xref: /qemu/hw/intc/xive.c (revision e3a6e0da)
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 /*
596  * kvm_irqchip_in_kernel() will cause the compiler to turn this
597  * info a nop if CONFIG_KVM isn't defined.
598  */
599 #define xive_in_kernel(xptr)                                            \
600     (kvm_irqchip_in_kernel() &&                                         \
601      ({                                                                 \
602          XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);      \
603          xpc->in_kernel ? xpc->in_kernel(xptr) : false;                 \
604      }))
605 
606 void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
607 {
608     int cpu_index;
609     int i;
610 
611     /* Skip partially initialized vCPUs. This can happen on sPAPR when vCPUs
612      * are hot plugged or unplugged.
613      */
614     if (!tctx) {
615         return;
616     }
617 
618     cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
619 
620     if (xive_in_kernel(tctx->xptr)) {
621         Error *local_err = NULL;
622 
623         kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
624         if (local_err) {
625             error_report_err(local_err);
626             return;
627         }
628     }
629 
630     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
631                    "  W2\n", cpu_index);
632 
633     for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
634         char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
635         monitor_printf(mon, "CPU[%04x]: %4s    %s\n", cpu_index,
636                        xive_tctx_ring_names[i], s);
637         g_free(s);
638     }
639 }
640 
641 void xive_tctx_reset(XiveTCTX *tctx)
642 {
643     memset(tctx->regs, 0, sizeof(tctx->regs));
644 
645     /* Set some defaults */
646     tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
647     tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
648     tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
649 
650     /*
651      * Initialize PIPR to 0xFF to avoid phantom interrupts when the
652      * CPPR is first set.
653      */
654     tctx->regs[TM_QW1_OS + TM_PIPR] =
655         ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
656     tctx->regs[TM_QW3_HV_PHYS + TM_PIPR] =
657         ipb_to_pipr(tctx->regs[TM_QW3_HV_PHYS + TM_IPB]);
658 }
659 
660 static void xive_tctx_realize(DeviceState *dev, Error **errp)
661 {
662     XiveTCTX *tctx = XIVE_TCTX(dev);
663     PowerPCCPU *cpu;
664     CPUPPCState *env;
665 
666     assert(tctx->cs);
667     assert(tctx->xptr);
668 
669     cpu = POWERPC_CPU(tctx->cs);
670     env = &cpu->env;
671     switch (PPC_INPUT(env)) {
672     case PPC_FLAGS_INPUT_POWER9:
673         tctx->hv_output = env->irq_inputs[POWER9_INPUT_HINT];
674         tctx->os_output = env->irq_inputs[POWER9_INPUT_INT];
675         break;
676 
677     default:
678         error_setg(errp, "XIVE interrupt controller does not support "
679                    "this CPU bus model");
680         return;
681     }
682 
683     /* Connect the presenter to the VCPU (required for CPU hotplug) */
684     if (xive_in_kernel(tctx->xptr)) {
685         if (kvmppc_xive_cpu_connect(tctx, errp) < 0) {
686             return;
687         }
688     }
689 }
690 
691 static int vmstate_xive_tctx_pre_save(void *opaque)
692 {
693     XiveTCTX *tctx = XIVE_TCTX(opaque);
694     Error *local_err = NULL;
695     int ret;
696 
697     if (xive_in_kernel(tctx->xptr)) {
698         ret = kvmppc_xive_cpu_get_state(tctx, &local_err);
699         if (ret < 0) {
700             error_report_err(local_err);
701             return ret;
702         }
703     }
704 
705     return 0;
706 }
707 
708 static int vmstate_xive_tctx_post_load(void *opaque, int version_id)
709 {
710     XiveTCTX *tctx = XIVE_TCTX(opaque);
711     Error *local_err = NULL;
712     int ret;
713 
714     if (xive_in_kernel(tctx->xptr)) {
715         /*
716          * Required for hotplugged CPU, for which the state comes
717          * after all states of the machine.
718          */
719         ret = kvmppc_xive_cpu_set_state(tctx, &local_err);
720         if (ret < 0) {
721             error_report_err(local_err);
722             return ret;
723         }
724     }
725 
726     return 0;
727 }
728 
729 static const VMStateDescription vmstate_xive_tctx = {
730     .name = TYPE_XIVE_TCTX,
731     .version_id = 1,
732     .minimum_version_id = 1,
733     .pre_save = vmstate_xive_tctx_pre_save,
734     .post_load = vmstate_xive_tctx_post_load,
735     .fields = (VMStateField[]) {
736         VMSTATE_BUFFER(regs, XiveTCTX),
737         VMSTATE_END_OF_LIST()
738     },
739 };
740 
741 static Property xive_tctx_properties[] = {
742     DEFINE_PROP_LINK("cpu", XiveTCTX, cs, TYPE_CPU, CPUState *),
743     DEFINE_PROP_LINK("presenter", XiveTCTX, xptr, TYPE_XIVE_PRESENTER,
744                      XivePresenter *),
745     DEFINE_PROP_END_OF_LIST(),
746 };
747 
748 static void xive_tctx_class_init(ObjectClass *klass, void *data)
749 {
750     DeviceClass *dc = DEVICE_CLASS(klass);
751 
752     dc->desc = "XIVE Interrupt Thread Context";
753     dc->realize = xive_tctx_realize;
754     dc->vmsd = &vmstate_xive_tctx;
755     device_class_set_props(dc, xive_tctx_properties);
756     /*
757      * Reason: part of XIVE interrupt controller, needs to be wired up
758      * by xive_tctx_create().
759      */
760     dc->user_creatable = false;
761 }
762 
763 static const TypeInfo xive_tctx_info = {
764     .name          = TYPE_XIVE_TCTX,
765     .parent        = TYPE_DEVICE,
766     .instance_size = sizeof(XiveTCTX),
767     .class_init    = xive_tctx_class_init,
768 };
769 
770 Object *xive_tctx_create(Object *cpu, XivePresenter *xptr, Error **errp)
771 {
772     Object *obj;
773 
774     obj = object_new(TYPE_XIVE_TCTX);
775     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj);
776     object_unref(obj);
777     object_property_set_link(obj, "cpu", cpu, &error_abort);
778     object_property_set_link(obj, "presenter", OBJECT(xptr), &error_abort);
779     if (!qdev_realize(DEVICE(obj), NULL, errp)) {
780         object_unparent(obj);
781         return NULL;
782     }
783     return obj;
784 }
785 
786 void xive_tctx_destroy(XiveTCTX *tctx)
787 {
788     Object *obj = OBJECT(tctx);
789 
790     object_unparent(obj);
791 }
792 
793 /*
794  * XIVE ESB helpers
795  */
796 
797 static uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
798 {
799     uint8_t old_pq = *pq & 0x3;
800 
801     *pq &= ~0x3;
802     *pq |= value & 0x3;
803 
804     return old_pq;
805 }
806 
807 static bool xive_esb_trigger(uint8_t *pq)
808 {
809     uint8_t old_pq = *pq & 0x3;
810 
811     switch (old_pq) {
812     case XIVE_ESB_RESET:
813         xive_esb_set(pq, XIVE_ESB_PENDING);
814         return true;
815     case XIVE_ESB_PENDING:
816     case XIVE_ESB_QUEUED:
817         xive_esb_set(pq, XIVE_ESB_QUEUED);
818         return false;
819     case XIVE_ESB_OFF:
820         xive_esb_set(pq, XIVE_ESB_OFF);
821         return false;
822     default:
823          g_assert_not_reached();
824     }
825 }
826 
827 static bool xive_esb_eoi(uint8_t *pq)
828 {
829     uint8_t old_pq = *pq & 0x3;
830 
831     switch (old_pq) {
832     case XIVE_ESB_RESET:
833     case XIVE_ESB_PENDING:
834         xive_esb_set(pq, XIVE_ESB_RESET);
835         return false;
836     case XIVE_ESB_QUEUED:
837         xive_esb_set(pq, XIVE_ESB_PENDING);
838         return true;
839     case XIVE_ESB_OFF:
840         xive_esb_set(pq, XIVE_ESB_OFF);
841         return false;
842     default:
843          g_assert_not_reached();
844     }
845 }
846 
847 /*
848  * XIVE Interrupt Source (or IVSE)
849  */
850 
851 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
852 {
853     assert(srcno < xsrc->nr_irqs);
854 
855     return xsrc->status[srcno] & 0x3;
856 }
857 
858 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
859 {
860     assert(srcno < xsrc->nr_irqs);
861 
862     return xive_esb_set(&xsrc->status[srcno], pq);
863 }
864 
865 /*
866  * Returns whether the event notification should be forwarded.
867  */
868 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
869 {
870     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
871 
872     xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
873 
874     switch (old_pq) {
875     case XIVE_ESB_RESET:
876         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
877         return true;
878     default:
879         return false;
880     }
881 }
882 
883 /*
884  * Returns whether the event notification should be forwarded.
885  */
886 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
887 {
888     bool ret;
889 
890     assert(srcno < xsrc->nr_irqs);
891 
892     ret = xive_esb_trigger(&xsrc->status[srcno]);
893 
894     if (xive_source_irq_is_lsi(xsrc, srcno) &&
895         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
896         qemu_log_mask(LOG_GUEST_ERROR,
897                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
898     }
899 
900     return ret;
901 }
902 
903 /*
904  * Returns whether the event notification should be forwarded.
905  */
906 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
907 {
908     bool ret;
909 
910     assert(srcno < xsrc->nr_irqs);
911 
912     ret = xive_esb_eoi(&xsrc->status[srcno]);
913 
914     /*
915      * LSI sources do not set the Q bit but they can still be
916      * asserted, in which case we should forward a new event
917      * notification
918      */
919     if (xive_source_irq_is_lsi(xsrc, srcno) &&
920         xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
921         ret = xive_source_lsi_trigger(xsrc, srcno);
922     }
923 
924     return ret;
925 }
926 
927 /*
928  * Forward the source event notification to the Router
929  */
930 static void xive_source_notify(XiveSource *xsrc, int srcno)
931 {
932     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
933 
934     if (xnc->notify) {
935         xnc->notify(xsrc->xive, srcno);
936     }
937 }
938 
939 /*
940  * In a two pages ESB MMIO setting, even page is the trigger page, odd
941  * page is for management
942  */
943 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
944 {
945     return !((addr >> shift) & 1);
946 }
947 
948 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
949 {
950     return xive_source_esb_has_2page(xsrc) &&
951         addr_is_even(addr, xsrc->esb_shift - 1);
952 }
953 
954 /*
955  * ESB MMIO loads
956  *                      Trigger page    Management/EOI page
957  *
958  * ESB MMIO setting     2 pages         1 or 2 pages
959  *
960  * 0x000 .. 0x3FF       -1              EOI and return 0|1
961  * 0x400 .. 0x7FF       -1              EOI and return 0|1
962  * 0x800 .. 0xBFF       -1              return PQ
963  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
964  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
965  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
966  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
967  */
968 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
969 {
970     XiveSource *xsrc = XIVE_SOURCE(opaque);
971     uint32_t offset = addr & 0xFFF;
972     uint32_t srcno = addr >> xsrc->esb_shift;
973     uint64_t ret = -1;
974 
975     /* In a two pages ESB MMIO setting, trigger page should not be read */
976     if (xive_source_is_trigger_page(xsrc, addr)) {
977         qemu_log_mask(LOG_GUEST_ERROR,
978                       "XIVE: invalid load on IRQ %d trigger page at "
979                       "0x%"HWADDR_PRIx"\n", srcno, addr);
980         return -1;
981     }
982 
983     switch (offset) {
984     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
985         ret = xive_source_esb_eoi(xsrc, srcno);
986 
987         /* Forward the source event notification for routing */
988         if (ret) {
989             xive_source_notify(xsrc, srcno);
990         }
991         break;
992 
993     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
994         ret = xive_source_esb_get(xsrc, srcno);
995         break;
996 
997     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
998     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
999     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1000     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1001         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1002         break;
1003     default:
1004         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
1005                       offset);
1006     }
1007 
1008     return ret;
1009 }
1010 
1011 /*
1012  * ESB MMIO stores
1013  *                      Trigger page    Management/EOI page
1014  *
1015  * ESB MMIO setting     2 pages         1 or 2 pages
1016  *
1017  * 0x000 .. 0x3FF       Trigger         Trigger
1018  * 0x400 .. 0x7FF       Trigger         EOI
1019  * 0x800 .. 0xBFF       Trigger         undefined
1020  * 0xC00 .. 0xCFF       Trigger         PQ=00
1021  * 0xD00 .. 0xDFF       Trigger         PQ=01
1022  * 0xE00 .. 0xDFF       Trigger         PQ=10
1023  * 0xF00 .. 0xDFF       Trigger         PQ=11
1024  */
1025 static void xive_source_esb_write(void *opaque, hwaddr addr,
1026                                   uint64_t value, unsigned size)
1027 {
1028     XiveSource *xsrc = XIVE_SOURCE(opaque);
1029     uint32_t offset = addr & 0xFFF;
1030     uint32_t srcno = addr >> xsrc->esb_shift;
1031     bool notify = false;
1032 
1033     /* In a two pages ESB MMIO setting, trigger page only triggers */
1034     if (xive_source_is_trigger_page(xsrc, addr)) {
1035         notify = xive_source_esb_trigger(xsrc, srcno);
1036         goto out;
1037     }
1038 
1039     switch (offset) {
1040     case 0 ... 0x3FF:
1041         notify = xive_source_esb_trigger(xsrc, srcno);
1042         break;
1043 
1044     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
1045         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
1046             qemu_log_mask(LOG_GUEST_ERROR,
1047                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
1048             return;
1049         }
1050 
1051         notify = xive_source_esb_eoi(xsrc, srcno);
1052         break;
1053 
1054     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1055     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1056     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1057     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1058         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1059         break;
1060 
1061     default:
1062         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
1063                       offset);
1064         return;
1065     }
1066 
1067 out:
1068     /* Forward the source event notification for routing */
1069     if (notify) {
1070         xive_source_notify(xsrc, srcno);
1071     }
1072 }
1073 
1074 static const MemoryRegionOps xive_source_esb_ops = {
1075     .read = xive_source_esb_read,
1076     .write = xive_source_esb_write,
1077     .endianness = DEVICE_BIG_ENDIAN,
1078     .valid = {
1079         .min_access_size = 8,
1080         .max_access_size = 8,
1081     },
1082     .impl = {
1083         .min_access_size = 8,
1084         .max_access_size = 8,
1085     },
1086 };
1087 
1088 void xive_source_set_irq(void *opaque, int srcno, int val)
1089 {
1090     XiveSource *xsrc = XIVE_SOURCE(opaque);
1091     bool notify = false;
1092 
1093     if (xive_source_irq_is_lsi(xsrc, srcno)) {
1094         if (val) {
1095             notify = xive_source_lsi_trigger(xsrc, srcno);
1096         } else {
1097             xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
1098         }
1099     } else {
1100         if (val) {
1101             notify = xive_source_esb_trigger(xsrc, srcno);
1102         }
1103     }
1104 
1105     /* Forward the source event notification for routing */
1106     if (notify) {
1107         xive_source_notify(xsrc, srcno);
1108     }
1109 }
1110 
1111 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
1112 {
1113     int i;
1114 
1115     for (i = 0; i < xsrc->nr_irqs; i++) {
1116         uint8_t pq = xive_source_esb_get(xsrc, i);
1117 
1118         if (pq == XIVE_ESB_OFF) {
1119             continue;
1120         }
1121 
1122         monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
1123                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
1124                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
1125                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1126                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ');
1127     }
1128 }
1129 
1130 static void xive_source_reset(void *dev)
1131 {
1132     XiveSource *xsrc = XIVE_SOURCE(dev);
1133 
1134     /* Do not clear the LSI bitmap */
1135 
1136     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
1137     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
1138 }
1139 
1140 static void xive_source_realize(DeviceState *dev, Error **errp)
1141 {
1142     XiveSource *xsrc = XIVE_SOURCE(dev);
1143     size_t esb_len = xive_source_esb_len(xsrc);
1144 
1145     assert(xsrc->xive);
1146 
1147     if (!xsrc->nr_irqs) {
1148         error_setg(errp, "Number of interrupt needs to be greater than 0");
1149         return;
1150     }
1151 
1152     if (xsrc->esb_shift != XIVE_ESB_4K &&
1153         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1154         xsrc->esb_shift != XIVE_ESB_64K &&
1155         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1156         error_setg(errp, "Invalid ESB shift setting");
1157         return;
1158     }
1159 
1160     xsrc->status = g_malloc0(xsrc->nr_irqs);
1161     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1162 
1163     memory_region_init(&xsrc->esb_mmio, OBJECT(xsrc), "xive.esb", esb_len);
1164     memory_region_init_io(&xsrc->esb_mmio_emulated, OBJECT(xsrc),
1165                           &xive_source_esb_ops, xsrc, "xive.esb-emulated",
1166                           esb_len);
1167     memory_region_add_subregion(&xsrc->esb_mmio, 0, &xsrc->esb_mmio_emulated);
1168 
1169     qemu_register_reset(xive_source_reset, dev);
1170 }
1171 
1172 static const VMStateDescription vmstate_xive_source = {
1173     .name = TYPE_XIVE_SOURCE,
1174     .version_id = 1,
1175     .minimum_version_id = 1,
1176     .fields = (VMStateField[]) {
1177         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1178         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1179         VMSTATE_END_OF_LIST()
1180     },
1181 };
1182 
1183 /*
1184  * The default XIVE interrupt source setting for the ESB MMIOs is two
1185  * 64k pages without Store EOI, to be in sync with KVM.
1186  */
1187 static Property xive_source_properties[] = {
1188     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1189     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1190     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1191     DEFINE_PROP_LINK("xive", XiveSource, xive, TYPE_XIVE_NOTIFIER,
1192                      XiveNotifier *),
1193     DEFINE_PROP_END_OF_LIST(),
1194 };
1195 
1196 static void xive_source_class_init(ObjectClass *klass, void *data)
1197 {
1198     DeviceClass *dc = DEVICE_CLASS(klass);
1199 
1200     dc->desc    = "XIVE Interrupt Source";
1201     device_class_set_props(dc, xive_source_properties);
1202     dc->realize = xive_source_realize;
1203     dc->vmsd    = &vmstate_xive_source;
1204     /*
1205      * Reason: part of XIVE interrupt controller, needs to be wired up,
1206      * e.g. by spapr_xive_instance_init().
1207      */
1208     dc->user_creatable = false;
1209 }
1210 
1211 static const TypeInfo xive_source_info = {
1212     .name          = TYPE_XIVE_SOURCE,
1213     .parent        = TYPE_DEVICE,
1214     .instance_size = sizeof(XiveSource),
1215     .class_init    = xive_source_class_init,
1216 };
1217 
1218 /*
1219  * XiveEND helpers
1220  */
1221 
1222 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
1223 {
1224     uint64_t qaddr_base = xive_end_qaddr(end);
1225     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1226     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1227     uint32_t qentries = 1 << (qsize + 10);
1228     int i;
1229 
1230     /*
1231      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1232      */
1233     monitor_printf(mon, " [ ");
1234     qindex = (qindex - (width - 1)) & (qentries - 1);
1235     for (i = 0; i < width; i++) {
1236         uint64_t qaddr = qaddr_base + (qindex << 2);
1237         uint32_t qdata = -1;
1238 
1239         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
1240                             sizeof(qdata))) {
1241             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1242                           HWADDR_PRIx "\n", qaddr);
1243             return;
1244         }
1245         monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1246                        be32_to_cpu(qdata));
1247         qindex = (qindex + 1) & (qentries - 1);
1248     }
1249     monitor_printf(mon, "]");
1250 }
1251 
1252 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1253 {
1254     uint64_t qaddr_base = xive_end_qaddr(end);
1255     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1256     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1257     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1258     uint32_t qentries = 1 << (qsize + 10);
1259 
1260     uint32_t nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1261     uint32_t nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1262     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1263     uint8_t pq;
1264 
1265     if (!xive_end_is_valid(end)) {
1266         return;
1267     }
1268 
1269     pq = xive_get_field32(END_W1_ESn, end->w1);
1270 
1271     monitor_printf(mon, "  %08x %c%c %c%c%c%c%c%c%c prio:%d nvt:%02x/%04x",
1272                    end_idx,
1273                    pq & XIVE_ESB_VAL_P ? 'P' : '-',
1274                    pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1275                    xive_end_is_valid(end)    ? 'v' : '-',
1276                    xive_end_is_enqueue(end)  ? 'q' : '-',
1277                    xive_end_is_notify(end)   ? 'n' : '-',
1278                    xive_end_is_backlog(end)  ? 'b' : '-',
1279                    xive_end_is_escalate(end) ? 'e' : '-',
1280                    xive_end_is_uncond_escalation(end)   ? 'u' : '-',
1281                    xive_end_is_silent_escalation(end)   ? 's' : '-',
1282                    priority, nvt_blk, nvt_idx);
1283 
1284     if (qaddr_base) {
1285         monitor_printf(mon, " eq:@%08"PRIx64"% 6d/%5d ^%d",
1286                        qaddr_base, qindex, qentries, qgen);
1287         xive_end_queue_pic_print_info(end, 6, mon);
1288     }
1289     monitor_printf(mon, "\n");
1290 }
1291 
1292 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1293 {
1294     uint64_t qaddr_base = xive_end_qaddr(end);
1295     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1296     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1297     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1298 
1299     uint64_t qaddr = qaddr_base + (qindex << 2);
1300     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1301     uint32_t qentries = 1 << (qsize + 10);
1302 
1303     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata))) {
1304         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1305                       HWADDR_PRIx "\n", qaddr);
1306         return;
1307     }
1308 
1309     qindex = (qindex + 1) & (qentries - 1);
1310     if (qindex == 0) {
1311         qgen ^= 1;
1312         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1313     }
1314     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1315 }
1316 
1317 void xive_end_eas_pic_print_info(XiveEND *end, uint32_t end_idx,
1318                                    Monitor *mon)
1319 {
1320     XiveEAS *eas = (XiveEAS *) &end->w4;
1321     uint8_t pq;
1322 
1323     if (!xive_end_is_escalate(end)) {
1324         return;
1325     }
1326 
1327     pq = xive_get_field32(END_W1_ESe, end->w1);
1328 
1329     monitor_printf(mon, "  %08x %c%c %c%c end:%02x/%04x data:%08x\n",
1330                    end_idx,
1331                    pq & XIVE_ESB_VAL_P ? 'P' : '-',
1332                    pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1333                    xive_eas_is_valid(eas) ? 'V' : ' ',
1334                    xive_eas_is_masked(eas) ? 'M' : ' ',
1335                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1336                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1337                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1338 }
1339 
1340 /*
1341  * XIVE Router (aka. Virtualization Controller or IVRE)
1342  */
1343 
1344 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1345                         XiveEAS *eas)
1346 {
1347     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1348 
1349     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1350 }
1351 
1352 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1353                         XiveEND *end)
1354 {
1355    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1356 
1357    return xrc->get_end(xrtr, end_blk, end_idx, end);
1358 }
1359 
1360 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1361                           XiveEND *end, uint8_t word_number)
1362 {
1363    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1364 
1365    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1366 }
1367 
1368 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1369                         XiveNVT *nvt)
1370 {
1371    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1372 
1373    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1374 }
1375 
1376 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1377                         XiveNVT *nvt, uint8_t word_number)
1378 {
1379    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1380 
1381    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1382 }
1383 
1384 static int xive_router_get_block_id(XiveRouter *xrtr)
1385 {
1386    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1387 
1388    return xrc->get_block_id(xrtr);
1389 }
1390 
1391 static void xive_router_realize(DeviceState *dev, Error **errp)
1392 {
1393     XiveRouter *xrtr = XIVE_ROUTER(dev);
1394 
1395     assert(xrtr->xfb);
1396 }
1397 
1398 /*
1399  * Encode the HW CAM line in the block group mode format :
1400  *
1401  *   chip << 19 | 0000000 0 0001 thread (7Bit)
1402  */
1403 static uint32_t xive_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
1404 {
1405     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1406     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1407     uint8_t blk = xive_router_get_block_id(XIVE_ROUTER(xptr));
1408 
1409     return xive_nvt_cam_line(blk, 1 << 7 | (pir & 0x7f));
1410 }
1411 
1412 /*
1413  * The thread context register words are in big-endian format.
1414  */
1415 int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1416                               uint8_t format,
1417                               uint8_t nvt_blk, uint32_t nvt_idx,
1418                               bool cam_ignore, uint32_t logic_serv)
1419 {
1420     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1421     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1422     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1423     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1424     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1425 
1426     /*
1427      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1428      * identifier are ignored in the "CAM" match.
1429      */
1430 
1431     if (format == 0) {
1432         if (cam_ignore == true) {
1433             /*
1434              * F=0 & i=1: Logical server notification (bits ignored at
1435              * the end of the NVT identifier)
1436              */
1437             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1438                           nvt_blk, nvt_idx);
1439              return -1;
1440         }
1441 
1442         /* F=0 & i=0: Specific NVT notification */
1443 
1444         /* PHYS ring */
1445         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1446             cam == xive_tctx_hw_cam_line(xptr, tctx)) {
1447             return TM_QW3_HV_PHYS;
1448         }
1449 
1450         /* HV POOL ring */
1451         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1452             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1453             return TM_QW2_HV_POOL;
1454         }
1455 
1456         /* OS ring */
1457         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1458             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1459             return TM_QW1_OS;
1460         }
1461     } else {
1462         /* F=1 : User level Event-Based Branch (EBB) notification */
1463 
1464         /* USER ring */
1465         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1466              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1467              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1468              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1469             return TM_QW0_USER;
1470         }
1471     }
1472     return -1;
1473 }
1474 
1475 /*
1476  * This is our simple Xive Presenter Engine model. It is merged in the
1477  * Router as it does not require an extra object.
1478  *
1479  * It receives notification requests sent by the IVRE to find one
1480  * matching NVT (or more) dispatched on the processor threads. In case
1481  * of a single NVT notification, the process is abreviated and the
1482  * thread is signaled if a match is found. In case of a logical server
1483  * notification (bits ignored at the end of the NVT identifier), the
1484  * IVPE and IVRE select a winning thread using different filters. This
1485  * involves 2 or 3 exchanges on the PowerBus that the model does not
1486  * support.
1487  *
1488  * The parameters represent what is sent on the PowerBus
1489  */
1490 static bool xive_presenter_notify(XiveFabric *xfb, uint8_t format,
1491                                   uint8_t nvt_blk, uint32_t nvt_idx,
1492                                   bool cam_ignore, uint8_t priority,
1493                                   uint32_t logic_serv)
1494 {
1495     XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xfb);
1496     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1497     int count;
1498 
1499     /*
1500      * Ask the machine to scan the interrupt controllers for a match
1501      */
1502     count = xfc->match_nvt(xfb, format, nvt_blk, nvt_idx, cam_ignore,
1503                            priority, logic_serv, &match);
1504     if (count < 0) {
1505         return false;
1506     }
1507 
1508     /* handle CPU exception delivery */
1509     if (count) {
1510         xive_tctx_ipb_update(match.tctx, match.ring, priority_to_ipb(priority));
1511     }
1512 
1513     return !!count;
1514 }
1515 
1516 /*
1517  * Notification using the END ESe/ESn bit (Event State Buffer for
1518  * escalation and notification). Provide further coalescing in the
1519  * Router.
1520  */
1521 static bool xive_router_end_es_notify(XiveRouter *xrtr, uint8_t end_blk,
1522                                       uint32_t end_idx, XiveEND *end,
1523                                       uint32_t end_esmask)
1524 {
1525     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1526     bool notify = xive_esb_trigger(&pq);
1527 
1528     if (pq != xive_get_field32(end_esmask, end->w1)) {
1529         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1530         xive_router_write_end(xrtr, end_blk, end_idx, end, 1);
1531     }
1532 
1533     /* ESe/n[Q]=1 : end of notification */
1534     return notify;
1535 }
1536 
1537 /*
1538  * An END trigger can come from an event trigger (IPI or HW) or from
1539  * another chip. We don't model the PowerBus but the END trigger
1540  * message has the same parameters than in the function below.
1541  */
1542 static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1543                                    uint32_t end_idx, uint32_t end_data)
1544 {
1545     XiveEND end;
1546     uint8_t priority;
1547     uint8_t format;
1548     uint8_t nvt_blk;
1549     uint32_t nvt_idx;
1550     XiveNVT nvt;
1551     bool found;
1552 
1553     /* END cache lookup */
1554     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1555         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1556                       end_idx);
1557         return;
1558     }
1559 
1560     if (!xive_end_is_valid(&end)) {
1561         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1562                       end_blk, end_idx);
1563         return;
1564     }
1565 
1566     if (xive_end_is_enqueue(&end)) {
1567         xive_end_enqueue(&end, end_data);
1568         /* Enqueuing event data modifies the EQ toggle and index */
1569         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1570     }
1571 
1572     /*
1573      * When the END is silent, we skip the notification part.
1574      */
1575     if (xive_end_is_silent_escalation(&end)) {
1576         goto do_escalation;
1577     }
1578 
1579     /*
1580      * The W7 format depends on the F bit in W6. It defines the type
1581      * of the notification :
1582      *
1583      *   F=0 : single or multiple NVT notification
1584      *   F=1 : User level Event-Based Branch (EBB) notification, no
1585      *         priority
1586      */
1587     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1588     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1589 
1590     /* The END is masked */
1591     if (format == 0 && priority == 0xff) {
1592         return;
1593     }
1594 
1595     /*
1596      * Check the END ESn (Event State Buffer for notification) for
1597      * even further coalescing in the Router
1598      */
1599     if (!xive_end_is_notify(&end)) {
1600         /* ESn[Q]=1 : end of notification */
1601         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1602                                        &end, END_W1_ESn)) {
1603             return;
1604         }
1605     }
1606 
1607     /*
1608      * Follows IVPE notification
1609      */
1610     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end.w6);
1611     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end.w6);
1612 
1613     /* NVT cache lookup */
1614     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1615         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1616                       nvt_blk, nvt_idx);
1617         return;
1618     }
1619 
1620     if (!xive_nvt_is_valid(&nvt)) {
1621         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1622                       nvt_blk, nvt_idx);
1623         return;
1624     }
1625 
1626     found = xive_presenter_notify(xrtr->xfb, format, nvt_blk, nvt_idx,
1627                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1628                           priority,
1629                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1630 
1631     /* TODO: Auto EOI. */
1632 
1633     if (found) {
1634         return;
1635     }
1636 
1637     /*
1638      * If no matching NVT is dispatched on a HW thread :
1639      * - specific VP: update the NVT structure if backlog is activated
1640      * - logical server : forward request to IVPE (not supported)
1641      */
1642     if (xive_end_is_backlog(&end)) {
1643         uint8_t ipb;
1644 
1645         if (format == 1) {
1646             qemu_log_mask(LOG_GUEST_ERROR,
1647                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1648                           end_blk, end_idx);
1649             return;
1650         }
1651         /*
1652          * Record the IPB in the associated NVT structure for later
1653          * use. The presenter will resend the interrupt when the vCPU
1654          * is dispatched again on a HW thread.
1655          */
1656         ipb = xive_get_field32(NVT_W4_IPB, nvt.w4) | priority_to_ipb(priority);
1657         nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, ipb);
1658         xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1659 
1660         /*
1661          * On HW, follows a "Broadcast Backlog" to IVPEs
1662          */
1663     }
1664 
1665 do_escalation:
1666     /*
1667      * If activated, escalate notification using the ESe PQ bits and
1668      * the EAS in w4-5
1669      */
1670     if (!xive_end_is_escalate(&end)) {
1671         return;
1672     }
1673 
1674     /*
1675      * Check the END ESe (Event State Buffer for escalation) for even
1676      * further coalescing in the Router
1677      */
1678     if (!xive_end_is_uncond_escalation(&end)) {
1679         /* ESe[Q]=1 : end of notification */
1680         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1681                                        &end, END_W1_ESe)) {
1682             return;
1683         }
1684     }
1685 
1686     /*
1687      * The END trigger becomes an Escalation trigger
1688      */
1689     xive_router_end_notify(xrtr,
1690                            xive_get_field32(END_W4_ESC_END_BLOCK, end.w4),
1691                            xive_get_field32(END_W4_ESC_END_INDEX, end.w4),
1692                            xive_get_field32(END_W5_ESC_END_DATA,  end.w5));
1693 }
1694 
1695 void xive_router_notify(XiveNotifier *xn, uint32_t lisn)
1696 {
1697     XiveRouter *xrtr = XIVE_ROUTER(xn);
1698     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1699     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1700     XiveEAS eas;
1701 
1702     /* EAS cache lookup */
1703     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1704         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1705         return;
1706     }
1707 
1708     /*
1709      * The IVRE checks the State Bit Cache at this point. We skip the
1710      * SBC lookup because the state bits of the sources are modeled
1711      * internally in QEMU.
1712      */
1713 
1714     if (!xive_eas_is_valid(&eas)) {
1715         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1716         return;
1717     }
1718 
1719     if (xive_eas_is_masked(&eas)) {
1720         /* Notification completed */
1721         return;
1722     }
1723 
1724     /*
1725      * The event trigger becomes an END trigger
1726      */
1727     xive_router_end_notify(xrtr,
1728                            xive_get_field64(EAS_END_BLOCK, eas.w),
1729                            xive_get_field64(EAS_END_INDEX, eas.w),
1730                            xive_get_field64(EAS_END_DATA,  eas.w));
1731 }
1732 
1733 static Property xive_router_properties[] = {
1734     DEFINE_PROP_LINK("xive-fabric", XiveRouter, xfb,
1735                      TYPE_XIVE_FABRIC, XiveFabric *),
1736     DEFINE_PROP_END_OF_LIST(),
1737 };
1738 
1739 static void xive_router_class_init(ObjectClass *klass, void *data)
1740 {
1741     DeviceClass *dc = DEVICE_CLASS(klass);
1742     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1743 
1744     dc->desc    = "XIVE Router Engine";
1745     device_class_set_props(dc, xive_router_properties);
1746     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1747     dc->realize = xive_router_realize;
1748     xnc->notify = xive_router_notify;
1749 }
1750 
1751 static const TypeInfo xive_router_info = {
1752     .name          = TYPE_XIVE_ROUTER,
1753     .parent        = TYPE_SYS_BUS_DEVICE,
1754     .abstract      = true,
1755     .instance_size = sizeof(XiveRouter),
1756     .class_size    = sizeof(XiveRouterClass),
1757     .class_init    = xive_router_class_init,
1758     .interfaces    = (InterfaceInfo[]) {
1759         { TYPE_XIVE_NOTIFIER },
1760         { TYPE_XIVE_PRESENTER },
1761         { }
1762     }
1763 };
1764 
1765 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1766 {
1767     if (!xive_eas_is_valid(eas)) {
1768         return;
1769     }
1770 
1771     monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1772                    lisn, xive_eas_is_masked(eas) ? "M" : " ",
1773                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1774                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1775                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1776 }
1777 
1778 /*
1779  * END ESB MMIO loads
1780  */
1781 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1782 {
1783     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1784     uint32_t offset = addr & 0xFFF;
1785     uint8_t end_blk;
1786     uint32_t end_idx;
1787     XiveEND end;
1788     uint32_t end_esmask;
1789     uint8_t pq;
1790     uint64_t ret = -1;
1791 
1792     /*
1793      * The block id should be deduced from the load address on the END
1794      * ESB MMIO but our model only supports a single block per XIVE chip.
1795      */
1796     end_blk = xive_router_get_block_id(xsrc->xrtr);
1797     end_idx = addr >> (xsrc->esb_shift + 1);
1798 
1799     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1800         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1801                       end_idx);
1802         return -1;
1803     }
1804 
1805     if (!xive_end_is_valid(&end)) {
1806         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1807                       end_blk, end_idx);
1808         return -1;
1809     }
1810 
1811     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1812     pq = xive_get_field32(end_esmask, end.w1);
1813 
1814     switch (offset) {
1815     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1816         ret = xive_esb_eoi(&pq);
1817 
1818         /* Forward the source event notification for routing ?? */
1819         break;
1820 
1821     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1822         ret = pq;
1823         break;
1824 
1825     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1826     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1827     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1828     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1829         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1830         break;
1831     default:
1832         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1833                       offset);
1834         return -1;
1835     }
1836 
1837     if (pq != xive_get_field32(end_esmask, end.w1)) {
1838         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1839         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1840     }
1841 
1842     return ret;
1843 }
1844 
1845 /*
1846  * END ESB MMIO stores are invalid
1847  */
1848 static void xive_end_source_write(void *opaque, hwaddr addr,
1849                                   uint64_t value, unsigned size)
1850 {
1851     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1852                   HWADDR_PRIx"\n", addr);
1853 }
1854 
1855 static const MemoryRegionOps xive_end_source_ops = {
1856     .read = xive_end_source_read,
1857     .write = xive_end_source_write,
1858     .endianness = DEVICE_BIG_ENDIAN,
1859     .valid = {
1860         .min_access_size = 8,
1861         .max_access_size = 8,
1862     },
1863     .impl = {
1864         .min_access_size = 8,
1865         .max_access_size = 8,
1866     },
1867 };
1868 
1869 static void xive_end_source_realize(DeviceState *dev, Error **errp)
1870 {
1871     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1872 
1873     assert(xsrc->xrtr);
1874 
1875     if (!xsrc->nr_ends) {
1876         error_setg(errp, "Number of interrupt needs to be greater than 0");
1877         return;
1878     }
1879 
1880     if (xsrc->esb_shift != XIVE_ESB_4K &&
1881         xsrc->esb_shift != XIVE_ESB_64K) {
1882         error_setg(errp, "Invalid ESB shift setting");
1883         return;
1884     }
1885 
1886     /*
1887      * Each END is assigned an even/odd pair of MMIO pages, the even page
1888      * manages the ESn field while the odd page manages the ESe field.
1889      */
1890     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1891                           &xive_end_source_ops, xsrc, "xive.end",
1892                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1893 }
1894 
1895 static Property xive_end_source_properties[] = {
1896     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
1897     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
1898     DEFINE_PROP_LINK("xive", XiveENDSource, xrtr, TYPE_XIVE_ROUTER,
1899                      XiveRouter *),
1900     DEFINE_PROP_END_OF_LIST(),
1901 };
1902 
1903 static void xive_end_source_class_init(ObjectClass *klass, void *data)
1904 {
1905     DeviceClass *dc = DEVICE_CLASS(klass);
1906 
1907     dc->desc    = "XIVE END Source";
1908     device_class_set_props(dc, xive_end_source_properties);
1909     dc->realize = xive_end_source_realize;
1910     /*
1911      * Reason: part of XIVE interrupt controller, needs to be wired up,
1912      * e.g. by spapr_xive_instance_init().
1913      */
1914     dc->user_creatable = false;
1915 }
1916 
1917 static const TypeInfo xive_end_source_info = {
1918     .name          = TYPE_XIVE_END_SOURCE,
1919     .parent        = TYPE_DEVICE,
1920     .instance_size = sizeof(XiveENDSource),
1921     .class_init    = xive_end_source_class_init,
1922 };
1923 
1924 /*
1925  * XIVE Notifier
1926  */
1927 static const TypeInfo xive_notifier_info = {
1928     .name = TYPE_XIVE_NOTIFIER,
1929     .parent = TYPE_INTERFACE,
1930     .class_size = sizeof(XiveNotifierClass),
1931 };
1932 
1933 /*
1934  * XIVE Presenter
1935  */
1936 static const TypeInfo xive_presenter_info = {
1937     .name = TYPE_XIVE_PRESENTER,
1938     .parent = TYPE_INTERFACE,
1939     .class_size = sizeof(XivePresenterClass),
1940 };
1941 
1942 /*
1943  * XIVE Fabric
1944  */
1945 static const TypeInfo xive_fabric_info = {
1946     .name = TYPE_XIVE_FABRIC,
1947     .parent = TYPE_INTERFACE,
1948     .class_size = sizeof(XiveFabricClass),
1949 };
1950 
1951 static void xive_register_types(void)
1952 {
1953     type_register_static(&xive_fabric_info);
1954     type_register_static(&xive_source_info);
1955     type_register_static(&xive_notifier_info);
1956     type_register_static(&xive_presenter_info);
1957     type_register_static(&xive_router_info);
1958     type_register_static(&xive_end_source_info);
1959     type_register_static(&xive_tctx_info);
1960 }
1961 
1962 type_init(xive_register_types)
1963