xref: /openbmc/qemu/hw/intc/xive.c (revision 2a8100cb61959e7a934049f2bd3a49a0f84a066b)
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     Error *local_err = NULL;
666 
667     assert(tctx->cs);
668     assert(tctx->xptr);
669 
670     cpu = POWERPC_CPU(tctx->cs);
671     env = &cpu->env;
672     switch (PPC_INPUT(env)) {
673     case PPC_FLAGS_INPUT_POWER9:
674         tctx->hv_output = env->irq_inputs[POWER9_INPUT_HINT];
675         tctx->os_output = env->irq_inputs[POWER9_INPUT_INT];
676         break;
677 
678     default:
679         error_setg(errp, "XIVE interrupt controller does not support "
680                    "this CPU bus model");
681         return;
682     }
683 
684     /* Connect the presenter to the VCPU (required for CPU hotplug) */
685     if (xive_in_kernel(tctx->xptr)) {
686         kvmppc_xive_cpu_connect(tctx, &local_err);
687         if (local_err) {
688             error_propagate(errp, local_err);
689             return;
690         }
691     }
692 }
693 
694 static int vmstate_xive_tctx_pre_save(void *opaque)
695 {
696     XiveTCTX *tctx = XIVE_TCTX(opaque);
697     Error *local_err = NULL;
698     int ret;
699 
700     if (xive_in_kernel(tctx->xptr)) {
701         ret = kvmppc_xive_cpu_get_state(tctx, &local_err);
702         if (ret < 0) {
703             error_report_err(local_err);
704             return ret;
705         }
706     }
707 
708     return 0;
709 }
710 
711 static int vmstate_xive_tctx_post_load(void *opaque, int version_id)
712 {
713     XiveTCTX *tctx = XIVE_TCTX(opaque);
714     Error *local_err = NULL;
715     int ret;
716 
717     if (xive_in_kernel(tctx->xptr)) {
718         /*
719          * Required for hotplugged CPU, for which the state comes
720          * after all states of the machine.
721          */
722         ret = kvmppc_xive_cpu_set_state(tctx, &local_err);
723         if (ret < 0) {
724             error_report_err(local_err);
725             return ret;
726         }
727     }
728 
729     return 0;
730 }
731 
732 static const VMStateDescription vmstate_xive_tctx = {
733     .name = TYPE_XIVE_TCTX,
734     .version_id = 1,
735     .minimum_version_id = 1,
736     .pre_save = vmstate_xive_tctx_pre_save,
737     .post_load = vmstate_xive_tctx_post_load,
738     .fields = (VMStateField[]) {
739         VMSTATE_BUFFER(regs, XiveTCTX),
740         VMSTATE_END_OF_LIST()
741     },
742 };
743 
744 static Property xive_tctx_properties[] = {
745     DEFINE_PROP_LINK("cpu", XiveTCTX, cs, TYPE_CPU, CPUState *),
746     DEFINE_PROP_LINK("presenter", XiveTCTX, xptr, TYPE_XIVE_PRESENTER,
747                      XivePresenter *),
748     DEFINE_PROP_END_OF_LIST(),
749 };
750 
751 static void xive_tctx_class_init(ObjectClass *klass, void *data)
752 {
753     DeviceClass *dc = DEVICE_CLASS(klass);
754 
755     dc->desc = "XIVE Interrupt Thread Context";
756     dc->realize = xive_tctx_realize;
757     dc->vmsd = &vmstate_xive_tctx;
758     device_class_set_props(dc, xive_tctx_properties);
759     /*
760      * Reason: part of XIVE interrupt controller, needs to be wired up
761      * by xive_tctx_create().
762      */
763     dc->user_creatable = false;
764 }
765 
766 static const TypeInfo xive_tctx_info = {
767     .name          = TYPE_XIVE_TCTX,
768     .parent        = TYPE_DEVICE,
769     .instance_size = sizeof(XiveTCTX),
770     .class_init    = xive_tctx_class_init,
771 };
772 
773 Object *xive_tctx_create(Object *cpu, XivePresenter *xptr, Error **errp)
774 {
775     Object *obj;
776 
777     obj = object_new(TYPE_XIVE_TCTX);
778     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj);
779     object_unref(obj);
780     object_property_set_link(obj, "cpu", cpu, &error_abort);
781     object_property_set_link(obj, "presenter", OBJECT(xptr), &error_abort);
782     if (!qdev_realize(DEVICE(obj), NULL, errp)) {
783         object_unparent(obj);
784         return NULL;
785     }
786     return obj;
787 }
788 
789 void xive_tctx_destroy(XiveTCTX *tctx)
790 {
791     Object *obj = OBJECT(tctx);
792 
793     object_unparent(obj);
794 }
795 
796 /*
797  * XIVE ESB helpers
798  */
799 
800 static uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
801 {
802     uint8_t old_pq = *pq & 0x3;
803 
804     *pq &= ~0x3;
805     *pq |= value & 0x3;
806 
807     return old_pq;
808 }
809 
810 static bool xive_esb_trigger(uint8_t *pq)
811 {
812     uint8_t old_pq = *pq & 0x3;
813 
814     switch (old_pq) {
815     case XIVE_ESB_RESET:
816         xive_esb_set(pq, XIVE_ESB_PENDING);
817         return true;
818     case XIVE_ESB_PENDING:
819     case XIVE_ESB_QUEUED:
820         xive_esb_set(pq, XIVE_ESB_QUEUED);
821         return false;
822     case XIVE_ESB_OFF:
823         xive_esb_set(pq, XIVE_ESB_OFF);
824         return false;
825     default:
826          g_assert_not_reached();
827     }
828 }
829 
830 static bool xive_esb_eoi(uint8_t *pq)
831 {
832     uint8_t old_pq = *pq & 0x3;
833 
834     switch (old_pq) {
835     case XIVE_ESB_RESET:
836     case XIVE_ESB_PENDING:
837         xive_esb_set(pq, XIVE_ESB_RESET);
838         return false;
839     case XIVE_ESB_QUEUED:
840         xive_esb_set(pq, XIVE_ESB_PENDING);
841         return true;
842     case XIVE_ESB_OFF:
843         xive_esb_set(pq, XIVE_ESB_OFF);
844         return false;
845     default:
846          g_assert_not_reached();
847     }
848 }
849 
850 /*
851  * XIVE Interrupt Source (or IVSE)
852  */
853 
854 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
855 {
856     assert(srcno < xsrc->nr_irqs);
857 
858     return xsrc->status[srcno] & 0x3;
859 }
860 
861 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
862 {
863     assert(srcno < xsrc->nr_irqs);
864 
865     return xive_esb_set(&xsrc->status[srcno], pq);
866 }
867 
868 /*
869  * Returns whether the event notification should be forwarded.
870  */
871 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
872 {
873     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
874 
875     xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
876 
877     switch (old_pq) {
878     case XIVE_ESB_RESET:
879         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
880         return true;
881     default:
882         return false;
883     }
884 }
885 
886 /*
887  * Returns whether the event notification should be forwarded.
888  */
889 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
890 {
891     bool ret;
892 
893     assert(srcno < xsrc->nr_irqs);
894 
895     ret = xive_esb_trigger(&xsrc->status[srcno]);
896 
897     if (xive_source_irq_is_lsi(xsrc, srcno) &&
898         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
899         qemu_log_mask(LOG_GUEST_ERROR,
900                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
901     }
902 
903     return ret;
904 }
905 
906 /*
907  * Returns whether the event notification should be forwarded.
908  */
909 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
910 {
911     bool ret;
912 
913     assert(srcno < xsrc->nr_irqs);
914 
915     ret = xive_esb_eoi(&xsrc->status[srcno]);
916 
917     /*
918      * LSI sources do not set the Q bit but they can still be
919      * asserted, in which case we should forward a new event
920      * notification
921      */
922     if (xive_source_irq_is_lsi(xsrc, srcno) &&
923         xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
924         ret = xive_source_lsi_trigger(xsrc, srcno);
925     }
926 
927     return ret;
928 }
929 
930 /*
931  * Forward the source event notification to the Router
932  */
933 static void xive_source_notify(XiveSource *xsrc, int srcno)
934 {
935     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
936 
937     if (xnc->notify) {
938         xnc->notify(xsrc->xive, srcno);
939     }
940 }
941 
942 /*
943  * In a two pages ESB MMIO setting, even page is the trigger page, odd
944  * page is for management
945  */
946 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
947 {
948     return !((addr >> shift) & 1);
949 }
950 
951 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
952 {
953     return xive_source_esb_has_2page(xsrc) &&
954         addr_is_even(addr, xsrc->esb_shift - 1);
955 }
956 
957 /*
958  * ESB MMIO loads
959  *                      Trigger page    Management/EOI page
960  *
961  * ESB MMIO setting     2 pages         1 or 2 pages
962  *
963  * 0x000 .. 0x3FF       -1              EOI and return 0|1
964  * 0x400 .. 0x7FF       -1              EOI and return 0|1
965  * 0x800 .. 0xBFF       -1              return PQ
966  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
967  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
968  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
969  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
970  */
971 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
972 {
973     XiveSource *xsrc = XIVE_SOURCE(opaque);
974     uint32_t offset = addr & 0xFFF;
975     uint32_t srcno = addr >> xsrc->esb_shift;
976     uint64_t ret = -1;
977 
978     /* In a two pages ESB MMIO setting, trigger page should not be read */
979     if (xive_source_is_trigger_page(xsrc, addr)) {
980         qemu_log_mask(LOG_GUEST_ERROR,
981                       "XIVE: invalid load on IRQ %d trigger page at "
982                       "0x%"HWADDR_PRIx"\n", srcno, addr);
983         return -1;
984     }
985 
986     switch (offset) {
987     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
988         ret = xive_source_esb_eoi(xsrc, srcno);
989 
990         /* Forward the source event notification for routing */
991         if (ret) {
992             xive_source_notify(xsrc, srcno);
993         }
994         break;
995 
996     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
997         ret = xive_source_esb_get(xsrc, srcno);
998         break;
999 
1000     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1001     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1002     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1003     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1004         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1005         break;
1006     default:
1007         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
1008                       offset);
1009     }
1010 
1011     return ret;
1012 }
1013 
1014 /*
1015  * ESB MMIO stores
1016  *                      Trigger page    Management/EOI page
1017  *
1018  * ESB MMIO setting     2 pages         1 or 2 pages
1019  *
1020  * 0x000 .. 0x3FF       Trigger         Trigger
1021  * 0x400 .. 0x7FF       Trigger         EOI
1022  * 0x800 .. 0xBFF       Trigger         undefined
1023  * 0xC00 .. 0xCFF       Trigger         PQ=00
1024  * 0xD00 .. 0xDFF       Trigger         PQ=01
1025  * 0xE00 .. 0xDFF       Trigger         PQ=10
1026  * 0xF00 .. 0xDFF       Trigger         PQ=11
1027  */
1028 static void xive_source_esb_write(void *opaque, hwaddr addr,
1029                                   uint64_t value, unsigned size)
1030 {
1031     XiveSource *xsrc = XIVE_SOURCE(opaque);
1032     uint32_t offset = addr & 0xFFF;
1033     uint32_t srcno = addr >> xsrc->esb_shift;
1034     bool notify = false;
1035 
1036     /* In a two pages ESB MMIO setting, trigger page only triggers */
1037     if (xive_source_is_trigger_page(xsrc, addr)) {
1038         notify = xive_source_esb_trigger(xsrc, srcno);
1039         goto out;
1040     }
1041 
1042     switch (offset) {
1043     case 0 ... 0x3FF:
1044         notify = xive_source_esb_trigger(xsrc, srcno);
1045         break;
1046 
1047     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
1048         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
1049             qemu_log_mask(LOG_GUEST_ERROR,
1050                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
1051             return;
1052         }
1053 
1054         notify = xive_source_esb_eoi(xsrc, srcno);
1055         break;
1056 
1057     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1058     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1059     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1060     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1061         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1062         break;
1063 
1064     default:
1065         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
1066                       offset);
1067         return;
1068     }
1069 
1070 out:
1071     /* Forward the source event notification for routing */
1072     if (notify) {
1073         xive_source_notify(xsrc, srcno);
1074     }
1075 }
1076 
1077 static const MemoryRegionOps xive_source_esb_ops = {
1078     .read = xive_source_esb_read,
1079     .write = xive_source_esb_write,
1080     .endianness = DEVICE_BIG_ENDIAN,
1081     .valid = {
1082         .min_access_size = 8,
1083         .max_access_size = 8,
1084     },
1085     .impl = {
1086         .min_access_size = 8,
1087         .max_access_size = 8,
1088     },
1089 };
1090 
1091 void xive_source_set_irq(void *opaque, int srcno, int val)
1092 {
1093     XiveSource *xsrc = XIVE_SOURCE(opaque);
1094     bool notify = false;
1095 
1096     if (xive_source_irq_is_lsi(xsrc, srcno)) {
1097         if (val) {
1098             notify = xive_source_lsi_trigger(xsrc, srcno);
1099         } else {
1100             xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
1101         }
1102     } else {
1103         if (val) {
1104             notify = xive_source_esb_trigger(xsrc, srcno);
1105         }
1106     }
1107 
1108     /* Forward the source event notification for routing */
1109     if (notify) {
1110         xive_source_notify(xsrc, srcno);
1111     }
1112 }
1113 
1114 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
1115 {
1116     int i;
1117 
1118     for (i = 0; i < xsrc->nr_irqs; i++) {
1119         uint8_t pq = xive_source_esb_get(xsrc, i);
1120 
1121         if (pq == XIVE_ESB_OFF) {
1122             continue;
1123         }
1124 
1125         monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
1126                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
1127                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
1128                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1129                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ');
1130     }
1131 }
1132 
1133 static void xive_source_reset(void *dev)
1134 {
1135     XiveSource *xsrc = XIVE_SOURCE(dev);
1136 
1137     /* Do not clear the LSI bitmap */
1138 
1139     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
1140     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
1141 }
1142 
1143 static void xive_source_realize(DeviceState *dev, Error **errp)
1144 {
1145     XiveSource *xsrc = XIVE_SOURCE(dev);
1146     size_t esb_len = xive_source_esb_len(xsrc);
1147 
1148     assert(xsrc->xive);
1149 
1150     if (!xsrc->nr_irqs) {
1151         error_setg(errp, "Number of interrupt needs to be greater than 0");
1152         return;
1153     }
1154 
1155     if (xsrc->esb_shift != XIVE_ESB_4K &&
1156         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1157         xsrc->esb_shift != XIVE_ESB_64K &&
1158         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1159         error_setg(errp, "Invalid ESB shift setting");
1160         return;
1161     }
1162 
1163     xsrc->status = g_malloc0(xsrc->nr_irqs);
1164     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1165 
1166     memory_region_init(&xsrc->esb_mmio, OBJECT(xsrc), "xive.esb", esb_len);
1167     memory_region_init_io(&xsrc->esb_mmio_emulated, OBJECT(xsrc),
1168                           &xive_source_esb_ops, xsrc, "xive.esb-emulated",
1169                           esb_len);
1170     memory_region_add_subregion(&xsrc->esb_mmio, 0, &xsrc->esb_mmio_emulated);
1171 
1172     qemu_register_reset(xive_source_reset, dev);
1173 }
1174 
1175 static const VMStateDescription vmstate_xive_source = {
1176     .name = TYPE_XIVE_SOURCE,
1177     .version_id = 1,
1178     .minimum_version_id = 1,
1179     .fields = (VMStateField[]) {
1180         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1181         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1182         VMSTATE_END_OF_LIST()
1183     },
1184 };
1185 
1186 /*
1187  * The default XIVE interrupt source setting for the ESB MMIOs is two
1188  * 64k pages without Store EOI, to be in sync with KVM.
1189  */
1190 static Property xive_source_properties[] = {
1191     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1192     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1193     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1194     DEFINE_PROP_LINK("xive", XiveSource, xive, TYPE_XIVE_NOTIFIER,
1195                      XiveNotifier *),
1196     DEFINE_PROP_END_OF_LIST(),
1197 };
1198 
1199 static void xive_source_class_init(ObjectClass *klass, void *data)
1200 {
1201     DeviceClass *dc = DEVICE_CLASS(klass);
1202 
1203     dc->desc    = "XIVE Interrupt Source";
1204     device_class_set_props(dc, xive_source_properties);
1205     dc->realize = xive_source_realize;
1206     dc->vmsd    = &vmstate_xive_source;
1207     /*
1208      * Reason: part of XIVE interrupt controller, needs to be wired up,
1209      * e.g. by spapr_xive_instance_init().
1210      */
1211     dc->user_creatable = false;
1212 }
1213 
1214 static const TypeInfo xive_source_info = {
1215     .name          = TYPE_XIVE_SOURCE,
1216     .parent        = TYPE_DEVICE,
1217     .instance_size = sizeof(XiveSource),
1218     .class_init    = xive_source_class_init,
1219 };
1220 
1221 /*
1222  * XiveEND helpers
1223  */
1224 
1225 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
1226 {
1227     uint64_t qaddr_base = xive_end_qaddr(end);
1228     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1229     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1230     uint32_t qentries = 1 << (qsize + 10);
1231     int i;
1232 
1233     /*
1234      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1235      */
1236     monitor_printf(mon, " [ ");
1237     qindex = (qindex - (width - 1)) & (qentries - 1);
1238     for (i = 0; i < width; i++) {
1239         uint64_t qaddr = qaddr_base + (qindex << 2);
1240         uint32_t qdata = -1;
1241 
1242         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
1243                             sizeof(qdata))) {
1244             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1245                           HWADDR_PRIx "\n", qaddr);
1246             return;
1247         }
1248         monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1249                        be32_to_cpu(qdata));
1250         qindex = (qindex + 1) & (qentries - 1);
1251     }
1252     monitor_printf(mon, "]");
1253 }
1254 
1255 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1256 {
1257     uint64_t qaddr_base = xive_end_qaddr(end);
1258     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1259     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1260     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1261     uint32_t qentries = 1 << (qsize + 10);
1262 
1263     uint32_t nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1264     uint32_t nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1265     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1266     uint8_t pq;
1267 
1268     if (!xive_end_is_valid(end)) {
1269         return;
1270     }
1271 
1272     pq = xive_get_field32(END_W1_ESn, end->w1);
1273 
1274     monitor_printf(mon, "  %08x %c%c %c%c%c%c%c%c%c prio:%d nvt:%02x/%04x",
1275                    end_idx,
1276                    pq & XIVE_ESB_VAL_P ? 'P' : '-',
1277                    pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1278                    xive_end_is_valid(end)    ? 'v' : '-',
1279                    xive_end_is_enqueue(end)  ? 'q' : '-',
1280                    xive_end_is_notify(end)   ? 'n' : '-',
1281                    xive_end_is_backlog(end)  ? 'b' : '-',
1282                    xive_end_is_escalate(end) ? 'e' : '-',
1283                    xive_end_is_uncond_escalation(end)   ? 'u' : '-',
1284                    xive_end_is_silent_escalation(end)   ? 's' : '-',
1285                    priority, nvt_blk, nvt_idx);
1286 
1287     if (qaddr_base) {
1288         monitor_printf(mon, " eq:@%08"PRIx64"% 6d/%5d ^%d",
1289                        qaddr_base, qindex, qentries, qgen);
1290         xive_end_queue_pic_print_info(end, 6, mon);
1291     }
1292     monitor_printf(mon, "\n");
1293 }
1294 
1295 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1296 {
1297     uint64_t qaddr_base = xive_end_qaddr(end);
1298     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1299     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1300     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1301 
1302     uint64_t qaddr = qaddr_base + (qindex << 2);
1303     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1304     uint32_t qentries = 1 << (qsize + 10);
1305 
1306     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata))) {
1307         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1308                       HWADDR_PRIx "\n", qaddr);
1309         return;
1310     }
1311 
1312     qindex = (qindex + 1) & (qentries - 1);
1313     if (qindex == 0) {
1314         qgen ^= 1;
1315         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1316     }
1317     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1318 }
1319 
1320 void xive_end_eas_pic_print_info(XiveEND *end, uint32_t end_idx,
1321                                    Monitor *mon)
1322 {
1323     XiveEAS *eas = (XiveEAS *) &end->w4;
1324     uint8_t pq;
1325 
1326     if (!xive_end_is_escalate(end)) {
1327         return;
1328     }
1329 
1330     pq = xive_get_field32(END_W1_ESe, end->w1);
1331 
1332     monitor_printf(mon, "  %08x %c%c %c%c end:%02x/%04x data:%08x\n",
1333                    end_idx,
1334                    pq & XIVE_ESB_VAL_P ? 'P' : '-',
1335                    pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1336                    xive_eas_is_valid(eas) ? 'V' : ' ',
1337                    xive_eas_is_masked(eas) ? 'M' : ' ',
1338                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1339                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1340                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1341 }
1342 
1343 /*
1344  * XIVE Router (aka. Virtualization Controller or IVRE)
1345  */
1346 
1347 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1348                         XiveEAS *eas)
1349 {
1350     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1351 
1352     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1353 }
1354 
1355 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1356                         XiveEND *end)
1357 {
1358    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1359 
1360    return xrc->get_end(xrtr, end_blk, end_idx, end);
1361 }
1362 
1363 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1364                           XiveEND *end, uint8_t word_number)
1365 {
1366    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1367 
1368    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1369 }
1370 
1371 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1372                         XiveNVT *nvt)
1373 {
1374    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1375 
1376    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1377 }
1378 
1379 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1380                         XiveNVT *nvt, uint8_t word_number)
1381 {
1382    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1383 
1384    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1385 }
1386 
1387 static int xive_router_get_block_id(XiveRouter *xrtr)
1388 {
1389    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1390 
1391    return xrc->get_block_id(xrtr);
1392 }
1393 
1394 static void xive_router_realize(DeviceState *dev, Error **errp)
1395 {
1396     XiveRouter *xrtr = XIVE_ROUTER(dev);
1397 
1398     assert(xrtr->xfb);
1399 }
1400 
1401 /*
1402  * Encode the HW CAM line in the block group mode format :
1403  *
1404  *   chip << 19 | 0000000 0 0001 thread (7Bit)
1405  */
1406 static uint32_t xive_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
1407 {
1408     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1409     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1410     uint8_t blk = xive_router_get_block_id(XIVE_ROUTER(xptr));
1411 
1412     return xive_nvt_cam_line(blk, 1 << 7 | (pir & 0x7f));
1413 }
1414 
1415 /*
1416  * The thread context register words are in big-endian format.
1417  */
1418 int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1419                               uint8_t format,
1420                               uint8_t nvt_blk, uint32_t nvt_idx,
1421                               bool cam_ignore, uint32_t logic_serv)
1422 {
1423     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1424     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1425     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1426     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1427     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1428 
1429     /*
1430      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1431      * identifier are ignored in the "CAM" match.
1432      */
1433 
1434     if (format == 0) {
1435         if (cam_ignore == true) {
1436             /*
1437              * F=0 & i=1: Logical server notification (bits ignored at
1438              * the end of the NVT identifier)
1439              */
1440             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1441                           nvt_blk, nvt_idx);
1442              return -1;
1443         }
1444 
1445         /* F=0 & i=0: Specific NVT notification */
1446 
1447         /* PHYS ring */
1448         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1449             cam == xive_tctx_hw_cam_line(xptr, tctx)) {
1450             return TM_QW3_HV_PHYS;
1451         }
1452 
1453         /* HV POOL ring */
1454         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1455             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1456             return TM_QW2_HV_POOL;
1457         }
1458 
1459         /* OS ring */
1460         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1461             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1462             return TM_QW1_OS;
1463         }
1464     } else {
1465         /* F=1 : User level Event-Based Branch (EBB) notification */
1466 
1467         /* USER ring */
1468         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1469              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1470              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1471              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1472             return TM_QW0_USER;
1473         }
1474     }
1475     return -1;
1476 }
1477 
1478 /*
1479  * This is our simple Xive Presenter Engine model. It is merged in the
1480  * Router as it does not require an extra object.
1481  *
1482  * It receives notification requests sent by the IVRE to find one
1483  * matching NVT (or more) dispatched on the processor threads. In case
1484  * of a single NVT notification, the process is abreviated and the
1485  * thread is signaled if a match is found. In case of a logical server
1486  * notification (bits ignored at the end of the NVT identifier), the
1487  * IVPE and IVRE select a winning thread using different filters. This
1488  * involves 2 or 3 exchanges on the PowerBus that the model does not
1489  * support.
1490  *
1491  * The parameters represent what is sent on the PowerBus
1492  */
1493 static bool xive_presenter_notify(XiveFabric *xfb, uint8_t format,
1494                                   uint8_t nvt_blk, uint32_t nvt_idx,
1495                                   bool cam_ignore, uint8_t priority,
1496                                   uint32_t logic_serv)
1497 {
1498     XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xfb);
1499     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1500     int count;
1501 
1502     /*
1503      * Ask the machine to scan the interrupt controllers for a match
1504      */
1505     count = xfc->match_nvt(xfb, format, nvt_blk, nvt_idx, cam_ignore,
1506                            priority, logic_serv, &match);
1507     if (count < 0) {
1508         return false;
1509     }
1510 
1511     /* handle CPU exception delivery */
1512     if (count) {
1513         xive_tctx_ipb_update(match.tctx, match.ring, priority_to_ipb(priority));
1514     }
1515 
1516     return !!count;
1517 }
1518 
1519 /*
1520  * Notification using the END ESe/ESn bit (Event State Buffer for
1521  * escalation and notification). Provide further coalescing in the
1522  * Router.
1523  */
1524 static bool xive_router_end_es_notify(XiveRouter *xrtr, uint8_t end_blk,
1525                                       uint32_t end_idx, XiveEND *end,
1526                                       uint32_t end_esmask)
1527 {
1528     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1529     bool notify = xive_esb_trigger(&pq);
1530 
1531     if (pq != xive_get_field32(end_esmask, end->w1)) {
1532         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1533         xive_router_write_end(xrtr, end_blk, end_idx, end, 1);
1534     }
1535 
1536     /* ESe/n[Q]=1 : end of notification */
1537     return notify;
1538 }
1539 
1540 /*
1541  * An END trigger can come from an event trigger (IPI or HW) or from
1542  * another chip. We don't model the PowerBus but the END trigger
1543  * message has the same parameters than in the function below.
1544  */
1545 static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1546                                    uint32_t end_idx, uint32_t end_data)
1547 {
1548     XiveEND end;
1549     uint8_t priority;
1550     uint8_t format;
1551     uint8_t nvt_blk;
1552     uint32_t nvt_idx;
1553     XiveNVT nvt;
1554     bool found;
1555 
1556     /* END cache lookup */
1557     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1558         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1559                       end_idx);
1560         return;
1561     }
1562 
1563     if (!xive_end_is_valid(&end)) {
1564         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1565                       end_blk, end_idx);
1566         return;
1567     }
1568 
1569     if (xive_end_is_enqueue(&end)) {
1570         xive_end_enqueue(&end, end_data);
1571         /* Enqueuing event data modifies the EQ toggle and index */
1572         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1573     }
1574 
1575     /*
1576      * When the END is silent, we skip the notification part.
1577      */
1578     if (xive_end_is_silent_escalation(&end)) {
1579         goto do_escalation;
1580     }
1581 
1582     /*
1583      * The W7 format depends on the F bit in W6. It defines the type
1584      * of the notification :
1585      *
1586      *   F=0 : single or multiple NVT notification
1587      *   F=1 : User level Event-Based Branch (EBB) notification, no
1588      *         priority
1589      */
1590     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1591     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1592 
1593     /* The END is masked */
1594     if (format == 0 && priority == 0xff) {
1595         return;
1596     }
1597 
1598     /*
1599      * Check the END ESn (Event State Buffer for notification) for
1600      * even further coalescing in the Router
1601      */
1602     if (!xive_end_is_notify(&end)) {
1603         /* ESn[Q]=1 : end of notification */
1604         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1605                                        &end, END_W1_ESn)) {
1606             return;
1607         }
1608     }
1609 
1610     /*
1611      * Follows IVPE notification
1612      */
1613     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end.w6);
1614     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end.w6);
1615 
1616     /* NVT cache lookup */
1617     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1618         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1619                       nvt_blk, nvt_idx);
1620         return;
1621     }
1622 
1623     if (!xive_nvt_is_valid(&nvt)) {
1624         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1625                       nvt_blk, nvt_idx);
1626         return;
1627     }
1628 
1629     found = xive_presenter_notify(xrtr->xfb, format, nvt_blk, nvt_idx,
1630                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1631                           priority,
1632                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1633 
1634     /* TODO: Auto EOI. */
1635 
1636     if (found) {
1637         return;
1638     }
1639 
1640     /*
1641      * If no matching NVT is dispatched on a HW thread :
1642      * - specific VP: update the NVT structure if backlog is activated
1643      * - logical server : forward request to IVPE (not supported)
1644      */
1645     if (xive_end_is_backlog(&end)) {
1646         uint8_t ipb;
1647 
1648         if (format == 1) {
1649             qemu_log_mask(LOG_GUEST_ERROR,
1650                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1651                           end_blk, end_idx);
1652             return;
1653         }
1654         /*
1655          * Record the IPB in the associated NVT structure for later
1656          * use. The presenter will resend the interrupt when the vCPU
1657          * is dispatched again on a HW thread.
1658          */
1659         ipb = xive_get_field32(NVT_W4_IPB, nvt.w4) | priority_to_ipb(priority);
1660         nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, ipb);
1661         xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1662 
1663         /*
1664          * On HW, follows a "Broadcast Backlog" to IVPEs
1665          */
1666     }
1667 
1668 do_escalation:
1669     /*
1670      * If activated, escalate notification using the ESe PQ bits and
1671      * the EAS in w4-5
1672      */
1673     if (!xive_end_is_escalate(&end)) {
1674         return;
1675     }
1676 
1677     /*
1678      * Check the END ESe (Event State Buffer for escalation) for even
1679      * further coalescing in the Router
1680      */
1681     if (!xive_end_is_uncond_escalation(&end)) {
1682         /* ESe[Q]=1 : end of notification */
1683         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1684                                        &end, END_W1_ESe)) {
1685             return;
1686         }
1687     }
1688 
1689     /*
1690      * The END trigger becomes an Escalation trigger
1691      */
1692     xive_router_end_notify(xrtr,
1693                            xive_get_field32(END_W4_ESC_END_BLOCK, end.w4),
1694                            xive_get_field32(END_W4_ESC_END_INDEX, end.w4),
1695                            xive_get_field32(END_W5_ESC_END_DATA,  end.w5));
1696 }
1697 
1698 void xive_router_notify(XiveNotifier *xn, uint32_t lisn)
1699 {
1700     XiveRouter *xrtr = XIVE_ROUTER(xn);
1701     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1702     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1703     XiveEAS eas;
1704 
1705     /* EAS cache lookup */
1706     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1707         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1708         return;
1709     }
1710 
1711     /*
1712      * The IVRE checks the State Bit Cache at this point. We skip the
1713      * SBC lookup because the state bits of the sources are modeled
1714      * internally in QEMU.
1715      */
1716 
1717     if (!xive_eas_is_valid(&eas)) {
1718         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1719         return;
1720     }
1721 
1722     if (xive_eas_is_masked(&eas)) {
1723         /* Notification completed */
1724         return;
1725     }
1726 
1727     /*
1728      * The event trigger becomes an END trigger
1729      */
1730     xive_router_end_notify(xrtr,
1731                            xive_get_field64(EAS_END_BLOCK, eas.w),
1732                            xive_get_field64(EAS_END_INDEX, eas.w),
1733                            xive_get_field64(EAS_END_DATA,  eas.w));
1734 }
1735 
1736 static Property xive_router_properties[] = {
1737     DEFINE_PROP_LINK("xive-fabric", XiveRouter, xfb,
1738                      TYPE_XIVE_FABRIC, XiveFabric *),
1739     DEFINE_PROP_END_OF_LIST(),
1740 };
1741 
1742 static void xive_router_class_init(ObjectClass *klass, void *data)
1743 {
1744     DeviceClass *dc = DEVICE_CLASS(klass);
1745     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1746 
1747     dc->desc    = "XIVE Router Engine";
1748     device_class_set_props(dc, xive_router_properties);
1749     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1750     dc->realize = xive_router_realize;
1751     xnc->notify = xive_router_notify;
1752 }
1753 
1754 static const TypeInfo xive_router_info = {
1755     .name          = TYPE_XIVE_ROUTER,
1756     .parent        = TYPE_SYS_BUS_DEVICE,
1757     .abstract      = true,
1758     .instance_size = sizeof(XiveRouter),
1759     .class_size    = sizeof(XiveRouterClass),
1760     .class_init    = xive_router_class_init,
1761     .interfaces    = (InterfaceInfo[]) {
1762         { TYPE_XIVE_NOTIFIER },
1763         { TYPE_XIVE_PRESENTER },
1764         { }
1765     }
1766 };
1767 
1768 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1769 {
1770     if (!xive_eas_is_valid(eas)) {
1771         return;
1772     }
1773 
1774     monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1775                    lisn, xive_eas_is_masked(eas) ? "M" : " ",
1776                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1777                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1778                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1779 }
1780 
1781 /*
1782  * END ESB MMIO loads
1783  */
1784 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1785 {
1786     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1787     uint32_t offset = addr & 0xFFF;
1788     uint8_t end_blk;
1789     uint32_t end_idx;
1790     XiveEND end;
1791     uint32_t end_esmask;
1792     uint8_t pq;
1793     uint64_t ret = -1;
1794 
1795     /*
1796      * The block id should be deduced from the load address on the END
1797      * ESB MMIO but our model only supports a single block per XIVE chip.
1798      */
1799     end_blk = xive_router_get_block_id(xsrc->xrtr);
1800     end_idx = addr >> (xsrc->esb_shift + 1);
1801 
1802     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1803         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1804                       end_idx);
1805         return -1;
1806     }
1807 
1808     if (!xive_end_is_valid(&end)) {
1809         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1810                       end_blk, end_idx);
1811         return -1;
1812     }
1813 
1814     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1815     pq = xive_get_field32(end_esmask, end.w1);
1816 
1817     switch (offset) {
1818     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1819         ret = xive_esb_eoi(&pq);
1820 
1821         /* Forward the source event notification for routing ?? */
1822         break;
1823 
1824     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1825         ret = pq;
1826         break;
1827 
1828     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1829     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1830     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1831     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1832         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1833         break;
1834     default:
1835         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1836                       offset);
1837         return -1;
1838     }
1839 
1840     if (pq != xive_get_field32(end_esmask, end.w1)) {
1841         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1842         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1843     }
1844 
1845     return ret;
1846 }
1847 
1848 /*
1849  * END ESB MMIO stores are invalid
1850  */
1851 static void xive_end_source_write(void *opaque, hwaddr addr,
1852                                   uint64_t value, unsigned size)
1853 {
1854     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1855                   HWADDR_PRIx"\n", addr);
1856 }
1857 
1858 static const MemoryRegionOps xive_end_source_ops = {
1859     .read = xive_end_source_read,
1860     .write = xive_end_source_write,
1861     .endianness = DEVICE_BIG_ENDIAN,
1862     .valid = {
1863         .min_access_size = 8,
1864         .max_access_size = 8,
1865     },
1866     .impl = {
1867         .min_access_size = 8,
1868         .max_access_size = 8,
1869     },
1870 };
1871 
1872 static void xive_end_source_realize(DeviceState *dev, Error **errp)
1873 {
1874     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1875 
1876     assert(xsrc->xrtr);
1877 
1878     if (!xsrc->nr_ends) {
1879         error_setg(errp, "Number of interrupt needs to be greater than 0");
1880         return;
1881     }
1882 
1883     if (xsrc->esb_shift != XIVE_ESB_4K &&
1884         xsrc->esb_shift != XIVE_ESB_64K) {
1885         error_setg(errp, "Invalid ESB shift setting");
1886         return;
1887     }
1888 
1889     /*
1890      * Each END is assigned an even/odd pair of MMIO pages, the even page
1891      * manages the ESn field while the odd page manages the ESe field.
1892      */
1893     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1894                           &xive_end_source_ops, xsrc, "xive.end",
1895                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1896 }
1897 
1898 static Property xive_end_source_properties[] = {
1899     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
1900     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
1901     DEFINE_PROP_LINK("xive", XiveENDSource, xrtr, TYPE_XIVE_ROUTER,
1902                      XiveRouter *),
1903     DEFINE_PROP_END_OF_LIST(),
1904 };
1905 
1906 static void xive_end_source_class_init(ObjectClass *klass, void *data)
1907 {
1908     DeviceClass *dc = DEVICE_CLASS(klass);
1909 
1910     dc->desc    = "XIVE END Source";
1911     device_class_set_props(dc, xive_end_source_properties);
1912     dc->realize = xive_end_source_realize;
1913     /*
1914      * Reason: part of XIVE interrupt controller, needs to be wired up,
1915      * e.g. by spapr_xive_instance_init().
1916      */
1917     dc->user_creatable = false;
1918 }
1919 
1920 static const TypeInfo xive_end_source_info = {
1921     .name          = TYPE_XIVE_END_SOURCE,
1922     .parent        = TYPE_DEVICE,
1923     .instance_size = sizeof(XiveENDSource),
1924     .class_init    = xive_end_source_class_init,
1925 };
1926 
1927 /*
1928  * XIVE Notifier
1929  */
1930 static const TypeInfo xive_notifier_info = {
1931     .name = TYPE_XIVE_NOTIFIER,
1932     .parent = TYPE_INTERFACE,
1933     .class_size = sizeof(XiveNotifierClass),
1934 };
1935 
1936 /*
1937  * XIVE Presenter
1938  */
1939 static const TypeInfo xive_presenter_info = {
1940     .name = TYPE_XIVE_PRESENTER,
1941     .parent = TYPE_INTERFACE,
1942     .class_size = sizeof(XivePresenterClass),
1943 };
1944 
1945 /*
1946  * XIVE Fabric
1947  */
1948 static const TypeInfo xive_fabric_info = {
1949     .name = TYPE_XIVE_FABRIC,
1950     .parent = TYPE_INTERFACE,
1951     .class_size = sizeof(XiveFabricClass),
1952 };
1953 
1954 static void xive_register_types(void)
1955 {
1956     type_register_static(&xive_fabric_info);
1957     type_register_static(&xive_source_info);
1958     type_register_static(&xive_notifier_info);
1959     type_register_static(&xive_presenter_info);
1960     type_register_static(&xive_router_info);
1961     type_register_static(&xive_end_source_info);
1962     type_register_static(&xive_tctx_info);
1963 }
1964 
1965 type_init(xive_register_types)
1966