xref: /openbmc/qemu/hw/intc/xive.c (revision 30b6852c)
1 /*
2  * QEMU PowerPC XIVE interrupt controller model
3  *
4  * Copyright (c) 2017-2018, IBM Corporation.
5  *
6  * This code is licensed under the GPL version 2 or later. See the
7  * COPYING file in the top-level directory.
8  */
9 
10 #include "qemu/osdep.h"
11 #include "qemu/log.h"
12 #include "qemu/module.h"
13 #include "qapi/error.h"
14 #include "target/ppc/cpu.h"
15 #include "sysemu/cpus.h"
16 #include "sysemu/dma.h"
17 #include "sysemu/reset.h"
18 #include "hw/qdev-properties.h"
19 #include "migration/vmstate.h"
20 #include "monitor/monitor.h"
21 #include "hw/irq.h"
22 #include "hw/ppc/xive.h"
23 #include "hw/ppc/xive_regs.h"
24 #include "trace.h"
25 
26 /*
27  * XIVE Thread Interrupt Management context
28  */
29 
30 /*
31  * Convert an Interrupt Pending Buffer (IPB) register to a Pending
32  * Interrupt Priority Register (PIPR), which contains the priority of
33  * the most favored pending notification.
34  */
35 static uint8_t ipb_to_pipr(uint8_t ibp)
36 {
37     return ibp ? clz32((uint32_t)ibp << 24) : 0xff;
38 }
39 
40 static uint8_t exception_mask(uint8_t ring)
41 {
42     switch (ring) {
43     case TM_QW1_OS:
44         return TM_QW1_NSR_EO;
45     case TM_QW3_HV_PHYS:
46         return TM_QW3_NSR_HE;
47     default:
48         g_assert_not_reached();
49     }
50 }
51 
52 static qemu_irq xive_tctx_output(XiveTCTX *tctx, uint8_t ring)
53 {
54         switch (ring) {
55         case TM_QW0_USER:
56                 return 0; /* Not supported */
57         case TM_QW1_OS:
58                 return tctx->os_output;
59         case TM_QW2_HV_POOL:
60         case TM_QW3_HV_PHYS:
61                 return tctx->hv_output;
62         default:
63                 return 0;
64         }
65 }
66 
67 static uint64_t xive_tctx_accept(XiveTCTX *tctx, uint8_t ring)
68 {
69     uint8_t *regs = &tctx->regs[ring];
70     uint8_t nsr = regs[TM_NSR];
71     uint8_t mask = exception_mask(ring);
72 
73     qemu_irq_lower(xive_tctx_output(tctx, ring));
74 
75     if (regs[TM_NSR] & mask) {
76         uint8_t cppr = regs[TM_PIPR];
77 
78         regs[TM_CPPR] = cppr;
79 
80         /* Reset the pending buffer bit */
81         regs[TM_IPB] &= ~xive_priority_to_ipb(cppr);
82         regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
83 
84         /* Drop Exception bit */
85         regs[TM_NSR] &= ~mask;
86 
87         trace_xive_tctx_accept(tctx->cs->cpu_index, ring,
88                                regs[TM_IPB], regs[TM_PIPR],
89                                regs[TM_CPPR], regs[TM_NSR]);
90     }
91 
92     return (nsr << 8) | regs[TM_CPPR];
93 }
94 
95 static void xive_tctx_notify(XiveTCTX *tctx, uint8_t ring)
96 {
97     uint8_t *regs = &tctx->regs[ring];
98 
99     if (regs[TM_PIPR] < regs[TM_CPPR]) {
100         switch (ring) {
101         case TM_QW1_OS:
102             regs[TM_NSR] |= TM_QW1_NSR_EO;
103             break;
104         case TM_QW3_HV_PHYS:
105             regs[TM_NSR] |= (TM_QW3_NSR_HE_PHYS << 6);
106             break;
107         default:
108             g_assert_not_reached();
109         }
110         trace_xive_tctx_notify(tctx->cs->cpu_index, ring,
111                                regs[TM_IPB], regs[TM_PIPR],
112                                regs[TM_CPPR], regs[TM_NSR]);
113         qemu_irq_raise(xive_tctx_output(tctx, ring));
114     }
115 }
116 
117 static void xive_tctx_set_cppr(XiveTCTX *tctx, uint8_t ring, uint8_t cppr)
118 {
119     uint8_t *regs = &tctx->regs[ring];
120 
121     trace_xive_tctx_set_cppr(tctx->cs->cpu_index, ring,
122                              regs[TM_IPB], regs[TM_PIPR],
123                              cppr, regs[TM_NSR]);
124 
125     if (cppr > XIVE_PRIORITY_MAX) {
126         cppr = 0xff;
127     }
128 
129     tctx->regs[ring + TM_CPPR] = cppr;
130 
131     /* CPPR has changed, check if we need to raise a pending exception */
132     xive_tctx_notify(tctx, ring);
133 }
134 
135 void xive_tctx_ipb_update(XiveTCTX *tctx, uint8_t ring, uint8_t ipb)
136 {
137     uint8_t *regs = &tctx->regs[ring];
138 
139     regs[TM_IPB] |= ipb;
140     regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
141     xive_tctx_notify(tctx, ring);
142 }
143 
144 /*
145  * XIVE Thread Interrupt Management Area (TIMA)
146  */
147 
148 static void xive_tm_set_hv_cppr(XivePresenter *xptr, XiveTCTX *tctx,
149                                 hwaddr offset, uint64_t value, unsigned size)
150 {
151     xive_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
152 }
153 
154 static uint64_t xive_tm_ack_hv_reg(XivePresenter *xptr, XiveTCTX *tctx,
155                                    hwaddr offset, unsigned size)
156 {
157     return xive_tctx_accept(tctx, TM_QW3_HV_PHYS);
158 }
159 
160 static uint64_t xive_tm_pull_pool_ctx(XivePresenter *xptr, XiveTCTX *tctx,
161                                       hwaddr offset, unsigned size)
162 {
163     uint32_t qw2w2_prev = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
164     uint32_t qw2w2;
165 
166     qw2w2 = xive_set_field32(TM_QW2W2_VP, qw2w2_prev, 0);
167     memcpy(&tctx->regs[TM_QW2_HV_POOL + TM_WORD2], &qw2w2, 4);
168     return qw2w2;
169 }
170 
171 static void xive_tm_vt_push(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
172                             uint64_t value, unsigned size)
173 {
174     tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] = value & 0xff;
175 }
176 
177 static uint64_t xive_tm_vt_poll(XivePresenter *xptr, XiveTCTX *tctx,
178                                 hwaddr offset, unsigned size)
179 {
180     return tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] & 0xff;
181 }
182 
183 /*
184  * Define an access map for each page of the TIMA that we will use in
185  * the memory region ops to filter values when doing loads and stores
186  * of raw registers values
187  *
188  * Registers accessibility bits :
189  *
190  *    0x0 - no access
191  *    0x1 - write only
192  *    0x2 - read only
193  *    0x3 - read/write
194  */
195 
196 static const uint8_t xive_tm_hw_view[] = {
197     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
198     3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
199     0, 0, 3, 3,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
200     3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   3, 3, 3, 0, /* QW-3 PHYS */
201 };
202 
203 static const uint8_t xive_tm_hv_view[] = {
204     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
205     3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
206     0, 0, 3, 3,   0, 0, 0, 0,   0, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
207     3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   0, 0, 0, 0, /* QW-3 PHYS */
208 };
209 
210 static const uint8_t xive_tm_os_view[] = {
211     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
212     2, 3, 2, 2,   2, 2, 0, 2,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
213     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
214     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
215 };
216 
217 static const uint8_t xive_tm_user_view[] = {
218     3, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-0 User */
219     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
220     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
221     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
222 };
223 
224 /*
225  * Overall TIMA access map for the thread interrupt management context
226  * registers
227  */
228 static const uint8_t *xive_tm_views[] = {
229     [XIVE_TM_HW_PAGE]   = xive_tm_hw_view,
230     [XIVE_TM_HV_PAGE]   = xive_tm_hv_view,
231     [XIVE_TM_OS_PAGE]   = xive_tm_os_view,
232     [XIVE_TM_USER_PAGE] = xive_tm_user_view,
233 };
234 
235 /*
236  * Computes a register access mask for a given offset in the TIMA
237  */
238 static uint64_t xive_tm_mask(hwaddr offset, unsigned size, bool write)
239 {
240     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
241     uint8_t reg_offset = offset & 0x3F;
242     uint8_t reg_mask = write ? 0x1 : 0x2;
243     uint64_t mask = 0x0;
244     int i;
245 
246     for (i = 0; i < size; i++) {
247         if (xive_tm_views[page_offset][reg_offset + i] & reg_mask) {
248             mask |= (uint64_t) 0xff << (8 * (size - i - 1));
249         }
250     }
251 
252     return mask;
253 }
254 
255 static void xive_tm_raw_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
256                               unsigned size)
257 {
258     uint8_t ring_offset = offset & 0x30;
259     uint8_t reg_offset = offset & 0x3F;
260     uint64_t mask = xive_tm_mask(offset, size, true);
261     int i;
262 
263     /*
264      * Only 4 or 8 bytes stores are allowed and the User ring is
265      * excluded
266      */
267     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
268         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA @%"
269                       HWADDR_PRIx"\n", offset);
270         return;
271     }
272 
273     /*
274      * Use the register offset for the raw values and filter out
275      * reserved values
276      */
277     for (i = 0; i < size; i++) {
278         uint8_t byte_mask = (mask >> (8 * (size - i - 1)));
279         if (byte_mask) {
280             tctx->regs[reg_offset + i] = (value >> (8 * (size - i - 1))) &
281                 byte_mask;
282         }
283     }
284 }
285 
286 static uint64_t xive_tm_raw_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
287 {
288     uint8_t ring_offset = offset & 0x30;
289     uint8_t reg_offset = offset & 0x3F;
290     uint64_t mask = xive_tm_mask(offset, size, false);
291     uint64_t ret;
292     int i;
293 
294     /*
295      * Only 4 or 8 bytes loads are allowed and the User ring is
296      * excluded
297      */
298     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
299         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access at TIMA @%"
300                       HWADDR_PRIx"\n", offset);
301         return -1;
302     }
303 
304     /* Use the register offset for the raw values */
305     ret = 0;
306     for (i = 0; i < size; i++) {
307         ret |= (uint64_t) tctx->regs[reg_offset + i] << (8 * (size - i - 1));
308     }
309 
310     /* filter out reserved values */
311     return ret & mask;
312 }
313 
314 /*
315  * The TM context is mapped twice within each page. Stores and loads
316  * to the first mapping below 2K write and read the specified values
317  * without modification. The second mapping above 2K performs specific
318  * state changes (side effects) in addition to setting/returning the
319  * interrupt management area context of the processor thread.
320  */
321 static uint64_t xive_tm_ack_os_reg(XivePresenter *xptr, XiveTCTX *tctx,
322                                    hwaddr offset, unsigned size)
323 {
324     return xive_tctx_accept(tctx, TM_QW1_OS);
325 }
326 
327 static void xive_tm_set_os_cppr(XivePresenter *xptr, XiveTCTX *tctx,
328                                 hwaddr offset, uint64_t value, unsigned size)
329 {
330     xive_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
331 }
332 
333 /*
334  * Adjust the IPB to allow a CPU to process event queues of other
335  * priorities during one physical interrupt cycle.
336  */
337 static void xive_tm_set_os_pending(XivePresenter *xptr, XiveTCTX *tctx,
338                                    hwaddr offset, uint64_t value, unsigned size)
339 {
340     xive_tctx_ipb_update(tctx, TM_QW1_OS, xive_priority_to_ipb(value & 0xff));
341 }
342 
343 static void xive_os_cam_decode(uint32_t cam, uint8_t *nvt_blk,
344                                uint32_t *nvt_idx, bool *vo)
345 {
346     if (nvt_blk) {
347         *nvt_blk = xive_nvt_blk(cam);
348     }
349     if (nvt_idx) {
350         *nvt_idx = xive_nvt_idx(cam);
351     }
352     if (vo) {
353         *vo = !!(cam & TM_QW1W2_VO);
354     }
355 }
356 
357 static uint32_t xive_tctx_get_os_cam(XiveTCTX *tctx, uint8_t *nvt_blk,
358                                      uint32_t *nvt_idx, bool *vo)
359 {
360     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
361     uint32_t cam = be32_to_cpu(qw1w2);
362 
363     xive_os_cam_decode(cam, nvt_blk, nvt_idx, vo);
364     return qw1w2;
365 }
366 
367 static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t qw1w2)
368 {
369     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
370 }
371 
372 static uint64_t xive_tm_pull_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
373                                     hwaddr offset, unsigned size)
374 {
375     uint32_t qw1w2;
376     uint32_t qw1w2_new;
377     uint8_t nvt_blk;
378     uint32_t nvt_idx;
379     bool vo;
380 
381     qw1w2 = xive_tctx_get_os_cam(tctx, &nvt_blk, &nvt_idx, &vo);
382 
383     if (!vo) {
384         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: pulling invalid NVT %x/%x !?\n",
385                       nvt_blk, nvt_idx);
386     }
387 
388     /* Invalidate CAM line */
389     qw1w2_new = xive_set_field32(TM_QW1W2_VO, qw1w2, 0);
390     xive_tctx_set_os_cam(tctx, qw1w2_new);
391     return qw1w2;
392 }
393 
394 static void xive_tctx_need_resend(XiveRouter *xrtr, XiveTCTX *tctx,
395                                   uint8_t nvt_blk, uint32_t nvt_idx)
396 {
397     XiveNVT nvt;
398     uint8_t ipb;
399 
400     /*
401      * Grab the associated NVT to pull the pending bits, and merge
402      * them with the IPB of the thread interrupt context registers
403      */
404     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
405         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVT %x/%x\n",
406                           nvt_blk, nvt_idx);
407         return;
408     }
409 
410     ipb = xive_get_field32(NVT_W4_IPB, nvt.w4);
411 
412     if (ipb) {
413         /* Reset the NVT value */
414         nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, 0);
415         xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
416 
417         /* Merge in current context */
418         xive_tctx_ipb_update(tctx, TM_QW1_OS, ipb);
419     }
420 }
421 
422 /*
423  * Updating the OS CAM line can trigger a resend of interrupt
424  */
425 static void xive_tm_push_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
426                                 hwaddr offset, uint64_t value, unsigned size)
427 {
428     uint32_t cam = value;
429     uint32_t qw1w2 = cpu_to_be32(cam);
430     uint8_t nvt_blk;
431     uint32_t nvt_idx;
432     bool vo;
433 
434     xive_os_cam_decode(cam, &nvt_blk, &nvt_idx, &vo);
435 
436     /* First update the registers */
437     xive_tctx_set_os_cam(tctx, qw1w2);
438 
439     /* Check the interrupt pending bits */
440     if (vo) {
441         xive_tctx_need_resend(XIVE_ROUTER(xptr), tctx, nvt_blk, nvt_idx);
442     }
443 }
444 
445 /*
446  * Define a mapping of "special" operations depending on the TIMA page
447  * offset and the size of the operation.
448  */
449 typedef struct XiveTmOp {
450     uint8_t  page_offset;
451     uint32_t op_offset;
452     unsigned size;
453     void     (*write_handler)(XivePresenter *xptr, XiveTCTX *tctx,
454                               hwaddr offset,
455                               uint64_t value, unsigned size);
456     uint64_t (*read_handler)(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
457                              unsigned size);
458 } XiveTmOp;
459 
460 static const XiveTmOp xive_tm_operations[] = {
461     /*
462      * MMIOs below 2K : raw values and special operations without side
463      * effects
464      */
465     { XIVE_TM_OS_PAGE, TM_QW1_OS + TM_CPPR,   1, xive_tm_set_os_cppr, NULL },
466     { XIVE_TM_HV_PAGE, TM_QW1_OS + TM_WORD2,     4, xive_tm_push_os_ctx, NULL },
467     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_CPPR, 1, xive_tm_set_hv_cppr, NULL },
468     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, xive_tm_vt_push, NULL },
469     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, NULL, xive_tm_vt_poll },
470 
471     /* MMIOs above 2K : special operations with side effects */
472     { XIVE_TM_OS_PAGE, TM_SPC_ACK_OS_REG,     2, NULL, xive_tm_ack_os_reg },
473     { XIVE_TM_OS_PAGE, TM_SPC_SET_OS_PENDING, 1, xive_tm_set_os_pending, NULL },
474     { XIVE_TM_HV_PAGE, TM_SPC_PULL_OS_CTX,    4, NULL, xive_tm_pull_os_ctx },
475     { XIVE_TM_HV_PAGE, TM_SPC_PULL_OS_CTX,    8, NULL, xive_tm_pull_os_ctx },
476     { XIVE_TM_HV_PAGE, TM_SPC_ACK_HV_REG,     2, NULL, xive_tm_ack_hv_reg },
477     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  4, NULL, xive_tm_pull_pool_ctx },
478     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  8, NULL, xive_tm_pull_pool_ctx },
479 };
480 
481 static const XiveTmOp *xive_tm_find_op(hwaddr offset, unsigned size, bool write)
482 {
483     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
484     uint32_t op_offset = offset & 0xFFF;
485     int i;
486 
487     for (i = 0; i < ARRAY_SIZE(xive_tm_operations); i++) {
488         const XiveTmOp *xto = &xive_tm_operations[i];
489 
490         /* Accesses done from a more privileged TIMA page is allowed */
491         if (xto->page_offset >= page_offset &&
492             xto->op_offset == op_offset &&
493             xto->size == size &&
494             ((write && xto->write_handler) || (!write && xto->read_handler))) {
495             return xto;
496         }
497     }
498     return NULL;
499 }
500 
501 /*
502  * TIMA MMIO handlers
503  */
504 void xive_tctx_tm_write(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
505                         uint64_t value, unsigned size)
506 {
507     const XiveTmOp *xto;
508 
509     trace_xive_tctx_tm_write(offset, size, value);
510 
511     /*
512      * TODO: check V bit in Q[0-3]W2
513      */
514 
515     /*
516      * First, check for special operations in the 2K region
517      */
518     if (offset & 0x800) {
519         xto = xive_tm_find_op(offset, size, true);
520         if (!xto) {
521             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA "
522                           "@%"HWADDR_PRIx"\n", offset);
523         } else {
524             xto->write_handler(xptr, tctx, offset, value, size);
525         }
526         return;
527     }
528 
529     /*
530      * Then, for special operations in the region below 2K.
531      */
532     xto = xive_tm_find_op(offset, size, true);
533     if (xto) {
534         xto->write_handler(xptr, tctx, offset, value, size);
535         return;
536     }
537 
538     /*
539      * Finish with raw access to the register values
540      */
541     xive_tm_raw_write(tctx, offset, value, size);
542 }
543 
544 uint64_t xive_tctx_tm_read(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
545                            unsigned size)
546 {
547     const XiveTmOp *xto;
548     uint64_t ret;
549 
550     /*
551      * TODO: check V bit in Q[0-3]W2
552      */
553 
554     /*
555      * First, check for special operations in the 2K region
556      */
557     if (offset & 0x800) {
558         xto = xive_tm_find_op(offset, size, false);
559         if (!xto) {
560             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access to TIMA"
561                           "@%"HWADDR_PRIx"\n", offset);
562             return -1;
563         }
564         ret = xto->read_handler(xptr, tctx, offset, size);
565         goto out;
566     }
567 
568     /*
569      * Then, for special operations in the region below 2K.
570      */
571     xto = xive_tm_find_op(offset, size, false);
572     if (xto) {
573         ret = xto->read_handler(xptr, tctx, offset, size);
574         goto out;
575     }
576 
577     /*
578      * Finish with raw access to the register values
579      */
580     ret = xive_tm_raw_read(tctx, offset, size);
581 out:
582     trace_xive_tctx_tm_read(offset, size, ret);
583     return ret;
584 }
585 
586 static char *xive_tctx_ring_print(uint8_t *ring)
587 {
588     uint32_t w2 = xive_tctx_word2(ring);
589 
590     return g_strdup_printf("%02x   %02x  %02x    %02x   %02x  "
591                    "%02x  %02x   %02x  %08x",
592                    ring[TM_NSR], ring[TM_CPPR], ring[TM_IPB], ring[TM_LSMFB],
593                    ring[TM_ACK_CNT], ring[TM_INC], ring[TM_AGE], ring[TM_PIPR],
594                    be32_to_cpu(w2));
595 }
596 
597 static const char * const xive_tctx_ring_names[] = {
598     "USER", "OS", "POOL", "PHYS",
599 };
600 
601 /*
602  * kvm_irqchip_in_kernel() will cause the compiler to turn this
603  * info a nop if CONFIG_KVM isn't defined.
604  */
605 #define xive_in_kernel(xptr)                                            \
606     (kvm_irqchip_in_kernel() &&                                         \
607      ({                                                                 \
608          XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);      \
609          xpc->in_kernel ? xpc->in_kernel(xptr) : false;                 \
610      }))
611 
612 void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
613 {
614     int cpu_index;
615     int i;
616 
617     /* Skip partially initialized vCPUs. This can happen on sPAPR when vCPUs
618      * are hot plugged or unplugged.
619      */
620     if (!tctx) {
621         return;
622     }
623 
624     cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
625 
626     if (xive_in_kernel(tctx->xptr)) {
627         Error *local_err = NULL;
628 
629         kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
630         if (local_err) {
631             error_report_err(local_err);
632             return;
633         }
634     }
635 
636     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
637                    "  W2\n", cpu_index);
638 
639     for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
640         char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
641         monitor_printf(mon, "CPU[%04x]: %4s    %s\n", cpu_index,
642                        xive_tctx_ring_names[i], s);
643         g_free(s);
644     }
645 }
646 
647 void xive_tctx_reset(XiveTCTX *tctx)
648 {
649     memset(tctx->regs, 0, sizeof(tctx->regs));
650 
651     /* Set some defaults */
652     tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
653     tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
654     tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
655 
656     /*
657      * Initialize PIPR to 0xFF to avoid phantom interrupts when the
658      * CPPR is first set.
659      */
660     tctx->regs[TM_QW1_OS + TM_PIPR] =
661         ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
662     tctx->regs[TM_QW3_HV_PHYS + TM_PIPR] =
663         ipb_to_pipr(tctx->regs[TM_QW3_HV_PHYS + TM_IPB]);
664 }
665 
666 static void xive_tctx_realize(DeviceState *dev, Error **errp)
667 {
668     XiveTCTX *tctx = XIVE_TCTX(dev);
669     PowerPCCPU *cpu;
670     CPUPPCState *env;
671 
672     assert(tctx->cs);
673     assert(tctx->xptr);
674 
675     cpu = POWERPC_CPU(tctx->cs);
676     env = &cpu->env;
677     switch (PPC_INPUT(env)) {
678     case PPC_FLAGS_INPUT_POWER9:
679         tctx->hv_output = env->irq_inputs[POWER9_INPUT_HINT];
680         tctx->os_output = env->irq_inputs[POWER9_INPUT_INT];
681         break;
682 
683     default:
684         error_setg(errp, "XIVE interrupt controller does not support "
685                    "this CPU bus model");
686         return;
687     }
688 
689     /* Connect the presenter to the VCPU (required for CPU hotplug) */
690     if (xive_in_kernel(tctx->xptr)) {
691         if (kvmppc_xive_cpu_connect(tctx, errp) < 0) {
692             return;
693         }
694     }
695 }
696 
697 static int vmstate_xive_tctx_pre_save(void *opaque)
698 {
699     XiveTCTX *tctx = XIVE_TCTX(opaque);
700     Error *local_err = NULL;
701     int ret;
702 
703     if (xive_in_kernel(tctx->xptr)) {
704         ret = kvmppc_xive_cpu_get_state(tctx, &local_err);
705         if (ret < 0) {
706             error_report_err(local_err);
707             return ret;
708         }
709     }
710 
711     return 0;
712 }
713 
714 static int vmstate_xive_tctx_post_load(void *opaque, int version_id)
715 {
716     XiveTCTX *tctx = XIVE_TCTX(opaque);
717     Error *local_err = NULL;
718     int ret;
719 
720     if (xive_in_kernel(tctx->xptr)) {
721         /*
722          * Required for hotplugged CPU, for which the state comes
723          * after all states of the machine.
724          */
725         ret = kvmppc_xive_cpu_set_state(tctx, &local_err);
726         if (ret < 0) {
727             error_report_err(local_err);
728             return ret;
729         }
730     }
731 
732     return 0;
733 }
734 
735 static const VMStateDescription vmstate_xive_tctx = {
736     .name = TYPE_XIVE_TCTX,
737     .version_id = 1,
738     .minimum_version_id = 1,
739     .pre_save = vmstate_xive_tctx_pre_save,
740     .post_load = vmstate_xive_tctx_post_load,
741     .fields = (VMStateField[]) {
742         VMSTATE_BUFFER(regs, XiveTCTX),
743         VMSTATE_END_OF_LIST()
744     },
745 };
746 
747 static Property xive_tctx_properties[] = {
748     DEFINE_PROP_LINK("cpu", XiveTCTX, cs, TYPE_CPU, CPUState *),
749     DEFINE_PROP_LINK("presenter", XiveTCTX, xptr, TYPE_XIVE_PRESENTER,
750                      XivePresenter *),
751     DEFINE_PROP_END_OF_LIST(),
752 };
753 
754 static void xive_tctx_class_init(ObjectClass *klass, void *data)
755 {
756     DeviceClass *dc = DEVICE_CLASS(klass);
757 
758     dc->desc = "XIVE Interrupt Thread Context";
759     dc->realize = xive_tctx_realize;
760     dc->vmsd = &vmstate_xive_tctx;
761     device_class_set_props(dc, xive_tctx_properties);
762     /*
763      * Reason: part of XIVE interrupt controller, needs to be wired up
764      * by xive_tctx_create().
765      */
766     dc->user_creatable = false;
767 }
768 
769 static const TypeInfo xive_tctx_info = {
770     .name          = TYPE_XIVE_TCTX,
771     .parent        = TYPE_DEVICE,
772     .instance_size = sizeof(XiveTCTX),
773     .class_init    = xive_tctx_class_init,
774 };
775 
776 Object *xive_tctx_create(Object *cpu, XivePresenter *xptr, Error **errp)
777 {
778     Object *obj;
779 
780     obj = object_new(TYPE_XIVE_TCTX);
781     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj);
782     object_unref(obj);
783     object_property_set_link(obj, "cpu", cpu, &error_abort);
784     object_property_set_link(obj, "presenter", OBJECT(xptr), &error_abort);
785     if (!qdev_realize(DEVICE(obj), NULL, errp)) {
786         object_unparent(obj);
787         return NULL;
788     }
789     return obj;
790 }
791 
792 void xive_tctx_destroy(XiveTCTX *tctx)
793 {
794     Object *obj = OBJECT(tctx);
795 
796     object_unparent(obj);
797 }
798 
799 /*
800  * XIVE ESB helpers
801  */
802 
803 uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
804 {
805     uint8_t old_pq = *pq & 0x3;
806 
807     *pq &= ~0x3;
808     *pq |= value & 0x3;
809 
810     return old_pq;
811 }
812 
813 bool xive_esb_trigger(uint8_t *pq)
814 {
815     uint8_t old_pq = *pq & 0x3;
816 
817     switch (old_pq) {
818     case XIVE_ESB_RESET:
819         xive_esb_set(pq, XIVE_ESB_PENDING);
820         return true;
821     case XIVE_ESB_PENDING:
822     case XIVE_ESB_QUEUED:
823         xive_esb_set(pq, XIVE_ESB_QUEUED);
824         return false;
825     case XIVE_ESB_OFF:
826         xive_esb_set(pq, XIVE_ESB_OFF);
827         return false;
828     default:
829          g_assert_not_reached();
830     }
831 }
832 
833 bool xive_esb_eoi(uint8_t *pq)
834 {
835     uint8_t old_pq = *pq & 0x3;
836 
837     switch (old_pq) {
838     case XIVE_ESB_RESET:
839     case XIVE_ESB_PENDING:
840         xive_esb_set(pq, XIVE_ESB_RESET);
841         return false;
842     case XIVE_ESB_QUEUED:
843         xive_esb_set(pq, XIVE_ESB_PENDING);
844         return true;
845     case XIVE_ESB_OFF:
846         xive_esb_set(pq, XIVE_ESB_OFF);
847         return false;
848     default:
849          g_assert_not_reached();
850     }
851 }
852 
853 /*
854  * XIVE Interrupt Source (or IVSE)
855  */
856 
857 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
858 {
859     assert(srcno < xsrc->nr_irqs);
860 
861     return xsrc->status[srcno] & 0x3;
862 }
863 
864 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
865 {
866     assert(srcno < xsrc->nr_irqs);
867 
868     return xive_esb_set(&xsrc->status[srcno], pq);
869 }
870 
871 /*
872  * Returns whether the event notification should be forwarded.
873  */
874 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
875 {
876     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
877 
878     xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
879 
880     switch (old_pq) {
881     case XIVE_ESB_RESET:
882         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
883         return true;
884     default:
885         return false;
886     }
887 }
888 
889 /*
890  * Returns whether the event notification should be forwarded.
891  */
892 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
893 {
894     bool ret;
895 
896     assert(srcno < xsrc->nr_irqs);
897 
898     ret = xive_esb_trigger(&xsrc->status[srcno]);
899 
900     if (xive_source_irq_is_lsi(xsrc, srcno) &&
901         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
902         qemu_log_mask(LOG_GUEST_ERROR,
903                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
904     }
905 
906     return ret;
907 }
908 
909 /*
910  * Returns whether the event notification should be forwarded.
911  */
912 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
913 {
914     bool ret;
915 
916     assert(srcno < xsrc->nr_irqs);
917 
918     ret = xive_esb_eoi(&xsrc->status[srcno]);
919 
920     /*
921      * LSI sources do not set the Q bit but they can still be
922      * asserted, in which case we should forward a new event
923      * notification
924      */
925     if (xive_source_irq_is_lsi(xsrc, srcno) &&
926         xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
927         ret = xive_source_lsi_trigger(xsrc, srcno);
928     }
929 
930     return ret;
931 }
932 
933 /*
934  * Forward the source event notification to the Router
935  */
936 static void xive_source_notify(XiveSource *xsrc, int srcno)
937 {
938     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
939 
940     if (xnc->notify) {
941         xnc->notify(xsrc->xive, srcno);
942     }
943 }
944 
945 /*
946  * In a two pages ESB MMIO setting, even page is the trigger page, odd
947  * page is for management
948  */
949 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
950 {
951     return !((addr >> shift) & 1);
952 }
953 
954 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
955 {
956     return xive_source_esb_has_2page(xsrc) &&
957         addr_is_even(addr, xsrc->esb_shift - 1);
958 }
959 
960 /*
961  * ESB MMIO loads
962  *                      Trigger page    Management/EOI page
963  *
964  * ESB MMIO setting     2 pages         1 or 2 pages
965  *
966  * 0x000 .. 0x3FF       -1              EOI and return 0|1
967  * 0x400 .. 0x7FF       -1              EOI and return 0|1
968  * 0x800 .. 0xBFF       -1              return PQ
969  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
970  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
971  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
972  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
973  */
974 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
975 {
976     XiveSource *xsrc = XIVE_SOURCE(opaque);
977     uint32_t offset = addr & 0xFFF;
978     uint32_t srcno = addr >> xsrc->esb_shift;
979     uint64_t ret = -1;
980 
981     /* In a two pages ESB MMIO setting, trigger page should not be read */
982     if (xive_source_is_trigger_page(xsrc, addr)) {
983         qemu_log_mask(LOG_GUEST_ERROR,
984                       "XIVE: invalid load on IRQ %d trigger page at "
985                       "0x%"HWADDR_PRIx"\n", srcno, addr);
986         return -1;
987     }
988 
989     switch (offset) {
990     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
991         ret = xive_source_esb_eoi(xsrc, srcno);
992 
993         /* Forward the source event notification for routing */
994         if (ret) {
995             xive_source_notify(xsrc, srcno);
996         }
997         break;
998 
999     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1000         ret = xive_source_esb_get(xsrc, srcno);
1001         break;
1002 
1003     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1004     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1005     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1006     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1007         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1008         break;
1009     default:
1010         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
1011                       offset);
1012     }
1013 
1014     trace_xive_source_esb_read(addr, srcno, ret);
1015 
1016     return ret;
1017 }
1018 
1019 /*
1020  * ESB MMIO stores
1021  *                      Trigger page    Management/EOI page
1022  *
1023  * ESB MMIO setting     2 pages         1 or 2 pages
1024  *
1025  * 0x000 .. 0x3FF       Trigger         Trigger
1026  * 0x400 .. 0x7FF       Trigger         EOI
1027  * 0x800 .. 0xBFF       Trigger         undefined
1028  * 0xC00 .. 0xCFF       Trigger         PQ=00
1029  * 0xD00 .. 0xDFF       Trigger         PQ=01
1030  * 0xE00 .. 0xDFF       Trigger         PQ=10
1031  * 0xF00 .. 0xDFF       Trigger         PQ=11
1032  */
1033 static void xive_source_esb_write(void *opaque, hwaddr addr,
1034                                   uint64_t value, unsigned size)
1035 {
1036     XiveSource *xsrc = XIVE_SOURCE(opaque);
1037     uint32_t offset = addr & 0xFFF;
1038     uint32_t srcno = addr >> xsrc->esb_shift;
1039     bool notify = false;
1040 
1041     trace_xive_source_esb_write(addr, srcno, value);
1042 
1043     /* In a two pages ESB MMIO setting, trigger page only triggers */
1044     if (xive_source_is_trigger_page(xsrc, addr)) {
1045         notify = xive_source_esb_trigger(xsrc, srcno);
1046         goto out;
1047     }
1048 
1049     switch (offset) {
1050     case 0 ... 0x3FF:
1051         notify = xive_source_esb_trigger(xsrc, srcno);
1052         break;
1053 
1054     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
1055         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
1056             qemu_log_mask(LOG_GUEST_ERROR,
1057                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
1058             return;
1059         }
1060 
1061         notify = xive_source_esb_eoi(xsrc, srcno);
1062         break;
1063 
1064     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1065     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1066     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1067     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1068         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1069         break;
1070 
1071     default:
1072         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
1073                       offset);
1074         return;
1075     }
1076 
1077 out:
1078     /* Forward the source event notification for routing */
1079     if (notify) {
1080         xive_source_notify(xsrc, srcno);
1081     }
1082 }
1083 
1084 static const MemoryRegionOps xive_source_esb_ops = {
1085     .read = xive_source_esb_read,
1086     .write = xive_source_esb_write,
1087     .endianness = DEVICE_BIG_ENDIAN,
1088     .valid = {
1089         .min_access_size = 8,
1090         .max_access_size = 8,
1091     },
1092     .impl = {
1093         .min_access_size = 8,
1094         .max_access_size = 8,
1095     },
1096 };
1097 
1098 void xive_source_set_irq(void *opaque, int srcno, int val)
1099 {
1100     XiveSource *xsrc = XIVE_SOURCE(opaque);
1101     bool notify = false;
1102 
1103     if (xive_source_irq_is_lsi(xsrc, srcno)) {
1104         if (val) {
1105             notify = xive_source_lsi_trigger(xsrc, srcno);
1106         } else {
1107             xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
1108         }
1109     } else {
1110         if (val) {
1111             notify = xive_source_esb_trigger(xsrc, srcno);
1112         }
1113     }
1114 
1115     /* Forward the source event notification for routing */
1116     if (notify) {
1117         xive_source_notify(xsrc, srcno);
1118     }
1119 }
1120 
1121 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
1122 {
1123     int i;
1124 
1125     for (i = 0; i < xsrc->nr_irqs; i++) {
1126         uint8_t pq = xive_source_esb_get(xsrc, i);
1127 
1128         if (pq == XIVE_ESB_OFF) {
1129             continue;
1130         }
1131 
1132         monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
1133                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
1134                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
1135                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1136                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ');
1137     }
1138 }
1139 
1140 static void xive_source_reset(void *dev)
1141 {
1142     XiveSource *xsrc = XIVE_SOURCE(dev);
1143 
1144     /* Do not clear the LSI bitmap */
1145 
1146     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
1147     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
1148 }
1149 
1150 static void xive_source_realize(DeviceState *dev, Error **errp)
1151 {
1152     XiveSource *xsrc = XIVE_SOURCE(dev);
1153     size_t esb_len = xive_source_esb_len(xsrc);
1154 
1155     assert(xsrc->xive);
1156 
1157     if (!xsrc->nr_irqs) {
1158         error_setg(errp, "Number of interrupt needs to be greater than 0");
1159         return;
1160     }
1161 
1162     if (xsrc->esb_shift != XIVE_ESB_4K &&
1163         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1164         xsrc->esb_shift != XIVE_ESB_64K &&
1165         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1166         error_setg(errp, "Invalid ESB shift setting");
1167         return;
1168     }
1169 
1170     xsrc->status = g_malloc0(xsrc->nr_irqs);
1171     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1172 
1173     memory_region_init(&xsrc->esb_mmio, OBJECT(xsrc), "xive.esb", esb_len);
1174     memory_region_init_io(&xsrc->esb_mmio_emulated, OBJECT(xsrc),
1175                           &xive_source_esb_ops, xsrc, "xive.esb-emulated",
1176                           esb_len);
1177     memory_region_add_subregion(&xsrc->esb_mmio, 0, &xsrc->esb_mmio_emulated);
1178 
1179     qemu_register_reset(xive_source_reset, dev);
1180 }
1181 
1182 static const VMStateDescription vmstate_xive_source = {
1183     .name = TYPE_XIVE_SOURCE,
1184     .version_id = 1,
1185     .minimum_version_id = 1,
1186     .fields = (VMStateField[]) {
1187         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1188         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1189         VMSTATE_END_OF_LIST()
1190     },
1191 };
1192 
1193 /*
1194  * The default XIVE interrupt source setting for the ESB MMIOs is two
1195  * 64k pages without Store EOI, to be in sync with KVM.
1196  */
1197 static Property xive_source_properties[] = {
1198     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1199     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1200     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1201     DEFINE_PROP_LINK("xive", XiveSource, xive, TYPE_XIVE_NOTIFIER,
1202                      XiveNotifier *),
1203     DEFINE_PROP_END_OF_LIST(),
1204 };
1205 
1206 static void xive_source_class_init(ObjectClass *klass, void *data)
1207 {
1208     DeviceClass *dc = DEVICE_CLASS(klass);
1209 
1210     dc->desc    = "XIVE Interrupt Source";
1211     device_class_set_props(dc, xive_source_properties);
1212     dc->realize = xive_source_realize;
1213     dc->vmsd    = &vmstate_xive_source;
1214     /*
1215      * Reason: part of XIVE interrupt controller, needs to be wired up,
1216      * e.g. by spapr_xive_instance_init().
1217      */
1218     dc->user_creatable = false;
1219 }
1220 
1221 static const TypeInfo xive_source_info = {
1222     .name          = TYPE_XIVE_SOURCE,
1223     .parent        = TYPE_DEVICE,
1224     .instance_size = sizeof(XiveSource),
1225     .class_init    = xive_source_class_init,
1226 };
1227 
1228 /*
1229  * XiveEND helpers
1230  */
1231 
1232 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
1233 {
1234     uint64_t qaddr_base = xive_end_qaddr(end);
1235     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1236     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1237     uint32_t qentries = 1 << (qsize + 10);
1238     int i;
1239 
1240     /*
1241      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1242      */
1243     monitor_printf(mon, " [ ");
1244     qindex = (qindex - (width - 1)) & (qentries - 1);
1245     for (i = 0; i < width; i++) {
1246         uint64_t qaddr = qaddr_base + (qindex << 2);
1247         uint32_t qdata = -1;
1248 
1249         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
1250                             sizeof(qdata))) {
1251             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1252                           HWADDR_PRIx "\n", qaddr);
1253             return;
1254         }
1255         monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1256                        be32_to_cpu(qdata));
1257         qindex = (qindex + 1) & (qentries - 1);
1258     }
1259     monitor_printf(mon, "]");
1260 }
1261 
1262 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1263 {
1264     uint64_t qaddr_base = xive_end_qaddr(end);
1265     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1266     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1267     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1268     uint32_t qentries = 1 << (qsize + 10);
1269 
1270     uint32_t nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1271     uint32_t nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1272     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1273     uint8_t pq;
1274 
1275     if (!xive_end_is_valid(end)) {
1276         return;
1277     }
1278 
1279     pq = xive_get_field32(END_W1_ESn, end->w1);
1280 
1281     monitor_printf(mon, "  %08x %c%c %c%c%c%c%c%c%c%c prio:%d nvt:%02x/%04x",
1282                    end_idx,
1283                    pq & XIVE_ESB_VAL_P ? 'P' : '-',
1284                    pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1285                    xive_end_is_valid(end)    ? 'v' : '-',
1286                    xive_end_is_enqueue(end)  ? 'q' : '-',
1287                    xive_end_is_notify(end)   ? 'n' : '-',
1288                    xive_end_is_backlog(end)  ? 'b' : '-',
1289                    xive_end_is_escalate(end) ? 'e' : '-',
1290                    xive_end_is_uncond_escalation(end)   ? 'u' : '-',
1291                    xive_end_is_silent_escalation(end)   ? 's' : '-',
1292                    xive_end_is_firmware(end)   ? 'f' : '-',
1293                    priority, nvt_blk, nvt_idx);
1294 
1295     if (qaddr_base) {
1296         monitor_printf(mon, " eq:@%08"PRIx64"% 6d/%5d ^%d",
1297                        qaddr_base, qindex, qentries, qgen);
1298         xive_end_queue_pic_print_info(end, 6, mon);
1299     }
1300     monitor_printf(mon, "\n");
1301 }
1302 
1303 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1304 {
1305     uint64_t qaddr_base = xive_end_qaddr(end);
1306     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1307     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1308     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1309 
1310     uint64_t qaddr = qaddr_base + (qindex << 2);
1311     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1312     uint32_t qentries = 1 << (qsize + 10);
1313 
1314     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata))) {
1315         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1316                       HWADDR_PRIx "\n", qaddr);
1317         return;
1318     }
1319 
1320     qindex = (qindex + 1) & (qentries - 1);
1321     if (qindex == 0) {
1322         qgen ^= 1;
1323         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1324     }
1325     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1326 }
1327 
1328 void xive_end_eas_pic_print_info(XiveEND *end, uint32_t end_idx,
1329                                    Monitor *mon)
1330 {
1331     XiveEAS *eas = (XiveEAS *) &end->w4;
1332     uint8_t pq;
1333 
1334     if (!xive_end_is_escalate(end)) {
1335         return;
1336     }
1337 
1338     pq = xive_get_field32(END_W1_ESe, end->w1);
1339 
1340     monitor_printf(mon, "  %08x %c%c %c%c end:%02x/%04x data:%08x\n",
1341                    end_idx,
1342                    pq & XIVE_ESB_VAL_P ? 'P' : '-',
1343                    pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1344                    xive_eas_is_valid(eas) ? 'V' : ' ',
1345                    xive_eas_is_masked(eas) ? 'M' : ' ',
1346                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1347                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1348                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1349 }
1350 
1351 /*
1352  * XIVE Router (aka. Virtualization Controller or IVRE)
1353  */
1354 
1355 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1356                         XiveEAS *eas)
1357 {
1358     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1359 
1360     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1361 }
1362 
1363 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1364                         XiveEND *end)
1365 {
1366    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1367 
1368    return xrc->get_end(xrtr, end_blk, end_idx, end);
1369 }
1370 
1371 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1372                           XiveEND *end, uint8_t word_number)
1373 {
1374    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1375 
1376    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1377 }
1378 
1379 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1380                         XiveNVT *nvt)
1381 {
1382    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1383 
1384    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1385 }
1386 
1387 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1388                         XiveNVT *nvt, uint8_t word_number)
1389 {
1390    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1391 
1392    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1393 }
1394 
1395 static int xive_router_get_block_id(XiveRouter *xrtr)
1396 {
1397    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1398 
1399    return xrc->get_block_id(xrtr);
1400 }
1401 
1402 static void xive_router_realize(DeviceState *dev, Error **errp)
1403 {
1404     XiveRouter *xrtr = XIVE_ROUTER(dev);
1405 
1406     assert(xrtr->xfb);
1407 }
1408 
1409 /*
1410  * Encode the HW CAM line in the block group mode format :
1411  *
1412  *   chip << 19 | 0000000 0 0001 thread (7Bit)
1413  */
1414 static uint32_t xive_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
1415 {
1416     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1417     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1418     uint8_t blk = xive_router_get_block_id(XIVE_ROUTER(xptr));
1419 
1420     return xive_nvt_cam_line(blk, 1 << 7 | (pir & 0x7f));
1421 }
1422 
1423 /*
1424  * The thread context register words are in big-endian format.
1425  */
1426 int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1427                               uint8_t format,
1428                               uint8_t nvt_blk, uint32_t nvt_idx,
1429                               bool cam_ignore, uint32_t logic_serv)
1430 {
1431     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1432     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1433     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1434     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1435     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1436 
1437     /*
1438      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1439      * identifier are ignored in the "CAM" match.
1440      */
1441 
1442     if (format == 0) {
1443         if (cam_ignore == true) {
1444             /*
1445              * F=0 & i=1: Logical server notification (bits ignored at
1446              * the end of the NVT identifier)
1447              */
1448             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1449                           nvt_blk, nvt_idx);
1450              return -1;
1451         }
1452 
1453         /* F=0 & i=0: Specific NVT notification */
1454 
1455         /* PHYS ring */
1456         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1457             cam == xive_tctx_hw_cam_line(xptr, tctx)) {
1458             return TM_QW3_HV_PHYS;
1459         }
1460 
1461         /* HV POOL ring */
1462         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1463             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1464             return TM_QW2_HV_POOL;
1465         }
1466 
1467         /* OS ring */
1468         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1469             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1470             return TM_QW1_OS;
1471         }
1472     } else {
1473         /* F=1 : User level Event-Based Branch (EBB) notification */
1474 
1475         /* USER ring */
1476         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1477              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1478              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1479              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1480             return TM_QW0_USER;
1481         }
1482     }
1483     return -1;
1484 }
1485 
1486 /*
1487  * This is our simple Xive Presenter Engine model. It is merged in the
1488  * Router as it does not require an extra object.
1489  *
1490  * It receives notification requests sent by the IVRE to find one
1491  * matching NVT (or more) dispatched on the processor threads. In case
1492  * of a single NVT notification, the process is abreviated and the
1493  * thread is signaled if a match is found. In case of a logical server
1494  * notification (bits ignored at the end of the NVT identifier), the
1495  * IVPE and IVRE select a winning thread using different filters. This
1496  * involves 2 or 3 exchanges on the PowerBus that the model does not
1497  * support.
1498  *
1499  * The parameters represent what is sent on the PowerBus
1500  */
1501 bool xive_presenter_notify(XiveFabric *xfb, uint8_t format,
1502                            uint8_t nvt_blk, uint32_t nvt_idx,
1503                            bool cam_ignore, uint8_t priority,
1504                            uint32_t logic_serv)
1505 {
1506     XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xfb);
1507     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1508     int count;
1509 
1510     /*
1511      * Ask the machine to scan the interrupt controllers for a match
1512      */
1513     count = xfc->match_nvt(xfb, format, nvt_blk, nvt_idx, cam_ignore,
1514                            priority, logic_serv, &match);
1515     if (count < 0) {
1516         return false;
1517     }
1518 
1519     /* handle CPU exception delivery */
1520     if (count) {
1521         trace_xive_presenter_notify(nvt_blk, nvt_idx, match.ring);
1522         xive_tctx_ipb_update(match.tctx, match.ring,
1523                              xive_priority_to_ipb(priority));
1524     }
1525 
1526     return !!count;
1527 }
1528 
1529 /*
1530  * Notification using the END ESe/ESn bit (Event State Buffer for
1531  * escalation and notification). Provide further coalescing in the
1532  * Router.
1533  */
1534 static bool xive_router_end_es_notify(XiveRouter *xrtr, uint8_t end_blk,
1535                                       uint32_t end_idx, XiveEND *end,
1536                                       uint32_t end_esmask)
1537 {
1538     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1539     bool notify = xive_esb_trigger(&pq);
1540 
1541     if (pq != xive_get_field32(end_esmask, end->w1)) {
1542         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1543         xive_router_write_end(xrtr, end_blk, end_idx, end, 1);
1544     }
1545 
1546     /* ESe/n[Q]=1 : end of notification */
1547     return notify;
1548 }
1549 
1550 /*
1551  * An END trigger can come from an event trigger (IPI or HW) or from
1552  * another chip. We don't model the PowerBus but the END trigger
1553  * message has the same parameters than in the function below.
1554  */
1555 static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1556                                    uint32_t end_idx, uint32_t end_data)
1557 {
1558     XiveEND end;
1559     uint8_t priority;
1560     uint8_t format;
1561     uint8_t nvt_blk;
1562     uint32_t nvt_idx;
1563     XiveNVT nvt;
1564     bool found;
1565 
1566     /* END cache lookup */
1567     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1568         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1569                       end_idx);
1570         return;
1571     }
1572 
1573     if (!xive_end_is_valid(&end)) {
1574         trace_xive_router_end_notify(end_blk, end_idx, end_data);
1575         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1576                       end_blk, end_idx);
1577         return;
1578     }
1579 
1580     if (xive_end_is_enqueue(&end)) {
1581         xive_end_enqueue(&end, end_data);
1582         /* Enqueuing event data modifies the EQ toggle and index */
1583         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1584     }
1585 
1586     /*
1587      * When the END is silent, we skip the notification part.
1588      */
1589     if (xive_end_is_silent_escalation(&end)) {
1590         goto do_escalation;
1591     }
1592 
1593     /*
1594      * The W7 format depends on the F bit in W6. It defines the type
1595      * of the notification :
1596      *
1597      *   F=0 : single or multiple NVT notification
1598      *   F=1 : User level Event-Based Branch (EBB) notification, no
1599      *         priority
1600      */
1601     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1602     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1603 
1604     /* The END is masked */
1605     if (format == 0 && priority == 0xff) {
1606         return;
1607     }
1608 
1609     /*
1610      * Check the END ESn (Event State Buffer for notification) for
1611      * even further coalescing in the Router
1612      */
1613     if (!xive_end_is_notify(&end)) {
1614         /* ESn[Q]=1 : end of notification */
1615         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1616                                        &end, END_W1_ESn)) {
1617             return;
1618         }
1619     }
1620 
1621     /*
1622      * Follows IVPE notification
1623      */
1624     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end.w6);
1625     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end.w6);
1626 
1627     /* NVT cache lookup */
1628     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1629         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1630                       nvt_blk, nvt_idx);
1631         return;
1632     }
1633 
1634     if (!xive_nvt_is_valid(&nvt)) {
1635         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1636                       nvt_blk, nvt_idx);
1637         return;
1638     }
1639 
1640     found = xive_presenter_notify(xrtr->xfb, format, nvt_blk, nvt_idx,
1641                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1642                           priority,
1643                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1644 
1645     /* TODO: Auto EOI. */
1646 
1647     if (found) {
1648         return;
1649     }
1650 
1651     /*
1652      * If no matching NVT is dispatched on a HW thread :
1653      * - specific VP: update the NVT structure if backlog is activated
1654      * - logical server : forward request to IVPE (not supported)
1655      */
1656     if (xive_end_is_backlog(&end)) {
1657         uint8_t ipb;
1658 
1659         if (format == 1) {
1660             qemu_log_mask(LOG_GUEST_ERROR,
1661                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1662                           end_blk, end_idx);
1663             return;
1664         }
1665         /*
1666          * Record the IPB in the associated NVT structure for later
1667          * use. The presenter will resend the interrupt when the vCPU
1668          * is dispatched again on a HW thread.
1669          */
1670         ipb = xive_get_field32(NVT_W4_IPB, nvt.w4) |
1671             xive_priority_to_ipb(priority);
1672         nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, ipb);
1673         xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1674 
1675         /*
1676          * On HW, follows a "Broadcast Backlog" to IVPEs
1677          */
1678     }
1679 
1680 do_escalation:
1681     /*
1682      * If activated, escalate notification using the ESe PQ bits and
1683      * the EAS in w4-5
1684      */
1685     if (!xive_end_is_escalate(&end)) {
1686         return;
1687     }
1688 
1689     /*
1690      * Check the END ESe (Event State Buffer for escalation) for even
1691      * further coalescing in the Router
1692      */
1693     if (!xive_end_is_uncond_escalation(&end)) {
1694         /* ESe[Q]=1 : end of notification */
1695         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1696                                        &end, END_W1_ESe)) {
1697             return;
1698         }
1699     }
1700 
1701     trace_xive_router_end_escalate(end_blk, end_idx,
1702            (uint8_t) xive_get_field32(END_W4_ESC_END_BLOCK, end.w4),
1703            (uint32_t) xive_get_field32(END_W4_ESC_END_INDEX, end.w4),
1704            (uint32_t) xive_get_field32(END_W5_ESC_END_DATA,  end.w5));
1705     /*
1706      * The END trigger becomes an Escalation trigger
1707      */
1708     xive_router_end_notify(xrtr,
1709                            xive_get_field32(END_W4_ESC_END_BLOCK, end.w4),
1710                            xive_get_field32(END_W4_ESC_END_INDEX, end.w4),
1711                            xive_get_field32(END_W5_ESC_END_DATA,  end.w5));
1712 }
1713 
1714 void xive_router_notify(XiveNotifier *xn, uint32_t lisn)
1715 {
1716     XiveRouter *xrtr = XIVE_ROUTER(xn);
1717     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1718     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1719     XiveEAS eas;
1720 
1721     /* EAS cache lookup */
1722     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1723         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1724         return;
1725     }
1726 
1727     /*
1728      * The IVRE checks the State Bit Cache at this point. We skip the
1729      * SBC lookup because the state bits of the sources are modeled
1730      * internally in QEMU.
1731      */
1732 
1733     if (!xive_eas_is_valid(&eas)) {
1734         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1735         return;
1736     }
1737 
1738     if (xive_eas_is_masked(&eas)) {
1739         /* Notification completed */
1740         return;
1741     }
1742 
1743     /*
1744      * The event trigger becomes an END trigger
1745      */
1746     xive_router_end_notify(xrtr,
1747                            xive_get_field64(EAS_END_BLOCK, eas.w),
1748                            xive_get_field64(EAS_END_INDEX, eas.w),
1749                            xive_get_field64(EAS_END_DATA,  eas.w));
1750 }
1751 
1752 static Property xive_router_properties[] = {
1753     DEFINE_PROP_LINK("xive-fabric", XiveRouter, xfb,
1754                      TYPE_XIVE_FABRIC, XiveFabric *),
1755     DEFINE_PROP_END_OF_LIST(),
1756 };
1757 
1758 static void xive_router_class_init(ObjectClass *klass, void *data)
1759 {
1760     DeviceClass *dc = DEVICE_CLASS(klass);
1761     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1762 
1763     dc->desc    = "XIVE Router Engine";
1764     device_class_set_props(dc, xive_router_properties);
1765     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1766     dc->realize = xive_router_realize;
1767     xnc->notify = xive_router_notify;
1768 }
1769 
1770 static const TypeInfo xive_router_info = {
1771     .name          = TYPE_XIVE_ROUTER,
1772     .parent        = TYPE_SYS_BUS_DEVICE,
1773     .abstract      = true,
1774     .instance_size = sizeof(XiveRouter),
1775     .class_size    = sizeof(XiveRouterClass),
1776     .class_init    = xive_router_class_init,
1777     .interfaces    = (InterfaceInfo[]) {
1778         { TYPE_XIVE_NOTIFIER },
1779         { TYPE_XIVE_PRESENTER },
1780         { }
1781     }
1782 };
1783 
1784 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1785 {
1786     if (!xive_eas_is_valid(eas)) {
1787         return;
1788     }
1789 
1790     monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1791                    lisn, xive_eas_is_masked(eas) ? "M" : " ",
1792                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1793                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1794                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1795 }
1796 
1797 /*
1798  * END ESB MMIO loads
1799  */
1800 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1801 {
1802     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1803     uint32_t offset = addr & 0xFFF;
1804     uint8_t end_blk;
1805     uint32_t end_idx;
1806     XiveEND end;
1807     uint32_t end_esmask;
1808     uint8_t pq;
1809     uint64_t ret = -1;
1810 
1811     /*
1812      * The block id should be deduced from the load address on the END
1813      * ESB MMIO but our model only supports a single block per XIVE chip.
1814      */
1815     end_blk = xive_router_get_block_id(xsrc->xrtr);
1816     end_idx = addr >> (xsrc->esb_shift + 1);
1817 
1818     trace_xive_end_source_read(end_blk, end_idx, addr);
1819 
1820     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1821         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1822                       end_idx);
1823         return -1;
1824     }
1825 
1826     if (!xive_end_is_valid(&end)) {
1827         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1828                       end_blk, end_idx);
1829         return -1;
1830     }
1831 
1832     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1833     pq = xive_get_field32(end_esmask, end.w1);
1834 
1835     switch (offset) {
1836     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1837         ret = xive_esb_eoi(&pq);
1838 
1839         /* Forward the source event notification for routing ?? */
1840         break;
1841 
1842     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1843         ret = pq;
1844         break;
1845 
1846     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1847     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1848     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1849     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1850         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1851         break;
1852     default:
1853         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1854                       offset);
1855         return -1;
1856     }
1857 
1858     if (pq != xive_get_field32(end_esmask, end.w1)) {
1859         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1860         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1861     }
1862 
1863     return ret;
1864 }
1865 
1866 /*
1867  * END ESB MMIO stores are invalid
1868  */
1869 static void xive_end_source_write(void *opaque, hwaddr addr,
1870                                   uint64_t value, unsigned size)
1871 {
1872     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1873                   HWADDR_PRIx"\n", addr);
1874 }
1875 
1876 static const MemoryRegionOps xive_end_source_ops = {
1877     .read = xive_end_source_read,
1878     .write = xive_end_source_write,
1879     .endianness = DEVICE_BIG_ENDIAN,
1880     .valid = {
1881         .min_access_size = 8,
1882         .max_access_size = 8,
1883     },
1884     .impl = {
1885         .min_access_size = 8,
1886         .max_access_size = 8,
1887     },
1888 };
1889 
1890 static void xive_end_source_realize(DeviceState *dev, Error **errp)
1891 {
1892     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1893 
1894     assert(xsrc->xrtr);
1895 
1896     if (!xsrc->nr_ends) {
1897         error_setg(errp, "Number of interrupt needs to be greater than 0");
1898         return;
1899     }
1900 
1901     if (xsrc->esb_shift != XIVE_ESB_4K &&
1902         xsrc->esb_shift != XIVE_ESB_64K) {
1903         error_setg(errp, "Invalid ESB shift setting");
1904         return;
1905     }
1906 
1907     /*
1908      * Each END is assigned an even/odd pair of MMIO pages, the even page
1909      * manages the ESn field while the odd page manages the ESe field.
1910      */
1911     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1912                           &xive_end_source_ops, xsrc, "xive.end",
1913                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1914 }
1915 
1916 static Property xive_end_source_properties[] = {
1917     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
1918     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
1919     DEFINE_PROP_LINK("xive", XiveENDSource, xrtr, TYPE_XIVE_ROUTER,
1920                      XiveRouter *),
1921     DEFINE_PROP_END_OF_LIST(),
1922 };
1923 
1924 static void xive_end_source_class_init(ObjectClass *klass, void *data)
1925 {
1926     DeviceClass *dc = DEVICE_CLASS(klass);
1927 
1928     dc->desc    = "XIVE END Source";
1929     device_class_set_props(dc, xive_end_source_properties);
1930     dc->realize = xive_end_source_realize;
1931     /*
1932      * Reason: part of XIVE interrupt controller, needs to be wired up,
1933      * e.g. by spapr_xive_instance_init().
1934      */
1935     dc->user_creatable = false;
1936 }
1937 
1938 static const TypeInfo xive_end_source_info = {
1939     .name          = TYPE_XIVE_END_SOURCE,
1940     .parent        = TYPE_DEVICE,
1941     .instance_size = sizeof(XiveENDSource),
1942     .class_init    = xive_end_source_class_init,
1943 };
1944 
1945 /*
1946  * XIVE Notifier
1947  */
1948 static const TypeInfo xive_notifier_info = {
1949     .name = TYPE_XIVE_NOTIFIER,
1950     .parent = TYPE_INTERFACE,
1951     .class_size = sizeof(XiveNotifierClass),
1952 };
1953 
1954 /*
1955  * XIVE Presenter
1956  */
1957 static const TypeInfo xive_presenter_info = {
1958     .name = TYPE_XIVE_PRESENTER,
1959     .parent = TYPE_INTERFACE,
1960     .class_size = sizeof(XivePresenterClass),
1961 };
1962 
1963 /*
1964  * XIVE Fabric
1965  */
1966 static const TypeInfo xive_fabric_info = {
1967     .name = TYPE_XIVE_FABRIC,
1968     .parent = TYPE_INTERFACE,
1969     .class_size = sizeof(XiveFabricClass),
1970 };
1971 
1972 static void xive_register_types(void)
1973 {
1974     type_register_static(&xive_fabric_info);
1975     type_register_static(&xive_source_info);
1976     type_register_static(&xive_notifier_info);
1977     type_register_static(&xive_presenter_info);
1978     type_register_static(&xive_router_info);
1979     type_register_static(&xive_end_source_info);
1980     type_register_static(&xive_tctx_info);
1981 }
1982 
1983 type_init(xive_register_types)
1984