xref: /openbmc/qemu/hw/intc/xive.c (revision 5cc8767d)
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 "hw/qdev-properties.h"
18 #include "monitor/monitor.h"
19 #include "hw/ppc/xive.h"
20 #include "hw/ppc/xive_regs.h"
21 
22 /*
23  * XIVE Thread Interrupt Management context
24  */
25 
26 /*
27  * Convert a priority number to an Interrupt Pending Buffer (IPB)
28  * register, which indicates a pending interrupt at the priority
29  * corresponding to the bit number
30  */
31 static uint8_t priority_to_ipb(uint8_t priority)
32 {
33     return priority > XIVE_PRIORITY_MAX ?
34         0 : 1 << (XIVE_PRIORITY_MAX - priority);
35 }
36 
37 /*
38  * Convert an Interrupt Pending Buffer (IPB) register to a Pending
39  * Interrupt Priority Register (PIPR), which contains the priority of
40  * the most favored pending notification.
41  */
42 static uint8_t ipb_to_pipr(uint8_t ibp)
43 {
44     return ibp ? clz32((uint32_t)ibp << 24) : 0xff;
45 }
46 
47 static void ipb_update(uint8_t *regs, uint8_t priority)
48 {
49     regs[TM_IPB] |= priority_to_ipb(priority);
50     regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
51 }
52 
53 static uint8_t exception_mask(uint8_t ring)
54 {
55     switch (ring) {
56     case TM_QW1_OS:
57         return TM_QW1_NSR_EO;
58     case TM_QW3_HV_PHYS:
59         return TM_QW3_NSR_HE;
60     default:
61         g_assert_not_reached();
62     }
63 }
64 
65 static qemu_irq xive_tctx_output(XiveTCTX *tctx, uint8_t ring)
66 {
67         switch (ring) {
68         case TM_QW0_USER:
69                 return 0; /* Not supported */
70         case TM_QW1_OS:
71                 return tctx->os_output;
72         case TM_QW2_HV_POOL:
73         case TM_QW3_HV_PHYS:
74                 return tctx->hv_output;
75         default:
76                 return 0;
77         }
78 }
79 
80 static uint64_t xive_tctx_accept(XiveTCTX *tctx, uint8_t ring)
81 {
82     uint8_t *regs = &tctx->regs[ring];
83     uint8_t nsr = regs[TM_NSR];
84     uint8_t mask = exception_mask(ring);
85 
86     qemu_irq_lower(xive_tctx_output(tctx, ring));
87 
88     if (regs[TM_NSR] & mask) {
89         uint8_t cppr = regs[TM_PIPR];
90 
91         regs[TM_CPPR] = cppr;
92 
93         /* Reset the pending buffer bit */
94         regs[TM_IPB] &= ~priority_to_ipb(cppr);
95         regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
96 
97         /* Drop Exception bit */
98         regs[TM_NSR] &= ~mask;
99     }
100 
101     return (nsr << 8) | regs[TM_CPPR];
102 }
103 
104 static void xive_tctx_notify(XiveTCTX *tctx, uint8_t ring)
105 {
106     uint8_t *regs = &tctx->regs[ring];
107 
108     if (regs[TM_PIPR] < regs[TM_CPPR]) {
109         switch (ring) {
110         case TM_QW1_OS:
111             regs[TM_NSR] |= TM_QW1_NSR_EO;
112             break;
113         case TM_QW3_HV_PHYS:
114             regs[TM_NSR] |= (TM_QW3_NSR_HE_PHYS << 6);
115             break;
116         default:
117             g_assert_not_reached();
118         }
119         qemu_irq_raise(xive_tctx_output(tctx, ring));
120     }
121 }
122 
123 static void xive_tctx_set_cppr(XiveTCTX *tctx, uint8_t ring, uint8_t cppr)
124 {
125     if (cppr > XIVE_PRIORITY_MAX) {
126         cppr = 0xff;
127     }
128 
129     tctx->regs[ring + TM_CPPR] = cppr;
130 
131     /* CPPR has changed, check if we need to raise a pending exception */
132     xive_tctx_notify(tctx, ring);
133 }
134 
135 static inline uint32_t xive_tctx_word2(uint8_t *ring)
136 {
137     return *((uint32_t *) &ring[TM_WORD2]);
138 }
139 
140 /*
141  * XIVE Thread Interrupt Management Area (TIMA)
142  */
143 
144 static void xive_tm_set_hv_cppr(XiveTCTX *tctx, hwaddr offset,
145                                 uint64_t value, unsigned size)
146 {
147     xive_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
148 }
149 
150 static uint64_t xive_tm_ack_hv_reg(XiveTCTX *tctx, hwaddr offset, unsigned size)
151 {
152     return xive_tctx_accept(tctx, TM_QW3_HV_PHYS);
153 }
154 
155 static uint64_t xive_tm_pull_pool_ctx(XiveTCTX *tctx, hwaddr offset,
156                                       unsigned size)
157 {
158     uint32_t qw2w2_prev = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
159     uint32_t qw2w2;
160 
161     qw2w2 = xive_set_field32(TM_QW2W2_VP, qw2w2_prev, 0);
162     memcpy(&tctx->regs[TM_QW2_HV_POOL + TM_WORD2], &qw2w2, 4);
163     return qw2w2;
164 }
165 
166 static void xive_tm_vt_push(XiveTCTX *tctx, hwaddr offset,
167                             uint64_t value, unsigned size)
168 {
169     tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] = value & 0xff;
170 }
171 
172 static uint64_t xive_tm_vt_poll(XiveTCTX *tctx, hwaddr offset, unsigned size)
173 {
174     return tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] & 0xff;
175 }
176 
177 /*
178  * Define an access map for each page of the TIMA that we will use in
179  * the memory region ops to filter values when doing loads and stores
180  * of raw registers values
181  *
182  * Registers accessibility bits :
183  *
184  *    0x0 - no access
185  *    0x1 - write only
186  *    0x2 - read only
187  *    0x3 - read/write
188  */
189 
190 static const uint8_t xive_tm_hw_view[] = {
191     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
192     3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
193     0, 0, 3, 3,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
194     3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   3, 3, 3, 0, /* QW-3 PHYS */
195 };
196 
197 static const uint8_t xive_tm_hv_view[] = {
198     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
199     3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
200     0, 0, 3, 3,   0, 0, 0, 0,   0, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
201     3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   0, 0, 0, 0, /* QW-3 PHYS */
202 };
203 
204 static const uint8_t xive_tm_os_view[] = {
205     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
206     2, 3, 2, 2,   2, 2, 0, 2,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
207     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
208     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
209 };
210 
211 static const uint8_t xive_tm_user_view[] = {
212     3, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-0 User */
213     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
214     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
215     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
216 };
217 
218 /*
219  * Overall TIMA access map for the thread interrupt management context
220  * registers
221  */
222 static const uint8_t *xive_tm_views[] = {
223     [XIVE_TM_HW_PAGE]   = xive_tm_hw_view,
224     [XIVE_TM_HV_PAGE]   = xive_tm_hv_view,
225     [XIVE_TM_OS_PAGE]   = xive_tm_os_view,
226     [XIVE_TM_USER_PAGE] = xive_tm_user_view,
227 };
228 
229 /*
230  * Computes a register access mask for a given offset in the TIMA
231  */
232 static uint64_t xive_tm_mask(hwaddr offset, unsigned size, bool write)
233 {
234     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
235     uint8_t reg_offset = offset & 0x3F;
236     uint8_t reg_mask = write ? 0x1 : 0x2;
237     uint64_t mask = 0x0;
238     int i;
239 
240     for (i = 0; i < size; i++) {
241         if (xive_tm_views[page_offset][reg_offset + i] & reg_mask) {
242             mask |= (uint64_t) 0xff << (8 * (size - i - 1));
243         }
244     }
245 
246     return mask;
247 }
248 
249 static void xive_tm_raw_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
250                               unsigned size)
251 {
252     uint8_t ring_offset = offset & 0x30;
253     uint8_t reg_offset = offset & 0x3F;
254     uint64_t mask = xive_tm_mask(offset, size, true);
255     int i;
256 
257     /*
258      * Only 4 or 8 bytes stores are allowed and the User ring is
259      * excluded
260      */
261     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
262         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA @%"
263                       HWADDR_PRIx"\n", offset);
264         return;
265     }
266 
267     /*
268      * Use the register offset for the raw values and filter out
269      * reserved values
270      */
271     for (i = 0; i < size; i++) {
272         uint8_t byte_mask = (mask >> (8 * (size - i - 1)));
273         if (byte_mask) {
274             tctx->regs[reg_offset + i] = (value >> (8 * (size - i - 1))) &
275                 byte_mask;
276         }
277     }
278 }
279 
280 static uint64_t xive_tm_raw_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
281 {
282     uint8_t ring_offset = offset & 0x30;
283     uint8_t reg_offset = offset & 0x3F;
284     uint64_t mask = xive_tm_mask(offset, size, false);
285     uint64_t ret;
286     int i;
287 
288     /*
289      * Only 4 or 8 bytes loads are allowed and the User ring is
290      * excluded
291      */
292     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
293         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access at TIMA @%"
294                       HWADDR_PRIx"\n", offset);
295         return -1;
296     }
297 
298     /* Use the register offset for the raw values */
299     ret = 0;
300     for (i = 0; i < size; i++) {
301         ret |= (uint64_t) tctx->regs[reg_offset + i] << (8 * (size - i - 1));
302     }
303 
304     /* filter out reserved values */
305     return ret & mask;
306 }
307 
308 /*
309  * The TM context is mapped twice within each page. Stores and loads
310  * to the first mapping below 2K write and read the specified values
311  * without modification. The second mapping above 2K performs specific
312  * state changes (side effects) in addition to setting/returning the
313  * interrupt management area context of the processor thread.
314  */
315 static uint64_t xive_tm_ack_os_reg(XiveTCTX *tctx, hwaddr offset, unsigned size)
316 {
317     return xive_tctx_accept(tctx, TM_QW1_OS);
318 }
319 
320 static void xive_tm_set_os_cppr(XiveTCTX *tctx, hwaddr offset,
321                                 uint64_t value, unsigned size)
322 {
323     xive_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
324 }
325 
326 /*
327  * Adjust the IPB to allow a CPU to process event queues of other
328  * priorities during one physical interrupt cycle.
329  */
330 static void xive_tm_set_os_pending(XiveTCTX *tctx, hwaddr offset,
331                                    uint64_t value, unsigned size)
332 {
333     ipb_update(&tctx->regs[TM_QW1_OS], value & 0xff);
334     xive_tctx_notify(tctx, TM_QW1_OS);
335 }
336 
337 /*
338  * Define a mapping of "special" operations depending on the TIMA page
339  * offset and the size of the operation.
340  */
341 typedef struct XiveTmOp {
342     uint8_t  page_offset;
343     uint32_t op_offset;
344     unsigned size;
345     void     (*write_handler)(XiveTCTX *tctx, hwaddr offset, uint64_t value,
346                               unsigned size);
347     uint64_t (*read_handler)(XiveTCTX *tctx, hwaddr offset, unsigned size);
348 } XiveTmOp;
349 
350 static const XiveTmOp xive_tm_operations[] = {
351     /*
352      * MMIOs below 2K : raw values and special operations without side
353      * effects
354      */
355     { XIVE_TM_OS_PAGE, TM_QW1_OS + TM_CPPR,   1, xive_tm_set_os_cppr, NULL },
356     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_CPPR, 1, xive_tm_set_hv_cppr, NULL },
357     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, xive_tm_vt_push, NULL },
358     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, NULL, xive_tm_vt_poll },
359 
360     /* MMIOs above 2K : special operations with side effects */
361     { XIVE_TM_OS_PAGE, TM_SPC_ACK_OS_REG,     2, NULL, xive_tm_ack_os_reg },
362     { XIVE_TM_OS_PAGE, TM_SPC_SET_OS_PENDING, 1, xive_tm_set_os_pending, NULL },
363     { XIVE_TM_HV_PAGE, TM_SPC_ACK_HV_REG,     2, NULL, xive_tm_ack_hv_reg },
364     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  4, NULL, xive_tm_pull_pool_ctx },
365     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  8, NULL, xive_tm_pull_pool_ctx },
366 };
367 
368 static const XiveTmOp *xive_tm_find_op(hwaddr offset, unsigned size, bool write)
369 {
370     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
371     uint32_t op_offset = offset & 0xFFF;
372     int i;
373 
374     for (i = 0; i < ARRAY_SIZE(xive_tm_operations); i++) {
375         const XiveTmOp *xto = &xive_tm_operations[i];
376 
377         /* Accesses done from a more privileged TIMA page is allowed */
378         if (xto->page_offset >= page_offset &&
379             xto->op_offset == op_offset &&
380             xto->size == size &&
381             ((write && xto->write_handler) || (!write && xto->read_handler))) {
382             return xto;
383         }
384     }
385     return NULL;
386 }
387 
388 /*
389  * TIMA MMIO handlers
390  */
391 void xive_tctx_tm_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
392                         unsigned size)
393 {
394     const XiveTmOp *xto;
395 
396     /*
397      * TODO: check V bit in Q[0-3]W2
398      */
399 
400     /*
401      * First, check for special operations in the 2K region
402      */
403     if (offset & 0x800) {
404         xto = xive_tm_find_op(offset, size, true);
405         if (!xto) {
406             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA"
407                           "@%"HWADDR_PRIx"\n", offset);
408         } else {
409             xto->write_handler(tctx, offset, value, size);
410         }
411         return;
412     }
413 
414     /*
415      * Then, for special operations in the region below 2K.
416      */
417     xto = xive_tm_find_op(offset, size, true);
418     if (xto) {
419         xto->write_handler(tctx, offset, value, size);
420         return;
421     }
422 
423     /*
424      * Finish with raw access to the register values
425      */
426     xive_tm_raw_write(tctx, offset, value, size);
427 }
428 
429 uint64_t xive_tctx_tm_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
430 {
431     const XiveTmOp *xto;
432 
433     /*
434      * TODO: check V bit in Q[0-3]W2
435      */
436 
437     /*
438      * First, check for special operations in the 2K region
439      */
440     if (offset & 0x800) {
441         xto = xive_tm_find_op(offset, size, false);
442         if (!xto) {
443             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access to TIMA"
444                           "@%"HWADDR_PRIx"\n", offset);
445             return -1;
446         }
447         return xto->read_handler(tctx, offset, size);
448     }
449 
450     /*
451      * Then, for special operations in the region below 2K.
452      */
453     xto = xive_tm_find_op(offset, size, false);
454     if (xto) {
455         return xto->read_handler(tctx, offset, size);
456     }
457 
458     /*
459      * Finish with raw access to the register values
460      */
461     return xive_tm_raw_read(tctx, offset, size);
462 }
463 
464 static void xive_tm_write(void *opaque, hwaddr offset,
465                           uint64_t value, unsigned size)
466 {
467     XiveTCTX *tctx = xive_router_get_tctx(XIVE_ROUTER(opaque), current_cpu);
468 
469     xive_tctx_tm_write(tctx, offset, value, size);
470 }
471 
472 static uint64_t xive_tm_read(void *opaque, hwaddr offset, unsigned size)
473 {
474     XiveTCTX *tctx = xive_router_get_tctx(XIVE_ROUTER(opaque), current_cpu);
475 
476     return xive_tctx_tm_read(tctx, offset, size);
477 }
478 
479 const MemoryRegionOps xive_tm_ops = {
480     .read = xive_tm_read,
481     .write = xive_tm_write,
482     .endianness = DEVICE_BIG_ENDIAN,
483     .valid = {
484         .min_access_size = 1,
485         .max_access_size = 8,
486     },
487     .impl = {
488         .min_access_size = 1,
489         .max_access_size = 8,
490     },
491 };
492 
493 static char *xive_tctx_ring_print(uint8_t *ring)
494 {
495     uint32_t w2 = xive_tctx_word2(ring);
496 
497     return g_strdup_printf("%02x   %02x  %02x    %02x   %02x  "
498                    "%02x  %02x   %02x  %08x",
499                    ring[TM_NSR], ring[TM_CPPR], ring[TM_IPB], ring[TM_LSMFB],
500                    ring[TM_ACK_CNT], ring[TM_INC], ring[TM_AGE], ring[TM_PIPR],
501                    be32_to_cpu(w2));
502 }
503 
504 static const char * const xive_tctx_ring_names[] = {
505     "USER", "OS", "POOL", "PHYS",
506 };
507 
508 void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
509 {
510     int cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
511     int i;
512 
513     if (kvm_irqchip_in_kernel()) {
514         Error *local_err = NULL;
515 
516         kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
517         if (local_err) {
518             error_report_err(local_err);
519             return;
520         }
521     }
522 
523     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
524                    "  W2\n", cpu_index);
525 
526     for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
527         char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
528         monitor_printf(mon, "CPU[%04x]: %4s    %s\n", cpu_index,
529                        xive_tctx_ring_names[i], s);
530         g_free(s);
531     }
532 }
533 
534 static void xive_tctx_reset(void *dev)
535 {
536     XiveTCTX *tctx = XIVE_TCTX(dev);
537 
538     memset(tctx->regs, 0, sizeof(tctx->regs));
539 
540     /* Set some defaults */
541     tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
542     tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
543     tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
544 
545     /*
546      * Initialize PIPR to 0xFF to avoid phantom interrupts when the
547      * CPPR is first set.
548      */
549     tctx->regs[TM_QW1_OS + TM_PIPR] =
550         ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
551     tctx->regs[TM_QW3_HV_PHYS + TM_PIPR] =
552         ipb_to_pipr(tctx->regs[TM_QW3_HV_PHYS + TM_IPB]);
553 }
554 
555 static void xive_tctx_realize(DeviceState *dev, Error **errp)
556 {
557     XiveTCTX *tctx = XIVE_TCTX(dev);
558     PowerPCCPU *cpu;
559     CPUPPCState *env;
560     Object *obj;
561     Error *local_err = NULL;
562 
563     obj = object_property_get_link(OBJECT(dev), "cpu", &local_err);
564     if (!obj) {
565         error_propagate(errp, local_err);
566         error_prepend(errp, "required link 'cpu' not found: ");
567         return;
568     }
569 
570     cpu = POWERPC_CPU(obj);
571     tctx->cs = CPU(obj);
572 
573     env = &cpu->env;
574     switch (PPC_INPUT(env)) {
575     case PPC_FLAGS_INPUT_POWER9:
576         tctx->hv_output = env->irq_inputs[POWER9_INPUT_HINT];
577         tctx->os_output = env->irq_inputs[POWER9_INPUT_INT];
578         break;
579 
580     default:
581         error_setg(errp, "XIVE interrupt controller does not support "
582                    "this CPU bus model");
583         return;
584     }
585 
586     /* Connect the presenter to the VCPU (required for CPU hotplug) */
587     if (kvm_irqchip_in_kernel()) {
588         kvmppc_xive_cpu_connect(tctx, &local_err);
589         if (local_err) {
590             error_propagate(errp, local_err);
591             return;
592         }
593     }
594 
595     qemu_register_reset(xive_tctx_reset, dev);
596 }
597 
598 static void xive_tctx_unrealize(DeviceState *dev, Error **errp)
599 {
600     qemu_unregister_reset(xive_tctx_reset, dev);
601 }
602 
603 static int vmstate_xive_tctx_pre_save(void *opaque)
604 {
605     Error *local_err = NULL;
606 
607     if (kvm_irqchip_in_kernel()) {
608         kvmppc_xive_cpu_get_state(XIVE_TCTX(opaque), &local_err);
609         if (local_err) {
610             error_report_err(local_err);
611             return -1;
612         }
613     }
614 
615     return 0;
616 }
617 
618 static const VMStateDescription vmstate_xive_tctx = {
619     .name = TYPE_XIVE_TCTX,
620     .version_id = 1,
621     .minimum_version_id = 1,
622     .pre_save = vmstate_xive_tctx_pre_save,
623     .post_load = NULL, /* handled by the sPAPRxive model */
624     .fields = (VMStateField[]) {
625         VMSTATE_BUFFER(regs, XiveTCTX),
626         VMSTATE_END_OF_LIST()
627     },
628 };
629 
630 static void xive_tctx_class_init(ObjectClass *klass, void *data)
631 {
632     DeviceClass *dc = DEVICE_CLASS(klass);
633 
634     dc->desc = "XIVE Interrupt Thread Context";
635     dc->realize = xive_tctx_realize;
636     dc->unrealize = xive_tctx_unrealize;
637     dc->vmsd = &vmstate_xive_tctx;
638 }
639 
640 static const TypeInfo xive_tctx_info = {
641     .name          = TYPE_XIVE_TCTX,
642     .parent        = TYPE_DEVICE,
643     .instance_size = sizeof(XiveTCTX),
644     .class_init    = xive_tctx_class_init,
645 };
646 
647 Object *xive_tctx_create(Object *cpu, XiveRouter *xrtr, Error **errp)
648 {
649     Error *local_err = NULL;
650     Object *obj;
651 
652     obj = object_new(TYPE_XIVE_TCTX);
653     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj, &error_abort);
654     object_unref(obj);
655     object_property_add_const_link(obj, "cpu", cpu, &error_abort);
656     object_property_set_bool(obj, true, "realized", &local_err);
657     if (local_err) {
658         goto error;
659     }
660 
661     return obj;
662 
663 error:
664     object_unparent(obj);
665     error_propagate(errp, local_err);
666     return NULL;
667 }
668 
669 /*
670  * XIVE ESB helpers
671  */
672 
673 static uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
674 {
675     uint8_t old_pq = *pq & 0x3;
676 
677     *pq &= ~0x3;
678     *pq |= value & 0x3;
679 
680     return old_pq;
681 }
682 
683 static bool xive_esb_trigger(uint8_t *pq)
684 {
685     uint8_t old_pq = *pq & 0x3;
686 
687     switch (old_pq) {
688     case XIVE_ESB_RESET:
689         xive_esb_set(pq, XIVE_ESB_PENDING);
690         return true;
691     case XIVE_ESB_PENDING:
692     case XIVE_ESB_QUEUED:
693         xive_esb_set(pq, XIVE_ESB_QUEUED);
694         return false;
695     case XIVE_ESB_OFF:
696         xive_esb_set(pq, XIVE_ESB_OFF);
697         return false;
698     default:
699          g_assert_not_reached();
700     }
701 }
702 
703 static bool xive_esb_eoi(uint8_t *pq)
704 {
705     uint8_t old_pq = *pq & 0x3;
706 
707     switch (old_pq) {
708     case XIVE_ESB_RESET:
709     case XIVE_ESB_PENDING:
710         xive_esb_set(pq, XIVE_ESB_RESET);
711         return false;
712     case XIVE_ESB_QUEUED:
713         xive_esb_set(pq, XIVE_ESB_PENDING);
714         return true;
715     case XIVE_ESB_OFF:
716         xive_esb_set(pq, XIVE_ESB_OFF);
717         return false;
718     default:
719          g_assert_not_reached();
720     }
721 }
722 
723 /*
724  * XIVE Interrupt Source (or IVSE)
725  */
726 
727 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
728 {
729     assert(srcno < xsrc->nr_irqs);
730 
731     return xsrc->status[srcno] & 0x3;
732 }
733 
734 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
735 {
736     assert(srcno < xsrc->nr_irqs);
737 
738     return xive_esb_set(&xsrc->status[srcno], pq);
739 }
740 
741 /*
742  * Returns whether the event notification should be forwarded.
743  */
744 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
745 {
746     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
747 
748     xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
749 
750     switch (old_pq) {
751     case XIVE_ESB_RESET:
752         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
753         return true;
754     default:
755         return false;
756     }
757 }
758 
759 /*
760  * Returns whether the event notification should be forwarded.
761  */
762 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
763 {
764     bool ret;
765 
766     assert(srcno < xsrc->nr_irqs);
767 
768     ret = xive_esb_trigger(&xsrc->status[srcno]);
769 
770     if (xive_source_irq_is_lsi(xsrc, srcno) &&
771         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
772         qemu_log_mask(LOG_GUEST_ERROR,
773                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
774     }
775 
776     return ret;
777 }
778 
779 /*
780  * Returns whether the event notification should be forwarded.
781  */
782 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
783 {
784     bool ret;
785 
786     assert(srcno < xsrc->nr_irqs);
787 
788     ret = xive_esb_eoi(&xsrc->status[srcno]);
789 
790     /*
791      * LSI sources do not set the Q bit but they can still be
792      * asserted, in which case we should forward a new event
793      * notification
794      */
795     if (xive_source_irq_is_lsi(xsrc, srcno) &&
796         xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
797         ret = xive_source_lsi_trigger(xsrc, srcno);
798     }
799 
800     return ret;
801 }
802 
803 /*
804  * Forward the source event notification to the Router
805  */
806 static void xive_source_notify(XiveSource *xsrc, int srcno)
807 {
808     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
809 
810     if (xnc->notify) {
811         xnc->notify(xsrc->xive, srcno);
812     }
813 }
814 
815 /*
816  * In a two pages ESB MMIO setting, even page is the trigger page, odd
817  * page is for management
818  */
819 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
820 {
821     return !((addr >> shift) & 1);
822 }
823 
824 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
825 {
826     return xive_source_esb_has_2page(xsrc) &&
827         addr_is_even(addr, xsrc->esb_shift - 1);
828 }
829 
830 /*
831  * ESB MMIO loads
832  *                      Trigger page    Management/EOI page
833  *
834  * ESB MMIO setting     2 pages         1 or 2 pages
835  *
836  * 0x000 .. 0x3FF       -1              EOI and return 0|1
837  * 0x400 .. 0x7FF       -1              EOI and return 0|1
838  * 0x800 .. 0xBFF       -1              return PQ
839  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
840  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
841  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
842  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
843  */
844 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
845 {
846     XiveSource *xsrc = XIVE_SOURCE(opaque);
847     uint32_t offset = addr & 0xFFF;
848     uint32_t srcno = addr >> xsrc->esb_shift;
849     uint64_t ret = -1;
850 
851     /* In a two pages ESB MMIO setting, trigger page should not be read */
852     if (xive_source_is_trigger_page(xsrc, addr)) {
853         qemu_log_mask(LOG_GUEST_ERROR,
854                       "XIVE: invalid load on IRQ %d trigger page at "
855                       "0x%"HWADDR_PRIx"\n", srcno, addr);
856         return -1;
857     }
858 
859     switch (offset) {
860     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
861         ret = xive_source_esb_eoi(xsrc, srcno);
862 
863         /* Forward the source event notification for routing */
864         if (ret) {
865             xive_source_notify(xsrc, srcno);
866         }
867         break;
868 
869     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
870         ret = xive_source_esb_get(xsrc, srcno);
871         break;
872 
873     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
874     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
875     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
876     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
877         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
878         break;
879     default:
880         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
881                       offset);
882     }
883 
884     return ret;
885 }
886 
887 /*
888  * ESB MMIO stores
889  *                      Trigger page    Management/EOI page
890  *
891  * ESB MMIO setting     2 pages         1 or 2 pages
892  *
893  * 0x000 .. 0x3FF       Trigger         Trigger
894  * 0x400 .. 0x7FF       Trigger         EOI
895  * 0x800 .. 0xBFF       Trigger         undefined
896  * 0xC00 .. 0xCFF       Trigger         PQ=00
897  * 0xD00 .. 0xDFF       Trigger         PQ=01
898  * 0xE00 .. 0xDFF       Trigger         PQ=10
899  * 0xF00 .. 0xDFF       Trigger         PQ=11
900  */
901 static void xive_source_esb_write(void *opaque, hwaddr addr,
902                                   uint64_t value, unsigned size)
903 {
904     XiveSource *xsrc = XIVE_SOURCE(opaque);
905     uint32_t offset = addr & 0xFFF;
906     uint32_t srcno = addr >> xsrc->esb_shift;
907     bool notify = false;
908 
909     /* In a two pages ESB MMIO setting, trigger page only triggers */
910     if (xive_source_is_trigger_page(xsrc, addr)) {
911         notify = xive_source_esb_trigger(xsrc, srcno);
912         goto out;
913     }
914 
915     switch (offset) {
916     case 0 ... 0x3FF:
917         notify = xive_source_esb_trigger(xsrc, srcno);
918         break;
919 
920     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
921         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
922             qemu_log_mask(LOG_GUEST_ERROR,
923                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
924             return;
925         }
926 
927         notify = xive_source_esb_eoi(xsrc, srcno);
928         break;
929 
930     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
931     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
932     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
933     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
934         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
935         break;
936 
937     default:
938         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
939                       offset);
940         return;
941     }
942 
943 out:
944     /* Forward the source event notification for routing */
945     if (notify) {
946         xive_source_notify(xsrc, srcno);
947     }
948 }
949 
950 static const MemoryRegionOps xive_source_esb_ops = {
951     .read = xive_source_esb_read,
952     .write = xive_source_esb_write,
953     .endianness = DEVICE_BIG_ENDIAN,
954     .valid = {
955         .min_access_size = 8,
956         .max_access_size = 8,
957     },
958     .impl = {
959         .min_access_size = 8,
960         .max_access_size = 8,
961     },
962 };
963 
964 void xive_source_set_irq(void *opaque, int srcno, int val)
965 {
966     XiveSource *xsrc = XIVE_SOURCE(opaque);
967     bool notify = false;
968 
969     if (xive_source_irq_is_lsi(xsrc, srcno)) {
970         if (val) {
971             notify = xive_source_lsi_trigger(xsrc, srcno);
972         } else {
973             xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
974         }
975     } else {
976         if (val) {
977             notify = xive_source_esb_trigger(xsrc, srcno);
978         }
979     }
980 
981     /* Forward the source event notification for routing */
982     if (notify) {
983         xive_source_notify(xsrc, srcno);
984     }
985 }
986 
987 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
988 {
989     int i;
990 
991     for (i = 0; i < xsrc->nr_irqs; i++) {
992         uint8_t pq = xive_source_esb_get(xsrc, i);
993 
994         if (pq == XIVE_ESB_OFF) {
995             continue;
996         }
997 
998         monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
999                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
1000                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
1001                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1002                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ');
1003     }
1004 }
1005 
1006 static void xive_source_reset(void *dev)
1007 {
1008     XiveSource *xsrc = XIVE_SOURCE(dev);
1009 
1010     /* Do not clear the LSI bitmap */
1011 
1012     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
1013     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
1014 }
1015 
1016 static void xive_source_realize(DeviceState *dev, Error **errp)
1017 {
1018     XiveSource *xsrc = XIVE_SOURCE(dev);
1019     Object *obj;
1020     Error *local_err = NULL;
1021 
1022     obj = object_property_get_link(OBJECT(dev), "xive", &local_err);
1023     if (!obj) {
1024         error_propagate(errp, local_err);
1025         error_prepend(errp, "required link 'xive' not found: ");
1026         return;
1027     }
1028 
1029     xsrc->xive = XIVE_NOTIFIER(obj);
1030 
1031     if (!xsrc->nr_irqs) {
1032         error_setg(errp, "Number of interrupt needs to be greater than 0");
1033         return;
1034     }
1035 
1036     if (xsrc->esb_shift != XIVE_ESB_4K &&
1037         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1038         xsrc->esb_shift != XIVE_ESB_64K &&
1039         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1040         error_setg(errp, "Invalid ESB shift setting");
1041         return;
1042     }
1043 
1044     xsrc->status = g_malloc0(xsrc->nr_irqs);
1045     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1046 
1047     if (!kvm_irqchip_in_kernel()) {
1048         memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1049                               &xive_source_esb_ops, xsrc, "xive.esb",
1050                               (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
1051     }
1052 
1053     qemu_register_reset(xive_source_reset, dev);
1054 }
1055 
1056 static const VMStateDescription vmstate_xive_source = {
1057     .name = TYPE_XIVE_SOURCE,
1058     .version_id = 1,
1059     .minimum_version_id = 1,
1060     .fields = (VMStateField[]) {
1061         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1062         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1063         VMSTATE_END_OF_LIST()
1064     },
1065 };
1066 
1067 /*
1068  * The default XIVE interrupt source setting for the ESB MMIOs is two
1069  * 64k pages without Store EOI, to be in sync with KVM.
1070  */
1071 static Property xive_source_properties[] = {
1072     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1073     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1074     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1075     DEFINE_PROP_END_OF_LIST(),
1076 };
1077 
1078 static void xive_source_class_init(ObjectClass *klass, void *data)
1079 {
1080     DeviceClass *dc = DEVICE_CLASS(klass);
1081 
1082     dc->desc    = "XIVE Interrupt Source";
1083     dc->props   = xive_source_properties;
1084     dc->realize = xive_source_realize;
1085     dc->vmsd    = &vmstate_xive_source;
1086 }
1087 
1088 static const TypeInfo xive_source_info = {
1089     .name          = TYPE_XIVE_SOURCE,
1090     .parent        = TYPE_DEVICE,
1091     .instance_size = sizeof(XiveSource),
1092     .class_init    = xive_source_class_init,
1093 };
1094 
1095 /*
1096  * XiveEND helpers
1097  */
1098 
1099 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
1100 {
1101     uint64_t qaddr_base = xive_end_qaddr(end);
1102     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1103     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1104     uint32_t qentries = 1 << (qsize + 10);
1105     int i;
1106 
1107     /*
1108      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1109      */
1110     monitor_printf(mon, " [ ");
1111     qindex = (qindex - (width - 1)) & (qentries - 1);
1112     for (i = 0; i < width; i++) {
1113         uint64_t qaddr = qaddr_base + (qindex << 2);
1114         uint32_t qdata = -1;
1115 
1116         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
1117                             sizeof(qdata))) {
1118             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1119                           HWADDR_PRIx "\n", qaddr);
1120             return;
1121         }
1122         monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1123                        be32_to_cpu(qdata));
1124         qindex = (qindex + 1) & (qentries - 1);
1125     }
1126 }
1127 
1128 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1129 {
1130     uint64_t qaddr_base = xive_end_qaddr(end);
1131     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1132     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1133     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1134     uint32_t qentries = 1 << (qsize + 10);
1135 
1136     uint32_t nvt = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1137     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1138 
1139     if (!xive_end_is_valid(end)) {
1140         return;
1141     }
1142 
1143     monitor_printf(mon, "  %08x %c%c%c%c%c prio:%d nvt:%04x eq:@%08"PRIx64
1144                    "% 6d/%5d ^%d", end_idx,
1145                    xive_end_is_valid(end)    ? 'v' : '-',
1146                    xive_end_is_enqueue(end)  ? 'q' : '-',
1147                    xive_end_is_notify(end)   ? 'n' : '-',
1148                    xive_end_is_backlog(end)  ? 'b' : '-',
1149                    xive_end_is_escalate(end) ? 'e' : '-',
1150                    priority, nvt, qaddr_base, qindex, qentries, qgen);
1151 
1152     xive_end_queue_pic_print_info(end, 6, mon);
1153     monitor_printf(mon, "]\n");
1154 }
1155 
1156 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1157 {
1158     uint64_t qaddr_base = xive_end_qaddr(end);
1159     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1160     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1161     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1162 
1163     uint64_t qaddr = qaddr_base + (qindex << 2);
1164     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1165     uint32_t qentries = 1 << (qsize + 10);
1166 
1167     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata))) {
1168         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1169                       HWADDR_PRIx "\n", qaddr);
1170         return;
1171     }
1172 
1173     qindex = (qindex + 1) & (qentries - 1);
1174     if (qindex == 0) {
1175         qgen ^= 1;
1176         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1177     }
1178     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1179 }
1180 
1181 /*
1182  * XIVE Router (aka. Virtualization Controller or IVRE)
1183  */
1184 
1185 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1186                         XiveEAS *eas)
1187 {
1188     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1189 
1190     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1191 }
1192 
1193 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1194                         XiveEND *end)
1195 {
1196    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1197 
1198    return xrc->get_end(xrtr, end_blk, end_idx, end);
1199 }
1200 
1201 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1202                           XiveEND *end, uint8_t word_number)
1203 {
1204    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1205 
1206    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1207 }
1208 
1209 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1210                         XiveNVT *nvt)
1211 {
1212    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1213 
1214    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1215 }
1216 
1217 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1218                         XiveNVT *nvt, uint8_t word_number)
1219 {
1220    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1221 
1222    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1223 }
1224 
1225 XiveTCTX *xive_router_get_tctx(XiveRouter *xrtr, CPUState *cs)
1226 {
1227     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1228 
1229     return xrc->get_tctx(xrtr, cs);
1230 }
1231 
1232 /*
1233  * Encode the HW CAM line in the block group mode format :
1234  *
1235  *   chip << 19 | 0000000 0 0001 thread (7Bit)
1236  */
1237 static uint32_t xive_tctx_hw_cam_line(XiveTCTX *tctx)
1238 {
1239     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1240     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1241 
1242     return xive_nvt_cam_line((pir >> 8) & 0xf, 1 << 7 | (pir & 0x7f));
1243 }
1244 
1245 /*
1246  * The thread context register words are in big-endian format.
1247  */
1248 static int xive_presenter_tctx_match(XiveTCTX *tctx, uint8_t format,
1249                                      uint8_t nvt_blk, uint32_t nvt_idx,
1250                                      bool cam_ignore, uint32_t logic_serv)
1251 {
1252     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1253     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1254     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1255     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1256     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1257 
1258     /*
1259      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1260      * identifier are ignored in the "CAM" match.
1261      */
1262 
1263     if (format == 0) {
1264         if (cam_ignore == true) {
1265             /*
1266              * F=0 & i=1: Logical server notification (bits ignored at
1267              * the end of the NVT identifier)
1268              */
1269             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1270                           nvt_blk, nvt_idx);
1271              return -1;
1272         }
1273 
1274         /* F=0 & i=0: Specific NVT notification */
1275 
1276         /* PHYS ring */
1277         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1278             cam == xive_tctx_hw_cam_line(tctx)) {
1279             return TM_QW3_HV_PHYS;
1280         }
1281 
1282         /* HV POOL ring */
1283         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1284             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1285             return TM_QW2_HV_POOL;
1286         }
1287 
1288         /* OS ring */
1289         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1290             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1291             return TM_QW1_OS;
1292         }
1293     } else {
1294         /* F=1 : User level Event-Based Branch (EBB) notification */
1295 
1296         /* USER ring */
1297         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1298              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1299              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1300              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1301             return TM_QW0_USER;
1302         }
1303     }
1304     return -1;
1305 }
1306 
1307 typedef struct XiveTCTXMatch {
1308     XiveTCTX *tctx;
1309     uint8_t ring;
1310 } XiveTCTXMatch;
1311 
1312 static bool xive_presenter_match(XiveRouter *xrtr, uint8_t format,
1313                                  uint8_t nvt_blk, uint32_t nvt_idx,
1314                                  bool cam_ignore, uint8_t priority,
1315                                  uint32_t logic_serv, XiveTCTXMatch *match)
1316 {
1317     CPUState *cs;
1318 
1319     /*
1320      * TODO (PowerNV): handle chip_id overwrite of block field for
1321      * hardwired CAM compares
1322      */
1323 
1324     CPU_FOREACH(cs) {
1325         XiveTCTX *tctx = xive_router_get_tctx(xrtr, cs);
1326         int ring;
1327 
1328         /*
1329          * HW checks that the CPU is enabled in the Physical Thread
1330          * Enable Register (PTER).
1331          */
1332 
1333         /*
1334          * Check the thread context CAM lines and record matches. We
1335          * will handle CPU exception delivery later
1336          */
1337         ring = xive_presenter_tctx_match(tctx, format, nvt_blk, nvt_idx,
1338                                          cam_ignore, logic_serv);
1339         /*
1340          * Save the context and follow on to catch duplicates, that we
1341          * don't support yet.
1342          */
1343         if (ring != -1) {
1344             if (match->tctx) {
1345                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
1346                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
1347                 return false;
1348             }
1349 
1350             match->ring = ring;
1351             match->tctx = tctx;
1352         }
1353     }
1354 
1355     if (!match->tctx) {
1356         qemu_log_mask(LOG_UNIMP, "XIVE: NVT %x/%x is not dispatched\n",
1357                       nvt_blk, nvt_idx);
1358         return false;
1359     }
1360 
1361     return true;
1362 }
1363 
1364 /*
1365  * This is our simple Xive Presenter Engine model. It is merged in the
1366  * Router as it does not require an extra object.
1367  *
1368  * It receives notification requests sent by the IVRE to find one
1369  * matching NVT (or more) dispatched on the processor threads. In case
1370  * of a single NVT notification, the process is abreviated and the
1371  * thread is signaled if a match is found. In case of a logical server
1372  * notification (bits ignored at the end of the NVT identifier), the
1373  * IVPE and IVRE select a winning thread using different filters. This
1374  * involves 2 or 3 exchanges on the PowerBus that the model does not
1375  * support.
1376  *
1377  * The parameters represent what is sent on the PowerBus
1378  */
1379 static void xive_presenter_notify(XiveRouter *xrtr, uint8_t format,
1380                                   uint8_t nvt_blk, uint32_t nvt_idx,
1381                                   bool cam_ignore, uint8_t priority,
1382                                   uint32_t logic_serv)
1383 {
1384     XiveNVT nvt;
1385     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1386     bool found;
1387 
1388     /* NVT cache lookup */
1389     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1390         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1391                       nvt_blk, nvt_idx);
1392         return;
1393     }
1394 
1395     if (!xive_nvt_is_valid(&nvt)) {
1396         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1397                       nvt_blk, nvt_idx);
1398         return;
1399     }
1400 
1401     found = xive_presenter_match(xrtr, format, nvt_blk, nvt_idx, cam_ignore,
1402                                  priority, logic_serv, &match);
1403     if (found) {
1404         ipb_update(&match.tctx->regs[match.ring], priority);
1405         xive_tctx_notify(match.tctx, match.ring);
1406         return;
1407     }
1408 
1409     /* Record the IPB in the associated NVT structure */
1410     ipb_update((uint8_t *) &nvt.w4, priority);
1411     xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1412 
1413     /*
1414      * If no matching NVT is dispatched on a HW thread :
1415      * - update the NVT structure if backlog is activated
1416      * - escalate (ESe PQ bits and EAS in w4-5) if escalation is
1417      *   activated
1418      */
1419 }
1420 
1421 /*
1422  * An END trigger can come from an event trigger (IPI or HW) or from
1423  * another chip. We don't model the PowerBus but the END trigger
1424  * message has the same parameters than in the function below.
1425  */
1426 static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1427                                    uint32_t end_idx, uint32_t end_data)
1428 {
1429     XiveEND end;
1430     uint8_t priority;
1431     uint8_t format;
1432 
1433     /* END cache lookup */
1434     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1435         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1436                       end_idx);
1437         return;
1438     }
1439 
1440     if (!xive_end_is_valid(&end)) {
1441         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1442                       end_blk, end_idx);
1443         return;
1444     }
1445 
1446     if (xive_end_is_enqueue(&end)) {
1447         xive_end_enqueue(&end, end_data);
1448         /* Enqueuing event data modifies the EQ toggle and index */
1449         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1450     }
1451 
1452     /*
1453      * The W7 format depends on the F bit in W6. It defines the type
1454      * of the notification :
1455      *
1456      *   F=0 : single or multiple NVT notification
1457      *   F=1 : User level Event-Based Branch (EBB) notification, no
1458      *         priority
1459      */
1460     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1461     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1462 
1463     /* The END is masked */
1464     if (format == 0 && priority == 0xff) {
1465         return;
1466     }
1467 
1468     /*
1469      * Check the END ESn (Event State Buffer for notification) for
1470      * even futher coalescing in the Router
1471      */
1472     if (!xive_end_is_notify(&end)) {
1473         uint8_t pq = xive_get_field32(END_W1_ESn, end.w1);
1474         bool notify = xive_esb_trigger(&pq);
1475 
1476         if (pq != xive_get_field32(END_W1_ESn, end.w1)) {
1477             end.w1 = xive_set_field32(END_W1_ESn, end.w1, pq);
1478             xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1479         }
1480 
1481         /* ESn[Q]=1 : end of notification */
1482         if (!notify) {
1483             return;
1484         }
1485     }
1486 
1487     /*
1488      * Follows IVPE notification
1489      */
1490     xive_presenter_notify(xrtr, format,
1491                           xive_get_field32(END_W6_NVT_BLOCK, end.w6),
1492                           xive_get_field32(END_W6_NVT_INDEX, end.w6),
1493                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1494                           priority,
1495                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1496 
1497     /* TODO: Auto EOI. */
1498 }
1499 
1500 void xive_router_notify(XiveNotifier *xn, uint32_t lisn)
1501 {
1502     XiveRouter *xrtr = XIVE_ROUTER(xn);
1503     uint8_t eas_blk = XIVE_SRCNO_BLOCK(lisn);
1504     uint32_t eas_idx = XIVE_SRCNO_INDEX(lisn);
1505     XiveEAS eas;
1506 
1507     /* EAS cache lookup */
1508     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1509         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1510         return;
1511     }
1512 
1513     /*
1514      * The IVRE checks the State Bit Cache at this point. We skip the
1515      * SBC lookup because the state bits of the sources are modeled
1516      * internally in QEMU.
1517      */
1518 
1519     if (!xive_eas_is_valid(&eas)) {
1520         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1521         return;
1522     }
1523 
1524     if (xive_eas_is_masked(&eas)) {
1525         /* Notification completed */
1526         return;
1527     }
1528 
1529     /*
1530      * The event trigger becomes an END trigger
1531      */
1532     xive_router_end_notify(xrtr,
1533                            xive_get_field64(EAS_END_BLOCK, eas.w),
1534                            xive_get_field64(EAS_END_INDEX, eas.w),
1535                            xive_get_field64(EAS_END_DATA,  eas.w));
1536 }
1537 
1538 static void xive_router_class_init(ObjectClass *klass, void *data)
1539 {
1540     DeviceClass *dc = DEVICE_CLASS(klass);
1541     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1542 
1543     dc->desc    = "XIVE Router Engine";
1544     xnc->notify = xive_router_notify;
1545 }
1546 
1547 static const TypeInfo xive_router_info = {
1548     .name          = TYPE_XIVE_ROUTER,
1549     .parent        = TYPE_SYS_BUS_DEVICE,
1550     .abstract      = true,
1551     .class_size    = sizeof(XiveRouterClass),
1552     .class_init    = xive_router_class_init,
1553     .interfaces    = (InterfaceInfo[]) {
1554         { TYPE_XIVE_NOTIFIER },
1555         { }
1556     }
1557 };
1558 
1559 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1560 {
1561     if (!xive_eas_is_valid(eas)) {
1562         return;
1563     }
1564 
1565     monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1566                    lisn, xive_eas_is_masked(eas) ? "M" : " ",
1567                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1568                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1569                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1570 }
1571 
1572 /*
1573  * END ESB MMIO loads
1574  */
1575 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1576 {
1577     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1578     uint32_t offset = addr & 0xFFF;
1579     uint8_t end_blk;
1580     uint32_t end_idx;
1581     XiveEND end;
1582     uint32_t end_esmask;
1583     uint8_t pq;
1584     uint64_t ret = -1;
1585 
1586     end_blk = xsrc->block_id;
1587     end_idx = addr >> (xsrc->esb_shift + 1);
1588 
1589     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1590         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1591                       end_idx);
1592         return -1;
1593     }
1594 
1595     if (!xive_end_is_valid(&end)) {
1596         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1597                       end_blk, end_idx);
1598         return -1;
1599     }
1600 
1601     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1602     pq = xive_get_field32(end_esmask, end.w1);
1603 
1604     switch (offset) {
1605     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1606         ret = xive_esb_eoi(&pq);
1607 
1608         /* Forward the source event notification for routing ?? */
1609         break;
1610 
1611     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1612         ret = pq;
1613         break;
1614 
1615     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1616     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1617     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1618     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1619         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1620         break;
1621     default:
1622         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1623                       offset);
1624         return -1;
1625     }
1626 
1627     if (pq != xive_get_field32(end_esmask, end.w1)) {
1628         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1629         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1630     }
1631 
1632     return ret;
1633 }
1634 
1635 /*
1636  * END ESB MMIO stores are invalid
1637  */
1638 static void xive_end_source_write(void *opaque, hwaddr addr,
1639                                   uint64_t value, unsigned size)
1640 {
1641     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1642                   HWADDR_PRIx"\n", addr);
1643 }
1644 
1645 static const MemoryRegionOps xive_end_source_ops = {
1646     .read = xive_end_source_read,
1647     .write = xive_end_source_write,
1648     .endianness = DEVICE_BIG_ENDIAN,
1649     .valid = {
1650         .min_access_size = 8,
1651         .max_access_size = 8,
1652     },
1653     .impl = {
1654         .min_access_size = 8,
1655         .max_access_size = 8,
1656     },
1657 };
1658 
1659 static void xive_end_source_realize(DeviceState *dev, Error **errp)
1660 {
1661     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1662     Object *obj;
1663     Error *local_err = NULL;
1664 
1665     obj = object_property_get_link(OBJECT(dev), "xive", &local_err);
1666     if (!obj) {
1667         error_propagate(errp, local_err);
1668         error_prepend(errp, "required link 'xive' not found: ");
1669         return;
1670     }
1671 
1672     xsrc->xrtr = XIVE_ROUTER(obj);
1673 
1674     if (!xsrc->nr_ends) {
1675         error_setg(errp, "Number of interrupt needs to be greater than 0");
1676         return;
1677     }
1678 
1679     if (xsrc->esb_shift != XIVE_ESB_4K &&
1680         xsrc->esb_shift != XIVE_ESB_64K) {
1681         error_setg(errp, "Invalid ESB shift setting");
1682         return;
1683     }
1684 
1685     /*
1686      * Each END is assigned an even/odd pair of MMIO pages, the even page
1687      * manages the ESn field while the odd page manages the ESe field.
1688      */
1689     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1690                           &xive_end_source_ops, xsrc, "xive.end",
1691                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1692 }
1693 
1694 static Property xive_end_source_properties[] = {
1695     DEFINE_PROP_UINT8("block-id", XiveENDSource, block_id, 0),
1696     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
1697     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
1698     DEFINE_PROP_END_OF_LIST(),
1699 };
1700 
1701 static void xive_end_source_class_init(ObjectClass *klass, void *data)
1702 {
1703     DeviceClass *dc = DEVICE_CLASS(klass);
1704 
1705     dc->desc    = "XIVE END Source";
1706     dc->props   = xive_end_source_properties;
1707     dc->realize = xive_end_source_realize;
1708 }
1709 
1710 static const TypeInfo xive_end_source_info = {
1711     .name          = TYPE_XIVE_END_SOURCE,
1712     .parent        = TYPE_DEVICE,
1713     .instance_size = sizeof(XiveENDSource),
1714     .class_init    = xive_end_source_class_init,
1715 };
1716 
1717 /*
1718  * XIVE Notifier
1719  */
1720 static const TypeInfo xive_notifier_info = {
1721     .name = TYPE_XIVE_NOTIFIER,
1722     .parent = TYPE_INTERFACE,
1723     .class_size = sizeof(XiveNotifierClass),
1724 };
1725 
1726 static void xive_register_types(void)
1727 {
1728     type_register_static(&xive_source_info);
1729     type_register_static(&xive_notifier_info);
1730     type_register_static(&xive_router_info);
1731     type_register_static(&xive_end_source_info);
1732     type_register_static(&xive_tctx_info);
1733 }
1734 
1735 type_init(xive_register_types)
1736