_router_end_notify(Xive2Router *xrtr, uint8_t end_blk,
1171                                     uint32_t end_idx, uint32_t end_data)
1172 {
1173     Xive2End end;
1174     uint8_t priority;
1175     uint8_t format;
1176     bool found, precluded;
1177     uint8_t nvp_blk;
1178     uint32_t nvp_idx;
1179 
1180     /* END cache lookup */
1181     if (xive2_router_get_end(xrtr, end_blk, end_idx, &end)) {
1182         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1183                       end_idx);
1184         return;
1185     }
1186 
1187     if (!xive2_end_is_valid(&end)) {
1188         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1189                       end_blk, end_idx);
1190         return;
1191     }
1192 
1193     if (xive2_end_is_enqueue(&end)) {
1194         xive2_end_enqueue(&end, end_data);
1195         /* Enqueuing event data modifies the EQ toggle and index */
1196         xive2_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1197     }
1198 
1199     /*
1200      * When the END is silent, we skip the notification part.
1201      */
1202     if (xive2_end_is_silent_escalation(&end)) {
1203         goto do_escalation;
1204     }
1205 
1206     /*
1207      * The W7 format depends on the F bit in W6. It defines the type
1208      * of the notification :
1209      *
1210      *   F=0 : single or multiple NVP notification
1211      *   F=1 : User level Event-Based Branch (EBB) notification, no
1212      *         priority
1213      */
1214     format = xive_get_field32(END2_W6_FORMAT_BIT, end.w6);
1215     priority = xive_get_field32(END2_W7_F0_PRIORITY, end.w7);
1216 
1217     /* The END is masked */
1218     if (format == 0 && priority == 0xff) {
1219         return;
1220     }
1221 
1222     /*
1223      * Check the END ESn (Event State Buffer for notification) for
1224      * even further coalescing in the Router
1225      */
1226     if (!xive2_end_is_notify(&end)) {
1227         /* ESn[Q]=1 : end of notification */
1228         if (!xive2_router_end_es_notify(xrtr, end_blk, end_idx,
1229                                        &end, END2_W1_ESn)) {
1230             return;
1231         }
1232     }
1233 
1234     /*
1235      * Follows IVPE notification
1236      */
1237     nvp_blk = xive_get_field32(END2_W6_VP_BLOCK, end.w6);
1238     nvp_idx = xive_get_field32(END2_W6_VP_OFFSET, end.w6);
1239 
1240     found = xive_presenter_notify(xrtr->xfb, format, nvp_blk, nvp_idx,
1241                           xive2_end_is_ignore(&end),
1242                           priority,
1243                           xive_get_field32(END2_W7_F1_LOG_SERVER_ID, end.w7),
1244                           &precluded);
1245 
1246     /* TODO: Auto EOI. */
1247 
1248     if (found) {
1249         return;
1250     }
1251 
1252     /*
1253      * If no matching NVP is dispatched on a HW thread :
1254      * - specific VP: update the NVP structure if backlog is activated
1255      * - VP-group: update the backlog counter for that priority in the NVG
1256      */
1257     if (xive2_end_is_backlog(&end)) {
1258 
1259         if (format == 1) {
1260             qemu_log_mask(LOG_GUEST_ERROR,
1261                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1262                           end_blk, end_idx);
1263             return;
1264         }
1265 
1266         if (!xive2_end_is_ignore(&end)) {
1267             uint8_t ipb;
1268             Xive2Nvp nvp;
1269 
1270             /* NVP cache lookup */
1271             if (xive2_router_get_nvp(xrtr, nvp_blk, nvp_idx, &nvp)) {
1272                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVP %x/%x\n",
1273                               nvp_blk, nvp_idx);
1274                 return;
1275             }
1276 
1277             if (!xive2_nvp_is_valid(&nvp)) {
1278                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVP %x/%x is invalid\n",
1279                               nvp_blk, nvp_idx);
1280                 return;
1281             }
1282 
1283             /*
1284              * Record the IPB in the associated NVP structure for later
1285              * use. The presenter will resend the interrupt when the vCPU
1286              * is dispatched again on a HW thread.
1287              */
1288             ipb = xive_get_field32(NVP2_W2_IPB, nvp.w2) |
1289                 xive_priority_to_ipb(priority);
1290             nvp.w2 = xive_set_field32(NVP2_W2_IPB, nvp.w2, ipb);
1291             xive2_router_write_nvp(xrtr, nvp_blk, nvp_idx, &nvp, 2);
1292         } else {
1293             Xive2Nvgc nvg;
1294             uint32_t backlog;
1295 
1296             /* For groups, the per-priority backlog counters are in the NVG */
1297             if (xive2_router_get_nvgc(xrtr, false, nvp_blk, nvp_idx, &nvg)) {
1298                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVG %x/%x\n",
1299                               nvp_blk, nvp_idx);
1300                 return;
1301             }
1302 
1303             if (!xive2_nvgc_is_valid(&nvg)) {
1304                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVG %x/%x is invalid\n",
1305                               nvp_blk, nvp_idx);
1306                 return;
1307             }
1308 
1309             /*
1310              * Increment the backlog counter for that priority.
1311              * We only call broadcast the first time the counter is
1312              * incremented. broadcast will set the LSMFB field of the TIMA of
1313              * relevant threads so that they know an interrupt is pending.
1314              */
1315             backlog = xive2_nvgc_get_backlog(&nvg, priority) + 1;
1316             xive2_nvgc_set_backlog(&nvg, priority, backlog);
1317             xive2_router_write_nvgc(xrtr, false, nvp_blk, nvp_idx, &nvg);
1318 
1319             if (backlog == 1) {
1320                 XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xrtr->xfb);
1321                 xfc->broadcast(xrtr->xfb, nvp_blk, nvp_idx, priority);
1322 
1323                 if (!xive2_end_is_precluded_escalation(&end)) {
1324                     /*
1325                      * The interrupt will be picked up when the
1326                      * matching thread lowers its priority level
1327                      */
1328                     return;
1329                 }
1330             }
1331         }
1332     }
1333 
1334 do_escalation:
1335     /*
1336      * If activated, escalate notification using the ESe PQ bits and
1337      * the EAS in w4-5
1338      */
1339     if (!xive2_end_is_escalate(&end)) {
1340         return;
1341     }
1342 
1343     /*
1344      * Check the END ESe (Event State Buffer for escalation) for even
1345      * further coalescing in the Router
1346      */
1347     if (!xive2_end_is_uncond_escalation(&end)) {
1348         /* ESe[Q]=1 : end of escalation notification */
1349         if (!xive2_router_end_es_notify(xrtr, end_blk, end_idx,
1350                                        &end, END2_W1_ESe)) {
1351             return;
1352         }
1353     }
1354 
1355     /*
1356      * The END trigger becomes an Escalation trigger
1357      */
1358     xive2_router_end_notify(xrtr,
1359                            xive_get_field32(END2_W4_END_BLOCK,     end.w4),
1360                            xive_get_field32(END2_W4_ESC_END_INDEX, end.w4),
1361                            xive_get_field32(END2_W5_ESC_END_DATA,  end.w5));
1362 }
1363 
1364 void xive2_router_notify(XiveNotifier *xn, uint32_t lisn, bool pq_checked)
1365 {
1366     Xive2Router *xrtr = XIVE2_ROUTER(xn);
1367     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1368     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1369     Xive2Eas eas;
1370 
1371     /* EAS cache lookup */
1372     if (xive2_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1373         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1374         return;
1375     }
1376 
1377     if (!pq_checked) {
1378         bool notify;
1379         uint8_t pq;
1380 
1381         /* PQ cache lookup */
1382         if (xive2_router_get_pq(xrtr, eas_blk, eas_idx, &pq)) {
1383             /* Set FIR */
1384             g_assert_not_reached();
1385         }
1386 
1387         notify = xive_esb_trigger(&pq);
1388 
1389         if (xive2_router_set_pq(xrtr, eas_blk, eas_idx, &pq)) {
1390             /* Set FIR */
1391             g_assert_not_reached();
1392         }
1393 
1394         if (!notify) {
1395             return;
1396         }
1397     }
1398 
1399     if (!xive2_eas_is_valid(&eas)) {
1400         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN %x\n", lisn);
1401         return;
1402     }
1403 
1404     if (xive2_eas_is_masked(&eas)) {
1405         /* Notification completed */
1406         return;
1407     }
1408 
1409     /*
1410      * The event trigger becomes an END trigger
1411      */
1412     xive2_router_end_notify(xrtr,
1413                              xive_get_field64(EAS2_END_BLOCK, eas.w),
1414                              xive_get_field64(EAS2_END_INDEX, eas.w),
1415                              xive_get_field64(EAS2_END_DATA,  eas.w));
1416 }
1417 
1418 static const Property xive2_router_properties[] = {
1419     DEFINE_PROP_LINK("xive-fabric", Xive2Router, xfb,
1420                      TYPE_XIVE_FABRIC, XiveFabric *),
1421 };
1422 
1423 static void xive2_router_class_init(ObjectClass *klass, void *data)
1424 {
1425     DeviceClass *dc = DEVICE_CLASS(klass);
1426     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1427 
1428     dc->desc    = "XIVE2 Router Engine";
1429     device_class_set_props(dc, xive2_router_properties);
1430     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1431     dc->realize = xive2_router_realize;
1432     xnc->notify = xive2_router_notify;
1433 }
1434 
1435 static const TypeInfo xive2_router_info = {
1436     .name          = TYPE_XIVE2_ROUTER,
1437     .parent        = TYPE_SYS_BUS_DEVICE,
1438     .abstract      = true,
1439     .instance_size = sizeof(Xive2Router),
1440     .class_size    = sizeof(Xive2RouterClass),
1441     .class_init    = xive2_router_class_init,
1442     .interfaces    = (InterfaceInfo[]) {
1443         { TYPE_XIVE_NOTIFIER },
1444         { TYPE_XIVE_PRESENTER },
1445         { }
1446     }
1447 };
1448 
1449 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
1450 {
1451     return !((addr >> shift) & 1);
1452 }
1453 
1454 static uint64_t xive2_end_source_read(void *opaque, hwaddr addr, unsigned size)
1455 {
1456     Xive2EndSource *xsrc = XIVE2_END_SOURCE(opaque);
1457     uint32_t offset = addr & 0xFFF;
1458     uint8_t end_blk;
1459     uint32_t end_idx;
1460     Xive2End end;
1461     uint32_t end_esmask;
1462     uint8_t pq;
1463     uint64_t ret;
1464 
1465     /*
1466      * The block id should be deduced from the load address on the END
1467      * ESB MMIO but our model only supports a single block per XIVE chip.
1468      */
1469     end_blk = xive2_router_get_block_id(xsrc->xrtr);
1470     end_idx = addr >> (xsrc->esb_shift + 1);
1471 
1472     if (xive2_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1473         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1474                       end_idx);
1475         return -1;
1476     }
1477 
1478     if (!xive2_end_is_valid(&end)) {
1479         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1480                       end_blk, end_idx);
1481         return -1;
1482     }
1483 
1484     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END2_W1_ESn :
1485         END2_W1_ESe;
1486     pq = xive_get_field32(end_esmask, end.w1);
1487 
1488     switch (offset) {
1489     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1490         ret = xive_esb_eoi(&pq);
1491 
1492         /* Forward the source event notification for routing ?? */
1493         break;
1494 
1495     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1496         ret = pq;
1497         break;
1498 
1499     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1500     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1501     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1502     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1503         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1504         break;
1505     default:
1506         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1507                       offset);
1508         return -1;
1509     }
1510 
1511     if (pq != xive_get_field32(end_esmask, end.w1)) {
1512         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1513         xive2_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1514     }
1515 
1516     return ret;
1517 }
1518 
1519 static void xive2_end_source_write(void *opaque, hwaddr addr,
1520                                    uint64_t value, unsigned size)
1521 {
1522     Xive2EndSource *xsrc = XIVE2_END_SOURCE(opaque);
1523     uint32_t offset = addr & 0xFFF;
1524     uint8_t end_blk;
1525     uint32_t end_idx;
1526     Xive2End end;
1527     uint32_t end_esmask;
1528     uint8_t pq;
1529     bool notify = false;
1530 
1531     /*
1532      * The block id should be deduced from the load address on the END
1533      * ESB MMIO but our model only supports a single block per XIVE chip.
1534      */
1535     end_blk = xive2_router_get_block_id(xsrc->xrtr);
1536     end_idx = addr >> (xsrc->esb_shift + 1);
1537 
1538     if (xive2_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1539         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1540                       end_idx);
1541         return;
1542     }
1543 
1544     if (!xive2_end_is_valid(&end)) {
1545         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1546                       end_blk, end_idx);
1547         return;
1548     }
1549 
1550     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END2_W1_ESn :
1551         END2_W1_ESe;
1552     pq = xive_get_field32(end_esmask, end.w1);
1553 
1554     switch (offset) {
1555     case 0 ... 0x3FF:
1556         notify = xive_esb_trigger(&pq);
1557         break;
1558 
1559     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
1560         /* TODO: can we check StoreEOI availability from the router ? */
1561         notify = xive_esb_eoi(&pq);
1562         break;
1563 
1564     case XIVE_ESB_INJECT ... XIVE_ESB_INJECT + 0x3FF:
1565         if (end_esmask == END2_W1_ESe) {
1566             qemu_log_mask(LOG_GUEST_ERROR,
1567                           "XIVE: END %x/%x can not EQ inject on ESe\n",
1568                            end_blk, end_idx);
1569             return;
1570         }
1571         notify = true;
1572         break;
1573 
1574     default:
1575         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB write addr %d\n",
1576                       offset);
1577         return;
1578     }
1579 
1580     if (pq != xive_get_field32(end_esmask, end.w1)) {
1581         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1582         xive2_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1583     }
1584 
1585     /* TODO: Forward the source event notification for routing */
1586     if (notify) {
1587         ;
1588     }
1589 }
1590 
1591 static const MemoryRegionOps xive2_end_source_ops = {
1592     .read = xive2_end_source_read,
1593     .write = xive2_end_source_write,
1594     .endianness = DEVICE_BIG_ENDIAN,
1595     .valid = {
1596         .min_access_size = 1,
1597         .max_access_size = 8,
1598     },
1599     .impl = {
1600         .min_access_size = 1,
1601         .max_access_size = 8,
1602     },
1603 };
1604 
1605 static void xive2_end_source_realize(DeviceState *dev, Error **errp)
1606 {
1607     Xive2EndSource *xsrc = XIVE2_END_SOURCE(dev);
1608 
1609     assert(xsrc->xrtr);
1610 
1611     if (!xsrc->nr_ends) {
1612         error_setg(errp, "Number of interrupt needs to be greater than 0");
1613         return;
1614     }
1615 
1616     if (xsrc->esb_shift != XIVE_ESB_4K &&
1617         xsrc->esb_shift != XIVE_ESB_64K) {
1618         error_setg(errp, "Invalid ESB shift setting");
1619         return;
1620     }
1621 
1622     /*
1623      * Each END is assigned an even/odd pair of MMIO pages, the even page
1624      * manages the ESn field while the odd page manages the ESe field.
1625      */
1626     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1627                           &xive2_end_source_ops, xsrc, "xive.end",
1628                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1629 }
1630 
1631 static const Property xive2_end_source_properties[] = {
1632     DEFINE_PROP_UINT32("nr-ends", Xive2EndSource, nr_ends, 0),
1633     DEFINE_PROP_UINT32("shift", Xive2EndSource, esb_shift, XIVE_ESB_64K),
1634     DEFINE_PROP_LINK("xive", Xive2EndSource, xrtr, TYPE_XIVE2_ROUTER,
1635                      Xive2Router *),
1636 };
1637 
1638 static void xive2_end_source_class_init(ObjectClass *klass, void *data)
1639 {
1640     DeviceClass *dc = DEVICE_CLASS(klass);
1641 
1642     dc->desc    = "XIVE END Source";
1643     device_class_set_props(dc, xive2_end_source_properties);
1644     dc->realize = xive2_end_source_realize;
1645     dc->user_creatable = false;
1646 }
1647 
1648 static const TypeInfo xive2_end_source_info = {
1649     .name          = TYPE_XIVE2_END_SOURCE,
1650     .parent        = TYPE_DEVICE,
1651     .instance_size = sizeof(Xive2EndSource),
1652     .class_init    = xive2_end_source_class_init,
1653 };
1654 
1655 static void xive2_register_types(void)
1656 {
1657     type_register_static(&xive2_router_info);
1658     type_register_static(&xive2_end_source_info);
1659 }
1660 
1661 type_init(xive2_register_types)
1662