pnv_phb3.c: add unique chassis and slot for pnv_phb3_root_port
[qemu.git] / hw / pci-host / pnv_phb3.c
blob3467bbb5d9d8a35806853675a5a080ccb86d5ce4
1 /*
2 * QEMU PowerPC PowerNV (POWER8) PHB3 model
4 * Copyright (c) 2014-2020, IBM Corporation.
6 * This code is licensed under the GPL version 2 or later. See the
7 * COPYING file in the top-level directory.
8 */
9 #include "qemu/osdep.h"
10 #include "qemu/log.h"
11 #include "qapi/visitor.h"
12 #include "qapi/error.h"
13 #include "qemu-common.h"
14 #include "hw/pci-host/pnv_phb3_regs.h"
15 #include "hw/pci-host/pnv_phb3.h"
16 #include "hw/pci/pcie_host.h"
17 #include "hw/pci/pcie_port.h"
18 #include "hw/ppc/pnv.h"
19 #include "hw/irq.h"
20 #include "hw/qdev-properties.h"
21 #include "qom/object.h"
23 #define phb3_error(phb, fmt, ...) \
24 qemu_log_mask(LOG_GUEST_ERROR, "phb3[%d:%d]: " fmt "\n", \
25 (phb)->chip_id, (phb)->phb_id, ## __VA_ARGS__)
27 static PCIDevice *pnv_phb3_find_cfg_dev(PnvPHB3 *phb)
29 PCIHostState *pci = PCI_HOST_BRIDGE(phb);
30 uint64_t addr = phb->regs[PHB_CONFIG_ADDRESS >> 3];
31 uint8_t bus, devfn;
33 if (!(addr >> 63)) {
34 return NULL;
36 bus = (addr >> 52) & 0xff;
37 devfn = (addr >> 44) & 0xff;
39 return pci_find_device(pci->bus, bus, devfn);
43 * The CONFIG_DATA register expects little endian accesses, but as the
44 * region is big endian, we have to swap the value.
46 static void pnv_phb3_config_write(PnvPHB3 *phb, unsigned off,
47 unsigned size, uint64_t val)
49 uint32_t cfg_addr, limit;
50 PCIDevice *pdev;
52 pdev = pnv_phb3_find_cfg_dev(phb);
53 if (!pdev) {
54 return;
56 cfg_addr = (phb->regs[PHB_CONFIG_ADDRESS >> 3] >> 32) & 0xffc;
57 cfg_addr |= off;
58 limit = pci_config_size(pdev);
59 if (limit <= cfg_addr) {
61 * conventional pci device can be behind pcie-to-pci bridge.
62 * 256 <= addr < 4K has no effects.
64 return;
66 switch (size) {
67 case 1:
68 break;
69 case 2:
70 val = bswap16(val);
71 break;
72 case 4:
73 val = bswap32(val);
74 break;
75 default:
76 g_assert_not_reached();
78 pci_host_config_write_common(pdev, cfg_addr, limit, val, size);
81 static uint64_t pnv_phb3_config_read(PnvPHB3 *phb, unsigned off,
82 unsigned size)
84 uint32_t cfg_addr, limit;
85 PCIDevice *pdev;
86 uint64_t val;
88 pdev = pnv_phb3_find_cfg_dev(phb);
89 if (!pdev) {
90 return ~0ull;
92 cfg_addr = (phb->regs[PHB_CONFIG_ADDRESS >> 3] >> 32) & 0xffc;
93 cfg_addr |= off;
94 limit = pci_config_size(pdev);
95 if (limit <= cfg_addr) {
97 * conventional pci device can be behind pcie-to-pci bridge.
98 * 256 <= addr < 4K has no effects.
100 return ~0ull;
102 val = pci_host_config_read_common(pdev, cfg_addr, limit, size);
103 switch (size) {
104 case 1:
105 return val;
106 case 2:
107 return bswap16(val);
108 case 4:
109 return bswap32(val);
110 default:
111 g_assert_not_reached();
115 static void pnv_phb3_check_m32(PnvPHB3 *phb)
117 uint64_t base, start, size;
118 MemoryRegion *parent;
119 PnvPBCQState *pbcq = &phb->pbcq;
121 if (memory_region_is_mapped(&phb->mr_m32)) {
122 memory_region_del_subregion(phb->mr_m32.container, &phb->mr_m32);
125 if (!(phb->regs[PHB_PHB3_CONFIG >> 3] & PHB_PHB3C_M32_EN)) {
126 return;
129 /* Grab geometry from registers */
130 base = phb->regs[PHB_M32_BASE_ADDR >> 3];
131 start = phb->regs[PHB_M32_START_ADDR >> 3];
132 size = ~(phb->regs[PHB_M32_BASE_MASK >> 3] | 0xfffc000000000000ull) + 1;
134 /* Check if it matches an enabled MMIO region in the PBCQ */
135 if (memory_region_is_mapped(&pbcq->mmbar0) &&
136 base >= pbcq->mmio0_base &&
137 (base + size) <= (pbcq->mmio0_base + pbcq->mmio0_size)) {
138 parent = &pbcq->mmbar0;
139 base -= pbcq->mmio0_base;
140 } else if (memory_region_is_mapped(&pbcq->mmbar1) &&
141 base >= pbcq->mmio1_base &&
142 (base + size) <= (pbcq->mmio1_base + pbcq->mmio1_size)) {
143 parent = &pbcq->mmbar1;
144 base -= pbcq->mmio1_base;
145 } else {
146 return;
149 /* Create alias */
150 memory_region_init_alias(&phb->mr_m32, OBJECT(phb), "phb3-m32",
151 &phb->pci_mmio, start, size);
152 memory_region_add_subregion(parent, base, &phb->mr_m32);
155 static void pnv_phb3_check_m64(PnvPHB3 *phb, uint32_t index)
157 uint64_t base, start, size, m64;
158 MemoryRegion *parent;
159 PnvPBCQState *pbcq = &phb->pbcq;
161 if (memory_region_is_mapped(&phb->mr_m64[index])) {
162 /* Should we destroy it in RCU friendly way... ? */
163 memory_region_del_subregion(phb->mr_m64[index].container,
164 &phb->mr_m64[index]);
167 /* Get table entry */
168 m64 = phb->ioda_M64BT[index];
170 if (!(m64 & IODA2_M64BT_ENABLE)) {
171 return;
174 /* Grab geometry from registers */
175 base = GETFIELD(IODA2_M64BT_BASE, m64) << 20;
176 if (m64 & IODA2_M64BT_SINGLE_PE) {
177 base &= ~0x1ffffffull;
179 size = GETFIELD(IODA2_M64BT_MASK, m64) << 20;
180 size |= 0xfffc000000000000ull;
181 size = ~size + 1;
182 start = base | (phb->regs[PHB_M64_UPPER_BITS >> 3]);
184 /* Check if it matches an enabled MMIO region in the PBCQ */
185 if (memory_region_is_mapped(&pbcq->mmbar0) &&
186 base >= pbcq->mmio0_base &&
187 (base + size) <= (pbcq->mmio0_base + pbcq->mmio0_size)) {
188 parent = &pbcq->mmbar0;
189 base -= pbcq->mmio0_base;
190 } else if (memory_region_is_mapped(&pbcq->mmbar1) &&
191 base >= pbcq->mmio1_base &&
192 (base + size) <= (pbcq->mmio1_base + pbcq->mmio1_size)) {
193 parent = &pbcq->mmbar1;
194 base -= pbcq->mmio1_base;
195 } else {
196 return;
199 /* Create alias */
200 memory_region_init_alias(&phb->mr_m64[index], OBJECT(phb), "phb3-m64",
201 &phb->pci_mmio, start, size);
202 memory_region_add_subregion(parent, base, &phb->mr_m64[index]);
205 static void pnv_phb3_check_all_m64s(PnvPHB3 *phb)
207 uint64_t i;
209 for (i = 0; i < PNV_PHB3_NUM_M64; i++) {
210 pnv_phb3_check_m64(phb, i);
214 static void pnv_phb3_lxivt_write(PnvPHB3 *phb, unsigned idx, uint64_t val)
216 uint8_t server, prio;
218 phb->ioda_LXIVT[idx] = val & (IODA2_LXIVT_SERVER |
219 IODA2_LXIVT_PRIORITY |
220 IODA2_LXIVT_NODE_ID);
221 server = GETFIELD(IODA2_LXIVT_SERVER, val);
222 prio = GETFIELD(IODA2_LXIVT_PRIORITY, val);
225 * The low order 2 bits are the link pointer (Type II interrupts).
226 * Shift back to get a valid IRQ server.
228 server >>= 2;
230 ics_write_xive(&phb->lsis, idx, server, prio, prio);
233 static uint64_t *pnv_phb3_ioda_access(PnvPHB3 *phb,
234 unsigned *out_table, unsigned *out_idx)
236 uint64_t adreg = phb->regs[PHB_IODA_ADDR >> 3];
237 unsigned int index = GETFIELD(PHB_IODA_AD_TADR, adreg);
238 unsigned int table = GETFIELD(PHB_IODA_AD_TSEL, adreg);
239 unsigned int mask;
240 uint64_t *tptr = NULL;
242 switch (table) {
243 case IODA2_TBL_LIST:
244 tptr = phb->ioda_LIST;
245 mask = 7;
246 break;
247 case IODA2_TBL_LXIVT:
248 tptr = phb->ioda_LXIVT;
249 mask = 7;
250 break;
251 case IODA2_TBL_IVC_CAM:
252 case IODA2_TBL_RBA:
253 mask = 31;
254 break;
255 case IODA2_TBL_RCAM:
256 mask = 63;
257 break;
258 case IODA2_TBL_MRT:
259 mask = 7;
260 break;
261 case IODA2_TBL_PESTA:
262 case IODA2_TBL_PESTB:
263 mask = 255;
264 break;
265 case IODA2_TBL_TVT:
266 tptr = phb->ioda_TVT;
267 mask = 511;
268 break;
269 case IODA2_TBL_TCAM:
270 case IODA2_TBL_TDR:
271 mask = 63;
272 break;
273 case IODA2_TBL_M64BT:
274 tptr = phb->ioda_M64BT;
275 mask = 15;
276 break;
277 case IODA2_TBL_M32DT:
278 tptr = phb->ioda_MDT;
279 mask = 255;
280 break;
281 case IODA2_TBL_PEEV:
282 tptr = phb->ioda_PEEV;
283 mask = 3;
284 break;
285 default:
286 phb3_error(phb, "invalid IODA table %d", table);
287 return NULL;
289 index &= mask;
290 if (out_idx) {
291 *out_idx = index;
293 if (out_table) {
294 *out_table = table;
296 if (tptr) {
297 tptr += index;
299 if (adreg & PHB_IODA_AD_AUTOINC) {
300 index = (index + 1) & mask;
301 adreg = SETFIELD(PHB_IODA_AD_TADR, adreg, index);
303 phb->regs[PHB_IODA_ADDR >> 3] = adreg;
304 return tptr;
307 static uint64_t pnv_phb3_ioda_read(PnvPHB3 *phb)
309 unsigned table;
310 uint64_t *tptr;
312 tptr = pnv_phb3_ioda_access(phb, &table, NULL);
313 if (!tptr) {
314 /* Return 0 on unsupported tables, not ff's */
315 return 0;
317 return *tptr;
320 static void pnv_phb3_ioda_write(PnvPHB3 *phb, uint64_t val)
322 unsigned table, idx;
323 uint64_t *tptr;
325 tptr = pnv_phb3_ioda_access(phb, &table, &idx);
326 if (!tptr) {
327 return;
330 /* Handle side effects */
331 switch (table) {
332 case IODA2_TBL_LXIVT:
333 pnv_phb3_lxivt_write(phb, idx, val);
334 break;
335 case IODA2_TBL_M64BT:
336 *tptr = val;
337 pnv_phb3_check_m64(phb, idx);
338 break;
339 default:
340 *tptr = val;
345 * This is called whenever the PHB LSI, MSI source ID register or
346 * the PBCQ irq filters are written.
348 void pnv_phb3_remap_irqs(PnvPHB3 *phb)
350 ICSState *ics = &phb->lsis;
351 uint32_t local, global, count, mask, comp;
352 uint64_t baren;
353 PnvPBCQState *pbcq = &phb->pbcq;
356 * First check if we are enabled. Unlike real HW we don't separate
357 * TX and RX so we enable if both are set
359 baren = pbcq->nest_regs[PBCQ_NEST_BAR_EN];
360 if (!(baren & PBCQ_NEST_BAR_EN_IRSN_RX) ||
361 !(baren & PBCQ_NEST_BAR_EN_IRSN_TX)) {
362 ics->offset = 0;
363 return;
366 /* Grab local LSI source ID */
367 local = GETFIELD(PHB_LSI_SRC_ID, phb->regs[PHB_LSI_SOURCE_ID >> 3]) << 3;
369 /* Grab global one and compare */
370 global = GETFIELD(PBCQ_NEST_LSI_SRC,
371 pbcq->nest_regs[PBCQ_NEST_LSI_SRC_ID]) << 3;
372 if (global != local) {
374 * This happens during initialization, let's come back when we
375 * are properly configured
377 ics->offset = 0;
378 return;
381 /* Get the base on the powerbus */
382 comp = GETFIELD(PBCQ_NEST_IRSN_COMP,
383 pbcq->nest_regs[PBCQ_NEST_IRSN_COMPARE]);
384 mask = GETFIELD(PBCQ_NEST_IRSN_COMP,
385 pbcq->nest_regs[PBCQ_NEST_IRSN_MASK]);
386 count = ((~mask) + 1) & 0x7ffff;
387 phb->total_irq = count;
389 /* Sanity checks */
390 if ((global + PNV_PHB3_NUM_LSI) > count) {
391 phb3_error(phb, "LSIs out of reach: LSI base=%d total irq=%d", global,
392 count);
395 if (count > 2048) {
396 phb3_error(phb, "More interrupts than supported: %d", count);
399 if ((comp & mask) != comp) {
400 phb3_error(phb, "IRQ compare bits not in mask: comp=0x%x mask=0x%x",
401 comp, mask);
402 comp &= mask;
404 /* Setup LSI offset */
405 ics->offset = comp + global;
407 /* Setup MSI offset */
408 pnv_phb3_msi_update_config(&phb->msis, comp, count - PNV_PHB3_NUM_LSI);
411 static void pnv_phb3_lsi_src_id_write(PnvPHB3 *phb, uint64_t val)
413 /* Sanitize content */
414 val &= PHB_LSI_SRC_ID;
415 phb->regs[PHB_LSI_SOURCE_ID >> 3] = val;
416 pnv_phb3_remap_irqs(phb);
419 static void pnv_phb3_rtc_invalidate(PnvPHB3 *phb, uint64_t val)
421 PnvPhb3DMASpace *ds;
423 /* Always invalidate all for now ... */
424 QLIST_FOREACH(ds, &phb->dma_spaces, list) {
425 ds->pe_num = PHB_INVALID_PE;
430 static void pnv_phb3_update_msi_regions(PnvPhb3DMASpace *ds)
432 uint64_t cfg = ds->phb->regs[PHB_PHB3_CONFIG >> 3];
434 if (cfg & PHB_PHB3C_32BIT_MSI_EN) {
435 if (!memory_region_is_mapped(&ds->msi32_mr)) {
436 memory_region_add_subregion(MEMORY_REGION(&ds->dma_mr),
437 0xffff0000, &ds->msi32_mr);
439 } else {
440 if (memory_region_is_mapped(&ds->msi32_mr)) {
441 memory_region_del_subregion(MEMORY_REGION(&ds->dma_mr),
442 &ds->msi32_mr);
446 if (cfg & PHB_PHB3C_64BIT_MSI_EN) {
447 if (!memory_region_is_mapped(&ds->msi64_mr)) {
448 memory_region_add_subregion(MEMORY_REGION(&ds->dma_mr),
449 (1ull << 60), &ds->msi64_mr);
451 } else {
452 if (memory_region_is_mapped(&ds->msi64_mr)) {
453 memory_region_del_subregion(MEMORY_REGION(&ds->dma_mr),
454 &ds->msi64_mr);
459 static void pnv_phb3_update_all_msi_regions(PnvPHB3 *phb)
461 PnvPhb3DMASpace *ds;
463 QLIST_FOREACH(ds, &phb->dma_spaces, list) {
464 pnv_phb3_update_msi_regions(ds);
468 void pnv_phb3_reg_write(void *opaque, hwaddr off, uint64_t val, unsigned size)
470 PnvPHB3 *phb = opaque;
471 bool changed;
473 /* Special case configuration data */
474 if ((off & 0xfffc) == PHB_CONFIG_DATA) {
475 pnv_phb3_config_write(phb, off & 0x3, size, val);
476 return;
479 /* Other registers are 64-bit only */
480 if (size != 8 || off & 0x7) {
481 phb3_error(phb, "Invalid register access, offset: 0x%"PRIx64" size: %d",
482 off, size);
483 return;
486 /* Handle masking & filtering */
487 switch (off) {
488 case PHB_M64_UPPER_BITS:
489 val &= 0xfffc000000000000ull;
490 break;
491 case PHB_Q_DMA_R:
493 * This is enough logic to make SW happy but we aren't actually
494 * quiescing the DMAs
496 if (val & PHB_Q_DMA_R_AUTORESET) {
497 val = 0;
498 } else {
499 val &= PHB_Q_DMA_R_QUIESCE_DMA;
501 break;
502 /* LEM stuff */
503 case PHB_LEM_FIR_AND_MASK:
504 phb->regs[PHB_LEM_FIR_ACCUM >> 3] &= val;
505 return;
506 case PHB_LEM_FIR_OR_MASK:
507 phb->regs[PHB_LEM_FIR_ACCUM >> 3] |= val;
508 return;
509 case PHB_LEM_ERROR_AND_MASK:
510 phb->regs[PHB_LEM_ERROR_MASK >> 3] &= val;
511 return;
512 case PHB_LEM_ERROR_OR_MASK:
513 phb->regs[PHB_LEM_ERROR_MASK >> 3] |= val;
514 return;
515 case PHB_LEM_WOF:
516 val = 0;
517 break;
520 /* Record whether it changed */
521 changed = phb->regs[off >> 3] != val;
523 /* Store in register cache first */
524 phb->regs[off >> 3] = val;
526 /* Handle side effects */
527 switch (off) {
528 case PHB_PHB3_CONFIG:
529 if (changed) {
530 pnv_phb3_update_all_msi_regions(phb);
532 /* fall through */
533 case PHB_M32_BASE_ADDR:
534 case PHB_M32_BASE_MASK:
535 case PHB_M32_START_ADDR:
536 if (changed) {
537 pnv_phb3_check_m32(phb);
539 break;
540 case PHB_M64_UPPER_BITS:
541 if (changed) {
542 pnv_phb3_check_all_m64s(phb);
544 break;
545 case PHB_LSI_SOURCE_ID:
546 if (changed) {
547 pnv_phb3_lsi_src_id_write(phb, val);
549 break;
551 /* IODA table accesses */
552 case PHB_IODA_DATA0:
553 pnv_phb3_ioda_write(phb, val);
554 break;
556 /* RTC invalidation */
557 case PHB_RTC_INVALIDATE:
558 pnv_phb3_rtc_invalidate(phb, val);
559 break;
561 /* FFI request */
562 case PHB_FFI_REQUEST:
563 pnv_phb3_msi_ffi(&phb->msis, val);
564 break;
566 /* Silent simple writes */
567 case PHB_CONFIG_ADDRESS:
568 case PHB_IODA_ADDR:
569 case PHB_TCE_KILL:
570 case PHB_TCE_SPEC_CTL:
571 case PHB_PEST_BAR:
572 case PHB_PELTV_BAR:
573 case PHB_RTT_BAR:
574 case PHB_RBA_BAR:
575 case PHB_IVT_BAR:
576 case PHB_FFI_LOCK:
577 case PHB_LEM_FIR_ACCUM:
578 case PHB_LEM_ERROR_MASK:
579 case PHB_LEM_ACTION0:
580 case PHB_LEM_ACTION1:
581 break;
583 /* Noise on anything else */
584 default:
585 qemu_log_mask(LOG_UNIMP, "phb3: reg_write 0x%"PRIx64"=%"PRIx64"\n",
586 off, val);
590 uint64_t pnv_phb3_reg_read(void *opaque, hwaddr off, unsigned size)
592 PnvPHB3 *phb = opaque;
593 PCIHostState *pci = PCI_HOST_BRIDGE(phb);
594 uint64_t val;
596 if ((off & 0xfffc) == PHB_CONFIG_DATA) {
597 return pnv_phb3_config_read(phb, off & 0x3, size);
600 /* Other registers are 64-bit only */
601 if (size != 8 || off & 0x7) {
602 phb3_error(phb, "Invalid register access, offset: 0x%"PRIx64" size: %d",
603 off, size);
604 return ~0ull;
607 /* Default read from cache */
608 val = phb->regs[off >> 3];
610 switch (off) {
611 /* Simulate venice DD2.0 */
612 case PHB_VERSION:
613 return 0x000000a300000005ull;
614 case PHB_PCIE_SYSTEM_CONFIG:
615 return 0x441100fc30000000;
617 /* IODA table accesses */
618 case PHB_IODA_DATA0:
619 return pnv_phb3_ioda_read(phb);
621 /* Link training always appears trained */
622 case PHB_PCIE_DLP_TRAIN_CTL:
623 if (!pci_find_device(pci->bus, 1, 0)) {
624 return 0;
626 return PHB_PCIE_DLP_INBAND_PRESENCE | PHB_PCIE_DLP_TC_DL_LINKACT;
628 /* FFI Lock */
629 case PHB_FFI_LOCK:
630 /* Set lock and return previous value */
631 phb->regs[off >> 3] |= PHB_FFI_LOCK_STATE;
632 return val;
634 /* DMA read sync: make it look like it's complete */
635 case PHB_DMARD_SYNC:
636 return PHB_DMARD_SYNC_COMPLETE;
638 /* Silent simple reads */
639 case PHB_PHB3_CONFIG:
640 case PHB_M32_BASE_ADDR:
641 case PHB_M32_BASE_MASK:
642 case PHB_M32_START_ADDR:
643 case PHB_CONFIG_ADDRESS:
644 case PHB_IODA_ADDR:
645 case PHB_RTC_INVALIDATE:
646 case PHB_TCE_KILL:
647 case PHB_TCE_SPEC_CTL:
648 case PHB_PEST_BAR:
649 case PHB_PELTV_BAR:
650 case PHB_RTT_BAR:
651 case PHB_RBA_BAR:
652 case PHB_IVT_BAR:
653 case PHB_M64_UPPER_BITS:
654 case PHB_LEM_FIR_ACCUM:
655 case PHB_LEM_ERROR_MASK:
656 case PHB_LEM_ACTION0:
657 case PHB_LEM_ACTION1:
658 break;
660 /* Noise on anything else */
661 default:
662 qemu_log_mask(LOG_UNIMP, "phb3: reg_read 0x%"PRIx64"=%"PRIx64"\n",
663 off, val);
665 return val;
668 static const MemoryRegionOps pnv_phb3_reg_ops = {
669 .read = pnv_phb3_reg_read,
670 .write = pnv_phb3_reg_write,
671 .valid.min_access_size = 1,
672 .valid.max_access_size = 8,
673 .impl.min_access_size = 1,
674 .impl.max_access_size = 8,
675 .endianness = DEVICE_BIG_ENDIAN,
678 static int pnv_phb3_map_irq(PCIDevice *pci_dev, int irq_num)
680 /* Check that out properly ... */
681 return irq_num & 3;
684 static void pnv_phb3_set_irq(void *opaque, int irq_num, int level)
686 PnvPHB3 *phb = opaque;
688 /* LSI only ... */
689 if (irq_num > 3) {
690 phb3_error(phb, "Unknown IRQ to set %d", irq_num);
692 qemu_set_irq(phb->qirqs[irq_num], level);
695 static bool pnv_phb3_resolve_pe(PnvPhb3DMASpace *ds)
697 uint64_t rtt, addr;
698 uint16_t rte;
699 int bus_num;
701 /* Already resolved ? */
702 if (ds->pe_num != PHB_INVALID_PE) {
703 return true;
706 /* We need to lookup the RTT */
707 rtt = ds->phb->regs[PHB_RTT_BAR >> 3];
708 if (!(rtt & PHB_RTT_BAR_ENABLE)) {
709 phb3_error(ds->phb, "DMA with RTT BAR disabled !");
710 /* Set error bits ? fence ? ... */
711 return false;
714 /* Read RTE */
715 bus_num = pci_bus_num(ds->bus);
716 addr = rtt & PHB_RTT_BASE_ADDRESS_MASK;
717 addr += 2 * ((bus_num << 8) | ds->devfn);
718 if (dma_memory_read(&address_space_memory, addr, &rte,
719 sizeof(rte), MEMTXATTRS_UNSPECIFIED)) {
720 phb3_error(ds->phb, "Failed to read RTT entry at 0x%"PRIx64, addr);
721 /* Set error bits ? fence ? ... */
722 return false;
724 rte = be16_to_cpu(rte);
726 /* Fail upon reading of invalid PE# */
727 if (rte >= PNV_PHB3_NUM_PE) {
728 phb3_error(ds->phb, "RTE for RID 0x%x invalid (%04x", ds->devfn, rte);
729 /* Set error bits ? fence ? ... */
730 return false;
732 ds->pe_num = rte;
733 return true;
736 static void pnv_phb3_translate_tve(PnvPhb3DMASpace *ds, hwaddr addr,
737 bool is_write, uint64_t tve,
738 IOMMUTLBEntry *tlb)
740 uint64_t tta = GETFIELD(IODA2_TVT_TABLE_ADDR, tve);
741 int32_t lev = GETFIELD(IODA2_TVT_NUM_LEVELS, tve);
742 uint32_t tts = GETFIELD(IODA2_TVT_TCE_TABLE_SIZE, tve);
743 uint32_t tps = GETFIELD(IODA2_TVT_IO_PSIZE, tve);
744 PnvPHB3 *phb = ds->phb;
746 /* Invalid levels */
747 if (lev > 4) {
748 phb3_error(phb, "Invalid #levels in TVE %d", lev);
749 return;
752 /* IO Page Size of 0 means untranslated, else use TCEs */
753 if (tps == 0) {
755 * We only support non-translate in top window.
757 * TODO: Venice/Murano support it on bottom window above 4G and
758 * Naples suports it on everything
760 if (!(tve & PPC_BIT(51))) {
761 phb3_error(phb, "xlate for invalid non-translate TVE");
762 return;
764 /* TODO: Handle boundaries */
766 /* Use 4k pages like q35 ... for now */
767 tlb->iova = addr & 0xfffffffffffff000ull;
768 tlb->translated_addr = addr & 0x0003fffffffff000ull;
769 tlb->addr_mask = 0xfffull;
770 tlb->perm = IOMMU_RW;
771 } else {
772 uint32_t tce_shift, tbl_shift, sh;
773 uint64_t base, taddr, tce, tce_mask;
775 /* TVE disabled ? */
776 if (tts == 0) {
777 phb3_error(phb, "xlate for invalid translated TVE");
778 return;
781 /* Address bits per bottom level TCE entry */
782 tce_shift = tps + 11;
784 /* Address bits per table level */
785 tbl_shift = tts + 8;
787 /* Top level table base address */
788 base = tta << 12;
790 /* Total shift to first level */
791 sh = tbl_shift * lev + tce_shift;
793 /* TODO: Multi-level untested */
794 while ((lev--) >= 0) {
795 /* Grab the TCE address */
796 taddr = base | (((addr >> sh) & ((1ul << tbl_shift) - 1)) << 3);
797 if (dma_memory_read(&address_space_memory, taddr, &tce,
798 sizeof(tce), MEMTXATTRS_UNSPECIFIED)) {
799 phb3_error(phb, "Failed to read TCE at 0x%"PRIx64, taddr);
800 return;
802 tce = be64_to_cpu(tce);
804 /* Check permission for indirect TCE */
805 if ((lev >= 0) && !(tce & 3)) {
806 phb3_error(phb, "Invalid indirect TCE at 0x%"PRIx64, taddr);
807 phb3_error(phb, " xlate %"PRIx64":%c TVE=%"PRIx64, addr,
808 is_write ? 'W' : 'R', tve);
809 phb3_error(phb, " tta=%"PRIx64" lev=%d tts=%d tps=%d",
810 tta, lev, tts, tps);
811 return;
813 sh -= tbl_shift;
814 base = tce & ~0xfffull;
817 /* We exit the loop with TCE being the final TCE */
818 tce_mask = ~((1ull << tce_shift) - 1);
819 tlb->iova = addr & tce_mask;
820 tlb->translated_addr = tce & tce_mask;
821 tlb->addr_mask = ~tce_mask;
822 tlb->perm = tce & 3;
823 if ((is_write & !(tce & 2)) || ((!is_write) && !(tce & 1))) {
824 phb3_error(phb, "TCE access fault at 0x%"PRIx64, taddr);
825 phb3_error(phb, " xlate %"PRIx64":%c TVE=%"PRIx64, addr,
826 is_write ? 'W' : 'R', tve);
827 phb3_error(phb, " tta=%"PRIx64" lev=%d tts=%d tps=%d",
828 tta, lev, tts, tps);
833 static IOMMUTLBEntry pnv_phb3_translate_iommu(IOMMUMemoryRegion *iommu,
834 hwaddr addr,
835 IOMMUAccessFlags flag,
836 int iommu_idx)
838 PnvPhb3DMASpace *ds = container_of(iommu, PnvPhb3DMASpace, dma_mr);
839 int tve_sel;
840 uint64_t tve, cfg;
841 IOMMUTLBEntry ret = {
842 .target_as = &address_space_memory,
843 .iova = addr,
844 .translated_addr = 0,
845 .addr_mask = ~(hwaddr)0,
846 .perm = IOMMU_NONE,
848 PnvPHB3 *phb = ds->phb;
850 /* Resolve PE# */
851 if (!pnv_phb3_resolve_pe(ds)) {
852 phb3_error(phb, "Failed to resolve PE# for bus @%p (%d) devfn 0x%x",
853 ds->bus, pci_bus_num(ds->bus), ds->devfn);
854 return ret;
857 /* Check top bits */
858 switch (addr >> 60) {
859 case 00:
860 /* DMA or 32-bit MSI ? */
861 cfg = ds->phb->regs[PHB_PHB3_CONFIG >> 3];
862 if ((cfg & PHB_PHB3C_32BIT_MSI_EN) &&
863 ((addr & 0xffffffffffff0000ull) == 0xffff0000ull)) {
864 phb3_error(phb, "xlate on 32-bit MSI region");
865 return ret;
867 /* Choose TVE XXX Use PHB3 Control Register */
868 tve_sel = (addr >> 59) & 1;
869 tve = ds->phb->ioda_TVT[ds->pe_num * 2 + tve_sel];
870 pnv_phb3_translate_tve(ds, addr, flag & IOMMU_WO, tve, &ret);
871 break;
872 case 01:
873 phb3_error(phb, "xlate on 64-bit MSI region");
874 break;
875 default:
876 phb3_error(phb, "xlate on unsupported address 0x%"PRIx64, addr);
878 return ret;
881 #define TYPE_PNV_PHB3_IOMMU_MEMORY_REGION "pnv-phb3-iommu-memory-region"
882 DECLARE_INSTANCE_CHECKER(IOMMUMemoryRegion, PNV_PHB3_IOMMU_MEMORY_REGION,
883 TYPE_PNV_PHB3_IOMMU_MEMORY_REGION)
885 static void pnv_phb3_iommu_memory_region_class_init(ObjectClass *klass,
886 void *data)
888 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
890 imrc->translate = pnv_phb3_translate_iommu;
893 static const TypeInfo pnv_phb3_iommu_memory_region_info = {
894 .parent = TYPE_IOMMU_MEMORY_REGION,
895 .name = TYPE_PNV_PHB3_IOMMU_MEMORY_REGION,
896 .class_init = pnv_phb3_iommu_memory_region_class_init,
900 * MSI/MSIX memory region implementation.
901 * The handler handles both MSI and MSIX.
903 static void pnv_phb3_msi_write(void *opaque, hwaddr addr,
904 uint64_t data, unsigned size)
906 PnvPhb3DMASpace *ds = opaque;
908 /* Resolve PE# */
909 if (!pnv_phb3_resolve_pe(ds)) {
910 phb3_error(ds->phb, "Failed to resolve PE# for bus @%p (%d) devfn 0x%x",
911 ds->bus, pci_bus_num(ds->bus), ds->devfn);
912 return;
915 pnv_phb3_msi_send(&ds->phb->msis, addr, data, ds->pe_num);
918 /* There is no .read as the read result is undefined by PCI spec */
919 static uint64_t pnv_phb3_msi_read(void *opaque, hwaddr addr, unsigned size)
921 PnvPhb3DMASpace *ds = opaque;
923 phb3_error(ds->phb, "invalid read @ 0x%" HWADDR_PRIx, addr);
924 return -1;
927 static const MemoryRegionOps pnv_phb3_msi_ops = {
928 .read = pnv_phb3_msi_read,
929 .write = pnv_phb3_msi_write,
930 .endianness = DEVICE_LITTLE_ENDIAN
933 static AddressSpace *pnv_phb3_dma_iommu(PCIBus *bus, void *opaque, int devfn)
935 PnvPHB3 *phb = opaque;
936 PnvPhb3DMASpace *ds;
938 QLIST_FOREACH(ds, &phb->dma_spaces, list) {
939 if (ds->bus == bus && ds->devfn == devfn) {
940 break;
944 if (ds == NULL) {
945 ds = g_malloc0(sizeof(PnvPhb3DMASpace));
946 ds->bus = bus;
947 ds->devfn = devfn;
948 ds->pe_num = PHB_INVALID_PE;
949 ds->phb = phb;
950 memory_region_init_iommu(&ds->dma_mr, sizeof(ds->dma_mr),
951 TYPE_PNV_PHB3_IOMMU_MEMORY_REGION,
952 OBJECT(phb), "phb3_iommu", UINT64_MAX);
953 address_space_init(&ds->dma_as, MEMORY_REGION(&ds->dma_mr),
954 "phb3_iommu");
955 memory_region_init_io(&ds->msi32_mr, OBJECT(phb), &pnv_phb3_msi_ops,
956 ds, "msi32", 0x10000);
957 memory_region_init_io(&ds->msi64_mr, OBJECT(phb), &pnv_phb3_msi_ops,
958 ds, "msi64", 0x100000);
959 pnv_phb3_update_msi_regions(ds);
961 QLIST_INSERT_HEAD(&phb->dma_spaces, ds, list);
963 return &ds->dma_as;
966 static void pnv_phb3_instance_init(Object *obj)
968 PnvPHB3 *phb = PNV_PHB3(obj);
970 QLIST_INIT(&phb->dma_spaces);
972 /* LSI sources */
973 object_initialize_child(obj, "lsi", &phb->lsis, TYPE_ICS);
975 /* Default init ... will be fixed by HW inits */
976 phb->lsis.offset = 0;
978 /* MSI sources */
979 object_initialize_child(obj, "msi", &phb->msis, TYPE_PHB3_MSI);
981 /* Power Bus Common Queue */
982 object_initialize_child(obj, "pbcq", &phb->pbcq, TYPE_PNV_PBCQ);
984 /* Root Port */
985 object_initialize_child(obj, "root", &phb->root, TYPE_PNV_PHB3_ROOT_PORT);
986 qdev_prop_set_int32(DEVICE(&phb->root), "addr", PCI_DEVFN(0, 0));
987 qdev_prop_set_bit(DEVICE(&phb->root), "multifunction", false);
990 static void pnv_phb3_realize(DeviceState *dev, Error **errp)
992 PnvPHB3 *phb = PNV_PHB3(dev);
993 PCIHostState *pci = PCI_HOST_BRIDGE(dev);
994 PnvMachineState *pnv = PNV_MACHINE(qdev_get_machine());
995 int i;
997 if (phb->phb_id >= PNV_CHIP_GET_CLASS(phb->chip)->num_phbs) {
998 error_setg(errp, "invalid PHB index: %d", phb->phb_id);
999 return;
1002 /* LSI sources */
1003 object_property_set_link(OBJECT(&phb->lsis), "xics", OBJECT(pnv),
1004 &error_abort);
1005 object_property_set_int(OBJECT(&phb->lsis), "nr-irqs", PNV_PHB3_NUM_LSI,
1006 &error_abort);
1007 if (!qdev_realize(DEVICE(&phb->lsis), NULL, errp)) {
1008 return;
1011 for (i = 0; i < phb->lsis.nr_irqs; i++) {
1012 ics_set_irq_type(&phb->lsis, i, true);
1015 phb->qirqs = qemu_allocate_irqs(ics_set_irq, &phb->lsis, phb->lsis.nr_irqs);
1017 /* MSI sources */
1018 object_property_set_link(OBJECT(&phb->msis), "phb", OBJECT(phb),
1019 &error_abort);
1020 object_property_set_link(OBJECT(&phb->msis), "xics", OBJECT(pnv),
1021 &error_abort);
1022 object_property_set_int(OBJECT(&phb->msis), "nr-irqs", PHB3_MAX_MSI,
1023 &error_abort);
1024 if (!qdev_realize(DEVICE(&phb->msis), NULL, errp)) {
1025 return;
1028 /* Power Bus Common Queue */
1029 object_property_set_link(OBJECT(&phb->pbcq), "phb", OBJECT(phb),
1030 &error_abort);
1031 if (!qdev_realize(DEVICE(&phb->pbcq), NULL, errp)) {
1032 return;
1035 /* Controller Registers */
1036 memory_region_init_io(&phb->mr_regs, OBJECT(phb), &pnv_phb3_reg_ops, phb,
1037 "phb3-regs", 0x1000);
1040 * PHB3 doesn't support IO space. However, qemu gets very upset if
1041 * we don't have an IO region to anchor IO BARs onto so we just
1042 * initialize one which we never hook up to anything
1044 memory_region_init(&phb->pci_io, OBJECT(phb), "pci-io", 0x10000);
1045 memory_region_init(&phb->pci_mmio, OBJECT(phb), "pci-mmio",
1046 PCI_MMIO_TOTAL_SIZE);
1048 pci->bus = pci_register_root_bus(dev,
1049 dev->id ? dev->id : NULL,
1050 pnv_phb3_set_irq, pnv_phb3_map_irq, phb,
1051 &phb->pci_mmio, &phb->pci_io,
1052 0, 4, TYPE_PNV_PHB3_ROOT_BUS);
1054 pci_setup_iommu(pci->bus, pnv_phb3_dma_iommu, phb);
1056 /* Add a single Root port */
1057 qdev_prop_set_uint8(DEVICE(&phb->root), "chassis", phb->chip_id);
1058 qdev_prop_set_uint16(DEVICE(&phb->root), "slot", phb->phb_id);
1059 qdev_realize(DEVICE(&phb->root), BUS(pci->bus), &error_fatal);
1062 void pnv_phb3_update_regions(PnvPHB3 *phb)
1064 PnvPBCQState *pbcq = &phb->pbcq;
1066 /* Unmap first always */
1067 if (memory_region_is_mapped(&phb->mr_regs)) {
1068 memory_region_del_subregion(&pbcq->phbbar, &phb->mr_regs);
1071 /* Map registers if enabled */
1072 if (memory_region_is_mapped(&pbcq->phbbar)) {
1073 /* TODO: We should use the PHB BAR 2 register but we don't ... */
1074 memory_region_add_subregion(&pbcq->phbbar, 0, &phb->mr_regs);
1077 /* Check/update m32 */
1078 if (memory_region_is_mapped(&phb->mr_m32)) {
1079 pnv_phb3_check_m32(phb);
1081 pnv_phb3_check_all_m64s(phb);
1084 static const char *pnv_phb3_root_bus_path(PCIHostState *host_bridge,
1085 PCIBus *rootbus)
1087 PnvPHB3 *phb = PNV_PHB3(host_bridge);
1089 snprintf(phb->bus_path, sizeof(phb->bus_path), "00%02x:%02x",
1090 phb->chip_id, phb->phb_id);
1091 return phb->bus_path;
1094 static Property pnv_phb3_properties[] = {
1095 DEFINE_PROP_UINT32("index", PnvPHB3, phb_id, 0),
1096 DEFINE_PROP_UINT32("chip-id", PnvPHB3, chip_id, 0),
1097 DEFINE_PROP_LINK("chip", PnvPHB3, chip, TYPE_PNV_CHIP, PnvChip *),
1098 DEFINE_PROP_END_OF_LIST(),
1101 static void pnv_phb3_class_init(ObjectClass *klass, void *data)
1103 PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
1104 DeviceClass *dc = DEVICE_CLASS(klass);
1106 hc->root_bus_path = pnv_phb3_root_bus_path;
1107 dc->realize = pnv_phb3_realize;
1108 device_class_set_props(dc, pnv_phb3_properties);
1109 set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
1110 dc->user_creatable = false;
1113 static const TypeInfo pnv_phb3_type_info = {
1114 .name = TYPE_PNV_PHB3,
1115 .parent = TYPE_PCIE_HOST_BRIDGE,
1116 .instance_size = sizeof(PnvPHB3),
1117 .class_init = pnv_phb3_class_init,
1118 .instance_init = pnv_phb3_instance_init,
1121 static void pnv_phb3_root_bus_class_init(ObjectClass *klass, void *data)
1123 BusClass *k = BUS_CLASS(klass);
1126 * PHB3 has only a single root complex. Enforce the limit on the
1127 * parent bus
1129 k->max_dev = 1;
1132 static const TypeInfo pnv_phb3_root_bus_info = {
1133 .name = TYPE_PNV_PHB3_ROOT_BUS,
1134 .parent = TYPE_PCIE_BUS,
1135 .class_init = pnv_phb3_root_bus_class_init,
1136 .interfaces = (InterfaceInfo[]) {
1137 { INTERFACE_PCIE_DEVICE },
1142 static void pnv_phb3_root_port_realize(DeviceState *dev, Error **errp)
1144 PCIERootPortClass *rpc = PCIE_ROOT_PORT_GET_CLASS(dev);
1145 PCIDevice *pci = PCI_DEVICE(dev);
1146 PCIBus *bus = pci_get_bus(pci);
1147 PnvPHB3 *phb = NULL;
1148 Error *local_err = NULL;
1150 phb = (PnvPHB3 *) object_dynamic_cast(OBJECT(bus->qbus.parent),
1151 TYPE_PNV_PHB3);
1153 if (!phb) {
1154 error_setg(errp,
1155 "pnv_phb3_root_port devices must be connected to pnv-phb3 buses");
1156 return;
1159 /* Set unique chassis/slot values for the root port */
1160 qdev_prop_set_uint8(&pci->qdev, "chassis", phb->chip_id);
1161 qdev_prop_set_uint16(&pci->qdev, "slot", phb->phb_id);
1163 rpc->parent_realize(dev, &local_err);
1164 if (local_err) {
1165 error_propagate(errp, local_err);
1166 return;
1170 static void pnv_phb3_root_port_class_init(ObjectClass *klass, void *data)
1172 DeviceClass *dc = DEVICE_CLASS(klass);
1173 PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1174 PCIERootPortClass *rpc = PCIE_ROOT_PORT_CLASS(klass);
1176 dc->desc = "IBM PHB3 PCIE Root Port";
1178 device_class_set_parent_realize(dc, pnv_phb3_root_port_realize,
1179 &rpc->parent_realize);
1180 dc->user_creatable = false;
1182 k->vendor_id = PCI_VENDOR_ID_IBM;
1183 k->device_id = 0x03dc;
1184 k->revision = 0;
1186 rpc->exp_offset = 0x48;
1187 rpc->aer_offset = 0x100;
1190 static const TypeInfo pnv_phb3_root_port_info = {
1191 .name = TYPE_PNV_PHB3_ROOT_PORT,
1192 .parent = TYPE_PCIE_ROOT_PORT,
1193 .instance_size = sizeof(PnvPHB3RootPort),
1194 .class_init = pnv_phb3_root_port_class_init,
1197 static void pnv_phb3_register_types(void)
1199 type_register_static(&pnv_phb3_root_bus_info);
1200 type_register_static(&pnv_phb3_root_port_info);
1201 type_register_static(&pnv_phb3_type_info);
1202 type_register_static(&pnv_phb3_iommu_memory_region_info);
1205 type_init(pnv_phb3_register_types)