xref: /linux/drivers/misc/cxl/pci.c (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
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/pnv-pci.h>
23 #include <asm/io.h>
24 #include <asm/reg.h>
25 
26 #include "cxl.h"
27 #include <misc/cxl.h>
28 
29 
30 #define CXL_PCI_VSEC_ID	0x1280
31 #define CXL_VSEC_MIN_SIZE 0x80
32 
33 #define CXL_READ_VSEC_LENGTH(dev, vsec, dest)			\
34 	{							\
35 		pci_read_config_word(dev, vsec + 0x6, dest);	\
36 		*dest >>= 4;					\
37 	}
38 #define CXL_READ_VSEC_NAFUS(dev, vsec, dest) \
39 	pci_read_config_byte(dev, vsec + 0x8, dest)
40 
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)
53 
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 uses this */
64 #define CXL_VSEC_PROTOCOL_ENABLE 0x01
65 
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)
74 
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 */
82 
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)
91 
92 
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))
99 
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)
120 
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_CLASS(0x120000, ~0), },
127 
128 	{ }
129 };
130 MODULE_DEVICE_TABLE(pci, cxl_pci_tbl);
131 
132 
133 /*
134  * Mostly using these wrappers to avoid confusion:
135  * priv 1 is BAR2, while priv 2 is BAR0
136  */
137 static inline resource_size_t p1_base(struct pci_dev *dev)
138 {
139 	return pci_resource_start(dev, 2);
140 }
141 
142 static inline resource_size_t p1_size(struct pci_dev *dev)
143 {
144 	return pci_resource_len(dev, 2);
145 }
146 
147 static inline resource_size_t p2_base(struct pci_dev *dev)
148 {
149 	return pci_resource_start(dev, 0);
150 }
151 
152 static inline resource_size_t p2_size(struct pci_dev *dev)
153 {
154 	return pci_resource_len(dev, 0);
155 }
156 
157 static int find_cxl_vsec(struct pci_dev *dev)
158 {
159 	int vsec = 0;
160 	u16 val;
161 
162 	while ((vsec = pci_find_next_ext_capability(dev, vsec, PCI_EXT_CAP_ID_VNDR))) {
163 		pci_read_config_word(dev, vsec + 0x4, &val);
164 		if (val == CXL_PCI_VSEC_ID)
165 			return vsec;
166 	}
167 	return 0;
168 
169 }
170 
171 static void dump_cxl_config_space(struct pci_dev *dev)
172 {
173 	int vsec;
174 	u32 val;
175 
176 	dev_info(&dev->dev, "dump_cxl_config_space\n");
177 
178 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_0, &val);
179 	dev_info(&dev->dev, "BAR0: %#.8x\n", val);
180 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_1, &val);
181 	dev_info(&dev->dev, "BAR1: %#.8x\n", val);
182 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_2, &val);
183 	dev_info(&dev->dev, "BAR2: %#.8x\n", val);
184 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_3, &val);
185 	dev_info(&dev->dev, "BAR3: %#.8x\n", val);
186 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_4, &val);
187 	dev_info(&dev->dev, "BAR4: %#.8x\n", val);
188 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_5, &val);
189 	dev_info(&dev->dev, "BAR5: %#.8x\n", val);
190 
191 	dev_info(&dev->dev, "p1 regs: %#llx, len: %#llx\n",
192 		p1_base(dev), p1_size(dev));
193 	dev_info(&dev->dev, "p2 regs: %#llx, len: %#llx\n",
194 		p2_base(dev), p2_size(dev));
195 	dev_info(&dev->dev, "BAR 4/5: %#llx, len: %#llx\n",
196 		pci_resource_start(dev, 4), pci_resource_len(dev, 4));
197 
198 	if (!(vsec = find_cxl_vsec(dev)))
199 		return;
200 
201 #define show_reg(name, what) \
202 	dev_info(&dev->dev, "cxl vsec: %30s: %#x\n", name, what)
203 
204 	pci_read_config_dword(dev, vsec + 0x0, &val);
205 	show_reg("Cap ID", (val >> 0) & 0xffff);
206 	show_reg("Cap Ver", (val >> 16) & 0xf);
207 	show_reg("Next Cap Ptr", (val >> 20) & 0xfff);
208 	pci_read_config_dword(dev, vsec + 0x4, &val);
209 	show_reg("VSEC ID", (val >> 0) & 0xffff);
210 	show_reg("VSEC Rev", (val >> 16) & 0xf);
211 	show_reg("VSEC Length",	(val >> 20) & 0xfff);
212 	pci_read_config_dword(dev, vsec + 0x8, &val);
213 	show_reg("Num AFUs", (val >> 0) & 0xff);
214 	show_reg("Status", (val >> 8) & 0xff);
215 	show_reg("Mode Control", (val >> 16) & 0xff);
216 	show_reg("Reserved", (val >> 24) & 0xff);
217 	pci_read_config_dword(dev, vsec + 0xc, &val);
218 	show_reg("PSL Rev", (val >> 0) & 0xffff);
219 	show_reg("CAIA Ver", (val >> 16) & 0xffff);
220 	pci_read_config_dword(dev, vsec + 0x10, &val);
221 	show_reg("Base Image Rev", (val >> 0) & 0xffff);
222 	show_reg("Reserved", (val >> 16) & 0x0fff);
223 	show_reg("Image Control", (val >> 28) & 0x3);
224 	show_reg("Reserved", (val >> 30) & 0x1);
225 	show_reg("Image Loaded", (val >> 31) & 0x1);
226 
227 	pci_read_config_dword(dev, vsec + 0x14, &val);
228 	show_reg("Reserved", val);
229 	pci_read_config_dword(dev, vsec + 0x18, &val);
230 	show_reg("Reserved", val);
231 	pci_read_config_dword(dev, vsec + 0x1c, &val);
232 	show_reg("Reserved", val);
233 
234 	pci_read_config_dword(dev, vsec + 0x20, &val);
235 	show_reg("AFU Descriptor Offset", val);
236 	pci_read_config_dword(dev, vsec + 0x24, &val);
237 	show_reg("AFU Descriptor Size", val);
238 	pci_read_config_dword(dev, vsec + 0x28, &val);
239 	show_reg("Problem State Offset", val);
240 	pci_read_config_dword(dev, vsec + 0x2c, &val);
241 	show_reg("Problem State Size", val);
242 
243 	pci_read_config_dword(dev, vsec + 0x30, &val);
244 	show_reg("Reserved", val);
245 	pci_read_config_dword(dev, vsec + 0x34, &val);
246 	show_reg("Reserved", val);
247 	pci_read_config_dword(dev, vsec + 0x38, &val);
248 	show_reg("Reserved", val);
249 	pci_read_config_dword(dev, vsec + 0x3c, &val);
250 	show_reg("Reserved", val);
251 
252 	pci_read_config_dword(dev, vsec + 0x40, &val);
253 	show_reg("PSL Programming Port", val);
254 	pci_read_config_dword(dev, vsec + 0x44, &val);
255 	show_reg("PSL Programming Control", val);
256 
257 	pci_read_config_dword(dev, vsec + 0x48, &val);
258 	show_reg("Reserved", val);
259 	pci_read_config_dword(dev, vsec + 0x4c, &val);
260 	show_reg("Reserved", val);
261 
262 	pci_read_config_dword(dev, vsec + 0x50, &val);
263 	show_reg("Flash Address Register", val);
264 	pci_read_config_dword(dev, vsec + 0x54, &val);
265 	show_reg("Flash Size Register", val);
266 	pci_read_config_dword(dev, vsec + 0x58, &val);
267 	show_reg("Flash Status/Control Register", val);
268 	pci_read_config_dword(dev, vsec + 0x58, &val);
269 	show_reg("Flash Data Port", val);
270 
271 #undef show_reg
272 }
273 
274 static void dump_afu_descriptor(struct cxl_afu *afu)
275 {
276 	u64 val, afu_cr_num, afu_cr_off, afu_cr_len;
277 	int i;
278 
279 #define show_reg(name, what) \
280 	dev_info(&afu->dev, "afu desc: %30s: %#llx\n", name, what)
281 
282 	val = AFUD_READ_INFO(afu);
283 	show_reg("num_ints_per_process", AFUD_NUM_INTS_PER_PROC(val));
284 	show_reg("num_of_processes", AFUD_NUM_PROCS(val));
285 	show_reg("num_of_afu_CRs", AFUD_NUM_CRS(val));
286 	show_reg("req_prog_mode", val & 0xffffULL);
287 	afu_cr_num = AFUD_NUM_CRS(val);
288 
289 	val = AFUD_READ(afu, 0x8);
290 	show_reg("Reserved", val);
291 	val = AFUD_READ(afu, 0x10);
292 	show_reg("Reserved", val);
293 	val = AFUD_READ(afu, 0x18);
294 	show_reg("Reserved", val);
295 
296 	val = AFUD_READ_CR(afu);
297 	show_reg("Reserved", (val >> (63-7)) & 0xff);
298 	show_reg("AFU_CR_len", AFUD_CR_LEN(val));
299 	afu_cr_len = AFUD_CR_LEN(val) * 256;
300 
301 	val = AFUD_READ_CR_OFF(afu);
302 	afu_cr_off = val;
303 	show_reg("AFU_CR_offset", val);
304 
305 	val = AFUD_READ_PPPSA(afu);
306 	show_reg("PerProcessPSA_control", (val >> (63-7)) & 0xff);
307 	show_reg("PerProcessPSA Length", AFUD_PPPSA_LEN(val));
308 
309 	val = AFUD_READ_PPPSA_OFF(afu);
310 	show_reg("PerProcessPSA_offset", val);
311 
312 	val = AFUD_READ_EB(afu);
313 	show_reg("Reserved", (val >> (63-7)) & 0xff);
314 	show_reg("AFU_EB_len", AFUD_EB_LEN(val));
315 
316 	val = AFUD_READ_EB_OFF(afu);
317 	show_reg("AFU_EB_offset", val);
318 
319 	for (i = 0; i < afu_cr_num; i++) {
320 		val = AFUD_READ_LE(afu, afu_cr_off + i * afu_cr_len);
321 		show_reg("CR Vendor", val & 0xffff);
322 		show_reg("CR Device", (val >> 16) & 0xffff);
323 	}
324 #undef show_reg
325 }
326 
327 #define CAPP_UNIT0_ID 0xBA
328 #define CAPP_UNIT1_ID 0XBE
329 
330 static u64 get_capp_unit_id(struct device_node *np)
331 {
332 	u32 phb_index;
333 
334 	/*
335 	 * For chips other than POWER8NVL, we only have CAPP 0,
336 	 * irrespective of which PHB is used.
337 	 */
338 	if (!pvr_version_is(PVR_POWER8NVL))
339 		return CAPP_UNIT0_ID;
340 
341 	/*
342 	 * For POWER8NVL, assume CAPP 0 is attached to PHB0 and
343 	 * CAPP 1 is attached to PHB1.
344 	 */
345 	if (of_property_read_u32(np, "ibm,phb-index", &phb_index))
346 		return 0;
347 
348 	if (phb_index == 0)
349 		return CAPP_UNIT0_ID;
350 
351 	if (phb_index == 1)
352 		return CAPP_UNIT1_ID;
353 
354 	return 0;
355 }
356 
357 static int calc_capp_routing(struct pci_dev *dev, u64 *chipid, u64 *capp_unit_id)
358 {
359 	struct device_node *np;
360 	const __be32 *prop;
361 
362 	if (!(np = pnv_pci_get_phb_node(dev)))
363 		return -ENODEV;
364 
365 	while (np && !(prop = of_get_property(np, "ibm,chip-id", NULL)))
366 		np = of_get_next_parent(np);
367 	if (!np)
368 		return -ENODEV;
369 	*chipid = be32_to_cpup(prop);
370 	*capp_unit_id = get_capp_unit_id(np);
371 	of_node_put(np);
372 	if (!*capp_unit_id) {
373 		pr_err("cxl: invalid capp unit id\n");
374 		return -ENODEV;
375 	}
376 
377 	return 0;
378 }
379 
380 static int init_implementation_adapter_psl_regs(struct cxl *adapter, struct pci_dev *dev)
381 {
382 	u64 psl_dsnctl, psl_fircntl;
383 	u64 chipid;
384 	u64 capp_unit_id;
385 	int rc;
386 
387 	rc = calc_capp_routing(dev, &chipid, &capp_unit_id);
388 	if (rc)
389 		return rc;
390 
391 	psl_dsnctl = 0x0000900000000000ULL; /* pteupd ttype, scdone */
392 	psl_dsnctl |= (0x2ULL << (63-38)); /* MMIO hang pulse: 256 us */
393 	/* Tell PSL where to route data to */
394 	psl_dsnctl |= (chipid << (63-5));
395 	psl_dsnctl |= (capp_unit_id << (63-13));
396 
397 	cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
398 	cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
399 	/* snoop write mask */
400 	cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
401 	/* set fir_cntl to recommended value for production env */
402 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
403 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
404 	psl_fircntl |= 0x1ULL; /* ce_thresh */
405 	cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, psl_fircntl);
406 	/* for debugging with trace arrays */
407 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
408 
409 	return 0;
410 }
411 
412 static int init_implementation_adapter_xsl_regs(struct cxl *adapter, struct pci_dev *dev)
413 {
414 	u64 xsl_dsnctl;
415 	u64 chipid;
416 	u64 capp_unit_id;
417 	int rc;
418 
419 	rc = calc_capp_routing(dev, &chipid, &capp_unit_id);
420 	if (rc)
421 		return rc;
422 
423 	/* Tell XSL where to route data to */
424 	xsl_dsnctl = 0x0000600000000000ULL | (chipid << (63-5));
425 	xsl_dsnctl |= (capp_unit_id << (63-13));
426 	cxl_p1_write(adapter, CXL_XSL_DSNCTL, xsl_dsnctl);
427 
428 	return 0;
429 }
430 
431 /* PSL & XSL */
432 #define TBSYNC_CAL(n) (((u64)n & 0x7) << (63-3))
433 #define TBSYNC_CNT(n) (((u64)n & 0x7) << (63-6))
434 /* For the PSL this is a multiple for 0 < n <= 7: */
435 #define PSL_2048_250MHZ_CYCLES 1
436 
437 static void write_timebase_ctrl_psl(struct cxl *adapter)
438 {
439 	cxl_p1_write(adapter, CXL_PSL_TB_CTLSTAT,
440 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
441 }
442 
443 /* XSL */
444 #define TBSYNC_ENA (1ULL << 63)
445 /* For the XSL this is 2**n * 2000 clocks for 0 < n <= 6: */
446 #define XSL_2000_CLOCKS 1
447 #define XSL_4000_CLOCKS 2
448 #define XSL_8000_CLOCKS 3
449 
450 static void write_timebase_ctrl_xsl(struct cxl *adapter)
451 {
452 	cxl_p1_write(adapter, CXL_XSL_TB_CTLSTAT,
453 		     TBSYNC_ENA |
454 		     TBSYNC_CAL(3) |
455 		     TBSYNC_CNT(XSL_4000_CLOCKS));
456 }
457 
458 static u64 timebase_read_psl(struct cxl *adapter)
459 {
460 	return cxl_p1_read(adapter, CXL_PSL_Timebase);
461 }
462 
463 static u64 timebase_read_xsl(struct cxl *adapter)
464 {
465 	return cxl_p1_read(adapter, CXL_XSL_Timebase);
466 }
467 
468 static void cxl_setup_psl_timebase(struct cxl *adapter, struct pci_dev *dev)
469 {
470 	u64 psl_tb;
471 	int delta;
472 	unsigned int retry = 0;
473 	struct device_node *np;
474 
475 	adapter->psl_timebase_synced = false;
476 
477 	if (!(np = pnv_pci_get_phb_node(dev)))
478 		return;
479 
480 	/* Do not fail when CAPP timebase sync is not supported by OPAL */
481 	of_node_get(np);
482 	if (! of_get_property(np, "ibm,capp-timebase-sync", NULL)) {
483 		of_node_put(np);
484 		dev_info(&dev->dev, "PSL timebase inactive: OPAL support missing\n");
485 		return;
486 	}
487 	of_node_put(np);
488 
489 	/*
490 	 * Setup PSL Timebase Control and Status register
491 	 * with the recommended Timebase Sync Count value
492 	 */
493 	adapter->native->sl_ops->write_timebase_ctrl(adapter);
494 
495 	/* Enable PSL Timebase */
496 	cxl_p1_write(adapter, CXL_PSL_Control, 0x0000000000000000);
497 	cxl_p1_write(adapter, CXL_PSL_Control, CXL_PSL_Control_tb);
498 
499 	/* Wait until CORE TB and PSL TB difference <= 16usecs */
500 	do {
501 		msleep(1);
502 		if (retry++ > 5) {
503 			dev_info(&dev->dev, "PSL timebase can't synchronize\n");
504 			return;
505 		}
506 		psl_tb = adapter->native->sl_ops->timebase_read(adapter);
507 		delta = mftb() - psl_tb;
508 		if (delta < 0)
509 			delta = -delta;
510 	} while (tb_to_ns(delta) > 16000);
511 
512 	adapter->psl_timebase_synced = true;
513 	return;
514 }
515 
516 static int init_implementation_afu_psl_regs(struct cxl_afu *afu)
517 {
518 	/* read/write masks for this slice */
519 	cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
520 	/* APC read/write masks for this slice */
521 	cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
522 	/* for debugging with trace arrays */
523 	cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
524 	cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
525 
526 	return 0;
527 }
528 
529 int cxl_pci_setup_irq(struct cxl *adapter, unsigned int hwirq,
530 		unsigned int virq)
531 {
532 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
533 
534 	return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
535 }
536 
537 int cxl_update_image_control(struct cxl *adapter)
538 {
539 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
540 	int rc;
541 	int vsec;
542 	u8 image_state;
543 
544 	if (!(vsec = find_cxl_vsec(dev))) {
545 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
546 		return -ENODEV;
547 	}
548 
549 	if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
550 		dev_err(&dev->dev, "failed to read image state: %i\n", rc);
551 		return rc;
552 	}
553 
554 	if (adapter->perst_loads_image)
555 		image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
556 	else
557 		image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
558 
559 	if (adapter->perst_select_user)
560 		image_state |= CXL_VSEC_PERST_SELECT_USER;
561 	else
562 		image_state &= ~CXL_VSEC_PERST_SELECT_USER;
563 
564 	if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
565 		dev_err(&dev->dev, "failed to update image control: %i\n", rc);
566 		return rc;
567 	}
568 
569 	return 0;
570 }
571 
572 int cxl_pci_alloc_one_irq(struct cxl *adapter)
573 {
574 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
575 
576 	return pnv_cxl_alloc_hwirqs(dev, 1);
577 }
578 
579 void cxl_pci_release_one_irq(struct cxl *adapter, int hwirq)
580 {
581 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
582 
583 	return pnv_cxl_release_hwirqs(dev, hwirq, 1);
584 }
585 
586 int cxl_pci_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
587 			struct cxl *adapter, unsigned int num)
588 {
589 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
590 
591 	return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
592 }
593 
594 void cxl_pci_release_irq_ranges(struct cxl_irq_ranges *irqs,
595 				struct cxl *adapter)
596 {
597 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
598 
599 	pnv_cxl_release_hwirq_ranges(irqs, dev);
600 }
601 
602 static int setup_cxl_bars(struct pci_dev *dev)
603 {
604 	/* Safety check in case we get backported to < 3.17 without M64 */
605 	if ((p1_base(dev) < 0x100000000ULL) ||
606 	    (p2_base(dev) < 0x100000000ULL)) {
607 		dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
608 		return -ENODEV;
609 	}
610 
611 	/*
612 	 * BAR 4/5 has a special meaning for CXL and must be programmed with a
613 	 * special value corresponding to the CXL protocol address range.
614 	 * For POWER 8 that means bits 48:49 must be set to 10
615 	 */
616 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
617 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
618 
619 	return 0;
620 }
621 
622 #ifdef CONFIG_CXL_BIMODAL
623 
624 struct cxl_switch_work {
625 	struct pci_dev *dev;
626 	struct work_struct work;
627 	int vsec;
628 	int mode;
629 };
630 
631 static void switch_card_to_cxl(struct work_struct *work)
632 {
633 	struct cxl_switch_work *switch_work =
634 		container_of(work, struct cxl_switch_work, work);
635 	struct pci_dev *dev = switch_work->dev;
636 	struct pci_bus *bus = dev->bus;
637 	struct pci_controller *hose = pci_bus_to_host(bus);
638 	struct pci_dev *bridge;
639 	struct pnv_php_slot *php_slot;
640 	unsigned int devfn;
641 	u8 val;
642 	int rc;
643 
644 	dev_info(&bus->dev, "cxl: Preparing for mode switch...\n");
645 	bridge = list_first_entry_or_null(&hose->bus->devices, struct pci_dev,
646 					  bus_list);
647 	if (!bridge) {
648 		dev_WARN(&bus->dev, "cxl: Couldn't find root port!\n");
649 		goto err_dev_put;
650 	}
651 
652 	php_slot = pnv_php_find_slot(pci_device_to_OF_node(bridge));
653 	if (!php_slot) {
654 		dev_err(&bus->dev, "cxl: Failed to find slot hotplug "
655 			           "information. You may need to upgrade "
656 			           "skiboot. Aborting.\n");
657 		goto err_dev_put;
658 	}
659 
660 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, switch_work->vsec, &val);
661 	if (rc) {
662 		dev_err(&bus->dev, "cxl: Failed to read CAPI mode control: %i\n", rc);
663 		goto err_dev_put;
664 	}
665 	devfn = dev->devfn;
666 
667 	/* Release the reference obtained in cxl_check_and_switch_mode() */
668 	pci_dev_put(dev);
669 
670 	dev_dbg(&bus->dev, "cxl: Removing PCI devices from kernel\n");
671 	pci_lock_rescan_remove();
672 	pci_hp_remove_devices(bridge->subordinate);
673 	pci_unlock_rescan_remove();
674 
675 	/* Switch the CXL protocol on the card */
676 	if (switch_work->mode == CXL_BIMODE_CXL) {
677 		dev_info(&bus->dev, "cxl: Switching card to CXL mode\n");
678 		val &= ~CXL_VSEC_PROTOCOL_MASK;
679 		val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
680 		rc = pnv_cxl_enable_phb_kernel_api(hose, true);
681 		if (rc) {
682 			dev_err(&bus->dev, "cxl: Failed to enable kernel API"
683 				           " on real PHB, aborting\n");
684 			goto err_free_work;
685 		}
686 	} else {
687 		dev_WARN(&bus->dev, "cxl: Switching card to PCI mode not supported!\n");
688 		goto err_free_work;
689 	}
690 
691 	rc = CXL_WRITE_VSEC_MODE_CONTROL_BUS(bus, devfn, switch_work->vsec, val);
692 	if (rc) {
693 		dev_err(&bus->dev, "cxl: Failed to configure CXL protocol: %i\n", rc);
694 		goto err_free_work;
695 	}
696 
697 	/*
698 	 * The CAIA spec (v1.1, Section 10.6 Bi-modal Device Support) states
699 	 * we must wait 100ms after this mode switch before touching PCIe config
700 	 * space.
701 	 */
702 	msleep(100);
703 
704 	/*
705 	 * Hot reset to cause the card to come back in cxl mode. A
706 	 * OPAL_RESET_PCI_LINK would be sufficient, but currently lacks support
707 	 * in skiboot, so we use a hot reset instead.
708 	 *
709 	 * We call pci_set_pcie_reset_state() on the bridge, as a CAPI card is
710 	 * guaranteed to sit directly under the root port, and setting the reset
711 	 * state on a device directly under the root port is equivalent to doing
712 	 * it on the root port iself.
713 	 */
714 	dev_info(&bus->dev, "cxl: Configuration write complete, resetting card\n");
715 	pci_set_pcie_reset_state(bridge, pcie_hot_reset);
716 	pci_set_pcie_reset_state(bridge, pcie_deassert_reset);
717 
718 	dev_dbg(&bus->dev, "cxl: Offlining slot\n");
719 	rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_OFFLINE);
720 	if (rc) {
721 		dev_err(&bus->dev, "cxl: OPAL offlining call failed: %i\n", rc);
722 		goto err_free_work;
723 	}
724 
725 	dev_dbg(&bus->dev, "cxl: Onlining and probing slot\n");
726 	rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_ONLINE);
727 	if (rc) {
728 		dev_err(&bus->dev, "cxl: OPAL onlining call failed: %i\n", rc);
729 		goto err_free_work;
730 	}
731 
732 	pci_lock_rescan_remove();
733 	pci_hp_add_devices(bridge->subordinate);
734 	pci_unlock_rescan_remove();
735 
736 	dev_info(&bus->dev, "cxl: CAPI mode switch completed\n");
737 	kfree(switch_work);
738 	return;
739 
740 err_dev_put:
741 	/* Release the reference obtained in cxl_check_and_switch_mode() */
742 	pci_dev_put(dev);
743 err_free_work:
744 	kfree(switch_work);
745 }
746 
747 int cxl_check_and_switch_mode(struct pci_dev *dev, int mode, int vsec)
748 {
749 	struct cxl_switch_work *work;
750 	u8 val;
751 	int rc;
752 
753 	if (!cpu_has_feature(CPU_FTR_HVMODE))
754 		return -ENODEV;
755 
756 	if (!vsec) {
757 		vsec = find_cxl_vsec(dev);
758 		if (!vsec) {
759 			dev_info(&dev->dev, "CXL VSEC not found\n");
760 			return -ENODEV;
761 		}
762 	}
763 
764 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
765 	if (rc) {
766 		dev_err(&dev->dev, "Failed to read current mode control: %i", rc);
767 		return rc;
768 	}
769 
770 	if (mode == CXL_BIMODE_PCI) {
771 		if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
772 			dev_info(&dev->dev, "Card is already in PCI mode\n");
773 			return 0;
774 		}
775 		/*
776 		 * TODO: Before it's safe to switch the card back to PCI mode
777 		 * we need to disable the CAPP and make sure any cachelines the
778 		 * card holds have been flushed out. Needs skiboot support.
779 		 */
780 		dev_WARN(&dev->dev, "CXL mode switch to PCI unsupported!\n");
781 		return -EIO;
782 	}
783 
784 	if (val & CXL_VSEC_PROTOCOL_ENABLE) {
785 		dev_info(&dev->dev, "Card is already in CXL mode\n");
786 		return 0;
787 	}
788 
789 	dev_info(&dev->dev, "Card is in PCI mode, scheduling kernel thread "
790 			    "to switch to CXL mode\n");
791 
792 	work = kmalloc(sizeof(struct cxl_switch_work), GFP_KERNEL);
793 	if (!work)
794 		return -ENOMEM;
795 
796 	pci_dev_get(dev);
797 	work->dev = dev;
798 	work->vsec = vsec;
799 	work->mode = mode;
800 	INIT_WORK(&work->work, switch_card_to_cxl);
801 
802 	schedule_work(&work->work);
803 
804 	/*
805 	 * We return a failure now to abort the driver init. Once the
806 	 * link has been cycled and the card is in cxl mode we will
807 	 * come back (possibly using the generic cxl driver), but
808 	 * return success as the card should then be in cxl mode.
809 	 *
810 	 * TODO: What if the card comes back in PCI mode even after
811 	 *       the switch?  Don't want to spin endlessly.
812 	 */
813 	return -EBUSY;
814 }
815 EXPORT_SYMBOL_GPL(cxl_check_and_switch_mode);
816 
817 #endif /* CONFIG_CXL_BIMODAL */
818 
819 static int setup_cxl_protocol_area(struct pci_dev *dev)
820 {
821 	u8 val;
822 	int rc;
823 	int vsec = find_cxl_vsec(dev);
824 
825 	if (!vsec) {
826 		dev_info(&dev->dev, "CXL VSEC not found\n");
827 		return -ENODEV;
828 	}
829 
830 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
831 	if (rc) {
832 		dev_err(&dev->dev, "Failed to read current mode control: %i\n", rc);
833 		return rc;
834 	}
835 
836 	if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
837 		dev_err(&dev->dev, "Card not in CAPI mode!\n");
838 		return -EIO;
839 	}
840 
841 	if ((val & CXL_VSEC_PROTOCOL_MASK) != CXL_VSEC_PROTOCOL_256TB) {
842 		val &= ~CXL_VSEC_PROTOCOL_MASK;
843 		val |= CXL_VSEC_PROTOCOL_256TB;
844 		rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val);
845 		if (rc) {
846 			dev_err(&dev->dev, "Failed to set CXL protocol area: %i\n", rc);
847 			return rc;
848 		}
849 	}
850 
851 	return 0;
852 }
853 
854 static int pci_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
855 {
856 	u64 p1n_base, p2n_base, afu_desc;
857 	const u64 p1n_size = 0x100;
858 	const u64 p2n_size = 0x1000;
859 
860 	p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
861 	p2n_base = p2_base(dev) + (afu->slice * p2n_size);
862 	afu->psn_phys = p2_base(dev) + (adapter->native->ps_off + (afu->slice * adapter->ps_size));
863 	afu_desc = p2_base(dev) + adapter->native->afu_desc_off + (afu->slice * adapter->native->afu_desc_size);
864 
865 	if (!(afu->native->p1n_mmio = ioremap(p1n_base, p1n_size)))
866 		goto err;
867 	if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
868 		goto err1;
869 	if (afu_desc) {
870 		if (!(afu->native->afu_desc_mmio = ioremap(afu_desc, adapter->native->afu_desc_size)))
871 			goto err2;
872 	}
873 
874 	return 0;
875 err2:
876 	iounmap(afu->p2n_mmio);
877 err1:
878 	iounmap(afu->native->p1n_mmio);
879 err:
880 	dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
881 	return -ENOMEM;
882 }
883 
884 static void pci_unmap_slice_regs(struct cxl_afu *afu)
885 {
886 	if (afu->p2n_mmio) {
887 		iounmap(afu->p2n_mmio);
888 		afu->p2n_mmio = NULL;
889 	}
890 	if (afu->native->p1n_mmio) {
891 		iounmap(afu->native->p1n_mmio);
892 		afu->native->p1n_mmio = NULL;
893 	}
894 	if (afu->native->afu_desc_mmio) {
895 		iounmap(afu->native->afu_desc_mmio);
896 		afu->native->afu_desc_mmio = NULL;
897 	}
898 }
899 
900 void cxl_pci_release_afu(struct device *dev)
901 {
902 	struct cxl_afu *afu = to_cxl_afu(dev);
903 
904 	pr_devel("%s\n", __func__);
905 
906 	idr_destroy(&afu->contexts_idr);
907 	cxl_release_spa(afu);
908 
909 	kfree(afu->native);
910 	kfree(afu);
911 }
912 
913 /* Expects AFU struct to have recently been zeroed out */
914 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
915 {
916 	u64 val;
917 
918 	val = AFUD_READ_INFO(afu);
919 	afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
920 	afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
921 	afu->crs_num = AFUD_NUM_CRS(val);
922 
923 	if (AFUD_AFU_DIRECTED(val))
924 		afu->modes_supported |= CXL_MODE_DIRECTED;
925 	if (AFUD_DEDICATED_PROCESS(val))
926 		afu->modes_supported |= CXL_MODE_DEDICATED;
927 	if (AFUD_TIME_SLICED(val))
928 		afu->modes_supported |= CXL_MODE_TIME_SLICED;
929 
930 	val = AFUD_READ_PPPSA(afu);
931 	afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
932 	afu->psa = AFUD_PPPSA_PSA(val);
933 	if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
934 		afu->native->pp_offset = AFUD_READ_PPPSA_OFF(afu);
935 
936 	val = AFUD_READ_CR(afu);
937 	afu->crs_len = AFUD_CR_LEN(val) * 256;
938 	afu->crs_offset = AFUD_READ_CR_OFF(afu);
939 
940 
941 	/* eb_len is in multiple of 4K */
942 	afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
943 	afu->eb_offset = AFUD_READ_EB_OFF(afu);
944 
945 	/* eb_off is 4K aligned so lower 12 bits are always zero */
946 	if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
947 		dev_warn(&afu->dev,
948 			 "Invalid AFU error buffer offset %Lx\n",
949 			 afu->eb_offset);
950 		dev_info(&afu->dev,
951 			 "Ignoring AFU error buffer in the descriptor\n");
952 		/* indicate that no afu buffer exists */
953 		afu->eb_len = 0;
954 	}
955 
956 	return 0;
957 }
958 
959 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
960 {
961 	int i, rc;
962 	u32 val;
963 
964 	if (afu->psa && afu->adapter->ps_size <
965 			(afu->native->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
966 		dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
967 		return -ENODEV;
968 	}
969 
970 	if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
971 		dev_warn(&afu->dev, "AFU uses < PAGE_SIZE per-process PSA!");
972 
973 	for (i = 0; i < afu->crs_num; i++) {
974 		rc = cxl_ops->afu_cr_read32(afu, i, 0, &val);
975 		if (rc || val == 0) {
976 			dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
977 			return -EINVAL;
978 		}
979 	}
980 
981 	if ((afu->modes_supported & ~CXL_MODE_DEDICATED) && afu->max_procs_virtualised == 0) {
982 		/*
983 		 * We could also check this for the dedicated process model
984 		 * since the architecture indicates it should be set to 1, but
985 		 * in that case we ignore the value and I'd rather not risk
986 		 * breaking any existing dedicated process AFUs that left it as
987 		 * 0 (not that I'm aware of any). It is clearly an error for an
988 		 * AFU directed AFU to set this to 0, and would have previously
989 		 * triggered a bug resulting in the maximum not being enforced
990 		 * at all since idr_alloc treats 0 as no maximum.
991 		 */
992 		dev_err(&afu->dev, "AFU does not support any processes\n");
993 		return -EINVAL;
994 	}
995 
996 	return 0;
997 }
998 
999 static int sanitise_afu_regs(struct cxl_afu *afu)
1000 {
1001 	u64 reg;
1002 
1003 	/*
1004 	 * Clear out any regs that contain either an IVTE or address or may be
1005 	 * waiting on an acknowledgement to try to be a bit safer as we bring
1006 	 * it online
1007 	 */
1008 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
1009 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
1010 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
1011 		if (cxl_ops->afu_reset(afu))
1012 			return -EIO;
1013 		if (cxl_afu_disable(afu))
1014 			return -EIO;
1015 		if (cxl_psl_purge(afu))
1016 			return -EIO;
1017 	}
1018 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
1019 	cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
1020 	cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
1021 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1022 	cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
1023 	cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
1024 	cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
1025 	cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
1026 	cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
1027 	cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
1028 	cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
1029 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1030 	if (reg) {
1031 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1032 		if (reg & CXL_PSL_DSISR_TRANS)
1033 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1034 		else
1035 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1036 	}
1037 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1038 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1039 		if (reg) {
1040 			if (reg & ~0xffff)
1041 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1042 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1043 		}
1044 	}
1045 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1046 	if (reg) {
1047 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1048 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1049 	}
1050 
1051 	return 0;
1052 }
1053 
1054 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
1055 /*
1056  * afu_eb_read:
1057  * Called from sysfs and reads the afu error info buffer. The h/w only supports
1058  * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
1059  * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
1060  */
1061 ssize_t cxl_pci_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
1062 				loff_t off, size_t count)
1063 {
1064 	loff_t aligned_start, aligned_end;
1065 	size_t aligned_length;
1066 	void *tbuf;
1067 	const void __iomem *ebuf = afu->native->afu_desc_mmio + afu->eb_offset;
1068 
1069 	if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
1070 		return 0;
1071 
1072 	/* calculate aligned read window */
1073 	count = min((size_t)(afu->eb_len - off), count);
1074 	aligned_start = round_down(off, 8);
1075 	aligned_end = round_up(off + count, 8);
1076 	aligned_length = aligned_end - aligned_start;
1077 
1078 	/* max we can copy in one read is PAGE_SIZE */
1079 	if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
1080 		aligned_length = ERR_BUFF_MAX_COPY_SIZE;
1081 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
1082 	}
1083 
1084 	/* use bounce buffer for copy */
1085 	tbuf = (void *)__get_free_page(GFP_TEMPORARY);
1086 	if (!tbuf)
1087 		return -ENOMEM;
1088 
1089 	/* perform aligned read from the mmio region */
1090 	memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
1091 	memcpy(buf, tbuf + (off & 0x7), count);
1092 
1093 	free_page((unsigned long)tbuf);
1094 
1095 	return count;
1096 }
1097 
1098 static int pci_configure_afu(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1099 {
1100 	int rc;
1101 
1102 	if ((rc = pci_map_slice_regs(afu, adapter, dev)))
1103 		return rc;
1104 
1105 	if ((rc = sanitise_afu_regs(afu)))
1106 		goto err1;
1107 
1108 	/* We need to reset the AFU before we can read the AFU descriptor */
1109 	if ((rc = cxl_ops->afu_reset(afu)))
1110 		goto err1;
1111 
1112 	if (cxl_verbose)
1113 		dump_afu_descriptor(afu);
1114 
1115 	if ((rc = cxl_read_afu_descriptor(afu)))
1116 		goto err1;
1117 
1118 	if ((rc = cxl_afu_descriptor_looks_ok(afu)))
1119 		goto err1;
1120 
1121 	if (adapter->native->sl_ops->afu_regs_init)
1122 		if ((rc = adapter->native->sl_ops->afu_regs_init(afu)))
1123 			goto err1;
1124 
1125 	if (adapter->native->sl_ops->register_serr_irq)
1126 		if ((rc = adapter->native->sl_ops->register_serr_irq(afu)))
1127 			goto err1;
1128 
1129 	if ((rc = cxl_native_register_psl_irq(afu)))
1130 		goto err2;
1131 
1132 	return 0;
1133 
1134 err2:
1135 	if (adapter->native->sl_ops->release_serr_irq)
1136 		adapter->native->sl_ops->release_serr_irq(afu);
1137 err1:
1138 	pci_unmap_slice_regs(afu);
1139 	return rc;
1140 }
1141 
1142 static void pci_deconfigure_afu(struct cxl_afu *afu)
1143 {
1144 	cxl_native_release_psl_irq(afu);
1145 	if (afu->adapter->native->sl_ops->release_serr_irq)
1146 		afu->adapter->native->sl_ops->release_serr_irq(afu);
1147 	pci_unmap_slice_regs(afu);
1148 }
1149 
1150 static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
1151 {
1152 	struct cxl_afu *afu;
1153 	int rc = -ENOMEM;
1154 
1155 	afu = cxl_alloc_afu(adapter, slice);
1156 	if (!afu)
1157 		return -ENOMEM;
1158 
1159 	afu->native = kzalloc(sizeof(struct cxl_afu_native), GFP_KERNEL);
1160 	if (!afu->native)
1161 		goto err_free_afu;
1162 
1163 	mutex_init(&afu->native->spa_mutex);
1164 
1165 	rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice);
1166 	if (rc)
1167 		goto err_free_native;
1168 
1169 	rc = pci_configure_afu(afu, adapter, dev);
1170 	if (rc)
1171 		goto err_free_native;
1172 
1173 	/* Don't care if this fails */
1174 	cxl_debugfs_afu_add(afu);
1175 
1176 	/*
1177 	 * After we call this function we must not free the afu directly, even
1178 	 * if it returns an error!
1179 	 */
1180 	if ((rc = cxl_register_afu(afu)))
1181 		goto err_put1;
1182 
1183 	if ((rc = cxl_sysfs_afu_add(afu)))
1184 		goto err_put1;
1185 
1186 	adapter->afu[afu->slice] = afu;
1187 
1188 	if ((rc = cxl_pci_vphb_add(afu)))
1189 		dev_info(&afu->dev, "Can't register vPHB\n");
1190 
1191 	return 0;
1192 
1193 err_put1:
1194 	pci_deconfigure_afu(afu);
1195 	cxl_debugfs_afu_remove(afu);
1196 	device_unregister(&afu->dev);
1197 	return rc;
1198 
1199 err_free_native:
1200 	kfree(afu->native);
1201 err_free_afu:
1202 	kfree(afu);
1203 	return rc;
1204 
1205 }
1206 
1207 static void cxl_pci_remove_afu(struct cxl_afu *afu)
1208 {
1209 	pr_devel("%s\n", __func__);
1210 
1211 	if (!afu)
1212 		return;
1213 
1214 	cxl_pci_vphb_remove(afu);
1215 	cxl_sysfs_afu_remove(afu);
1216 	cxl_debugfs_afu_remove(afu);
1217 
1218 	spin_lock(&afu->adapter->afu_list_lock);
1219 	afu->adapter->afu[afu->slice] = NULL;
1220 	spin_unlock(&afu->adapter->afu_list_lock);
1221 
1222 	cxl_context_detach_all(afu);
1223 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1224 
1225 	pci_deconfigure_afu(afu);
1226 	device_unregister(&afu->dev);
1227 }
1228 
1229 int cxl_pci_reset(struct cxl *adapter)
1230 {
1231 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1232 	int rc;
1233 
1234 	if (adapter->perst_same_image) {
1235 		dev_warn(&dev->dev,
1236 			 "cxl: refusing to reset/reflash when perst_reloads_same_image is set.\n");
1237 		return -EINVAL;
1238 	}
1239 
1240 	dev_info(&dev->dev, "CXL reset\n");
1241 
1242 	/* the adapter is about to be reset, so ignore errors */
1243 	cxl_data_cache_flush(adapter);
1244 
1245 	/* pcie_warm_reset requests a fundamental pci reset which includes a
1246 	 * PERST assert/deassert.  PERST triggers a loading of the image
1247 	 * if "user" or "factory" is selected in sysfs */
1248 	if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
1249 		dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
1250 		return rc;
1251 	}
1252 
1253 	return rc;
1254 }
1255 
1256 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
1257 {
1258 	if (pci_request_region(dev, 2, "priv 2 regs"))
1259 		goto err1;
1260 	if (pci_request_region(dev, 0, "priv 1 regs"))
1261 		goto err2;
1262 
1263 	pr_devel("cxl_map_adapter_regs: p1: %#016llx %#llx, p2: %#016llx %#llx",
1264 			p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
1265 
1266 	if (!(adapter->native->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
1267 		goto err3;
1268 
1269 	if (!(adapter->native->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
1270 		goto err4;
1271 
1272 	return 0;
1273 
1274 err4:
1275 	iounmap(adapter->native->p1_mmio);
1276 	adapter->native->p1_mmio = NULL;
1277 err3:
1278 	pci_release_region(dev, 0);
1279 err2:
1280 	pci_release_region(dev, 2);
1281 err1:
1282 	return -ENOMEM;
1283 }
1284 
1285 static void cxl_unmap_adapter_regs(struct cxl *adapter)
1286 {
1287 	if (adapter->native->p1_mmio) {
1288 		iounmap(adapter->native->p1_mmio);
1289 		adapter->native->p1_mmio = NULL;
1290 		pci_release_region(to_pci_dev(adapter->dev.parent), 2);
1291 	}
1292 	if (adapter->native->p2_mmio) {
1293 		iounmap(adapter->native->p2_mmio);
1294 		adapter->native->p2_mmio = NULL;
1295 		pci_release_region(to_pci_dev(adapter->dev.parent), 0);
1296 	}
1297 }
1298 
1299 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
1300 {
1301 	int vsec;
1302 	u32 afu_desc_off, afu_desc_size;
1303 	u32 ps_off, ps_size;
1304 	u16 vseclen;
1305 	u8 image_state;
1306 
1307 	if (!(vsec = find_cxl_vsec(dev))) {
1308 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
1309 		return -ENODEV;
1310 	}
1311 
1312 	CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
1313 	if (vseclen < CXL_VSEC_MIN_SIZE) {
1314 		dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
1315 		return -EINVAL;
1316 	}
1317 
1318 	CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
1319 	CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
1320 	CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
1321 	CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
1322 	CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
1323 	CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
1324 	adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1325 	adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1326 
1327 	CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
1328 	CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
1329 	CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
1330 	CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
1331 	CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
1332 
1333 	/* Convert everything to bytes, because there is NO WAY I'd look at the
1334 	 * code a month later and forget what units these are in ;-) */
1335 	adapter->native->ps_off = ps_off * 64 * 1024;
1336 	adapter->ps_size = ps_size * 64 * 1024;
1337 	adapter->native->afu_desc_off = afu_desc_off * 64 * 1024;
1338 	adapter->native->afu_desc_size = afu_desc_size * 64 * 1024;
1339 
1340 	/* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
1341 	adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
1342 
1343 	return 0;
1344 }
1345 
1346 /*
1347  * Workaround a PCIe Host Bridge defect on some cards, that can cause
1348  * malformed Transaction Layer Packet (TLP) errors to be erroneously
1349  * reported. Mask this error in the Uncorrectable Error Mask Register.
1350  *
1351  * The upper nibble of the PSL revision is used to distinguish between
1352  * different cards. The affected ones have it set to 0.
1353  */
1354 static void cxl_fixup_malformed_tlp(struct cxl *adapter, struct pci_dev *dev)
1355 {
1356 	int aer;
1357 	u32 data;
1358 
1359 	if (adapter->psl_rev & 0xf000)
1360 		return;
1361 	if (!(aer = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR)))
1362 		return;
1363 	pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &data);
1364 	if (data & PCI_ERR_UNC_MALF_TLP)
1365 		if (data & PCI_ERR_UNC_INTN)
1366 			return;
1367 	data |= PCI_ERR_UNC_MALF_TLP;
1368 	data |= PCI_ERR_UNC_INTN;
1369 	pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, data);
1370 }
1371 
1372 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
1373 {
1374 	if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
1375 		return -EBUSY;
1376 
1377 	if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
1378 		dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
1379 		return -EINVAL;
1380 	}
1381 
1382 	if (!adapter->slices) {
1383 		/* Once we support dynamic reprogramming we can use the card if
1384 		 * it supports loadable AFUs */
1385 		dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
1386 		return -EINVAL;
1387 	}
1388 
1389 	if (!adapter->native->afu_desc_off || !adapter->native->afu_desc_size) {
1390 		dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
1391 		return -EINVAL;
1392 	}
1393 
1394 	if (adapter->ps_size > p2_size(dev) - adapter->native->ps_off) {
1395 		dev_err(&dev->dev, "ABORTING: Problem state size larger than "
1396 				   "available in BAR2: 0x%llx > 0x%llx\n",
1397 			 adapter->ps_size, p2_size(dev) - adapter->native->ps_off);
1398 		return -EINVAL;
1399 	}
1400 
1401 	return 0;
1402 }
1403 
1404 ssize_t cxl_pci_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1405 {
1406 	return pci_read_vpd(to_pci_dev(adapter->dev.parent), 0, len, buf);
1407 }
1408 
1409 static void cxl_release_adapter(struct device *dev)
1410 {
1411 	struct cxl *adapter = to_cxl_adapter(dev);
1412 
1413 	pr_devel("cxl_release_adapter\n");
1414 
1415 	cxl_remove_adapter_nr(adapter);
1416 
1417 	kfree(adapter->native);
1418 	kfree(adapter);
1419 }
1420 
1421 #define CXL_PSL_ErrIVTE_tberror (0x1ull << (63-31))
1422 
1423 static int sanitise_adapter_regs(struct cxl *adapter)
1424 {
1425 	/* Clear PSL tberror bit by writing 1 to it */
1426 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, CXL_PSL_ErrIVTE_tberror);
1427 	return cxl_tlb_slb_invalidate(adapter);
1428 }
1429 
1430 /* This should contain *only* operations that can safely be done in
1431  * both creation and recovery.
1432  */
1433 static int cxl_configure_adapter(struct cxl *adapter, struct pci_dev *dev)
1434 {
1435 	int rc;
1436 
1437 	adapter->dev.parent = &dev->dev;
1438 	adapter->dev.release = cxl_release_adapter;
1439 	pci_set_drvdata(dev, adapter);
1440 
1441 	rc = pci_enable_device(dev);
1442 	if (rc) {
1443 		dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1444 		return rc;
1445 	}
1446 
1447 	if ((rc = cxl_read_vsec(adapter, dev)))
1448 		return rc;
1449 
1450 	if ((rc = cxl_vsec_looks_ok(adapter, dev)))
1451 	        return rc;
1452 
1453 	cxl_fixup_malformed_tlp(adapter, dev);
1454 
1455 	if ((rc = setup_cxl_bars(dev)))
1456 		return rc;
1457 
1458 	if ((rc = setup_cxl_protocol_area(dev)))
1459 		return rc;
1460 
1461 	if ((rc = cxl_update_image_control(adapter)))
1462 		return rc;
1463 
1464 	if ((rc = cxl_map_adapter_regs(adapter, dev)))
1465 		return rc;
1466 
1467 	if ((rc = sanitise_adapter_regs(adapter)))
1468 		goto err;
1469 
1470 	if ((rc = adapter->native->sl_ops->adapter_regs_init(adapter, dev)))
1471 		goto err;
1472 
1473 	/* Required for devices using CAPP DMA mode, harmless for others */
1474 	pci_set_master(dev);
1475 
1476 	if ((rc = pnv_phb_to_cxl_mode(dev, adapter->native->sl_ops->capi_mode)))
1477 		goto err;
1478 
1479 	/* If recovery happened, the last step is to turn on snooping.
1480 	 * In the non-recovery case this has no effect */
1481 	if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON)))
1482 		goto err;
1483 
1484 	/* Ignore error, adapter init is not dependant on timebase sync */
1485 	cxl_setup_psl_timebase(adapter, dev);
1486 
1487 	if ((rc = cxl_native_register_psl_err_irq(adapter)))
1488 		goto err;
1489 
1490 	/* Release the context lock as adapter is configured */
1491 	cxl_adapter_context_unlock(adapter);
1492 	return 0;
1493 
1494 err:
1495 	cxl_unmap_adapter_regs(adapter);
1496 	return rc;
1497 
1498 }
1499 
1500 static void cxl_deconfigure_adapter(struct cxl *adapter)
1501 {
1502 	struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1503 
1504 	cxl_native_release_psl_err_irq(adapter);
1505 	cxl_unmap_adapter_regs(adapter);
1506 
1507 	pci_disable_device(pdev);
1508 }
1509 
1510 static const struct cxl_service_layer_ops psl_ops = {
1511 	.adapter_regs_init = init_implementation_adapter_psl_regs,
1512 	.afu_regs_init = init_implementation_afu_psl_regs,
1513 	.register_serr_irq = cxl_native_register_serr_irq,
1514 	.release_serr_irq = cxl_native_release_serr_irq,
1515 	.debugfs_add_adapter_sl_regs = cxl_debugfs_add_adapter_psl_regs,
1516 	.debugfs_add_afu_sl_regs = cxl_debugfs_add_afu_psl_regs,
1517 	.psl_irq_dump_registers = cxl_native_psl_irq_dump_regs,
1518 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs,
1519 	.debugfs_stop_trace = cxl_stop_trace,
1520 	.write_timebase_ctrl = write_timebase_ctrl_psl,
1521 	.timebase_read = timebase_read_psl,
1522 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1523 	.needs_reset_before_disable = true,
1524 };
1525 
1526 static const struct cxl_service_layer_ops xsl_ops = {
1527 	.adapter_regs_init = init_implementation_adapter_xsl_regs,
1528 	.debugfs_add_adapter_sl_regs = cxl_debugfs_add_adapter_xsl_regs,
1529 	.write_timebase_ctrl = write_timebase_ctrl_xsl,
1530 	.timebase_read = timebase_read_xsl,
1531 	.capi_mode = OPAL_PHB_CAPI_MODE_DMA,
1532 };
1533 
1534 static void set_sl_ops(struct cxl *adapter, struct pci_dev *dev)
1535 {
1536 	if (dev->vendor == PCI_VENDOR_ID_MELLANOX && dev->device == 0x1013) {
1537 		/* Mellanox CX-4 */
1538 		dev_info(&dev->dev, "Device uses an XSL\n");
1539 		adapter->native->sl_ops = &xsl_ops;
1540 		adapter->min_pe = 1; /* Workaround for CX-4 hardware bug */
1541 	} else {
1542 		dev_info(&dev->dev, "Device uses a PSL\n");
1543 		adapter->native->sl_ops = &psl_ops;
1544 	}
1545 }
1546 
1547 
1548 static struct cxl *cxl_pci_init_adapter(struct pci_dev *dev)
1549 {
1550 	struct cxl *adapter;
1551 	int rc;
1552 
1553 	adapter = cxl_alloc_adapter();
1554 	if (!adapter)
1555 		return ERR_PTR(-ENOMEM);
1556 
1557 	adapter->native = kzalloc(sizeof(struct cxl_native), GFP_KERNEL);
1558 	if (!adapter->native) {
1559 		rc = -ENOMEM;
1560 		goto err_release;
1561 	}
1562 
1563 	set_sl_ops(adapter, dev);
1564 
1565 	/* Set defaults for parameters which need to persist over
1566 	 * configure/reconfigure
1567 	 */
1568 	adapter->perst_loads_image = true;
1569 	adapter->perst_same_image = false;
1570 
1571 	rc = cxl_configure_adapter(adapter, dev);
1572 	if (rc) {
1573 		pci_disable_device(dev);
1574 		goto err_release;
1575 	}
1576 
1577 	/* Don't care if this one fails: */
1578 	cxl_debugfs_adapter_add(adapter);
1579 
1580 	/*
1581 	 * After we call this function we must not free the adapter directly,
1582 	 * even if it returns an error!
1583 	 */
1584 	if ((rc = cxl_register_adapter(adapter)))
1585 		goto err_put1;
1586 
1587 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1588 		goto err_put1;
1589 
1590 	return adapter;
1591 
1592 err_put1:
1593 	/* This should mirror cxl_remove_adapter, except without the
1594 	 * sysfs parts
1595 	 */
1596 	cxl_debugfs_adapter_remove(adapter);
1597 	cxl_deconfigure_adapter(adapter);
1598 	device_unregister(&adapter->dev);
1599 	return ERR_PTR(rc);
1600 
1601 err_release:
1602 	cxl_release_adapter(&adapter->dev);
1603 	return ERR_PTR(rc);
1604 }
1605 
1606 static void cxl_pci_remove_adapter(struct cxl *adapter)
1607 {
1608 	pr_devel("cxl_remove_adapter\n");
1609 
1610 	cxl_sysfs_adapter_remove(adapter);
1611 	cxl_debugfs_adapter_remove(adapter);
1612 
1613 	cxl_deconfigure_adapter(adapter);
1614 
1615 	device_unregister(&adapter->dev);
1616 }
1617 
1618 #define CXL_MAX_PCIEX_PARENT 2
1619 
1620 static int cxl_slot_is_switched(struct pci_dev *dev)
1621 {
1622 	struct device_node *np;
1623 	int depth = 0;
1624 	const __be32 *prop;
1625 
1626 	if (!(np = pci_device_to_OF_node(dev))) {
1627 		pr_err("cxl: np = NULL\n");
1628 		return -ENODEV;
1629 	}
1630 	of_node_get(np);
1631 	while (np) {
1632 		np = of_get_next_parent(np);
1633 		prop = of_get_property(np, "device_type", NULL);
1634 		if (!prop || strcmp((char *)prop, "pciex"))
1635 			break;
1636 		depth++;
1637 	}
1638 	of_node_put(np);
1639 	return (depth > CXL_MAX_PCIEX_PARENT);
1640 }
1641 
1642 bool cxl_slot_is_supported(struct pci_dev *dev, int flags)
1643 {
1644 	if (!cpu_has_feature(CPU_FTR_HVMODE))
1645 		return false;
1646 
1647 	if ((flags & CXL_SLOT_FLAG_DMA) && (!pvr_version_is(PVR_POWER8NVL))) {
1648 		/*
1649 		 * CAPP DMA mode is technically supported on regular P8, but
1650 		 * will EEH if the card attempts to access memory < 4GB, which
1651 		 * we cannot realistically avoid. We might be able to work
1652 		 * around the issue, but until then return unsupported:
1653 		 */
1654 		return false;
1655 	}
1656 
1657 	if (cxl_slot_is_switched(dev))
1658 		return false;
1659 
1660 	/*
1661 	 * XXX: This gets a little tricky on regular P8 (not POWER8NVL) since
1662 	 * the CAPP can be connected to PHB 0, 1 or 2 on a first come first
1663 	 * served basis, which is racy to check from here. If we need to
1664 	 * support this in future we might need to consider having this
1665 	 * function effectively reserve it ahead of time.
1666 	 *
1667 	 * Currently, the only user of this API is the Mellanox CX4, which is
1668 	 * only supported on P8NVL due to the above mentioned limitation of
1669 	 * CAPP DMA mode and therefore does not need to worry about this. If the
1670 	 * issue with CAPP DMA mode is later worked around on P8 we might need
1671 	 * to revisit this.
1672 	 */
1673 
1674 	return true;
1675 }
1676 EXPORT_SYMBOL_GPL(cxl_slot_is_supported);
1677 
1678 
1679 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
1680 {
1681 	struct cxl *adapter;
1682 	int slice;
1683 	int rc;
1684 
1685 	if (cxl_pci_is_vphb_device(dev)) {
1686 		dev_dbg(&dev->dev, "cxl_init_adapter: Ignoring cxl vphb device\n");
1687 		return -ENODEV;
1688 	}
1689 
1690 	if (cxl_slot_is_switched(dev)) {
1691 		dev_info(&dev->dev, "Ignoring card on incompatible PCI slot\n");
1692 		return -ENODEV;
1693 	}
1694 
1695 	if (cxl_verbose)
1696 		dump_cxl_config_space(dev);
1697 
1698 	adapter = cxl_pci_init_adapter(dev);
1699 	if (IS_ERR(adapter)) {
1700 		dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
1701 		return PTR_ERR(adapter);
1702 	}
1703 
1704 	for (slice = 0; slice < adapter->slices; slice++) {
1705 		if ((rc = pci_init_afu(adapter, slice, dev))) {
1706 			dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
1707 			continue;
1708 		}
1709 
1710 		rc = cxl_afu_select_best_mode(adapter->afu[slice]);
1711 		if (rc)
1712 			dev_err(&dev->dev, "AFU %i failed to start: %i\n", slice, rc);
1713 	}
1714 
1715 	if (pnv_pci_on_cxl_phb(dev) && adapter->slices >= 1)
1716 		pnv_cxl_phb_set_peer_afu(dev, adapter->afu[0]);
1717 
1718 	return 0;
1719 }
1720 
1721 static void cxl_remove(struct pci_dev *dev)
1722 {
1723 	struct cxl *adapter = pci_get_drvdata(dev);
1724 	struct cxl_afu *afu;
1725 	int i;
1726 
1727 	/*
1728 	 * Lock to prevent someone grabbing a ref through the adapter list as
1729 	 * we are removing it
1730 	 */
1731 	for (i = 0; i < adapter->slices; i++) {
1732 		afu = adapter->afu[i];
1733 		cxl_pci_remove_afu(afu);
1734 	}
1735 	cxl_pci_remove_adapter(adapter);
1736 }
1737 
1738 static pci_ers_result_t cxl_vphb_error_detected(struct cxl_afu *afu,
1739 						pci_channel_state_t state)
1740 {
1741 	struct pci_dev *afu_dev;
1742 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
1743 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
1744 
1745 	/* There should only be one entry, but go through the list
1746 	 * anyway
1747 	 */
1748 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
1749 		if (!afu_dev->driver)
1750 			continue;
1751 
1752 		afu_dev->error_state = state;
1753 
1754 		if (afu_dev->driver->err_handler)
1755 			afu_result = afu_dev->driver->err_handler->error_detected(afu_dev,
1756 										  state);
1757 		/* Disconnect trumps all, NONE trumps NEED_RESET */
1758 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
1759 			result = PCI_ERS_RESULT_DISCONNECT;
1760 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
1761 			 (result == PCI_ERS_RESULT_NEED_RESET))
1762 			result = PCI_ERS_RESULT_NONE;
1763 	}
1764 	return result;
1765 }
1766 
1767 static pci_ers_result_t cxl_pci_error_detected(struct pci_dev *pdev,
1768 					       pci_channel_state_t state)
1769 {
1770 	struct cxl *adapter = pci_get_drvdata(pdev);
1771 	struct cxl_afu *afu;
1772 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
1773 	int i;
1774 
1775 	/* At this point, we could still have an interrupt pending.
1776 	 * Let's try to get them out of the way before they do
1777 	 * anything we don't like.
1778 	 */
1779 	schedule();
1780 
1781 	/* If we're permanently dead, give up. */
1782 	if (state == pci_channel_io_perm_failure) {
1783 		/* Tell the AFU drivers; but we don't care what they
1784 		 * say, we're going away.
1785 		 */
1786 		for (i = 0; i < adapter->slices; i++) {
1787 			afu = adapter->afu[i];
1788 			/* Only participate in EEH if we are on a virtual PHB */
1789 			if (afu->phb == NULL)
1790 				return PCI_ERS_RESULT_NONE;
1791 			cxl_vphb_error_detected(afu, state);
1792 		}
1793 		return PCI_ERS_RESULT_DISCONNECT;
1794 	}
1795 
1796 	/* Are we reflashing?
1797 	 *
1798 	 * If we reflash, we could come back as something entirely
1799 	 * different, including a non-CAPI card. As such, by default
1800 	 * we don't participate in the process. We'll be unbound and
1801 	 * the slot re-probed. (TODO: check EEH doesn't blindly rebind
1802 	 * us!)
1803 	 *
1804 	 * However, this isn't the entire story: for reliablity
1805 	 * reasons, we usually want to reflash the FPGA on PERST in
1806 	 * order to get back to a more reliable known-good state.
1807 	 *
1808 	 * This causes us a bit of a problem: if we reflash we can't
1809 	 * trust that we'll come back the same - we could have a new
1810 	 * image and been PERSTed in order to load that
1811 	 * image. However, most of the time we actually *will* come
1812 	 * back the same - for example a regular EEH event.
1813 	 *
1814 	 * Therefore, we allow the user to assert that the image is
1815 	 * indeed the same and that we should continue on into EEH
1816 	 * anyway.
1817 	 */
1818 	if (adapter->perst_loads_image && !adapter->perst_same_image) {
1819 		/* TODO take the PHB out of CXL mode */
1820 		dev_info(&pdev->dev, "reflashing, so opting out of EEH!\n");
1821 		return PCI_ERS_RESULT_NONE;
1822 	}
1823 
1824 	/*
1825 	 * At this point, we want to try to recover.  We'll always
1826 	 * need a complete slot reset: we don't trust any other reset.
1827 	 *
1828 	 * Now, we go through each AFU:
1829 	 *  - We send the driver, if bound, an error_detected callback.
1830 	 *    We expect it to clean up, but it can also tell us to give
1831 	 *    up and permanently detach the card. To simplify things, if
1832 	 *    any bound AFU driver doesn't support EEH, we give up on EEH.
1833 	 *
1834 	 *  - We detach all contexts associated with the AFU. This
1835 	 *    does not free them, but puts them into a CLOSED state
1836 	 *    which causes any the associated files to return useful
1837 	 *    errors to userland. It also unmaps, but does not free,
1838 	 *    any IRQs.
1839 	 *
1840 	 *  - We clean up our side: releasing and unmapping resources we hold
1841 	 *    so we can wire them up again when the hardware comes back up.
1842 	 *
1843 	 * Driver authors should note:
1844 	 *
1845 	 *  - Any contexts you create in your kernel driver (except
1846 	 *    those associated with anonymous file descriptors) are
1847 	 *    your responsibility to free and recreate. Likewise with
1848 	 *    any attached resources.
1849 	 *
1850 	 *  - We will take responsibility for re-initialising the
1851 	 *    device context (the one set up for you in
1852 	 *    cxl_pci_enable_device_hook and accessed through
1853 	 *    cxl_get_context). If you've attached IRQs or other
1854 	 *    resources to it, they remains yours to free.
1855 	 *
1856 	 * You can call the same functions to release resources as you
1857 	 * normally would: we make sure that these functions continue
1858 	 * to work when the hardware is down.
1859 	 *
1860 	 * Two examples:
1861 	 *
1862 	 * 1) If you normally free all your resources at the end of
1863 	 *    each request, or if you use anonymous FDs, your
1864 	 *    error_detected callback can simply set a flag to tell
1865 	 *    your driver not to start any new calls. You can then
1866 	 *    clear the flag in the resume callback.
1867 	 *
1868 	 * 2) If you normally allocate your resources on startup:
1869 	 *     * Set a flag in error_detected as above.
1870 	 *     * Let CXL detach your contexts.
1871 	 *     * In slot_reset, free the old resources and allocate new ones.
1872 	 *     * In resume, clear the flag to allow things to start.
1873 	 */
1874 	for (i = 0; i < adapter->slices; i++) {
1875 		afu = adapter->afu[i];
1876 
1877 		result = cxl_vphb_error_detected(afu, state);
1878 
1879 		/* Only continue if everyone agrees on NEED_RESET */
1880 		if (result != PCI_ERS_RESULT_NEED_RESET)
1881 			return result;
1882 
1883 		cxl_context_detach_all(afu);
1884 		cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1885 		pci_deconfigure_afu(afu);
1886 	}
1887 	cxl_deconfigure_adapter(adapter);
1888 
1889 	return result;
1890 }
1891 
1892 static pci_ers_result_t cxl_pci_slot_reset(struct pci_dev *pdev)
1893 {
1894 	struct cxl *adapter = pci_get_drvdata(pdev);
1895 	struct cxl_afu *afu;
1896 	struct cxl_context *ctx;
1897 	struct pci_dev *afu_dev;
1898 	pci_ers_result_t afu_result = PCI_ERS_RESULT_RECOVERED;
1899 	pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
1900 	int i;
1901 
1902 	if (cxl_configure_adapter(adapter, pdev))
1903 		goto err;
1904 
1905 	for (i = 0; i < adapter->slices; i++) {
1906 		afu = adapter->afu[i];
1907 
1908 		if (pci_configure_afu(afu, adapter, pdev))
1909 			goto err;
1910 
1911 		if (cxl_afu_select_best_mode(afu))
1912 			goto err;
1913 
1914 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
1915 			/* Reset the device context.
1916 			 * TODO: make this less disruptive
1917 			 */
1918 			ctx = cxl_get_context(afu_dev);
1919 
1920 			if (ctx && cxl_release_context(ctx))
1921 				goto err;
1922 
1923 			ctx = cxl_dev_context_init(afu_dev);
1924 			if (!ctx)
1925 				goto err;
1926 
1927 			afu_dev->dev.archdata.cxl_ctx = ctx;
1928 
1929 			if (cxl_ops->afu_check_and_enable(afu))
1930 				goto err;
1931 
1932 			afu_dev->error_state = pci_channel_io_normal;
1933 
1934 			/* If there's a driver attached, allow it to
1935 			 * chime in on recovery. Drivers should check
1936 			 * if everything has come back OK, but
1937 			 * shouldn't start new work until we call
1938 			 * their resume function.
1939 			 */
1940 			if (!afu_dev->driver)
1941 				continue;
1942 
1943 			if (afu_dev->driver->err_handler &&
1944 			    afu_dev->driver->err_handler->slot_reset)
1945 				afu_result = afu_dev->driver->err_handler->slot_reset(afu_dev);
1946 
1947 			if (afu_result == PCI_ERS_RESULT_DISCONNECT)
1948 				result = PCI_ERS_RESULT_DISCONNECT;
1949 		}
1950 	}
1951 	return result;
1952 
1953 err:
1954 	/* All the bits that happen in both error_detected and cxl_remove
1955 	 * should be idempotent, so we don't need to worry about leaving a mix
1956 	 * of unconfigured and reconfigured resources.
1957 	 */
1958 	dev_err(&pdev->dev, "EEH recovery failed. Asking to be disconnected.\n");
1959 	return PCI_ERS_RESULT_DISCONNECT;
1960 }
1961 
1962 static void cxl_pci_resume(struct pci_dev *pdev)
1963 {
1964 	struct cxl *adapter = pci_get_drvdata(pdev);
1965 	struct cxl_afu *afu;
1966 	struct pci_dev *afu_dev;
1967 	int i;
1968 
1969 	/* Everything is back now. Drivers should restart work now.
1970 	 * This is not the place to be checking if everything came back up
1971 	 * properly, because there's no return value: do that in slot_reset.
1972 	 */
1973 	for (i = 0; i < adapter->slices; i++) {
1974 		afu = adapter->afu[i];
1975 
1976 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
1977 			if (afu_dev->driver && afu_dev->driver->err_handler &&
1978 			    afu_dev->driver->err_handler->resume)
1979 				afu_dev->driver->err_handler->resume(afu_dev);
1980 		}
1981 	}
1982 }
1983 
1984 static const struct pci_error_handlers cxl_err_handler = {
1985 	.error_detected = cxl_pci_error_detected,
1986 	.slot_reset = cxl_pci_slot_reset,
1987 	.resume = cxl_pci_resume,
1988 };
1989 
1990 struct pci_driver cxl_pci_driver = {
1991 	.name = "cxl-pci",
1992 	.id_table = cxl_pci_tbl,
1993 	.probe = cxl_probe,
1994 	.remove = cxl_remove,
1995 	.shutdown = cxl_remove,
1996 	.err_handler = &cxl_err_handler,
1997 };
1998