1 /*
2 * Copyright 2014 IBM Corp.
3 *
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.
8 */
9
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>
18 #include <linux/of.h>
19 #include <linux/delay.h>
20 #include <asm/opal.h>
21 #include <asm/msi_bitmap.h>
22 #include <asm/pci-bridge.h> /* for struct pci_controller */
23 #include <asm/pnv-pci.h>
24 #include <asm/io.h>
25
26 #include "cxl.h"
27
28
29 #define CXL_PCI_VSEC_ID 0x1280
30 #define CXL_VSEC_MIN_SIZE 0x80
31
32 #define CXL_READ_VSEC_LENGTH(dev, vsec, dest) \
33 { \
34 pci_read_config_word(dev, vsec + 0x6, dest); \
35 *dest >>= 4; \
36 }
37 #define CXL_READ_VSEC_NAFUS(dev, vsec, dest) \
38 pci_read_config_byte(dev, vsec + 0x8, dest)
39
40 #define CXL_READ_VSEC_STATUS(dev, vsec, dest) \
41 pci_read_config_byte(dev, vsec + 0x9, dest)
42 #define CXL_STATUS_SECOND_PORT 0x80
43 #define CXL_STATUS_MSI_X_FULL 0x40
44 #define CXL_STATUS_MSI_X_SINGLE 0x20
45 #define CXL_STATUS_FLASH_RW 0x08
46 #define CXL_STATUS_FLASH_RO 0x04
47 #define CXL_STATUS_LOADABLE_AFU 0x02
48 #define CXL_STATUS_LOADABLE_PSL 0x01
49 /* If we see these features we won't try to use the card */
50 #define CXL_UNSUPPORTED_FEATURES \
51 (CXL_STATUS_MSI_X_FULL | CXL_STATUS_MSI_X_SINGLE)
52
53 #define CXL_READ_VSEC_MODE_CONTROL(dev, vsec, dest) \
54 pci_read_config_byte(dev, vsec + 0xa, dest)
55 #define CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val) \
56 pci_write_config_byte(dev, vsec + 0xa, val)
57 #define CXL_VSEC_PROTOCOL_MASK 0xe0
58 #define CXL_VSEC_PROTOCOL_1024TB 0x80
59 #define CXL_VSEC_PROTOCOL_512TB 0x40
60 #define CXL_VSEC_PROTOCOL_256TB 0x20 /* Power 8 uses this */
61 #define CXL_VSEC_PROTOCOL_ENABLE 0x01
62
63 #define CXL_READ_VSEC_PSL_REVISION(dev, vsec, dest) \
64 pci_read_config_word(dev, vsec + 0xc, dest)
65 #define CXL_READ_VSEC_CAIA_MINOR(dev, vsec, dest) \
66 pci_read_config_byte(dev, vsec + 0xe, dest)
67 #define CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, dest) \
68 pci_read_config_byte(dev, vsec + 0xf, dest)
69 #define CXL_READ_VSEC_BASE_IMAGE(dev, vsec, dest) \
70 pci_read_config_word(dev, vsec + 0x10, dest)
71
72 #define CXL_READ_VSEC_IMAGE_STATE(dev, vsec, dest) \
73 pci_read_config_byte(dev, vsec + 0x13, dest)
74 #define CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, val) \
75 pci_write_config_byte(dev, vsec + 0x13, val)
76 #define CXL_VSEC_USER_IMAGE_LOADED 0x80 /* RO */
77 #define CXL_VSEC_PERST_LOADS_IMAGE 0x20 /* RW */
78 #define CXL_VSEC_PERST_SELECT_USER 0x10 /* RW */
79
80 #define CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, dest) \
81 pci_read_config_dword(dev, vsec + 0x20, dest)
82 #define CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, dest) \
83 pci_read_config_dword(dev, vsec + 0x24, dest)
84 #define CXL_READ_VSEC_PS_OFF(dev, vsec, dest) \
85 pci_read_config_dword(dev, vsec + 0x28, dest)
86 #define CXL_READ_VSEC_PS_SIZE(dev, vsec, dest) \
87 pci_read_config_dword(dev, vsec + 0x2c, dest)
88
89
90 /* This works a little different than the p1/p2 register accesses to make it
91 * easier to pull out individual fields */
92 #define AFUD_READ(afu, off) in_be64(afu->afu_desc_mmio + off)
93 #define EXTRACT_PPC_BIT(val, bit) (!!(val & PPC_BIT(bit)))
94 #define EXTRACT_PPC_BITS(val, bs, be) ((val & PPC_BITMASK(bs, be)) >> PPC_BITLSHIFT(be))
95
96 #define AFUD_READ_INFO(afu) AFUD_READ(afu, 0x0)
97 #define AFUD_NUM_INTS_PER_PROC(val) EXTRACT_PPC_BITS(val, 0, 15)
98 #define AFUD_NUM_PROCS(val) EXTRACT_PPC_BITS(val, 16, 31)
99 #define AFUD_NUM_CRS(val) EXTRACT_PPC_BITS(val, 32, 47)
100 #define AFUD_MULTIMODE(val) EXTRACT_PPC_BIT(val, 48)
101 #define AFUD_PUSH_BLOCK_TRANSFER(val) EXTRACT_PPC_BIT(val, 55)
102 #define AFUD_DEDICATED_PROCESS(val) EXTRACT_PPC_BIT(val, 59)
103 #define AFUD_AFU_DIRECTED(val) EXTRACT_PPC_BIT(val, 61)
104 #define AFUD_TIME_SLICED(val) EXTRACT_PPC_BIT(val, 63)
105 #define AFUD_READ_CR(afu) AFUD_READ(afu, 0x20)
106 #define AFUD_CR_LEN(val) EXTRACT_PPC_BITS(val, 8, 63)
107 #define AFUD_READ_CR_OFF(afu) AFUD_READ(afu, 0x28)
108 #define AFUD_READ_PPPSA(afu) AFUD_READ(afu, 0x30)
109 #define AFUD_PPPSA_PP(val) EXTRACT_PPC_BIT(val, 6)
110 #define AFUD_PPPSA_PSA(val) EXTRACT_PPC_BIT(val, 7)
111 #define AFUD_PPPSA_LEN(val) EXTRACT_PPC_BITS(val, 8, 63)
112 #define AFUD_READ_PPPSA_OFF(afu) AFUD_READ(afu, 0x38)
113 #define AFUD_READ_EB(afu) AFUD_READ(afu, 0x40)
114 #define AFUD_EB_LEN(val) EXTRACT_PPC_BITS(val, 8, 63)
115 #define AFUD_READ_EB_OFF(afu) AFUD_READ(afu, 0x48)
116
cxl_afu_cr_read16(struct cxl_afu * afu,int cr,u64 off)117 u16 cxl_afu_cr_read16(struct cxl_afu *afu, int cr, u64 off)
118 {
119 u64 aligned_off = off & ~0x3L;
120 u32 val;
121
122 val = cxl_afu_cr_read32(afu, cr, aligned_off);
123 return (val >> ((off & 0x2) * 8)) & 0xffff;
124 }
125
cxl_afu_cr_read8(struct cxl_afu * afu,int cr,u64 off)126 u8 cxl_afu_cr_read8(struct cxl_afu *afu, int cr, u64 off)
127 {
128 u64 aligned_off = off & ~0x3L;
129 u32 val;
130
131 val = cxl_afu_cr_read32(afu, cr, aligned_off);
132 return (val >> ((off & 0x3) * 8)) & 0xff;
133 }
134
135 static DEFINE_PCI_DEVICE_TABLE(cxl_pci_tbl) = {
136 { PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0477), },
137 { PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x044b), },
138 { PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x04cf), },
139 { PCI_DEVICE_CLASS(0x120000, ~0), },
140
141 { }
142 };
143 MODULE_DEVICE_TABLE(pci, cxl_pci_tbl);
144
145
146 /*
147 * Mostly using these wrappers to avoid confusion:
148 * priv 1 is BAR2, while priv 2 is BAR0
149 */
p1_base(struct pci_dev * dev)150 static inline resource_size_t p1_base(struct pci_dev *dev)
151 {
152 return pci_resource_start(dev, 2);
153 }
154
p1_size(struct pci_dev * dev)155 static inline resource_size_t p1_size(struct pci_dev *dev)
156 {
157 return pci_resource_len(dev, 2);
158 }
159
p2_base(struct pci_dev * dev)160 static inline resource_size_t p2_base(struct pci_dev *dev)
161 {
162 return pci_resource_start(dev, 0);
163 }
164
p2_size(struct pci_dev * dev)165 static inline resource_size_t p2_size(struct pci_dev *dev)
166 {
167 return pci_resource_len(dev, 0);
168 }
169
find_cxl_vsec(struct pci_dev * dev)170 static int find_cxl_vsec(struct pci_dev *dev)
171 {
172 int vsec = 0;
173 u16 val;
174
175 while ((vsec = pci_find_next_ext_capability(dev, vsec, PCI_EXT_CAP_ID_VNDR))) {
176 pci_read_config_word(dev, vsec + 0x4, &val);
177 if (val == CXL_PCI_VSEC_ID)
178 return vsec;
179 }
180 return 0;
181
182 }
183
dump_cxl_config_space(struct pci_dev * dev)184 static void dump_cxl_config_space(struct pci_dev *dev)
185 {
186 int vsec;
187 u32 val;
188
189 dev_info(&dev->dev, "dump_cxl_config_space\n");
190
191 pci_read_config_dword(dev, PCI_BASE_ADDRESS_0, &val);
192 dev_info(&dev->dev, "BAR0: %#.8x\n", val);
193 pci_read_config_dword(dev, PCI_BASE_ADDRESS_1, &val);
194 dev_info(&dev->dev, "BAR1: %#.8x\n", val);
195 pci_read_config_dword(dev, PCI_BASE_ADDRESS_2, &val);
196 dev_info(&dev->dev, "BAR2: %#.8x\n", val);
197 pci_read_config_dword(dev, PCI_BASE_ADDRESS_3, &val);
198 dev_info(&dev->dev, "BAR3: %#.8x\n", val);
199 pci_read_config_dword(dev, PCI_BASE_ADDRESS_4, &val);
200 dev_info(&dev->dev, "BAR4: %#.8x\n", val);
201 pci_read_config_dword(dev, PCI_BASE_ADDRESS_5, &val);
202 dev_info(&dev->dev, "BAR5: %#.8x\n", val);
203
204 dev_info(&dev->dev, "p1 regs: %#llx, len: %#llx\n",
205 p1_base(dev), p1_size(dev));
206 dev_info(&dev->dev, "p2 regs: %#llx, len: %#llx\n",
207 p1_base(dev), p2_size(dev));
208 dev_info(&dev->dev, "BAR 4/5: %#llx, len: %#llx\n",
209 pci_resource_start(dev, 4), pci_resource_len(dev, 4));
210
211 if (!(vsec = find_cxl_vsec(dev)))
212 return;
213
214 #define show_reg(name, what) \
215 dev_info(&dev->dev, "cxl vsec: %30s: %#x\n", name, what)
216
217 pci_read_config_dword(dev, vsec + 0x0, &val);
218 show_reg("Cap ID", (val >> 0) & 0xffff);
219 show_reg("Cap Ver", (val >> 16) & 0xf);
220 show_reg("Next Cap Ptr", (val >> 20) & 0xfff);
221 pci_read_config_dword(dev, vsec + 0x4, &val);
222 show_reg("VSEC ID", (val >> 0) & 0xffff);
223 show_reg("VSEC Rev", (val >> 16) & 0xf);
224 show_reg("VSEC Length", (val >> 20) & 0xfff);
225 pci_read_config_dword(dev, vsec + 0x8, &val);
226 show_reg("Num AFUs", (val >> 0) & 0xff);
227 show_reg("Status", (val >> 8) & 0xff);
228 show_reg("Mode Control", (val >> 16) & 0xff);
229 show_reg("Reserved", (val >> 24) & 0xff);
230 pci_read_config_dword(dev, vsec + 0xc, &val);
231 show_reg("PSL Rev", (val >> 0) & 0xffff);
232 show_reg("CAIA Ver", (val >> 16) & 0xffff);
233 pci_read_config_dword(dev, vsec + 0x10, &val);
234 show_reg("Base Image Rev", (val >> 0) & 0xffff);
235 show_reg("Reserved", (val >> 16) & 0x0fff);
236 show_reg("Image Control", (val >> 28) & 0x3);
237 show_reg("Reserved", (val >> 30) & 0x1);
238 show_reg("Image Loaded", (val >> 31) & 0x1);
239
240 pci_read_config_dword(dev, vsec + 0x14, &val);
241 show_reg("Reserved", val);
242 pci_read_config_dword(dev, vsec + 0x18, &val);
243 show_reg("Reserved", val);
244 pci_read_config_dword(dev, vsec + 0x1c, &val);
245 show_reg("Reserved", val);
246
247 pci_read_config_dword(dev, vsec + 0x20, &val);
248 show_reg("AFU Descriptor Offset", val);
249 pci_read_config_dword(dev, vsec + 0x24, &val);
250 show_reg("AFU Descriptor Size", val);
251 pci_read_config_dword(dev, vsec + 0x28, &val);
252 show_reg("Problem State Offset", val);
253 pci_read_config_dword(dev, vsec + 0x2c, &val);
254 show_reg("Problem State Size", val);
255
256 pci_read_config_dword(dev, vsec + 0x30, &val);
257 show_reg("Reserved", val);
258 pci_read_config_dword(dev, vsec + 0x34, &val);
259 show_reg("Reserved", val);
260 pci_read_config_dword(dev, vsec + 0x38, &val);
261 show_reg("Reserved", val);
262 pci_read_config_dword(dev, vsec + 0x3c, &val);
263 show_reg("Reserved", val);
264
265 pci_read_config_dword(dev, vsec + 0x40, &val);
266 show_reg("PSL Programming Port", val);
267 pci_read_config_dword(dev, vsec + 0x44, &val);
268 show_reg("PSL Programming Control", val);
269
270 pci_read_config_dword(dev, vsec + 0x48, &val);
271 show_reg("Reserved", val);
272 pci_read_config_dword(dev, vsec + 0x4c, &val);
273 show_reg("Reserved", val);
274
275 pci_read_config_dword(dev, vsec + 0x50, &val);
276 show_reg("Flash Address Register", val);
277 pci_read_config_dword(dev, vsec + 0x54, &val);
278 show_reg("Flash Size Register", val);
279 pci_read_config_dword(dev, vsec + 0x58, &val);
280 show_reg("Flash Status/Control Register", val);
281 pci_read_config_dword(dev, vsec + 0x58, &val);
282 show_reg("Flash Data Port", val);
283
284 #undef show_reg
285 }
286
dump_afu_descriptor(struct cxl_afu * afu)287 static void dump_afu_descriptor(struct cxl_afu *afu)
288 {
289 u64 val;
290
291 #define show_reg(name, what) \
292 dev_info(&afu->dev, "afu desc: %30s: %#llx\n", name, what)
293
294 val = AFUD_READ_INFO(afu);
295 show_reg("num_ints_per_process", AFUD_NUM_INTS_PER_PROC(val));
296 show_reg("num_of_processes", AFUD_NUM_PROCS(val));
297 show_reg("num_of_afu_CRs", AFUD_NUM_CRS(val));
298 show_reg("req_prog_mode", val & 0xffffULL);
299
300 val = AFUD_READ(afu, 0x8);
301 show_reg("Reserved", val);
302 val = AFUD_READ(afu, 0x10);
303 show_reg("Reserved", val);
304 val = AFUD_READ(afu, 0x18);
305 show_reg("Reserved", val);
306
307 val = AFUD_READ_CR(afu);
308 show_reg("Reserved", (val >> (63-7)) & 0xff);
309 show_reg("AFU_CR_len", AFUD_CR_LEN(val));
310
311 val = AFUD_READ_CR_OFF(afu);
312 show_reg("AFU_CR_offset", val);
313
314 val = AFUD_READ_PPPSA(afu);
315 show_reg("PerProcessPSA_control", (val >> (63-7)) & 0xff);
316 show_reg("PerProcessPSA Length", AFUD_PPPSA_LEN(val));
317
318 val = AFUD_READ_PPPSA_OFF(afu);
319 show_reg("PerProcessPSA_offset", val);
320
321 val = AFUD_READ_EB(afu);
322 show_reg("Reserved", (val >> (63-7)) & 0xff);
323 show_reg("AFU_EB_len", AFUD_EB_LEN(val));
324
325 val = AFUD_READ_EB_OFF(afu);
326 show_reg("AFU_EB_offset", val);
327
328 #undef show_reg
329 }
330
init_implementation_adapter_regs(struct cxl * adapter,struct pci_dev * dev)331 static int init_implementation_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
332 {
333 struct device_node *np;
334 const __be32 *prop;
335 u64 psl_dsnctl;
336 u64 chipid;
337
338 if (!(np = pnv_pci_get_phb_node(dev)))
339 return -ENODEV;
340
341 while (np && !(prop = of_get_property(np, "ibm,chip-id", NULL)))
342 np = of_get_next_parent(np);
343 if (!np)
344 return -ENODEV;
345 chipid = be32_to_cpup(prop);
346 of_node_put(np);
347
348 /* Tell PSL where to route data to */
349 psl_dsnctl = 0x02E8900002000000ULL | (chipid << (63-5));
350 cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
351 cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
352 /* snoop write mask */
353 cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
354 /* set fir_accum */
355 cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, 0x0800000000000000ULL);
356 /* for debugging with trace arrays */
357 cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
358
359 return 0;
360 }
361
init_implementation_afu_regs(struct cxl_afu * afu)362 static int init_implementation_afu_regs(struct cxl_afu *afu)
363 {
364 /* read/write masks for this slice */
365 cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
366 /* APC read/write masks for this slice */
367 cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
368 /* for debugging with trace arrays */
369 cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
370 cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
371
372 return 0;
373 }
374
cxl_setup_irq(struct cxl * adapter,unsigned int hwirq,unsigned int virq)375 int cxl_setup_irq(struct cxl *adapter, unsigned int hwirq,
376 unsigned int virq)
377 {
378 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
379
380 return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
381 }
382
cxl_update_image_control(struct cxl * adapter)383 int cxl_update_image_control(struct cxl *adapter)
384 {
385 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
386 int rc;
387 int vsec;
388 u8 image_state;
389
390 if (!(vsec = find_cxl_vsec(dev))) {
391 dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
392 return -ENODEV;
393 }
394
395 if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
396 dev_err(&dev->dev, "failed to read image state: %i\n", rc);
397 return rc;
398 }
399
400 if (adapter->perst_loads_image)
401 image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
402 else
403 image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
404
405 if (adapter->perst_select_user)
406 image_state |= CXL_VSEC_PERST_SELECT_USER;
407 else
408 image_state &= ~CXL_VSEC_PERST_SELECT_USER;
409
410 if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
411 dev_err(&dev->dev, "failed to update image control: %i\n", rc);
412 return rc;
413 }
414
415 return 0;
416 }
417
cxl_alloc_one_irq(struct cxl * adapter)418 int cxl_alloc_one_irq(struct cxl *adapter)
419 {
420 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
421
422 return pnv_cxl_alloc_hwirqs(dev, 1);
423 }
424
cxl_release_one_irq(struct cxl * adapter,int hwirq)425 void cxl_release_one_irq(struct cxl *adapter, int hwirq)
426 {
427 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
428
429 return pnv_cxl_release_hwirqs(dev, hwirq, 1);
430 }
431
cxl_alloc_irq_ranges(struct cxl_irq_ranges * irqs,struct cxl * adapter,unsigned int num)432 int cxl_alloc_irq_ranges(struct cxl_irq_ranges *irqs, struct cxl *adapter, unsigned int num)
433 {
434 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
435
436 return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
437 }
438
cxl_release_irq_ranges(struct cxl_irq_ranges * irqs,struct cxl * adapter)439 void cxl_release_irq_ranges(struct cxl_irq_ranges *irqs, struct cxl *adapter)
440 {
441 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
442
443 pnv_cxl_release_hwirq_ranges(irqs, dev);
444 }
445
setup_cxl_bars(struct pci_dev * dev)446 static int setup_cxl_bars(struct pci_dev *dev)
447 {
448 /* Safety check in case we get backported to < 3.17 without M64 */
449 if ((p1_base(dev) < 0x100000000ULL) ||
450 (p2_base(dev) < 0x100000000ULL)) {
451 dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
452 return -ENODEV;
453 }
454
455 /*
456 * BAR 4/5 has a special meaning for CXL and must be programmed with a
457 * special value corresponding to the CXL protocol address range.
458 * For POWER 8 that means bits 48:49 must be set to 10
459 */
460 pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
461 pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
462
463 return 0;
464 }
465
466 /* pciex node: ibm,opal-m64-window = <0x3d058 0x0 0x3d058 0x0 0x8 0x0>; */
switch_card_to_cxl(struct pci_dev * dev)467 static int switch_card_to_cxl(struct pci_dev *dev)
468 {
469 int vsec;
470 u8 val;
471 int rc;
472
473 dev_info(&dev->dev, "switch card to CXL\n");
474
475 if (!(vsec = find_cxl_vsec(dev))) {
476 dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
477 return -ENODEV;
478 }
479
480 if ((rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val))) {
481 dev_err(&dev->dev, "failed to read current mode control: %i", rc);
482 return rc;
483 }
484 val &= ~CXL_VSEC_PROTOCOL_MASK;
485 val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
486 if ((rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val))) {
487 dev_err(&dev->dev, "failed to enable CXL protocol: %i", rc);
488 return rc;
489 }
490 /*
491 * The CAIA spec (v0.12 11.6 Bi-modal Device Support) states
492 * we must wait 100ms after this mode switch before touching
493 * PCIe config space.
494 */
495 msleep(100);
496
497 return 0;
498 }
499
cxl_map_slice_regs(struct cxl_afu * afu,struct cxl * adapter,struct pci_dev * dev)500 static int cxl_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
501 {
502 u64 p1n_base, p2n_base, afu_desc;
503 const u64 p1n_size = 0x100;
504 const u64 p2n_size = 0x1000;
505
506 p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
507 p2n_base = p2_base(dev) + (afu->slice * p2n_size);
508 afu->psn_phys = p2_base(dev) + (adapter->ps_off + (afu->slice * adapter->ps_size));
509 afu_desc = p2_base(dev) + adapter->afu_desc_off + (afu->slice * adapter->afu_desc_size);
510
511 if (!(afu->p1n_mmio = ioremap(p1n_base, p1n_size)))
512 goto err;
513 if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
514 goto err1;
515 if (afu_desc) {
516 if (!(afu->afu_desc_mmio = ioremap(afu_desc, adapter->afu_desc_size)))
517 goto err2;
518 }
519
520 return 0;
521 err2:
522 iounmap(afu->p2n_mmio);
523 err1:
524 iounmap(afu->p1n_mmio);
525 err:
526 dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
527 return -ENOMEM;
528 }
529
cxl_unmap_slice_regs(struct cxl_afu * afu)530 static void cxl_unmap_slice_regs(struct cxl_afu *afu)
531 {
532 if (afu->p1n_mmio)
533 iounmap(afu->p2n_mmio);
534 if (afu->p1n_mmio)
535 iounmap(afu->p1n_mmio);
536 }
537
cxl_release_afu(struct device * dev)538 static void cxl_release_afu(struct device *dev)
539 {
540 struct cxl_afu *afu = to_cxl_afu(dev);
541
542 pr_devel("cxl_release_afu\n");
543
544 kfree(afu);
545 }
546
cxl_alloc_afu(struct cxl * adapter,int slice)547 static struct cxl_afu *cxl_alloc_afu(struct cxl *adapter, int slice)
548 {
549 struct cxl_afu *afu;
550
551 if (!(afu = kzalloc(sizeof(struct cxl_afu), GFP_KERNEL)))
552 return NULL;
553
554 afu->adapter = adapter;
555 afu->dev.parent = &adapter->dev;
556 afu->dev.release = cxl_release_afu;
557 afu->slice = slice;
558 idr_init(&afu->contexts_idr);
559 mutex_init(&afu->contexts_lock);
560 spin_lock_init(&afu->afu_cntl_lock);
561 mutex_init(&afu->spa_mutex);
562
563 afu->prefault_mode = CXL_PREFAULT_NONE;
564 afu->irqs_max = afu->adapter->user_irqs;
565
566 return afu;
567 }
568
569 /* Expects AFU struct to have recently been zeroed out */
cxl_read_afu_descriptor(struct cxl_afu * afu)570 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
571 {
572 u64 val;
573
574 val = AFUD_READ_INFO(afu);
575 afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
576 afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
577 afu->crs_num = AFUD_NUM_CRS(val);
578
579 if (AFUD_AFU_DIRECTED(val))
580 afu->modes_supported |= CXL_MODE_DIRECTED;
581 if (AFUD_DEDICATED_PROCESS(val))
582 afu->modes_supported |= CXL_MODE_DEDICATED;
583 if (AFUD_TIME_SLICED(val))
584 afu->modes_supported |= CXL_MODE_TIME_SLICED;
585
586 val = AFUD_READ_PPPSA(afu);
587 afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
588 afu->psa = AFUD_PPPSA_PSA(val);
589 if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
590 afu->pp_offset = AFUD_READ_PPPSA_OFF(afu);
591
592 val = AFUD_READ_CR(afu);
593 afu->crs_len = AFUD_CR_LEN(val) * 256;
594 afu->crs_offset = AFUD_READ_CR_OFF(afu);
595
596 return 0;
597 }
598
cxl_afu_descriptor_looks_ok(struct cxl_afu * afu)599 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
600 {
601 int i;
602
603 if (afu->psa && afu->adapter->ps_size <
604 (afu->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
605 dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
606 return -ENODEV;
607 }
608
609 if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
610 dev_warn(&afu->dev, "AFU uses < PAGE_SIZE per-process PSA!");
611
612 for (i = 0; i < afu->crs_num; i++) {
613 if ((cxl_afu_cr_read32(afu, i, 0) == 0)) {
614 dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
615 return -EINVAL;
616 }
617 }
618
619 return 0;
620 }
621
sanitise_afu_regs(struct cxl_afu * afu)622 static int sanitise_afu_regs(struct cxl_afu *afu)
623 {
624 u64 reg;
625
626 /*
627 * Clear out any regs that contain either an IVTE or address or may be
628 * waiting on an acknowledgement to try to be a bit safer as we bring
629 * it online
630 */
631 reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
632 if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
633 dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#.16llx\n", reg);
634 if (cxl_afu_reset(afu))
635 return -EIO;
636 if (cxl_afu_disable(afu))
637 return -EIO;
638 if (cxl_psl_purge(afu))
639 return -EIO;
640 }
641 cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
642 cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
643 cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
644 cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
645 cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
646 cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
647 cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
648 cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
649 cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
650 cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
651 cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
652 reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
653 if (reg) {
654 dev_warn(&afu->dev, "AFU had pending DSISR: %#.16llx\n", reg);
655 if (reg & CXL_PSL_DSISR_TRANS)
656 cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
657 else
658 cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
659 }
660 reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
661 if (reg) {
662 if (reg & ~0xffff)
663 dev_warn(&afu->dev, "AFU had pending SERR: %#.16llx\n", reg);
664 cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
665 }
666 reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
667 if (reg) {
668 dev_warn(&afu->dev, "AFU had pending error status: %#.16llx\n", reg);
669 cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
670 }
671
672 return 0;
673 }
674
cxl_init_afu(struct cxl * adapter,int slice,struct pci_dev * dev)675 static int cxl_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
676 {
677 struct cxl_afu *afu;
678 bool free = true;
679 int rc;
680
681 if (!(afu = cxl_alloc_afu(adapter, slice)))
682 return -ENOMEM;
683
684 if ((rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice)))
685 goto err1;
686
687 if ((rc = cxl_map_slice_regs(afu, adapter, dev)))
688 goto err1;
689
690 if ((rc = sanitise_afu_regs(afu)))
691 goto err2;
692
693 /* We need to reset the AFU before we can read the AFU descriptor */
694 if ((rc = cxl_afu_reset(afu)))
695 goto err2;
696
697 if (cxl_verbose)
698 dump_afu_descriptor(afu);
699
700 if ((rc = cxl_read_afu_descriptor(afu)))
701 goto err2;
702
703 if ((rc = cxl_afu_descriptor_looks_ok(afu)))
704 goto err2;
705
706 if ((rc = init_implementation_afu_regs(afu)))
707 goto err2;
708
709 if ((rc = cxl_register_serr_irq(afu)))
710 goto err2;
711
712 if ((rc = cxl_register_psl_irq(afu)))
713 goto err3;
714
715 /* Don't care if this fails */
716 cxl_debugfs_afu_add(afu);
717
718 /*
719 * After we call this function we must not free the afu directly, even
720 * if it returns an error!
721 */
722 if ((rc = cxl_register_afu(afu)))
723 goto err_put1;
724
725 if ((rc = cxl_sysfs_afu_add(afu)))
726 goto err_put1;
727
728
729 if ((rc = cxl_afu_select_best_mode(afu)))
730 goto err_put2;
731
732 adapter->afu[afu->slice] = afu;
733
734 return 0;
735
736 err_put2:
737 cxl_sysfs_afu_remove(afu);
738 err_put1:
739 device_unregister(&afu->dev);
740 free = false;
741 cxl_debugfs_afu_remove(afu);
742 cxl_release_psl_irq(afu);
743 err3:
744 cxl_release_serr_irq(afu);
745 err2:
746 cxl_unmap_slice_regs(afu);
747 err1:
748 if (free)
749 kfree(afu);
750 return rc;
751 }
752
cxl_remove_afu(struct cxl_afu * afu)753 static void cxl_remove_afu(struct cxl_afu *afu)
754 {
755 pr_devel("cxl_remove_afu\n");
756
757 if (!afu)
758 return;
759
760 cxl_sysfs_afu_remove(afu);
761 cxl_debugfs_afu_remove(afu);
762
763 spin_lock(&afu->adapter->afu_list_lock);
764 afu->adapter->afu[afu->slice] = NULL;
765 spin_unlock(&afu->adapter->afu_list_lock);
766
767 cxl_context_detach_all(afu);
768 cxl_afu_deactivate_mode(afu);
769
770 cxl_release_psl_irq(afu);
771 cxl_release_serr_irq(afu);
772 cxl_unmap_slice_regs(afu);
773
774 device_unregister(&afu->dev);
775 }
776
cxl_reset(struct cxl * adapter)777 int cxl_reset(struct cxl *adapter)
778 {
779 struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
780 int rc;
781
782 dev_info(&dev->dev, "CXL reset\n");
783
784 /* pcie_warm_reset requests a fundamental pci reset which includes a
785 * PERST assert/deassert. PERST triggers a loading of the image
786 * if "user" or "factory" is selected in sysfs */
787 if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
788 dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
789 return rc;
790 }
791
792 return rc;
793 }
794
cxl_map_adapter_regs(struct cxl * adapter,struct pci_dev * dev)795 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
796 {
797 if (pci_request_region(dev, 2, "priv 2 regs"))
798 goto err1;
799 if (pci_request_region(dev, 0, "priv 1 regs"))
800 goto err2;
801
802 pr_devel("cxl_map_adapter_regs: p1: %#.16llx %#llx, p2: %#.16llx %#llx",
803 p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
804
805 if (!(adapter->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
806 goto err3;
807
808 if (!(adapter->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
809 goto err4;
810
811 return 0;
812
813 err4:
814 iounmap(adapter->p1_mmio);
815 adapter->p1_mmio = NULL;
816 err3:
817 pci_release_region(dev, 0);
818 err2:
819 pci_release_region(dev, 2);
820 err1:
821 return -ENOMEM;
822 }
823
cxl_unmap_adapter_regs(struct cxl * adapter)824 static void cxl_unmap_adapter_regs(struct cxl *adapter)
825 {
826 if (adapter->p1_mmio)
827 iounmap(adapter->p1_mmio);
828 if (adapter->p2_mmio)
829 iounmap(adapter->p2_mmio);
830 }
831
cxl_read_vsec(struct cxl * adapter,struct pci_dev * dev)832 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
833 {
834 int vsec;
835 u32 afu_desc_off, afu_desc_size;
836 u32 ps_off, ps_size;
837 u16 vseclen;
838 u8 image_state;
839
840 if (!(vsec = find_cxl_vsec(dev))) {
841 dev_err(&adapter->dev, "ABORTING: CXL VSEC not found!\n");
842 return -ENODEV;
843 }
844
845 CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
846 if (vseclen < CXL_VSEC_MIN_SIZE) {
847 pr_err("ABORTING: CXL VSEC too short\n");
848 return -EINVAL;
849 }
850
851 CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
852 CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
853 CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
854 CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
855 CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
856 CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
857 adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
858 adapter->perst_loads_image = true;
859 adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
860
861 CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
862 CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
863 CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
864 CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
865 CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
866
867 /* Convert everything to bytes, because there is NO WAY I'd look at the
868 * code a month later and forget what units these are in ;-) */
869 adapter->ps_off = ps_off * 64 * 1024;
870 adapter->ps_size = ps_size * 64 * 1024;
871 adapter->afu_desc_off = afu_desc_off * 64 * 1024;
872 adapter->afu_desc_size = afu_desc_size *64 * 1024;
873
874 /* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
875 adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
876
877 return 0;
878 }
879
cxl_vsec_looks_ok(struct cxl * adapter,struct pci_dev * dev)880 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
881 {
882 if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
883 return -EBUSY;
884
885 if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
886 dev_err(&adapter->dev, "ABORTING: CXL requires unsupported features\n");
887 return -EINVAL;
888 }
889
890 if (!adapter->slices) {
891 /* Once we support dynamic reprogramming we can use the card if
892 * it supports loadable AFUs */
893 dev_err(&adapter->dev, "ABORTING: Device has no AFUs\n");
894 return -EINVAL;
895 }
896
897 if (!adapter->afu_desc_off || !adapter->afu_desc_size) {
898 dev_err(&adapter->dev, "ABORTING: VSEC shows no AFU descriptors\n");
899 return -EINVAL;
900 }
901
902 if (adapter->ps_size > p2_size(dev) - adapter->ps_off) {
903 dev_err(&adapter->dev, "ABORTING: Problem state size larger than "
904 "available in BAR2: 0x%llx > 0x%llx\n",
905 adapter->ps_size, p2_size(dev) - adapter->ps_off);
906 return -EINVAL;
907 }
908
909 return 0;
910 }
911
cxl_release_adapter(struct device * dev)912 static void cxl_release_adapter(struct device *dev)
913 {
914 struct cxl *adapter = to_cxl_adapter(dev);
915
916 pr_devel("cxl_release_adapter\n");
917
918 kfree(adapter);
919 }
920
cxl_alloc_adapter(struct pci_dev * dev)921 static struct cxl *cxl_alloc_adapter(struct pci_dev *dev)
922 {
923 struct cxl *adapter;
924
925 if (!(adapter = kzalloc(sizeof(struct cxl), GFP_KERNEL)))
926 return NULL;
927
928 adapter->dev.parent = &dev->dev;
929 adapter->dev.release = cxl_release_adapter;
930 pci_set_drvdata(dev, adapter);
931 spin_lock_init(&adapter->afu_list_lock);
932
933 return adapter;
934 }
935
sanitise_adapter_regs(struct cxl * adapter)936 static int sanitise_adapter_regs(struct cxl *adapter)
937 {
938 cxl_p1_write(adapter, CXL_PSL_ErrIVTE, 0x0000000000000000);
939 return cxl_tlb_slb_invalidate(adapter);
940 }
941
cxl_init_adapter(struct pci_dev * dev)942 static struct cxl *cxl_init_adapter(struct pci_dev *dev)
943 {
944 struct cxl *adapter;
945 bool free = true;
946 int rc;
947
948
949 if (!(adapter = cxl_alloc_adapter(dev)))
950 return ERR_PTR(-ENOMEM);
951
952 if ((rc = switch_card_to_cxl(dev)))
953 goto err1;
954
955 if ((rc = cxl_alloc_adapter_nr(adapter)))
956 goto err1;
957
958 if ((rc = dev_set_name(&adapter->dev, "card%i", adapter->adapter_num)))
959 goto err2;
960
961 if ((rc = cxl_read_vsec(adapter, dev)))
962 goto err2;
963
964 if ((rc = cxl_vsec_looks_ok(adapter, dev)))
965 goto err2;
966
967 if ((rc = cxl_update_image_control(adapter)))
968 goto err2;
969
970 if ((rc = cxl_map_adapter_regs(adapter, dev)))
971 goto err2;
972
973 if ((rc = sanitise_adapter_regs(adapter)))
974 goto err2;
975
976 if ((rc = init_implementation_adapter_regs(adapter, dev)))
977 goto err3;
978
979 if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_CAPI)))
980 goto err3;
981
982 /* If recovery happened, the last step is to turn on snooping.
983 * In the non-recovery case this has no effect */
984 if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON))) {
985 goto err3;
986 }
987
988 if ((rc = cxl_register_psl_err_irq(adapter)))
989 goto err3;
990
991 /* Don't care if this one fails: */
992 cxl_debugfs_adapter_add(adapter);
993
994 /*
995 * After we call this function we must not free the adapter directly,
996 * even if it returns an error!
997 */
998 if ((rc = cxl_register_adapter(adapter)))
999 goto err_put1;
1000
1001 if ((rc = cxl_sysfs_adapter_add(adapter)))
1002 goto err_put1;
1003
1004 return adapter;
1005
1006 err_put1:
1007 device_unregister(&adapter->dev);
1008 free = false;
1009 cxl_debugfs_adapter_remove(adapter);
1010 cxl_release_psl_err_irq(adapter);
1011 err3:
1012 cxl_unmap_adapter_regs(adapter);
1013 err2:
1014 cxl_remove_adapter_nr(adapter);
1015 err1:
1016 if (free)
1017 kfree(adapter);
1018 return ERR_PTR(rc);
1019 }
1020
cxl_remove_adapter(struct cxl * adapter)1021 static void cxl_remove_adapter(struct cxl *adapter)
1022 {
1023 struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1024
1025 pr_devel("cxl_release_adapter\n");
1026
1027 cxl_sysfs_adapter_remove(adapter);
1028 cxl_debugfs_adapter_remove(adapter);
1029 cxl_release_psl_err_irq(adapter);
1030 cxl_unmap_adapter_regs(adapter);
1031 cxl_remove_adapter_nr(adapter);
1032
1033 device_unregister(&adapter->dev);
1034
1035 pci_release_region(pdev, 0);
1036 pci_release_region(pdev, 2);
1037 pci_disable_device(pdev);
1038 }
1039
cxl_probe(struct pci_dev * dev,const struct pci_device_id * id)1040 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
1041 {
1042 struct cxl *adapter;
1043 int slice;
1044 int rc;
1045
1046 if (cxl_verbose)
1047 dump_cxl_config_space(dev);
1048
1049 if ((rc = setup_cxl_bars(dev)))
1050 return rc;
1051
1052 if ((rc = pci_enable_device(dev))) {
1053 dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1054 return rc;
1055 }
1056
1057 adapter = cxl_init_adapter(dev);
1058 if (IS_ERR(adapter)) {
1059 dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
1060 return PTR_ERR(adapter);
1061 }
1062
1063 for (slice = 0; slice < adapter->slices; slice++) {
1064 if ((rc = cxl_init_afu(adapter, slice, dev)))
1065 dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
1066 }
1067
1068 return 0;
1069 }
1070
cxl_remove(struct pci_dev * dev)1071 static void cxl_remove(struct pci_dev *dev)
1072 {
1073 struct cxl *adapter = pci_get_drvdata(dev);
1074 int afu;
1075
1076 dev_warn(&dev->dev, "pci remove\n");
1077
1078 /*
1079 * Lock to prevent someone grabbing a ref through the adapter list as
1080 * we are removing it
1081 */
1082 for (afu = 0; afu < adapter->slices; afu++)
1083 cxl_remove_afu(adapter->afu[afu]);
1084 cxl_remove_adapter(adapter);
1085 }
1086
1087 struct pci_driver cxl_pci_driver = {
1088 .name = "cxl-pci",
1089 .id_table = cxl_pci_tbl,
1090 .probe = cxl_probe,
1091 .remove = cxl_remove,
1092 };
1093