xref: /openbmc/qemu/hw/intc/xive.c (revision 37677d7d)
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 uint64_t xive_tctx_accept(XiveTCTX *tctx, uint8_t ring)
66 {
67     uint8_t *regs = &tctx->regs[ring];
68     uint8_t nsr = regs[TM_NSR];
69     uint8_t mask = exception_mask(ring);
70 
71     qemu_irq_lower(tctx->output);
72 
73     if (regs[TM_NSR] & mask) {
74         uint8_t cppr = regs[TM_PIPR];
75 
76         regs[TM_CPPR] = cppr;
77 
78         /* Reset the pending buffer bit */
79         regs[TM_IPB] &= ~priority_to_ipb(cppr);
80         regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
81 
82         /* Drop Exception bit */
83         regs[TM_NSR] &= ~mask;
84     }
85 
86     return (nsr << 8) | regs[TM_CPPR];
87 }
88 
89 static void xive_tctx_notify(XiveTCTX *tctx, uint8_t ring)
90 {
91     uint8_t *regs = &tctx->regs[ring];
92 
93     if (regs[TM_PIPR] < regs[TM_CPPR]) {
94         switch (ring) {
95         case TM_QW1_OS:
96             regs[TM_NSR] |= TM_QW1_NSR_EO;
97             break;
98         case TM_QW3_HV_PHYS:
99             regs[TM_NSR] |= (TM_QW3_NSR_HE_PHYS << 6);
100             break;
101         default:
102             g_assert_not_reached();
103         }
104         qemu_irq_raise(tctx->output);
105     }
106 }
107 
108 static void xive_tctx_set_cppr(XiveTCTX *tctx, uint8_t ring, uint8_t cppr)
109 {
110     if (cppr > XIVE_PRIORITY_MAX) {
111         cppr = 0xff;
112     }
113 
114     tctx->regs[ring + TM_CPPR] = cppr;
115 
116     /* CPPR has changed, check if we need to raise a pending exception */
117     xive_tctx_notify(tctx, ring);
118 }
119 
120 /*
121  * XIVE Thread Interrupt Management Area (TIMA)
122  */
123 
124 static void xive_tm_set_hv_cppr(XiveTCTX *tctx, hwaddr offset,
125                                 uint64_t value, unsigned size)
126 {
127     xive_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
128 }
129 
130 static uint64_t xive_tm_ack_hv_reg(XiveTCTX *tctx, hwaddr offset, unsigned size)
131 {
132     return xive_tctx_accept(tctx, TM_QW3_HV_PHYS);
133 }
134 
135 static uint64_t xive_tm_pull_pool_ctx(XiveTCTX *tctx, hwaddr offset,
136                                       unsigned size)
137 {
138     uint64_t ret;
139 
140     ret = tctx->regs[TM_QW2_HV_POOL + TM_WORD2] & TM_QW2W2_POOL_CAM;
141     tctx->regs[TM_QW2_HV_POOL + TM_WORD2] &= ~TM_QW2W2_POOL_CAM;
142     return ret;
143 }
144 
145 static void xive_tm_vt_push(XiveTCTX *tctx, hwaddr offset,
146                             uint64_t value, unsigned size)
147 {
148     tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] = value & 0xff;
149 }
150 
151 static uint64_t xive_tm_vt_poll(XiveTCTX *tctx, hwaddr offset, unsigned size)
152 {
153     return tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] & 0xff;
154 }
155 
156 /*
157  * Define an access map for each page of the TIMA that we will use in
158  * the memory region ops to filter values when doing loads and stores
159  * of raw registers values
160  *
161  * Registers accessibility bits :
162  *
163  *    0x0 - no access
164  *    0x1 - write only
165  *    0x2 - read only
166  *    0x3 - read/write
167  */
168 
169 static const uint8_t xive_tm_hw_view[] = {
170     /* QW-0 User */   3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0,
171     /* QW-1 OS   */   3, 3, 3, 3,   3, 3, 0, 3,   3, 3, 3, 3,   0, 0, 0, 0,
172     /* QW-2 POOL */   0, 0, 3, 3,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0,
173     /* QW-3 PHYS */   3, 3, 3, 3,   0, 3, 0, 3,   3, 0, 0, 3,   3, 3, 3, 0,
174 };
175 
176 static const uint8_t xive_tm_hv_view[] = {
177     /* QW-0 User */   3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0,
178     /* QW-1 OS   */   3, 3, 3, 3,   3, 3, 0, 3,   3, 3, 3, 3,   0, 0, 0, 0,
179     /* QW-2 POOL */   0, 0, 3, 3,   0, 0, 0, 0,   0, 3, 3, 3,   0, 0, 0, 0,
180     /* QW-3 PHYS */   3, 3, 3, 3,   0, 3, 0, 3,   3, 0, 0, 3,   0, 0, 0, 0,
181 };
182 
183 static const uint8_t xive_tm_os_view[] = {
184     /* QW-0 User */   3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0,
185     /* QW-1 OS   */   2, 3, 2, 2,   2, 2, 0, 2,   0, 0, 0, 0,   0, 0, 0, 0,
186     /* QW-2 POOL */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
187     /* QW-3 PHYS */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
188 };
189 
190 static const uint8_t xive_tm_user_view[] = {
191     /* QW-0 User */   3, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
192     /* QW-1 OS   */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
193     /* QW-2 POOL */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
194     /* QW-3 PHYS */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
195 };
196 
197 /*
198  * Overall TIMA access map for the thread interrupt management context
199  * registers
200  */
201 static const uint8_t *xive_tm_views[] = {
202     [XIVE_TM_HW_PAGE]   = xive_tm_hw_view,
203     [XIVE_TM_HV_PAGE]   = xive_tm_hv_view,
204     [XIVE_TM_OS_PAGE]   = xive_tm_os_view,
205     [XIVE_TM_USER_PAGE] = xive_tm_user_view,
206 };
207 
208 /*
209  * Computes a register access mask for a given offset in the TIMA
210  */
211 static uint64_t xive_tm_mask(hwaddr offset, unsigned size, bool write)
212 {
213     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
214     uint8_t reg_offset = offset & 0x3F;
215     uint8_t reg_mask = write ? 0x1 : 0x2;
216     uint64_t mask = 0x0;
217     int i;
218 
219     for (i = 0; i < size; i++) {
220         if (xive_tm_views[page_offset][reg_offset + i] & reg_mask) {
221             mask |= (uint64_t) 0xff << (8 * (size - i - 1));
222         }
223     }
224 
225     return mask;
226 }
227 
228 static void xive_tm_raw_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
229                               unsigned size)
230 {
231     uint8_t ring_offset = offset & 0x30;
232     uint8_t reg_offset = offset & 0x3F;
233     uint64_t mask = xive_tm_mask(offset, size, true);
234     int i;
235 
236     /*
237      * Only 4 or 8 bytes stores are allowed and the User ring is
238      * excluded
239      */
240     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
241         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA @%"
242                       HWADDR_PRIx"\n", offset);
243         return;
244     }
245 
246     /*
247      * Use the register offset for the raw values and filter out
248      * reserved values
249      */
250     for (i = 0; i < size; i++) {
251         uint8_t byte_mask = (mask >> (8 * (size - i - 1)));
252         if (byte_mask) {
253             tctx->regs[reg_offset + i] = (value >> (8 * (size - i - 1))) &
254                 byte_mask;
255         }
256     }
257 }
258 
259 static uint64_t xive_tm_raw_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
260 {
261     uint8_t ring_offset = offset & 0x30;
262     uint8_t reg_offset = offset & 0x3F;
263     uint64_t mask = xive_tm_mask(offset, size, false);
264     uint64_t ret;
265     int i;
266 
267     /*
268      * Only 4 or 8 bytes loads are allowed and the User ring is
269      * excluded
270      */
271     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
272         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access at TIMA @%"
273                       HWADDR_PRIx"\n", offset);
274         return -1;
275     }
276 
277     /* Use the register offset for the raw values */
278     ret = 0;
279     for (i = 0; i < size; i++) {
280         ret |= (uint64_t) tctx->regs[reg_offset + i] << (8 * (size - i - 1));
281     }
282 
283     /* filter out reserved values */
284     return ret & mask;
285 }
286 
287 /*
288  * The TM context is mapped twice within each page. Stores and loads
289  * to the first mapping below 2K write and read the specified values
290  * without modification. The second mapping above 2K performs specific
291  * state changes (side effects) in addition to setting/returning the
292  * interrupt management area context of the processor thread.
293  */
294 static uint64_t xive_tm_ack_os_reg(XiveTCTX *tctx, hwaddr offset, unsigned size)
295 {
296     return xive_tctx_accept(tctx, TM_QW1_OS);
297 }
298 
299 static void xive_tm_set_os_cppr(XiveTCTX *tctx, hwaddr offset,
300                                 uint64_t value, unsigned size)
301 {
302     xive_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
303 }
304 
305 /*
306  * Adjust the IPB to allow a CPU to process event queues of other
307  * priorities during one physical interrupt cycle.
308  */
309 static void xive_tm_set_os_pending(XiveTCTX *tctx, hwaddr offset,
310                                    uint64_t value, unsigned size)
311 {
312     ipb_update(&tctx->regs[TM_QW1_OS], value & 0xff);
313     xive_tctx_notify(tctx, TM_QW1_OS);
314 }
315 
316 /*
317  * Define a mapping of "special" operations depending on the TIMA page
318  * offset and the size of the operation.
319  */
320 typedef struct XiveTmOp {
321     uint8_t  page_offset;
322     uint32_t op_offset;
323     unsigned size;
324     void     (*write_handler)(XiveTCTX *tctx, hwaddr offset, uint64_t value,
325                               unsigned size);
326     uint64_t (*read_handler)(XiveTCTX *tctx, hwaddr offset, unsigned size);
327 } XiveTmOp;
328 
329 static const XiveTmOp xive_tm_operations[] = {
330     /*
331      * MMIOs below 2K : raw values and special operations without side
332      * effects
333      */
334     { XIVE_TM_OS_PAGE, TM_QW1_OS + TM_CPPR,   1, xive_tm_set_os_cppr, NULL },
335     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_CPPR, 1, xive_tm_set_hv_cppr, NULL },
336     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, xive_tm_vt_push, NULL },
337     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, NULL, xive_tm_vt_poll },
338 
339     /* MMIOs above 2K : special operations with side effects */
340     { XIVE_TM_OS_PAGE, TM_SPC_ACK_OS_REG,     2, NULL, xive_tm_ack_os_reg },
341     { XIVE_TM_OS_PAGE, TM_SPC_SET_OS_PENDING, 1, xive_tm_set_os_pending, NULL },
342     { XIVE_TM_HV_PAGE, TM_SPC_ACK_HV_REG,     2, NULL, xive_tm_ack_hv_reg },
343     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  4, NULL, xive_tm_pull_pool_ctx },
344     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  8, NULL, xive_tm_pull_pool_ctx },
345 };
346 
347 static const XiveTmOp *xive_tm_find_op(hwaddr offset, unsigned size, bool write)
348 {
349     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
350     uint32_t op_offset = offset & 0xFFF;
351     int i;
352 
353     for (i = 0; i < ARRAY_SIZE(xive_tm_operations); i++) {
354         const XiveTmOp *xto = &xive_tm_operations[i];
355 
356         /* Accesses done from a more privileged TIMA page is allowed */
357         if (xto->page_offset >= page_offset &&
358             xto->op_offset == op_offset &&
359             xto->size == size &&
360             ((write && xto->write_handler) || (!write && xto->read_handler))) {
361             return xto;
362         }
363     }
364     return NULL;
365 }
366 
367 /*
368  * TIMA MMIO handlers
369  */
370 void xive_tctx_tm_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
371                         unsigned size)
372 {
373     const XiveTmOp *xto;
374 
375     /*
376      * TODO: check V bit in Q[0-3]W2
377      */
378 
379     /*
380      * First, check for special operations in the 2K region
381      */
382     if (offset & 0x800) {
383         xto = xive_tm_find_op(offset, size, true);
384         if (!xto) {
385             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA"
386                           "@%"HWADDR_PRIx"\n", offset);
387         } else {
388             xto->write_handler(tctx, offset, value, size);
389         }
390         return;
391     }
392 
393     /*
394      * Then, for special operations in the region below 2K.
395      */
396     xto = xive_tm_find_op(offset, size, true);
397     if (xto) {
398         xto->write_handler(tctx, offset, value, size);
399         return;
400     }
401 
402     /*
403      * Finish with raw access to the register values
404      */
405     xive_tm_raw_write(tctx, offset, value, size);
406 }
407 
408 uint64_t xive_tctx_tm_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
409 {
410     const XiveTmOp *xto;
411 
412     /*
413      * TODO: check V bit in Q[0-3]W2
414      */
415 
416     /*
417      * First, check for special operations in the 2K region
418      */
419     if (offset & 0x800) {
420         xto = xive_tm_find_op(offset, size, false);
421         if (!xto) {
422             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access to TIMA"
423                           "@%"HWADDR_PRIx"\n", offset);
424             return -1;
425         }
426         return xto->read_handler(tctx, offset, size);
427     }
428 
429     /*
430      * Then, for special operations in the region below 2K.
431      */
432     xto = xive_tm_find_op(offset, size, false);
433     if (xto) {
434         return xto->read_handler(tctx, offset, size);
435     }
436 
437     /*
438      * Finish with raw access to the register values
439      */
440     return xive_tm_raw_read(tctx, offset, size);
441 }
442 
443 static void xive_tm_write(void *opaque, hwaddr offset,
444                           uint64_t value, unsigned size)
445 {
446     XiveTCTX *tctx = xive_router_get_tctx(XIVE_ROUTER(opaque), current_cpu);
447 
448     xive_tctx_tm_write(tctx, offset, value, size);
449 }
450 
451 static uint64_t xive_tm_read(void *opaque, hwaddr offset, unsigned size)
452 {
453     XiveTCTX *tctx = xive_router_get_tctx(XIVE_ROUTER(opaque), current_cpu);
454 
455     return xive_tctx_tm_read(tctx, offset, size);
456 }
457 
458 const MemoryRegionOps xive_tm_ops = {
459     .read = xive_tm_read,
460     .write = xive_tm_write,
461     .endianness = DEVICE_BIG_ENDIAN,
462     .valid = {
463         .min_access_size = 1,
464         .max_access_size = 8,
465     },
466     .impl = {
467         .min_access_size = 1,
468         .max_access_size = 8,
469     },
470 };
471 
472 static inline uint32_t xive_tctx_word2(uint8_t *ring)
473 {
474     return *((uint32_t *) &ring[TM_WORD2]);
475 }
476 
477 static char *xive_tctx_ring_print(uint8_t *ring)
478 {
479     uint32_t w2 = xive_tctx_word2(ring);
480 
481     return g_strdup_printf("%02x   %02x  %02x    %02x   %02x  "
482                    "%02x  %02x   %02x  %08x",
483                    ring[TM_NSR], ring[TM_CPPR], ring[TM_IPB], ring[TM_LSMFB],
484                    ring[TM_ACK_CNT], ring[TM_INC], ring[TM_AGE], ring[TM_PIPR],
485                    be32_to_cpu(w2));
486 }
487 
488 static const char * const xive_tctx_ring_names[] = {
489     "USER", "OS", "POOL", "PHYS",
490 };
491 
492 void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
493 {
494     int cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
495     int i;
496 
497     if (kvm_irqchip_in_kernel()) {
498         Error *local_err = NULL;
499 
500         kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
501         if (local_err) {
502             error_report_err(local_err);
503             return;
504         }
505     }
506 
507     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
508                    "  W2\n", cpu_index);
509 
510     for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
511         char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
512         monitor_printf(mon, "CPU[%04x]: %4s    %s\n", cpu_index,
513                        xive_tctx_ring_names[i], s);
514         g_free(s);
515     }
516 }
517 
518 static void xive_tctx_reset(void *dev)
519 {
520     XiveTCTX *tctx = XIVE_TCTX(dev);
521 
522     memset(tctx->regs, 0, sizeof(tctx->regs));
523 
524     /* Set some defaults */
525     tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
526     tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
527     tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
528 
529     /*
530      * Initialize PIPR to 0xFF to avoid phantom interrupts when the
531      * CPPR is first set.
532      */
533     tctx->regs[TM_QW1_OS + TM_PIPR] =
534         ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
535     tctx->regs[TM_QW3_HV_PHYS + TM_PIPR] =
536         ipb_to_pipr(tctx->regs[TM_QW3_HV_PHYS + TM_IPB]);
537 }
538 
539 static void xive_tctx_realize(DeviceState *dev, Error **errp)
540 {
541     XiveTCTX *tctx = XIVE_TCTX(dev);
542     PowerPCCPU *cpu;
543     CPUPPCState *env;
544     Object *obj;
545     Error *local_err = NULL;
546 
547     obj = object_property_get_link(OBJECT(dev), "cpu", &local_err);
548     if (!obj) {
549         error_propagate(errp, local_err);
550         error_prepend(errp, "required link 'cpu' not found: ");
551         return;
552     }
553 
554     cpu = POWERPC_CPU(obj);
555     tctx->cs = CPU(obj);
556 
557     env = &cpu->env;
558     switch (PPC_INPUT(env)) {
559     case PPC_FLAGS_INPUT_POWER9:
560         tctx->output = env->irq_inputs[POWER9_INPUT_INT];
561         break;
562 
563     default:
564         error_setg(errp, "XIVE interrupt controller does not support "
565                    "this CPU bus model");
566         return;
567     }
568 
569     /* Connect the presenter to the VCPU (required for CPU hotplug) */
570     if (kvm_irqchip_in_kernel()) {
571         kvmppc_xive_cpu_connect(tctx, &local_err);
572         if (local_err) {
573             error_propagate(errp, local_err);
574             return;
575         }
576     }
577 
578     qemu_register_reset(xive_tctx_reset, dev);
579 }
580 
581 static void xive_tctx_unrealize(DeviceState *dev, Error **errp)
582 {
583     qemu_unregister_reset(xive_tctx_reset, dev);
584 }
585 
586 static int vmstate_xive_tctx_pre_save(void *opaque)
587 {
588     Error *local_err = NULL;
589 
590     if (kvm_irqchip_in_kernel()) {
591         kvmppc_xive_cpu_get_state(XIVE_TCTX(opaque), &local_err);
592         if (local_err) {
593             error_report_err(local_err);
594             return -1;
595         }
596     }
597 
598     return 0;
599 }
600 
601 static const VMStateDescription vmstate_xive_tctx = {
602     .name = TYPE_XIVE_TCTX,
603     .version_id = 1,
604     .minimum_version_id = 1,
605     .pre_save = vmstate_xive_tctx_pre_save,
606     .post_load = NULL, /* handled by the sPAPRxive model */
607     .fields = (VMStateField[]) {
608         VMSTATE_BUFFER(regs, XiveTCTX),
609         VMSTATE_END_OF_LIST()
610     },
611 };
612 
613 static void xive_tctx_class_init(ObjectClass *klass, void *data)
614 {
615     DeviceClass *dc = DEVICE_CLASS(klass);
616 
617     dc->desc = "XIVE Interrupt Thread Context";
618     dc->realize = xive_tctx_realize;
619     dc->unrealize = xive_tctx_unrealize;
620     dc->vmsd = &vmstate_xive_tctx;
621 }
622 
623 static const TypeInfo xive_tctx_info = {
624     .name          = TYPE_XIVE_TCTX,
625     .parent        = TYPE_DEVICE,
626     .instance_size = sizeof(XiveTCTX),
627     .class_init    = xive_tctx_class_init,
628 };
629 
630 Object *xive_tctx_create(Object *cpu, XiveRouter *xrtr, Error **errp)
631 {
632     Error *local_err = NULL;
633     Object *obj;
634 
635     obj = object_new(TYPE_XIVE_TCTX);
636     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj, &error_abort);
637     object_unref(obj);
638     object_property_add_const_link(obj, "cpu", cpu, &error_abort);
639     object_property_set_bool(obj, true, "realized", &local_err);
640     if (local_err) {
641         goto error;
642     }
643 
644     return obj;
645 
646 error:
647     object_unparent(obj);
648     error_propagate(errp, local_err);
649     return NULL;
650 }
651 
652 /*
653  * XIVE ESB helpers
654  */
655 
656 static uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
657 {
658     uint8_t old_pq = *pq & 0x3;
659 
660     *pq &= ~0x3;
661     *pq |= value & 0x3;
662 
663     return old_pq;
664 }
665 
666 static bool xive_esb_trigger(uint8_t *pq)
667 {
668     uint8_t old_pq = *pq & 0x3;
669 
670     switch (old_pq) {
671     case XIVE_ESB_RESET:
672         xive_esb_set(pq, XIVE_ESB_PENDING);
673         return true;
674     case XIVE_ESB_PENDING:
675     case XIVE_ESB_QUEUED:
676         xive_esb_set(pq, XIVE_ESB_QUEUED);
677         return false;
678     case XIVE_ESB_OFF:
679         xive_esb_set(pq, XIVE_ESB_OFF);
680         return false;
681     default:
682          g_assert_not_reached();
683     }
684 }
685 
686 static bool xive_esb_eoi(uint8_t *pq)
687 {
688     uint8_t old_pq = *pq & 0x3;
689 
690     switch (old_pq) {
691     case XIVE_ESB_RESET:
692     case XIVE_ESB_PENDING:
693         xive_esb_set(pq, XIVE_ESB_RESET);
694         return false;
695     case XIVE_ESB_QUEUED:
696         xive_esb_set(pq, XIVE_ESB_PENDING);
697         return true;
698     case XIVE_ESB_OFF:
699         xive_esb_set(pq, XIVE_ESB_OFF);
700         return false;
701     default:
702          g_assert_not_reached();
703     }
704 }
705 
706 /*
707  * XIVE Interrupt Source (or IVSE)
708  */
709 
710 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
711 {
712     assert(srcno < xsrc->nr_irqs);
713 
714     return xsrc->status[srcno] & 0x3;
715 }
716 
717 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
718 {
719     assert(srcno < xsrc->nr_irqs);
720 
721     return xive_esb_set(&xsrc->status[srcno], pq);
722 }
723 
724 /*
725  * Returns whether the event notification should be forwarded.
726  */
727 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
728 {
729     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
730 
731     xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
732 
733     switch (old_pq) {
734     case XIVE_ESB_RESET:
735         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
736         return true;
737     default:
738         return false;
739     }
740 }
741 
742 /*
743  * Returns whether the event notification should be forwarded.
744  */
745 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
746 {
747     bool ret;
748 
749     assert(srcno < xsrc->nr_irqs);
750 
751     ret = xive_esb_trigger(&xsrc->status[srcno]);
752 
753     if (xive_source_irq_is_lsi(xsrc, srcno) &&
754         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
755         qemu_log_mask(LOG_GUEST_ERROR,
756                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
757     }
758 
759     return ret;
760 }
761 
762 /*
763  * Returns whether the event notification should be forwarded.
764  */
765 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
766 {
767     bool ret;
768 
769     assert(srcno < xsrc->nr_irqs);
770 
771     ret = xive_esb_eoi(&xsrc->status[srcno]);
772 
773     /*
774      * LSI sources do not set the Q bit but they can still be
775      * asserted, in which case we should forward a new event
776      * notification
777      */
778     if (xive_source_irq_is_lsi(xsrc, srcno) &&
779         xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
780         ret = xive_source_lsi_trigger(xsrc, srcno);
781     }
782 
783     return ret;
784 }
785 
786 /*
787  * Forward the source event notification to the Router
788  */
789 static void xive_source_notify(XiveSource *xsrc, int srcno)
790 {
791     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
792 
793     if (xnc->notify) {
794         xnc->notify(xsrc->xive, srcno);
795     }
796 }
797 
798 /*
799  * In a two pages ESB MMIO setting, even page is the trigger page, odd
800  * page is for management
801  */
802 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
803 {
804     return !((addr >> shift) & 1);
805 }
806 
807 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
808 {
809     return xive_source_esb_has_2page(xsrc) &&
810         addr_is_even(addr, xsrc->esb_shift - 1);
811 }
812 
813 /*
814  * ESB MMIO loads
815  *                      Trigger page    Management/EOI page
816  *
817  * ESB MMIO setting     2 pages         1 or 2 pages
818  *
819  * 0x000 .. 0x3FF       -1              EOI and return 0|1
820  * 0x400 .. 0x7FF       -1              EOI and return 0|1
821  * 0x800 .. 0xBFF       -1              return PQ
822  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
823  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
824  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
825  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
826  */
827 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
828 {
829     XiveSource *xsrc = XIVE_SOURCE(opaque);
830     uint32_t offset = addr & 0xFFF;
831     uint32_t srcno = addr >> xsrc->esb_shift;
832     uint64_t ret = -1;
833 
834     /* In a two pages ESB MMIO setting, trigger page should not be read */
835     if (xive_source_is_trigger_page(xsrc, addr)) {
836         qemu_log_mask(LOG_GUEST_ERROR,
837                       "XIVE: invalid load on IRQ %d trigger page at "
838                       "0x%"HWADDR_PRIx"\n", srcno, addr);
839         return -1;
840     }
841 
842     switch (offset) {
843     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
844         ret = xive_source_esb_eoi(xsrc, srcno);
845 
846         /* Forward the source event notification for routing */
847         if (ret) {
848             xive_source_notify(xsrc, srcno);
849         }
850         break;
851 
852     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
853         ret = xive_source_esb_get(xsrc, srcno);
854         break;
855 
856     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
857     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
858     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
859     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
860         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
861         break;
862     default:
863         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
864                       offset);
865     }
866 
867     return ret;
868 }
869 
870 /*
871  * ESB MMIO stores
872  *                      Trigger page    Management/EOI page
873  *
874  * ESB MMIO setting     2 pages         1 or 2 pages
875  *
876  * 0x000 .. 0x3FF       Trigger         Trigger
877  * 0x400 .. 0x7FF       Trigger         EOI
878  * 0x800 .. 0xBFF       Trigger         undefined
879  * 0xC00 .. 0xCFF       Trigger         PQ=00
880  * 0xD00 .. 0xDFF       Trigger         PQ=01
881  * 0xE00 .. 0xDFF       Trigger         PQ=10
882  * 0xF00 .. 0xDFF       Trigger         PQ=11
883  */
884 static void xive_source_esb_write(void *opaque, hwaddr addr,
885                                   uint64_t value, unsigned size)
886 {
887     XiveSource *xsrc = XIVE_SOURCE(opaque);
888     uint32_t offset = addr & 0xFFF;
889     uint32_t srcno = addr >> xsrc->esb_shift;
890     bool notify = false;
891 
892     /* In a two pages ESB MMIO setting, trigger page only triggers */
893     if (xive_source_is_trigger_page(xsrc, addr)) {
894         notify = xive_source_esb_trigger(xsrc, srcno);
895         goto out;
896     }
897 
898     switch (offset) {
899     case 0 ... 0x3FF:
900         notify = xive_source_esb_trigger(xsrc, srcno);
901         break;
902 
903     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
904         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
905             qemu_log_mask(LOG_GUEST_ERROR,
906                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
907             return;
908         }
909 
910         notify = xive_source_esb_eoi(xsrc, srcno);
911         break;
912 
913     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
914     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
915     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
916     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
917         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
918         break;
919 
920     default:
921         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
922                       offset);
923         return;
924     }
925 
926 out:
927     /* Forward the source event notification for routing */
928     if (notify) {
929         xive_source_notify(xsrc, srcno);
930     }
931 }
932 
933 static const MemoryRegionOps xive_source_esb_ops = {
934     .read = xive_source_esb_read,
935     .write = xive_source_esb_write,
936     .endianness = DEVICE_BIG_ENDIAN,
937     .valid = {
938         .min_access_size = 8,
939         .max_access_size = 8,
940     },
941     .impl = {
942         .min_access_size = 8,
943         .max_access_size = 8,
944     },
945 };
946 
947 void xive_source_set_irq(void *opaque, int srcno, int val)
948 {
949     XiveSource *xsrc = XIVE_SOURCE(opaque);
950     bool notify = false;
951 
952     if (xive_source_irq_is_lsi(xsrc, srcno)) {
953         if (val) {
954             notify = xive_source_lsi_trigger(xsrc, srcno);
955         } else {
956             xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
957         }
958     } else {
959         if (val) {
960             notify = xive_source_esb_trigger(xsrc, srcno);
961         }
962     }
963 
964     /* Forward the source event notification for routing */
965     if (notify) {
966         xive_source_notify(xsrc, srcno);
967     }
968 }
969 
970 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
971 {
972     int i;
973 
974     for (i = 0; i < xsrc->nr_irqs; i++) {
975         uint8_t pq = xive_source_esb_get(xsrc, i);
976 
977         if (pq == XIVE_ESB_OFF) {
978             continue;
979         }
980 
981         monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
982                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
983                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
984                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
985                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ');
986     }
987 }
988 
989 static void xive_source_reset(void *dev)
990 {
991     XiveSource *xsrc = XIVE_SOURCE(dev);
992 
993     /* Do not clear the LSI bitmap */
994 
995     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
996     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
997 }
998 
999 static void xive_source_realize(DeviceState *dev, Error **errp)
1000 {
1001     XiveSource *xsrc = XIVE_SOURCE(dev);
1002     Object *obj;
1003     Error *local_err = NULL;
1004 
1005     obj = object_property_get_link(OBJECT(dev), "xive", &local_err);
1006     if (!obj) {
1007         error_propagate(errp, local_err);
1008         error_prepend(errp, "required link 'xive' not found: ");
1009         return;
1010     }
1011 
1012     xsrc->xive = XIVE_NOTIFIER(obj);
1013 
1014     if (!xsrc->nr_irqs) {
1015         error_setg(errp, "Number of interrupt needs to be greater than 0");
1016         return;
1017     }
1018 
1019     if (xsrc->esb_shift != XIVE_ESB_4K &&
1020         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1021         xsrc->esb_shift != XIVE_ESB_64K &&
1022         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1023         error_setg(errp, "Invalid ESB shift setting");
1024         return;
1025     }
1026 
1027     xsrc->status = g_malloc0(xsrc->nr_irqs);
1028     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1029 
1030     if (!kvm_irqchip_in_kernel()) {
1031         memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1032                               &xive_source_esb_ops, xsrc, "xive.esb",
1033                               (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
1034     }
1035 
1036     qemu_register_reset(xive_source_reset, dev);
1037 }
1038 
1039 static const VMStateDescription vmstate_xive_source = {
1040     .name = TYPE_XIVE_SOURCE,
1041     .version_id = 1,
1042     .minimum_version_id = 1,
1043     .fields = (VMStateField[]) {
1044         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1045         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1046         VMSTATE_END_OF_LIST()
1047     },
1048 };
1049 
1050 /*
1051  * The default XIVE interrupt source setting for the ESB MMIOs is two
1052  * 64k pages without Store EOI, to be in sync with KVM.
1053  */
1054 static Property xive_source_properties[] = {
1055     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1056     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1057     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1058     DEFINE_PROP_END_OF_LIST(),
1059 };
1060 
1061 static void xive_source_class_init(ObjectClass *klass, void *data)
1062 {
1063     DeviceClass *dc = DEVICE_CLASS(klass);
1064 
1065     dc->desc    = "XIVE Interrupt Source";
1066     dc->props   = xive_source_properties;
1067     dc->realize = xive_source_realize;
1068     dc->vmsd    = &vmstate_xive_source;
1069 }
1070 
1071 static const TypeInfo xive_source_info = {
1072     .name          = TYPE_XIVE_SOURCE,
1073     .parent        = TYPE_DEVICE,
1074     .instance_size = sizeof(XiveSource),
1075     .class_init    = xive_source_class_init,
1076 };
1077 
1078 /*
1079  * XiveEND helpers
1080  */
1081 
1082 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
1083 {
1084     uint64_t qaddr_base = xive_end_qaddr(end);
1085     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1086     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1087     uint32_t qentries = 1 << (qsize + 10);
1088     int i;
1089 
1090     /*
1091      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1092      */
1093     monitor_printf(mon, " [ ");
1094     qindex = (qindex - (width - 1)) & (qentries - 1);
1095     for (i = 0; i < width; i++) {
1096         uint64_t qaddr = qaddr_base + (qindex << 2);
1097         uint32_t qdata = -1;
1098 
1099         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
1100                             sizeof(qdata))) {
1101             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1102                           HWADDR_PRIx "\n", qaddr);
1103             return;
1104         }
1105         monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1106                        be32_to_cpu(qdata));
1107         qindex = (qindex + 1) & (qentries - 1);
1108     }
1109 }
1110 
1111 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1112 {
1113     uint64_t qaddr_base = xive_end_qaddr(end);
1114     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1115     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1116     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1117     uint32_t qentries = 1 << (qsize + 10);
1118 
1119     uint32_t nvt = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1120     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1121 
1122     if (!xive_end_is_valid(end)) {
1123         return;
1124     }
1125 
1126     monitor_printf(mon, "  %08x %c%c%c%c%c prio:%d nvt:%04x eq:@%08"PRIx64
1127                    "% 6d/%5d ^%d", end_idx,
1128                    xive_end_is_valid(end)    ? 'v' : '-',
1129                    xive_end_is_enqueue(end)  ? 'q' : '-',
1130                    xive_end_is_notify(end)   ? 'n' : '-',
1131                    xive_end_is_backlog(end)  ? 'b' : '-',
1132                    xive_end_is_escalate(end) ? 'e' : '-',
1133                    priority, nvt, qaddr_base, qindex, qentries, qgen);
1134 
1135     xive_end_queue_pic_print_info(end, 6, mon);
1136     monitor_printf(mon, "]\n");
1137 }
1138 
1139 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1140 {
1141     uint64_t qaddr_base = xive_end_qaddr(end);
1142     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1143     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1144     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1145 
1146     uint64_t qaddr = qaddr_base + (qindex << 2);
1147     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1148     uint32_t qentries = 1 << (qsize + 10);
1149 
1150     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata))) {
1151         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1152                       HWADDR_PRIx "\n", qaddr);
1153         return;
1154     }
1155 
1156     qindex = (qindex + 1) & (qentries - 1);
1157     if (qindex == 0) {
1158         qgen ^= 1;
1159         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1160     }
1161     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1162 }
1163 
1164 /*
1165  * XIVE Router (aka. Virtualization Controller or IVRE)
1166  */
1167 
1168 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1169                         XiveEAS *eas)
1170 {
1171     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1172 
1173     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1174 }
1175 
1176 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1177                         XiveEND *end)
1178 {
1179    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1180 
1181    return xrc->get_end(xrtr, end_blk, end_idx, end);
1182 }
1183 
1184 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1185                           XiveEND *end, uint8_t word_number)
1186 {
1187    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1188 
1189    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1190 }
1191 
1192 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1193                         XiveNVT *nvt)
1194 {
1195    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1196 
1197    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1198 }
1199 
1200 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1201                         XiveNVT *nvt, uint8_t word_number)
1202 {
1203    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1204 
1205    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1206 }
1207 
1208 XiveTCTX *xive_router_get_tctx(XiveRouter *xrtr, CPUState *cs)
1209 {
1210     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1211 
1212     return xrc->get_tctx(xrtr, cs);
1213 }
1214 
1215 /*
1216  * By default on P9, the HW CAM line (23bits) is hardwired to :
1217  *
1218  *   0x000||0b1||4Bit chip number||7Bit Thread number.
1219  *
1220  * When the block grouping is enabled, the CAM line is changed to :
1221  *
1222  *   4Bit chip number||0x001||7Bit Thread number.
1223  */
1224 static uint32_t hw_cam_line(uint8_t chip_id, uint8_t tid)
1225 {
1226     return 1 << 11 | (chip_id & 0xf) << 7 | (tid & 0x7f);
1227 }
1228 
1229 static bool xive_presenter_tctx_match_hw(XiveTCTX *tctx,
1230                                          uint8_t nvt_blk, uint32_t nvt_idx)
1231 {
1232     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1233     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1234 
1235     return hw_cam_line((pir >> 8) & 0xf, pir & 0x7f) ==
1236         hw_cam_line(nvt_blk, nvt_idx);
1237 }
1238 
1239 /*
1240  * The thread context register words are in big-endian format.
1241  */
1242 static int xive_presenter_tctx_match(XiveTCTX *tctx, uint8_t format,
1243                                      uint8_t nvt_blk, uint32_t nvt_idx,
1244                                      bool cam_ignore, uint32_t logic_serv)
1245 {
1246     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1247     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1248     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1249     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1250     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1251 
1252     /*
1253      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1254      * identifier are ignored in the "CAM" match.
1255      */
1256 
1257     if (format == 0) {
1258         if (cam_ignore == true) {
1259             /*
1260              * F=0 & i=1: Logical server notification (bits ignored at
1261              * the end of the NVT identifier)
1262              */
1263             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1264                           nvt_blk, nvt_idx);
1265              return -1;
1266         }
1267 
1268         /* F=0 & i=0: Specific NVT notification */
1269 
1270         /* PHYS ring */
1271         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1272             xive_presenter_tctx_match_hw(tctx, nvt_blk, nvt_idx)) {
1273             return TM_QW3_HV_PHYS;
1274         }
1275 
1276         /* HV POOL ring */
1277         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1278             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1279             return TM_QW2_HV_POOL;
1280         }
1281 
1282         /* OS ring */
1283         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1284             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1285             return TM_QW1_OS;
1286         }
1287     } else {
1288         /* F=1 : User level Event-Based Branch (EBB) notification */
1289 
1290         /* USER ring */
1291         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1292              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1293              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1294              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1295             return TM_QW0_USER;
1296         }
1297     }
1298     return -1;
1299 }
1300 
1301 typedef struct XiveTCTXMatch {
1302     XiveTCTX *tctx;
1303     uint8_t ring;
1304 } XiveTCTXMatch;
1305 
1306 static bool xive_presenter_match(XiveRouter *xrtr, uint8_t format,
1307                                  uint8_t nvt_blk, uint32_t nvt_idx,
1308                                  bool cam_ignore, uint8_t priority,
1309                                  uint32_t logic_serv, XiveTCTXMatch *match)
1310 {
1311     CPUState *cs;
1312 
1313     /*
1314      * TODO (PowerNV): handle chip_id overwrite of block field for
1315      * hardwired CAM compares
1316      */
1317 
1318     CPU_FOREACH(cs) {
1319         XiveTCTX *tctx = xive_router_get_tctx(xrtr, cs);
1320         int ring;
1321 
1322         /*
1323          * HW checks that the CPU is enabled in the Physical Thread
1324          * Enable Register (PTER).
1325          */
1326 
1327         /*
1328          * Check the thread context CAM lines and record matches. We
1329          * will handle CPU exception delivery later
1330          */
1331         ring = xive_presenter_tctx_match(tctx, format, nvt_blk, nvt_idx,
1332                                          cam_ignore, logic_serv);
1333         /*
1334          * Save the context and follow on to catch duplicates, that we
1335          * don't support yet.
1336          */
1337         if (ring != -1) {
1338             if (match->tctx) {
1339                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
1340                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
1341                 return false;
1342             }
1343 
1344             match->ring = ring;
1345             match->tctx = tctx;
1346         }
1347     }
1348 
1349     if (!match->tctx) {
1350         qemu_log_mask(LOG_UNIMP, "XIVE: NVT %x/%x is not dispatched\n",
1351                       nvt_blk, nvt_idx);
1352         return false;
1353     }
1354 
1355     return true;
1356 }
1357 
1358 /*
1359  * This is our simple Xive Presenter Engine model. It is merged in the
1360  * Router as it does not require an extra object.
1361  *
1362  * It receives notification requests sent by the IVRE to find one
1363  * matching NVT (or more) dispatched on the processor threads. In case
1364  * of a single NVT notification, the process is abreviated and the
1365  * thread is signaled if a match is found. In case of a logical server
1366  * notification (bits ignored at the end of the NVT identifier), the
1367  * IVPE and IVRE select a winning thread using different filters. This
1368  * involves 2 or 3 exchanges on the PowerBus that the model does not
1369  * support.
1370  *
1371  * The parameters represent what is sent on the PowerBus
1372  */
1373 static void xive_presenter_notify(XiveRouter *xrtr, uint8_t format,
1374                                   uint8_t nvt_blk, uint32_t nvt_idx,
1375                                   bool cam_ignore, uint8_t priority,
1376                                   uint32_t logic_serv)
1377 {
1378     XiveNVT nvt;
1379     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1380     bool found;
1381 
1382     /* NVT cache lookup */
1383     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1384         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1385                       nvt_blk, nvt_idx);
1386         return;
1387     }
1388 
1389     if (!xive_nvt_is_valid(&nvt)) {
1390         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1391                       nvt_blk, nvt_idx);
1392         return;
1393     }
1394 
1395     found = xive_presenter_match(xrtr, format, nvt_blk, nvt_idx, cam_ignore,
1396                                  priority, logic_serv, &match);
1397     if (found) {
1398         ipb_update(&match.tctx->regs[match.ring], priority);
1399         xive_tctx_notify(match.tctx, match.ring);
1400         return;
1401     }
1402 
1403     /* Record the IPB in the associated NVT structure */
1404     ipb_update((uint8_t *) &nvt.w4, priority);
1405     xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1406 
1407     /*
1408      * If no matching NVT is dispatched on a HW thread :
1409      * - update the NVT structure if backlog is activated
1410      * - escalate (ESe PQ bits and EAS in w4-5) if escalation is
1411      *   activated
1412      */
1413 }
1414 
1415 /*
1416  * An END trigger can come from an event trigger (IPI or HW) or from
1417  * another chip. We don't model the PowerBus but the END trigger
1418  * message has the same parameters than in the function below.
1419  */
1420 static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1421                                    uint32_t end_idx, uint32_t end_data)
1422 {
1423     XiveEND end;
1424     uint8_t priority;
1425     uint8_t format;
1426 
1427     /* END cache lookup */
1428     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1429         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1430                       end_idx);
1431         return;
1432     }
1433 
1434     if (!xive_end_is_valid(&end)) {
1435         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1436                       end_blk, end_idx);
1437         return;
1438     }
1439 
1440     if (xive_end_is_enqueue(&end)) {
1441         xive_end_enqueue(&end, end_data);
1442         /* Enqueuing event data modifies the EQ toggle and index */
1443         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1444     }
1445 
1446     /*
1447      * The W7 format depends on the F bit in W6. It defines the type
1448      * of the notification :
1449      *
1450      *   F=0 : single or multiple NVT notification
1451      *   F=1 : User level Event-Based Branch (EBB) notification, no
1452      *         priority
1453      */
1454     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1455     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1456 
1457     /* The END is masked */
1458     if (format == 0 && priority == 0xff) {
1459         return;
1460     }
1461 
1462     /*
1463      * Check the END ESn (Event State Buffer for notification) for
1464      * even futher coalescing in the Router
1465      */
1466     if (!xive_end_is_notify(&end)) {
1467         uint8_t pq = xive_get_field32(END_W1_ESn, end.w1);
1468         bool notify = xive_esb_trigger(&pq);
1469 
1470         if (pq != xive_get_field32(END_W1_ESn, end.w1)) {
1471             end.w1 = xive_set_field32(END_W1_ESn, end.w1, pq);
1472             xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1473         }
1474 
1475         /* ESn[Q]=1 : end of notification */
1476         if (!notify) {
1477             return;
1478         }
1479     }
1480 
1481     /*
1482      * Follows IVPE notification
1483      */
1484     xive_presenter_notify(xrtr, format,
1485                           xive_get_field32(END_W6_NVT_BLOCK, end.w6),
1486                           xive_get_field32(END_W6_NVT_INDEX, end.w6),
1487                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1488                           priority,
1489                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1490 
1491     /* TODO: Auto EOI. */
1492 }
1493 
1494 void xive_router_notify(XiveNotifier *xn, uint32_t lisn)
1495 {
1496     XiveRouter *xrtr = XIVE_ROUTER(xn);
1497     uint8_t eas_blk = XIVE_SRCNO_BLOCK(lisn);
1498     uint32_t eas_idx = XIVE_SRCNO_INDEX(lisn);
1499     XiveEAS eas;
1500 
1501     /* EAS cache lookup */
1502     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1503         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1504         return;
1505     }
1506 
1507     /*
1508      * The IVRE checks the State Bit Cache at this point. We skip the
1509      * SBC lookup because the state bits of the sources are modeled
1510      * internally in QEMU.
1511      */
1512 
1513     if (!xive_eas_is_valid(&eas)) {
1514         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1515         return;
1516     }
1517 
1518     if (xive_eas_is_masked(&eas)) {
1519         /* Notification completed */
1520         return;
1521     }
1522 
1523     /*
1524      * The event trigger becomes an END trigger
1525      */
1526     xive_router_end_notify(xrtr,
1527                            xive_get_field64(EAS_END_BLOCK, eas.w),
1528                            xive_get_field64(EAS_END_INDEX, eas.w),
1529                            xive_get_field64(EAS_END_DATA,  eas.w));
1530 }
1531 
1532 static void xive_router_class_init(ObjectClass *klass, void *data)
1533 {
1534     DeviceClass *dc = DEVICE_CLASS(klass);
1535     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1536 
1537     dc->desc    = "XIVE Router Engine";
1538     xnc->notify = xive_router_notify;
1539 }
1540 
1541 static const TypeInfo xive_router_info = {
1542     .name          = TYPE_XIVE_ROUTER,
1543     .parent        = TYPE_SYS_BUS_DEVICE,
1544     .abstract      = true,
1545     .class_size    = sizeof(XiveRouterClass),
1546     .class_init    = xive_router_class_init,
1547     .interfaces    = (InterfaceInfo[]) {
1548         { TYPE_XIVE_NOTIFIER },
1549         { }
1550     }
1551 };
1552 
1553 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1554 {
1555     if (!xive_eas_is_valid(eas)) {
1556         return;
1557     }
1558 
1559     monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1560                    lisn, xive_eas_is_masked(eas) ? "M" : " ",
1561                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1562                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1563                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1564 }
1565 
1566 /*
1567  * END ESB MMIO loads
1568  */
1569 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1570 {
1571     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1572     uint32_t offset = addr & 0xFFF;
1573     uint8_t end_blk;
1574     uint32_t end_idx;
1575     XiveEND end;
1576     uint32_t end_esmask;
1577     uint8_t pq;
1578     uint64_t ret = -1;
1579 
1580     end_blk = xsrc->block_id;
1581     end_idx = addr >> (xsrc->esb_shift + 1);
1582 
1583     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1584         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1585                       end_idx);
1586         return -1;
1587     }
1588 
1589     if (!xive_end_is_valid(&end)) {
1590         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1591                       end_blk, end_idx);
1592         return -1;
1593     }
1594 
1595     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1596     pq = xive_get_field32(end_esmask, end.w1);
1597 
1598     switch (offset) {
1599     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1600         ret = xive_esb_eoi(&pq);
1601 
1602         /* Forward the source event notification for routing ?? */
1603         break;
1604 
1605     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1606         ret = pq;
1607         break;
1608 
1609     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1610     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1611     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1612     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1613         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1614         break;
1615     default:
1616         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1617                       offset);
1618         return -1;
1619     }
1620 
1621     if (pq != xive_get_field32(end_esmask, end.w1)) {
1622         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1623         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1624     }
1625 
1626     return ret;
1627 }
1628 
1629 /*
1630  * END ESB MMIO stores are invalid
1631  */
1632 static void xive_end_source_write(void *opaque, hwaddr addr,
1633                                   uint64_t value, unsigned size)
1634 {
1635     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1636                   HWADDR_PRIx"\n", addr);
1637 }
1638 
1639 static const MemoryRegionOps xive_end_source_ops = {
1640     .read = xive_end_source_read,
1641     .write = xive_end_source_write,
1642     .endianness = DEVICE_BIG_ENDIAN,
1643     .valid = {
1644         .min_access_size = 8,
1645         .max_access_size = 8,
1646     },
1647     .impl = {
1648         .min_access_size = 8,
1649         .max_access_size = 8,
1650     },
1651 };
1652 
1653 static void xive_end_source_realize(DeviceState *dev, Error **errp)
1654 {
1655     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1656     Object *obj;
1657     Error *local_err = NULL;
1658 
1659     obj = object_property_get_link(OBJECT(dev), "xive", &local_err);
1660     if (!obj) {
1661         error_propagate(errp, local_err);
1662         error_prepend(errp, "required link 'xive' not found: ");
1663         return;
1664     }
1665 
1666     xsrc->xrtr = XIVE_ROUTER(obj);
1667 
1668     if (!xsrc->nr_ends) {
1669         error_setg(errp, "Number of interrupt needs to be greater than 0");
1670         return;
1671     }
1672 
1673     if (xsrc->esb_shift != XIVE_ESB_4K &&
1674         xsrc->esb_shift != XIVE_ESB_64K) {
1675         error_setg(errp, "Invalid ESB shift setting");
1676         return;
1677     }
1678 
1679     /*
1680      * Each END is assigned an even/odd pair of MMIO pages, the even page
1681      * manages the ESn field while the odd page manages the ESe field.
1682      */
1683     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1684                           &xive_end_source_ops, xsrc, "xive.end",
1685                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1686 }
1687 
1688 static Property xive_end_source_properties[] = {
1689     DEFINE_PROP_UINT8("block-id", XiveENDSource, block_id, 0),
1690     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
1691     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
1692     DEFINE_PROP_END_OF_LIST(),
1693 };
1694 
1695 static void xive_end_source_class_init(ObjectClass *klass, void *data)
1696 {
1697     DeviceClass *dc = DEVICE_CLASS(klass);
1698 
1699     dc->desc    = "XIVE END Source";
1700     dc->props   = xive_end_source_properties;
1701     dc->realize = xive_end_source_realize;
1702 }
1703 
1704 static const TypeInfo xive_end_source_info = {
1705     .name          = TYPE_XIVE_END_SOURCE,
1706     .parent        = TYPE_DEVICE,
1707     .instance_size = sizeof(XiveENDSource),
1708     .class_init    = xive_end_source_class_init,
1709 };
1710 
1711 /*
1712  * XIVE Notifier
1713  */
1714 static const TypeInfo xive_notifier_info = {
1715     .name = TYPE_XIVE_NOTIFIER,
1716     .parent = TYPE_INTERFACE,
1717     .class_size = sizeof(XiveNotifierClass),
1718 };
1719 
1720 static void xive_register_types(void)
1721 {
1722     type_register_static(&xive_source_info);
1723     type_register_static(&xive_notifier_info);
1724     type_register_static(&xive_router_info);
1725     type_register_static(&xive_end_source_info);
1726     type_register_static(&xive_tctx_info);
1727 }
1728 
1729 type_init(xive_register_types)
1730