2 * Copyright 2014 IBM Corp.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
10 #include <linux/pci_regs.h>
11 #include <linux/pci_ids.h>
12 #include <linux/device.h>
13 #include <linux/module.h>
14 #include <linux/kernel.h>
15 #include <linux/slab.h>
16 #include <linux/sort.h>
17 #include <linux/pci.h>
19 #include <linux/delay.h>
21 #include <asm/msi_bitmap.h>
22 #include <asm/pnv-pci.h>
30 #define CXL_PCI_VSEC_ID 0x1280
31 #define CXL_VSEC_MIN_SIZE 0x80
33 #define CXL_READ_VSEC_LENGTH(dev, vsec, dest) \
35 pci_read_config_word(dev, vsec + 0x6, dest); \
38 #define CXL_READ_VSEC_NAFUS(dev, vsec, dest) \
39 pci_read_config_byte(dev, vsec + 0x8, dest)
41 #define CXL_READ_VSEC_STATUS(dev, vsec, dest) \
42 pci_read_config_byte(dev, vsec + 0x9, dest)
43 #define CXL_STATUS_SECOND_PORT 0x80
44 #define CXL_STATUS_MSI_X_FULL 0x40
45 #define CXL_STATUS_MSI_X_SINGLE 0x20
46 #define CXL_STATUS_FLASH_RW 0x08
47 #define CXL_STATUS_FLASH_RO 0x04
48 #define CXL_STATUS_LOADABLE_AFU 0x02
49 #define CXL_STATUS_LOADABLE_PSL 0x01
50 /* If we see these features we won't try to use the card */
51 #define CXL_UNSUPPORTED_FEATURES \
52 (CXL_STATUS_MSI_X_FULL | CXL_STATUS_MSI_X_SINGLE)
54 #define CXL_READ_VSEC_MODE_CONTROL(dev, vsec, dest) \
55 pci_read_config_byte(dev, vsec + 0xa, dest)
56 #define CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val) \
57 pci_write_config_byte(dev, vsec + 0xa, val)
58 #define CXL_WRITE_VSEC_MODE_CONTROL_BUS(bus, devfn, vsec, val) \
59 pci_bus_write_config_byte(bus, devfn, vsec + 0xa, val)
60 #define CXL_VSEC_PROTOCOL_MASK 0xe0
61 #define CXL_VSEC_PROTOCOL_1024TB 0x80
62 #define CXL_VSEC_PROTOCOL_512TB 0x40
63 #define CXL_VSEC_PROTOCOL_256TB 0x20 /* Power 8/9 uses this */
64 #define CXL_VSEC_PROTOCOL_ENABLE 0x01
66 #define CXL_READ_VSEC_PSL_REVISION(dev, vsec, dest) \
67 pci_read_config_word(dev, vsec + 0xc, dest)
68 #define CXL_READ_VSEC_CAIA_MINOR(dev, vsec, dest) \
69 pci_read_config_byte(dev, vsec + 0xe, dest)
70 #define CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, dest) \
71 pci_read_config_byte(dev, vsec + 0xf, dest)
72 #define CXL_READ_VSEC_BASE_IMAGE(dev, vsec, dest) \
73 pci_read_config_word(dev, vsec + 0x10, dest)
75 #define CXL_READ_VSEC_IMAGE_STATE(dev, vsec, dest) \
76 pci_read_config_byte(dev, vsec + 0x13, dest)
77 #define CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, val) \
78 pci_write_config_byte(dev, vsec + 0x13, val)
79 #define CXL_VSEC_USER_IMAGE_LOADED 0x80 /* RO */
80 #define CXL_VSEC_PERST_LOADS_IMAGE 0x20 /* RW */
81 #define CXL_VSEC_PERST_SELECT_USER 0x10 /* RW */
83 #define CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, dest) \
84 pci_read_config_dword(dev, vsec + 0x20, dest)
85 #define CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, dest) \
86 pci_read_config_dword(dev, vsec + 0x24, dest)
87 #define CXL_READ_VSEC_PS_OFF(dev, vsec, dest) \
88 pci_read_config_dword(dev, vsec + 0x28, dest)
89 #define CXL_READ_VSEC_PS_SIZE(dev, vsec, dest) \
90 pci_read_config_dword(dev, vsec + 0x2c, dest)
93 /* This works a little different than the p1/p2 register accesses to make it
94 * easier to pull out individual fields */
95 #define AFUD_READ(afu, off) in_be64(afu->native->afu_desc_mmio + off)
96 #define AFUD_READ_LE(afu, off) in_le64(afu->native->afu_desc_mmio + off)
97 #define EXTRACT_PPC_BIT(val, bit) (!!(val & PPC_BIT(bit)))
98 #define EXTRACT_PPC_BITS(val, bs, be) ((val & PPC_BITMASK(bs, be)) >> PPC_BITLSHIFT(be))
100 #define AFUD_READ_INFO(afu) AFUD_READ(afu, 0x0)
101 #define AFUD_NUM_INTS_PER_PROC(val) EXTRACT_PPC_BITS(val, 0, 15)
102 #define AFUD_NUM_PROCS(val) EXTRACT_PPC_BITS(val, 16, 31)
103 #define AFUD_NUM_CRS(val) EXTRACT_PPC_BITS(val, 32, 47)
104 #define AFUD_MULTIMODE(val) EXTRACT_PPC_BIT(val, 48)
105 #define AFUD_PUSH_BLOCK_TRANSFER(val) EXTRACT_PPC_BIT(val, 55)
106 #define AFUD_DEDICATED_PROCESS(val) EXTRACT_PPC_BIT(val, 59)
107 #define AFUD_AFU_DIRECTED(val) EXTRACT_PPC_BIT(val, 61)
108 #define AFUD_TIME_SLICED(val) EXTRACT_PPC_BIT(val, 63)
109 #define AFUD_READ_CR(afu) AFUD_READ(afu, 0x20)
110 #define AFUD_CR_LEN(val) EXTRACT_PPC_BITS(val, 8, 63)
111 #define AFUD_READ_CR_OFF(afu) AFUD_READ(afu, 0x28)
112 #define AFUD_READ_PPPSA(afu) AFUD_READ(afu, 0x30)
113 #define AFUD_PPPSA_PP(val) EXTRACT_PPC_BIT(val, 6)
114 #define AFUD_PPPSA_PSA(val) EXTRACT_PPC_BIT(val, 7)
115 #define AFUD_PPPSA_LEN(val) EXTRACT_PPC_BITS(val, 8, 63)
116 #define AFUD_READ_PPPSA_OFF(afu) AFUD_READ(afu, 0x38)
117 #define AFUD_READ_EB(afu) AFUD_READ(afu, 0x40)
118 #define AFUD_EB_LEN(val) EXTRACT_PPC_BITS(val, 8, 63)
119 #define AFUD_READ_EB_OFF(afu) AFUD_READ(afu, 0x48)
121 static const struct pci_device_id cxl_pci_tbl[] = {
122 { PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0477), },
123 { PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x044b), },
124 { PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x04cf), },
125 { PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0601), },
126 { PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0623), },
127 { PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0628), },
128 { PCI_DEVICE_CLASS(0x120000, ~0), },
132 MODULE_DEVICE_TABLE(pci, cxl_pci_tbl);
136 * Mostly using these wrappers to avoid confusion:
137 * priv 1 is BAR2, while priv 2 is BAR0
139 static inline resource_size_t p1_base(struct pci_dev *dev)
141 return pci_resource_start(dev, 2);
144 static inline resource_size_t p1_size(struct pci_dev *dev)
146 return pci_resource_len(dev, 2);
149 static inline resource_size_t p2_base(struct pci_dev *dev)
151 return pci_resource_start(dev, 0);
154 static inline resource_size_t p2_size(struct pci_dev *dev)
156 return pci_resource_len(dev, 0);
159 static int find_cxl_vsec(struct pci_dev *dev)
164 while ((vsec = pci_find_next_ext_capability(dev, vsec, PCI_EXT_CAP_ID_VNDR))) {
165 pci_read_config_word(dev, vsec + 0x4, &val);
166 if (val == CXL_PCI_VSEC_ID)
173 static void dump_cxl_config_space(struct pci_dev *dev)
178 dev_info(&dev->dev, "dump_cxl_config_space\n");
180 pci_read_config_dword(dev, PCI_BASE_ADDRESS_0, &val);
181 dev_info(&dev->dev, "BAR0: %#.8x\n", val);
182 pci_read_config_dword(dev, PCI_BASE_ADDRESS_1, &val);
183 dev_info(&dev->dev, "BAR1: %#.8x\n", val);
184 pci_read_config_dword(dev, PCI_BASE_ADDRESS_2, &val);
185 dev_info(&dev->dev, "BAR2: %#.8x\n", val);
186 pci_read_config_dword(dev, PCI_BASE_ADDRESS_3, &val);
187 dev_info(&dev->dev, "BAR3: %#.8x\n", val);
188 pci_read_config_dword(dev, PCI_BASE_ADDRESS_4, &val);
189 dev_info(&dev->dev, "BAR4: %#.8x\n", val);
190 pci_read_config_dword(dev, PCI_BASE_ADDRESS_5, &val);
191 dev_info(&dev->dev, "BAR5: %#.8x\n", val);
193 dev_info(&dev->dev, "p1 regs: %#llx, len: %#llx\n",
194 p1_base(dev), p1_size(dev));
195 dev_info(&dev->dev, "p2 regs: %#llx, len: %#llx\n",
196 p2_base(dev), p2_size(dev));
197 dev_info(&dev->dev, "BAR 4/5: %#llx, len: %#llx\n",
198 pci_resource_start(dev, 4), pci_resource_len(dev, 4));
200 if (!(vsec = find_cxl_vsec(dev)))
203 #define show_reg(name, what) \
204 dev_info(&dev->dev, "cxl vsec: %30s: %#x\n", name, what)
206 pci_read_config_dword(dev, vsec + 0x0, &val);
207 show_reg("Cap ID", (val >> 0) & 0xffff);
208 show_reg("Cap Ver", (val >> 16) & 0xf);
209 show_reg("Next Cap Ptr", (val >> 20) & 0xfff);
210 pci_read_config_dword(dev, vsec + 0x4, &val);
211 show_reg("VSEC ID", (val >> 0) & 0xffff);
212 show_reg("VSEC Rev", (val >> 16) & 0xf);
213 show_reg("VSEC Length", (val >> 20) & 0xfff);
214 pci_read_config_dword(dev, vsec + 0x8, &val);
215 show_reg("Num AFUs", (val >> 0) & 0xff);
216 show_reg("Status", (val >> 8) & 0xff);
217 show_reg("Mode Control", (val >> 16) & 0xff);
218 show_reg("Reserved", (val >> 24) & 0xff);
219 pci_read_config_dword(dev, vsec + 0xc, &val);
220 show_reg("PSL Rev", (val >> 0) & 0xffff);
221 show_reg("CAIA Ver", (val >> 16) & 0xffff);
222 pci_read_config_dword(dev, vsec + 0x10, &val);
223 show_reg("Base Image Rev", (val >> 0) & 0xffff);
224 show_reg("Reserved", (val >> 16) & 0x0fff);
225 show_reg("Image Control", (val >> 28) & 0x3);
226 show_reg("Reserved", (val >> 30) & 0x1);
227 show_reg("Image Loaded", (val >> 31) & 0x1);
229 pci_read_config_dword(dev, vsec + 0x14, &val);
230 show_reg("Reserved", val);
231 pci_read_config_dword(dev, vsec + 0x18, &val);
232 show_reg("Reserved", val);
233 pci_read_config_dword(dev, vsec + 0x1c, &val);
234 show_reg("Reserved", val);
236 pci_read_config_dword(dev, vsec + 0x20, &val);
237 show_reg("AFU Descriptor Offset", val);
238 pci_read_config_dword(dev, vsec + 0x24, &val);
239 show_reg("AFU Descriptor Size", val);
240 pci_read_config_dword(dev, vsec + 0x28, &val);
241 show_reg("Problem State Offset", val);
242 pci_read_config_dword(dev, vsec + 0x2c, &val);
243 show_reg("Problem State Size", val);
245 pci_read_config_dword(dev, vsec + 0x30, &val);
246 show_reg("Reserved", val);
247 pci_read_config_dword(dev, vsec + 0x34, &val);
248 show_reg("Reserved", val);
249 pci_read_config_dword(dev, vsec + 0x38, &val);
250 show_reg("Reserved", val);
251 pci_read_config_dword(dev, vsec + 0x3c, &val);
252 show_reg("Reserved", val);
254 pci_read_config_dword(dev, vsec + 0x40, &val);
255 show_reg("PSL Programming Port", val);
256 pci_read_config_dword(dev, vsec + 0x44, &val);
257 show_reg("PSL Programming Control", val);
259 pci_read_config_dword(dev, vsec + 0x48, &val);
260 show_reg("Reserved", val);
261 pci_read_config_dword(dev, vsec + 0x4c, &val);
262 show_reg("Reserved", val);
264 pci_read_config_dword(dev, vsec + 0x50, &val);
265 show_reg("Flash Address Register", val);
266 pci_read_config_dword(dev, vsec + 0x54, &val);
267 show_reg("Flash Size Register", val);
268 pci_read_config_dword(dev, vsec + 0x58, &val);
269 show_reg("Flash Status/Control Register", val);
270 pci_read_config_dword(dev, vsec + 0x58, &val);
271 show_reg("Flash Data Port", val);
276 static void dump_afu_descriptor(struct cxl_afu *afu)
278 u64 val, afu_cr_num, afu_cr_off, afu_cr_len;
281 #define show_reg(name, what) \
282 dev_info(&afu->dev, "afu desc: %30s: %#llx\n", name, what)
284 val = AFUD_READ_INFO(afu);
285 show_reg("num_ints_per_process", AFUD_NUM_INTS_PER_PROC(val));
286 show_reg("num_of_processes", AFUD_NUM_PROCS(val));
287 show_reg("num_of_afu_CRs", AFUD_NUM_CRS(val));
288 show_reg("req_prog_mode", val & 0xffffULL);
289 afu_cr_num = AFUD_NUM_CRS(val);
291 val = AFUD_READ(afu, 0x8);
292 show_reg("Reserved", val);
293 val = AFUD_READ(afu, 0x10);
294 show_reg("Reserved", val);
295 val = AFUD_READ(afu, 0x18);
296 show_reg("Reserved", val);
298 val = AFUD_READ_CR(afu);
299 show_reg("Reserved", (val >> (63-7)) & 0xff);
300 show_reg("AFU_CR_len", AFUD_CR_LEN(val));
301 afu_cr_len = AFUD_CR_LEN(val) * 256;
303 val = AFUD_READ_CR_OFF(afu);
305 show_reg("AFU_CR_offset", val);
307 val = AFUD_READ_PPPSA(afu);
308 show_reg("PerProcessPSA_control", (val >> (63-7)) & 0xff);
309 show_reg("PerProcessPSA Length", AFUD_PPPSA_LEN(val));
311 val = AFUD_READ_PPPSA_OFF(afu);
312 show_reg("PerProcessPSA_offset", val);
314 val = AFUD_READ_EB(afu);
315 show_reg("Reserved", (val >> (63-7)) & 0xff);
316 show_reg("AFU_EB_len", AFUD_EB_LEN(val));
318 val = AFUD_READ_EB_OFF(afu);
319 show_reg("AFU_EB_offset", val);
321 for (i = 0; i < afu_cr_num; i++) {
322 val = AFUD_READ_LE(afu, afu_cr_off + i * afu_cr_len);
323 show_reg("CR Vendor", val & 0xffff);
324 show_reg("CR Device", (val >> 16) & 0xffff);
329 #define P8_CAPP_UNIT0_ID 0xBA
330 #define P8_CAPP_UNIT1_ID 0XBE
331 #define P9_CAPP_UNIT0_ID 0xC0
332 #define P9_CAPP_UNIT1_ID 0xE0
334 static int get_phb_index(struct device_node *np, u32 *phb_index)
336 if (of_property_read_u32(np, "ibm,phb-index", phb_index))
341 static u64 get_capp_unit_id(struct device_node *np, u32 phb_index)
345 * - For chips other than POWER8NVL, we only have CAPP 0,
346 * irrespective of which PHB is used.
347 * - For POWER8NVL, assume CAPP 0 is attached to PHB0 and
348 * CAPP 1 is attached to PHB1.
350 if (cxl_is_power8()) {
351 if (!pvr_version_is(PVR_POWER8NVL))
352 return P8_CAPP_UNIT0_ID;
355 return P8_CAPP_UNIT0_ID;
358 return P8_CAPP_UNIT1_ID;
363 * PEC0 (PHB0). Capp ID = CAPP0 (0b1100_0000)
364 * PEC1 (PHB1 - PHB2). No capi mode
365 * PEC2 (PHB3 - PHB4 - PHB5): Capi mode on PHB3 only. Capp ID = CAPP1 (0b1110_0000)
367 if (cxl_is_power9()) {
369 return P9_CAPP_UNIT0_ID;
372 return P9_CAPP_UNIT1_ID;
378 int cxl_calc_capp_routing(struct pci_dev *dev, u64 *chipid,
379 u32 *phb_index, u64 *capp_unit_id)
382 struct device_node *np;
385 if (!(np = pnv_pci_get_phb_node(dev)))
388 while (np && !(prop = of_get_property(np, "ibm,chip-id", NULL)))
389 np = of_get_next_parent(np);
393 *chipid = be32_to_cpup(prop);
395 rc = get_phb_index(np, phb_index);
397 pr_err("cxl: invalid phb index\n");
402 *capp_unit_id = get_capp_unit_id(np, *phb_index);
404 if (!*capp_unit_id) {
405 pr_err("cxl: invalid capp unit id\n");
412 int cxl_get_xsl9_dsnctl(u64 capp_unit_id, u64 *reg)
417 * CAPI Identifier bits [0:7]
418 * bit 61:60 MSI bits --> 0
419 * bit 59 TVT selector --> 0
423 * Tell XSL where to route data to.
424 * The field chipid should match the PHB CAPI_CMPM register
426 xsl_dsnctl = ((u64)0x2 << (63-7)); /* Bit 57 */
427 xsl_dsnctl |= (capp_unit_id << (63-15));
429 /* nMMU_ID Defaults to: b’000001001’*/
430 xsl_dsnctl |= ((u64)0x09 << (63-28));
432 if (!(cxl_is_power9_dd1())) {
434 * Used to identify CAPI packets which should be sorted into
435 * the Non-Blocking queues by the PHB. This field should match
436 * the PHB PBL_NBW_CMPM register
437 * nbwind=0x03, bits [57:58], must include capi indicator.
438 * Not supported on P9 DD1.
440 xsl_dsnctl |= ((u64)0x03 << (63-47));
443 * Upper 16b address bits of ASB_Notify messages sent to the
444 * system. Need to match the PHB’s ASN Compare/Mask Register.
445 * Not supported on P9 DD1.
447 xsl_dsnctl |= ((u64)0x04 << (63-55));
454 static int init_implementation_adapter_regs_psl9(struct cxl *adapter,
457 u64 xsl_dsnctl, psl_fircntl;
464 rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
468 rc = cxl_get_xsl9_dsnctl(capp_unit_id, &xsl_dsnctl);
472 cxl_p1_write(adapter, CXL_XSL9_DSNCTL, xsl_dsnctl);
474 /* Set fir_cntl to recommended value for production env */
475 psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
476 psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
477 psl_fircntl |= 0x1ULL; /* ce_thresh */
478 cxl_p1_write(adapter, CXL_PSL9_FIR_CNTL, psl_fircntl);
480 /* vccredits=0x1 pcklat=0x4 */
481 cxl_p1_write(adapter, CXL_PSL9_DSNDCTL, 0x0000000000001810ULL);
484 * For debugging with trace arrays.
485 * Configure RX trace 0 segmented mode.
486 * Configure CT trace 0 segmented mode.
487 * Configure LA0 trace 0 segmented mode.
488 * Configure LA1 trace 0 segmented mode.
490 cxl_p1_write(adapter, CXL_PSL9_TRACECFG, 0x8040800080000000ULL);
491 cxl_p1_write(adapter, CXL_PSL9_TRACECFG, 0x8040800080000003ULL);
492 cxl_p1_write(adapter, CXL_PSL9_TRACECFG, 0x8040800080000005ULL);
493 cxl_p1_write(adapter, CXL_PSL9_TRACECFG, 0x8040800080000006ULL);
496 * A response to an ASB_Notify request is returned by the
497 * system as an MMIO write to the address defined in
498 * the PSL_TNR_ADDR register
503 cxl_p1_write(adapter, CXL_PSL9_DEBUG, 0x8000000000000000ULL);
505 /* allocate the apc machines */
506 cxl_p1_write(adapter, CXL_PSL9_APCDEDTYPE, 0x40000003FFFF0000ULL);
508 /* Disable vc dd1 fix */
509 if (cxl_is_power9_dd1())
510 cxl_p1_write(adapter, CXL_PSL9_GP_CT, 0x0400000000000001ULL);
513 * Check if PSL has data-cache. We need to flush adapter datacache
514 * when as its about to be removed.
516 psl_debug = cxl_p1_read(adapter, CXL_PSL9_DEBUG);
517 if (psl_debug & CXL_PSL_DEBUG_CDC) {
518 dev_dbg(&dev->dev, "No data-cache present\n");
519 adapter->native->no_data_cache = true;
525 static int init_implementation_adapter_regs_psl8(struct cxl *adapter, struct pci_dev *dev)
527 u64 psl_dsnctl, psl_fircntl;
533 rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
537 psl_dsnctl = 0x0000900000000000ULL; /* pteupd ttype, scdone */
538 psl_dsnctl |= (0x2ULL << (63-38)); /* MMIO hang pulse: 256 us */
539 /* Tell PSL where to route data to */
540 psl_dsnctl |= (chipid << (63-5));
541 psl_dsnctl |= (capp_unit_id << (63-13));
543 cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
544 cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
545 /* snoop write mask */
546 cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
547 /* set fir_cntl to recommended value for production env */
548 psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
549 psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
550 psl_fircntl |= 0x1ULL; /* ce_thresh */
551 cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, psl_fircntl);
552 /* for debugging with trace arrays */
553 cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
558 static int init_implementation_adapter_regs_xsl(struct cxl *adapter, struct pci_dev *dev)
566 rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
570 /* Tell XSL where to route data to */
571 xsl_dsnctl = 0x0000600000000000ULL | (chipid << (63-5));
572 xsl_dsnctl |= (capp_unit_id << (63-13));
573 cxl_p1_write(adapter, CXL_XSL_DSNCTL, xsl_dsnctl);
579 #define TBSYNC_CAL(n) (((u64)n & 0x7) << (63-3))
580 #define TBSYNC_CNT(n) (((u64)n & 0x7) << (63-6))
581 /* For the PSL this is a multiple for 0 < n <= 7: */
582 #define PSL_2048_250MHZ_CYCLES 1
584 static void write_timebase_ctrl_psl9(struct cxl *adapter)
586 cxl_p1_write(adapter, CXL_PSL9_TB_CTLSTAT,
587 TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
590 static void write_timebase_ctrl_psl8(struct cxl *adapter)
592 cxl_p1_write(adapter, CXL_PSL_TB_CTLSTAT,
593 TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
597 #define TBSYNC_ENA (1ULL << 63)
598 /* For the XSL this is 2**n * 2000 clocks for 0 < n <= 6: */
599 #define XSL_2000_CLOCKS 1
600 #define XSL_4000_CLOCKS 2
601 #define XSL_8000_CLOCKS 3
603 static void write_timebase_ctrl_xsl(struct cxl *adapter)
605 cxl_p1_write(adapter, CXL_XSL_TB_CTLSTAT,
608 TBSYNC_CNT(XSL_4000_CLOCKS));
611 static u64 timebase_read_psl9(struct cxl *adapter)
613 return cxl_p1_read(adapter, CXL_PSL9_Timebase);
616 static u64 timebase_read_psl8(struct cxl *adapter)
618 return cxl_p1_read(adapter, CXL_PSL_Timebase);
621 static u64 timebase_read_xsl(struct cxl *adapter)
623 return cxl_p1_read(adapter, CXL_XSL_Timebase);
626 static void cxl_setup_psl_timebase(struct cxl *adapter, struct pci_dev *dev)
630 unsigned int retry = 0;
631 struct device_node *np;
633 adapter->psl_timebase_synced = false;
635 if (!(np = pnv_pci_get_phb_node(dev)))
638 /* Do not fail when CAPP timebase sync is not supported by OPAL */
640 if (! of_get_property(np, "ibm,capp-timebase-sync", NULL)) {
642 dev_info(&dev->dev, "PSL timebase inactive: OPAL support missing\n");
648 * Setup PSL Timebase Control and Status register
649 * with the recommended Timebase Sync Count value
651 adapter->native->sl_ops->write_timebase_ctrl(adapter);
653 /* Enable PSL Timebase */
654 cxl_p1_write(adapter, CXL_PSL_Control, 0x0000000000000000);
655 cxl_p1_write(adapter, CXL_PSL_Control, CXL_PSL_Control_tb);
657 /* Wait until CORE TB and PSL TB difference <= 16usecs */
661 dev_info(&dev->dev, "PSL timebase can't synchronize\n");
664 psl_tb = adapter->native->sl_ops->timebase_read(adapter);
665 delta = mftb() - psl_tb;
668 } while (tb_to_ns(delta) > 16000);
670 adapter->psl_timebase_synced = true;
674 static int init_implementation_afu_regs_psl9(struct cxl_afu *afu)
679 static int init_implementation_afu_regs_psl8(struct cxl_afu *afu)
681 /* read/write masks for this slice */
682 cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
683 /* APC read/write masks for this slice */
684 cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
685 /* for debugging with trace arrays */
686 cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
687 cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
692 int cxl_pci_setup_irq(struct cxl *adapter, unsigned int hwirq,
695 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
697 return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
700 int cxl_update_image_control(struct cxl *adapter)
702 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
707 if (!(vsec = find_cxl_vsec(dev))) {
708 dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
712 if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
713 dev_err(&dev->dev, "failed to read image state: %i\n", rc);
717 if (adapter->perst_loads_image)
718 image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
720 image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
722 if (adapter->perst_select_user)
723 image_state |= CXL_VSEC_PERST_SELECT_USER;
725 image_state &= ~CXL_VSEC_PERST_SELECT_USER;
727 if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
728 dev_err(&dev->dev, "failed to update image control: %i\n", rc);
735 int cxl_pci_alloc_one_irq(struct cxl *adapter)
737 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
739 return pnv_cxl_alloc_hwirqs(dev, 1);
742 void cxl_pci_release_one_irq(struct cxl *adapter, int hwirq)
744 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
746 return pnv_cxl_release_hwirqs(dev, hwirq, 1);
749 int cxl_pci_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
750 struct cxl *adapter, unsigned int num)
752 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
754 return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
757 void cxl_pci_release_irq_ranges(struct cxl_irq_ranges *irqs,
760 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
762 pnv_cxl_release_hwirq_ranges(irqs, dev);
765 static int setup_cxl_bars(struct pci_dev *dev)
767 /* Safety check in case we get backported to < 3.17 without M64 */
768 if ((p1_base(dev) < 0x100000000ULL) ||
769 (p2_base(dev) < 0x100000000ULL)) {
770 dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
775 * BAR 4/5 has a special meaning for CXL and must be programmed with a
776 * special value corresponding to the CXL protocol address range.
777 * For POWER 8/9 that means bits 48:49 must be set to 10
779 pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
780 pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
785 #ifdef CONFIG_CXL_BIMODAL
787 struct cxl_switch_work {
789 struct work_struct work;
794 static void switch_card_to_cxl(struct work_struct *work)
796 struct cxl_switch_work *switch_work =
797 container_of(work, struct cxl_switch_work, work);
798 struct pci_dev *dev = switch_work->dev;
799 struct pci_bus *bus = dev->bus;
800 struct pci_controller *hose = pci_bus_to_host(bus);
801 struct pci_dev *bridge;
802 struct pnv_php_slot *php_slot;
807 dev_info(&bus->dev, "cxl: Preparing for mode switch...\n");
808 bridge = list_first_entry_or_null(&hose->bus->devices, struct pci_dev,
811 dev_WARN(&bus->dev, "cxl: Couldn't find root port!\n");
815 php_slot = pnv_php_find_slot(pci_device_to_OF_node(bridge));
817 dev_err(&bus->dev, "cxl: Failed to find slot hotplug "
818 "information. You may need to upgrade "
819 "skiboot. Aborting.\n");
823 rc = CXL_READ_VSEC_MODE_CONTROL(dev, switch_work->vsec, &val);
825 dev_err(&bus->dev, "cxl: Failed to read CAPI mode control: %i\n", rc);
830 /* Release the reference obtained in cxl_check_and_switch_mode() */
833 dev_dbg(&bus->dev, "cxl: Removing PCI devices from kernel\n");
834 pci_lock_rescan_remove();
835 pci_hp_remove_devices(bridge->subordinate);
836 pci_unlock_rescan_remove();
838 /* Switch the CXL protocol on the card */
839 if (switch_work->mode == CXL_BIMODE_CXL) {
840 dev_info(&bus->dev, "cxl: Switching card to CXL mode\n");
841 val &= ~CXL_VSEC_PROTOCOL_MASK;
842 val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
843 rc = pnv_cxl_enable_phb_kernel_api(hose, true);
845 dev_err(&bus->dev, "cxl: Failed to enable kernel API"
846 " on real PHB, aborting\n");
850 dev_WARN(&bus->dev, "cxl: Switching card to PCI mode not supported!\n");
854 rc = CXL_WRITE_VSEC_MODE_CONTROL_BUS(bus, devfn, switch_work->vsec, val);
856 dev_err(&bus->dev, "cxl: Failed to configure CXL protocol: %i\n", rc);
861 * The CAIA spec (v1.1, Section 10.6 Bi-modal Device Support) states
862 * we must wait 100ms after this mode switch before touching PCIe config
868 * Hot reset to cause the card to come back in cxl mode. A
869 * OPAL_RESET_PCI_LINK would be sufficient, but currently lacks support
870 * in skiboot, so we use a hot reset instead.
872 * We call pci_set_pcie_reset_state() on the bridge, as a CAPI card is
873 * guaranteed to sit directly under the root port, and setting the reset
874 * state on a device directly under the root port is equivalent to doing
875 * it on the root port iself.
877 dev_info(&bus->dev, "cxl: Configuration write complete, resetting card\n");
878 pci_set_pcie_reset_state(bridge, pcie_hot_reset);
879 pci_set_pcie_reset_state(bridge, pcie_deassert_reset);
881 dev_dbg(&bus->dev, "cxl: Offlining slot\n");
882 rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_OFFLINE);
884 dev_err(&bus->dev, "cxl: OPAL offlining call failed: %i\n", rc);
888 dev_dbg(&bus->dev, "cxl: Onlining and probing slot\n");
889 rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_ONLINE);
891 dev_err(&bus->dev, "cxl: OPAL onlining call failed: %i\n", rc);
895 pci_lock_rescan_remove();
896 pci_hp_add_devices(bridge->subordinate);
897 pci_unlock_rescan_remove();
899 dev_info(&bus->dev, "cxl: CAPI mode switch completed\n");
904 /* Release the reference obtained in cxl_check_and_switch_mode() */
910 int cxl_check_and_switch_mode(struct pci_dev *dev, int mode, int vsec)
912 struct cxl_switch_work *work;
916 if (!cpu_has_feature(CPU_FTR_HVMODE))
920 vsec = find_cxl_vsec(dev);
922 dev_info(&dev->dev, "CXL VSEC not found\n");
927 rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
929 dev_err(&dev->dev, "Failed to read current mode control: %i", rc);
933 if (mode == CXL_BIMODE_PCI) {
934 if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
935 dev_info(&dev->dev, "Card is already in PCI mode\n");
939 * TODO: Before it's safe to switch the card back to PCI mode
940 * we need to disable the CAPP and make sure any cachelines the
941 * card holds have been flushed out. Needs skiboot support.
943 dev_WARN(&dev->dev, "CXL mode switch to PCI unsupported!\n");
947 if (val & CXL_VSEC_PROTOCOL_ENABLE) {
948 dev_info(&dev->dev, "Card is already in CXL mode\n");
952 dev_info(&dev->dev, "Card is in PCI mode, scheduling kernel thread "
953 "to switch to CXL mode\n");
955 work = kmalloc(sizeof(struct cxl_switch_work), GFP_KERNEL);
963 INIT_WORK(&work->work, switch_card_to_cxl);
965 schedule_work(&work->work);
968 * We return a failure now to abort the driver init. Once the
969 * link has been cycled and the card is in cxl mode we will
970 * come back (possibly using the generic cxl driver), but
971 * return success as the card should then be in cxl mode.
973 * TODO: What if the card comes back in PCI mode even after
974 * the switch? Don't want to spin endlessly.
978 EXPORT_SYMBOL_GPL(cxl_check_and_switch_mode);
980 #endif /* CONFIG_CXL_BIMODAL */
982 static int setup_cxl_protocol_area(struct pci_dev *dev)
986 int vsec = find_cxl_vsec(dev);
989 dev_info(&dev->dev, "CXL VSEC not found\n");
993 rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
995 dev_err(&dev->dev, "Failed to read current mode control: %i\n", rc);
999 if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
1000 dev_err(&dev->dev, "Card not in CAPI mode!\n");
1004 if ((val & CXL_VSEC_PROTOCOL_MASK) != CXL_VSEC_PROTOCOL_256TB) {
1005 val &= ~CXL_VSEC_PROTOCOL_MASK;
1006 val |= CXL_VSEC_PROTOCOL_256TB;
1007 rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val);
1009 dev_err(&dev->dev, "Failed to set CXL protocol area: %i\n", rc);
1017 static int pci_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1019 u64 p1n_base, p2n_base, afu_desc;
1020 const u64 p1n_size = 0x100;
1021 const u64 p2n_size = 0x1000;
1023 p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
1024 p2n_base = p2_base(dev) + (afu->slice * p2n_size);
1025 afu->psn_phys = p2_base(dev) + (adapter->native->ps_off + (afu->slice * adapter->ps_size));
1026 afu_desc = p2_base(dev) + adapter->native->afu_desc_off + (afu->slice * adapter->native->afu_desc_size);
1028 if (!(afu->native->p1n_mmio = ioremap(p1n_base, p1n_size)))
1030 if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
1033 if (!(afu->native->afu_desc_mmio = ioremap(afu_desc, adapter->native->afu_desc_size)))
1039 iounmap(afu->p2n_mmio);
1041 iounmap(afu->native->p1n_mmio);
1043 dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
1047 static void pci_unmap_slice_regs(struct cxl_afu *afu)
1049 if (afu->p2n_mmio) {
1050 iounmap(afu->p2n_mmio);
1051 afu->p2n_mmio = NULL;
1053 if (afu->native->p1n_mmio) {
1054 iounmap(afu->native->p1n_mmio);
1055 afu->native->p1n_mmio = NULL;
1057 if (afu->native->afu_desc_mmio) {
1058 iounmap(afu->native->afu_desc_mmio);
1059 afu->native->afu_desc_mmio = NULL;
1063 void cxl_pci_release_afu(struct device *dev)
1065 struct cxl_afu *afu = to_cxl_afu(dev);
1067 pr_devel("%s\n", __func__);
1069 idr_destroy(&afu->contexts_idr);
1070 cxl_release_spa(afu);
1076 /* Expects AFU struct to have recently been zeroed out */
1077 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
1081 val = AFUD_READ_INFO(afu);
1082 afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
1083 afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
1084 afu->crs_num = AFUD_NUM_CRS(val);
1086 if (AFUD_AFU_DIRECTED(val))
1087 afu->modes_supported |= CXL_MODE_DIRECTED;
1088 if (AFUD_DEDICATED_PROCESS(val))
1089 afu->modes_supported |= CXL_MODE_DEDICATED;
1090 if (AFUD_TIME_SLICED(val))
1091 afu->modes_supported |= CXL_MODE_TIME_SLICED;
1093 val = AFUD_READ_PPPSA(afu);
1094 afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
1095 afu->psa = AFUD_PPPSA_PSA(val);
1096 if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
1097 afu->native->pp_offset = AFUD_READ_PPPSA_OFF(afu);
1099 val = AFUD_READ_CR(afu);
1100 afu->crs_len = AFUD_CR_LEN(val) * 256;
1101 afu->crs_offset = AFUD_READ_CR_OFF(afu);
1104 /* eb_len is in multiple of 4K */
1105 afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
1106 afu->eb_offset = AFUD_READ_EB_OFF(afu);
1108 /* eb_off is 4K aligned so lower 12 bits are always zero */
1109 if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
1111 "Invalid AFU error buffer offset %Lx\n",
1114 "Ignoring AFU error buffer in the descriptor\n");
1115 /* indicate that no afu buffer exists */
1122 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
1127 if (afu->psa && afu->adapter->ps_size <
1128 (afu->native->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
1129 dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
1133 if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
1134 dev_warn(&afu->dev, "AFU uses pp_size(%#016llx) < PAGE_SIZE per-process PSA!\n", afu->pp_size);
1136 for (i = 0; i < afu->crs_num; i++) {
1137 rc = cxl_ops->afu_cr_read32(afu, i, 0, &val);
1138 if (rc || val == 0) {
1139 dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
1144 if ((afu->modes_supported & ~CXL_MODE_DEDICATED) && afu->max_procs_virtualised == 0) {
1146 * We could also check this for the dedicated process model
1147 * since the architecture indicates it should be set to 1, but
1148 * in that case we ignore the value and I'd rather not risk
1149 * breaking any existing dedicated process AFUs that left it as
1150 * 0 (not that I'm aware of any). It is clearly an error for an
1151 * AFU directed AFU to set this to 0, and would have previously
1152 * triggered a bug resulting in the maximum not being enforced
1153 * at all since idr_alloc treats 0 as no maximum.
1155 dev_err(&afu->dev, "AFU does not support any processes\n");
1162 static int sanitise_afu_regs_psl9(struct cxl_afu *afu)
1167 * Clear out any regs that contain either an IVTE or address or may be
1168 * waiting on an acknowledgment to try to be a bit safer as we bring
1171 reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
1172 if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
1173 dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
1174 if (cxl_ops->afu_reset(afu))
1176 if (cxl_afu_disable(afu))
1178 if (cxl_psl_purge(afu))
1181 cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
1182 cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1183 reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1185 dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1186 if (reg & CXL_PSL9_DSISR_An_TF)
1187 cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1189 cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1191 if (afu->adapter->native->sl_ops->register_serr_irq) {
1192 reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1194 if (reg & ~0x000000007fffffff)
1195 dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1196 cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1199 reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1201 dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1202 cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1208 static int sanitise_afu_regs_psl8(struct cxl_afu *afu)
1213 * Clear out any regs that contain either an IVTE or address or may be
1214 * waiting on an acknowledgement to try to be a bit safer as we bring
1217 reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
1218 if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
1219 dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
1220 if (cxl_ops->afu_reset(afu))
1222 if (cxl_afu_disable(afu))
1224 if (cxl_psl_purge(afu))
1227 cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
1228 cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
1229 cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
1230 cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1231 cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
1232 cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
1233 cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
1234 cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
1235 cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
1236 cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
1237 cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
1238 reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1240 dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1241 if (reg & CXL_PSL_DSISR_TRANS)
1242 cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1244 cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1246 if (afu->adapter->native->sl_ops->register_serr_irq) {
1247 reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1250 dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1251 cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1254 reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1256 dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1257 cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1263 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
1266 * Called from sysfs and reads the afu error info buffer. The h/w only supports
1267 * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
1268 * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
1270 ssize_t cxl_pci_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
1271 loff_t off, size_t count)
1273 loff_t aligned_start, aligned_end;
1274 size_t aligned_length;
1276 const void __iomem *ebuf = afu->native->afu_desc_mmio + afu->eb_offset;
1278 if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
1281 /* calculate aligned read window */
1282 count = min((size_t)(afu->eb_len - off), count);
1283 aligned_start = round_down(off, 8);
1284 aligned_end = round_up(off + count, 8);
1285 aligned_length = aligned_end - aligned_start;
1287 /* max we can copy in one read is PAGE_SIZE */
1288 if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
1289 aligned_length = ERR_BUFF_MAX_COPY_SIZE;
1290 count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
1293 /* use bounce buffer for copy */
1294 tbuf = (void *)__get_free_page(GFP_KERNEL);
1298 /* perform aligned read from the mmio region */
1299 memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
1300 memcpy(buf, tbuf + (off & 0x7), count);
1302 free_page((unsigned long)tbuf);
1307 static int pci_configure_afu(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1311 if ((rc = pci_map_slice_regs(afu, adapter, dev)))
1314 if (adapter->native->sl_ops->sanitise_afu_regs) {
1315 rc = adapter->native->sl_ops->sanitise_afu_regs(afu);
1320 /* We need to reset the AFU before we can read the AFU descriptor */
1321 if ((rc = cxl_ops->afu_reset(afu)))
1325 dump_afu_descriptor(afu);
1327 if ((rc = cxl_read_afu_descriptor(afu)))
1330 if ((rc = cxl_afu_descriptor_looks_ok(afu)))
1333 if (adapter->native->sl_ops->afu_regs_init)
1334 if ((rc = adapter->native->sl_ops->afu_regs_init(afu)))
1337 if (adapter->native->sl_ops->register_serr_irq)
1338 if ((rc = adapter->native->sl_ops->register_serr_irq(afu)))
1341 if ((rc = cxl_native_register_psl_irq(afu)))
1344 atomic_set(&afu->configured_state, 0);
1348 if (adapter->native->sl_ops->release_serr_irq)
1349 adapter->native->sl_ops->release_serr_irq(afu);
1351 pci_unmap_slice_regs(afu);
1355 static void pci_deconfigure_afu(struct cxl_afu *afu)
1358 * It's okay to deconfigure when AFU is already locked, otherwise wait
1359 * until there are no readers
1361 if (atomic_read(&afu->configured_state) != -1) {
1362 while (atomic_cmpxchg(&afu->configured_state, 0, -1) != -1)
1365 cxl_native_release_psl_irq(afu);
1366 if (afu->adapter->native->sl_ops->release_serr_irq)
1367 afu->adapter->native->sl_ops->release_serr_irq(afu);
1368 pci_unmap_slice_regs(afu);
1371 static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
1373 struct cxl_afu *afu;
1376 afu = cxl_alloc_afu(adapter, slice);
1380 afu->native = kzalloc(sizeof(struct cxl_afu_native), GFP_KERNEL);
1384 mutex_init(&afu->native->spa_mutex);
1386 rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice);
1388 goto err_free_native;
1390 rc = pci_configure_afu(afu, adapter, dev);
1392 goto err_free_native;
1394 /* Don't care if this fails */
1395 cxl_debugfs_afu_add(afu);
1398 * After we call this function we must not free the afu directly, even
1399 * if it returns an error!
1401 if ((rc = cxl_register_afu(afu)))
1404 if ((rc = cxl_sysfs_afu_add(afu)))
1407 adapter->afu[afu->slice] = afu;
1409 if ((rc = cxl_pci_vphb_add(afu)))
1410 dev_info(&afu->dev, "Can't register vPHB\n");
1415 device_del(&afu->dev);
1417 pci_deconfigure_afu(afu);
1418 cxl_debugfs_afu_remove(afu);
1419 put_device(&afu->dev);
1430 static void cxl_pci_remove_afu(struct cxl_afu *afu)
1432 pr_devel("%s\n", __func__);
1437 cxl_pci_vphb_remove(afu);
1438 cxl_sysfs_afu_remove(afu);
1439 cxl_debugfs_afu_remove(afu);
1441 spin_lock(&afu->adapter->afu_list_lock);
1442 afu->adapter->afu[afu->slice] = NULL;
1443 spin_unlock(&afu->adapter->afu_list_lock);
1445 cxl_context_detach_all(afu);
1446 cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1448 pci_deconfigure_afu(afu);
1449 device_unregister(&afu->dev);
1452 int cxl_pci_reset(struct cxl *adapter)
1454 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1457 if (adapter->perst_same_image) {
1459 "cxl: refusing to reset/reflash when perst_reloads_same_image is set.\n");
1463 dev_info(&dev->dev, "CXL reset\n");
1466 * The adapter is about to be reset, so ignore errors.
1468 cxl_data_cache_flush(adapter);
1470 /* pcie_warm_reset requests a fundamental pci reset which includes a
1471 * PERST assert/deassert. PERST triggers a loading of the image
1472 * if "user" or "factory" is selected in sysfs */
1473 if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
1474 dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
1481 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
1483 if (pci_request_region(dev, 2, "priv 2 regs"))
1485 if (pci_request_region(dev, 0, "priv 1 regs"))
1488 pr_devel("cxl_map_adapter_regs: p1: %#016llx %#llx, p2: %#016llx %#llx",
1489 p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
1491 if (!(adapter->native->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
1494 if (!(adapter->native->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
1500 iounmap(adapter->native->p1_mmio);
1501 adapter->native->p1_mmio = NULL;
1503 pci_release_region(dev, 0);
1505 pci_release_region(dev, 2);
1510 static void cxl_unmap_adapter_regs(struct cxl *adapter)
1512 if (adapter->native->p1_mmio) {
1513 iounmap(adapter->native->p1_mmio);
1514 adapter->native->p1_mmio = NULL;
1515 pci_release_region(to_pci_dev(adapter->dev.parent), 2);
1517 if (adapter->native->p2_mmio) {
1518 iounmap(adapter->native->p2_mmio);
1519 adapter->native->p2_mmio = NULL;
1520 pci_release_region(to_pci_dev(adapter->dev.parent), 0);
1524 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
1527 u32 afu_desc_off, afu_desc_size;
1528 u32 ps_off, ps_size;
1532 if (!(vsec = find_cxl_vsec(dev))) {
1533 dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
1537 CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
1538 if (vseclen < CXL_VSEC_MIN_SIZE) {
1539 dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
1543 CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
1544 CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
1545 CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
1546 CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
1547 CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
1548 CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
1549 adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1550 adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1551 adapter->perst_loads_image = !!(image_state & CXL_VSEC_PERST_LOADS_IMAGE);
1553 CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
1554 CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
1555 CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
1556 CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
1557 CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
1559 /* Convert everything to bytes, because there is NO WAY I'd look at the
1560 * code a month later and forget what units these are in ;-) */
1561 adapter->native->ps_off = ps_off * 64 * 1024;
1562 adapter->ps_size = ps_size * 64 * 1024;
1563 adapter->native->afu_desc_off = afu_desc_off * 64 * 1024;
1564 adapter->native->afu_desc_size = afu_desc_size * 64 * 1024;
1566 /* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
1567 adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
1573 * Workaround a PCIe Host Bridge defect on some cards, that can cause
1574 * malformed Transaction Layer Packet (TLP) errors to be erroneously
1575 * reported. Mask this error in the Uncorrectable Error Mask Register.
1577 * The upper nibble of the PSL revision is used to distinguish between
1578 * different cards. The affected ones have it set to 0.
1580 static void cxl_fixup_malformed_tlp(struct cxl *adapter, struct pci_dev *dev)
1585 if (adapter->psl_rev & 0xf000)
1587 if (!(aer = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR)))
1589 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &data);
1590 if (data & PCI_ERR_UNC_MALF_TLP)
1591 if (data & PCI_ERR_UNC_INTN)
1593 data |= PCI_ERR_UNC_MALF_TLP;
1594 data |= PCI_ERR_UNC_INTN;
1595 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, data);
1598 static bool cxl_compatible_caia_version(struct cxl *adapter)
1600 if (cxl_is_power8() && (adapter->caia_major == 1))
1603 if (cxl_is_power9() && (adapter->caia_major == 2))
1609 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
1611 if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
1614 if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
1615 dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
1619 if (!cxl_compatible_caia_version(adapter)) {
1620 dev_info(&dev->dev, "Ignoring card. PSL type is not supported (caia version: %d)\n",
1621 adapter->caia_major);
1625 if (!adapter->slices) {
1626 /* Once we support dynamic reprogramming we can use the card if
1627 * it supports loadable AFUs */
1628 dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
1632 if (!adapter->native->afu_desc_off || !adapter->native->afu_desc_size) {
1633 dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
1637 if (adapter->ps_size > p2_size(dev) - adapter->native->ps_off) {
1638 dev_err(&dev->dev, "ABORTING: Problem state size larger than "
1639 "available in BAR2: 0x%llx > 0x%llx\n",
1640 adapter->ps_size, p2_size(dev) - adapter->native->ps_off);
1647 ssize_t cxl_pci_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1649 return pci_read_vpd(to_pci_dev(adapter->dev.parent), 0, len, buf);
1652 static void cxl_release_adapter(struct device *dev)
1654 struct cxl *adapter = to_cxl_adapter(dev);
1656 pr_devel("cxl_release_adapter\n");
1658 cxl_remove_adapter_nr(adapter);
1660 kfree(adapter->native);
1664 #define CXL_PSL_ErrIVTE_tberror (0x1ull << (63-31))
1666 static int sanitise_adapter_regs(struct cxl *adapter)
1670 /* Clear PSL tberror bit by writing 1 to it */
1671 cxl_p1_write(adapter, CXL_PSL_ErrIVTE, CXL_PSL_ErrIVTE_tberror);
1673 if (adapter->native->sl_ops->invalidate_all) {
1674 /* do not invalidate ERAT entries when not reloading on PERST */
1675 if (cxl_is_power9() && (adapter->perst_loads_image))
1677 rc = adapter->native->sl_ops->invalidate_all(adapter);
1683 /* This should contain *only* operations that can safely be done in
1684 * both creation and recovery.
1686 static int cxl_configure_adapter(struct cxl *adapter, struct pci_dev *dev)
1690 adapter->dev.parent = &dev->dev;
1691 adapter->dev.release = cxl_release_adapter;
1692 pci_set_drvdata(dev, adapter);
1694 rc = pci_enable_device(dev);
1696 dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1700 if ((rc = cxl_read_vsec(adapter, dev)))
1703 if ((rc = cxl_vsec_looks_ok(adapter, dev)))
1706 cxl_fixup_malformed_tlp(adapter, dev);
1708 if ((rc = setup_cxl_bars(dev)))
1711 if ((rc = setup_cxl_protocol_area(dev)))
1714 if ((rc = cxl_update_image_control(adapter)))
1717 if ((rc = cxl_map_adapter_regs(adapter, dev)))
1720 if ((rc = sanitise_adapter_regs(adapter)))
1723 if ((rc = adapter->native->sl_ops->adapter_regs_init(adapter, dev)))
1726 /* Required for devices using CAPP DMA mode, harmless for others */
1727 pci_set_master(dev);
1729 if ((rc = pnv_phb_to_cxl_mode(dev, adapter->native->sl_ops->capi_mode)))
1732 /* If recovery happened, the last step is to turn on snooping.
1733 * In the non-recovery case this has no effect */
1734 if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON)))
1737 /* Ignore error, adapter init is not dependant on timebase sync */
1738 cxl_setup_psl_timebase(adapter, dev);
1740 if ((rc = cxl_native_register_psl_err_irq(adapter)))
1746 cxl_unmap_adapter_regs(adapter);
1751 static void cxl_deconfigure_adapter(struct cxl *adapter)
1753 struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1755 cxl_native_release_psl_err_irq(adapter);
1756 cxl_unmap_adapter_regs(adapter);
1758 pci_disable_device(pdev);
1761 static const struct cxl_service_layer_ops psl9_ops = {
1762 .adapter_regs_init = init_implementation_adapter_regs_psl9,
1763 .invalidate_all = cxl_invalidate_all_psl9,
1764 .afu_regs_init = init_implementation_afu_regs_psl9,
1765 .sanitise_afu_regs = sanitise_afu_regs_psl9,
1766 .register_serr_irq = cxl_native_register_serr_irq,
1767 .release_serr_irq = cxl_native_release_serr_irq,
1768 .handle_interrupt = cxl_irq_psl9,
1769 .fail_irq = cxl_fail_irq_psl,
1770 .activate_dedicated_process = cxl_activate_dedicated_process_psl9,
1771 .attach_afu_directed = cxl_attach_afu_directed_psl9,
1772 .attach_dedicated_process = cxl_attach_dedicated_process_psl9,
1773 .update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl9,
1774 .debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl9,
1775 .debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl9,
1776 .psl_irq_dump_registers = cxl_native_irq_dump_regs_psl9,
1777 .debugfs_stop_trace = cxl_stop_trace_psl9,
1778 .write_timebase_ctrl = write_timebase_ctrl_psl9,
1779 .timebase_read = timebase_read_psl9,
1780 .capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1781 .needs_reset_before_disable = true,
1784 static const struct cxl_service_layer_ops psl8_ops = {
1785 .adapter_regs_init = init_implementation_adapter_regs_psl8,
1786 .invalidate_all = cxl_invalidate_all_psl8,
1787 .afu_regs_init = init_implementation_afu_regs_psl8,
1788 .sanitise_afu_regs = sanitise_afu_regs_psl8,
1789 .register_serr_irq = cxl_native_register_serr_irq,
1790 .release_serr_irq = cxl_native_release_serr_irq,
1791 .handle_interrupt = cxl_irq_psl8,
1792 .fail_irq = cxl_fail_irq_psl,
1793 .activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1794 .attach_afu_directed = cxl_attach_afu_directed_psl8,
1795 .attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1796 .update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1797 .debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl8,
1798 .debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl8,
1799 .psl_irq_dump_registers = cxl_native_irq_dump_regs_psl8,
1800 .err_irq_dump_registers = cxl_native_err_irq_dump_regs,
1801 .debugfs_stop_trace = cxl_stop_trace_psl8,
1802 .write_timebase_ctrl = write_timebase_ctrl_psl8,
1803 .timebase_read = timebase_read_psl8,
1804 .capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1805 .needs_reset_before_disable = true,
1808 static const struct cxl_service_layer_ops xsl_ops = {
1809 .adapter_regs_init = init_implementation_adapter_regs_xsl,
1810 .invalidate_all = cxl_invalidate_all_psl8,
1811 .sanitise_afu_regs = sanitise_afu_regs_psl8,
1812 .handle_interrupt = cxl_irq_psl8,
1813 .fail_irq = cxl_fail_irq_psl,
1814 .activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1815 .attach_afu_directed = cxl_attach_afu_directed_psl8,
1816 .attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1817 .update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1818 .debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_xsl,
1819 .write_timebase_ctrl = write_timebase_ctrl_xsl,
1820 .timebase_read = timebase_read_xsl,
1821 .capi_mode = OPAL_PHB_CAPI_MODE_DMA,
1824 static void set_sl_ops(struct cxl *adapter, struct pci_dev *dev)
1826 if (dev->vendor == PCI_VENDOR_ID_MELLANOX && dev->device == 0x1013) {
1828 dev_info(&dev->dev, "Device uses an XSL\n");
1829 adapter->native->sl_ops = &xsl_ops;
1830 adapter->min_pe = 1; /* Workaround for CX-4 hardware bug */
1832 if (cxl_is_power8()) {
1833 dev_info(&dev->dev, "Device uses a PSL8\n");
1834 adapter->native->sl_ops = &psl8_ops;
1836 dev_info(&dev->dev, "Device uses a PSL9\n");
1837 adapter->native->sl_ops = &psl9_ops;
1843 static struct cxl *cxl_pci_init_adapter(struct pci_dev *dev)
1845 struct cxl *adapter;
1848 adapter = cxl_alloc_adapter();
1850 return ERR_PTR(-ENOMEM);
1852 adapter->native = kzalloc(sizeof(struct cxl_native), GFP_KERNEL);
1853 if (!adapter->native) {
1858 set_sl_ops(adapter, dev);
1860 /* Set defaults for parameters which need to persist over
1861 * configure/reconfigure
1863 adapter->perst_loads_image = true;
1864 adapter->perst_same_image = false;
1866 rc = cxl_configure_adapter(adapter, dev);
1868 pci_disable_device(dev);
1872 /* Don't care if this one fails: */
1873 cxl_debugfs_adapter_add(adapter);
1876 * After we call this function we must not free the adapter directly,
1877 * even if it returns an error!
1879 if ((rc = cxl_register_adapter(adapter)))
1882 if ((rc = cxl_sysfs_adapter_add(adapter)))
1885 /* Release the context lock as adapter is configured */
1886 cxl_adapter_context_unlock(adapter);
1891 device_del(&adapter->dev);
1893 /* This should mirror cxl_remove_adapter, except without the
1896 cxl_debugfs_adapter_remove(adapter);
1897 cxl_deconfigure_adapter(adapter);
1898 put_device(&adapter->dev);
1902 cxl_release_adapter(&adapter->dev);
1906 static void cxl_pci_remove_adapter(struct cxl *adapter)
1908 pr_devel("cxl_remove_adapter\n");
1910 cxl_sysfs_adapter_remove(adapter);
1911 cxl_debugfs_adapter_remove(adapter);
1914 * Flush adapter datacache as its about to be removed.
1916 cxl_data_cache_flush(adapter);
1918 cxl_deconfigure_adapter(adapter);
1920 device_unregister(&adapter->dev);
1923 #define CXL_MAX_PCIEX_PARENT 2
1925 int cxl_slot_is_switched(struct pci_dev *dev)
1927 struct device_node *np;
1931 if (!(np = pci_device_to_OF_node(dev))) {
1932 pr_err("cxl: np = NULL\n");
1937 np = of_get_next_parent(np);
1938 prop = of_get_property(np, "device_type", NULL);
1939 if (!prop || strcmp((char *)prop, "pciex"))
1944 return (depth > CXL_MAX_PCIEX_PARENT);
1947 bool cxl_slot_is_supported(struct pci_dev *dev, int flags)
1949 if (!cpu_has_feature(CPU_FTR_HVMODE))
1952 if ((flags & CXL_SLOT_FLAG_DMA) && (!pvr_version_is(PVR_POWER8NVL))) {
1954 * CAPP DMA mode is technically supported on regular P8, but
1955 * will EEH if the card attempts to access memory < 4GB, which
1956 * we cannot realistically avoid. We might be able to work
1957 * around the issue, but until then return unsupported:
1962 if (cxl_slot_is_switched(dev))
1966 * XXX: This gets a little tricky on regular P8 (not POWER8NVL) since
1967 * the CAPP can be connected to PHB 0, 1 or 2 on a first come first
1968 * served basis, which is racy to check from here. If we need to
1969 * support this in future we might need to consider having this
1970 * function effectively reserve it ahead of time.
1972 * Currently, the only user of this API is the Mellanox CX4, which is
1973 * only supported on P8NVL due to the above mentioned limitation of
1974 * CAPP DMA mode and therefore does not need to worry about this. If the
1975 * issue with CAPP DMA mode is later worked around on P8 we might need
1981 EXPORT_SYMBOL_GPL(cxl_slot_is_supported);
1984 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
1986 struct cxl *adapter;
1990 if (cxl_pci_is_vphb_device(dev)) {
1991 dev_dbg(&dev->dev, "cxl_init_adapter: Ignoring cxl vphb device\n");
1995 if (cxl_slot_is_switched(dev)) {
1996 dev_info(&dev->dev, "Ignoring card on incompatible PCI slot\n");
2000 if (cxl_is_power9() && !radix_enabled()) {
2001 dev_info(&dev->dev, "Only Radix mode supported\n");
2006 dump_cxl_config_space(dev);
2008 adapter = cxl_pci_init_adapter(dev);
2009 if (IS_ERR(adapter)) {
2010 dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
2011 return PTR_ERR(adapter);
2014 for (slice = 0; slice < adapter->slices; slice++) {
2015 if ((rc = pci_init_afu(adapter, slice, dev))) {
2016 dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
2020 rc = cxl_afu_select_best_mode(adapter->afu[slice]);
2022 dev_err(&dev->dev, "AFU %i failed to start: %i\n", slice, rc);
2025 if (pnv_pci_on_cxl_phb(dev) && adapter->slices >= 1)
2026 pnv_cxl_phb_set_peer_afu(dev, adapter->afu[0]);
2031 static void cxl_remove(struct pci_dev *dev)
2033 struct cxl *adapter = pci_get_drvdata(dev);
2034 struct cxl_afu *afu;
2038 * Lock to prevent someone grabbing a ref through the adapter list as
2039 * we are removing it
2041 for (i = 0; i < adapter->slices; i++) {
2042 afu = adapter->afu[i];
2043 cxl_pci_remove_afu(afu);
2045 cxl_pci_remove_adapter(adapter);
2048 static pci_ers_result_t cxl_vphb_error_detected(struct cxl_afu *afu,
2049 pci_channel_state_t state)
2051 struct pci_dev *afu_dev;
2052 pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
2053 pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
2055 /* There should only be one entry, but go through the list
2058 if (afu == NULL || afu->phb == NULL)
2061 list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2062 if (!afu_dev->driver)
2065 afu_dev->error_state = state;
2067 if (afu_dev->driver->err_handler)
2068 afu_result = afu_dev->driver->err_handler->error_detected(afu_dev,
2070 /* Disconnect trumps all, NONE trumps NEED_RESET */
2071 if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2072 result = PCI_ERS_RESULT_DISCONNECT;
2073 else if ((afu_result == PCI_ERS_RESULT_NONE) &&
2074 (result == PCI_ERS_RESULT_NEED_RESET))
2075 result = PCI_ERS_RESULT_NONE;
2080 static pci_ers_result_t cxl_pci_error_detected(struct pci_dev *pdev,
2081 pci_channel_state_t state)
2083 struct cxl *adapter = pci_get_drvdata(pdev);
2084 struct cxl_afu *afu;
2085 pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
2086 pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
2089 /* At this point, we could still have an interrupt pending.
2090 * Let's try to get them out of the way before they do
2091 * anything we don't like.
2095 /* If we're permanently dead, give up. */
2096 if (state == pci_channel_io_perm_failure) {
2097 spin_lock(&adapter->afu_list_lock);
2098 for (i = 0; i < adapter->slices; i++) {
2099 afu = adapter->afu[i];
2101 * Tell the AFU drivers; but we don't care what they
2102 * say, we're going away.
2104 cxl_vphb_error_detected(afu, state);
2106 spin_unlock(&adapter->afu_list_lock);
2107 return PCI_ERS_RESULT_DISCONNECT;
2110 /* Are we reflashing?
2112 * If we reflash, we could come back as something entirely
2113 * different, including a non-CAPI card. As such, by default
2114 * we don't participate in the process. We'll be unbound and
2115 * the slot re-probed. (TODO: check EEH doesn't blindly rebind
2118 * However, this isn't the entire story: for reliablity
2119 * reasons, we usually want to reflash the FPGA on PERST in
2120 * order to get back to a more reliable known-good state.
2122 * This causes us a bit of a problem: if we reflash we can't
2123 * trust that we'll come back the same - we could have a new
2124 * image and been PERSTed in order to load that
2125 * image. However, most of the time we actually *will* come
2126 * back the same - for example a regular EEH event.
2128 * Therefore, we allow the user to assert that the image is
2129 * indeed the same and that we should continue on into EEH
2132 if (adapter->perst_loads_image && !adapter->perst_same_image) {
2133 /* TODO take the PHB out of CXL mode */
2134 dev_info(&pdev->dev, "reflashing, so opting out of EEH!\n");
2135 return PCI_ERS_RESULT_NONE;
2139 * At this point, we want to try to recover. We'll always
2140 * need a complete slot reset: we don't trust any other reset.
2142 * Now, we go through each AFU:
2143 * - We send the driver, if bound, an error_detected callback.
2144 * We expect it to clean up, but it can also tell us to give
2145 * up and permanently detach the card. To simplify things, if
2146 * any bound AFU driver doesn't support EEH, we give up on EEH.
2148 * - We detach all contexts associated with the AFU. This
2149 * does not free them, but puts them into a CLOSED state
2150 * which causes any the associated files to return useful
2151 * errors to userland. It also unmaps, but does not free,
2154 * - We clean up our side: releasing and unmapping resources we hold
2155 * so we can wire them up again when the hardware comes back up.
2157 * Driver authors should note:
2159 * - Any contexts you create in your kernel driver (except
2160 * those associated with anonymous file descriptors) are
2161 * your responsibility to free and recreate. Likewise with
2162 * any attached resources.
2164 * - We will take responsibility for re-initialising the
2165 * device context (the one set up for you in
2166 * cxl_pci_enable_device_hook and accessed through
2167 * cxl_get_context). If you've attached IRQs or other
2168 * resources to it, they remains yours to free.
2170 * You can call the same functions to release resources as you
2171 * normally would: we make sure that these functions continue
2172 * to work when the hardware is down.
2176 * 1) If you normally free all your resources at the end of
2177 * each request, or if you use anonymous FDs, your
2178 * error_detected callback can simply set a flag to tell
2179 * your driver not to start any new calls. You can then
2180 * clear the flag in the resume callback.
2182 * 2) If you normally allocate your resources on startup:
2183 * * Set a flag in error_detected as above.
2184 * * Let CXL detach your contexts.
2185 * * In slot_reset, free the old resources and allocate new ones.
2186 * * In resume, clear the flag to allow things to start.
2189 /* Make sure no one else changes the afu list */
2190 spin_lock(&adapter->afu_list_lock);
2192 for (i = 0; i < adapter->slices; i++) {
2193 afu = adapter->afu[i];
2198 afu_result = cxl_vphb_error_detected(afu, state);
2199 cxl_context_detach_all(afu);
2200 cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
2201 pci_deconfigure_afu(afu);
2203 /* Disconnect trumps all, NONE trumps NEED_RESET */
2204 if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2205 result = PCI_ERS_RESULT_DISCONNECT;
2206 else if ((afu_result == PCI_ERS_RESULT_NONE) &&
2207 (result == PCI_ERS_RESULT_NEED_RESET))
2208 result = PCI_ERS_RESULT_NONE;
2210 spin_unlock(&adapter->afu_list_lock);
2212 /* should take the context lock here */
2213 if (cxl_adapter_context_lock(adapter) != 0)
2214 dev_warn(&adapter->dev,
2215 "Couldn't take context lock with %d active-contexts\n",
2216 atomic_read(&adapter->contexts_num));
2218 cxl_deconfigure_adapter(adapter);
2223 static pci_ers_result_t cxl_pci_slot_reset(struct pci_dev *pdev)
2225 struct cxl *adapter = pci_get_drvdata(pdev);
2226 struct cxl_afu *afu;
2227 struct cxl_context *ctx;
2228 struct pci_dev *afu_dev;
2229 pci_ers_result_t afu_result = PCI_ERS_RESULT_RECOVERED;
2230 pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
2233 if (cxl_configure_adapter(adapter, pdev))
2237 * Unlock context activation for the adapter. Ideally this should be
2238 * done in cxl_pci_resume but cxlflash module tries to activate the
2239 * master context as part of slot_reset callback.
2241 cxl_adapter_context_unlock(adapter);
2243 spin_lock(&adapter->afu_list_lock);
2244 for (i = 0; i < adapter->slices; i++) {
2245 afu = adapter->afu[i];
2250 if (pci_configure_afu(afu, adapter, pdev))
2253 if (cxl_afu_select_best_mode(afu))
2256 if (afu->phb == NULL)
2259 list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2260 /* Reset the device context.
2261 * TODO: make this less disruptive
2263 ctx = cxl_get_context(afu_dev);
2265 if (ctx && cxl_release_context(ctx))
2268 ctx = cxl_dev_context_init(afu_dev);
2272 afu_dev->dev.archdata.cxl_ctx = ctx;
2274 if (cxl_ops->afu_check_and_enable(afu))
2277 afu_dev->error_state = pci_channel_io_normal;
2279 /* If there's a driver attached, allow it to
2280 * chime in on recovery. Drivers should check
2281 * if everything has come back OK, but
2282 * shouldn't start new work until we call
2283 * their resume function.
2285 if (!afu_dev->driver)
2288 if (afu_dev->driver->err_handler &&
2289 afu_dev->driver->err_handler->slot_reset)
2290 afu_result = afu_dev->driver->err_handler->slot_reset(afu_dev);
2292 if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2293 result = PCI_ERS_RESULT_DISCONNECT;
2297 spin_unlock(&adapter->afu_list_lock);
2301 spin_unlock(&adapter->afu_list_lock);
2304 /* All the bits that happen in both error_detected and cxl_remove
2305 * should be idempotent, so we don't need to worry about leaving a mix
2306 * of unconfigured and reconfigured resources.
2308 dev_err(&pdev->dev, "EEH recovery failed. Asking to be disconnected.\n");
2309 return PCI_ERS_RESULT_DISCONNECT;
2312 static void cxl_pci_resume(struct pci_dev *pdev)
2314 struct cxl *adapter = pci_get_drvdata(pdev);
2315 struct cxl_afu *afu;
2316 struct pci_dev *afu_dev;
2319 /* Everything is back now. Drivers should restart work now.
2320 * This is not the place to be checking if everything came back up
2321 * properly, because there's no return value: do that in slot_reset.
2323 spin_lock(&adapter->afu_list_lock);
2324 for (i = 0; i < adapter->slices; i++) {
2325 afu = adapter->afu[i];
2327 if (afu == NULL || afu->phb == NULL)
2330 list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2331 if (afu_dev->driver && afu_dev->driver->err_handler &&
2332 afu_dev->driver->err_handler->resume)
2333 afu_dev->driver->err_handler->resume(afu_dev);
2336 spin_unlock(&adapter->afu_list_lock);
2339 static const struct pci_error_handlers cxl_err_handler = {
2340 .error_detected = cxl_pci_error_detected,
2341 .slot_reset = cxl_pci_slot_reset,
2342 .resume = cxl_pci_resume,
2345 struct pci_driver cxl_pci_driver = {
2347 .id_table = cxl_pci_tbl,
2349 .remove = cxl_remove,
2350 .shutdown = cxl_remove,
2351 .err_handler = &cxl_err_handler,