xref: /openbmc/qemu/hw/cxl/cxl-component-utils.c (revision 55abfc1f)
1 /*
2  * CXL Utility library for components
3  *
4  * Copyright(C) 2020 Intel Corporation.
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2. See the
7  * COPYING file in the top-level directory.
8  */
9 
10 #include "qemu/osdep.h"
11 #include "qemu/log.h"
12 #include "qapi/error.h"
13 #include "hw/pci/pci.h"
14 #include "hw/cxl/cxl.h"
15 
16 /* CXL r3.0 Section 8.2.4.19.1 CXL HDM Decoder Capability Register */
17 int cxl_decoder_count_enc(int count)
18 {
19     switch (count) {
20     case 1: return 0x0;
21     case 2: return 0x1;
22     case 4: return 0x2;
23     case 6: return 0x3;
24     case 8: return 0x4;
25     case 10: return 0x5;
26     /* Switches and Host Bridges may have more than 10 decoders */
27     case 12: return 0x6;
28     case 14: return 0x7;
29     case 16: return 0x8;
30     case 20: return 0x9;
31     case 24: return 0xa;
32     case 28: return 0xb;
33     case 32: return 0xc;
34     }
35     return 0;
36 }
37 
38 int cxl_decoder_count_dec(int enc_cnt)
39 {
40     switch (enc_cnt) {
41     case 0x0: return 1;
42     case 0x1: return 2;
43     case 0x2: return 4;
44     case 0x3: return 6;
45     case 0x4: return 8;
46     case 0x5: return 10;
47     /* Switches and Host Bridges may have more than 10 decoders */
48     case 0x6: return 12;
49     case 0x7: return 14;
50     case 0x8: return 16;
51     case 0x9: return 20;
52     case 0xa: return 24;
53     case 0xb: return 28;
54     case 0xc: return 32;
55     }
56     return 0;
57 }
58 
59 hwaddr cxl_decode_ig(int ig)
60 {
61     return 1ULL << (ig + 8);
62 }
63 
64 static uint64_t cxl_cache_mem_read_reg(void *opaque, hwaddr offset,
65                                        unsigned size)
66 {
67     CXLComponentState *cxl_cstate = opaque;
68     ComponentRegisters *cregs = &cxl_cstate->crb;
69 
70     switch (size) {
71     case 4:
72         if (cregs->special_ops && cregs->special_ops->read) {
73             return cregs->special_ops->read(cxl_cstate, offset, 4);
74         } else {
75             QEMU_BUILD_BUG_ON(sizeof(*cregs->cache_mem_registers) != 4);
76             return cregs->cache_mem_registers[offset / 4];
77         }
78     case 8:
79         qemu_log_mask(LOG_UNIMP,
80                       "CXL 8 byte cache mem registers not implemented\n");
81         return 0;
82     default:
83         /*
84          * In line with specification limitaions on access sizes, this
85          * routine is not called with other sizes.
86          */
87         g_assert_not_reached();
88     }
89 }
90 
91 static void dumb_hdm_handler(CXLComponentState *cxl_cstate, hwaddr offset,
92                              uint32_t value)
93 {
94     ComponentRegisters *cregs = &cxl_cstate->crb;
95     uint32_t *cache_mem = cregs->cache_mem_registers;
96     bool should_commit = false;
97     bool should_uncommit = false;
98 
99     switch (offset) {
100     case A_CXL_HDM_DECODER0_CTRL:
101     case A_CXL_HDM_DECODER1_CTRL:
102     case A_CXL_HDM_DECODER2_CTRL:
103     case A_CXL_HDM_DECODER3_CTRL:
104         should_commit = FIELD_EX32(value, CXL_HDM_DECODER0_CTRL, COMMIT);
105         should_uncommit = !should_commit;
106         break;
107     default:
108         break;
109     }
110 
111     if (should_commit) {
112         value = FIELD_DP32(value, CXL_HDM_DECODER0_CTRL, ERR, 0);
113         value = FIELD_DP32(value, CXL_HDM_DECODER0_CTRL, COMMITTED, 1);
114     } else if (should_uncommit) {
115         value = FIELD_DP32(value, CXL_HDM_DECODER0_CTRL, ERR, 0);
116         value = FIELD_DP32(value, CXL_HDM_DECODER0_CTRL, COMMITTED, 0);
117     }
118     stl_le_p((uint8_t *)cache_mem + offset, value);
119 }
120 
121 static void cxl_cache_mem_write_reg(void *opaque, hwaddr offset, uint64_t value,
122                                     unsigned size)
123 {
124     CXLComponentState *cxl_cstate = opaque;
125     ComponentRegisters *cregs = &cxl_cstate->crb;
126     uint32_t mask;
127 
128     switch (size) {
129     case 4: {
130         QEMU_BUILD_BUG_ON(sizeof(*cregs->cache_mem_regs_write_mask) != 4);
131         QEMU_BUILD_BUG_ON(sizeof(*cregs->cache_mem_registers) != 4);
132         mask = cregs->cache_mem_regs_write_mask[offset / 4];
133         value &= mask;
134         /* RO bits should remain constant. Done by reading existing value */
135         value |= ~mask & cregs->cache_mem_registers[offset / 4];
136         if (cregs->special_ops && cregs->special_ops->write) {
137             cregs->special_ops->write(cxl_cstate, offset, value, size);
138             return;
139         }
140 
141         if (offset >= A_CXL_HDM_DECODER_CAPABILITY &&
142             offset <= A_CXL_HDM_DECODER3_TARGET_LIST_HI) {
143             dumb_hdm_handler(cxl_cstate, offset, value);
144         } else {
145             cregs->cache_mem_registers[offset / 4] = value;
146         }
147         return;
148     }
149     case 8:
150         qemu_log_mask(LOG_UNIMP,
151                       "CXL 8 byte cache mem registers not implemented\n");
152         return;
153     default:
154         /*
155          * In line with specification limitaions on access sizes, this
156          * routine is not called with other sizes.
157          */
158         g_assert_not_reached();
159     }
160 }
161 
162 /*
163  * 8.2.3
164  *   The access restrictions specified in Section 8.2.2 also apply to CXL 2.0
165  *   Component Registers.
166  *
167  * 8.2.2
168  *   • A 32 bit register shall be accessed as a 4 Bytes quantity. Partial
169  *   reads are not permitted.
170  *   • A 64 bit register shall be accessed as a 8 Bytes quantity. Partial
171  *   reads are not permitted.
172  *
173  * As of the spec defined today, only 4 byte registers exist.
174  */
175 static const MemoryRegionOps cache_mem_ops = {
176     .read = cxl_cache_mem_read_reg,
177     .write = cxl_cache_mem_write_reg,
178     .endianness = DEVICE_LITTLE_ENDIAN,
179     .valid = {
180         .min_access_size = 4,
181         .max_access_size = 8,
182         .unaligned = false,
183     },
184     .impl = {
185         .min_access_size = 4,
186         .max_access_size = 8,
187     },
188 };
189 
190 void cxl_component_register_block_init(Object *obj,
191                                        CXLComponentState *cxl_cstate,
192                                        const char *type)
193 {
194     ComponentRegisters *cregs = &cxl_cstate->crb;
195 
196     memory_region_init(&cregs->component_registers, obj, type,
197                        CXL2_COMPONENT_BLOCK_SIZE);
198 
199     /* io registers controls link which we don't care about in QEMU */
200     memory_region_init_io(&cregs->io, obj, NULL, cregs, ".io",
201                           CXL2_COMPONENT_IO_REGION_SIZE);
202     memory_region_init_io(&cregs->cache_mem, obj, &cache_mem_ops, cregs,
203                           ".cache_mem", CXL2_COMPONENT_CM_REGION_SIZE);
204 
205     memory_region_add_subregion(&cregs->component_registers, 0, &cregs->io);
206     memory_region_add_subregion(&cregs->component_registers,
207                                 CXL2_COMPONENT_IO_REGION_SIZE,
208                                 &cregs->cache_mem);
209 }
210 
211 static void ras_init_common(uint32_t *reg_state, uint32_t *write_msk)
212 {
213     /*
214      * Error status is RW1C but given bits are not yet set, it can
215      * be handled as RO.
216      */
217     stl_le_p(reg_state + R_CXL_RAS_UNC_ERR_STATUS, 0);
218     stl_le_p(write_msk + R_CXL_RAS_UNC_ERR_STATUS, 0x1cfff);
219     /* Bits 12-13 and 17-31 reserved in CXL 2.0 */
220     stl_le_p(reg_state + R_CXL_RAS_UNC_ERR_MASK, 0x1cfff);
221     stl_le_p(write_msk + R_CXL_RAS_UNC_ERR_MASK, 0x1cfff);
222     stl_le_p(reg_state + R_CXL_RAS_UNC_ERR_SEVERITY, 0x1cfff);
223     stl_le_p(write_msk + R_CXL_RAS_UNC_ERR_SEVERITY, 0x1cfff);
224     stl_le_p(reg_state + R_CXL_RAS_COR_ERR_STATUS, 0);
225     stl_le_p(write_msk + R_CXL_RAS_COR_ERR_STATUS, 0x7f);
226     stl_le_p(reg_state + R_CXL_RAS_COR_ERR_MASK, 0x7f);
227     stl_le_p(write_msk + R_CXL_RAS_COR_ERR_MASK, 0x7f);
228     /* CXL switches and devices must set */
229     stl_le_p(reg_state + R_CXL_RAS_ERR_CAP_CTRL, 0x200);
230 }
231 
232 static void hdm_init_common(uint32_t *reg_state, uint32_t *write_msk,
233                             enum reg_type type)
234 {
235     int decoder_count = CXL_HDM_DECODER_COUNT;
236     int hdm_inc = R_CXL_HDM_DECODER1_BASE_LO - R_CXL_HDM_DECODER0_BASE_LO;
237     int i;
238 
239     ARRAY_FIELD_DP32(reg_state, CXL_HDM_DECODER_CAPABILITY, DECODER_COUNT,
240                      cxl_decoder_count_enc(decoder_count));
241     ARRAY_FIELD_DP32(reg_state, CXL_HDM_DECODER_CAPABILITY, TARGET_COUNT, 1);
242     ARRAY_FIELD_DP32(reg_state, CXL_HDM_DECODER_CAPABILITY, INTERLEAVE_256B, 1);
243     ARRAY_FIELD_DP32(reg_state, CXL_HDM_DECODER_CAPABILITY, INTERLEAVE_4K, 1);
244     ARRAY_FIELD_DP32(reg_state, CXL_HDM_DECODER_CAPABILITY,
245                      POISON_ON_ERR_CAP, 0);
246     ARRAY_FIELD_DP32(reg_state, CXL_HDM_DECODER_GLOBAL_CONTROL,
247                      HDM_DECODER_ENABLE, 0);
248     write_msk[R_CXL_HDM_DECODER_GLOBAL_CONTROL] = 0x3;
249     for (i = 0; i < decoder_count; i++) {
250         write_msk[R_CXL_HDM_DECODER0_BASE_LO + i * hdm_inc] = 0xf0000000;
251         write_msk[R_CXL_HDM_DECODER0_BASE_HI + i * hdm_inc] = 0xffffffff;
252         write_msk[R_CXL_HDM_DECODER0_SIZE_LO + i * hdm_inc] = 0xf0000000;
253         write_msk[R_CXL_HDM_DECODER0_SIZE_HI + i * hdm_inc] = 0xffffffff;
254         write_msk[R_CXL_HDM_DECODER0_CTRL + i * hdm_inc] = 0x13ff;
255         if (type == CXL2_DEVICE ||
256             type == CXL2_TYPE3_DEVICE ||
257             type == CXL2_LOGICAL_DEVICE) {
258             write_msk[R_CXL_HDM_DECODER0_TARGET_LIST_LO + i * hdm_inc] =
259                 0xf0000000;
260         } else {
261             write_msk[R_CXL_HDM_DECODER0_TARGET_LIST_LO + i * hdm_inc] =
262                 0xffffffff;
263         }
264         write_msk[R_CXL_HDM_DECODER0_TARGET_LIST_HI + i * hdm_inc] = 0xffffffff;
265     }
266 }
267 
268 void cxl_component_register_init_common(uint32_t *reg_state,
269                                         uint32_t *write_msk,
270                                         enum reg_type type)
271 {
272     int caps = 0;
273 
274     /*
275      * In CXL 2.0 the capabilities required for each CXL component are such
276      * that, with the ordering chosen here, a single number can be used to
277      * define which capabilities should be provided.
278      */
279     switch (type) {
280     case CXL2_DOWNSTREAM_PORT:
281     case CXL2_DEVICE:
282         /* RAS, Link */
283         caps = 2;
284         break;
285     case CXL2_UPSTREAM_PORT:
286     case CXL2_TYPE3_DEVICE:
287     case CXL2_LOGICAL_DEVICE:
288         /* + HDM */
289         caps = 3;
290         break;
291     case CXL2_ROOT_PORT:
292         /* + Extended Security, + Snoop */
293         caps = 5;
294         break;
295     default:
296         abort();
297     }
298 
299     memset(reg_state, 0, CXL2_COMPONENT_CM_REGION_SIZE);
300 
301     /* CXL Capability Header Register */
302     ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, ID, 1);
303     ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, VERSION, 1);
304     ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, CACHE_MEM_VERSION, 1);
305     ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, ARRAY_SIZE, caps);
306 
307 #define init_cap_reg(reg, id, version)                                        \
308     do {                                                                      \
309         int which = R_CXL_##reg##_CAPABILITY_HEADER;                          \
310         reg_state[which] = FIELD_DP32(reg_state[which],                       \
311                                       CXL_##reg##_CAPABILITY_HEADER, ID, id); \
312         reg_state[which] =                                                    \
313             FIELD_DP32(reg_state[which], CXL_##reg##_CAPABILITY_HEADER,       \
314                        VERSION, version);                                     \
315         reg_state[which] =                                                    \
316             FIELD_DP32(reg_state[which], CXL_##reg##_CAPABILITY_HEADER, PTR,  \
317                        CXL_##reg##_REGISTERS_OFFSET);                         \
318     } while (0)
319 
320     init_cap_reg(RAS, 2, 2);
321     ras_init_common(reg_state, write_msk);
322 
323     init_cap_reg(LINK, 4, 2);
324 
325     if (caps < 3) {
326         return;
327     }
328 
329     init_cap_reg(HDM, 5, 1);
330     hdm_init_common(reg_state, write_msk, type);
331 
332     if (caps < 5) {
333         return;
334     }
335 
336     init_cap_reg(EXTSEC, 6, 1);
337     init_cap_reg(SNOOP, 8, 1);
338 
339 #undef init_cap_reg
340 }
341 
342 /*
343  * Helper to creates a DVSEC header for a CXL entity. The caller is responsible
344  * for tracking the valid offset.
345  *
346  * This function will build the DVSEC header on behalf of the caller and then
347  * copy in the remaining data for the vendor specific bits.
348  * It will also set up appropriate write masks.
349  */
350 void cxl_component_create_dvsec(CXLComponentState *cxl,
351                                 enum reg_type cxl_dev_type, uint16_t length,
352                                 uint16_t type, uint8_t rev, uint8_t *body)
353 {
354     PCIDevice *pdev = cxl->pdev;
355     uint16_t offset = cxl->dvsec_offset;
356     uint8_t *wmask = pdev->wmask;
357 
358     assert(offset >= PCI_CFG_SPACE_SIZE &&
359            ((offset + length) < PCI_CFG_SPACE_EXP_SIZE));
360     assert((length & 0xf000) == 0);
361     assert((rev & ~0xf) == 0);
362 
363     /* Create the DVSEC in the MCFG space */
364     pcie_add_capability(pdev, PCI_EXT_CAP_ID_DVSEC, 1, offset, length);
365     pci_set_long(pdev->config + offset + PCIE_DVSEC_HEADER1_OFFSET,
366                  (length << 20) | (rev << 16) | CXL_VENDOR_ID);
367     pci_set_word(pdev->config + offset + PCIE_DVSEC_ID_OFFSET, type);
368     memcpy(pdev->config + offset + sizeof(DVSECHeader),
369            body + sizeof(DVSECHeader),
370            length - sizeof(DVSECHeader));
371 
372     /* Configure write masks */
373     switch (type) {
374     case PCIE_CXL_DEVICE_DVSEC:
375         /* Cntrl RW Lock - so needs explicit blocking when lock is set */
376         wmask[offset + offsetof(CXLDVSECDevice, ctrl)] = 0xFD;
377         wmask[offset + offsetof(CXLDVSECDevice, ctrl) + 1] = 0x4F;
378         /* Status is RW1CS */
379         wmask[offset + offsetof(CXLDVSECDevice, ctrl2)] = 0x0F;
380        /* Lock is RW Once */
381         wmask[offset + offsetof(CXLDVSECDevice, lock)] = 0x01;
382         /* range1/2_base_high/low is RW Lock */
383         wmask[offset + offsetof(CXLDVSECDevice, range1_base_hi)] = 0xFF;
384         wmask[offset + offsetof(CXLDVSECDevice, range1_base_hi) + 1] = 0xFF;
385         wmask[offset + offsetof(CXLDVSECDevice, range1_base_hi) + 2] = 0xFF;
386         wmask[offset + offsetof(CXLDVSECDevice, range1_base_hi) + 3] = 0xFF;
387         wmask[offset + offsetof(CXLDVSECDevice, range1_base_lo) + 3] = 0xF0;
388         wmask[offset + offsetof(CXLDVSECDevice, range2_base_hi)] = 0xFF;
389         wmask[offset + offsetof(CXLDVSECDevice, range2_base_hi) + 1] = 0xFF;
390         wmask[offset + offsetof(CXLDVSECDevice, range2_base_hi) + 2] = 0xFF;
391         wmask[offset + offsetof(CXLDVSECDevice, range2_base_hi) + 3] = 0xFF;
392         wmask[offset + offsetof(CXLDVSECDevice, range2_base_lo) + 3] = 0xF0;
393         break;
394     case NON_CXL_FUNCTION_MAP_DVSEC:
395         break; /* Not yet implemented */
396     case EXTENSIONS_PORT_DVSEC:
397         wmask[offset + offsetof(CXLDVSECPortExt, control)] = 0x0F;
398         wmask[offset + offsetof(CXLDVSECPortExt, control) + 1] = 0x40;
399         wmask[offset + offsetof(CXLDVSECPortExt, alt_bus_base)] = 0xFF;
400         wmask[offset + offsetof(CXLDVSECPortExt, alt_bus_limit)] = 0xFF;
401         wmask[offset + offsetof(CXLDVSECPortExt, alt_memory_base)] = 0xF0;
402         wmask[offset + offsetof(CXLDVSECPortExt, alt_memory_base) + 1] = 0xFF;
403         wmask[offset + offsetof(CXLDVSECPortExt, alt_memory_limit)] = 0xF0;
404         wmask[offset + offsetof(CXLDVSECPortExt, alt_memory_limit) + 1] = 0xFF;
405         wmask[offset + offsetof(CXLDVSECPortExt, alt_prefetch_base)] = 0xF0;
406         wmask[offset + offsetof(CXLDVSECPortExt, alt_prefetch_base) + 1] = 0xFF;
407         wmask[offset + offsetof(CXLDVSECPortExt, alt_prefetch_limit)] = 0xF0;
408         wmask[offset + offsetof(CXLDVSECPortExt, alt_prefetch_limit) + 1] =
409             0xFF;
410         wmask[offset + offsetof(CXLDVSECPortExt, alt_prefetch_base_high)] =
411             0xFF;
412         wmask[offset + offsetof(CXLDVSECPortExt, alt_prefetch_base_high) + 1] =
413             0xFF;
414         wmask[offset + offsetof(CXLDVSECPortExt, alt_prefetch_base_high) + 2] =
415             0xFF;
416         wmask[offset + offsetof(CXLDVSECPortExt, alt_prefetch_base_high) + 3] =
417             0xFF;
418         wmask[offset + offsetof(CXLDVSECPortExt, alt_prefetch_limit_high)] =
419             0xFF;
420         wmask[offset + offsetof(CXLDVSECPortExt, alt_prefetch_limit_high) + 1] =
421             0xFF;
422         wmask[offset + offsetof(CXLDVSECPortExt, alt_prefetch_limit_high) + 2] =
423             0xFF;
424         wmask[offset + offsetof(CXLDVSECPortExt, alt_prefetch_limit_high) + 3] =
425             0xFF;
426         break;
427     case GPF_PORT_DVSEC:
428         wmask[offset + offsetof(CXLDVSECPortGPF, phase1_ctrl)] = 0x0F;
429         wmask[offset + offsetof(CXLDVSECPortGPF, phase1_ctrl) + 1] = 0x0F;
430         wmask[offset + offsetof(CXLDVSECPortGPF, phase2_ctrl)] = 0x0F;
431         wmask[offset + offsetof(CXLDVSECPortGPF, phase2_ctrl) + 1] = 0x0F;
432         break;
433     case GPF_DEVICE_DVSEC:
434         wmask[offset + offsetof(CXLDVSECDeviceGPF, phase2_duration)] = 0x0F;
435         wmask[offset + offsetof(CXLDVSECDeviceGPF, phase2_duration) + 1] = 0x0F;
436         wmask[offset + offsetof(CXLDVSECDeviceGPF, phase2_power)] = 0xFF;
437         wmask[offset + offsetof(CXLDVSECDeviceGPF, phase2_power) + 1] = 0xFF;
438         wmask[offset + offsetof(CXLDVSECDeviceGPF, phase2_power) + 2] = 0xFF;
439         wmask[offset + offsetof(CXLDVSECDeviceGPF, phase2_power) + 3] = 0xFF;
440         break;
441     case PCIE_FLEXBUS_PORT_DVSEC:
442         switch (cxl_dev_type) {
443         case CXL2_ROOT_PORT:
444             /* No MLD */
445             wmask[offset + offsetof(CXLDVSECPortFlexBus, ctrl)] = 0xbd;
446             break;
447         case CXL2_DOWNSTREAM_PORT:
448             wmask[offset + offsetof(CXLDVSECPortFlexBus, ctrl)] = 0xfd;
449             break;
450         default: /* Registers are RO for other component types */
451             break;
452         }
453         /* There are rw1cs bits in the status register but never set */
454         break;
455     }
456 
457     /* Update state for future DVSEC additions */
458     range_init_nofail(&cxl->dvsecs[type], cxl->dvsec_offset, length);
459     cxl->dvsec_offset += length;
460 }
461 
462 /* CXL r3.0 Section 8.2.4.19.7 CXL HDM Decoder n Control Register */
463 uint8_t cxl_interleave_ways_enc(int iw, Error **errp)
464 {
465     switch (iw) {
466     case 1: return 0x0;
467     case 2: return 0x1;
468     case 4: return 0x2;
469     case 8: return 0x3;
470     case 16: return 0x4;
471     case 3: return 0x8;
472     case 6: return 0x9;
473     case 12: return 0xa;
474     default:
475         error_setg(errp, "Interleave ways: %d not supported", iw);
476         return 0;
477     }
478 }
479 
480 int cxl_interleave_ways_dec(uint8_t iw_enc, Error **errp)
481 {
482     switch (iw_enc) {
483     case 0x0: return 1;
484     case 0x1: return 2;
485     case 0x2: return 4;
486     case 0x3: return 8;
487     case 0x4: return 16;
488     case 0x8: return 3;
489     case 0x9: return 6;
490     case 0xa: return 12;
491     default:
492         error_setg(errp, "Encoded interleave ways: %d not supported", iw_enc);
493         return 0;
494     }
495 }
496 
497 uint8_t cxl_interleave_granularity_enc(uint64_t gran, Error **errp)
498 {
499     switch (gran) {
500     case 256: return 0;
501     case 512: return 1;
502     case 1024: return 2;
503     case 2048: return 3;
504     case 4096: return 4;
505     case 8192: return 5;
506     case 16384: return 6;
507     default:
508         error_setg(errp, "Interleave granularity: %" PRIu64 " invalid", gran);
509         return 0;
510     }
511 }
512