xref: /openbmc/qemu/hw/intc/xive.c (revision 1c8f85d9)
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 
658     cpu = POWERPC_CPU(tctx->cs);
659     env = &cpu->env;
660     switch (PPC_INPUT(env)) {
661     case PPC_FLAGS_INPUT_POWER9:
662         tctx->hv_output = env->irq_inputs[POWER9_INPUT_HINT];
663         tctx->os_output = env->irq_inputs[POWER9_INPUT_INT];
664         break;
665 
666     default:
667         error_setg(errp, "XIVE interrupt controller does not support "
668                    "this CPU bus model");
669         return;
670     }
671 
672     /* Connect the presenter to the VCPU (required for CPU hotplug) */
673     if (kvm_irqchip_in_kernel()) {
674         kvmppc_xive_cpu_connect(tctx, &local_err);
675         if (local_err) {
676             error_propagate(errp, local_err);
677             return;
678         }
679     }
680 }
681 
682 static int vmstate_xive_tctx_pre_save(void *opaque)
683 {
684     Error *local_err = NULL;
685 
686     if (kvm_irqchip_in_kernel()) {
687         kvmppc_xive_cpu_get_state(XIVE_TCTX(opaque), &local_err);
688         if (local_err) {
689             error_report_err(local_err);
690             return -1;
691         }
692     }
693 
694     return 0;
695 }
696 
697 static int vmstate_xive_tctx_post_load(void *opaque, int version_id)
698 {
699     Error *local_err = NULL;
700 
701     if (kvm_irqchip_in_kernel()) {
702         /*
703          * Required for hotplugged CPU, for which the state comes
704          * after all states of the machine.
705          */
706         kvmppc_xive_cpu_set_state(XIVE_TCTX(opaque), &local_err);
707         if (local_err) {
708             error_report_err(local_err);
709             return -1;
710         }
711     }
712 
713     return 0;
714 }
715 
716 static const VMStateDescription vmstate_xive_tctx = {
717     .name = TYPE_XIVE_TCTX,
718     .version_id = 1,
719     .minimum_version_id = 1,
720     .pre_save = vmstate_xive_tctx_pre_save,
721     .post_load = vmstate_xive_tctx_post_load,
722     .fields = (VMStateField[]) {
723         VMSTATE_BUFFER(regs, XiveTCTX),
724         VMSTATE_END_OF_LIST()
725     },
726 };
727 
728 static Property xive_tctx_properties[] = {
729     DEFINE_PROP_LINK("cpu", XiveTCTX, cs, TYPE_CPU, CPUState *),
730     DEFINE_PROP_END_OF_LIST(),
731 };
732 
733 static void xive_tctx_class_init(ObjectClass *klass, void *data)
734 {
735     DeviceClass *dc = DEVICE_CLASS(klass);
736 
737     dc->desc = "XIVE Interrupt Thread Context";
738     dc->realize = xive_tctx_realize;
739     dc->vmsd = &vmstate_xive_tctx;
740     dc->props = xive_tctx_properties;
741     /*
742      * Reason: part of XIVE interrupt controller, needs to be wired up
743      * by xive_tctx_create().
744      */
745     dc->user_creatable = false;
746 }
747 
748 static const TypeInfo xive_tctx_info = {
749     .name          = TYPE_XIVE_TCTX,
750     .parent        = TYPE_DEVICE,
751     .instance_size = sizeof(XiveTCTX),
752     .class_init    = xive_tctx_class_init,
753 };
754 
755 Object *xive_tctx_create(Object *cpu, XiveRouter *xrtr, Error **errp)
756 {
757     Error *local_err = NULL;
758     Object *obj;
759 
760     obj = object_new(TYPE_XIVE_TCTX);
761     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj, &error_abort);
762     object_unref(obj);
763     object_property_set_link(obj, cpu, "cpu", &error_abort);
764     object_property_set_bool(obj, true, "realized", &local_err);
765     if (local_err) {
766         goto error;
767     }
768 
769     return obj;
770 
771 error:
772     object_unparent(obj);
773     error_propagate(errp, local_err);
774     return NULL;
775 }
776 
777 void xive_tctx_destroy(XiveTCTX *tctx)
778 {
779     Object *obj = OBJECT(tctx);
780 
781     object_unparent(obj);
782 }
783 
784 /*
785  * XIVE ESB helpers
786  */
787 
788 static uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
789 {
790     uint8_t old_pq = *pq & 0x3;
791 
792     *pq &= ~0x3;
793     *pq |= value & 0x3;
794 
795     return old_pq;
796 }
797 
798 static bool xive_esb_trigger(uint8_t *pq)
799 {
800     uint8_t old_pq = *pq & 0x3;
801 
802     switch (old_pq) {
803     case XIVE_ESB_RESET:
804         xive_esb_set(pq, XIVE_ESB_PENDING);
805         return true;
806     case XIVE_ESB_PENDING:
807     case XIVE_ESB_QUEUED:
808         xive_esb_set(pq, XIVE_ESB_QUEUED);
809         return false;
810     case XIVE_ESB_OFF:
811         xive_esb_set(pq, XIVE_ESB_OFF);
812         return false;
813     default:
814          g_assert_not_reached();
815     }
816 }
817 
818 static bool xive_esb_eoi(uint8_t *pq)
819 {
820     uint8_t old_pq = *pq & 0x3;
821 
822     switch (old_pq) {
823     case XIVE_ESB_RESET:
824     case XIVE_ESB_PENDING:
825         xive_esb_set(pq, XIVE_ESB_RESET);
826         return false;
827     case XIVE_ESB_QUEUED:
828         xive_esb_set(pq, XIVE_ESB_PENDING);
829         return true;
830     case XIVE_ESB_OFF:
831         xive_esb_set(pq, XIVE_ESB_OFF);
832         return false;
833     default:
834          g_assert_not_reached();
835     }
836 }
837 
838 /*
839  * XIVE Interrupt Source (or IVSE)
840  */
841 
842 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
843 {
844     assert(srcno < xsrc->nr_irqs);
845 
846     return xsrc->status[srcno] & 0x3;
847 }
848 
849 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
850 {
851     assert(srcno < xsrc->nr_irqs);
852 
853     return xive_esb_set(&xsrc->status[srcno], pq);
854 }
855 
856 /*
857  * Returns whether the event notification should be forwarded.
858  */
859 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
860 {
861     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
862 
863     xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
864 
865     switch (old_pq) {
866     case XIVE_ESB_RESET:
867         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
868         return true;
869     default:
870         return false;
871     }
872 }
873 
874 /*
875  * Returns whether the event notification should be forwarded.
876  */
877 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
878 {
879     bool ret;
880 
881     assert(srcno < xsrc->nr_irqs);
882 
883     ret = xive_esb_trigger(&xsrc->status[srcno]);
884 
885     if (xive_source_irq_is_lsi(xsrc, srcno) &&
886         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
887         qemu_log_mask(LOG_GUEST_ERROR,
888                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
889     }
890 
891     return ret;
892 }
893 
894 /*
895  * Returns whether the event notification should be forwarded.
896  */
897 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
898 {
899     bool ret;
900 
901     assert(srcno < xsrc->nr_irqs);
902 
903     ret = xive_esb_eoi(&xsrc->status[srcno]);
904 
905     /*
906      * LSI sources do not set the Q bit but they can still be
907      * asserted, in which case we should forward a new event
908      * notification
909      */
910     if (xive_source_irq_is_lsi(xsrc, srcno) &&
911         xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
912         ret = xive_source_lsi_trigger(xsrc, srcno);
913     }
914 
915     return ret;
916 }
917 
918 /*
919  * Forward the source event notification to the Router
920  */
921 static void xive_source_notify(XiveSource *xsrc, int srcno)
922 {
923     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
924 
925     if (xnc->notify) {
926         xnc->notify(xsrc->xive, srcno);
927     }
928 }
929 
930 /*
931  * In a two pages ESB MMIO setting, even page is the trigger page, odd
932  * page is for management
933  */
934 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
935 {
936     return !((addr >> shift) & 1);
937 }
938 
939 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
940 {
941     return xive_source_esb_has_2page(xsrc) &&
942         addr_is_even(addr, xsrc->esb_shift - 1);
943 }
944 
945 /*
946  * ESB MMIO loads
947  *                      Trigger page    Management/EOI page
948  *
949  * ESB MMIO setting     2 pages         1 or 2 pages
950  *
951  * 0x000 .. 0x3FF       -1              EOI and return 0|1
952  * 0x400 .. 0x7FF       -1              EOI and return 0|1
953  * 0x800 .. 0xBFF       -1              return PQ
954  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
955  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
956  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
957  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
958  */
959 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
960 {
961     XiveSource *xsrc = XIVE_SOURCE(opaque);
962     uint32_t offset = addr & 0xFFF;
963     uint32_t srcno = addr >> xsrc->esb_shift;
964     uint64_t ret = -1;
965 
966     /* In a two pages ESB MMIO setting, trigger page should not be read */
967     if (xive_source_is_trigger_page(xsrc, addr)) {
968         qemu_log_mask(LOG_GUEST_ERROR,
969                       "XIVE: invalid load on IRQ %d trigger page at "
970                       "0x%"HWADDR_PRIx"\n", srcno, addr);
971         return -1;
972     }
973 
974     switch (offset) {
975     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
976         ret = xive_source_esb_eoi(xsrc, srcno);
977 
978         /* Forward the source event notification for routing */
979         if (ret) {
980             xive_source_notify(xsrc, srcno);
981         }
982         break;
983 
984     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
985         ret = xive_source_esb_get(xsrc, srcno);
986         break;
987 
988     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
989     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
990     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
991     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
992         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
993         break;
994     default:
995         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
996                       offset);
997     }
998 
999     return ret;
1000 }
1001 
1002 /*
1003  * ESB MMIO stores
1004  *                      Trigger page    Management/EOI page
1005  *
1006  * ESB MMIO setting     2 pages         1 or 2 pages
1007  *
1008  * 0x000 .. 0x3FF       Trigger         Trigger
1009  * 0x400 .. 0x7FF       Trigger         EOI
1010  * 0x800 .. 0xBFF       Trigger         undefined
1011  * 0xC00 .. 0xCFF       Trigger         PQ=00
1012  * 0xD00 .. 0xDFF       Trigger         PQ=01
1013  * 0xE00 .. 0xDFF       Trigger         PQ=10
1014  * 0xF00 .. 0xDFF       Trigger         PQ=11
1015  */
1016 static void xive_source_esb_write(void *opaque, hwaddr addr,
1017                                   uint64_t value, unsigned size)
1018 {
1019     XiveSource *xsrc = XIVE_SOURCE(opaque);
1020     uint32_t offset = addr & 0xFFF;
1021     uint32_t srcno = addr >> xsrc->esb_shift;
1022     bool notify = false;
1023 
1024     /* In a two pages ESB MMIO setting, trigger page only triggers */
1025     if (xive_source_is_trigger_page(xsrc, addr)) {
1026         notify = xive_source_esb_trigger(xsrc, srcno);
1027         goto out;
1028     }
1029 
1030     switch (offset) {
1031     case 0 ... 0x3FF:
1032         notify = xive_source_esb_trigger(xsrc, srcno);
1033         break;
1034 
1035     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
1036         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
1037             qemu_log_mask(LOG_GUEST_ERROR,
1038                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
1039             return;
1040         }
1041 
1042         notify = xive_source_esb_eoi(xsrc, srcno);
1043         break;
1044 
1045     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1046     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1047     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1048     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1049         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1050         break;
1051 
1052     default:
1053         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
1054                       offset);
1055         return;
1056     }
1057 
1058 out:
1059     /* Forward the source event notification for routing */
1060     if (notify) {
1061         xive_source_notify(xsrc, srcno);
1062     }
1063 }
1064 
1065 static const MemoryRegionOps xive_source_esb_ops = {
1066     .read = xive_source_esb_read,
1067     .write = xive_source_esb_write,
1068     .endianness = DEVICE_BIG_ENDIAN,
1069     .valid = {
1070         .min_access_size = 8,
1071         .max_access_size = 8,
1072     },
1073     .impl = {
1074         .min_access_size = 8,
1075         .max_access_size = 8,
1076     },
1077 };
1078 
1079 void xive_source_set_irq(void *opaque, int srcno, int val)
1080 {
1081     XiveSource *xsrc = XIVE_SOURCE(opaque);
1082     bool notify = false;
1083 
1084     if (xive_source_irq_is_lsi(xsrc, srcno)) {
1085         if (val) {
1086             notify = xive_source_lsi_trigger(xsrc, srcno);
1087         } else {
1088             xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
1089         }
1090     } else {
1091         if (val) {
1092             notify = xive_source_esb_trigger(xsrc, srcno);
1093         }
1094     }
1095 
1096     /* Forward the source event notification for routing */
1097     if (notify) {
1098         xive_source_notify(xsrc, srcno);
1099     }
1100 }
1101 
1102 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
1103 {
1104     int i;
1105 
1106     for (i = 0; i < xsrc->nr_irqs; i++) {
1107         uint8_t pq = xive_source_esb_get(xsrc, i);
1108 
1109         if (pq == XIVE_ESB_OFF) {
1110             continue;
1111         }
1112 
1113         monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
1114                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
1115                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
1116                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1117                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ');
1118     }
1119 }
1120 
1121 static void xive_source_reset(void *dev)
1122 {
1123     XiveSource *xsrc = XIVE_SOURCE(dev);
1124 
1125     /* Do not clear the LSI bitmap */
1126 
1127     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
1128     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
1129 }
1130 
1131 static void xive_source_realize(DeviceState *dev, Error **errp)
1132 {
1133     XiveSource *xsrc = XIVE_SOURCE(dev);
1134 
1135     assert(xsrc->xive);
1136 
1137     if (!xsrc->nr_irqs) {
1138         error_setg(errp, "Number of interrupt needs to be greater than 0");
1139         return;
1140     }
1141 
1142     if (xsrc->esb_shift != XIVE_ESB_4K &&
1143         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1144         xsrc->esb_shift != XIVE_ESB_64K &&
1145         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1146         error_setg(errp, "Invalid ESB shift setting");
1147         return;
1148     }
1149 
1150     xsrc->status = g_malloc0(xsrc->nr_irqs);
1151     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1152 
1153     if (!kvm_irqchip_in_kernel()) {
1154         memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1155                               &xive_source_esb_ops, xsrc, "xive.esb",
1156                               (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
1157     }
1158 
1159     qemu_register_reset(xive_source_reset, dev);
1160 }
1161 
1162 static const VMStateDescription vmstate_xive_source = {
1163     .name = TYPE_XIVE_SOURCE,
1164     .version_id = 1,
1165     .minimum_version_id = 1,
1166     .fields = (VMStateField[]) {
1167         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1168         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1169         VMSTATE_END_OF_LIST()
1170     },
1171 };
1172 
1173 /*
1174  * The default XIVE interrupt source setting for the ESB MMIOs is two
1175  * 64k pages without Store EOI, to be in sync with KVM.
1176  */
1177 static Property xive_source_properties[] = {
1178     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1179     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1180     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1181     DEFINE_PROP_LINK("xive", XiveSource, xive, TYPE_XIVE_NOTIFIER,
1182                      XiveNotifier *),
1183     DEFINE_PROP_END_OF_LIST(),
1184 };
1185 
1186 static void xive_source_class_init(ObjectClass *klass, void *data)
1187 {
1188     DeviceClass *dc = DEVICE_CLASS(klass);
1189 
1190     dc->desc    = "XIVE Interrupt Source";
1191     dc->props   = xive_source_properties;
1192     dc->realize = xive_source_realize;
1193     dc->vmsd    = &vmstate_xive_source;
1194     /*
1195      * Reason: part of XIVE interrupt controller, needs to be wired up,
1196      * e.g. by spapr_xive_instance_init().
1197      */
1198     dc->user_creatable = false;
1199 }
1200 
1201 static const TypeInfo xive_source_info = {
1202     .name          = TYPE_XIVE_SOURCE,
1203     .parent        = TYPE_DEVICE,
1204     .instance_size = sizeof(XiveSource),
1205     .class_init    = xive_source_class_init,
1206 };
1207 
1208 /*
1209  * XiveEND helpers
1210  */
1211 
1212 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
1213 {
1214     uint64_t qaddr_base = xive_end_qaddr(end);
1215     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1216     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1217     uint32_t qentries = 1 << (qsize + 10);
1218     int i;
1219 
1220     /*
1221      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1222      */
1223     monitor_printf(mon, " [ ");
1224     qindex = (qindex - (width - 1)) & (qentries - 1);
1225     for (i = 0; i < width; i++) {
1226         uint64_t qaddr = qaddr_base + (qindex << 2);
1227         uint32_t qdata = -1;
1228 
1229         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
1230                             sizeof(qdata))) {
1231             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1232                           HWADDR_PRIx "\n", qaddr);
1233             return;
1234         }
1235         monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1236                        be32_to_cpu(qdata));
1237         qindex = (qindex + 1) & (qentries - 1);
1238     }
1239     monitor_printf(mon, "]");
1240 }
1241 
1242 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1243 {
1244     uint64_t qaddr_base = xive_end_qaddr(end);
1245     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1246     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1247     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1248     uint32_t qentries = 1 << (qsize + 10);
1249 
1250     uint32_t nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1251     uint32_t nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1252     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1253     uint8_t pq;
1254 
1255     if (!xive_end_is_valid(end)) {
1256         return;
1257     }
1258 
1259     pq = xive_get_field32(END_W1_ESn, end->w1);
1260 
1261     monitor_printf(mon, "  %08x %c%c %c%c%c%c%c%c%c prio:%d nvt:%02x/%04x",
1262                    end_idx,
1263                    pq & XIVE_ESB_VAL_P ? 'P' : '-',
1264                    pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1265                    xive_end_is_valid(end)    ? 'v' : '-',
1266                    xive_end_is_enqueue(end)  ? 'q' : '-',
1267                    xive_end_is_notify(end)   ? 'n' : '-',
1268                    xive_end_is_backlog(end)  ? 'b' : '-',
1269                    xive_end_is_escalate(end) ? 'e' : '-',
1270                    xive_end_is_uncond_escalation(end)   ? 'u' : '-',
1271                    xive_end_is_silent_escalation(end)   ? 's' : '-',
1272                    priority, nvt_blk, nvt_idx);
1273 
1274     if (qaddr_base) {
1275         monitor_printf(mon, " eq:@%08"PRIx64"% 6d/%5d ^%d",
1276                        qaddr_base, qindex, qentries, qgen);
1277         xive_end_queue_pic_print_info(end, 6, mon);
1278     }
1279     monitor_printf(mon, "\n");
1280 }
1281 
1282 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1283 {
1284     uint64_t qaddr_base = xive_end_qaddr(end);
1285     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1286     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1287     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1288 
1289     uint64_t qaddr = qaddr_base + (qindex << 2);
1290     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1291     uint32_t qentries = 1 << (qsize + 10);
1292 
1293     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata))) {
1294         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1295                       HWADDR_PRIx "\n", qaddr);
1296         return;
1297     }
1298 
1299     qindex = (qindex + 1) & (qentries - 1);
1300     if (qindex == 0) {
1301         qgen ^= 1;
1302         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1303     }
1304     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1305 }
1306 
1307 void xive_end_eas_pic_print_info(XiveEND *end, uint32_t end_idx,
1308                                    Monitor *mon)
1309 {
1310     XiveEAS *eas = (XiveEAS *) &end->w4;
1311     uint8_t pq;
1312 
1313     if (!xive_end_is_escalate(end)) {
1314         return;
1315     }
1316 
1317     pq = xive_get_field32(END_W1_ESe, end->w1);
1318 
1319     monitor_printf(mon, "  %08x %c%c %c%c end:%02x/%04x data:%08x\n",
1320                    end_idx,
1321                    pq & XIVE_ESB_VAL_P ? 'P' : '-',
1322                    pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1323                    xive_eas_is_valid(eas) ? 'V' : ' ',
1324                    xive_eas_is_masked(eas) ? 'M' : ' ',
1325                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1326                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1327                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1328 }
1329 
1330 /*
1331  * XIVE Router (aka. Virtualization Controller or IVRE)
1332  */
1333 
1334 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1335                         XiveEAS *eas)
1336 {
1337     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1338 
1339     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1340 }
1341 
1342 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1343                         XiveEND *end)
1344 {
1345    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1346 
1347    return xrc->get_end(xrtr, end_blk, end_idx, end);
1348 }
1349 
1350 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1351                           XiveEND *end, uint8_t word_number)
1352 {
1353    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1354 
1355    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1356 }
1357 
1358 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1359                         XiveNVT *nvt)
1360 {
1361    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1362 
1363    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1364 }
1365 
1366 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1367                         XiveNVT *nvt, uint8_t word_number)
1368 {
1369    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1370 
1371    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1372 }
1373 
1374 static int xive_router_get_block_id(XiveRouter *xrtr)
1375 {
1376    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1377 
1378    return xrc->get_block_id(xrtr);
1379 }
1380 
1381 /*
1382  * Encode the HW CAM line in the block group mode format :
1383  *
1384  *   chip << 19 | 0000000 0 0001 thread (7Bit)
1385  */
1386 static uint32_t xive_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
1387 {
1388     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1389     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1390     uint8_t blk = xive_router_get_block_id(XIVE_ROUTER(xptr));
1391 
1392     return xive_nvt_cam_line(blk, 1 << 7 | (pir & 0x7f));
1393 }
1394 
1395 /*
1396  * The thread context register words are in big-endian format.
1397  */
1398 int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1399                               uint8_t format,
1400                               uint8_t nvt_blk, uint32_t nvt_idx,
1401                               bool cam_ignore, uint32_t logic_serv)
1402 {
1403     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1404     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1405     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1406     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1407     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1408 
1409     /*
1410      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1411      * identifier are ignored in the "CAM" match.
1412      */
1413 
1414     if (format == 0) {
1415         if (cam_ignore == true) {
1416             /*
1417              * F=0 & i=1: Logical server notification (bits ignored at
1418              * the end of the NVT identifier)
1419              */
1420             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1421                           nvt_blk, nvt_idx);
1422              return -1;
1423         }
1424 
1425         /* F=0 & i=0: Specific NVT notification */
1426 
1427         /* PHYS ring */
1428         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1429             cam == xive_tctx_hw_cam_line(xptr, tctx)) {
1430             return TM_QW3_HV_PHYS;
1431         }
1432 
1433         /* HV POOL ring */
1434         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1435             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1436             return TM_QW2_HV_POOL;
1437         }
1438 
1439         /* OS ring */
1440         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1441             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1442             return TM_QW1_OS;
1443         }
1444     } else {
1445         /* F=1 : User level Event-Based Branch (EBB) notification */
1446 
1447         /* USER ring */
1448         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1449              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1450              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1451              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1452             return TM_QW0_USER;
1453         }
1454     }
1455     return -1;
1456 }
1457 
1458 /*
1459  * This is our simple Xive Presenter Engine model. It is merged in the
1460  * Router as it does not require an extra object.
1461  *
1462  * It receives notification requests sent by the IVRE to find one
1463  * matching NVT (or more) dispatched on the processor threads. In case
1464  * of a single NVT notification, the process is abreviated and the
1465  * thread is signaled if a match is found. In case of a logical server
1466  * notification (bits ignored at the end of the NVT identifier), the
1467  * IVPE and IVRE select a winning thread using different filters. This
1468  * involves 2 or 3 exchanges on the PowerBus that the model does not
1469  * support.
1470  *
1471  * The parameters represent what is sent on the PowerBus
1472  */
1473 static bool xive_presenter_notify(uint8_t format,
1474                                   uint8_t nvt_blk, uint32_t nvt_idx,
1475                                   bool cam_ignore, uint8_t priority,
1476                                   uint32_t logic_serv)
1477 {
1478     XiveFabric *xfb = XIVE_FABRIC(qdev_get_machine());
1479     XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xfb);
1480     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1481     int count;
1482 
1483     /*
1484      * Ask the machine to scan the interrupt controllers for a match
1485      */
1486     count = xfc->match_nvt(xfb, format, nvt_blk, nvt_idx, cam_ignore,
1487                            priority, logic_serv, &match);
1488     if (count < 0) {
1489         return false;
1490     }
1491 
1492     /* handle CPU exception delivery */
1493     if (count) {
1494         xive_tctx_ipb_update(match.tctx, match.ring, priority_to_ipb(priority));
1495     }
1496 
1497     return !!count;
1498 }
1499 
1500 /*
1501  * Notification using the END ESe/ESn bit (Event State Buffer for
1502  * escalation and notification). Profide futher coalescing in the
1503  * Router.
1504  */
1505 static bool xive_router_end_es_notify(XiveRouter *xrtr, uint8_t end_blk,
1506                                       uint32_t end_idx, XiveEND *end,
1507                                       uint32_t end_esmask)
1508 {
1509     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1510     bool notify = xive_esb_trigger(&pq);
1511 
1512     if (pq != xive_get_field32(end_esmask, end->w1)) {
1513         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1514         xive_router_write_end(xrtr, end_blk, end_idx, end, 1);
1515     }
1516 
1517     /* ESe/n[Q]=1 : end of notification */
1518     return notify;
1519 }
1520 
1521 /*
1522  * An END trigger can come from an event trigger (IPI or HW) or from
1523  * another chip. We don't model the PowerBus but the END trigger
1524  * message has the same parameters than in the function below.
1525  */
1526 static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1527                                    uint32_t end_idx, uint32_t end_data)
1528 {
1529     XiveEND end;
1530     uint8_t priority;
1531     uint8_t format;
1532     uint8_t nvt_blk;
1533     uint32_t nvt_idx;
1534     XiveNVT nvt;
1535     bool found;
1536 
1537     /* END cache lookup */
1538     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1539         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1540                       end_idx);
1541         return;
1542     }
1543 
1544     if (!xive_end_is_valid(&end)) {
1545         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1546                       end_blk, end_idx);
1547         return;
1548     }
1549 
1550     if (xive_end_is_enqueue(&end)) {
1551         xive_end_enqueue(&end, end_data);
1552         /* Enqueuing event data modifies the EQ toggle and index */
1553         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1554     }
1555 
1556     /*
1557      * When the END is silent, we skip the notification part.
1558      */
1559     if (xive_end_is_silent_escalation(&end)) {
1560         goto do_escalation;
1561     }
1562 
1563     /*
1564      * The W7 format depends on the F bit in W6. It defines the type
1565      * of the notification :
1566      *
1567      *   F=0 : single or multiple NVT notification
1568      *   F=1 : User level Event-Based Branch (EBB) notification, no
1569      *         priority
1570      */
1571     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1572     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1573 
1574     /* The END is masked */
1575     if (format == 0 && priority == 0xff) {
1576         return;
1577     }
1578 
1579     /*
1580      * Check the END ESn (Event State Buffer for notification) for
1581      * even futher coalescing in the Router
1582      */
1583     if (!xive_end_is_notify(&end)) {
1584         /* ESn[Q]=1 : end of notification */
1585         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1586                                        &end, END_W1_ESn)) {
1587             return;
1588         }
1589     }
1590 
1591     /*
1592      * Follows IVPE notification
1593      */
1594     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end.w6);
1595     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end.w6);
1596 
1597     /* NVT cache lookup */
1598     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1599         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1600                       nvt_blk, nvt_idx);
1601         return;
1602     }
1603 
1604     if (!xive_nvt_is_valid(&nvt)) {
1605         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1606                       nvt_blk, nvt_idx);
1607         return;
1608     }
1609 
1610     found = xive_presenter_notify(format, nvt_blk, nvt_idx,
1611                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1612                           priority,
1613                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1614 
1615     /* TODO: Auto EOI. */
1616 
1617     if (found) {
1618         return;
1619     }
1620 
1621     /*
1622      * If no matching NVT is dispatched on a HW thread :
1623      * - specific VP: update the NVT structure if backlog is activated
1624      * - logical server : forward request to IVPE (not supported)
1625      */
1626     if (xive_end_is_backlog(&end)) {
1627         uint8_t ipb;
1628 
1629         if (format == 1) {
1630             qemu_log_mask(LOG_GUEST_ERROR,
1631                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1632                           end_blk, end_idx);
1633             return;
1634         }
1635         /*
1636          * Record the IPB in the associated NVT structure for later
1637          * use. The presenter will resend the interrupt when the vCPU
1638          * is dispatched again on a HW thread.
1639          */
1640         ipb = xive_get_field32(NVT_W4_IPB, nvt.w4) | priority_to_ipb(priority);
1641         nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, ipb);
1642         xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1643 
1644         /*
1645          * On HW, follows a "Broadcast Backlog" to IVPEs
1646          */
1647     }
1648 
1649 do_escalation:
1650     /*
1651      * If activated, escalate notification using the ESe PQ bits and
1652      * the EAS in w4-5
1653      */
1654     if (!xive_end_is_escalate(&end)) {
1655         return;
1656     }
1657 
1658     /*
1659      * Check the END ESe (Event State Buffer for escalation) for even
1660      * futher coalescing in the Router
1661      */
1662     if (!xive_end_is_uncond_escalation(&end)) {
1663         /* ESe[Q]=1 : end of notification */
1664         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1665                                        &end, END_W1_ESe)) {
1666             return;
1667         }
1668     }
1669 
1670     /*
1671      * The END trigger becomes an Escalation trigger
1672      */
1673     xive_router_end_notify(xrtr,
1674                            xive_get_field32(END_W4_ESC_END_BLOCK, end.w4),
1675                            xive_get_field32(END_W4_ESC_END_INDEX, end.w4),
1676                            xive_get_field32(END_W5_ESC_END_DATA,  end.w5));
1677 }
1678 
1679 void xive_router_notify(XiveNotifier *xn, uint32_t lisn)
1680 {
1681     XiveRouter *xrtr = XIVE_ROUTER(xn);
1682     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1683     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1684     XiveEAS eas;
1685 
1686     /* EAS cache lookup */
1687     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1688         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1689         return;
1690     }
1691 
1692     /*
1693      * The IVRE checks the State Bit Cache at this point. We skip the
1694      * SBC lookup because the state bits of the sources are modeled
1695      * internally in QEMU.
1696      */
1697 
1698     if (!xive_eas_is_valid(&eas)) {
1699         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1700         return;
1701     }
1702 
1703     if (xive_eas_is_masked(&eas)) {
1704         /* Notification completed */
1705         return;
1706     }
1707 
1708     /*
1709      * The event trigger becomes an END trigger
1710      */
1711     xive_router_end_notify(xrtr,
1712                            xive_get_field64(EAS_END_BLOCK, eas.w),
1713                            xive_get_field64(EAS_END_INDEX, eas.w),
1714                            xive_get_field64(EAS_END_DATA,  eas.w));
1715 }
1716 
1717 static void xive_router_class_init(ObjectClass *klass, void *data)
1718 {
1719     DeviceClass *dc = DEVICE_CLASS(klass);
1720     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1721 
1722     dc->desc    = "XIVE Router Engine";
1723     xnc->notify = xive_router_notify;
1724 }
1725 
1726 static const TypeInfo xive_router_info = {
1727     .name          = TYPE_XIVE_ROUTER,
1728     .parent        = TYPE_SYS_BUS_DEVICE,
1729     .abstract      = true,
1730     .class_size    = sizeof(XiveRouterClass),
1731     .class_init    = xive_router_class_init,
1732     .interfaces    = (InterfaceInfo[]) {
1733         { TYPE_XIVE_NOTIFIER },
1734         { TYPE_XIVE_PRESENTER },
1735         { }
1736     }
1737 };
1738 
1739 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1740 {
1741     if (!xive_eas_is_valid(eas)) {
1742         return;
1743     }
1744 
1745     monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1746                    lisn, xive_eas_is_masked(eas) ? "M" : " ",
1747                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1748                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1749                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1750 }
1751 
1752 /*
1753  * END ESB MMIO loads
1754  */
1755 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1756 {
1757     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1758     uint32_t offset = addr & 0xFFF;
1759     uint8_t end_blk;
1760     uint32_t end_idx;
1761     XiveEND end;
1762     uint32_t end_esmask;
1763     uint8_t pq;
1764     uint64_t ret = -1;
1765 
1766     /*
1767      * The block id should be deduced from the load address on the END
1768      * ESB MMIO but our model only supports a single block per XIVE chip.
1769      */
1770     end_blk = xive_router_get_block_id(xsrc->xrtr);
1771     end_idx = addr >> (xsrc->esb_shift + 1);
1772 
1773     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1774         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1775                       end_idx);
1776         return -1;
1777     }
1778 
1779     if (!xive_end_is_valid(&end)) {
1780         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1781                       end_blk, end_idx);
1782         return -1;
1783     }
1784 
1785     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1786     pq = xive_get_field32(end_esmask, end.w1);
1787 
1788     switch (offset) {
1789     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1790         ret = xive_esb_eoi(&pq);
1791 
1792         /* Forward the source event notification for routing ?? */
1793         break;
1794 
1795     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1796         ret = pq;
1797         break;
1798 
1799     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1800     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1801     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1802     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1803         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1804         break;
1805     default:
1806         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1807                       offset);
1808         return -1;
1809     }
1810 
1811     if (pq != xive_get_field32(end_esmask, end.w1)) {
1812         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1813         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1814     }
1815 
1816     return ret;
1817 }
1818 
1819 /*
1820  * END ESB MMIO stores are invalid
1821  */
1822 static void xive_end_source_write(void *opaque, hwaddr addr,
1823                                   uint64_t value, unsigned size)
1824 {
1825     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1826                   HWADDR_PRIx"\n", addr);
1827 }
1828 
1829 static const MemoryRegionOps xive_end_source_ops = {
1830     .read = xive_end_source_read,
1831     .write = xive_end_source_write,
1832     .endianness = DEVICE_BIG_ENDIAN,
1833     .valid = {
1834         .min_access_size = 8,
1835         .max_access_size = 8,
1836     },
1837     .impl = {
1838         .min_access_size = 8,
1839         .max_access_size = 8,
1840     },
1841 };
1842 
1843 static void xive_end_source_realize(DeviceState *dev, Error **errp)
1844 {
1845     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1846 
1847     assert(xsrc->xrtr);
1848 
1849     if (!xsrc->nr_ends) {
1850         error_setg(errp, "Number of interrupt needs to be greater than 0");
1851         return;
1852     }
1853 
1854     if (xsrc->esb_shift != XIVE_ESB_4K &&
1855         xsrc->esb_shift != XIVE_ESB_64K) {
1856         error_setg(errp, "Invalid ESB shift setting");
1857         return;
1858     }
1859 
1860     /*
1861      * Each END is assigned an even/odd pair of MMIO pages, the even page
1862      * manages the ESn field while the odd page manages the ESe field.
1863      */
1864     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1865                           &xive_end_source_ops, xsrc, "xive.end",
1866                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1867 }
1868 
1869 static Property xive_end_source_properties[] = {
1870     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
1871     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
1872     DEFINE_PROP_LINK("xive", XiveENDSource, xrtr, TYPE_XIVE_ROUTER,
1873                      XiveRouter *),
1874     DEFINE_PROP_END_OF_LIST(),
1875 };
1876 
1877 static void xive_end_source_class_init(ObjectClass *klass, void *data)
1878 {
1879     DeviceClass *dc = DEVICE_CLASS(klass);
1880 
1881     dc->desc    = "XIVE END Source";
1882     dc->props   = xive_end_source_properties;
1883     dc->realize = xive_end_source_realize;
1884     /*
1885      * Reason: part of XIVE interrupt controller, needs to be wired up,
1886      * e.g. by spapr_xive_instance_init().
1887      */
1888     dc->user_creatable = false;
1889 }
1890 
1891 static const TypeInfo xive_end_source_info = {
1892     .name          = TYPE_XIVE_END_SOURCE,
1893     .parent        = TYPE_DEVICE,
1894     .instance_size = sizeof(XiveENDSource),
1895     .class_init    = xive_end_source_class_init,
1896 };
1897 
1898 /*
1899  * XIVE Notifier
1900  */
1901 static const TypeInfo xive_notifier_info = {
1902     .name = TYPE_XIVE_NOTIFIER,
1903     .parent = TYPE_INTERFACE,
1904     .class_size = sizeof(XiveNotifierClass),
1905 };
1906 
1907 /*
1908  * XIVE Presenter
1909  */
1910 static const TypeInfo xive_presenter_info = {
1911     .name = TYPE_XIVE_PRESENTER,
1912     .parent = TYPE_INTERFACE,
1913     .class_size = sizeof(XivePresenterClass),
1914 };
1915 
1916 /*
1917  * XIVE Fabric
1918  */
1919 static const TypeInfo xive_fabric_info = {
1920     .name = TYPE_XIVE_FABRIC,
1921     .parent = TYPE_INTERFACE,
1922     .class_size = sizeof(XiveFabricClass),
1923 };
1924 
1925 static void xive_register_types(void)
1926 {
1927     type_register_static(&xive_fabric_info);
1928     type_register_static(&xive_source_info);
1929     type_register_static(&xive_notifier_info);
1930     type_register_static(&xive_presenter_info);
1931     type_register_static(&xive_router_info);
1932     type_register_static(&xive_end_source_info);
1933     type_register_static(&xive_tctx_info);
1934 }
1935 
1936 type_init(xive_register_types)
1937