xref: /openbmc/qemu/hw/intc/xive.c (revision b0e70950)
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 
25 /*
26  * XIVE Thread Interrupt Management context
27  */
28 
29 /*
30  * Convert a priority number to an Interrupt Pending Buffer (IPB)
31  * register, which indicates a pending interrupt at the priority
32  * corresponding to the bit number
33  */
34 static uint8_t priority_to_ipb(uint8_t priority)
35 {
36     return priority > XIVE_PRIORITY_MAX ?
37         0 : 1 << (XIVE_PRIORITY_MAX - priority);
38 }
39 
40 /*
41  * Convert an Interrupt Pending Buffer (IPB) register to a Pending
42  * Interrupt Priority Register (PIPR), which contains the priority of
43  * the most favored pending notification.
44  */
45 static uint8_t ipb_to_pipr(uint8_t ibp)
46 {
47     return ibp ? clz32((uint32_t)ibp << 24) : 0xff;
48 }
49 
50 static uint8_t exception_mask(uint8_t ring)
51 {
52     switch (ring) {
53     case TM_QW1_OS:
54         return TM_QW1_NSR_EO;
55     case TM_QW3_HV_PHYS:
56         return TM_QW3_NSR_HE;
57     default:
58         g_assert_not_reached();
59     }
60 }
61 
62 static qemu_irq xive_tctx_output(XiveTCTX *tctx, uint8_t ring)
63 {
64         switch (ring) {
65         case TM_QW0_USER:
66                 return 0; /* Not supported */
67         case TM_QW1_OS:
68                 return tctx->os_output;
69         case TM_QW2_HV_POOL:
70         case TM_QW3_HV_PHYS:
71                 return tctx->hv_output;
72         default:
73                 return 0;
74         }
75 }
76 
77 static uint64_t xive_tctx_accept(XiveTCTX *tctx, uint8_t ring)
78 {
79     uint8_t *regs = &tctx->regs[ring];
80     uint8_t nsr = regs[TM_NSR];
81     uint8_t mask = exception_mask(ring);
82 
83     qemu_irq_lower(xive_tctx_output(tctx, ring));
84 
85     if (regs[TM_NSR] & mask) {
86         uint8_t cppr = regs[TM_PIPR];
87 
88         regs[TM_CPPR] = cppr;
89 
90         /* Reset the pending buffer bit */
91         regs[TM_IPB] &= ~priority_to_ipb(cppr);
92         regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
93 
94         /* Drop Exception bit */
95         regs[TM_NSR] &= ~mask;
96     }
97 
98     return (nsr << 8) | regs[TM_CPPR];
99 }
100 
101 static void xive_tctx_notify(XiveTCTX *tctx, uint8_t ring)
102 {
103     uint8_t *regs = &tctx->regs[ring];
104 
105     if (regs[TM_PIPR] < regs[TM_CPPR]) {
106         switch (ring) {
107         case TM_QW1_OS:
108             regs[TM_NSR] |= TM_QW1_NSR_EO;
109             break;
110         case TM_QW3_HV_PHYS:
111             regs[TM_NSR] |= (TM_QW3_NSR_HE_PHYS << 6);
112             break;
113         default:
114             g_assert_not_reached();
115         }
116         qemu_irq_raise(xive_tctx_output(tctx, ring));
117     }
118 }
119 
120 static void xive_tctx_set_cppr(XiveTCTX *tctx, uint8_t ring, uint8_t cppr)
121 {
122     if (cppr > XIVE_PRIORITY_MAX) {
123         cppr = 0xff;
124     }
125 
126     tctx->regs[ring + TM_CPPR] = cppr;
127 
128     /* CPPR has changed, check if we need to raise a pending exception */
129     xive_tctx_notify(tctx, ring);
130 }
131 
132 void xive_tctx_ipb_update(XiveTCTX *tctx, uint8_t ring, uint8_t ipb)
133 {
134     uint8_t *regs = &tctx->regs[ring];
135 
136     regs[TM_IPB] |= ipb;
137     regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
138     xive_tctx_notify(tctx, ring);
139 }
140 
141 static inline uint32_t xive_tctx_word2(uint8_t *ring)
142 {
143     return *((uint32_t *) &ring[TM_WORD2]);
144 }
145 
146 /*
147  * XIVE Thread Interrupt Management Area (TIMA)
148  */
149 
150 static void xive_tm_set_hv_cppr(XivePresenter *xptr, XiveTCTX *tctx,
151                                 hwaddr offset, uint64_t value, unsigned size)
152 {
153     xive_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
154 }
155 
156 static uint64_t xive_tm_ack_hv_reg(XivePresenter *xptr, XiveTCTX *tctx,
157                                    hwaddr offset, unsigned size)
158 {
159     return xive_tctx_accept(tctx, TM_QW3_HV_PHYS);
160 }
161 
162 static uint64_t xive_tm_pull_pool_ctx(XivePresenter *xptr, XiveTCTX *tctx,
163                                       hwaddr offset, unsigned size)
164 {
165     uint32_t qw2w2_prev = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
166     uint32_t qw2w2;
167 
168     qw2w2 = xive_set_field32(TM_QW2W2_VP, qw2w2_prev, 0);
169     memcpy(&tctx->regs[TM_QW2_HV_POOL + TM_WORD2], &qw2w2, 4);
170     return qw2w2;
171 }
172 
173 static void xive_tm_vt_push(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
174                             uint64_t value, unsigned size)
175 {
176     tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] = value & 0xff;
177 }
178 
179 static uint64_t xive_tm_vt_poll(XivePresenter *xptr, XiveTCTX *tctx,
180                                 hwaddr offset, unsigned size)
181 {
182     return tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] & 0xff;
183 }
184 
185 /*
186  * Define an access map for each page of the TIMA that we will use in
187  * the memory region ops to filter values when doing loads and stores
188  * of raw registers values
189  *
190  * Registers accessibility bits :
191  *
192  *    0x0 - no access
193  *    0x1 - write only
194  *    0x2 - read only
195  *    0x3 - read/write
196  */
197 
198 static const uint8_t xive_tm_hw_view[] = {
199     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
200     3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
201     0, 0, 3, 3,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
202     3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   3, 3, 3, 0, /* QW-3 PHYS */
203 };
204 
205 static const uint8_t xive_tm_hv_view[] = {
206     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
207     3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
208     0, 0, 3, 3,   0, 0, 0, 0,   0, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
209     3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   0, 0, 0, 0, /* QW-3 PHYS */
210 };
211 
212 static const uint8_t xive_tm_os_view[] = {
213     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
214     2, 3, 2, 2,   2, 2, 0, 2,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
215     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
216     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
217 };
218 
219 static const uint8_t xive_tm_user_view[] = {
220     3, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-0 User */
221     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
222     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
223     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
224 };
225 
226 /*
227  * Overall TIMA access map for the thread interrupt management context
228  * registers
229  */
230 static const uint8_t *xive_tm_views[] = {
231     [XIVE_TM_HW_PAGE]   = xive_tm_hw_view,
232     [XIVE_TM_HV_PAGE]   = xive_tm_hv_view,
233     [XIVE_TM_OS_PAGE]   = xive_tm_os_view,
234     [XIVE_TM_USER_PAGE] = xive_tm_user_view,
235 };
236 
237 /*
238  * Computes a register access mask for a given offset in the TIMA
239  */
240 static uint64_t xive_tm_mask(hwaddr offset, unsigned size, bool write)
241 {
242     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
243     uint8_t reg_offset = offset & 0x3F;
244     uint8_t reg_mask = write ? 0x1 : 0x2;
245     uint64_t mask = 0x0;
246     int i;
247 
248     for (i = 0; i < size; i++) {
249         if (xive_tm_views[page_offset][reg_offset + i] & reg_mask) {
250             mask |= (uint64_t) 0xff << (8 * (size - i - 1));
251         }
252     }
253 
254     return mask;
255 }
256 
257 static void xive_tm_raw_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
258                               unsigned size)
259 {
260     uint8_t ring_offset = offset & 0x30;
261     uint8_t reg_offset = offset & 0x3F;
262     uint64_t mask = xive_tm_mask(offset, size, true);
263     int i;
264 
265     /*
266      * Only 4 or 8 bytes stores are allowed and the User ring is
267      * excluded
268      */
269     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
270         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA @%"
271                       HWADDR_PRIx"\n", offset);
272         return;
273     }
274 
275     /*
276      * Use the register offset for the raw values and filter out
277      * reserved values
278      */
279     for (i = 0; i < size; i++) {
280         uint8_t byte_mask = (mask >> (8 * (size - i - 1)));
281         if (byte_mask) {
282             tctx->regs[reg_offset + i] = (value >> (8 * (size - i - 1))) &
283                 byte_mask;
284         }
285     }
286 }
287 
288 static uint64_t xive_tm_raw_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
289 {
290     uint8_t ring_offset = offset & 0x30;
291     uint8_t reg_offset = offset & 0x3F;
292     uint64_t mask = xive_tm_mask(offset, size, false);
293     uint64_t ret;
294     int i;
295 
296     /*
297      * Only 4 or 8 bytes loads are allowed and the User ring is
298      * excluded
299      */
300     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
301         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access at TIMA @%"
302                       HWADDR_PRIx"\n", offset);
303         return -1;
304     }
305 
306     /* Use the register offset for the raw values */
307     ret = 0;
308     for (i = 0; i < size; i++) {
309         ret |= (uint64_t) tctx->regs[reg_offset + i] << (8 * (size - i - 1));
310     }
311 
312     /* filter out reserved values */
313     return ret & mask;
314 }
315 
316 /*
317  * The TM context is mapped twice within each page. Stores and loads
318  * to the first mapping below 2K write and read the specified values
319  * without modification. The second mapping above 2K performs specific
320  * state changes (side effects) in addition to setting/returning the
321  * interrupt management area context of the processor thread.
322  */
323 static uint64_t xive_tm_ack_os_reg(XivePresenter *xptr, XiveTCTX *tctx,
324                                    hwaddr offset, unsigned size)
325 {
326     return xive_tctx_accept(tctx, TM_QW1_OS);
327 }
328 
329 static void xive_tm_set_os_cppr(XivePresenter *xptr, XiveTCTX *tctx,
330                                 hwaddr offset, uint64_t value, unsigned size)
331 {
332     xive_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
333 }
334 
335 /*
336  * Adjust the IPB to allow a CPU to process event queues of other
337  * priorities during one physical interrupt cycle.
338  */
339 static void xive_tm_set_os_pending(XivePresenter *xptr, XiveTCTX *tctx,
340                                    hwaddr offset, uint64_t value, unsigned size)
341 {
342     xive_tctx_ipb_update(tctx, TM_QW1_OS, priority_to_ipb(value & 0xff));
343 }
344 
345 static void xive_os_cam_decode(uint32_t cam, uint8_t *nvt_blk,
346                                uint32_t *nvt_idx, bool *vo)
347 {
348     if (nvt_blk) {
349         *nvt_blk = xive_nvt_blk(cam);
350     }
351     if (nvt_idx) {
352         *nvt_idx = xive_nvt_idx(cam);
353     }
354     if (vo) {
355         *vo = !!(cam & TM_QW1W2_VO);
356     }
357 }
358 
359 static uint32_t xive_tctx_get_os_cam(XiveTCTX *tctx, uint8_t *nvt_blk,
360                                      uint32_t *nvt_idx, bool *vo)
361 {
362     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
363     uint32_t cam = be32_to_cpu(qw1w2);
364 
365     xive_os_cam_decode(cam, nvt_blk, nvt_idx, vo);
366     return qw1w2;
367 }
368 
369 static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t qw1w2)
370 {
371     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
372 }
373 
374 static uint64_t xive_tm_pull_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
375                                     hwaddr offset, unsigned size)
376 {
377     uint32_t qw1w2;
378     uint32_t qw1w2_new;
379     uint8_t nvt_blk;
380     uint32_t nvt_idx;
381     bool vo;
382 
383     qw1w2 = xive_tctx_get_os_cam(tctx, &nvt_blk, &nvt_idx, &vo);
384 
385     if (!vo) {
386         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: pulling invalid NVT %x/%x !?\n",
387                       nvt_blk, nvt_idx);
388     }
389 
390     /* Invalidate CAM line */
391     qw1w2_new = xive_set_field32(TM_QW1W2_VO, qw1w2, 0);
392     xive_tctx_set_os_cam(tctx, qw1w2_new);
393     return qw1w2;
394 }
395 
396 static void xive_tctx_need_resend(XiveRouter *xrtr, XiveTCTX *tctx,
397                                   uint8_t nvt_blk, uint32_t nvt_idx)
398 {
399     XiveNVT nvt;
400     uint8_t ipb;
401 
402     /*
403      * Grab the associated NVT to pull the pending bits, and merge
404      * them with the IPB of the thread interrupt context registers
405      */
406     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
407         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVT %x/%x\n",
408                           nvt_blk, nvt_idx);
409         return;
410     }
411 
412     ipb = xive_get_field32(NVT_W4_IPB, nvt.w4);
413 
414     if (ipb) {
415         /* Reset the NVT value */
416         nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, 0);
417         xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
418 
419         /* Merge in current context */
420         xive_tctx_ipb_update(tctx, TM_QW1_OS, ipb);
421     }
422 }
423 
424 /*
425  * Updating the OS CAM line can trigger a resend of interrupt
426  */
427 static void xive_tm_push_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
428                                 hwaddr offset, uint64_t value, unsigned size)
429 {
430     uint32_t cam = value;
431     uint32_t qw1w2 = cpu_to_be32(cam);
432     uint8_t nvt_blk;
433     uint32_t nvt_idx;
434     bool vo;
435 
436     xive_os_cam_decode(cam, &nvt_blk, &nvt_idx, &vo);
437 
438     /* First update the registers */
439     xive_tctx_set_os_cam(tctx, qw1w2);
440 
441     /* Check the interrupt pending bits */
442     if (vo) {
443         xive_tctx_need_resend(XIVE_ROUTER(xptr), tctx, nvt_blk, nvt_idx);
444     }
445 }
446 
447 /*
448  * Define a mapping of "special" operations depending on the TIMA page
449  * offset and the size of the operation.
450  */
451 typedef struct XiveTmOp {
452     uint8_t  page_offset;
453     uint32_t op_offset;
454     unsigned size;
455     void     (*write_handler)(XivePresenter *xptr, XiveTCTX *tctx,
456                               hwaddr offset,
457                               uint64_t value, unsigned size);
458     uint64_t (*read_handler)(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
459                              unsigned size);
460 } XiveTmOp;
461 
462 static const XiveTmOp xive_tm_operations[] = {
463     /*
464      * MMIOs below 2K : raw values and special operations without side
465      * effects
466      */
467     { XIVE_TM_OS_PAGE, TM_QW1_OS + TM_CPPR,   1, xive_tm_set_os_cppr, NULL },
468     { XIVE_TM_HV_PAGE, TM_QW1_OS + TM_WORD2,     4, xive_tm_push_os_ctx, NULL },
469     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_CPPR, 1, xive_tm_set_hv_cppr, NULL },
470     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, xive_tm_vt_push, NULL },
471     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, NULL, xive_tm_vt_poll },
472 
473     /* MMIOs above 2K : special operations with side effects */
474     { XIVE_TM_OS_PAGE, TM_SPC_ACK_OS_REG,     2, NULL, xive_tm_ack_os_reg },
475     { XIVE_TM_OS_PAGE, TM_SPC_SET_OS_PENDING, 1, xive_tm_set_os_pending, NULL },
476     { XIVE_TM_HV_PAGE, TM_SPC_PULL_OS_CTX,    4, NULL, xive_tm_pull_os_ctx },
477     { XIVE_TM_HV_PAGE, TM_SPC_PULL_OS_CTX,    8, NULL, xive_tm_pull_os_ctx },
478     { XIVE_TM_HV_PAGE, TM_SPC_ACK_HV_REG,     2, NULL, xive_tm_ack_hv_reg },
479     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  4, NULL, xive_tm_pull_pool_ctx },
480     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  8, NULL, xive_tm_pull_pool_ctx },
481 };
482 
483 static const XiveTmOp *xive_tm_find_op(hwaddr offset, unsigned size, bool write)
484 {
485     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
486     uint32_t op_offset = offset & 0xFFF;
487     int i;
488 
489     for (i = 0; i < ARRAY_SIZE(xive_tm_operations); i++) {
490         const XiveTmOp *xto = &xive_tm_operations[i];
491 
492         /* Accesses done from a more privileged TIMA page is allowed */
493         if (xto->page_offset >= page_offset &&
494             xto->op_offset == op_offset &&
495             xto->size == size &&
496             ((write && xto->write_handler) || (!write && xto->read_handler))) {
497             return xto;
498         }
499     }
500     return NULL;
501 }
502 
503 /*
504  * TIMA MMIO handlers
505  */
506 void xive_tctx_tm_write(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
507                         uint64_t value, unsigned size)
508 {
509     const XiveTmOp *xto;
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 
549     /*
550      * TODO: check V bit in Q[0-3]W2
551      */
552 
553     /*
554      * First, check for special operations in the 2K region
555      */
556     if (offset & 0x800) {
557         xto = xive_tm_find_op(offset, size, false);
558         if (!xto) {
559             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access to TIMA"
560                           "@%"HWADDR_PRIx"\n", offset);
561             return -1;
562         }
563         return xto->read_handler(xptr, tctx, offset, size);
564     }
565 
566     /*
567      * Then, for special operations in the region below 2K.
568      */
569     xto = xive_tm_find_op(offset, size, false);
570     if (xto) {
571         return xto->read_handler(xptr, tctx, offset, size);
572     }
573 
574     /*
575      * Finish with raw access to the register values
576      */
577     return xive_tm_raw_read(tctx, offset, size);
578 }
579 
580 static char *xive_tctx_ring_print(uint8_t *ring)
581 {
582     uint32_t w2 = xive_tctx_word2(ring);
583 
584     return g_strdup_printf("%02x   %02x  %02x    %02x   %02x  "
585                    "%02x  %02x   %02x  %08x",
586                    ring[TM_NSR], ring[TM_CPPR], ring[TM_IPB], ring[TM_LSMFB],
587                    ring[TM_ACK_CNT], ring[TM_INC], ring[TM_AGE], ring[TM_PIPR],
588                    be32_to_cpu(w2));
589 }
590 
591 static const char * const xive_tctx_ring_names[] = {
592     "USER", "OS", "POOL", "PHYS",
593 };
594 
595 void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
596 {
597     int cpu_index;
598     int i;
599 
600     /* Skip partially initialized vCPUs. This can happen on sPAPR when vCPUs
601      * are hot plugged or unplugged.
602      */
603     if (!tctx) {
604         return;
605     }
606 
607     cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
608 
609     if (kvm_irqchip_in_kernel()) {
610         Error *local_err = NULL;
611 
612         kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
613         if (local_err) {
614             error_report_err(local_err);
615             return;
616         }
617     }
618 
619     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
620                    "  W2\n", cpu_index);
621 
622     for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
623         char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
624         monitor_printf(mon, "CPU[%04x]: %4s    %s\n", cpu_index,
625                        xive_tctx_ring_names[i], s);
626         g_free(s);
627     }
628 }
629 
630 void xive_tctx_reset(XiveTCTX *tctx)
631 {
632     memset(tctx->regs, 0, sizeof(tctx->regs));
633 
634     /* Set some defaults */
635     tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
636     tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
637     tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
638 
639     /*
640      * Initialize PIPR to 0xFF to avoid phantom interrupts when the
641      * CPPR is first set.
642      */
643     tctx->regs[TM_QW1_OS + TM_PIPR] =
644         ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
645     tctx->regs[TM_QW3_HV_PHYS + TM_PIPR] =
646         ipb_to_pipr(tctx->regs[TM_QW3_HV_PHYS + TM_IPB]);
647 }
648 
649 static void xive_tctx_realize(DeviceState *dev, Error **errp)
650 {
651     XiveTCTX *tctx = XIVE_TCTX(dev);
652     PowerPCCPU *cpu;
653     CPUPPCState *env;
654     Error *local_err = NULL;
655 
656     assert(tctx->cs);
657     assert(tctx->xptr);
658 
659     cpu = POWERPC_CPU(tctx->cs);
660     env = &cpu->env;
661     switch (PPC_INPUT(env)) {
662     case PPC_FLAGS_INPUT_POWER9:
663         tctx->hv_output = env->irq_inputs[POWER9_INPUT_HINT];
664         tctx->os_output = env->irq_inputs[POWER9_INPUT_INT];
665         break;
666 
667     default:
668         error_setg(errp, "XIVE interrupt controller does not support "
669                    "this CPU bus model");
670         return;
671     }
672 
673     /* Connect the presenter to the VCPU (required for CPU hotplug) */
674     if (kvm_irqchip_in_kernel()) {
675         kvmppc_xive_cpu_connect(tctx, &local_err);
676         if (local_err) {
677             error_propagate(errp, local_err);
678             return;
679         }
680     }
681 }
682 
683 static int vmstate_xive_tctx_pre_save(void *opaque)
684 {
685     Error *local_err = NULL;
686 
687     if (kvm_irqchip_in_kernel()) {
688         kvmppc_xive_cpu_get_state(XIVE_TCTX(opaque), &local_err);
689         if (local_err) {
690             error_report_err(local_err);
691             return -1;
692         }
693     }
694 
695     return 0;
696 }
697 
698 static int vmstate_xive_tctx_post_load(void *opaque, int version_id)
699 {
700     Error *local_err = NULL;
701 
702     if (kvm_irqchip_in_kernel()) {
703         /*
704          * Required for hotplugged CPU, for which the state comes
705          * after all states of the machine.
706          */
707         kvmppc_xive_cpu_set_state(XIVE_TCTX(opaque), &local_err);
708         if (local_err) {
709             error_report_err(local_err);
710             return -1;
711         }
712     }
713 
714     return 0;
715 }
716 
717 static const VMStateDescription vmstate_xive_tctx = {
718     .name = TYPE_XIVE_TCTX,
719     .version_id = 1,
720     .minimum_version_id = 1,
721     .pre_save = vmstate_xive_tctx_pre_save,
722     .post_load = vmstate_xive_tctx_post_load,
723     .fields = (VMStateField[]) {
724         VMSTATE_BUFFER(regs, XiveTCTX),
725         VMSTATE_END_OF_LIST()
726     },
727 };
728 
729 static Property xive_tctx_properties[] = {
730     DEFINE_PROP_LINK("cpu", XiveTCTX, cs, TYPE_CPU, CPUState *),
731     DEFINE_PROP_LINK("presenter", XiveTCTX, xptr, TYPE_XIVE_PRESENTER,
732                      XivePresenter *),
733     DEFINE_PROP_END_OF_LIST(),
734 };
735 
736 static void xive_tctx_class_init(ObjectClass *klass, void *data)
737 {
738     DeviceClass *dc = DEVICE_CLASS(klass);
739 
740     dc->desc = "XIVE Interrupt Thread Context";
741     dc->realize = xive_tctx_realize;
742     dc->vmsd = &vmstate_xive_tctx;
743     device_class_set_props(dc, xive_tctx_properties);
744     /*
745      * Reason: part of XIVE interrupt controller, needs to be wired up
746      * by xive_tctx_create().
747      */
748     dc->user_creatable = false;
749 }
750 
751 static const TypeInfo xive_tctx_info = {
752     .name          = TYPE_XIVE_TCTX,
753     .parent        = TYPE_DEVICE,
754     .instance_size = sizeof(XiveTCTX),
755     .class_init    = xive_tctx_class_init,
756 };
757 
758 Object *xive_tctx_create(Object *cpu, XivePresenter *xptr, Error **errp)
759 {
760     Error *local_err = NULL;
761     Object *obj;
762 
763     obj = object_new(TYPE_XIVE_TCTX);
764     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj, &error_abort);
765     object_unref(obj);
766     object_property_set_link(obj, cpu, "cpu", &error_abort);
767     object_property_set_link(obj, OBJECT(xptr), "presenter", &error_abort);
768     object_property_set_bool(obj, true, "realized", &local_err);
769     if (local_err) {
770         goto error;
771     }
772 
773     return obj;
774 
775 error:
776     object_unparent(obj);
777     error_propagate(errp, local_err);
778     return NULL;
779 }
780 
781 void xive_tctx_destroy(XiveTCTX *tctx)
782 {
783     Object *obj = OBJECT(tctx);
784 
785     object_unparent(obj);
786 }
787 
788 /*
789  * XIVE ESB helpers
790  */
791 
792 static uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
793 {
794     uint8_t old_pq = *pq & 0x3;
795 
796     *pq &= ~0x3;
797     *pq |= value & 0x3;
798 
799     return old_pq;
800 }
801 
802 static bool xive_esb_trigger(uint8_t *pq)
803 {
804     uint8_t old_pq = *pq & 0x3;
805 
806     switch (old_pq) {
807     case XIVE_ESB_RESET:
808         xive_esb_set(pq, XIVE_ESB_PENDING);
809         return true;
810     case XIVE_ESB_PENDING:
811     case XIVE_ESB_QUEUED:
812         xive_esb_set(pq, XIVE_ESB_QUEUED);
813         return false;
814     case XIVE_ESB_OFF:
815         xive_esb_set(pq, XIVE_ESB_OFF);
816         return false;
817     default:
818          g_assert_not_reached();
819     }
820 }
821 
822 static bool xive_esb_eoi(uint8_t *pq)
823 {
824     uint8_t old_pq = *pq & 0x3;
825 
826     switch (old_pq) {
827     case XIVE_ESB_RESET:
828     case XIVE_ESB_PENDING:
829         xive_esb_set(pq, XIVE_ESB_RESET);
830         return false;
831     case XIVE_ESB_QUEUED:
832         xive_esb_set(pq, XIVE_ESB_PENDING);
833         return true;
834     case XIVE_ESB_OFF:
835         xive_esb_set(pq, XIVE_ESB_OFF);
836         return false;
837     default:
838          g_assert_not_reached();
839     }
840 }
841 
842 /*
843  * XIVE Interrupt Source (or IVSE)
844  */
845 
846 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
847 {
848     assert(srcno < xsrc->nr_irqs);
849 
850     return xsrc->status[srcno] & 0x3;
851 }
852 
853 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
854 {
855     assert(srcno < xsrc->nr_irqs);
856 
857     return xive_esb_set(&xsrc->status[srcno], pq);
858 }
859 
860 /*
861  * Returns whether the event notification should be forwarded.
862  */
863 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
864 {
865     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
866 
867     xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
868 
869     switch (old_pq) {
870     case XIVE_ESB_RESET:
871         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
872         return true;
873     default:
874         return false;
875     }
876 }
877 
878 /*
879  * Returns whether the event notification should be forwarded.
880  */
881 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
882 {
883     bool ret;
884 
885     assert(srcno < xsrc->nr_irqs);
886 
887     ret = xive_esb_trigger(&xsrc->status[srcno]);
888 
889     if (xive_source_irq_is_lsi(xsrc, srcno) &&
890         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
891         qemu_log_mask(LOG_GUEST_ERROR,
892                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
893     }
894 
895     return ret;
896 }
897 
898 /*
899  * Returns whether the event notification should be forwarded.
900  */
901 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
902 {
903     bool ret;
904 
905     assert(srcno < xsrc->nr_irqs);
906 
907     ret = xive_esb_eoi(&xsrc->status[srcno]);
908 
909     /*
910      * LSI sources do not set the Q bit but they can still be
911      * asserted, in which case we should forward a new event
912      * notification
913      */
914     if (xive_source_irq_is_lsi(xsrc, srcno) &&
915         xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
916         ret = xive_source_lsi_trigger(xsrc, srcno);
917     }
918 
919     return ret;
920 }
921 
922 /*
923  * Forward the source event notification to the Router
924  */
925 static void xive_source_notify(XiveSource *xsrc, int srcno)
926 {
927     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
928 
929     if (xnc->notify) {
930         xnc->notify(xsrc->xive, srcno);
931     }
932 }
933 
934 /*
935  * In a two pages ESB MMIO setting, even page is the trigger page, odd
936  * page is for management
937  */
938 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
939 {
940     return !((addr >> shift) & 1);
941 }
942 
943 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
944 {
945     return xive_source_esb_has_2page(xsrc) &&
946         addr_is_even(addr, xsrc->esb_shift - 1);
947 }
948 
949 /*
950  * ESB MMIO loads
951  *                      Trigger page    Management/EOI page
952  *
953  * ESB MMIO setting     2 pages         1 or 2 pages
954  *
955  * 0x000 .. 0x3FF       -1              EOI and return 0|1
956  * 0x400 .. 0x7FF       -1              EOI and return 0|1
957  * 0x800 .. 0xBFF       -1              return PQ
958  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
959  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
960  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
961  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
962  */
963 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
964 {
965     XiveSource *xsrc = XIVE_SOURCE(opaque);
966     uint32_t offset = addr & 0xFFF;
967     uint32_t srcno = addr >> xsrc->esb_shift;
968     uint64_t ret = -1;
969 
970     /* In a two pages ESB MMIO setting, trigger page should not be read */
971     if (xive_source_is_trigger_page(xsrc, addr)) {
972         qemu_log_mask(LOG_GUEST_ERROR,
973                       "XIVE: invalid load on IRQ %d trigger page at "
974                       "0x%"HWADDR_PRIx"\n", srcno, addr);
975         return -1;
976     }
977 
978     switch (offset) {
979     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
980         ret = xive_source_esb_eoi(xsrc, srcno);
981 
982         /* Forward the source event notification for routing */
983         if (ret) {
984             xive_source_notify(xsrc, srcno);
985         }
986         break;
987 
988     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
989         ret = xive_source_esb_get(xsrc, srcno);
990         break;
991 
992     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
993     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
994     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
995     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
996         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
997         break;
998     default:
999         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
1000                       offset);
1001     }
1002 
1003     return ret;
1004 }
1005 
1006 /*
1007  * ESB MMIO stores
1008  *                      Trigger page    Management/EOI page
1009  *
1010  * ESB MMIO setting     2 pages         1 or 2 pages
1011  *
1012  * 0x000 .. 0x3FF       Trigger         Trigger
1013  * 0x400 .. 0x7FF       Trigger         EOI
1014  * 0x800 .. 0xBFF       Trigger         undefined
1015  * 0xC00 .. 0xCFF       Trigger         PQ=00
1016  * 0xD00 .. 0xDFF       Trigger         PQ=01
1017  * 0xE00 .. 0xDFF       Trigger         PQ=10
1018  * 0xF00 .. 0xDFF       Trigger         PQ=11
1019  */
1020 static void xive_source_esb_write(void *opaque, hwaddr addr,
1021                                   uint64_t value, unsigned size)
1022 {
1023     XiveSource *xsrc = XIVE_SOURCE(opaque);
1024     uint32_t offset = addr & 0xFFF;
1025     uint32_t srcno = addr >> xsrc->esb_shift;
1026     bool notify = false;
1027 
1028     /* In a two pages ESB MMIO setting, trigger page only triggers */
1029     if (xive_source_is_trigger_page(xsrc, addr)) {
1030         notify = xive_source_esb_trigger(xsrc, srcno);
1031         goto out;
1032     }
1033 
1034     switch (offset) {
1035     case 0 ... 0x3FF:
1036         notify = xive_source_esb_trigger(xsrc, srcno);
1037         break;
1038 
1039     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
1040         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
1041             qemu_log_mask(LOG_GUEST_ERROR,
1042                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
1043             return;
1044         }
1045 
1046         notify = xive_source_esb_eoi(xsrc, srcno);
1047         break;
1048 
1049     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1050     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1051     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1052     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1053         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1054         break;
1055 
1056     default:
1057         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
1058                       offset);
1059         return;
1060     }
1061 
1062 out:
1063     /* Forward the source event notification for routing */
1064     if (notify) {
1065         xive_source_notify(xsrc, srcno);
1066     }
1067 }
1068 
1069 static const MemoryRegionOps xive_source_esb_ops = {
1070     .read = xive_source_esb_read,
1071     .write = xive_source_esb_write,
1072     .endianness = DEVICE_BIG_ENDIAN,
1073     .valid = {
1074         .min_access_size = 8,
1075         .max_access_size = 8,
1076     },
1077     .impl = {
1078         .min_access_size = 8,
1079         .max_access_size = 8,
1080     },
1081 };
1082 
1083 void xive_source_set_irq(void *opaque, int srcno, int val)
1084 {
1085     XiveSource *xsrc = XIVE_SOURCE(opaque);
1086     bool notify = false;
1087 
1088     if (xive_source_irq_is_lsi(xsrc, srcno)) {
1089         if (val) {
1090             notify = xive_source_lsi_trigger(xsrc, srcno);
1091         } else {
1092             xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
1093         }
1094     } else {
1095         if (val) {
1096             notify = xive_source_esb_trigger(xsrc, srcno);
1097         }
1098     }
1099 
1100     /* Forward the source event notification for routing */
1101     if (notify) {
1102         xive_source_notify(xsrc, srcno);
1103     }
1104 }
1105 
1106 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
1107 {
1108     int i;
1109 
1110     for (i = 0; i < xsrc->nr_irqs; i++) {
1111         uint8_t pq = xive_source_esb_get(xsrc, i);
1112 
1113         if (pq == XIVE_ESB_OFF) {
1114             continue;
1115         }
1116 
1117         monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
1118                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
1119                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
1120                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1121                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ');
1122     }
1123 }
1124 
1125 static void xive_source_reset(void *dev)
1126 {
1127     XiveSource *xsrc = XIVE_SOURCE(dev);
1128 
1129     /* Do not clear the LSI bitmap */
1130 
1131     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
1132     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
1133 }
1134 
1135 static void xive_source_realize(DeviceState *dev, Error **errp)
1136 {
1137     XiveSource *xsrc = XIVE_SOURCE(dev);
1138 
1139     assert(xsrc->xive);
1140 
1141     if (!xsrc->nr_irqs) {
1142         error_setg(errp, "Number of interrupt needs to be greater than 0");
1143         return;
1144     }
1145 
1146     if (xsrc->esb_shift != XIVE_ESB_4K &&
1147         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1148         xsrc->esb_shift != XIVE_ESB_64K &&
1149         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1150         error_setg(errp, "Invalid ESB shift setting");
1151         return;
1152     }
1153 
1154     xsrc->status = g_malloc0(xsrc->nr_irqs);
1155     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1156 
1157     if (!kvm_irqchip_in_kernel()) {
1158         memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1159                               &xive_source_esb_ops, xsrc, "xive.esb",
1160                               (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
1161     }
1162 
1163     qemu_register_reset(xive_source_reset, dev);
1164 }
1165 
1166 static const VMStateDescription vmstate_xive_source = {
1167     .name = TYPE_XIVE_SOURCE,
1168     .version_id = 1,
1169     .minimum_version_id = 1,
1170     .fields = (VMStateField[]) {
1171         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1172         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1173         VMSTATE_END_OF_LIST()
1174     },
1175 };
1176 
1177 /*
1178  * The default XIVE interrupt source setting for the ESB MMIOs is two
1179  * 64k pages without Store EOI, to be in sync with KVM.
1180  */
1181 static Property xive_source_properties[] = {
1182     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1183     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1184     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1185     DEFINE_PROP_LINK("xive", XiveSource, xive, TYPE_XIVE_NOTIFIER,
1186                      XiveNotifier *),
1187     DEFINE_PROP_END_OF_LIST(),
1188 };
1189 
1190 static void xive_source_class_init(ObjectClass *klass, void *data)
1191 {
1192     DeviceClass *dc = DEVICE_CLASS(klass);
1193 
1194     dc->desc    = "XIVE Interrupt Source";
1195     device_class_set_props(dc, xive_source_properties);
1196     dc->realize = xive_source_realize;
1197     dc->vmsd    = &vmstate_xive_source;
1198     /*
1199      * Reason: part of XIVE interrupt controller, needs to be wired up,
1200      * e.g. by spapr_xive_instance_init().
1201      */
1202     dc->user_creatable = false;
1203 }
1204 
1205 static const TypeInfo xive_source_info = {
1206     .name          = TYPE_XIVE_SOURCE,
1207     .parent        = TYPE_DEVICE,
1208     .instance_size = sizeof(XiveSource),
1209     .class_init    = xive_source_class_init,
1210 };
1211 
1212 /*
1213  * XiveEND helpers
1214  */
1215 
1216 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
1217 {
1218     uint64_t qaddr_base = xive_end_qaddr(end);
1219     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1220     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1221     uint32_t qentries = 1 << (qsize + 10);
1222     int i;
1223 
1224     /*
1225      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1226      */
1227     monitor_printf(mon, " [ ");
1228     qindex = (qindex - (width - 1)) & (qentries - 1);
1229     for (i = 0; i < width; i++) {
1230         uint64_t qaddr = qaddr_base + (qindex << 2);
1231         uint32_t qdata = -1;
1232 
1233         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
1234                             sizeof(qdata))) {
1235             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1236                           HWADDR_PRIx "\n", qaddr);
1237             return;
1238         }
1239         monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1240                        be32_to_cpu(qdata));
1241         qindex = (qindex + 1) & (qentries - 1);
1242     }
1243     monitor_printf(mon, "]");
1244 }
1245 
1246 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1247 {
1248     uint64_t qaddr_base = xive_end_qaddr(end);
1249     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1250     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1251     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1252     uint32_t qentries = 1 << (qsize + 10);
1253 
1254     uint32_t nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1255     uint32_t nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1256     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1257     uint8_t pq;
1258 
1259     if (!xive_end_is_valid(end)) {
1260         return;
1261     }
1262 
1263     pq = xive_get_field32(END_W1_ESn, end->w1);
1264 
1265     monitor_printf(mon, "  %08x %c%c %c%c%c%c%c%c%c prio:%d nvt:%02x/%04x",
1266                    end_idx,
1267                    pq & XIVE_ESB_VAL_P ? 'P' : '-',
1268                    pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1269                    xive_end_is_valid(end)    ? 'v' : '-',
1270                    xive_end_is_enqueue(end)  ? 'q' : '-',
1271                    xive_end_is_notify(end)   ? 'n' : '-',
1272                    xive_end_is_backlog(end)  ? 'b' : '-',
1273                    xive_end_is_escalate(end) ? 'e' : '-',
1274                    xive_end_is_uncond_escalation(end)   ? 'u' : '-',
1275                    xive_end_is_silent_escalation(end)   ? 's' : '-',
1276                    priority, nvt_blk, nvt_idx);
1277 
1278     if (qaddr_base) {
1279         monitor_printf(mon, " eq:@%08"PRIx64"% 6d/%5d ^%d",
1280                        qaddr_base, qindex, qentries, qgen);
1281         xive_end_queue_pic_print_info(end, 6, mon);
1282     }
1283     monitor_printf(mon, "\n");
1284 }
1285 
1286 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1287 {
1288     uint64_t qaddr_base = xive_end_qaddr(end);
1289     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1290     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1291     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1292 
1293     uint64_t qaddr = qaddr_base + (qindex << 2);
1294     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1295     uint32_t qentries = 1 << (qsize + 10);
1296 
1297     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata))) {
1298         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1299                       HWADDR_PRIx "\n", qaddr);
1300         return;
1301     }
1302 
1303     qindex = (qindex + 1) & (qentries - 1);
1304     if (qindex == 0) {
1305         qgen ^= 1;
1306         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1307     }
1308     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1309 }
1310 
1311 void xive_end_eas_pic_print_info(XiveEND *end, uint32_t end_idx,
1312                                    Monitor *mon)
1313 {
1314     XiveEAS *eas = (XiveEAS *) &end->w4;
1315     uint8_t pq;
1316 
1317     if (!xive_end_is_escalate(end)) {
1318         return;
1319     }
1320 
1321     pq = xive_get_field32(END_W1_ESe, end->w1);
1322 
1323     monitor_printf(mon, "  %08x %c%c %c%c end:%02x/%04x data:%08x\n",
1324                    end_idx,
1325                    pq & XIVE_ESB_VAL_P ? 'P' : '-',
1326                    pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1327                    xive_eas_is_valid(eas) ? 'V' : ' ',
1328                    xive_eas_is_masked(eas) ? 'M' : ' ',
1329                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1330                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1331                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1332 }
1333 
1334 /*
1335  * XIVE Router (aka. Virtualization Controller or IVRE)
1336  */
1337 
1338 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1339                         XiveEAS *eas)
1340 {
1341     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1342 
1343     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1344 }
1345 
1346 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1347                         XiveEND *end)
1348 {
1349    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1350 
1351    return xrc->get_end(xrtr, end_blk, end_idx, end);
1352 }
1353 
1354 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1355                           XiveEND *end, uint8_t word_number)
1356 {
1357    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1358 
1359    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1360 }
1361 
1362 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1363                         XiveNVT *nvt)
1364 {
1365    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1366 
1367    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1368 }
1369 
1370 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1371                         XiveNVT *nvt, uint8_t word_number)
1372 {
1373    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1374 
1375    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1376 }
1377 
1378 static int xive_router_get_block_id(XiveRouter *xrtr)
1379 {
1380    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1381 
1382    return xrc->get_block_id(xrtr);
1383 }
1384 
1385 static void xive_router_realize(DeviceState *dev, Error **errp)
1386 {
1387     XiveRouter *xrtr = XIVE_ROUTER(dev);
1388 
1389     assert(xrtr->xfb);
1390 }
1391 
1392 /*
1393  * Encode the HW CAM line in the block group mode format :
1394  *
1395  *   chip << 19 | 0000000 0 0001 thread (7Bit)
1396  */
1397 static uint32_t xive_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
1398 {
1399     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1400     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1401     uint8_t blk = xive_router_get_block_id(XIVE_ROUTER(xptr));
1402 
1403     return xive_nvt_cam_line(blk, 1 << 7 | (pir & 0x7f));
1404 }
1405 
1406 /*
1407  * The thread context register words are in big-endian format.
1408  */
1409 int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1410                               uint8_t format,
1411                               uint8_t nvt_blk, uint32_t nvt_idx,
1412                               bool cam_ignore, uint32_t logic_serv)
1413 {
1414     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1415     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1416     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1417     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1418     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1419 
1420     /*
1421      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1422      * identifier are ignored in the "CAM" match.
1423      */
1424 
1425     if (format == 0) {
1426         if (cam_ignore == true) {
1427             /*
1428              * F=0 & i=1: Logical server notification (bits ignored at
1429              * the end of the NVT identifier)
1430              */
1431             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1432                           nvt_blk, nvt_idx);
1433              return -1;
1434         }
1435 
1436         /* F=0 & i=0: Specific NVT notification */
1437 
1438         /* PHYS ring */
1439         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1440             cam == xive_tctx_hw_cam_line(xptr, tctx)) {
1441             return TM_QW3_HV_PHYS;
1442         }
1443 
1444         /* HV POOL ring */
1445         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1446             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1447             return TM_QW2_HV_POOL;
1448         }
1449 
1450         /* OS ring */
1451         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1452             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1453             return TM_QW1_OS;
1454         }
1455     } else {
1456         /* F=1 : User level Event-Based Branch (EBB) notification */
1457 
1458         /* USER ring */
1459         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1460              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1461              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1462              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1463             return TM_QW0_USER;
1464         }
1465     }
1466     return -1;
1467 }
1468 
1469 /*
1470  * This is our simple Xive Presenter Engine model. It is merged in the
1471  * Router as it does not require an extra object.
1472  *
1473  * It receives notification requests sent by the IVRE to find one
1474  * matching NVT (or more) dispatched on the processor threads. In case
1475  * of a single NVT notification, the process is abreviated and the
1476  * thread is signaled if a match is found. In case of a logical server
1477  * notification (bits ignored at the end of the NVT identifier), the
1478  * IVPE and IVRE select a winning thread using different filters. This
1479  * involves 2 or 3 exchanges on the PowerBus that the model does not
1480  * support.
1481  *
1482  * The parameters represent what is sent on the PowerBus
1483  */
1484 static bool xive_presenter_notify(XiveFabric *xfb, uint8_t format,
1485                                   uint8_t nvt_blk, uint32_t nvt_idx,
1486                                   bool cam_ignore, uint8_t priority,
1487                                   uint32_t logic_serv)
1488 {
1489     XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xfb);
1490     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1491     int count;
1492 
1493     /*
1494      * Ask the machine to scan the interrupt controllers for a match
1495      */
1496     count = xfc->match_nvt(xfb, format, nvt_blk, nvt_idx, cam_ignore,
1497                            priority, logic_serv, &match);
1498     if (count < 0) {
1499         return false;
1500     }
1501 
1502     /* handle CPU exception delivery */
1503     if (count) {
1504         xive_tctx_ipb_update(match.tctx, match.ring, priority_to_ipb(priority));
1505     }
1506 
1507     return !!count;
1508 }
1509 
1510 /*
1511  * Notification using the END ESe/ESn bit (Event State Buffer for
1512  * escalation and notification). Profide futher coalescing in the
1513  * Router.
1514  */
1515 static bool xive_router_end_es_notify(XiveRouter *xrtr, uint8_t end_blk,
1516                                       uint32_t end_idx, XiveEND *end,
1517                                       uint32_t end_esmask)
1518 {
1519     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1520     bool notify = xive_esb_trigger(&pq);
1521 
1522     if (pq != xive_get_field32(end_esmask, end->w1)) {
1523         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1524         xive_router_write_end(xrtr, end_blk, end_idx, end, 1);
1525     }
1526 
1527     /* ESe/n[Q]=1 : end of notification */
1528     return notify;
1529 }
1530 
1531 /*
1532  * An END trigger can come from an event trigger (IPI or HW) or from
1533  * another chip. We don't model the PowerBus but the END trigger
1534  * message has the same parameters than in the function below.
1535  */
1536 static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1537                                    uint32_t end_idx, uint32_t end_data)
1538 {
1539     XiveEND end;
1540     uint8_t priority;
1541     uint8_t format;
1542     uint8_t nvt_blk;
1543     uint32_t nvt_idx;
1544     XiveNVT nvt;
1545     bool found;
1546 
1547     /* END cache lookup */
1548     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1549         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1550                       end_idx);
1551         return;
1552     }
1553 
1554     if (!xive_end_is_valid(&end)) {
1555         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1556                       end_blk, end_idx);
1557         return;
1558     }
1559 
1560     if (xive_end_is_enqueue(&end)) {
1561         xive_end_enqueue(&end, end_data);
1562         /* Enqueuing event data modifies the EQ toggle and index */
1563         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1564     }
1565 
1566     /*
1567      * When the END is silent, we skip the notification part.
1568      */
1569     if (xive_end_is_silent_escalation(&end)) {
1570         goto do_escalation;
1571     }
1572 
1573     /*
1574      * The W7 format depends on the F bit in W6. It defines the type
1575      * of the notification :
1576      *
1577      *   F=0 : single or multiple NVT notification
1578      *   F=1 : User level Event-Based Branch (EBB) notification, no
1579      *         priority
1580      */
1581     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1582     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1583 
1584     /* The END is masked */
1585     if (format == 0 && priority == 0xff) {
1586         return;
1587     }
1588 
1589     /*
1590      * Check the END ESn (Event State Buffer for notification) for
1591      * even futher coalescing in the Router
1592      */
1593     if (!xive_end_is_notify(&end)) {
1594         /* ESn[Q]=1 : end of notification */
1595         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1596                                        &end, END_W1_ESn)) {
1597             return;
1598         }
1599     }
1600 
1601     /*
1602      * Follows IVPE notification
1603      */
1604     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end.w6);
1605     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end.w6);
1606 
1607     /* NVT cache lookup */
1608     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1609         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1610                       nvt_blk, nvt_idx);
1611         return;
1612     }
1613 
1614     if (!xive_nvt_is_valid(&nvt)) {
1615         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1616                       nvt_blk, nvt_idx);
1617         return;
1618     }
1619 
1620     found = xive_presenter_notify(xrtr->xfb, format, nvt_blk, nvt_idx,
1621                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1622                           priority,
1623                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1624 
1625     /* TODO: Auto EOI. */
1626 
1627     if (found) {
1628         return;
1629     }
1630 
1631     /*
1632      * If no matching NVT is dispatched on a HW thread :
1633      * - specific VP: update the NVT structure if backlog is activated
1634      * - logical server : forward request to IVPE (not supported)
1635      */
1636     if (xive_end_is_backlog(&end)) {
1637         uint8_t ipb;
1638 
1639         if (format == 1) {
1640             qemu_log_mask(LOG_GUEST_ERROR,
1641                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1642                           end_blk, end_idx);
1643             return;
1644         }
1645         /*
1646          * Record the IPB in the associated NVT structure for later
1647          * use. The presenter will resend the interrupt when the vCPU
1648          * is dispatched again on a HW thread.
1649          */
1650         ipb = xive_get_field32(NVT_W4_IPB, nvt.w4) | priority_to_ipb(priority);
1651         nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, ipb);
1652         xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1653 
1654         /*
1655          * On HW, follows a "Broadcast Backlog" to IVPEs
1656          */
1657     }
1658 
1659 do_escalation:
1660     /*
1661      * If activated, escalate notification using the ESe PQ bits and
1662      * the EAS in w4-5
1663      */
1664     if (!xive_end_is_escalate(&end)) {
1665         return;
1666     }
1667 
1668     /*
1669      * Check the END ESe (Event State Buffer for escalation) for even
1670      * futher coalescing in the Router
1671      */
1672     if (!xive_end_is_uncond_escalation(&end)) {
1673         /* ESe[Q]=1 : end of notification */
1674         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1675                                        &end, END_W1_ESe)) {
1676             return;
1677         }
1678     }
1679 
1680     /*
1681      * The END trigger becomes an Escalation trigger
1682      */
1683     xive_router_end_notify(xrtr,
1684                            xive_get_field32(END_W4_ESC_END_BLOCK, end.w4),
1685                            xive_get_field32(END_W4_ESC_END_INDEX, end.w4),
1686                            xive_get_field32(END_W5_ESC_END_DATA,  end.w5));
1687 }
1688 
1689 void xive_router_notify(XiveNotifier *xn, uint32_t lisn)
1690 {
1691     XiveRouter *xrtr = XIVE_ROUTER(xn);
1692     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1693     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1694     XiveEAS eas;
1695 
1696     /* EAS cache lookup */
1697     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1698         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1699         return;
1700     }
1701 
1702     /*
1703      * The IVRE checks the State Bit Cache at this point. We skip the
1704      * SBC lookup because the state bits of the sources are modeled
1705      * internally in QEMU.
1706      */
1707 
1708     if (!xive_eas_is_valid(&eas)) {
1709         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1710         return;
1711     }
1712 
1713     if (xive_eas_is_masked(&eas)) {
1714         /* Notification completed */
1715         return;
1716     }
1717 
1718     /*
1719      * The event trigger becomes an END trigger
1720      */
1721     xive_router_end_notify(xrtr,
1722                            xive_get_field64(EAS_END_BLOCK, eas.w),
1723                            xive_get_field64(EAS_END_INDEX, eas.w),
1724                            xive_get_field64(EAS_END_DATA,  eas.w));
1725 }
1726 
1727 static Property xive_router_properties[] = {
1728     DEFINE_PROP_LINK("xive-fabric", XiveRouter, xfb,
1729                      TYPE_XIVE_FABRIC, XiveFabric *),
1730     DEFINE_PROP_END_OF_LIST(),
1731 };
1732 
1733 static void xive_router_class_init(ObjectClass *klass, void *data)
1734 {
1735     DeviceClass *dc = DEVICE_CLASS(klass);
1736     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1737 
1738     dc->desc    = "XIVE Router Engine";
1739     device_class_set_props(dc, xive_router_properties);
1740     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1741     dc->realize = xive_router_realize;
1742     xnc->notify = xive_router_notify;
1743 }
1744 
1745 static const TypeInfo xive_router_info = {
1746     .name          = TYPE_XIVE_ROUTER,
1747     .parent        = TYPE_SYS_BUS_DEVICE,
1748     .abstract      = true,
1749     .instance_size = sizeof(XiveRouter),
1750     .class_size    = sizeof(XiveRouterClass),
1751     .class_init    = xive_router_class_init,
1752     .interfaces    = (InterfaceInfo[]) {
1753         { TYPE_XIVE_NOTIFIER },
1754         { TYPE_XIVE_PRESENTER },
1755         { }
1756     }
1757 };
1758 
1759 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1760 {
1761     if (!xive_eas_is_valid(eas)) {
1762         return;
1763     }
1764 
1765     monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1766                    lisn, xive_eas_is_masked(eas) ? "M" : " ",
1767                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1768                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1769                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1770 }
1771 
1772 /*
1773  * END ESB MMIO loads
1774  */
1775 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1776 {
1777     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1778     uint32_t offset = addr & 0xFFF;
1779     uint8_t end_blk;
1780     uint32_t end_idx;
1781     XiveEND end;
1782     uint32_t end_esmask;
1783     uint8_t pq;
1784     uint64_t ret = -1;
1785 
1786     /*
1787      * The block id should be deduced from the load address on the END
1788      * ESB MMIO but our model only supports a single block per XIVE chip.
1789      */
1790     end_blk = xive_router_get_block_id(xsrc->xrtr);
1791     end_idx = addr >> (xsrc->esb_shift + 1);
1792 
1793     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1794         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1795                       end_idx);
1796         return -1;
1797     }
1798 
1799     if (!xive_end_is_valid(&end)) {
1800         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1801                       end_blk, end_idx);
1802         return -1;
1803     }
1804 
1805     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1806     pq = xive_get_field32(end_esmask, end.w1);
1807 
1808     switch (offset) {
1809     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1810         ret = xive_esb_eoi(&pq);
1811 
1812         /* Forward the source event notification for routing ?? */
1813         break;
1814 
1815     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1816         ret = pq;
1817         break;
1818 
1819     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1820     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1821     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1822     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1823         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1824         break;
1825     default:
1826         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1827                       offset);
1828         return -1;
1829     }
1830 
1831     if (pq != xive_get_field32(end_esmask, end.w1)) {
1832         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1833         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1834     }
1835 
1836     return ret;
1837 }
1838 
1839 /*
1840  * END ESB MMIO stores are invalid
1841  */
1842 static void xive_end_source_write(void *opaque, hwaddr addr,
1843                                   uint64_t value, unsigned size)
1844 {
1845     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1846                   HWADDR_PRIx"\n", addr);
1847 }
1848 
1849 static const MemoryRegionOps xive_end_source_ops = {
1850     .read = xive_end_source_read,
1851     .write = xive_end_source_write,
1852     .endianness = DEVICE_BIG_ENDIAN,
1853     .valid = {
1854         .min_access_size = 8,
1855         .max_access_size = 8,
1856     },
1857     .impl = {
1858         .min_access_size = 8,
1859         .max_access_size = 8,
1860     },
1861 };
1862 
1863 static void xive_end_source_realize(DeviceState *dev, Error **errp)
1864 {
1865     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1866 
1867     assert(xsrc->xrtr);
1868 
1869     if (!xsrc->nr_ends) {
1870         error_setg(errp, "Number of interrupt needs to be greater than 0");
1871         return;
1872     }
1873 
1874     if (xsrc->esb_shift != XIVE_ESB_4K &&
1875         xsrc->esb_shift != XIVE_ESB_64K) {
1876         error_setg(errp, "Invalid ESB shift setting");
1877         return;
1878     }
1879 
1880     /*
1881      * Each END is assigned an even/odd pair of MMIO pages, the even page
1882      * manages the ESn field while the odd page manages the ESe field.
1883      */
1884     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1885                           &xive_end_source_ops, xsrc, "xive.end",
1886                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1887 }
1888 
1889 static Property xive_end_source_properties[] = {
1890     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
1891     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
1892     DEFINE_PROP_LINK("xive", XiveENDSource, xrtr, TYPE_XIVE_ROUTER,
1893                      XiveRouter *),
1894     DEFINE_PROP_END_OF_LIST(),
1895 };
1896 
1897 static void xive_end_source_class_init(ObjectClass *klass, void *data)
1898 {
1899     DeviceClass *dc = DEVICE_CLASS(klass);
1900 
1901     dc->desc    = "XIVE END Source";
1902     device_class_set_props(dc, xive_end_source_properties);
1903     dc->realize = xive_end_source_realize;
1904     /*
1905      * Reason: part of XIVE interrupt controller, needs to be wired up,
1906      * e.g. by spapr_xive_instance_init().
1907      */
1908     dc->user_creatable = false;
1909 }
1910 
1911 static const TypeInfo xive_end_source_info = {
1912     .name          = TYPE_XIVE_END_SOURCE,
1913     .parent        = TYPE_DEVICE,
1914     .instance_size = sizeof(XiveENDSource),
1915     .class_init    = xive_end_source_class_init,
1916 };
1917 
1918 /*
1919  * XIVE Notifier
1920  */
1921 static const TypeInfo xive_notifier_info = {
1922     .name = TYPE_XIVE_NOTIFIER,
1923     .parent = TYPE_INTERFACE,
1924     .class_size = sizeof(XiveNotifierClass),
1925 };
1926 
1927 /*
1928  * XIVE Presenter
1929  */
1930 static const TypeInfo xive_presenter_info = {
1931     .name = TYPE_XIVE_PRESENTER,
1932     .parent = TYPE_INTERFACE,
1933     .class_size = sizeof(XivePresenterClass),
1934 };
1935 
1936 /*
1937  * XIVE Fabric
1938  */
1939 static const TypeInfo xive_fabric_info = {
1940     .name = TYPE_XIVE_FABRIC,
1941     .parent = TYPE_INTERFACE,
1942     .class_size = sizeof(XiveFabricClass),
1943 };
1944 
1945 static void xive_register_types(void)
1946 {
1947     type_register_static(&xive_fabric_info);
1948     type_register_static(&xive_source_info);
1949     type_register_static(&xive_notifier_info);
1950     type_register_static(&xive_presenter_info);
1951     type_register_static(&xive_router_info);
1952     type_register_static(&xive_end_source_info);
1953     type_register_static(&xive_tctx_info);
1954 }
1955 
1956 type_init(xive_register_types)
1957