xref: /openbmc/qemu/hw/intc/xive.c (revision a6896ebc)
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 "hw/irq.h"
21 #include "hw/ppc/xive.h"
22 #include "hw/ppc/xive2.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 an Interrupt Pending Buffer (IPB) register to a Pending
32  * Interrupt Priority Register (PIPR), which contains the priority of
33  * the most favored pending notification.
34  */
35 static uint8_t ipb_to_pipr(uint8_t ibp)
36 {
37     return ibp ? clz32((uint32_t)ibp << 24) : 0xff;
38 }
39 
40 static uint8_t exception_mask(uint8_t ring)
41 {
42     switch (ring) {
43     case TM_QW1_OS:
44         return TM_QW1_NSR_EO;
45     case TM_QW3_HV_PHYS:
46         return TM_QW3_NSR_HE;
47     default:
48         g_assert_not_reached();
49     }
50 }
51 
52 static qemu_irq xive_tctx_output(XiveTCTX *tctx, uint8_t ring)
53 {
54         switch (ring) {
55         case TM_QW0_USER:
56                 return 0; /* Not supported */
57         case TM_QW1_OS:
58                 return tctx->os_output;
59         case TM_QW2_HV_POOL:
60         case TM_QW3_HV_PHYS:
61                 return tctx->hv_output;
62         default:
63                 return 0;
64         }
65 }
66 
67 static uint64_t xive_tctx_accept(XiveTCTX *tctx, uint8_t ring)
68 {
69     uint8_t *regs = &tctx->regs[ring];
70     uint8_t nsr = regs[TM_NSR];
71     uint8_t mask = exception_mask(ring);
72 
73     qemu_irq_lower(xive_tctx_output(tctx, ring));
74 
75     if (regs[TM_NSR] & mask) {
76         uint8_t cppr = regs[TM_PIPR];
77 
78         regs[TM_CPPR] = cppr;
79 
80         /* Reset the pending buffer bit */
81         regs[TM_IPB] &= ~xive_priority_to_ipb(cppr);
82         regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
83 
84         /* Drop Exception bit */
85         regs[TM_NSR] &= ~mask;
86 
87         trace_xive_tctx_accept(tctx->cs->cpu_index, ring,
88                                regs[TM_IPB], regs[TM_PIPR],
89                                regs[TM_CPPR], regs[TM_NSR]);
90     }
91 
92     return (nsr << 8) | regs[TM_CPPR];
93 }
94 
95 static void xive_tctx_notify(XiveTCTX *tctx, uint8_t ring)
96 {
97     uint8_t *regs = &tctx->regs[ring];
98 
99     if (regs[TM_PIPR] < regs[TM_CPPR]) {
100         switch (ring) {
101         case TM_QW1_OS:
102             regs[TM_NSR] |= TM_QW1_NSR_EO;
103             break;
104         case TM_QW3_HV_PHYS:
105             regs[TM_NSR] |= (TM_QW3_NSR_HE_PHYS << 6);
106             break;
107         default:
108             g_assert_not_reached();
109         }
110         trace_xive_tctx_notify(tctx->cs->cpu_index, ring,
111                                regs[TM_IPB], regs[TM_PIPR],
112                                regs[TM_CPPR], regs[TM_NSR]);
113         qemu_irq_raise(xive_tctx_output(tctx, ring));
114     }
115 }
116 
117 void xive_tctx_reset_os_signal(XiveTCTX *tctx)
118 {
119     /*
120      * Lower the External interrupt. Used when pulling an OS
121      * context. It is necessary to avoid catching it in the hypervisor
122      * context. It should be raised again when re-pushing the OS
123      * context.
124      */
125     qemu_irq_lower(xive_tctx_output(tctx, TM_QW1_OS));
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 /*
156  * XIVE Thread Interrupt Management Area (TIMA)
157  */
158 
159 static void xive_tm_set_hv_cppr(XivePresenter *xptr, XiveTCTX *tctx,
160                                 hwaddr offset, uint64_t value, unsigned size)
161 {
162     xive_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
163 }
164 
165 static uint64_t xive_tm_ack_hv_reg(XivePresenter *xptr, XiveTCTX *tctx,
166                                    hwaddr offset, unsigned size)
167 {
168     return xive_tctx_accept(tctx, TM_QW3_HV_PHYS);
169 }
170 
171 static uint64_t xive_tm_pull_pool_ctx(XivePresenter *xptr, XiveTCTX *tctx,
172                                       hwaddr offset, unsigned size)
173 {
174     uint32_t qw2w2_prev = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
175     uint32_t qw2w2;
176 
177     qw2w2 = xive_set_field32(TM_QW2W2_VP, qw2w2_prev, 0);
178     memcpy(&tctx->regs[TM_QW2_HV_POOL + TM_WORD2], &qw2w2, 4);
179     return qw2w2;
180 }
181 
182 static void xive_tm_vt_push(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
183                             uint64_t value, unsigned size)
184 {
185     tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] = value & 0xff;
186 }
187 
188 static uint64_t xive_tm_vt_poll(XivePresenter *xptr, XiveTCTX *tctx,
189                                 hwaddr offset, unsigned size)
190 {
191     return tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] & 0xff;
192 }
193 
194 /*
195  * Define an access map for each page of the TIMA that we will use in
196  * the memory region ops to filter values when doing loads and stores
197  * of raw registers values
198  *
199  * Registers accessibility bits :
200  *
201  *    0x0 - no access
202  *    0x1 - write only
203  *    0x2 - read only
204  *    0x3 - read/write
205  */
206 
207 static const uint8_t xive_tm_hw_view[] = {
208     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
209     3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
210     0, 0, 3, 3,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
211     3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   3, 3, 3, 0, /* QW-3 PHYS */
212 };
213 
214 static const uint8_t xive_tm_hv_view[] = {
215     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
216     3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
217     0, 0, 3, 3,   0, 0, 0, 0,   0, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
218     3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   0, 0, 0, 0, /* QW-3 PHYS */
219 };
220 
221 static const uint8_t xive_tm_os_view[] = {
222     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
223     2, 3, 2, 2,   2, 2, 0, 2,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
224     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
225     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
226 };
227 
228 static const uint8_t xive_tm_user_view[] = {
229     3, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-0 User */
230     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
231     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
232     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
233 };
234 
235 /*
236  * Overall TIMA access map for the thread interrupt management context
237  * registers
238  */
239 static const uint8_t *xive_tm_views[] = {
240     [XIVE_TM_HW_PAGE]   = xive_tm_hw_view,
241     [XIVE_TM_HV_PAGE]   = xive_tm_hv_view,
242     [XIVE_TM_OS_PAGE]   = xive_tm_os_view,
243     [XIVE_TM_USER_PAGE] = xive_tm_user_view,
244 };
245 
246 /*
247  * Computes a register access mask for a given offset in the TIMA
248  */
249 static uint64_t xive_tm_mask(hwaddr offset, unsigned size, bool write)
250 {
251     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
252     uint8_t reg_offset = offset & TM_REG_OFFSET;
253     uint8_t reg_mask = write ? 0x1 : 0x2;
254     uint64_t mask = 0x0;
255     int i;
256 
257     for (i = 0; i < size; i++) {
258         if (xive_tm_views[page_offset][reg_offset + i] & reg_mask) {
259             mask |= (uint64_t) 0xff << (8 * (size - i - 1));
260         }
261     }
262 
263     return mask;
264 }
265 
266 static void xive_tm_raw_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
267                               unsigned size)
268 {
269     uint8_t ring_offset = offset & TM_RING_OFFSET;
270     uint8_t reg_offset = offset & TM_REG_OFFSET;
271     uint64_t mask = xive_tm_mask(offset, size, true);
272     int i;
273 
274     /*
275      * Only 4 or 8 bytes stores are allowed and the User ring is
276      * excluded
277      */
278     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
279         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA @%"
280                       HWADDR_PRIx"\n", offset);
281         return;
282     }
283 
284     /*
285      * Use the register offset for the raw values and filter out
286      * reserved values
287      */
288     for (i = 0; i < size; i++) {
289         uint8_t byte_mask = (mask >> (8 * (size - i - 1)));
290         if (byte_mask) {
291             tctx->regs[reg_offset + i] = (value >> (8 * (size - i - 1))) &
292                 byte_mask;
293         }
294     }
295 }
296 
297 static uint64_t xive_tm_raw_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
298 {
299     uint8_t ring_offset = offset & TM_RING_OFFSET;
300     uint8_t reg_offset = offset & TM_REG_OFFSET;
301     uint64_t mask = xive_tm_mask(offset, size, false);
302     uint64_t ret;
303     int i;
304 
305     /*
306      * Only 4 or 8 bytes loads are allowed and the User ring is
307      * excluded
308      */
309     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
310         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access at TIMA @%"
311                       HWADDR_PRIx"\n", offset);
312         return -1;
313     }
314 
315     /* Use the register offset for the raw values */
316     ret = 0;
317     for (i = 0; i < size; i++) {
318         ret |= (uint64_t) tctx->regs[reg_offset + i] << (8 * (size - i - 1));
319     }
320 
321     /* filter out reserved values */
322     return ret & mask;
323 }
324 
325 /*
326  * The TM context is mapped twice within each page. Stores and loads
327  * to the first mapping below 2K write and read the specified values
328  * without modification. The second mapping above 2K performs specific
329  * state changes (side effects) in addition to setting/returning the
330  * interrupt management area context of the processor thread.
331  */
332 static uint64_t xive_tm_ack_os_reg(XivePresenter *xptr, XiveTCTX *tctx,
333                                    hwaddr offset, unsigned size)
334 {
335     return xive_tctx_accept(tctx, TM_QW1_OS);
336 }
337 
338 static void xive_tm_set_os_cppr(XivePresenter *xptr, XiveTCTX *tctx,
339                                 hwaddr offset, uint64_t value, unsigned size)
340 {
341     xive_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
342 }
343 
344 /*
345  * Adjust the IPB to allow a CPU to process event queues of other
346  * priorities during one physical interrupt cycle.
347  */
348 static void xive_tm_set_os_pending(XivePresenter *xptr, XiveTCTX *tctx,
349                                    hwaddr offset, uint64_t value, unsigned size)
350 {
351     xive_tctx_ipb_update(tctx, TM_QW1_OS, xive_priority_to_ipb(value & 0xff));
352 }
353 
354 static void xive_os_cam_decode(uint32_t cam, uint8_t *nvt_blk,
355                                uint32_t *nvt_idx, bool *vo)
356 {
357     if (nvt_blk) {
358         *nvt_blk = xive_nvt_blk(cam);
359     }
360     if (nvt_idx) {
361         *nvt_idx = xive_nvt_idx(cam);
362     }
363     if (vo) {
364         *vo = !!(cam & TM_QW1W2_VO);
365     }
366 }
367 
368 static uint32_t xive_tctx_get_os_cam(XiveTCTX *tctx, uint8_t *nvt_blk,
369                                      uint32_t *nvt_idx, bool *vo)
370 {
371     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
372     uint32_t cam = be32_to_cpu(qw1w2);
373 
374     xive_os_cam_decode(cam, nvt_blk, nvt_idx, vo);
375     return qw1w2;
376 }
377 
378 static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t qw1w2)
379 {
380     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
381 }
382 
383 static uint64_t xive_tm_pull_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
384                                     hwaddr offset, unsigned size)
385 {
386     uint32_t qw1w2;
387     uint32_t qw1w2_new;
388     uint8_t nvt_blk;
389     uint32_t nvt_idx;
390     bool vo;
391 
392     qw1w2 = xive_tctx_get_os_cam(tctx, &nvt_blk, &nvt_idx, &vo);
393 
394     if (!vo) {
395         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: pulling invalid NVT %x/%x !?\n",
396                       nvt_blk, nvt_idx);
397     }
398 
399     /* Invalidate CAM line */
400     qw1w2_new = xive_set_field32(TM_QW1W2_VO, qw1w2, 0);
401     xive_tctx_set_os_cam(tctx, qw1w2_new);
402 
403     xive_tctx_reset_os_signal(tctx);
404     return qw1w2;
405 }
406 
407 static void xive_tctx_need_resend(XiveRouter *xrtr, XiveTCTX *tctx,
408                                   uint8_t nvt_blk, uint32_t nvt_idx)
409 {
410     XiveNVT nvt;
411     uint8_t ipb;
412 
413     /*
414      * Grab the associated NVT to pull the pending bits, and merge
415      * them with the IPB of the thread interrupt context registers
416      */
417     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
418         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVT %x/%x\n",
419                           nvt_blk, nvt_idx);
420         return;
421     }
422 
423     ipb = xive_get_field32(NVT_W4_IPB, nvt.w4);
424 
425     if (ipb) {
426         /* Reset the NVT value */
427         nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, 0);
428         xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
429     }
430     /*
431      * Always call xive_tctx_ipb_update(). Even if there were no
432      * escalation triggered, there could be a pending interrupt which
433      * was saved when the context was pulled and that we need to take
434      * into account by recalculating the PIPR (which is not
435      * saved/restored).
436      * It will also raise the External interrupt signal if needed.
437      */
438     xive_tctx_ipb_update(tctx, TM_QW1_OS, ipb);
439 }
440 
441 /*
442  * Updating the OS CAM line can trigger a resend of interrupt
443  */
444 static void xive_tm_push_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
445                                 hwaddr offset, uint64_t value, unsigned size)
446 {
447     uint32_t cam = value;
448     uint32_t qw1w2 = cpu_to_be32(cam);
449     uint8_t nvt_blk;
450     uint32_t nvt_idx;
451     bool vo;
452 
453     xive_os_cam_decode(cam, &nvt_blk, &nvt_idx, &vo);
454 
455     /* First update the registers */
456     xive_tctx_set_os_cam(tctx, qw1w2);
457 
458     /* Check the interrupt pending bits */
459     if (vo) {
460         xive_tctx_need_resend(XIVE_ROUTER(xptr), tctx, nvt_blk, nvt_idx);
461     }
462 }
463 
464 static uint32_t xive_presenter_get_config(XivePresenter *xptr)
465 {
466     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
467 
468     return xpc->get_config(xptr);
469 }
470 
471 /*
472  * Define a mapping of "special" operations depending on the TIMA page
473  * offset and the size of the operation.
474  */
475 typedef struct XiveTmOp {
476     uint8_t  page_offset;
477     uint32_t op_offset;
478     unsigned size;
479     void     (*write_handler)(XivePresenter *xptr, XiveTCTX *tctx,
480                               hwaddr offset,
481                               uint64_t value, unsigned size);
482     uint64_t (*read_handler)(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
483                              unsigned size);
484 } XiveTmOp;
485 
486 static const XiveTmOp xive_tm_operations[] = {
487     /*
488      * MMIOs below 2K : raw values and special operations without side
489      * effects
490      */
491     { XIVE_TM_OS_PAGE, TM_QW1_OS + TM_CPPR,   1, xive_tm_set_os_cppr, NULL },
492     { XIVE_TM_HV_PAGE, TM_QW1_OS + TM_WORD2,     4, xive_tm_push_os_ctx, NULL },
493     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_CPPR, 1, xive_tm_set_hv_cppr, NULL },
494     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, xive_tm_vt_push, NULL },
495     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, NULL, xive_tm_vt_poll },
496 
497     /* MMIOs above 2K : special operations with side effects */
498     { XIVE_TM_OS_PAGE, TM_SPC_ACK_OS_REG,     2, NULL, xive_tm_ack_os_reg },
499     { XIVE_TM_OS_PAGE, TM_SPC_SET_OS_PENDING, 1, xive_tm_set_os_pending, NULL },
500     { XIVE_TM_HV_PAGE, TM_SPC_PULL_OS_CTX,    4, NULL, xive_tm_pull_os_ctx },
501     { XIVE_TM_HV_PAGE, TM_SPC_PULL_OS_CTX,    8, NULL, xive_tm_pull_os_ctx },
502     { XIVE_TM_HV_PAGE, TM_SPC_ACK_HV_REG,     2, NULL, xive_tm_ack_hv_reg },
503     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  4, NULL, xive_tm_pull_pool_ctx },
504     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  8, NULL, xive_tm_pull_pool_ctx },
505 };
506 
507 static const XiveTmOp xive2_tm_operations[] = {
508     /*
509      * MMIOs below 2K : raw values and special operations without side
510      * effects
511      */
512     { XIVE_TM_OS_PAGE, TM_QW1_OS + TM_CPPR,   1, xive_tm_set_os_cppr, NULL },
513     { XIVE_TM_HV_PAGE, TM_QW1_OS + TM_WORD2,  4, xive2_tm_push_os_ctx, NULL },
514     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_CPPR, 1, xive_tm_set_hv_cppr, NULL },
515     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, xive_tm_vt_push, NULL },
516     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, NULL, xive_tm_vt_poll },
517 
518     /* MMIOs above 2K : special operations with side effects */
519     { XIVE_TM_OS_PAGE, TM_SPC_ACK_OS_REG,     2, NULL, xive_tm_ack_os_reg },
520     { XIVE_TM_OS_PAGE, TM_SPC_SET_OS_PENDING, 1, xive_tm_set_os_pending, NULL },
521     { XIVE_TM_HV_PAGE, TM_SPC_PULL_OS_CTX,    4, NULL, xive2_tm_pull_os_ctx },
522     { XIVE_TM_HV_PAGE, TM_SPC_PULL_OS_CTX,    8, NULL, xive2_tm_pull_os_ctx },
523     { XIVE_TM_HV_PAGE, TM_SPC_ACK_HV_REG,     2, NULL, xive_tm_ack_hv_reg },
524     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  4, NULL, xive_tm_pull_pool_ctx },
525     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  8, NULL, xive_tm_pull_pool_ctx },
526 };
527 
528 static const XiveTmOp *xive_tm_find_op(XivePresenter *xptr, hwaddr offset,
529                                        unsigned size, bool write)
530 {
531     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
532     uint32_t op_offset = offset & TM_ADDRESS_MASK;
533     const XiveTmOp *tm_ops;
534     int i, tm_ops_count;
535     uint32_t cfg;
536 
537     cfg = xive_presenter_get_config(xptr);
538     if (cfg & XIVE_PRESENTER_GEN1_TIMA_OS) {
539         tm_ops = xive_tm_operations;
540         tm_ops_count = ARRAY_SIZE(xive_tm_operations);
541     } else {
542         tm_ops = xive2_tm_operations;
543         tm_ops_count = ARRAY_SIZE(xive2_tm_operations);
544     }
545 
546     for (i = 0; i < tm_ops_count; i++) {
547         const XiveTmOp *xto = &tm_ops[i];
548 
549         /* Accesses done from a more privileged TIMA page is allowed */
550         if (xto->page_offset >= page_offset &&
551             xto->op_offset == op_offset &&
552             xto->size == size &&
553             ((write && xto->write_handler) || (!write && xto->read_handler))) {
554             return xto;
555         }
556     }
557     return NULL;
558 }
559 
560 /*
561  * TIMA MMIO handlers
562  */
563 void xive_tctx_tm_write(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
564                         uint64_t value, unsigned size)
565 {
566     const XiveTmOp *xto;
567 
568     trace_xive_tctx_tm_write(tctx->cs->cpu_index, offset, size, value);
569 
570     /*
571      * TODO: check V bit in Q[0-3]W2
572      */
573 
574     /*
575      * First, check for special operations in the 2K region
576      */
577     if (offset & TM_SPECIAL_OP) {
578         xto = xive_tm_find_op(tctx->xptr, offset, size, true);
579         if (!xto) {
580             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA "
581                           "@%"HWADDR_PRIx"\n", offset);
582         } else {
583             xto->write_handler(xptr, tctx, offset, value, size);
584         }
585         return;
586     }
587 
588     /*
589      * Then, for special operations in the region below 2K.
590      */
591     xto = xive_tm_find_op(tctx->xptr, offset, size, true);
592     if (xto) {
593         xto->write_handler(xptr, tctx, offset, value, size);
594         return;
595     }
596 
597     /*
598      * Finish with raw access to the register values
599      */
600     xive_tm_raw_write(tctx, offset, value, size);
601 }
602 
603 uint64_t xive_tctx_tm_read(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
604                            unsigned size)
605 {
606     const XiveTmOp *xto;
607     uint64_t ret;
608 
609     /*
610      * TODO: check V bit in Q[0-3]W2
611      */
612 
613     /*
614      * First, check for special operations in the 2K region
615      */
616     if (offset & TM_SPECIAL_OP) {
617         xto = xive_tm_find_op(tctx->xptr, offset, size, false);
618         if (!xto) {
619             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access to TIMA"
620                           "@%"HWADDR_PRIx"\n", offset);
621             return -1;
622         }
623         ret = xto->read_handler(xptr, tctx, offset, size);
624         goto out;
625     }
626 
627     /*
628      * Then, for special operations in the region below 2K.
629      */
630     xto = xive_tm_find_op(tctx->xptr, offset, size, false);
631     if (xto) {
632         ret = xto->read_handler(xptr, tctx, offset, size);
633         goto out;
634     }
635 
636     /*
637      * Finish with raw access to the register values
638      */
639     ret = xive_tm_raw_read(tctx, offset, size);
640 out:
641     trace_xive_tctx_tm_read(tctx->cs->cpu_index, offset, size, ret);
642     return ret;
643 }
644 
645 static char *xive_tctx_ring_print(uint8_t *ring)
646 {
647     uint32_t w2 = xive_tctx_word2(ring);
648 
649     return g_strdup_printf("%02x   %02x  %02x    %02x   %02x  "
650                    "%02x  %02x   %02x  %08x",
651                    ring[TM_NSR], ring[TM_CPPR], ring[TM_IPB], ring[TM_LSMFB],
652                    ring[TM_ACK_CNT], ring[TM_INC], ring[TM_AGE], ring[TM_PIPR],
653                    be32_to_cpu(w2));
654 }
655 
656 static const char * const xive_tctx_ring_names[] = {
657     "USER", "OS", "POOL", "PHYS",
658 };
659 
660 /*
661  * kvm_irqchip_in_kernel() will cause the compiler to turn this
662  * info a nop if CONFIG_KVM isn't defined.
663  */
664 #define xive_in_kernel(xptr)                                            \
665     (kvm_irqchip_in_kernel() &&                                         \
666      ({                                                                 \
667          XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);      \
668          xpc->in_kernel ? xpc->in_kernel(xptr) : false;                 \
669      }))
670 
671 void xive_tctx_pic_print_info(XiveTCTX *tctx, GString *buf)
672 {
673     int cpu_index;
674     int i;
675 
676     /* Skip partially initialized vCPUs. This can happen on sPAPR when vCPUs
677      * are hot plugged or unplugged.
678      */
679     if (!tctx) {
680         return;
681     }
682 
683     cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
684 
685     if (xive_in_kernel(tctx->xptr)) {
686         Error *local_err = NULL;
687 
688         kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
689         if (local_err) {
690             error_report_err(local_err);
691             return;
692         }
693     }
694 
695     if (xive_presenter_get_config(tctx->xptr) & XIVE_PRESENTER_GEN1_TIMA_OS) {
696         g_string_append_printf(buf, "CPU[%04x]:   "
697                                "QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
698                                "  W2\n", cpu_index);
699     } else {
700         g_string_append_printf(buf, "CPU[%04x]:   "
701                                "QW   NSR CPPR IPB LSMFB   -  LGS  T  PIPR"
702                                "  W2\n", cpu_index);
703     }
704 
705     for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
706         char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
707         g_string_append_printf(buf, "CPU[%04x]: %4s    %s\n",
708                                cpu_index, xive_tctx_ring_names[i], s);
709         g_free(s);
710     }
711 }
712 
713 void xive_tctx_reset(XiveTCTX *tctx)
714 {
715     memset(tctx->regs, 0, sizeof(tctx->regs));
716 
717     /* Set some defaults */
718     tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
719     tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
720     tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
721 
722     /*
723      * Initialize PIPR to 0xFF to avoid phantom interrupts when the
724      * CPPR is first set.
725      */
726     tctx->regs[TM_QW1_OS + TM_PIPR] =
727         ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
728     tctx->regs[TM_QW3_HV_PHYS + TM_PIPR] =
729         ipb_to_pipr(tctx->regs[TM_QW3_HV_PHYS + TM_IPB]);
730 }
731 
732 static void xive_tctx_realize(DeviceState *dev, Error **errp)
733 {
734     XiveTCTX *tctx = XIVE_TCTX(dev);
735     PowerPCCPU *cpu;
736     CPUPPCState *env;
737 
738     assert(tctx->cs);
739     assert(tctx->xptr);
740 
741     cpu = POWERPC_CPU(tctx->cs);
742     env = &cpu->env;
743     switch (PPC_INPUT(env)) {
744     case PPC_FLAGS_INPUT_POWER9:
745         tctx->hv_output = qdev_get_gpio_in(DEVICE(cpu), POWER9_INPUT_HINT);
746         tctx->os_output = qdev_get_gpio_in(DEVICE(cpu), POWER9_INPUT_INT);
747         break;
748 
749     default:
750         error_setg(errp, "XIVE interrupt controller does not support "
751                    "this CPU bus model");
752         return;
753     }
754 
755     /* Connect the presenter to the VCPU (required for CPU hotplug) */
756     if (xive_in_kernel(tctx->xptr)) {
757         if (kvmppc_xive_cpu_connect(tctx, errp) < 0) {
758             return;
759         }
760     }
761 }
762 
763 static int vmstate_xive_tctx_pre_save(void *opaque)
764 {
765     XiveTCTX *tctx = XIVE_TCTX(opaque);
766     Error *local_err = NULL;
767     int ret;
768 
769     if (xive_in_kernel(tctx->xptr)) {
770         ret = kvmppc_xive_cpu_get_state(tctx, &local_err);
771         if (ret < 0) {
772             error_report_err(local_err);
773             return ret;
774         }
775     }
776 
777     return 0;
778 }
779 
780 static int vmstate_xive_tctx_post_load(void *opaque, int version_id)
781 {
782     XiveTCTX *tctx = XIVE_TCTX(opaque);
783     Error *local_err = NULL;
784     int ret;
785 
786     if (xive_in_kernel(tctx->xptr)) {
787         /*
788          * Required for hotplugged CPU, for which the state comes
789          * after all states of the machine.
790          */
791         ret = kvmppc_xive_cpu_set_state(tctx, &local_err);
792         if (ret < 0) {
793             error_report_err(local_err);
794             return ret;
795         }
796     }
797 
798     return 0;
799 }
800 
801 static const VMStateDescription vmstate_xive_tctx = {
802     .name = TYPE_XIVE_TCTX,
803     .version_id = 1,
804     .minimum_version_id = 1,
805     .pre_save = vmstate_xive_tctx_pre_save,
806     .post_load = vmstate_xive_tctx_post_load,
807     .fields = (const VMStateField[]) {
808         VMSTATE_BUFFER(regs, XiveTCTX),
809         VMSTATE_END_OF_LIST()
810     },
811 };
812 
813 static Property xive_tctx_properties[] = {
814     DEFINE_PROP_LINK("cpu", XiveTCTX, cs, TYPE_CPU, CPUState *),
815     DEFINE_PROP_LINK("presenter", XiveTCTX, xptr, TYPE_XIVE_PRESENTER,
816                      XivePresenter *),
817     DEFINE_PROP_END_OF_LIST(),
818 };
819 
820 static void xive_tctx_class_init(ObjectClass *klass, void *data)
821 {
822     DeviceClass *dc = DEVICE_CLASS(klass);
823 
824     dc->desc = "XIVE Interrupt Thread Context";
825     dc->realize = xive_tctx_realize;
826     dc->vmsd = &vmstate_xive_tctx;
827     device_class_set_props(dc, xive_tctx_properties);
828     /*
829      * Reason: part of XIVE interrupt controller, needs to be wired up
830      * by xive_tctx_create().
831      */
832     dc->user_creatable = false;
833 }
834 
835 static const TypeInfo xive_tctx_info = {
836     .name          = TYPE_XIVE_TCTX,
837     .parent        = TYPE_DEVICE,
838     .instance_size = sizeof(XiveTCTX),
839     .class_init    = xive_tctx_class_init,
840 };
841 
842 Object *xive_tctx_create(Object *cpu, XivePresenter *xptr, Error **errp)
843 {
844     Object *obj;
845 
846     obj = object_new(TYPE_XIVE_TCTX);
847     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj);
848     object_unref(obj);
849     object_property_set_link(obj, "cpu", cpu, &error_abort);
850     object_property_set_link(obj, "presenter", OBJECT(xptr), &error_abort);
851     if (!qdev_realize(DEVICE(obj), NULL, errp)) {
852         object_unparent(obj);
853         return NULL;
854     }
855     return obj;
856 }
857 
858 void xive_tctx_destroy(XiveTCTX *tctx)
859 {
860     Object *obj = OBJECT(tctx);
861 
862     object_unparent(obj);
863 }
864 
865 /*
866  * XIVE ESB helpers
867  */
868 
869 uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
870 {
871     uint8_t old_pq = *pq & 0x3;
872 
873     *pq &= ~0x3;
874     *pq |= value & 0x3;
875 
876     return old_pq;
877 }
878 
879 bool xive_esb_trigger(uint8_t *pq)
880 {
881     uint8_t old_pq = *pq & 0x3;
882 
883     switch (old_pq) {
884     case XIVE_ESB_RESET:
885         xive_esb_set(pq, XIVE_ESB_PENDING);
886         return true;
887     case XIVE_ESB_PENDING:
888     case XIVE_ESB_QUEUED:
889         xive_esb_set(pq, XIVE_ESB_QUEUED);
890         return false;
891     case XIVE_ESB_OFF:
892         xive_esb_set(pq, XIVE_ESB_OFF);
893         return false;
894     default:
895          g_assert_not_reached();
896     }
897 }
898 
899 bool xive_esb_eoi(uint8_t *pq)
900 {
901     uint8_t old_pq = *pq & 0x3;
902 
903     switch (old_pq) {
904     case XIVE_ESB_RESET:
905     case XIVE_ESB_PENDING:
906         xive_esb_set(pq, XIVE_ESB_RESET);
907         return false;
908     case XIVE_ESB_QUEUED:
909         xive_esb_set(pq, XIVE_ESB_PENDING);
910         return true;
911     case XIVE_ESB_OFF:
912         xive_esb_set(pq, XIVE_ESB_OFF);
913         return false;
914     default:
915          g_assert_not_reached();
916     }
917 }
918 
919 /*
920  * XIVE Interrupt Source (or IVSE)
921  */
922 
923 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
924 {
925     assert(srcno < xsrc->nr_irqs);
926 
927     return xsrc->status[srcno] & 0x3;
928 }
929 
930 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
931 {
932     assert(srcno < xsrc->nr_irqs);
933 
934     return xive_esb_set(&xsrc->status[srcno], pq);
935 }
936 
937 /*
938  * Returns whether the event notification should be forwarded.
939  */
940 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
941 {
942     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
943 
944     xive_source_set_asserted(xsrc, srcno, true);
945 
946     switch (old_pq) {
947     case XIVE_ESB_RESET:
948         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
949         return true;
950     default:
951         return false;
952     }
953 }
954 
955 /*
956  * Sources can be configured with PQ offloading in which case the check
957  * on the PQ state bits of MSIs is disabled
958  */
959 static bool xive_source_esb_disabled(XiveSource *xsrc, uint32_t srcno)
960 {
961     return (xsrc->esb_flags & XIVE_SRC_PQ_DISABLE) &&
962         !xive_source_irq_is_lsi(xsrc, srcno);
963 }
964 
965 /*
966  * Returns whether the event notification should be forwarded.
967  */
968 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
969 {
970     bool ret;
971 
972     assert(srcno < xsrc->nr_irqs);
973 
974     if (xive_source_esb_disabled(xsrc, srcno)) {
975         return true;
976     }
977 
978     ret = xive_esb_trigger(&xsrc->status[srcno]);
979 
980     if (xive_source_irq_is_lsi(xsrc, srcno) &&
981         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
982         qemu_log_mask(LOG_GUEST_ERROR,
983                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
984     }
985 
986     return ret;
987 }
988 
989 /*
990  * Returns whether the event notification should be forwarded.
991  */
992 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
993 {
994     bool ret;
995 
996     assert(srcno < xsrc->nr_irqs);
997 
998     if (xive_source_esb_disabled(xsrc, srcno)) {
999         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EOI for IRQ %d\n", srcno);
1000         return false;
1001     }
1002 
1003     ret = xive_esb_eoi(&xsrc->status[srcno]);
1004 
1005     /*
1006      * LSI sources do not set the Q bit but they can still be
1007      * asserted, in which case we should forward a new event
1008      * notification
1009      */
1010     if (xive_source_irq_is_lsi(xsrc, srcno) &&
1011         xive_source_is_asserted(xsrc, srcno)) {
1012         ret = xive_source_lsi_trigger(xsrc, srcno);
1013     }
1014 
1015     return ret;
1016 }
1017 
1018 /*
1019  * Forward the source event notification to the Router
1020  */
1021 static void xive_source_notify(XiveSource *xsrc, int srcno)
1022 {
1023     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
1024     bool pq_checked = !xive_source_esb_disabled(xsrc, srcno);
1025 
1026     if (xnc->notify) {
1027         xnc->notify(xsrc->xive, srcno, pq_checked);
1028     }
1029 }
1030 
1031 /*
1032  * In a two pages ESB MMIO setting, even page is the trigger page, odd
1033  * page is for management
1034  */
1035 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
1036 {
1037     return !((addr >> shift) & 1);
1038 }
1039 
1040 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
1041 {
1042     return xive_source_esb_has_2page(xsrc) &&
1043         addr_is_even(addr, xsrc->esb_shift - 1);
1044 }
1045 
1046 /*
1047  * ESB MMIO loads
1048  *                      Trigger page    Management/EOI page
1049  *
1050  * ESB MMIO setting     2 pages         1 or 2 pages
1051  *
1052  * 0x000 .. 0x3FF       -1              EOI and return 0|1
1053  * 0x400 .. 0x7FF       -1              EOI and return 0|1
1054  * 0x800 .. 0xBFF       -1              return PQ
1055  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
1056  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
1057  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
1058  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
1059  */
1060 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
1061 {
1062     XiveSource *xsrc = XIVE_SOURCE(opaque);
1063     uint32_t offset = addr & 0xFFF;
1064     uint32_t srcno = addr >> xsrc->esb_shift;
1065     uint64_t ret = -1;
1066 
1067     /* In a two pages ESB MMIO setting, trigger page should not be read */
1068     if (xive_source_is_trigger_page(xsrc, addr)) {
1069         qemu_log_mask(LOG_GUEST_ERROR,
1070                       "XIVE: invalid load on IRQ %d trigger page at "
1071                       "0x%"HWADDR_PRIx"\n", srcno, addr);
1072         return -1;
1073     }
1074 
1075     switch (offset) {
1076     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1077         ret = xive_source_esb_eoi(xsrc, srcno);
1078 
1079         /* Forward the source event notification for routing */
1080         if (ret) {
1081             xive_source_notify(xsrc, srcno);
1082         }
1083         break;
1084 
1085     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1086         ret = xive_source_esb_get(xsrc, srcno);
1087         break;
1088 
1089     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1090     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1091     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1092     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1093         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1094         break;
1095     default:
1096         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
1097                       offset);
1098     }
1099 
1100     trace_xive_source_esb_read(addr, srcno, ret);
1101 
1102     return ret;
1103 }
1104 
1105 /*
1106  * ESB MMIO stores
1107  *                      Trigger page    Management/EOI page
1108  *
1109  * ESB MMIO setting     2 pages         1 or 2 pages
1110  *
1111  * 0x000 .. 0x3FF       Trigger         Trigger
1112  * 0x400 .. 0x7FF       Trigger         EOI
1113  * 0x800 .. 0xBFF       Trigger         undefined
1114  * 0xC00 .. 0xCFF       Trigger         PQ=00
1115  * 0xD00 .. 0xDFF       Trigger         PQ=01
1116  * 0xE00 .. 0xDFF       Trigger         PQ=10
1117  * 0xF00 .. 0xDFF       Trigger         PQ=11
1118  */
1119 static void xive_source_esb_write(void *opaque, hwaddr addr,
1120                                   uint64_t value, unsigned size)
1121 {
1122     XiveSource *xsrc = XIVE_SOURCE(opaque);
1123     uint32_t offset = addr & 0xFFF;
1124     uint32_t srcno = addr >> xsrc->esb_shift;
1125     bool notify = false;
1126 
1127     trace_xive_source_esb_write(addr, srcno, value);
1128 
1129     /* In a two pages ESB MMIO setting, trigger page only triggers */
1130     if (xive_source_is_trigger_page(xsrc, addr)) {
1131         notify = xive_source_esb_trigger(xsrc, srcno);
1132         goto out;
1133     }
1134 
1135     switch (offset) {
1136     case 0 ... 0x3FF:
1137         notify = xive_source_esb_trigger(xsrc, srcno);
1138         break;
1139 
1140     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
1141         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
1142             qemu_log_mask(LOG_GUEST_ERROR,
1143                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
1144             return;
1145         }
1146 
1147         notify = xive_source_esb_eoi(xsrc, srcno);
1148         break;
1149 
1150     /*
1151      * This is an internal offset used to inject triggers when the PQ
1152      * state bits are not controlled locally. Such as for LSIs when
1153      * under ABT mode.
1154      */
1155     case XIVE_ESB_INJECT ... XIVE_ESB_INJECT + 0x3FF:
1156         notify = true;
1157         break;
1158 
1159     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1160     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1161     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1162     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1163         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1164         break;
1165 
1166     default:
1167         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
1168                       offset);
1169         return;
1170     }
1171 
1172 out:
1173     /* Forward the source event notification for routing */
1174     if (notify) {
1175         xive_source_notify(xsrc, srcno);
1176     }
1177 }
1178 
1179 static const MemoryRegionOps xive_source_esb_ops = {
1180     .read = xive_source_esb_read,
1181     .write = xive_source_esb_write,
1182     .endianness = DEVICE_BIG_ENDIAN,
1183     .valid = {
1184         .min_access_size = 1,
1185         .max_access_size = 8,
1186     },
1187     .impl = {
1188         .min_access_size = 1,
1189         .max_access_size = 8,
1190     },
1191 };
1192 
1193 void xive_source_set_irq(void *opaque, int srcno, int val)
1194 {
1195     XiveSource *xsrc = XIVE_SOURCE(opaque);
1196     bool notify = false;
1197 
1198     if (xive_source_irq_is_lsi(xsrc, srcno)) {
1199         if (val) {
1200             notify = xive_source_lsi_trigger(xsrc, srcno);
1201         } else {
1202             xive_source_set_asserted(xsrc, srcno, false);
1203         }
1204     } else {
1205         if (val) {
1206             notify = xive_source_esb_trigger(xsrc, srcno);
1207         }
1208     }
1209 
1210     /* Forward the source event notification for routing */
1211     if (notify) {
1212         xive_source_notify(xsrc, srcno);
1213     }
1214 }
1215 
1216 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, GString *buf)
1217 {
1218     for (unsigned i = 0; i < xsrc->nr_irqs; i++) {
1219         uint8_t pq = xive_source_esb_get(xsrc, i);
1220 
1221         if (pq == XIVE_ESB_OFF) {
1222             continue;
1223         }
1224 
1225         g_string_append_printf(buf, "  %08x %s %c%c%c\n", i + offset,
1226                                xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
1227                                pq & XIVE_ESB_VAL_P ? 'P' : '-',
1228                                pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1229                                xive_source_is_asserted(xsrc, i) ? 'A' : ' ');
1230     }
1231 }
1232 
1233 static void xive_source_reset(void *dev)
1234 {
1235     XiveSource *xsrc = XIVE_SOURCE(dev);
1236 
1237     /* Do not clear the LSI bitmap */
1238 
1239     memset(xsrc->status, xsrc->reset_pq, xsrc->nr_irqs);
1240 }
1241 
1242 static void xive_source_realize(DeviceState *dev, Error **errp)
1243 {
1244     XiveSource *xsrc = XIVE_SOURCE(dev);
1245     size_t esb_len = xive_source_esb_len(xsrc);
1246 
1247     assert(xsrc->xive);
1248 
1249     if (!xsrc->nr_irqs) {
1250         error_setg(errp, "Number of interrupt needs to be greater than 0");
1251         return;
1252     }
1253 
1254     if (xsrc->esb_shift != XIVE_ESB_4K &&
1255         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1256         xsrc->esb_shift != XIVE_ESB_64K &&
1257         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1258         error_setg(errp, "Invalid ESB shift setting");
1259         return;
1260     }
1261 
1262     xsrc->status = g_malloc0(xsrc->nr_irqs);
1263     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1264 
1265     memory_region_init(&xsrc->esb_mmio, OBJECT(xsrc), "xive.esb", esb_len);
1266     memory_region_init_io(&xsrc->esb_mmio_emulated, OBJECT(xsrc),
1267                           &xive_source_esb_ops, xsrc, "xive.esb-emulated",
1268                           esb_len);
1269     memory_region_add_subregion(&xsrc->esb_mmio, 0, &xsrc->esb_mmio_emulated);
1270 
1271     qemu_register_reset(xive_source_reset, dev);
1272 }
1273 
1274 static const VMStateDescription vmstate_xive_source = {
1275     .name = TYPE_XIVE_SOURCE,
1276     .version_id = 1,
1277     .minimum_version_id = 1,
1278     .fields = (const VMStateField[]) {
1279         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1280         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1281         VMSTATE_END_OF_LIST()
1282     },
1283 };
1284 
1285 /*
1286  * The default XIVE interrupt source setting for the ESB MMIOs is two
1287  * 64k pages without Store EOI, to be in sync with KVM.
1288  */
1289 static Property xive_source_properties[] = {
1290     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1291     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1292     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1293     /*
1294      * By default, PQs are initialized to 0b01 (Q=1) which corresponds
1295      * to "ints off"
1296      */
1297     DEFINE_PROP_UINT8("reset-pq", XiveSource, reset_pq, XIVE_ESB_OFF),
1298     DEFINE_PROP_LINK("xive", XiveSource, xive, TYPE_XIVE_NOTIFIER,
1299                      XiveNotifier *),
1300     DEFINE_PROP_END_OF_LIST(),
1301 };
1302 
1303 static void xive_source_class_init(ObjectClass *klass, void *data)
1304 {
1305     DeviceClass *dc = DEVICE_CLASS(klass);
1306 
1307     dc->desc    = "XIVE Interrupt Source";
1308     device_class_set_props(dc, xive_source_properties);
1309     dc->realize = xive_source_realize;
1310     dc->vmsd    = &vmstate_xive_source;
1311     /*
1312      * Reason: part of XIVE interrupt controller, needs to be wired up,
1313      * e.g. by spapr_xive_instance_init().
1314      */
1315     dc->user_creatable = false;
1316 }
1317 
1318 static const TypeInfo xive_source_info = {
1319     .name          = TYPE_XIVE_SOURCE,
1320     .parent        = TYPE_DEVICE,
1321     .instance_size = sizeof(XiveSource),
1322     .class_init    = xive_source_class_init,
1323 };
1324 
1325 /*
1326  * XiveEND helpers
1327  */
1328 
1329 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, GString *buf)
1330 {
1331     uint64_t qaddr_base = xive_end_qaddr(end);
1332     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1333     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1334     uint32_t qentries = 1 << (qsize + 10);
1335     int i;
1336 
1337     /*
1338      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1339      */
1340     g_string_append_printf(buf, " [ ");
1341     qindex = (qindex - (width - 1)) & (qentries - 1);
1342     for (i = 0; i < width; i++) {
1343         uint64_t qaddr = qaddr_base + (qindex << 2);
1344         uint32_t qdata = -1;
1345 
1346         if (dma_memory_read(&address_space_memory, qaddr,
1347                             &qdata, sizeof(qdata), MEMTXATTRS_UNSPECIFIED)) {
1348             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1349                           HWADDR_PRIx "\n", qaddr);
1350             return;
1351         }
1352         g_string_append_printf(buf, "%s%08x ", i == width - 1 ? "^" : "",
1353                                be32_to_cpu(qdata));
1354         qindex = (qindex + 1) & (qentries - 1);
1355     }
1356     g_string_append_c(buf, ']');
1357 }
1358 
1359 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, GString *buf)
1360 {
1361     uint64_t qaddr_base = xive_end_qaddr(end);
1362     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1363     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1364     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1365     uint32_t qentries = 1 << (qsize + 10);
1366 
1367     uint32_t nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1368     uint32_t nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1369     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1370     uint8_t pq;
1371 
1372     if (!xive_end_is_valid(end)) {
1373         return;
1374     }
1375 
1376     pq = xive_get_field32(END_W1_ESn, end->w1);
1377 
1378     g_string_append_printf(buf,
1379                            "  %08x %c%c %c%c%c%c%c%c%c%c prio:%d nvt:%02x/%04x",
1380                            end_idx,
1381                            pq & XIVE_ESB_VAL_P ? 'P' : '-',
1382                            pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1383                            xive_end_is_valid(end)    ? 'v' : '-',
1384                            xive_end_is_enqueue(end)  ? 'q' : '-',
1385                            xive_end_is_notify(end)   ? 'n' : '-',
1386                            xive_end_is_backlog(end)  ? 'b' : '-',
1387                            xive_end_is_escalate(end) ? 'e' : '-',
1388                            xive_end_is_uncond_escalation(end)   ? 'u' : '-',
1389                            xive_end_is_silent_escalation(end)   ? 's' : '-',
1390                            xive_end_is_firmware(end)   ? 'f' : '-',
1391                            priority, nvt_blk, nvt_idx);
1392 
1393     if (qaddr_base) {
1394         g_string_append_printf(buf, " eq:@%08"PRIx64"% 6d/%5d ^%d",
1395                                qaddr_base, qindex, qentries, qgen);
1396         xive_end_queue_pic_print_info(end, 6, buf);
1397     }
1398     g_string_append_c(buf, '\n');
1399 }
1400 
1401 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1402 {
1403     uint64_t qaddr_base = xive_end_qaddr(end);
1404     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1405     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1406     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1407 
1408     uint64_t qaddr = qaddr_base + (qindex << 2);
1409     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1410     uint32_t qentries = 1 << (qsize + 10);
1411 
1412     if (dma_memory_write(&address_space_memory, qaddr,
1413                          &qdata, sizeof(qdata), MEMTXATTRS_UNSPECIFIED)) {
1414         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1415                       HWADDR_PRIx "\n", qaddr);
1416         return;
1417     }
1418 
1419     qindex = (qindex + 1) & (qentries - 1);
1420     if (qindex == 0) {
1421         qgen ^= 1;
1422         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1423     }
1424     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1425 }
1426 
1427 void xive_end_eas_pic_print_info(XiveEND *end, uint32_t end_idx, GString *buf)
1428 {
1429     XiveEAS *eas = (XiveEAS *) &end->w4;
1430     uint8_t pq;
1431 
1432     if (!xive_end_is_escalate(end)) {
1433         return;
1434     }
1435 
1436     pq = xive_get_field32(END_W1_ESe, end->w1);
1437 
1438     g_string_append_printf(buf, "  %08x %c%c %c%c end:%02x/%04x data:%08x\n",
1439                            end_idx,
1440                            pq & XIVE_ESB_VAL_P ? 'P' : '-',
1441                            pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1442                            xive_eas_is_valid(eas) ? 'V' : ' ',
1443                            xive_eas_is_masked(eas) ? 'M' : ' ',
1444                            (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1445                            (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1446                            (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1447 }
1448 
1449 /*
1450  * XIVE Router (aka. Virtualization Controller or IVRE)
1451  */
1452 
1453 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1454                         XiveEAS *eas)
1455 {
1456     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1457 
1458     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1459 }
1460 
1461 static
1462 int xive_router_get_pq(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1463                        uint8_t *pq)
1464 {
1465     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1466 
1467     return xrc->get_pq(xrtr, eas_blk, eas_idx, pq);
1468 }
1469 
1470 static
1471 int xive_router_set_pq(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1472                        uint8_t *pq)
1473 {
1474     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1475 
1476     return xrc->set_pq(xrtr, eas_blk, eas_idx, pq);
1477 }
1478 
1479 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1480                         XiveEND *end)
1481 {
1482    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1483 
1484    return xrc->get_end(xrtr, end_blk, end_idx, end);
1485 }
1486 
1487 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1488                           XiveEND *end, uint8_t word_number)
1489 {
1490    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1491 
1492    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1493 }
1494 
1495 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1496                         XiveNVT *nvt)
1497 {
1498    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1499 
1500    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1501 }
1502 
1503 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1504                         XiveNVT *nvt, uint8_t word_number)
1505 {
1506    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1507 
1508    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1509 }
1510 
1511 static int xive_router_get_block_id(XiveRouter *xrtr)
1512 {
1513    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1514 
1515    return xrc->get_block_id(xrtr);
1516 }
1517 
1518 static void xive_router_realize(DeviceState *dev, Error **errp)
1519 {
1520     XiveRouter *xrtr = XIVE_ROUTER(dev);
1521 
1522     assert(xrtr->xfb);
1523 }
1524 
1525 static void xive_router_end_notify_handler(XiveRouter *xrtr, XiveEAS *eas)
1526 {
1527     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1528 
1529     return xrc->end_notify(xrtr, eas);
1530 }
1531 
1532 /*
1533  * Encode the HW CAM line in the block group mode format :
1534  *
1535  *   chip << 19 | 0000000 0 0001 thread (7Bit)
1536  */
1537 static uint32_t xive_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
1538 {
1539     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1540     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1541     uint8_t blk = xive_router_get_block_id(XIVE_ROUTER(xptr));
1542 
1543     return xive_nvt_cam_line(blk, 1 << 7 | (pir & 0x7f));
1544 }
1545 
1546 /*
1547  * The thread context register words are in big-endian format.
1548  */
1549 int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1550                               uint8_t format,
1551                               uint8_t nvt_blk, uint32_t nvt_idx,
1552                               bool cam_ignore, uint32_t logic_serv)
1553 {
1554     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1555     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1556     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1557     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1558     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1559 
1560     /*
1561      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1562      * identifier are ignored in the "CAM" match.
1563      */
1564 
1565     if (format == 0) {
1566         if (cam_ignore == true) {
1567             /*
1568              * F=0 & i=1: Logical server notification (bits ignored at
1569              * the end of the NVT identifier)
1570              */
1571             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1572                           nvt_blk, nvt_idx);
1573              return -1;
1574         }
1575 
1576         /* F=0 & i=0: Specific NVT notification */
1577 
1578         /* PHYS ring */
1579         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1580             cam == xive_tctx_hw_cam_line(xptr, tctx)) {
1581             return TM_QW3_HV_PHYS;
1582         }
1583 
1584         /* HV POOL ring */
1585         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1586             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1587             return TM_QW2_HV_POOL;
1588         }
1589 
1590         /* OS ring */
1591         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1592             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1593             return TM_QW1_OS;
1594         }
1595     } else {
1596         /* F=1 : User level Event-Based Branch (EBB) notification */
1597 
1598         /* USER ring */
1599         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1600              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1601              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1602              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1603             return TM_QW0_USER;
1604         }
1605     }
1606     return -1;
1607 }
1608 
1609 /*
1610  * This is our simple Xive Presenter Engine model. It is merged in the
1611  * Router as it does not require an extra object.
1612  *
1613  * It receives notification requests sent by the IVRE to find one
1614  * matching NVT (or more) dispatched on the processor threads. In case
1615  * of a single NVT notification, the process is abbreviated and the
1616  * thread is signaled if a match is found. In case of a logical server
1617  * notification (bits ignored at the end of the NVT identifier), the
1618  * IVPE and IVRE select a winning thread using different filters. This
1619  * involves 2 or 3 exchanges on the PowerBus that the model does not
1620  * support.
1621  *
1622  * The parameters represent what is sent on the PowerBus
1623  */
1624 bool xive_presenter_notify(XiveFabric *xfb, uint8_t format,
1625                            uint8_t nvt_blk, uint32_t nvt_idx,
1626                            bool cam_ignore, uint8_t priority,
1627                            uint32_t logic_serv)
1628 {
1629     XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xfb);
1630     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1631     int count;
1632 
1633     /*
1634      * Ask the machine to scan the interrupt controllers for a match
1635      */
1636     count = xfc->match_nvt(xfb, format, nvt_blk, nvt_idx, cam_ignore,
1637                            priority, logic_serv, &match);
1638     if (count < 0) {
1639         return false;
1640     }
1641 
1642     /* handle CPU exception delivery */
1643     if (count) {
1644         trace_xive_presenter_notify(nvt_blk, nvt_idx, match.ring);
1645         xive_tctx_ipb_update(match.tctx, match.ring,
1646                              xive_priority_to_ipb(priority));
1647     }
1648 
1649     return !!count;
1650 }
1651 
1652 /*
1653  * Notification using the END ESe/ESn bit (Event State Buffer for
1654  * escalation and notification). Provide further coalescing in the
1655  * Router.
1656  */
1657 static bool xive_router_end_es_notify(XiveRouter *xrtr, uint8_t end_blk,
1658                                       uint32_t end_idx, XiveEND *end,
1659                                       uint32_t end_esmask)
1660 {
1661     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1662     bool notify = xive_esb_trigger(&pq);
1663 
1664     if (pq != xive_get_field32(end_esmask, end->w1)) {
1665         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1666         xive_router_write_end(xrtr, end_blk, end_idx, end, 1);
1667     }
1668 
1669     /* ESe/n[Q]=1 : end of notification */
1670     return notify;
1671 }
1672 
1673 /*
1674  * An END trigger can come from an event trigger (IPI or HW) or from
1675  * another chip. We don't model the PowerBus but the END trigger
1676  * message has the same parameters than in the function below.
1677  */
1678 void xive_router_end_notify(XiveRouter *xrtr, XiveEAS *eas)
1679 {
1680     XiveEND end;
1681     uint8_t priority;
1682     uint8_t format;
1683     uint8_t nvt_blk;
1684     uint32_t nvt_idx;
1685     XiveNVT nvt;
1686     bool found;
1687 
1688     uint8_t end_blk = xive_get_field64(EAS_END_BLOCK, eas->w);
1689     uint32_t end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
1690     uint32_t end_data = xive_get_field64(EAS_END_DATA,  eas->w);
1691 
1692     /* END cache lookup */
1693     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1694         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1695                       end_idx);
1696         return;
1697     }
1698 
1699     if (!xive_end_is_valid(&end)) {
1700         trace_xive_router_end_notify(end_blk, end_idx, end_data);
1701         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1702                       end_blk, end_idx);
1703         return;
1704     }
1705 
1706     if (xive_end_is_enqueue(&end)) {
1707         xive_end_enqueue(&end, end_data);
1708         /* Enqueuing event data modifies the EQ toggle and index */
1709         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1710     }
1711 
1712     /*
1713      * When the END is silent, we skip the notification part.
1714      */
1715     if (xive_end_is_silent_escalation(&end)) {
1716         goto do_escalation;
1717     }
1718 
1719     /*
1720      * The W7 format depends on the F bit in W6. It defines the type
1721      * of the notification :
1722      *
1723      *   F=0 : single or multiple NVT notification
1724      *   F=1 : User level Event-Based Branch (EBB) notification, no
1725      *         priority
1726      */
1727     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1728     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1729 
1730     /* The END is masked */
1731     if (format == 0 && priority == 0xff) {
1732         return;
1733     }
1734 
1735     /*
1736      * Check the END ESn (Event State Buffer for notification) for
1737      * even further coalescing in the Router
1738      */
1739     if (!xive_end_is_notify(&end)) {
1740         /* ESn[Q]=1 : end of notification */
1741         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1742                                        &end, END_W1_ESn)) {
1743             return;
1744         }
1745     }
1746 
1747     /*
1748      * Follows IVPE notification
1749      */
1750     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end.w6);
1751     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end.w6);
1752 
1753     /* NVT cache lookup */
1754     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1755         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1756                       nvt_blk, nvt_idx);
1757         return;
1758     }
1759 
1760     if (!xive_nvt_is_valid(&nvt)) {
1761         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1762                       nvt_blk, nvt_idx);
1763         return;
1764     }
1765 
1766     found = xive_presenter_notify(xrtr->xfb, format, nvt_blk, nvt_idx,
1767                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1768                           priority,
1769                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1770 
1771     /* TODO: Auto EOI. */
1772 
1773     if (found) {
1774         return;
1775     }
1776 
1777     /*
1778      * If no matching NVT is dispatched on a HW thread :
1779      * - specific VP: update the NVT structure if backlog is activated
1780      * - logical server : forward request to IVPE (not supported)
1781      */
1782     if (xive_end_is_backlog(&end)) {
1783         uint8_t ipb;
1784 
1785         if (format == 1) {
1786             qemu_log_mask(LOG_GUEST_ERROR,
1787                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1788                           end_blk, end_idx);
1789             return;
1790         }
1791         /*
1792          * Record the IPB in the associated NVT structure for later
1793          * use. The presenter will resend the interrupt when the vCPU
1794          * is dispatched again on a HW thread.
1795          */
1796         ipb = xive_get_field32(NVT_W4_IPB, nvt.w4) |
1797             xive_priority_to_ipb(priority);
1798         nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, ipb);
1799         xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1800 
1801         /*
1802          * On HW, follows a "Broadcast Backlog" to IVPEs
1803          */
1804     }
1805 
1806 do_escalation:
1807     /*
1808      * If activated, escalate notification using the ESe PQ bits and
1809      * the EAS in w4-5
1810      */
1811     if (!xive_end_is_escalate(&end)) {
1812         return;
1813     }
1814 
1815     /*
1816      * Check the END ESe (Event State Buffer for escalation) for even
1817      * further coalescing in the Router
1818      */
1819     if (!xive_end_is_uncond_escalation(&end)) {
1820         /* ESe[Q]=1 : end of notification */
1821         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1822                                        &end, END_W1_ESe)) {
1823             return;
1824         }
1825     }
1826 
1827     trace_xive_router_end_escalate(end_blk, end_idx,
1828            (uint8_t) xive_get_field32(END_W4_ESC_END_BLOCK, end.w4),
1829            (uint32_t) xive_get_field32(END_W4_ESC_END_INDEX, end.w4),
1830            (uint32_t) xive_get_field32(END_W5_ESC_END_DATA,  end.w5));
1831     /*
1832      * The END trigger becomes an Escalation trigger
1833      */
1834     xive_router_end_notify_handler(xrtr, (XiveEAS *) &end.w4);
1835 }
1836 
1837 void xive_router_notify(XiveNotifier *xn, uint32_t lisn, bool pq_checked)
1838 {
1839     XiveRouter *xrtr = XIVE_ROUTER(xn);
1840     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1841     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1842     XiveEAS eas;
1843 
1844     /* EAS cache lookup */
1845     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1846         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1847         return;
1848     }
1849 
1850     if (!pq_checked) {
1851         bool notify;
1852         uint8_t pq;
1853 
1854         /* PQ cache lookup */
1855         if (xive_router_get_pq(xrtr, eas_blk, eas_idx, &pq)) {
1856             /* Set FIR */
1857             g_assert_not_reached();
1858         }
1859 
1860         notify = xive_esb_trigger(&pq);
1861 
1862         if (xive_router_set_pq(xrtr, eas_blk, eas_idx, &pq)) {
1863             /* Set FIR */
1864             g_assert_not_reached();
1865         }
1866 
1867         if (!notify) {
1868             return;
1869         }
1870     }
1871 
1872     if (!xive_eas_is_valid(&eas)) {
1873         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1874         return;
1875     }
1876 
1877     if (xive_eas_is_masked(&eas)) {
1878         /* Notification completed */
1879         return;
1880     }
1881 
1882     /*
1883      * The event trigger becomes an END trigger
1884      */
1885     xive_router_end_notify_handler(xrtr, &eas);
1886 }
1887 
1888 static Property xive_router_properties[] = {
1889     DEFINE_PROP_LINK("xive-fabric", XiveRouter, xfb,
1890                      TYPE_XIVE_FABRIC, XiveFabric *),
1891     DEFINE_PROP_END_OF_LIST(),
1892 };
1893 
1894 static void xive_router_class_init(ObjectClass *klass, void *data)
1895 {
1896     DeviceClass *dc = DEVICE_CLASS(klass);
1897     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1898     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
1899 
1900     dc->desc    = "XIVE Router Engine";
1901     device_class_set_props(dc, xive_router_properties);
1902     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1903     dc->realize = xive_router_realize;
1904     xnc->notify = xive_router_notify;
1905 
1906     /* By default, the router handles END triggers locally */
1907     xrc->end_notify = xive_router_end_notify;
1908 }
1909 
1910 static const TypeInfo xive_router_info = {
1911     .name          = TYPE_XIVE_ROUTER,
1912     .parent        = TYPE_SYS_BUS_DEVICE,
1913     .abstract      = true,
1914     .instance_size = sizeof(XiveRouter),
1915     .class_size    = sizeof(XiveRouterClass),
1916     .class_init    = xive_router_class_init,
1917     .interfaces    = (InterfaceInfo[]) {
1918         { TYPE_XIVE_NOTIFIER },
1919         { TYPE_XIVE_PRESENTER },
1920         { }
1921     }
1922 };
1923 
1924 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, GString *buf)
1925 {
1926     if (!xive_eas_is_valid(eas)) {
1927         return;
1928     }
1929 
1930     g_string_append_printf(buf, "  %08x %s end:%02x/%04x data:%08x\n",
1931                            lisn, xive_eas_is_masked(eas) ? "M" : " ",
1932                            (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1933                            (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1934                            (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1935 }
1936 
1937 /*
1938  * END ESB MMIO loads
1939  */
1940 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1941 {
1942     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1943     uint32_t offset = addr & 0xFFF;
1944     uint8_t end_blk;
1945     uint32_t end_idx;
1946     XiveEND end;
1947     uint32_t end_esmask;
1948     uint8_t pq;
1949     uint64_t ret = -1;
1950 
1951     /*
1952      * The block id should be deduced from the load address on the END
1953      * ESB MMIO but our model only supports a single block per XIVE chip.
1954      */
1955     end_blk = xive_router_get_block_id(xsrc->xrtr);
1956     end_idx = addr >> (xsrc->esb_shift + 1);
1957 
1958     trace_xive_end_source_read(end_blk, end_idx, addr);
1959 
1960     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1961         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1962                       end_idx);
1963         return -1;
1964     }
1965 
1966     if (!xive_end_is_valid(&end)) {
1967         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1968                       end_blk, end_idx);
1969         return -1;
1970     }
1971 
1972     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1973     pq = xive_get_field32(end_esmask, end.w1);
1974 
1975     switch (offset) {
1976     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1977         ret = xive_esb_eoi(&pq);
1978 
1979         /* Forward the source event notification for routing ?? */
1980         break;
1981 
1982     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1983         ret = pq;
1984         break;
1985 
1986     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1987     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1988     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1989     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1990         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1991         break;
1992     default:
1993         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1994                       offset);
1995         return -1;
1996     }
1997 
1998     if (pq != xive_get_field32(end_esmask, end.w1)) {
1999         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
2000         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
2001     }
2002 
2003     return ret;
2004 }
2005 
2006 /*
2007  * END ESB MMIO stores are invalid
2008  */
2009 static void xive_end_source_write(void *opaque, hwaddr addr,
2010                                   uint64_t value, unsigned size)
2011 {
2012     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
2013                   HWADDR_PRIx"\n", addr);
2014 }
2015 
2016 static const MemoryRegionOps xive_end_source_ops = {
2017     .read = xive_end_source_read,
2018     .write = xive_end_source_write,
2019     .endianness = DEVICE_BIG_ENDIAN,
2020     .valid = {
2021         .min_access_size = 1,
2022         .max_access_size = 8,
2023     },
2024     .impl = {
2025         .min_access_size = 1,
2026         .max_access_size = 8,
2027     },
2028 };
2029 
2030 static void xive_end_source_realize(DeviceState *dev, Error **errp)
2031 {
2032     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
2033 
2034     assert(xsrc->xrtr);
2035 
2036     if (!xsrc->nr_ends) {
2037         error_setg(errp, "Number of interrupt needs to be greater than 0");
2038         return;
2039     }
2040 
2041     if (xsrc->esb_shift != XIVE_ESB_4K &&
2042         xsrc->esb_shift != XIVE_ESB_64K) {
2043         error_setg(errp, "Invalid ESB shift setting");
2044         return;
2045     }
2046 
2047     /*
2048      * Each END is assigned an even/odd pair of MMIO pages, the even page
2049      * manages the ESn field while the odd page manages the ESe field.
2050      */
2051     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
2052                           &xive_end_source_ops, xsrc, "xive.end",
2053                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
2054 }
2055 
2056 static Property xive_end_source_properties[] = {
2057     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
2058     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
2059     DEFINE_PROP_LINK("xive", XiveENDSource, xrtr, TYPE_XIVE_ROUTER,
2060                      XiveRouter *),
2061     DEFINE_PROP_END_OF_LIST(),
2062 };
2063 
2064 static void xive_end_source_class_init(ObjectClass *klass, void *data)
2065 {
2066     DeviceClass *dc = DEVICE_CLASS(klass);
2067 
2068     dc->desc    = "XIVE END Source";
2069     device_class_set_props(dc, xive_end_source_properties);
2070     dc->realize = xive_end_source_realize;
2071     /*
2072      * Reason: part of XIVE interrupt controller, needs to be wired up,
2073      * e.g. by spapr_xive_instance_init().
2074      */
2075     dc->user_creatable = false;
2076 }
2077 
2078 static const TypeInfo xive_end_source_info = {
2079     .name          = TYPE_XIVE_END_SOURCE,
2080     .parent        = TYPE_DEVICE,
2081     .instance_size = sizeof(XiveENDSource),
2082     .class_init    = xive_end_source_class_init,
2083 };
2084 
2085 /*
2086  * XIVE Notifier
2087  */
2088 static const TypeInfo xive_notifier_info = {
2089     .name = TYPE_XIVE_NOTIFIER,
2090     .parent = TYPE_INTERFACE,
2091     .class_size = sizeof(XiveNotifierClass),
2092 };
2093 
2094 /*
2095  * XIVE Presenter
2096  */
2097 static const TypeInfo xive_presenter_info = {
2098     .name = TYPE_XIVE_PRESENTER,
2099     .parent = TYPE_INTERFACE,
2100     .class_size = sizeof(XivePresenterClass),
2101 };
2102 
2103 /*
2104  * XIVE Fabric
2105  */
2106 static const TypeInfo xive_fabric_info = {
2107     .name = TYPE_XIVE_FABRIC,
2108     .parent = TYPE_INTERFACE,
2109     .class_size = sizeof(XiveFabricClass),
2110 };
2111 
2112 static void xive_register_types(void)
2113 {
2114     type_register_static(&xive_fabric_info);
2115     type_register_static(&xive_source_info);
2116     type_register_static(&xive_notifier_info);
2117     type_register_static(&xive_presenter_info);
2118     type_register_static(&xive_router_info);
2119     type_register_static(&xive_end_source_info);
2120     type_register_static(&xive_tctx_info);
2121 }
2122 
2123 type_init(xive_register_types)
2124