xref: /qemu/hw/intc/xive.c (revision b355f08a)
1 /*
2  * QEMU PowerPC XIVE interrupt controller model
3  *
4  * Copyright (c) 2017-2018, IBM Corporation.
5  *
6  * This code is licensed under the GPL version 2 or later. See the
7  * COPYING file in the top-level directory.
8  */
9 
10 #include "qemu/osdep.h"
11 #include "qemu/log.h"
12 #include "qemu/module.h"
13 #include "qapi/error.h"
14 #include "target/ppc/cpu.h"
15 #include "sysemu/cpus.h"
16 #include "sysemu/dma.h"
17 #include "sysemu/reset.h"
18 #include "hw/qdev-properties.h"
19 #include "migration/vmstate.h"
20 #include "monitor/monitor.h"
21 #include "hw/irq.h"
22 #include "hw/ppc/xive.h"
23 #include "hw/ppc/xive_regs.h"
24 #include "trace.h"
25 
26 /*
27  * XIVE Thread Interrupt Management context
28  */
29 
30 /*
31  * Convert a priority number to an Interrupt Pending Buffer (IPB)
32  * register, which indicates a pending interrupt at the priority
33  * corresponding to the bit number
34  */
35 static uint8_t priority_to_ipb(uint8_t priority)
36 {
37     return priority > XIVE_PRIORITY_MAX ?
38         0 : 1 << (XIVE_PRIORITY_MAX - priority);
39 }
40 
41 /*
42  * Convert an Interrupt Pending Buffer (IPB) register to a Pending
43  * Interrupt Priority Register (PIPR), which contains the priority of
44  * the most favored pending notification.
45  */
46 static uint8_t ipb_to_pipr(uint8_t ibp)
47 {
48     return ibp ? clz32((uint32_t)ibp << 24) : 0xff;
49 }
50 
51 static uint8_t exception_mask(uint8_t ring)
52 {
53     switch (ring) {
54     case TM_QW1_OS:
55         return TM_QW1_NSR_EO;
56     case TM_QW3_HV_PHYS:
57         return TM_QW3_NSR_HE;
58     default:
59         g_assert_not_reached();
60     }
61 }
62 
63 static qemu_irq xive_tctx_output(XiveTCTX *tctx, uint8_t ring)
64 {
65         switch (ring) {
66         case TM_QW0_USER:
67                 return 0; /* Not supported */
68         case TM_QW1_OS:
69                 return tctx->os_output;
70         case TM_QW2_HV_POOL:
71         case TM_QW3_HV_PHYS:
72                 return tctx->hv_output;
73         default:
74                 return 0;
75         }
76 }
77 
78 static uint64_t xive_tctx_accept(XiveTCTX *tctx, uint8_t ring)
79 {
80     uint8_t *regs = &tctx->regs[ring];
81     uint8_t nsr = regs[TM_NSR];
82     uint8_t mask = exception_mask(ring);
83 
84     qemu_irq_lower(xive_tctx_output(tctx, ring));
85 
86     if (regs[TM_NSR] & mask) {
87         uint8_t cppr = regs[TM_PIPR];
88 
89         regs[TM_CPPR] = cppr;
90 
91         /* Reset the pending buffer bit */
92         regs[TM_IPB] &= ~priority_to_ipb(cppr);
93         regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
94 
95         /* Drop Exception bit */
96         regs[TM_NSR] &= ~mask;
97 
98         trace_xive_tctx_accept(tctx->cs->cpu_index, ring,
99                                regs[TM_IPB], regs[TM_PIPR],
100                                regs[TM_CPPR], regs[TM_NSR]);
101     }
102 
103     return (nsr << 8) | regs[TM_CPPR];
104 }
105 
106 static void xive_tctx_notify(XiveTCTX *tctx, uint8_t ring)
107 {
108     uint8_t *regs = &tctx->regs[ring];
109 
110     if (regs[TM_PIPR] < regs[TM_CPPR]) {
111         switch (ring) {
112         case TM_QW1_OS:
113             regs[TM_NSR] |= TM_QW1_NSR_EO;
114             break;
115         case TM_QW3_HV_PHYS:
116             regs[TM_NSR] |= (TM_QW3_NSR_HE_PHYS << 6);
117             break;
118         default:
119             g_assert_not_reached();
120         }
121         trace_xive_tctx_notify(tctx->cs->cpu_index, ring,
122                                regs[TM_IPB], regs[TM_PIPR],
123                                regs[TM_CPPR], regs[TM_NSR]);
124         qemu_irq_raise(xive_tctx_output(tctx, ring));
125     }
126 }
127 
128 static void xive_tctx_set_cppr(XiveTCTX *tctx, uint8_t ring, uint8_t cppr)
129 {
130     uint8_t *regs = &tctx->regs[ring];
131 
132     trace_xive_tctx_set_cppr(tctx->cs->cpu_index, ring,
133                              regs[TM_IPB], regs[TM_PIPR],
134                              cppr, regs[TM_NSR]);
135 
136     if (cppr > XIVE_PRIORITY_MAX) {
137         cppr = 0xff;
138     }
139 
140     tctx->regs[ring + TM_CPPR] = cppr;
141 
142     /* CPPR has changed, check if we need to raise a pending exception */
143     xive_tctx_notify(tctx, ring);
144 }
145 
146 void xive_tctx_ipb_update(XiveTCTX *tctx, uint8_t ring, uint8_t ipb)
147 {
148     uint8_t *regs = &tctx->regs[ring];
149 
150     regs[TM_IPB] |= ipb;
151     regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
152     xive_tctx_notify(tctx, ring);
153 }
154 
155 static inline uint32_t xive_tctx_word2(uint8_t *ring)
156 {
157     return *((uint32_t *) &ring[TM_WORD2]);
158 }
159 
160 /*
161  * XIVE Thread Interrupt Management Area (TIMA)
162  */
163 
164 static void xive_tm_set_hv_cppr(XivePresenter *xptr, XiveTCTX *tctx,
165                                 hwaddr offset, uint64_t value, unsigned size)
166 {
167     xive_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
168 }
169 
170 static uint64_t xive_tm_ack_hv_reg(XivePresenter *xptr, XiveTCTX *tctx,
171                                    hwaddr offset, unsigned size)
172 {
173     return xive_tctx_accept(tctx, TM_QW3_HV_PHYS);
174 }
175 
176 static uint64_t xive_tm_pull_pool_ctx(XivePresenter *xptr, XiveTCTX *tctx,
177                                       hwaddr offset, unsigned size)
178 {
179     uint32_t qw2w2_prev = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
180     uint32_t qw2w2;
181 
182     qw2w2 = xive_set_field32(TM_QW2W2_VP, qw2w2_prev, 0);
183     memcpy(&tctx->regs[TM_QW2_HV_POOL + TM_WORD2], &qw2w2, 4);
184     return qw2w2;
185 }
186 
187 static void xive_tm_vt_push(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
188                             uint64_t value, unsigned size)
189 {
190     tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] = value & 0xff;
191 }
192 
193 static uint64_t xive_tm_vt_poll(XivePresenter *xptr, XiveTCTX *tctx,
194                                 hwaddr offset, unsigned size)
195 {
196     return tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] & 0xff;
197 }
198 
199 /*
200  * Define an access map for each page of the TIMA that we will use in
201  * the memory region ops to filter values when doing loads and stores
202  * of raw registers values
203  *
204  * Registers accessibility bits :
205  *
206  *    0x0 - no access
207  *    0x1 - write only
208  *    0x2 - read only
209  *    0x3 - read/write
210  */
211 
212 static const uint8_t xive_tm_hw_view[] = {
213     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
214     3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
215     0, 0, 3, 3,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
216     3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   3, 3, 3, 0, /* QW-3 PHYS */
217 };
218 
219 static const uint8_t xive_tm_hv_view[] = {
220     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
221     3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
222     0, 0, 3, 3,   0, 0, 0, 0,   0, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
223     3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   0, 0, 0, 0, /* QW-3 PHYS */
224 };
225 
226 static const uint8_t xive_tm_os_view[] = {
227     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
228     2, 3, 2, 2,   2, 2, 0, 2,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
229     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
230     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
231 };
232 
233 static const uint8_t xive_tm_user_view[] = {
234     3, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-0 User */
235     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
236     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
237     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
238 };
239 
240 /*
241  * Overall TIMA access map for the thread interrupt management context
242  * registers
243  */
244 static const uint8_t *xive_tm_views[] = {
245     [XIVE_TM_HW_PAGE]   = xive_tm_hw_view,
246     [XIVE_TM_HV_PAGE]   = xive_tm_hv_view,
247     [XIVE_TM_OS_PAGE]   = xive_tm_os_view,
248     [XIVE_TM_USER_PAGE] = xive_tm_user_view,
249 };
250 
251 /*
252  * Computes a register access mask for a given offset in the TIMA
253  */
254 static uint64_t xive_tm_mask(hwaddr offset, unsigned size, bool write)
255 {
256     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
257     uint8_t reg_offset = offset & 0x3F;
258     uint8_t reg_mask = write ? 0x1 : 0x2;
259     uint64_t mask = 0x0;
260     int i;
261 
262     for (i = 0; i < size; i++) {
263         if (xive_tm_views[page_offset][reg_offset + i] & reg_mask) {
264             mask |= (uint64_t) 0xff << (8 * (size - i - 1));
265         }
266     }
267 
268     return mask;
269 }
270 
271 static void xive_tm_raw_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
272                               unsigned size)
273 {
274     uint8_t ring_offset = offset & 0x30;
275     uint8_t reg_offset = offset & 0x3F;
276     uint64_t mask = xive_tm_mask(offset, size, true);
277     int i;
278 
279     /*
280      * Only 4 or 8 bytes stores are allowed and the User ring is
281      * excluded
282      */
283     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
284         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA @%"
285                       HWADDR_PRIx"\n", offset);
286         return;
287     }
288 
289     /*
290      * Use the register offset for the raw values and filter out
291      * reserved values
292      */
293     for (i = 0; i < size; i++) {
294         uint8_t byte_mask = (mask >> (8 * (size - i - 1)));
295         if (byte_mask) {
296             tctx->regs[reg_offset + i] = (value >> (8 * (size - i - 1))) &
297                 byte_mask;
298         }
299     }
300 }
301 
302 static uint64_t xive_tm_raw_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
303 {
304     uint8_t ring_offset = offset & 0x30;
305     uint8_t reg_offset = offset & 0x3F;
306     uint64_t mask = xive_tm_mask(offset, size, false);
307     uint64_t ret;
308     int i;
309 
310     /*
311      * Only 4 or 8 bytes loads are allowed and the User ring is
312      * excluded
313      */
314     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
315         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access at TIMA @%"
316                       HWADDR_PRIx"\n", offset);
317         return -1;
318     }
319 
320     /* Use the register offset for the raw values */
321     ret = 0;
322     for (i = 0; i < size; i++) {
323         ret |= (uint64_t) tctx->regs[reg_offset + i] << (8 * (size - i - 1));
324     }
325 
326     /* filter out reserved values */
327     return ret & mask;
328 }
329 
330 /*
331  * The TM context is mapped twice within each page. Stores and loads
332  * to the first mapping below 2K write and read the specified values
333  * without modification. The second mapping above 2K performs specific
334  * state changes (side effects) in addition to setting/returning the
335  * interrupt management area context of the processor thread.
336  */
337 static uint64_t xive_tm_ack_os_reg(XivePresenter *xptr, XiveTCTX *tctx,
338                                    hwaddr offset, unsigned size)
339 {
340     return xive_tctx_accept(tctx, TM_QW1_OS);
341 }
342 
343 static void xive_tm_set_os_cppr(XivePresenter *xptr, XiveTCTX *tctx,
344                                 hwaddr offset, uint64_t value, unsigned size)
345 {
346     xive_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
347 }
348 
349 /*
350  * Adjust the IPB to allow a CPU to process event queues of other
351  * priorities during one physical interrupt cycle.
352  */
353 static void xive_tm_set_os_pending(XivePresenter *xptr, XiveTCTX *tctx,
354                                    hwaddr offset, uint64_t value, unsigned size)
355 {
356     xive_tctx_ipb_update(tctx, TM_QW1_OS, priority_to_ipb(value & 0xff));
357 }
358 
359 static void xive_os_cam_decode(uint32_t cam, uint8_t *nvt_blk,
360                                uint32_t *nvt_idx, bool *vo)
361 {
362     if (nvt_blk) {
363         *nvt_blk = xive_nvt_blk(cam);
364     }
365     if (nvt_idx) {
366         *nvt_idx = xive_nvt_idx(cam);
367     }
368     if (vo) {
369         *vo = !!(cam & TM_QW1W2_VO);
370     }
371 }
372 
373 static uint32_t xive_tctx_get_os_cam(XiveTCTX *tctx, uint8_t *nvt_blk,
374                                      uint32_t *nvt_idx, bool *vo)
375 {
376     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
377     uint32_t cam = be32_to_cpu(qw1w2);
378 
379     xive_os_cam_decode(cam, nvt_blk, nvt_idx, vo);
380     return qw1w2;
381 }
382 
383 static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t qw1w2)
384 {
385     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
386 }
387 
388 static uint64_t xive_tm_pull_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
389                                     hwaddr offset, unsigned size)
390 {
391     uint32_t qw1w2;
392     uint32_t qw1w2_new;
393     uint8_t nvt_blk;
394     uint32_t nvt_idx;
395     bool vo;
396 
397     qw1w2 = xive_tctx_get_os_cam(tctx, &nvt_blk, &nvt_idx, &vo);
398 
399     if (!vo) {
400         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: pulling invalid NVT %x/%x !?\n",
401                       nvt_blk, nvt_idx);
402     }
403 
404     /* Invalidate CAM line */
405     qw1w2_new = xive_set_field32(TM_QW1W2_VO, qw1w2, 0);
406     xive_tctx_set_os_cam(tctx, qw1w2_new);
407     return qw1w2;
408 }
409 
410 static void xive_tctx_need_resend(XiveRouter *xrtr, XiveTCTX *tctx,
411                                   uint8_t nvt_blk, uint32_t nvt_idx)
412 {
413     XiveNVT nvt;
414     uint8_t ipb;
415 
416     /*
417      * Grab the associated NVT to pull the pending bits, and merge
418      * them with the IPB of the thread interrupt context registers
419      */
420     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
421         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVT %x/%x\n",
422                           nvt_blk, nvt_idx);
423         return;
424     }
425 
426     ipb = xive_get_field32(NVT_W4_IPB, nvt.w4);
427 
428     if (ipb) {
429         /* Reset the NVT value */
430         nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, 0);
431         xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
432 
433         /* Merge in current context */
434         xive_tctx_ipb_update(tctx, TM_QW1_OS, ipb);
435     }
436 }
437 
438 /*
439  * Updating the OS CAM line can trigger a resend of interrupt
440  */
441 static void xive_tm_push_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
442                                 hwaddr offset, uint64_t value, unsigned size)
443 {
444     uint32_t cam = value;
445     uint32_t qw1w2 = cpu_to_be32(cam);
446     uint8_t nvt_blk;
447     uint32_t nvt_idx;
448     bool vo;
449 
450     xive_os_cam_decode(cam, &nvt_blk, &nvt_idx, &vo);
451 
452     /* First update the registers */
453     xive_tctx_set_os_cam(tctx, qw1w2);
454 
455     /* Check the interrupt pending bits */
456     if (vo) {
457         xive_tctx_need_resend(XIVE_ROUTER(xptr), tctx, nvt_blk, nvt_idx);
458     }
459 }
460 
461 /*
462  * Define a mapping of "special" operations depending on the TIMA page
463  * offset and the size of the operation.
464  */
465 typedef struct XiveTmOp {
466     uint8_t  page_offset;
467     uint32_t op_offset;
468     unsigned size;
469     void     (*write_handler)(XivePresenter *xptr, XiveTCTX *tctx,
470                               hwaddr offset,
471                               uint64_t value, unsigned size);
472     uint64_t (*read_handler)(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
473                              unsigned size);
474 } XiveTmOp;
475 
476 static const XiveTmOp xive_tm_operations[] = {
477     /*
478      * MMIOs below 2K : raw values and special operations without side
479      * effects
480      */
481     { XIVE_TM_OS_PAGE, TM_QW1_OS + TM_CPPR,   1, xive_tm_set_os_cppr, NULL },
482     { XIVE_TM_HV_PAGE, TM_QW1_OS + TM_WORD2,     4, xive_tm_push_os_ctx, NULL },
483     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_CPPR, 1, xive_tm_set_hv_cppr, NULL },
484     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, xive_tm_vt_push, NULL },
485     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, NULL, xive_tm_vt_poll },
486 
487     /* MMIOs above 2K : special operations with side effects */
488     { XIVE_TM_OS_PAGE, TM_SPC_ACK_OS_REG,     2, NULL, xive_tm_ack_os_reg },
489     { XIVE_TM_OS_PAGE, TM_SPC_SET_OS_PENDING, 1, xive_tm_set_os_pending, NULL },
490     { XIVE_TM_HV_PAGE, TM_SPC_PULL_OS_CTX,    4, NULL, xive_tm_pull_os_ctx },
491     { XIVE_TM_HV_PAGE, TM_SPC_PULL_OS_CTX,    8, NULL, xive_tm_pull_os_ctx },
492     { XIVE_TM_HV_PAGE, TM_SPC_ACK_HV_REG,     2, NULL, xive_tm_ack_hv_reg },
493     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  4, NULL, xive_tm_pull_pool_ctx },
494     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  8, NULL, xive_tm_pull_pool_ctx },
495 };
496 
497 static const XiveTmOp *xive_tm_find_op(hwaddr offset, unsigned size, bool write)
498 {
499     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
500     uint32_t op_offset = offset & 0xFFF;
501     int i;
502 
503     for (i = 0; i < ARRAY_SIZE(xive_tm_operations); i++) {
504         const XiveTmOp *xto = &xive_tm_operations[i];
505 
506         /* Accesses done from a more privileged TIMA page is allowed */
507         if (xto->page_offset >= page_offset &&
508             xto->op_offset == op_offset &&
509             xto->size == size &&
510             ((write && xto->write_handler) || (!write && xto->read_handler))) {
511             return xto;
512         }
513     }
514     return NULL;
515 }
516 
517 /*
518  * TIMA MMIO handlers
519  */
520 void xive_tctx_tm_write(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
521                         uint64_t value, unsigned size)
522 {
523     const XiveTmOp *xto;
524 
525     trace_xive_tctx_tm_write(offset, size, value);
526 
527     /*
528      * TODO: check V bit in Q[0-3]W2
529      */
530 
531     /*
532      * First, check for special operations in the 2K region
533      */
534     if (offset & 0x800) {
535         xto = xive_tm_find_op(offset, size, true);
536         if (!xto) {
537             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA "
538                           "@%"HWADDR_PRIx"\n", offset);
539         } else {
540             xto->write_handler(xptr, tctx, offset, value, size);
541         }
542         return;
543     }
544 
545     /*
546      * Then, for special operations in the region below 2K.
547      */
548     xto = xive_tm_find_op(offset, size, true);
549     if (xto) {
550         xto->write_handler(xptr, tctx, offset, value, size);
551         return;
552     }
553 
554     /*
555      * Finish with raw access to the register values
556      */
557     xive_tm_raw_write(tctx, offset, value, size);
558 }
559 
560 uint64_t xive_tctx_tm_read(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
561                            unsigned size)
562 {
563     const XiveTmOp *xto;
564     uint64_t ret;
565 
566     /*
567      * TODO: check V bit in Q[0-3]W2
568      */
569 
570     /*
571      * First, check for special operations in the 2K region
572      */
573     if (offset & 0x800) {
574         xto = xive_tm_find_op(offset, size, false);
575         if (!xto) {
576             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access to TIMA"
577                           "@%"HWADDR_PRIx"\n", offset);
578             return -1;
579         }
580         ret = xto->read_handler(xptr, tctx, offset, size);
581         goto out;
582     }
583 
584     /*
585      * Then, for special operations in the region below 2K.
586      */
587     xto = xive_tm_find_op(offset, size, false);
588     if (xto) {
589         ret = xto->read_handler(xptr, tctx, offset, size);
590         goto out;
591     }
592 
593     /*
594      * Finish with raw access to the register values
595      */
596     ret = xive_tm_raw_read(tctx, offset, size);
597 out:
598     trace_xive_tctx_tm_read(offset, size, ret);
599     return ret;
600 }
601 
602 static char *xive_tctx_ring_print(uint8_t *ring)
603 {
604     uint32_t w2 = xive_tctx_word2(ring);
605 
606     return g_strdup_printf("%02x   %02x  %02x    %02x   %02x  "
607                    "%02x  %02x   %02x  %08x",
608                    ring[TM_NSR], ring[TM_CPPR], ring[TM_IPB], ring[TM_LSMFB],
609                    ring[TM_ACK_CNT], ring[TM_INC], ring[TM_AGE], ring[TM_PIPR],
610                    be32_to_cpu(w2));
611 }
612 
613 static const char * const xive_tctx_ring_names[] = {
614     "USER", "OS", "POOL", "PHYS",
615 };
616 
617 /*
618  * kvm_irqchip_in_kernel() will cause the compiler to turn this
619  * info a nop if CONFIG_KVM isn't defined.
620  */
621 #define xive_in_kernel(xptr)                                            \
622     (kvm_irqchip_in_kernel() &&                                         \
623      ({                                                                 \
624          XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);      \
625          xpc->in_kernel ? xpc->in_kernel(xptr) : false;                 \
626      }))
627 
628 void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
629 {
630     int cpu_index;
631     int i;
632 
633     /* Skip partially initialized vCPUs. This can happen on sPAPR when vCPUs
634      * are hot plugged or unplugged.
635      */
636     if (!tctx) {
637         return;
638     }
639 
640     cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
641 
642     if (xive_in_kernel(tctx->xptr)) {
643         Error *local_err = NULL;
644 
645         kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
646         if (local_err) {
647             error_report_err(local_err);
648             return;
649         }
650     }
651 
652     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
653                    "  W2\n", cpu_index);
654 
655     for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
656         char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
657         monitor_printf(mon, "CPU[%04x]: %4s    %s\n", cpu_index,
658                        xive_tctx_ring_names[i], s);
659         g_free(s);
660     }
661 }
662 
663 void xive_tctx_reset(XiveTCTX *tctx)
664 {
665     memset(tctx->regs, 0, sizeof(tctx->regs));
666 
667     /* Set some defaults */
668     tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
669     tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
670     tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
671 
672     /*
673      * Initialize PIPR to 0xFF to avoid phantom interrupts when the
674      * CPPR is first set.
675      */
676     tctx->regs[TM_QW1_OS + TM_PIPR] =
677         ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
678     tctx->regs[TM_QW3_HV_PHYS + TM_PIPR] =
679         ipb_to_pipr(tctx->regs[TM_QW3_HV_PHYS + TM_IPB]);
680 }
681 
682 static void xive_tctx_realize(DeviceState *dev, Error **errp)
683 {
684     XiveTCTX *tctx = XIVE_TCTX(dev);
685     PowerPCCPU *cpu;
686     CPUPPCState *env;
687 
688     assert(tctx->cs);
689     assert(tctx->xptr);
690 
691     cpu = POWERPC_CPU(tctx->cs);
692     env = &cpu->env;
693     switch (PPC_INPUT(env)) {
694     case PPC_FLAGS_INPUT_POWER9:
695         tctx->hv_output = env->irq_inputs[POWER9_INPUT_HINT];
696         tctx->os_output = env->irq_inputs[POWER9_INPUT_INT];
697         break;
698 
699     default:
700         error_setg(errp, "XIVE interrupt controller does not support "
701                    "this CPU bus model");
702         return;
703     }
704 
705     /* Connect the presenter to the VCPU (required for CPU hotplug) */
706     if (xive_in_kernel(tctx->xptr)) {
707         if (kvmppc_xive_cpu_connect(tctx, errp) < 0) {
708             return;
709         }
710     }
711 }
712 
713 static int vmstate_xive_tctx_pre_save(void *opaque)
714 {
715     XiveTCTX *tctx = XIVE_TCTX(opaque);
716     Error *local_err = NULL;
717     int ret;
718 
719     if (xive_in_kernel(tctx->xptr)) {
720         ret = kvmppc_xive_cpu_get_state(tctx, &local_err);
721         if (ret < 0) {
722             error_report_err(local_err);
723             return ret;
724         }
725     }
726 
727     return 0;
728 }
729 
730 static int vmstate_xive_tctx_post_load(void *opaque, int version_id)
731 {
732     XiveTCTX *tctx = XIVE_TCTX(opaque);
733     Error *local_err = NULL;
734     int ret;
735 
736     if (xive_in_kernel(tctx->xptr)) {
737         /*
738          * Required for hotplugged CPU, for which the state comes
739          * after all states of the machine.
740          */
741         ret = kvmppc_xive_cpu_set_state(tctx, &local_err);
742         if (ret < 0) {
743             error_report_err(local_err);
744             return ret;
745         }
746     }
747 
748     return 0;
749 }
750 
751 static const VMStateDescription vmstate_xive_tctx = {
752     .name = TYPE_XIVE_TCTX,
753     .version_id = 1,
754     .minimum_version_id = 1,
755     .pre_save = vmstate_xive_tctx_pre_save,
756     .post_load = vmstate_xive_tctx_post_load,
757     .fields = (VMStateField[]) {
758         VMSTATE_BUFFER(regs, XiveTCTX),
759         VMSTATE_END_OF_LIST()
760     },
761 };
762 
763 static Property xive_tctx_properties[] = {
764     DEFINE_PROP_LINK("cpu", XiveTCTX, cs, TYPE_CPU, CPUState *),
765     DEFINE_PROP_LINK("presenter", XiveTCTX, xptr, TYPE_XIVE_PRESENTER,
766                      XivePresenter *),
767     DEFINE_PROP_END_OF_LIST(),
768 };
769 
770 static void xive_tctx_class_init(ObjectClass *klass, void *data)
771 {
772     DeviceClass *dc = DEVICE_CLASS(klass);
773 
774     dc->desc = "XIVE Interrupt Thread Context";
775     dc->realize = xive_tctx_realize;
776     dc->vmsd = &vmstate_xive_tctx;
777     device_class_set_props(dc, xive_tctx_properties);
778     /*
779      * Reason: part of XIVE interrupt controller, needs to be wired up
780      * by xive_tctx_create().
781      */
782     dc->user_creatable = false;
783 }
784 
785 static const TypeInfo xive_tctx_info = {
786     .name          = TYPE_XIVE_TCTX,
787     .parent        = TYPE_DEVICE,
788     .instance_size = sizeof(XiveTCTX),
789     .class_init    = xive_tctx_class_init,
790 };
791 
792 Object *xive_tctx_create(Object *cpu, XivePresenter *xptr, Error **errp)
793 {
794     Object *obj;
795 
796     obj = object_new(TYPE_XIVE_TCTX);
797     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj);
798     object_unref(obj);
799     object_property_set_link(obj, "cpu", cpu, &error_abort);
800     object_property_set_link(obj, "presenter", OBJECT(xptr), &error_abort);
801     if (!qdev_realize(DEVICE(obj), NULL, errp)) {
802         object_unparent(obj);
803         return NULL;
804     }
805     return obj;
806 }
807 
808 void xive_tctx_destroy(XiveTCTX *tctx)
809 {
810     Object *obj = OBJECT(tctx);
811 
812     object_unparent(obj);
813 }
814 
815 /*
816  * XIVE ESB helpers
817  */
818 
819 uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
820 {
821     uint8_t old_pq = *pq & 0x3;
822 
823     *pq &= ~0x3;
824     *pq |= value & 0x3;
825 
826     return old_pq;
827 }
828 
829 bool xive_esb_trigger(uint8_t *pq)
830 {
831     uint8_t old_pq = *pq & 0x3;
832 
833     switch (old_pq) {
834     case XIVE_ESB_RESET:
835         xive_esb_set(pq, XIVE_ESB_PENDING);
836         return true;
837     case XIVE_ESB_PENDING:
838     case XIVE_ESB_QUEUED:
839         xive_esb_set(pq, XIVE_ESB_QUEUED);
840         return false;
841     case XIVE_ESB_OFF:
842         xive_esb_set(pq, XIVE_ESB_OFF);
843         return false;
844     default:
845          g_assert_not_reached();
846     }
847 }
848 
849 bool xive_esb_eoi(uint8_t *pq)
850 {
851     uint8_t old_pq = *pq & 0x3;
852 
853     switch (old_pq) {
854     case XIVE_ESB_RESET:
855     case XIVE_ESB_PENDING:
856         xive_esb_set(pq, XIVE_ESB_RESET);
857         return false;
858     case XIVE_ESB_QUEUED:
859         xive_esb_set(pq, XIVE_ESB_PENDING);
860         return true;
861     case XIVE_ESB_OFF:
862         xive_esb_set(pq, XIVE_ESB_OFF);
863         return false;
864     default:
865          g_assert_not_reached();
866     }
867 }
868 
869 /*
870  * XIVE Interrupt Source (or IVSE)
871  */
872 
873 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
874 {
875     assert(srcno < xsrc->nr_irqs);
876 
877     return xsrc->status[srcno] & 0x3;
878 }
879 
880 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
881 {
882     assert(srcno < xsrc->nr_irqs);
883 
884     return xive_esb_set(&xsrc->status[srcno], pq);
885 }
886 
887 /*
888  * Returns whether the event notification should be forwarded.
889  */
890 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
891 {
892     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
893 
894     xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
895 
896     switch (old_pq) {
897     case XIVE_ESB_RESET:
898         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
899         return true;
900     default:
901         return false;
902     }
903 }
904 
905 /*
906  * Returns whether the event notification should be forwarded.
907  */
908 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
909 {
910     bool ret;
911 
912     assert(srcno < xsrc->nr_irqs);
913 
914     ret = xive_esb_trigger(&xsrc->status[srcno]);
915 
916     if (xive_source_irq_is_lsi(xsrc, srcno) &&
917         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
918         qemu_log_mask(LOG_GUEST_ERROR,
919                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
920     }
921 
922     return ret;
923 }
924 
925 /*
926  * Returns whether the event notification should be forwarded.
927  */
928 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
929 {
930     bool ret;
931 
932     assert(srcno < xsrc->nr_irqs);
933 
934     ret = xive_esb_eoi(&xsrc->status[srcno]);
935 
936     /*
937      * LSI sources do not set the Q bit but they can still be
938      * asserted, in which case we should forward a new event
939      * notification
940      */
941     if (xive_source_irq_is_lsi(xsrc, srcno) &&
942         xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
943         ret = xive_source_lsi_trigger(xsrc, srcno);
944     }
945 
946     return ret;
947 }
948 
949 /*
950  * Forward the source event notification to the Router
951  */
952 static void xive_source_notify(XiveSource *xsrc, int srcno)
953 {
954     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
955 
956     if (xnc->notify) {
957         xnc->notify(xsrc->xive, srcno);
958     }
959 }
960 
961 /*
962  * In a two pages ESB MMIO setting, even page is the trigger page, odd
963  * page is for management
964  */
965 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
966 {
967     return !((addr >> shift) & 1);
968 }
969 
970 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
971 {
972     return xive_source_esb_has_2page(xsrc) &&
973         addr_is_even(addr, xsrc->esb_shift - 1);
974 }
975 
976 /*
977  * ESB MMIO loads
978  *                      Trigger page    Management/EOI page
979  *
980  * ESB MMIO setting     2 pages         1 or 2 pages
981  *
982  * 0x000 .. 0x3FF       -1              EOI and return 0|1
983  * 0x400 .. 0x7FF       -1              EOI and return 0|1
984  * 0x800 .. 0xBFF       -1              return PQ
985  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
986  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
987  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
988  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
989  */
990 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
991 {
992     XiveSource *xsrc = XIVE_SOURCE(opaque);
993     uint32_t offset = addr & 0xFFF;
994     uint32_t srcno = addr >> xsrc->esb_shift;
995     uint64_t ret = -1;
996 
997     /* In a two pages ESB MMIO setting, trigger page should not be read */
998     if (xive_source_is_trigger_page(xsrc, addr)) {
999         qemu_log_mask(LOG_GUEST_ERROR,
1000                       "XIVE: invalid load on IRQ %d trigger page at "
1001                       "0x%"HWADDR_PRIx"\n", srcno, addr);
1002         return -1;
1003     }
1004 
1005     switch (offset) {
1006     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1007         ret = xive_source_esb_eoi(xsrc, srcno);
1008 
1009         /* Forward the source event notification for routing */
1010         if (ret) {
1011             xive_source_notify(xsrc, srcno);
1012         }
1013         break;
1014 
1015     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1016         ret = xive_source_esb_get(xsrc, srcno);
1017         break;
1018 
1019     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1020     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1021     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1022     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1023         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1024         break;
1025     default:
1026         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
1027                       offset);
1028     }
1029 
1030     trace_xive_source_esb_read(addr, srcno, ret);
1031 
1032     return ret;
1033 }
1034 
1035 /*
1036  * ESB MMIO stores
1037  *                      Trigger page    Management/EOI page
1038  *
1039  * ESB MMIO setting     2 pages         1 or 2 pages
1040  *
1041  * 0x000 .. 0x3FF       Trigger         Trigger
1042  * 0x400 .. 0x7FF       Trigger         EOI
1043  * 0x800 .. 0xBFF       Trigger         undefined
1044  * 0xC00 .. 0xCFF       Trigger         PQ=00
1045  * 0xD00 .. 0xDFF       Trigger         PQ=01
1046  * 0xE00 .. 0xDFF       Trigger         PQ=10
1047  * 0xF00 .. 0xDFF       Trigger         PQ=11
1048  */
1049 static void xive_source_esb_write(void *opaque, hwaddr addr,
1050                                   uint64_t value, unsigned size)
1051 {
1052     XiveSource *xsrc = XIVE_SOURCE(opaque);
1053     uint32_t offset = addr & 0xFFF;
1054     uint32_t srcno = addr >> xsrc->esb_shift;
1055     bool notify = false;
1056 
1057     trace_xive_source_esb_write(addr, srcno, value);
1058 
1059     /* In a two pages ESB MMIO setting, trigger page only triggers */
1060     if (xive_source_is_trigger_page(xsrc, addr)) {
1061         notify = xive_source_esb_trigger(xsrc, srcno);
1062         goto out;
1063     }
1064 
1065     switch (offset) {
1066     case 0 ... 0x3FF:
1067         notify = xive_source_esb_trigger(xsrc, srcno);
1068         break;
1069 
1070     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
1071         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
1072             qemu_log_mask(LOG_GUEST_ERROR,
1073                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
1074             return;
1075         }
1076 
1077         notify = xive_source_esb_eoi(xsrc, srcno);
1078         break;
1079 
1080     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1081     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1082     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1083     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1084         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1085         break;
1086 
1087     default:
1088         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
1089                       offset);
1090         return;
1091     }
1092 
1093 out:
1094     /* Forward the source event notification for routing */
1095     if (notify) {
1096         xive_source_notify(xsrc, srcno);
1097     }
1098 }
1099 
1100 static const MemoryRegionOps xive_source_esb_ops = {
1101     .read = xive_source_esb_read,
1102     .write = xive_source_esb_write,
1103     .endianness = DEVICE_BIG_ENDIAN,
1104     .valid = {
1105         .min_access_size = 8,
1106         .max_access_size = 8,
1107     },
1108     .impl = {
1109         .min_access_size = 8,
1110         .max_access_size = 8,
1111     },
1112 };
1113 
1114 void xive_source_set_irq(void *opaque, int srcno, int val)
1115 {
1116     XiveSource *xsrc = XIVE_SOURCE(opaque);
1117     bool notify = false;
1118 
1119     if (xive_source_irq_is_lsi(xsrc, srcno)) {
1120         if (val) {
1121             notify = xive_source_lsi_trigger(xsrc, srcno);
1122         } else {
1123             xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
1124         }
1125     } else {
1126         if (val) {
1127             notify = xive_source_esb_trigger(xsrc, srcno);
1128         }
1129     }
1130 
1131     /* Forward the source event notification for routing */
1132     if (notify) {
1133         xive_source_notify(xsrc, srcno);
1134     }
1135 }
1136 
1137 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
1138 {
1139     int i;
1140 
1141     for (i = 0; i < xsrc->nr_irqs; i++) {
1142         uint8_t pq = xive_source_esb_get(xsrc, i);
1143 
1144         if (pq == XIVE_ESB_OFF) {
1145             continue;
1146         }
1147 
1148         monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
1149                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
1150                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
1151                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1152                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ');
1153     }
1154 }
1155 
1156 static void xive_source_reset(void *dev)
1157 {
1158     XiveSource *xsrc = XIVE_SOURCE(dev);
1159 
1160     /* Do not clear the LSI bitmap */
1161 
1162     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
1163     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
1164 }
1165 
1166 static void xive_source_realize(DeviceState *dev, Error **errp)
1167 {
1168     XiveSource *xsrc = XIVE_SOURCE(dev);
1169     size_t esb_len = xive_source_esb_len(xsrc);
1170 
1171     assert(xsrc->xive);
1172 
1173     if (!xsrc->nr_irqs) {
1174         error_setg(errp, "Number of interrupt needs to be greater than 0");
1175         return;
1176     }
1177 
1178     if (xsrc->esb_shift != XIVE_ESB_4K &&
1179         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1180         xsrc->esb_shift != XIVE_ESB_64K &&
1181         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1182         error_setg(errp, "Invalid ESB shift setting");
1183         return;
1184     }
1185 
1186     xsrc->status = g_malloc0(xsrc->nr_irqs);
1187     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1188 
1189     memory_region_init(&xsrc->esb_mmio, OBJECT(xsrc), "xive.esb", esb_len);
1190     memory_region_init_io(&xsrc->esb_mmio_emulated, OBJECT(xsrc),
1191                           &xive_source_esb_ops, xsrc, "xive.esb-emulated",
1192                           esb_len);
1193     memory_region_add_subregion(&xsrc->esb_mmio, 0, &xsrc->esb_mmio_emulated);
1194 
1195     qemu_register_reset(xive_source_reset, dev);
1196 }
1197 
1198 static const VMStateDescription vmstate_xive_source = {
1199     .name = TYPE_XIVE_SOURCE,
1200     .version_id = 1,
1201     .minimum_version_id = 1,
1202     .fields = (VMStateField[]) {
1203         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1204         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1205         VMSTATE_END_OF_LIST()
1206     },
1207 };
1208 
1209 /*
1210  * The default XIVE interrupt source setting for the ESB MMIOs is two
1211  * 64k pages without Store EOI, to be in sync with KVM.
1212  */
1213 static Property xive_source_properties[] = {
1214     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1215     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1216     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1217     DEFINE_PROP_LINK("xive", XiveSource, xive, TYPE_XIVE_NOTIFIER,
1218                      XiveNotifier *),
1219     DEFINE_PROP_END_OF_LIST(),
1220 };
1221 
1222 static void xive_source_class_init(ObjectClass *klass, void *data)
1223 {
1224     DeviceClass *dc = DEVICE_CLASS(klass);
1225 
1226     dc->desc    = "XIVE Interrupt Source";
1227     device_class_set_props(dc, xive_source_properties);
1228     dc->realize = xive_source_realize;
1229     dc->vmsd    = &vmstate_xive_source;
1230     /*
1231      * Reason: part of XIVE interrupt controller, needs to be wired up,
1232      * e.g. by spapr_xive_instance_init().
1233      */
1234     dc->user_creatable = false;
1235 }
1236 
1237 static const TypeInfo xive_source_info = {
1238     .name          = TYPE_XIVE_SOURCE,
1239     .parent        = TYPE_DEVICE,
1240     .instance_size = sizeof(XiveSource),
1241     .class_init    = xive_source_class_init,
1242 };
1243 
1244 /*
1245  * XiveEND helpers
1246  */
1247 
1248 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
1249 {
1250     uint64_t qaddr_base = xive_end_qaddr(end);
1251     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1252     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1253     uint32_t qentries = 1 << (qsize + 10);
1254     int i;
1255 
1256     /*
1257      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1258      */
1259     monitor_printf(mon, " [ ");
1260     qindex = (qindex - (width - 1)) & (qentries - 1);
1261     for (i = 0; i < width; i++) {
1262         uint64_t qaddr = qaddr_base + (qindex << 2);
1263         uint32_t qdata = -1;
1264 
1265         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
1266                             sizeof(qdata))) {
1267             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1268                           HWADDR_PRIx "\n", qaddr);
1269             return;
1270         }
1271         monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1272                        be32_to_cpu(qdata));
1273         qindex = (qindex + 1) & (qentries - 1);
1274     }
1275     monitor_printf(mon, "]");
1276 }
1277 
1278 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1279 {
1280     uint64_t qaddr_base = xive_end_qaddr(end);
1281     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1282     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1283     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1284     uint32_t qentries = 1 << (qsize + 10);
1285 
1286     uint32_t nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1287     uint32_t nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1288     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1289     uint8_t pq;
1290 
1291     if (!xive_end_is_valid(end)) {
1292         return;
1293     }
1294 
1295     pq = xive_get_field32(END_W1_ESn, end->w1);
1296 
1297     monitor_printf(mon, "  %08x %c%c %c%c%c%c%c%c%c%c prio:%d nvt:%02x/%04x",
1298                    end_idx,
1299                    pq & XIVE_ESB_VAL_P ? 'P' : '-',
1300                    pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1301                    xive_end_is_valid(end)    ? 'v' : '-',
1302                    xive_end_is_enqueue(end)  ? 'q' : '-',
1303                    xive_end_is_notify(end)   ? 'n' : '-',
1304                    xive_end_is_backlog(end)  ? 'b' : '-',
1305                    xive_end_is_escalate(end) ? 'e' : '-',
1306                    xive_end_is_uncond_escalation(end)   ? 'u' : '-',
1307                    xive_end_is_silent_escalation(end)   ? 's' : '-',
1308                    xive_end_is_firmware(end)   ? 'f' : '-',
1309                    priority, nvt_blk, nvt_idx);
1310 
1311     if (qaddr_base) {
1312         monitor_printf(mon, " eq:@%08"PRIx64"% 6d/%5d ^%d",
1313                        qaddr_base, qindex, qentries, qgen);
1314         xive_end_queue_pic_print_info(end, 6, mon);
1315     }
1316     monitor_printf(mon, "\n");
1317 }
1318 
1319 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1320 {
1321     uint64_t qaddr_base = xive_end_qaddr(end);
1322     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1323     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1324     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1325 
1326     uint64_t qaddr = qaddr_base + (qindex << 2);
1327     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1328     uint32_t qentries = 1 << (qsize + 10);
1329 
1330     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata))) {
1331         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1332                       HWADDR_PRIx "\n", qaddr);
1333         return;
1334     }
1335 
1336     qindex = (qindex + 1) & (qentries - 1);
1337     if (qindex == 0) {
1338         qgen ^= 1;
1339         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1340     }
1341     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1342 }
1343 
1344 void xive_end_eas_pic_print_info(XiveEND *end, uint32_t end_idx,
1345                                    Monitor *mon)
1346 {
1347     XiveEAS *eas = (XiveEAS *) &end->w4;
1348     uint8_t pq;
1349 
1350     if (!xive_end_is_escalate(end)) {
1351         return;
1352     }
1353 
1354     pq = xive_get_field32(END_W1_ESe, end->w1);
1355 
1356     monitor_printf(mon, "  %08x %c%c %c%c end:%02x/%04x data:%08x\n",
1357                    end_idx,
1358                    pq & XIVE_ESB_VAL_P ? 'P' : '-',
1359                    pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1360                    xive_eas_is_valid(eas) ? 'V' : ' ',
1361                    xive_eas_is_masked(eas) ? 'M' : ' ',
1362                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1363                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1364                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1365 }
1366 
1367 /*
1368  * XIVE Router (aka. Virtualization Controller or IVRE)
1369  */
1370 
1371 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1372                         XiveEAS *eas)
1373 {
1374     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1375 
1376     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1377 }
1378 
1379 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1380                         XiveEND *end)
1381 {
1382    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1383 
1384    return xrc->get_end(xrtr, end_blk, end_idx, end);
1385 }
1386 
1387 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1388                           XiveEND *end, uint8_t word_number)
1389 {
1390    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1391 
1392    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1393 }
1394 
1395 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1396                         XiveNVT *nvt)
1397 {
1398    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1399 
1400    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1401 }
1402 
1403 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1404                         XiveNVT *nvt, uint8_t word_number)
1405 {
1406    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1407 
1408    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1409 }
1410 
1411 static int xive_router_get_block_id(XiveRouter *xrtr)
1412 {
1413    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1414 
1415    return xrc->get_block_id(xrtr);
1416 }
1417 
1418 static void xive_router_realize(DeviceState *dev, Error **errp)
1419 {
1420     XiveRouter *xrtr = XIVE_ROUTER(dev);
1421 
1422     assert(xrtr->xfb);
1423 }
1424 
1425 /*
1426  * Encode the HW CAM line in the block group mode format :
1427  *
1428  *   chip << 19 | 0000000 0 0001 thread (7Bit)
1429  */
1430 static uint32_t xive_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
1431 {
1432     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1433     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1434     uint8_t blk = xive_router_get_block_id(XIVE_ROUTER(xptr));
1435 
1436     return xive_nvt_cam_line(blk, 1 << 7 | (pir & 0x7f));
1437 }
1438 
1439 /*
1440  * The thread context register words are in big-endian format.
1441  */
1442 int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1443                               uint8_t format,
1444                               uint8_t nvt_blk, uint32_t nvt_idx,
1445                               bool cam_ignore, uint32_t logic_serv)
1446 {
1447     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1448     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1449     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1450     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1451     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1452 
1453     /*
1454      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1455      * identifier are ignored in the "CAM" match.
1456      */
1457 
1458     if (format == 0) {
1459         if (cam_ignore == true) {
1460             /*
1461              * F=0 & i=1: Logical server notification (bits ignored at
1462              * the end of the NVT identifier)
1463              */
1464             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1465                           nvt_blk, nvt_idx);
1466              return -1;
1467         }
1468 
1469         /* F=0 & i=0: Specific NVT notification */
1470 
1471         /* PHYS ring */
1472         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1473             cam == xive_tctx_hw_cam_line(xptr, tctx)) {
1474             return TM_QW3_HV_PHYS;
1475         }
1476 
1477         /* HV POOL ring */
1478         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1479             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1480             return TM_QW2_HV_POOL;
1481         }
1482 
1483         /* OS ring */
1484         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1485             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1486             return TM_QW1_OS;
1487         }
1488     } else {
1489         /* F=1 : User level Event-Based Branch (EBB) notification */
1490 
1491         /* USER ring */
1492         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1493              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1494              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1495              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1496             return TM_QW0_USER;
1497         }
1498     }
1499     return -1;
1500 }
1501 
1502 /*
1503  * This is our simple Xive Presenter Engine model. It is merged in the
1504  * Router as it does not require an extra object.
1505  *
1506  * It receives notification requests sent by the IVRE to find one
1507  * matching NVT (or more) dispatched on the processor threads. In case
1508  * of a single NVT notification, the process is abreviated and the
1509  * thread is signaled if a match is found. In case of a logical server
1510  * notification (bits ignored at the end of the NVT identifier), the
1511  * IVPE and IVRE select a winning thread using different filters. This
1512  * involves 2 or 3 exchanges on the PowerBus that the model does not
1513  * support.
1514  *
1515  * The parameters represent what is sent on the PowerBus
1516  */
1517 bool xive_presenter_notify(XiveFabric *xfb, uint8_t format,
1518                            uint8_t nvt_blk, uint32_t nvt_idx,
1519                            bool cam_ignore, uint8_t priority,
1520                            uint32_t logic_serv)
1521 {
1522     XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xfb);
1523     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1524     int count;
1525 
1526     /*
1527      * Ask the machine to scan the interrupt controllers for a match
1528      */
1529     count = xfc->match_nvt(xfb, format, nvt_blk, nvt_idx, cam_ignore,
1530                            priority, logic_serv, &match);
1531     if (count < 0) {
1532         return false;
1533     }
1534 
1535     /* handle CPU exception delivery */
1536     if (count) {
1537         trace_xive_presenter_notify(nvt_blk, nvt_idx, match.ring);
1538         xive_tctx_ipb_update(match.tctx, match.ring, priority_to_ipb(priority));
1539     }
1540 
1541     return !!count;
1542 }
1543 
1544 /*
1545  * Notification using the END ESe/ESn bit (Event State Buffer for
1546  * escalation and notification). Provide further coalescing in the
1547  * Router.
1548  */
1549 static bool xive_router_end_es_notify(XiveRouter *xrtr, uint8_t end_blk,
1550                                       uint32_t end_idx, XiveEND *end,
1551                                       uint32_t end_esmask)
1552 {
1553     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1554     bool notify = xive_esb_trigger(&pq);
1555 
1556     if (pq != xive_get_field32(end_esmask, end->w1)) {
1557         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1558         xive_router_write_end(xrtr, end_blk, end_idx, end, 1);
1559     }
1560 
1561     /* ESe/n[Q]=1 : end of notification */
1562     return notify;
1563 }
1564 
1565 /*
1566  * An END trigger can come from an event trigger (IPI or HW) or from
1567  * another chip. We don't model the PowerBus but the END trigger
1568  * message has the same parameters than in the function below.
1569  */
1570 static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1571                                    uint32_t end_idx, uint32_t end_data)
1572 {
1573     XiveEND end;
1574     uint8_t priority;
1575     uint8_t format;
1576     uint8_t nvt_blk;
1577     uint32_t nvt_idx;
1578     XiveNVT nvt;
1579     bool found;
1580 
1581     /* END cache lookup */
1582     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1583         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1584                       end_idx);
1585         return;
1586     }
1587 
1588     if (!xive_end_is_valid(&end)) {
1589         trace_xive_router_end_notify(end_blk, end_idx, end_data);
1590         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1591                       end_blk, end_idx);
1592         return;
1593     }
1594 
1595     if (xive_end_is_enqueue(&end)) {
1596         xive_end_enqueue(&end, end_data);
1597         /* Enqueuing event data modifies the EQ toggle and index */
1598         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1599     }
1600 
1601     /*
1602      * When the END is silent, we skip the notification part.
1603      */
1604     if (xive_end_is_silent_escalation(&end)) {
1605         goto do_escalation;
1606     }
1607 
1608     /*
1609      * The W7 format depends on the F bit in W6. It defines the type
1610      * of the notification :
1611      *
1612      *   F=0 : single or multiple NVT notification
1613      *   F=1 : User level Event-Based Branch (EBB) notification, no
1614      *         priority
1615      */
1616     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1617     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1618 
1619     /* The END is masked */
1620     if (format == 0 && priority == 0xff) {
1621         return;
1622     }
1623 
1624     /*
1625      * Check the END ESn (Event State Buffer for notification) for
1626      * even further coalescing in the Router
1627      */
1628     if (!xive_end_is_notify(&end)) {
1629         /* ESn[Q]=1 : end of notification */
1630         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1631                                        &end, END_W1_ESn)) {
1632             return;
1633         }
1634     }
1635 
1636     /*
1637      * Follows IVPE notification
1638      */
1639     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end.w6);
1640     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end.w6);
1641 
1642     /* NVT cache lookup */
1643     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1644         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1645                       nvt_blk, nvt_idx);
1646         return;
1647     }
1648 
1649     if (!xive_nvt_is_valid(&nvt)) {
1650         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1651                       nvt_blk, nvt_idx);
1652         return;
1653     }
1654 
1655     found = xive_presenter_notify(xrtr->xfb, format, nvt_blk, nvt_idx,
1656                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1657                           priority,
1658                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1659 
1660     /* TODO: Auto EOI. */
1661 
1662     if (found) {
1663         return;
1664     }
1665 
1666     /*
1667      * If no matching NVT is dispatched on a HW thread :
1668      * - specific VP: update the NVT structure if backlog is activated
1669      * - logical server : forward request to IVPE (not supported)
1670      */
1671     if (xive_end_is_backlog(&end)) {
1672         uint8_t ipb;
1673 
1674         if (format == 1) {
1675             qemu_log_mask(LOG_GUEST_ERROR,
1676                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1677                           end_blk, end_idx);
1678             return;
1679         }
1680         /*
1681          * Record the IPB in the associated NVT structure for later
1682          * use. The presenter will resend the interrupt when the vCPU
1683          * is dispatched again on a HW thread.
1684          */
1685         ipb = xive_get_field32(NVT_W4_IPB, nvt.w4) | priority_to_ipb(priority);
1686         nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, ipb);
1687         xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1688 
1689         /*
1690          * On HW, follows a "Broadcast Backlog" to IVPEs
1691          */
1692     }
1693 
1694 do_escalation:
1695     /*
1696      * If activated, escalate notification using the ESe PQ bits and
1697      * the EAS in w4-5
1698      */
1699     if (!xive_end_is_escalate(&end)) {
1700         return;
1701     }
1702 
1703     /*
1704      * Check the END ESe (Event State Buffer for escalation) for even
1705      * further coalescing in the Router
1706      */
1707     if (!xive_end_is_uncond_escalation(&end)) {
1708         /* ESe[Q]=1 : end of notification */
1709         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1710                                        &end, END_W1_ESe)) {
1711             return;
1712         }
1713     }
1714 
1715     trace_xive_router_end_escalate(end_blk, end_idx,
1716            (uint8_t) xive_get_field32(END_W4_ESC_END_BLOCK, end.w4),
1717            (uint32_t) xive_get_field32(END_W4_ESC_END_INDEX, end.w4),
1718            (uint32_t) xive_get_field32(END_W5_ESC_END_DATA,  end.w5));
1719     /*
1720      * The END trigger becomes an Escalation trigger
1721      */
1722     xive_router_end_notify(xrtr,
1723                            xive_get_field32(END_W4_ESC_END_BLOCK, end.w4),
1724                            xive_get_field32(END_W4_ESC_END_INDEX, end.w4),
1725                            xive_get_field32(END_W5_ESC_END_DATA,  end.w5));
1726 }
1727 
1728 void xive_router_notify(XiveNotifier *xn, uint32_t lisn)
1729 {
1730     XiveRouter *xrtr = XIVE_ROUTER(xn);
1731     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1732     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1733     XiveEAS eas;
1734 
1735     /* EAS cache lookup */
1736     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1737         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1738         return;
1739     }
1740 
1741     /*
1742      * The IVRE checks the State Bit Cache at this point. We skip the
1743      * SBC lookup because the state bits of the sources are modeled
1744      * internally in QEMU.
1745      */
1746 
1747     if (!xive_eas_is_valid(&eas)) {
1748         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1749         return;
1750     }
1751 
1752     if (xive_eas_is_masked(&eas)) {
1753         /* Notification completed */
1754         return;
1755     }
1756 
1757     /*
1758      * The event trigger becomes an END trigger
1759      */
1760     xive_router_end_notify(xrtr,
1761                            xive_get_field64(EAS_END_BLOCK, eas.w),
1762                            xive_get_field64(EAS_END_INDEX, eas.w),
1763                            xive_get_field64(EAS_END_DATA,  eas.w));
1764 }
1765 
1766 static Property xive_router_properties[] = {
1767     DEFINE_PROP_LINK("xive-fabric", XiveRouter, xfb,
1768                      TYPE_XIVE_FABRIC, XiveFabric *),
1769     DEFINE_PROP_END_OF_LIST(),
1770 };
1771 
1772 static void xive_router_class_init(ObjectClass *klass, void *data)
1773 {
1774     DeviceClass *dc = DEVICE_CLASS(klass);
1775     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1776 
1777     dc->desc    = "XIVE Router Engine";
1778     device_class_set_props(dc, xive_router_properties);
1779     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1780     dc->realize = xive_router_realize;
1781     xnc->notify = xive_router_notify;
1782 }
1783 
1784 static const TypeInfo xive_router_info = {
1785     .name          = TYPE_XIVE_ROUTER,
1786     .parent        = TYPE_SYS_BUS_DEVICE,
1787     .abstract      = true,
1788     .instance_size = sizeof(XiveRouter),
1789     .class_size    = sizeof(XiveRouterClass),
1790     .class_init    = xive_router_class_init,
1791     .interfaces    = (InterfaceInfo[]) {
1792         { TYPE_XIVE_NOTIFIER },
1793         { TYPE_XIVE_PRESENTER },
1794         { }
1795     }
1796 };
1797 
1798 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1799 {
1800     if (!xive_eas_is_valid(eas)) {
1801         return;
1802     }
1803 
1804     monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1805                    lisn, xive_eas_is_masked(eas) ? "M" : " ",
1806                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1807                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1808                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1809 }
1810 
1811 /*
1812  * END ESB MMIO loads
1813  */
1814 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1815 {
1816     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1817     uint32_t offset = addr & 0xFFF;
1818     uint8_t end_blk;
1819     uint32_t end_idx;
1820     XiveEND end;
1821     uint32_t end_esmask;
1822     uint8_t pq;
1823     uint64_t ret = -1;
1824 
1825     /*
1826      * The block id should be deduced from the load address on the END
1827      * ESB MMIO but our model only supports a single block per XIVE chip.
1828      */
1829     end_blk = xive_router_get_block_id(xsrc->xrtr);
1830     end_idx = addr >> (xsrc->esb_shift + 1);
1831 
1832     trace_xive_end_source_read(end_blk, end_idx, addr);
1833 
1834     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1835         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1836                       end_idx);
1837         return -1;
1838     }
1839 
1840     if (!xive_end_is_valid(&end)) {
1841         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1842                       end_blk, end_idx);
1843         return -1;
1844     }
1845 
1846     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1847     pq = xive_get_field32(end_esmask, end.w1);
1848 
1849     switch (offset) {
1850     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1851         ret = xive_esb_eoi(&pq);
1852 
1853         /* Forward the source event notification for routing ?? */
1854         break;
1855 
1856     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1857         ret = pq;
1858         break;
1859 
1860     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1861     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1862     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1863     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1864         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1865         break;
1866     default:
1867         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1868                       offset);
1869         return -1;
1870     }
1871 
1872     if (pq != xive_get_field32(end_esmask, end.w1)) {
1873         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1874         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1875     }
1876 
1877     return ret;
1878 }
1879 
1880 /*
1881  * END ESB MMIO stores are invalid
1882  */
1883 static void xive_end_source_write(void *opaque, hwaddr addr,
1884                                   uint64_t value, unsigned size)
1885 {
1886     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1887                   HWADDR_PRIx"\n", addr);
1888 }
1889 
1890 static const MemoryRegionOps xive_end_source_ops = {
1891     .read = xive_end_source_read,
1892     .write = xive_end_source_write,
1893     .endianness = DEVICE_BIG_ENDIAN,
1894     .valid = {
1895         .min_access_size = 8,
1896         .max_access_size = 8,
1897     },
1898     .impl = {
1899         .min_access_size = 8,
1900         .max_access_size = 8,
1901     },
1902 };
1903 
1904 static void xive_end_source_realize(DeviceState *dev, Error **errp)
1905 {
1906     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1907 
1908     assert(xsrc->xrtr);
1909 
1910     if (!xsrc->nr_ends) {
1911         error_setg(errp, "Number of interrupt needs to be greater than 0");
1912         return;
1913     }
1914 
1915     if (xsrc->esb_shift != XIVE_ESB_4K &&
1916         xsrc->esb_shift != XIVE_ESB_64K) {
1917         error_setg(errp, "Invalid ESB shift setting");
1918         return;
1919     }
1920 
1921     /*
1922      * Each END is assigned an even/odd pair of MMIO pages, the even page
1923      * manages the ESn field while the odd page manages the ESe field.
1924      */
1925     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1926                           &xive_end_source_ops, xsrc, "xive.end",
1927                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1928 }
1929 
1930 static Property xive_end_source_properties[] = {
1931     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
1932     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
1933     DEFINE_PROP_LINK("xive", XiveENDSource, xrtr, TYPE_XIVE_ROUTER,
1934                      XiveRouter *),
1935     DEFINE_PROP_END_OF_LIST(),
1936 };
1937 
1938 static void xive_end_source_class_init(ObjectClass *klass, void *data)
1939 {
1940     DeviceClass *dc = DEVICE_CLASS(klass);
1941 
1942     dc->desc    = "XIVE END Source";
1943     device_class_set_props(dc, xive_end_source_properties);
1944     dc->realize = xive_end_source_realize;
1945     /*
1946      * Reason: part of XIVE interrupt controller, needs to be wired up,
1947      * e.g. by spapr_xive_instance_init().
1948      */
1949     dc->user_creatable = false;
1950 }
1951 
1952 static const TypeInfo xive_end_source_info = {
1953     .name          = TYPE_XIVE_END_SOURCE,
1954     .parent        = TYPE_DEVICE,
1955     .instance_size = sizeof(XiveENDSource),
1956     .class_init    = xive_end_source_class_init,
1957 };
1958 
1959 /*
1960  * XIVE Notifier
1961  */
1962 static const TypeInfo xive_notifier_info = {
1963     .name = TYPE_XIVE_NOTIFIER,
1964     .parent = TYPE_INTERFACE,
1965     .class_size = sizeof(XiveNotifierClass),
1966 };
1967 
1968 /*
1969  * XIVE Presenter
1970  */
1971 static const TypeInfo xive_presenter_info = {
1972     .name = TYPE_XIVE_PRESENTER,
1973     .parent = TYPE_INTERFACE,
1974     .class_size = sizeof(XivePresenterClass),
1975 };
1976 
1977 /*
1978  * XIVE Fabric
1979  */
1980 static const TypeInfo xive_fabric_info = {
1981     .name = TYPE_XIVE_FABRIC,
1982     .parent = TYPE_INTERFACE,
1983     .class_size = sizeof(XiveFabricClass),
1984 };
1985 
1986 static void xive_register_types(void)
1987 {
1988     type_register_static(&xive_fabric_info);
1989     type_register_static(&xive_source_info);
1990     type_register_static(&xive_notifier_info);
1991     type_register_static(&xive_presenter_info);
1992     type_register_static(&xive_router_info);
1993     type_register_static(&xive_end_source_info);
1994     type_register_static(&xive_tctx_info);
1995 }
1996 
1997 type_init(xive_register_types)
1998