xref: /openbmc/qemu/hw/intc/xive.c (revision 7c08eefc)
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     g_string_append_printf(buf, "CPU[%04x]:   "
696                            "QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR  W2\n",
697                            cpu_index);
698 
699     for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
700         char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
701         g_string_append_printf(buf, "CPU[%04x]: %4s    %s\n",
702                                cpu_index, xive_tctx_ring_names[i], s);
703         g_free(s);
704     }
705 }
706 
707 void xive_tctx_reset(XiveTCTX *tctx)
708 {
709     memset(tctx->regs, 0, sizeof(tctx->regs));
710 
711     /* Set some defaults */
712     tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
713     tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
714     tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
715 
716     /*
717      * Initialize PIPR to 0xFF to avoid phantom interrupts when the
718      * CPPR is first set.
719      */
720     tctx->regs[TM_QW1_OS + TM_PIPR] =
721         ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
722     tctx->regs[TM_QW3_HV_PHYS + TM_PIPR] =
723         ipb_to_pipr(tctx->regs[TM_QW3_HV_PHYS + TM_IPB]);
724 }
725 
726 static void xive_tctx_realize(DeviceState *dev, Error **errp)
727 {
728     XiveTCTX *tctx = XIVE_TCTX(dev);
729     PowerPCCPU *cpu;
730     CPUPPCState *env;
731 
732     assert(tctx->cs);
733     assert(tctx->xptr);
734 
735     cpu = POWERPC_CPU(tctx->cs);
736     env = &cpu->env;
737     switch (PPC_INPUT(env)) {
738     case PPC_FLAGS_INPUT_POWER9:
739         tctx->hv_output = qdev_get_gpio_in(DEVICE(cpu), POWER9_INPUT_HINT);
740         tctx->os_output = qdev_get_gpio_in(DEVICE(cpu), POWER9_INPUT_INT);
741         break;
742 
743     default:
744         error_setg(errp, "XIVE interrupt controller does not support "
745                    "this CPU bus model");
746         return;
747     }
748 
749     /* Connect the presenter to the VCPU (required for CPU hotplug) */
750     if (xive_in_kernel(tctx->xptr)) {
751         if (kvmppc_xive_cpu_connect(tctx, errp) < 0) {
752             return;
753         }
754     }
755 }
756 
757 static int vmstate_xive_tctx_pre_save(void *opaque)
758 {
759     XiveTCTX *tctx = XIVE_TCTX(opaque);
760     Error *local_err = NULL;
761     int ret;
762 
763     if (xive_in_kernel(tctx->xptr)) {
764         ret = kvmppc_xive_cpu_get_state(tctx, &local_err);
765         if (ret < 0) {
766             error_report_err(local_err);
767             return ret;
768         }
769     }
770 
771     return 0;
772 }
773 
774 static int vmstate_xive_tctx_post_load(void *opaque, int version_id)
775 {
776     XiveTCTX *tctx = XIVE_TCTX(opaque);
777     Error *local_err = NULL;
778     int ret;
779 
780     if (xive_in_kernel(tctx->xptr)) {
781         /*
782          * Required for hotplugged CPU, for which the state comes
783          * after all states of the machine.
784          */
785         ret = kvmppc_xive_cpu_set_state(tctx, &local_err);
786         if (ret < 0) {
787             error_report_err(local_err);
788             return ret;
789         }
790     }
791 
792     return 0;
793 }
794 
795 static const VMStateDescription vmstate_xive_tctx = {
796     .name = TYPE_XIVE_TCTX,
797     .version_id = 1,
798     .minimum_version_id = 1,
799     .pre_save = vmstate_xive_tctx_pre_save,
800     .post_load = vmstate_xive_tctx_post_load,
801     .fields = (const VMStateField[]) {
802         VMSTATE_BUFFER(regs, XiveTCTX),
803         VMSTATE_END_OF_LIST()
804     },
805 };
806 
807 static Property xive_tctx_properties[] = {
808     DEFINE_PROP_LINK("cpu", XiveTCTX, cs, TYPE_CPU, CPUState *),
809     DEFINE_PROP_LINK("presenter", XiveTCTX, xptr, TYPE_XIVE_PRESENTER,
810                      XivePresenter *),
811     DEFINE_PROP_END_OF_LIST(),
812 };
813 
814 static void xive_tctx_class_init(ObjectClass *klass, void *data)
815 {
816     DeviceClass *dc = DEVICE_CLASS(klass);
817 
818     dc->desc = "XIVE Interrupt Thread Context";
819     dc->realize = xive_tctx_realize;
820     dc->vmsd = &vmstate_xive_tctx;
821     device_class_set_props(dc, xive_tctx_properties);
822     /*
823      * Reason: part of XIVE interrupt controller, needs to be wired up
824      * by xive_tctx_create().
825      */
826     dc->user_creatable = false;
827 }
828 
829 static const TypeInfo xive_tctx_info = {
830     .name          = TYPE_XIVE_TCTX,
831     .parent        = TYPE_DEVICE,
832     .instance_size = sizeof(XiveTCTX),
833     .class_init    = xive_tctx_class_init,
834 };
835 
836 Object *xive_tctx_create(Object *cpu, XivePresenter *xptr, Error **errp)
837 {
838     Object *obj;
839 
840     obj = object_new(TYPE_XIVE_TCTX);
841     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj);
842     object_unref(obj);
843     object_property_set_link(obj, "cpu", cpu, &error_abort);
844     object_property_set_link(obj, "presenter", OBJECT(xptr), &error_abort);
845     if (!qdev_realize(DEVICE(obj), NULL, errp)) {
846         object_unparent(obj);
847         return NULL;
848     }
849     return obj;
850 }
851 
852 void xive_tctx_destroy(XiveTCTX *tctx)
853 {
854     Object *obj = OBJECT(tctx);
855 
856     object_unparent(obj);
857 }
858 
859 /*
860  * XIVE ESB helpers
861  */
862 
863 uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
864 {
865     uint8_t old_pq = *pq & 0x3;
866 
867     *pq &= ~0x3;
868     *pq |= value & 0x3;
869 
870     return old_pq;
871 }
872 
873 bool xive_esb_trigger(uint8_t *pq)
874 {
875     uint8_t old_pq = *pq & 0x3;
876 
877     switch (old_pq) {
878     case XIVE_ESB_RESET:
879         xive_esb_set(pq, XIVE_ESB_PENDING);
880         return true;
881     case XIVE_ESB_PENDING:
882     case XIVE_ESB_QUEUED:
883         xive_esb_set(pq, XIVE_ESB_QUEUED);
884         return false;
885     case XIVE_ESB_OFF:
886         xive_esb_set(pq, XIVE_ESB_OFF);
887         return false;
888     default:
889          g_assert_not_reached();
890     }
891 }
892 
893 bool xive_esb_eoi(uint8_t *pq)
894 {
895     uint8_t old_pq = *pq & 0x3;
896 
897     switch (old_pq) {
898     case XIVE_ESB_RESET:
899     case XIVE_ESB_PENDING:
900         xive_esb_set(pq, XIVE_ESB_RESET);
901         return false;
902     case XIVE_ESB_QUEUED:
903         xive_esb_set(pq, XIVE_ESB_PENDING);
904         return true;
905     case XIVE_ESB_OFF:
906         xive_esb_set(pq, XIVE_ESB_OFF);
907         return false;
908     default:
909          g_assert_not_reached();
910     }
911 }
912 
913 /*
914  * XIVE Interrupt Source (or IVSE)
915  */
916 
917 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
918 {
919     assert(srcno < xsrc->nr_irqs);
920 
921     return xsrc->status[srcno] & 0x3;
922 }
923 
924 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
925 {
926     assert(srcno < xsrc->nr_irqs);
927 
928     return xive_esb_set(&xsrc->status[srcno], pq);
929 }
930 
931 /*
932  * Returns whether the event notification should be forwarded.
933  */
934 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
935 {
936     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
937 
938     xive_source_set_asserted(xsrc, srcno, true);
939 
940     switch (old_pq) {
941     case XIVE_ESB_RESET:
942         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
943         return true;
944     default:
945         return false;
946     }
947 }
948 
949 /*
950  * Sources can be configured with PQ offloading in which case the check
951  * on the PQ state bits of MSIs is disabled
952  */
953 static bool xive_source_esb_disabled(XiveSource *xsrc, uint32_t srcno)
954 {
955     return (xsrc->esb_flags & XIVE_SRC_PQ_DISABLE) &&
956         !xive_source_irq_is_lsi(xsrc, srcno);
957 }
958 
959 /*
960  * Returns whether the event notification should be forwarded.
961  */
962 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
963 {
964     bool ret;
965 
966     assert(srcno < xsrc->nr_irqs);
967 
968     if (xive_source_esb_disabled(xsrc, srcno)) {
969         return true;
970     }
971 
972     ret = xive_esb_trigger(&xsrc->status[srcno]);
973 
974     if (xive_source_irq_is_lsi(xsrc, srcno) &&
975         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
976         qemu_log_mask(LOG_GUEST_ERROR,
977                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
978     }
979 
980     return ret;
981 }
982 
983 /*
984  * Returns whether the event notification should be forwarded.
985  */
986 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
987 {
988     bool ret;
989 
990     assert(srcno < xsrc->nr_irqs);
991 
992     if (xive_source_esb_disabled(xsrc, srcno)) {
993         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EOI for IRQ %d\n", srcno);
994         return false;
995     }
996 
997     ret = xive_esb_eoi(&xsrc->status[srcno]);
998 
999     /*
1000      * LSI sources do not set the Q bit but they can still be
1001      * asserted, in which case we should forward a new event
1002      * notification
1003      */
1004     if (xive_source_irq_is_lsi(xsrc, srcno) &&
1005         xive_source_is_asserted(xsrc, srcno)) {
1006         ret = xive_source_lsi_trigger(xsrc, srcno);
1007     }
1008 
1009     return ret;
1010 }
1011 
1012 /*
1013  * Forward the source event notification to the Router
1014  */
1015 static void xive_source_notify(XiveSource *xsrc, int srcno)
1016 {
1017     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
1018     bool pq_checked = !xive_source_esb_disabled(xsrc, srcno);
1019 
1020     if (xnc->notify) {
1021         xnc->notify(xsrc->xive, srcno, pq_checked);
1022     }
1023 }
1024 
1025 /*
1026  * In a two pages ESB MMIO setting, even page is the trigger page, odd
1027  * page is for management
1028  */
1029 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
1030 {
1031     return !((addr >> shift) & 1);
1032 }
1033 
1034 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
1035 {
1036     return xive_source_esb_has_2page(xsrc) &&
1037         addr_is_even(addr, xsrc->esb_shift - 1);
1038 }
1039 
1040 /*
1041  * ESB MMIO loads
1042  *                      Trigger page    Management/EOI page
1043  *
1044  * ESB MMIO setting     2 pages         1 or 2 pages
1045  *
1046  * 0x000 .. 0x3FF       -1              EOI and return 0|1
1047  * 0x400 .. 0x7FF       -1              EOI and return 0|1
1048  * 0x800 .. 0xBFF       -1              return PQ
1049  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
1050  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
1051  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
1052  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
1053  */
1054 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
1055 {
1056     XiveSource *xsrc = XIVE_SOURCE(opaque);
1057     uint32_t offset = addr & 0xFFF;
1058     uint32_t srcno = addr >> xsrc->esb_shift;
1059     uint64_t ret = -1;
1060 
1061     /* In a two pages ESB MMIO setting, trigger page should not be read */
1062     if (xive_source_is_trigger_page(xsrc, addr)) {
1063         qemu_log_mask(LOG_GUEST_ERROR,
1064                       "XIVE: invalid load on IRQ %d trigger page at "
1065                       "0x%"HWADDR_PRIx"\n", srcno, addr);
1066         return -1;
1067     }
1068 
1069     switch (offset) {
1070     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1071         ret = xive_source_esb_eoi(xsrc, srcno);
1072 
1073         /* Forward the source event notification for routing */
1074         if (ret) {
1075             xive_source_notify(xsrc, srcno);
1076         }
1077         break;
1078 
1079     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1080         ret = xive_source_esb_get(xsrc, srcno);
1081         break;
1082 
1083     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1084     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1085     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1086     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1087         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1088         break;
1089     default:
1090         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
1091                       offset);
1092     }
1093 
1094     trace_xive_source_esb_read(addr, srcno, ret);
1095 
1096     return ret;
1097 }
1098 
1099 /*
1100  * ESB MMIO stores
1101  *                      Trigger page    Management/EOI page
1102  *
1103  * ESB MMIO setting     2 pages         1 or 2 pages
1104  *
1105  * 0x000 .. 0x3FF       Trigger         Trigger
1106  * 0x400 .. 0x7FF       Trigger         EOI
1107  * 0x800 .. 0xBFF       Trigger         undefined
1108  * 0xC00 .. 0xCFF       Trigger         PQ=00
1109  * 0xD00 .. 0xDFF       Trigger         PQ=01
1110  * 0xE00 .. 0xDFF       Trigger         PQ=10
1111  * 0xF00 .. 0xDFF       Trigger         PQ=11
1112  */
1113 static void xive_source_esb_write(void *opaque, hwaddr addr,
1114                                   uint64_t value, unsigned size)
1115 {
1116     XiveSource *xsrc = XIVE_SOURCE(opaque);
1117     uint32_t offset = addr & 0xFFF;
1118     uint32_t srcno = addr >> xsrc->esb_shift;
1119     bool notify = false;
1120 
1121     trace_xive_source_esb_write(addr, srcno, value);
1122 
1123     /* In a two pages ESB MMIO setting, trigger page only triggers */
1124     if (xive_source_is_trigger_page(xsrc, addr)) {
1125         notify = xive_source_esb_trigger(xsrc, srcno);
1126         goto out;
1127     }
1128 
1129     switch (offset) {
1130     case 0 ... 0x3FF:
1131         notify = xive_source_esb_trigger(xsrc, srcno);
1132         break;
1133 
1134     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
1135         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
1136             qemu_log_mask(LOG_GUEST_ERROR,
1137                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
1138             return;
1139         }
1140 
1141         notify = xive_source_esb_eoi(xsrc, srcno);
1142         break;
1143 
1144     /*
1145      * This is an internal offset used to inject triggers when the PQ
1146      * state bits are not controlled locally. Such as for LSIs when
1147      * under ABT mode.
1148      */
1149     case XIVE_ESB_INJECT ... XIVE_ESB_INJECT + 0x3FF:
1150         notify = true;
1151         break;
1152 
1153     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1154     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1155     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1156     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1157         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1158         break;
1159 
1160     default:
1161         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
1162                       offset);
1163         return;
1164     }
1165 
1166 out:
1167     /* Forward the source event notification for routing */
1168     if (notify) {
1169         xive_source_notify(xsrc, srcno);
1170     }
1171 }
1172 
1173 static const MemoryRegionOps xive_source_esb_ops = {
1174     .read = xive_source_esb_read,
1175     .write = xive_source_esb_write,
1176     .endianness = DEVICE_BIG_ENDIAN,
1177     .valid = {
1178         .min_access_size = 1,
1179         .max_access_size = 8,
1180     },
1181     .impl = {
1182         .min_access_size = 1,
1183         .max_access_size = 8,
1184     },
1185 };
1186 
1187 void xive_source_set_irq(void *opaque, int srcno, int val)
1188 {
1189     XiveSource *xsrc = XIVE_SOURCE(opaque);
1190     bool notify = false;
1191 
1192     if (xive_source_irq_is_lsi(xsrc, srcno)) {
1193         if (val) {
1194             notify = xive_source_lsi_trigger(xsrc, srcno);
1195         } else {
1196             xive_source_set_asserted(xsrc, srcno, false);
1197         }
1198     } else {
1199         if (val) {
1200             notify = xive_source_esb_trigger(xsrc, srcno);
1201         }
1202     }
1203 
1204     /* Forward the source event notification for routing */
1205     if (notify) {
1206         xive_source_notify(xsrc, srcno);
1207     }
1208 }
1209 
1210 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, GString *buf)
1211 {
1212     for (unsigned i = 0; i < xsrc->nr_irqs; i++) {
1213         uint8_t pq = xive_source_esb_get(xsrc, i);
1214 
1215         if (pq == XIVE_ESB_OFF) {
1216             continue;
1217         }
1218 
1219         g_string_append_printf(buf, "  %08x %s %c%c%c\n", i + offset,
1220                                xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
1221                                pq & XIVE_ESB_VAL_P ? 'P' : '-',
1222                                pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1223                                xive_source_is_asserted(xsrc, i) ? 'A' : ' ');
1224     }
1225 }
1226 
1227 static void xive_source_reset(void *dev)
1228 {
1229     XiveSource *xsrc = XIVE_SOURCE(dev);
1230 
1231     /* Do not clear the LSI bitmap */
1232 
1233     memset(xsrc->status, xsrc->reset_pq, xsrc->nr_irqs);
1234 }
1235 
1236 static void xive_source_realize(DeviceState *dev, Error **errp)
1237 {
1238     XiveSource *xsrc = XIVE_SOURCE(dev);
1239     size_t esb_len = xive_source_esb_len(xsrc);
1240 
1241     assert(xsrc->xive);
1242 
1243     if (!xsrc->nr_irqs) {
1244         error_setg(errp, "Number of interrupt needs to be greater than 0");
1245         return;
1246     }
1247 
1248     if (xsrc->esb_shift != XIVE_ESB_4K &&
1249         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1250         xsrc->esb_shift != XIVE_ESB_64K &&
1251         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1252         error_setg(errp, "Invalid ESB shift setting");
1253         return;
1254     }
1255 
1256     xsrc->status = g_malloc0(xsrc->nr_irqs);
1257     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1258 
1259     memory_region_init(&xsrc->esb_mmio, OBJECT(xsrc), "xive.esb", esb_len);
1260     memory_region_init_io(&xsrc->esb_mmio_emulated, OBJECT(xsrc),
1261                           &xive_source_esb_ops, xsrc, "xive.esb-emulated",
1262                           esb_len);
1263     memory_region_add_subregion(&xsrc->esb_mmio, 0, &xsrc->esb_mmio_emulated);
1264 
1265     qemu_register_reset(xive_source_reset, dev);
1266 }
1267 
1268 static const VMStateDescription vmstate_xive_source = {
1269     .name = TYPE_XIVE_SOURCE,
1270     .version_id = 1,
1271     .minimum_version_id = 1,
1272     .fields = (const VMStateField[]) {
1273         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1274         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1275         VMSTATE_END_OF_LIST()
1276     },
1277 };
1278 
1279 /*
1280  * The default XIVE interrupt source setting for the ESB MMIOs is two
1281  * 64k pages without Store EOI, to be in sync with KVM.
1282  */
1283 static Property xive_source_properties[] = {
1284     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1285     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1286     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1287     /*
1288      * By default, PQs are initialized to 0b01 (Q=1) which corresponds
1289      * to "ints off"
1290      */
1291     DEFINE_PROP_UINT8("reset-pq", XiveSource, reset_pq, XIVE_ESB_OFF),
1292     DEFINE_PROP_LINK("xive", XiveSource, xive, TYPE_XIVE_NOTIFIER,
1293                      XiveNotifier *),
1294     DEFINE_PROP_END_OF_LIST(),
1295 };
1296 
1297 static void xive_source_class_init(ObjectClass *klass, void *data)
1298 {
1299     DeviceClass *dc = DEVICE_CLASS(klass);
1300 
1301     dc->desc    = "XIVE Interrupt Source";
1302     device_class_set_props(dc, xive_source_properties);
1303     dc->realize = xive_source_realize;
1304     dc->vmsd    = &vmstate_xive_source;
1305     /*
1306      * Reason: part of XIVE interrupt controller, needs to be wired up,
1307      * e.g. by spapr_xive_instance_init().
1308      */
1309     dc->user_creatable = false;
1310 }
1311 
1312 static const TypeInfo xive_source_info = {
1313     .name          = TYPE_XIVE_SOURCE,
1314     .parent        = TYPE_DEVICE,
1315     .instance_size = sizeof(XiveSource),
1316     .class_init    = xive_source_class_init,
1317 };
1318 
1319 /*
1320  * XiveEND helpers
1321  */
1322 
1323 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, GString *buf)
1324 {
1325     uint64_t qaddr_base = xive_end_qaddr(end);
1326     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1327     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1328     uint32_t qentries = 1 << (qsize + 10);
1329     int i;
1330 
1331     /*
1332      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1333      */
1334     g_string_append_printf(buf, " [ ");
1335     qindex = (qindex - (width - 1)) & (qentries - 1);
1336     for (i = 0; i < width; i++) {
1337         uint64_t qaddr = qaddr_base + (qindex << 2);
1338         uint32_t qdata = -1;
1339 
1340         if (dma_memory_read(&address_space_memory, qaddr,
1341                             &qdata, sizeof(qdata), MEMTXATTRS_UNSPECIFIED)) {
1342             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1343                           HWADDR_PRIx "\n", qaddr);
1344             return;
1345         }
1346         g_string_append_printf(buf, "%s%08x ", i == width - 1 ? "^" : "",
1347                                be32_to_cpu(qdata));
1348         qindex = (qindex + 1) & (qentries - 1);
1349     }
1350     g_string_append_c(buf, ']');
1351 }
1352 
1353 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, GString *buf)
1354 {
1355     uint64_t qaddr_base = xive_end_qaddr(end);
1356     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1357     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1358     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1359     uint32_t qentries = 1 << (qsize + 10);
1360 
1361     uint32_t nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1362     uint32_t nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1363     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1364     uint8_t pq;
1365 
1366     if (!xive_end_is_valid(end)) {
1367         return;
1368     }
1369 
1370     pq = xive_get_field32(END_W1_ESn, end->w1);
1371 
1372     g_string_append_printf(buf,
1373                            "  %08x %c%c %c%c%c%c%c%c%c%c prio:%d nvt:%02x/%04x",
1374                            end_idx,
1375                            pq & XIVE_ESB_VAL_P ? 'P' : '-',
1376                            pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1377                            xive_end_is_valid(end)    ? 'v' : '-',
1378                            xive_end_is_enqueue(end)  ? 'q' : '-',
1379                            xive_end_is_notify(end)   ? 'n' : '-',
1380                            xive_end_is_backlog(end)  ? 'b' : '-',
1381                            xive_end_is_escalate(end) ? 'e' : '-',
1382                            xive_end_is_uncond_escalation(end)   ? 'u' : '-',
1383                            xive_end_is_silent_escalation(end)   ? 's' : '-',
1384                            xive_end_is_firmware(end)   ? 'f' : '-',
1385                            priority, nvt_blk, nvt_idx);
1386 
1387     if (qaddr_base) {
1388         g_string_append_printf(buf, " eq:@%08"PRIx64"% 6d/%5d ^%d",
1389                                qaddr_base, qindex, qentries, qgen);
1390         xive_end_queue_pic_print_info(end, 6, buf);
1391     }
1392     g_string_append_c(buf, '\n');
1393 }
1394 
1395 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1396 {
1397     uint64_t qaddr_base = xive_end_qaddr(end);
1398     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1399     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1400     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1401 
1402     uint64_t qaddr = qaddr_base + (qindex << 2);
1403     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1404     uint32_t qentries = 1 << (qsize + 10);
1405 
1406     if (dma_memory_write(&address_space_memory, qaddr,
1407                          &qdata, sizeof(qdata), MEMTXATTRS_UNSPECIFIED)) {
1408         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1409                       HWADDR_PRIx "\n", qaddr);
1410         return;
1411     }
1412 
1413     qindex = (qindex + 1) & (qentries - 1);
1414     if (qindex == 0) {
1415         qgen ^= 1;
1416         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1417     }
1418     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1419 }
1420 
1421 void xive_end_eas_pic_print_info(XiveEND *end, uint32_t end_idx, GString *buf)
1422 {
1423     XiveEAS *eas = (XiveEAS *) &end->w4;
1424     uint8_t pq;
1425 
1426     if (!xive_end_is_escalate(end)) {
1427         return;
1428     }
1429 
1430     pq = xive_get_field32(END_W1_ESe, end->w1);
1431 
1432     g_string_append_printf(buf, "  %08x %c%c %c%c end:%02x/%04x data:%08x\n",
1433                            end_idx,
1434                            pq & XIVE_ESB_VAL_P ? 'P' : '-',
1435                            pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1436                            xive_eas_is_valid(eas) ? 'V' : ' ',
1437                            xive_eas_is_masked(eas) ? 'M' : ' ',
1438                            (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1439                            (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1440                            (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1441 }
1442 
1443 /*
1444  * XIVE Router (aka. Virtualization Controller or IVRE)
1445  */
1446 
1447 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1448                         XiveEAS *eas)
1449 {
1450     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1451 
1452     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1453 }
1454 
1455 static
1456 int xive_router_get_pq(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1457                        uint8_t *pq)
1458 {
1459     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1460 
1461     return xrc->get_pq(xrtr, eas_blk, eas_idx, pq);
1462 }
1463 
1464 static
1465 int xive_router_set_pq(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1466                        uint8_t *pq)
1467 {
1468     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1469 
1470     return xrc->set_pq(xrtr, eas_blk, eas_idx, pq);
1471 }
1472 
1473 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1474                         XiveEND *end)
1475 {
1476    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1477 
1478    return xrc->get_end(xrtr, end_blk, end_idx, end);
1479 }
1480 
1481 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1482                           XiveEND *end, uint8_t word_number)
1483 {
1484    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1485 
1486    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1487 }
1488 
1489 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1490                         XiveNVT *nvt)
1491 {
1492    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1493 
1494    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1495 }
1496 
1497 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1498                         XiveNVT *nvt, uint8_t word_number)
1499 {
1500    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1501 
1502    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1503 }
1504 
1505 static int xive_router_get_block_id(XiveRouter *xrtr)
1506 {
1507    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1508 
1509    return xrc->get_block_id(xrtr);
1510 }
1511 
1512 static void xive_router_realize(DeviceState *dev, Error **errp)
1513 {
1514     XiveRouter *xrtr = XIVE_ROUTER(dev);
1515 
1516     assert(xrtr->xfb);
1517 }
1518 
1519 static void xive_router_end_notify_handler(XiveRouter *xrtr, XiveEAS *eas)
1520 {
1521     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1522 
1523     return xrc->end_notify(xrtr, eas);
1524 }
1525 
1526 /*
1527  * Encode the HW CAM line in the block group mode format :
1528  *
1529  *   chip << 19 | 0000000 0 0001 thread (7Bit)
1530  */
1531 static uint32_t xive_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
1532 {
1533     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1534     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1535     uint8_t blk = xive_router_get_block_id(XIVE_ROUTER(xptr));
1536 
1537     return xive_nvt_cam_line(blk, 1 << 7 | (pir & 0x7f));
1538 }
1539 
1540 /*
1541  * The thread context register words are in big-endian format.
1542  */
1543 int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1544                               uint8_t format,
1545                               uint8_t nvt_blk, uint32_t nvt_idx,
1546                               bool cam_ignore, uint32_t logic_serv)
1547 {
1548     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1549     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1550     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1551     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1552     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1553 
1554     /*
1555      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1556      * identifier are ignored in the "CAM" match.
1557      */
1558 
1559     if (format == 0) {
1560         if (cam_ignore == true) {
1561             /*
1562              * F=0 & i=1: Logical server notification (bits ignored at
1563              * the end of the NVT identifier)
1564              */
1565             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1566                           nvt_blk, nvt_idx);
1567              return -1;
1568         }
1569 
1570         /* F=0 & i=0: Specific NVT notification */
1571 
1572         /* PHYS ring */
1573         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1574             cam == xive_tctx_hw_cam_line(xptr, tctx)) {
1575             return TM_QW3_HV_PHYS;
1576         }
1577 
1578         /* HV POOL ring */
1579         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1580             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1581             return TM_QW2_HV_POOL;
1582         }
1583 
1584         /* OS ring */
1585         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1586             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1587             return TM_QW1_OS;
1588         }
1589     } else {
1590         /* F=1 : User level Event-Based Branch (EBB) notification */
1591 
1592         /* USER ring */
1593         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1594              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1595              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1596              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1597             return TM_QW0_USER;
1598         }
1599     }
1600     return -1;
1601 }
1602 
1603 /*
1604  * This is our simple Xive Presenter Engine model. It is merged in the
1605  * Router as it does not require an extra object.
1606  *
1607  * It receives notification requests sent by the IVRE to find one
1608  * matching NVT (or more) dispatched on the processor threads. In case
1609  * of a single NVT notification, the process is abbreviated and the
1610  * thread is signaled if a match is found. In case of a logical server
1611  * notification (bits ignored at the end of the NVT identifier), the
1612  * IVPE and IVRE select a winning thread using different filters. This
1613  * involves 2 or 3 exchanges on the PowerBus that the model does not
1614  * support.
1615  *
1616  * The parameters represent what is sent on the PowerBus
1617  */
1618 bool xive_presenter_notify(XiveFabric *xfb, uint8_t format,
1619                            uint8_t nvt_blk, uint32_t nvt_idx,
1620                            bool cam_ignore, uint8_t priority,
1621                            uint32_t logic_serv)
1622 {
1623     XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xfb);
1624     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1625     int count;
1626 
1627     /*
1628      * Ask the machine to scan the interrupt controllers for a match
1629      */
1630     count = xfc->match_nvt(xfb, format, nvt_blk, nvt_idx, cam_ignore,
1631                            priority, logic_serv, &match);
1632     if (count < 0) {
1633         return false;
1634     }
1635 
1636     /* handle CPU exception delivery */
1637     if (count) {
1638         trace_xive_presenter_notify(nvt_blk, nvt_idx, match.ring);
1639         xive_tctx_ipb_update(match.tctx, match.ring,
1640                              xive_priority_to_ipb(priority));
1641     }
1642 
1643     return !!count;
1644 }
1645 
1646 /*
1647  * Notification using the END ESe/ESn bit (Event State Buffer for
1648  * escalation and notification). Provide further coalescing in the
1649  * Router.
1650  */
1651 static bool xive_router_end_es_notify(XiveRouter *xrtr, uint8_t end_blk,
1652                                       uint32_t end_idx, XiveEND *end,
1653                                       uint32_t end_esmask)
1654 {
1655     uint8_t pq = xive_get_field32(end_esmask, end->w1);
1656     bool notify = xive_esb_trigger(&pq);
1657 
1658     if (pq != xive_get_field32(end_esmask, end->w1)) {
1659         end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1660         xive_router_write_end(xrtr, end_blk, end_idx, end, 1);
1661     }
1662 
1663     /* ESe/n[Q]=1 : end of notification */
1664     return notify;
1665 }
1666 
1667 /*
1668  * An END trigger can come from an event trigger (IPI or HW) or from
1669  * another chip. We don't model the PowerBus but the END trigger
1670  * message has the same parameters than in the function below.
1671  */
1672 void xive_router_end_notify(XiveRouter *xrtr, XiveEAS *eas)
1673 {
1674     XiveEND end;
1675     uint8_t priority;
1676     uint8_t format;
1677     uint8_t nvt_blk;
1678     uint32_t nvt_idx;
1679     XiveNVT nvt;
1680     bool found;
1681 
1682     uint8_t end_blk = xive_get_field64(EAS_END_BLOCK, eas->w);
1683     uint32_t end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
1684     uint32_t end_data = xive_get_field64(EAS_END_DATA,  eas->w);
1685 
1686     /* END cache lookup */
1687     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1688         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1689                       end_idx);
1690         return;
1691     }
1692 
1693     if (!xive_end_is_valid(&end)) {
1694         trace_xive_router_end_notify(end_blk, end_idx, end_data);
1695         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1696                       end_blk, end_idx);
1697         return;
1698     }
1699 
1700     if (xive_end_is_enqueue(&end)) {
1701         xive_end_enqueue(&end, end_data);
1702         /* Enqueuing event data modifies the EQ toggle and index */
1703         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1704     }
1705 
1706     /*
1707      * When the END is silent, we skip the notification part.
1708      */
1709     if (xive_end_is_silent_escalation(&end)) {
1710         goto do_escalation;
1711     }
1712 
1713     /*
1714      * The W7 format depends on the F bit in W6. It defines the type
1715      * of the notification :
1716      *
1717      *   F=0 : single or multiple NVT notification
1718      *   F=1 : User level Event-Based Branch (EBB) notification, no
1719      *         priority
1720      */
1721     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1722     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1723 
1724     /* The END is masked */
1725     if (format == 0 && priority == 0xff) {
1726         return;
1727     }
1728 
1729     /*
1730      * Check the END ESn (Event State Buffer for notification) for
1731      * even further coalescing in the Router
1732      */
1733     if (!xive_end_is_notify(&end)) {
1734         /* ESn[Q]=1 : end of notification */
1735         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1736                                        &end, END_W1_ESn)) {
1737             return;
1738         }
1739     }
1740 
1741     /*
1742      * Follows IVPE notification
1743      */
1744     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end.w6);
1745     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end.w6);
1746 
1747     /* NVT cache lookup */
1748     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1749         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1750                       nvt_blk, nvt_idx);
1751         return;
1752     }
1753 
1754     if (!xive_nvt_is_valid(&nvt)) {
1755         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1756                       nvt_blk, nvt_idx);
1757         return;
1758     }
1759 
1760     found = xive_presenter_notify(xrtr->xfb, format, nvt_blk, nvt_idx,
1761                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1762                           priority,
1763                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1764 
1765     /* TODO: Auto EOI. */
1766 
1767     if (found) {
1768         return;
1769     }
1770 
1771     /*
1772      * If no matching NVT is dispatched on a HW thread :
1773      * - specific VP: update the NVT structure if backlog is activated
1774      * - logical server : forward request to IVPE (not supported)
1775      */
1776     if (xive_end_is_backlog(&end)) {
1777         uint8_t ipb;
1778 
1779         if (format == 1) {
1780             qemu_log_mask(LOG_GUEST_ERROR,
1781                           "XIVE: END %x/%x invalid config: F1 & backlog\n",
1782                           end_blk, end_idx);
1783             return;
1784         }
1785         /*
1786          * Record the IPB in the associated NVT structure for later
1787          * use. The presenter will resend the interrupt when the vCPU
1788          * is dispatched again on a HW thread.
1789          */
1790         ipb = xive_get_field32(NVT_W4_IPB, nvt.w4) |
1791             xive_priority_to_ipb(priority);
1792         nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, ipb);
1793         xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1794 
1795         /*
1796          * On HW, follows a "Broadcast Backlog" to IVPEs
1797          */
1798     }
1799 
1800 do_escalation:
1801     /*
1802      * If activated, escalate notification using the ESe PQ bits and
1803      * the EAS in w4-5
1804      */
1805     if (!xive_end_is_escalate(&end)) {
1806         return;
1807     }
1808 
1809     /*
1810      * Check the END ESe (Event State Buffer for escalation) for even
1811      * further coalescing in the Router
1812      */
1813     if (!xive_end_is_uncond_escalation(&end)) {
1814         /* ESe[Q]=1 : end of notification */
1815         if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1816                                        &end, END_W1_ESe)) {
1817             return;
1818         }
1819     }
1820 
1821     trace_xive_router_end_escalate(end_blk, end_idx,
1822            (uint8_t) xive_get_field32(END_W4_ESC_END_BLOCK, end.w4),
1823            (uint32_t) xive_get_field32(END_W4_ESC_END_INDEX, end.w4),
1824            (uint32_t) xive_get_field32(END_W5_ESC_END_DATA,  end.w5));
1825     /*
1826      * The END trigger becomes an Escalation trigger
1827      */
1828     xive_router_end_notify_handler(xrtr, (XiveEAS *) &end.w4);
1829 }
1830 
1831 void xive_router_notify(XiveNotifier *xn, uint32_t lisn, bool pq_checked)
1832 {
1833     XiveRouter *xrtr = XIVE_ROUTER(xn);
1834     uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1835     uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1836     XiveEAS eas;
1837 
1838     /* EAS cache lookup */
1839     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1840         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1841         return;
1842     }
1843 
1844     if (!pq_checked) {
1845         bool notify;
1846         uint8_t pq;
1847 
1848         /* PQ cache lookup */
1849         if (xive_router_get_pq(xrtr, eas_blk, eas_idx, &pq)) {
1850             /* Set FIR */
1851             g_assert_not_reached();
1852         }
1853 
1854         notify = xive_esb_trigger(&pq);
1855 
1856         if (xive_router_set_pq(xrtr, eas_blk, eas_idx, &pq)) {
1857             /* Set FIR */
1858             g_assert_not_reached();
1859         }
1860 
1861         if (!notify) {
1862             return;
1863         }
1864     }
1865 
1866     if (!xive_eas_is_valid(&eas)) {
1867         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1868         return;
1869     }
1870 
1871     if (xive_eas_is_masked(&eas)) {
1872         /* Notification completed */
1873         return;
1874     }
1875 
1876     /*
1877      * The event trigger becomes an END trigger
1878      */
1879     xive_router_end_notify_handler(xrtr, &eas);
1880 }
1881 
1882 static Property xive_router_properties[] = {
1883     DEFINE_PROP_LINK("xive-fabric", XiveRouter, xfb,
1884                      TYPE_XIVE_FABRIC, XiveFabric *),
1885     DEFINE_PROP_END_OF_LIST(),
1886 };
1887 
1888 static void xive_router_class_init(ObjectClass *klass, void *data)
1889 {
1890     DeviceClass *dc = DEVICE_CLASS(klass);
1891     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1892     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
1893 
1894     dc->desc    = "XIVE Router Engine";
1895     device_class_set_props(dc, xive_router_properties);
1896     /* Parent is SysBusDeviceClass. No need to call its realize hook */
1897     dc->realize = xive_router_realize;
1898     xnc->notify = xive_router_notify;
1899 
1900     /* By default, the router handles END triggers locally */
1901     xrc->end_notify = xive_router_end_notify;
1902 }
1903 
1904 static const TypeInfo xive_router_info = {
1905     .name          = TYPE_XIVE_ROUTER,
1906     .parent        = TYPE_SYS_BUS_DEVICE,
1907     .abstract      = true,
1908     .instance_size = sizeof(XiveRouter),
1909     .class_size    = sizeof(XiveRouterClass),
1910     .class_init    = xive_router_class_init,
1911     .interfaces    = (InterfaceInfo[]) {
1912         { TYPE_XIVE_NOTIFIER },
1913         { TYPE_XIVE_PRESENTER },
1914         { }
1915     }
1916 };
1917 
1918 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, GString *buf)
1919 {
1920     if (!xive_eas_is_valid(eas)) {
1921         return;
1922     }
1923 
1924     g_string_append_printf(buf, "  %08x %s end:%02x/%04x data:%08x\n",
1925                            lisn, xive_eas_is_masked(eas) ? "M" : " ",
1926                            (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1927                            (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1928                            (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1929 }
1930 
1931 /*
1932  * END ESB MMIO loads
1933  */
1934 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1935 {
1936     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1937     uint32_t offset = addr & 0xFFF;
1938     uint8_t end_blk;
1939     uint32_t end_idx;
1940     XiveEND end;
1941     uint32_t end_esmask;
1942     uint8_t pq;
1943     uint64_t ret = -1;
1944 
1945     /*
1946      * The block id should be deduced from the load address on the END
1947      * ESB MMIO but our model only supports a single block per XIVE chip.
1948      */
1949     end_blk = xive_router_get_block_id(xsrc->xrtr);
1950     end_idx = addr >> (xsrc->esb_shift + 1);
1951 
1952     trace_xive_end_source_read(end_blk, end_idx, addr);
1953 
1954     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1955         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1956                       end_idx);
1957         return -1;
1958     }
1959 
1960     if (!xive_end_is_valid(&end)) {
1961         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1962                       end_blk, end_idx);
1963         return -1;
1964     }
1965 
1966     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1967     pq = xive_get_field32(end_esmask, end.w1);
1968 
1969     switch (offset) {
1970     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1971         ret = xive_esb_eoi(&pq);
1972 
1973         /* Forward the source event notification for routing ?? */
1974         break;
1975 
1976     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1977         ret = pq;
1978         break;
1979 
1980     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1981     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1982     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1983     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1984         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1985         break;
1986     default:
1987         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1988                       offset);
1989         return -1;
1990     }
1991 
1992     if (pq != xive_get_field32(end_esmask, end.w1)) {
1993         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1994         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1995     }
1996 
1997     return ret;
1998 }
1999 
2000 /*
2001  * END ESB MMIO stores are invalid
2002  */
2003 static void xive_end_source_write(void *opaque, hwaddr addr,
2004                                   uint64_t value, unsigned size)
2005 {
2006     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
2007                   HWADDR_PRIx"\n", addr);
2008 }
2009 
2010 static const MemoryRegionOps xive_end_source_ops = {
2011     .read = xive_end_source_read,
2012     .write = xive_end_source_write,
2013     .endianness = DEVICE_BIG_ENDIAN,
2014     .valid = {
2015         .min_access_size = 1,
2016         .max_access_size = 8,
2017     },
2018     .impl = {
2019         .min_access_size = 1,
2020         .max_access_size = 8,
2021     },
2022 };
2023 
2024 static void xive_end_source_realize(DeviceState *dev, Error **errp)
2025 {
2026     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
2027 
2028     assert(xsrc->xrtr);
2029 
2030     if (!xsrc->nr_ends) {
2031         error_setg(errp, "Number of interrupt needs to be greater than 0");
2032         return;
2033     }
2034 
2035     if (xsrc->esb_shift != XIVE_ESB_4K &&
2036         xsrc->esb_shift != XIVE_ESB_64K) {
2037         error_setg(errp, "Invalid ESB shift setting");
2038         return;
2039     }
2040 
2041     /*
2042      * Each END is assigned an even/odd pair of MMIO pages, the even page
2043      * manages the ESn field while the odd page manages the ESe field.
2044      */
2045     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
2046                           &xive_end_source_ops, xsrc, "xive.end",
2047                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
2048 }
2049 
2050 static Property xive_end_source_properties[] = {
2051     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
2052     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
2053     DEFINE_PROP_LINK("xive", XiveENDSource, xrtr, TYPE_XIVE_ROUTER,
2054                      XiveRouter *),
2055     DEFINE_PROP_END_OF_LIST(),
2056 };
2057 
2058 static void xive_end_source_class_init(ObjectClass *klass, void *data)
2059 {
2060     DeviceClass *dc = DEVICE_CLASS(klass);
2061 
2062     dc->desc    = "XIVE END Source";
2063     device_class_set_props(dc, xive_end_source_properties);
2064     dc->realize = xive_end_source_realize;
2065     /*
2066      * Reason: part of XIVE interrupt controller, needs to be wired up,
2067      * e.g. by spapr_xive_instance_init().
2068      */
2069     dc->user_creatable = false;
2070 }
2071 
2072 static const TypeInfo xive_end_source_info = {
2073     .name          = TYPE_XIVE_END_SOURCE,
2074     .parent        = TYPE_DEVICE,
2075     .instance_size = sizeof(XiveENDSource),
2076     .class_init    = xive_end_source_class_init,
2077 };
2078 
2079 /*
2080  * XIVE Notifier
2081  */
2082 static const TypeInfo xive_notifier_info = {
2083     .name = TYPE_XIVE_NOTIFIER,
2084     .parent = TYPE_INTERFACE,
2085     .class_size = sizeof(XiveNotifierClass),
2086 };
2087 
2088 /*
2089  * XIVE Presenter
2090  */
2091 static const TypeInfo xive_presenter_info = {
2092     .name = TYPE_XIVE_PRESENTER,
2093     .parent = TYPE_INTERFACE,
2094     .class_size = sizeof(XivePresenterClass),
2095 };
2096 
2097 /*
2098  * XIVE Fabric
2099  */
2100 static const TypeInfo xive_fabric_info = {
2101     .name = TYPE_XIVE_FABRIC,
2102     .parent = TYPE_INTERFACE,
2103     .class_size = sizeof(XiveFabricClass),
2104 };
2105 
2106 static void xive_register_types(void)
2107 {
2108     type_register_static(&xive_fabric_info);
2109     type_register_static(&xive_source_info);
2110     type_register_static(&xive_notifier_info);
2111     type_register_static(&xive_presenter_info);
2112     type_register_static(&xive_router_info);
2113     type_register_static(&xive_end_source_info);
2114     type_register_static(&xive_tctx_info);
2115 }
2116 
2117 type_init(xive_register_types)
2118