xref: /linux/drivers/misc/cxl/pci.c (revision b6ebbac51bedf9e98e837688bc838f400196da5e)
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;
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_accum */
402 	cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, 0x0800000000000000ULL);
403 	/* for debugging with trace arrays */
404 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
405 
406 	return 0;
407 }
408 
409 static int init_implementation_adapter_xsl_regs(struct cxl *adapter, struct pci_dev *dev)
410 {
411 	u64 xsl_dsnctl;
412 	u64 chipid;
413 	u64 capp_unit_id;
414 	int rc;
415 
416 	rc = calc_capp_routing(dev, &chipid, &capp_unit_id);
417 	if (rc)
418 		return rc;
419 
420 	/* Tell XSL where to route data to */
421 	xsl_dsnctl = 0x0000600000000000ULL | (chipid << (63-5));
422 	xsl_dsnctl |= (capp_unit_id << (63-13));
423 	cxl_p1_write(adapter, CXL_XSL_DSNCTL, xsl_dsnctl);
424 
425 	return 0;
426 }
427 
428 /* PSL & XSL */
429 #define TBSYNC_CAL(n) (((u64)n & 0x7) << (63-3))
430 #define TBSYNC_CNT(n) (((u64)n & 0x7) << (63-6))
431 /* For the PSL this is a multiple for 0 < n <= 7: */
432 #define PSL_2048_250MHZ_CYCLES 1
433 
434 static void write_timebase_ctrl_psl(struct cxl *adapter)
435 {
436 	cxl_p1_write(adapter, CXL_PSL_TB_CTLSTAT,
437 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
438 }
439 
440 /* XSL */
441 #define TBSYNC_ENA (1ULL << 63)
442 /* For the XSL this is 2**n * 2000 clocks for 0 < n <= 6: */
443 #define XSL_2000_CLOCKS 1
444 #define XSL_4000_CLOCKS 2
445 #define XSL_8000_CLOCKS 3
446 
447 static void write_timebase_ctrl_xsl(struct cxl *adapter)
448 {
449 	cxl_p1_write(adapter, CXL_XSL_TB_CTLSTAT,
450 		     TBSYNC_ENA |
451 		     TBSYNC_CAL(3) |
452 		     TBSYNC_CNT(XSL_4000_CLOCKS));
453 }
454 
455 static u64 timebase_read_psl(struct cxl *adapter)
456 {
457 	return cxl_p1_read(adapter, CXL_PSL_Timebase);
458 }
459 
460 static u64 timebase_read_xsl(struct cxl *adapter)
461 {
462 	return cxl_p1_read(adapter, CXL_XSL_Timebase);
463 }
464 
465 static void cxl_setup_psl_timebase(struct cxl *adapter, struct pci_dev *dev)
466 {
467 	u64 psl_tb;
468 	int delta;
469 	unsigned int retry = 0;
470 	struct device_node *np;
471 
472 	adapter->psl_timebase_synced = false;
473 
474 	if (!(np = pnv_pci_get_phb_node(dev)))
475 		return;
476 
477 	/* Do not fail when CAPP timebase sync is not supported by OPAL */
478 	of_node_get(np);
479 	if (! of_get_property(np, "ibm,capp-timebase-sync", NULL)) {
480 		of_node_put(np);
481 		dev_info(&dev->dev, "PSL timebase inactive: OPAL support missing\n");
482 		return;
483 	}
484 	of_node_put(np);
485 
486 	/*
487 	 * Setup PSL Timebase Control and Status register
488 	 * with the recommended Timebase Sync Count value
489 	 */
490 	adapter->native->sl_ops->write_timebase_ctrl(adapter);
491 
492 	/* Enable PSL Timebase */
493 	cxl_p1_write(adapter, CXL_PSL_Control, 0x0000000000000000);
494 	cxl_p1_write(adapter, CXL_PSL_Control, CXL_PSL_Control_tb);
495 
496 	/* Wait until CORE TB and PSL TB difference <= 16usecs */
497 	do {
498 		msleep(1);
499 		if (retry++ > 5) {
500 			dev_info(&dev->dev, "PSL timebase can't synchronize\n");
501 			return;
502 		}
503 		psl_tb = adapter->native->sl_ops->timebase_read(adapter);
504 		delta = mftb() - psl_tb;
505 		if (delta < 0)
506 			delta = -delta;
507 	} while (tb_to_ns(delta) > 16000);
508 
509 	adapter->psl_timebase_synced = true;
510 	return;
511 }
512 
513 static int init_implementation_afu_psl_regs(struct cxl_afu *afu)
514 {
515 	/* read/write masks for this slice */
516 	cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
517 	/* APC read/write masks for this slice */
518 	cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
519 	/* for debugging with trace arrays */
520 	cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
521 	cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
522 
523 	return 0;
524 }
525 
526 int cxl_pci_setup_irq(struct cxl *adapter, unsigned int hwirq,
527 		unsigned int virq)
528 {
529 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
530 
531 	return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
532 }
533 
534 int cxl_update_image_control(struct cxl *adapter)
535 {
536 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
537 	int rc;
538 	int vsec;
539 	u8 image_state;
540 
541 	if (!(vsec = find_cxl_vsec(dev))) {
542 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
543 		return -ENODEV;
544 	}
545 
546 	if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
547 		dev_err(&dev->dev, "failed to read image state: %i\n", rc);
548 		return rc;
549 	}
550 
551 	if (adapter->perst_loads_image)
552 		image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
553 	else
554 		image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
555 
556 	if (adapter->perst_select_user)
557 		image_state |= CXL_VSEC_PERST_SELECT_USER;
558 	else
559 		image_state &= ~CXL_VSEC_PERST_SELECT_USER;
560 
561 	if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
562 		dev_err(&dev->dev, "failed to update image control: %i\n", rc);
563 		return rc;
564 	}
565 
566 	return 0;
567 }
568 
569 int cxl_pci_alloc_one_irq(struct cxl *adapter)
570 {
571 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
572 
573 	return pnv_cxl_alloc_hwirqs(dev, 1);
574 }
575 
576 void cxl_pci_release_one_irq(struct cxl *adapter, int hwirq)
577 {
578 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
579 
580 	return pnv_cxl_release_hwirqs(dev, hwirq, 1);
581 }
582 
583 int cxl_pci_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
584 			struct cxl *adapter, unsigned int num)
585 {
586 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
587 
588 	return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
589 }
590 
591 void cxl_pci_release_irq_ranges(struct cxl_irq_ranges *irqs,
592 				struct cxl *adapter)
593 {
594 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
595 
596 	pnv_cxl_release_hwirq_ranges(irqs, dev);
597 }
598 
599 static int setup_cxl_bars(struct pci_dev *dev)
600 {
601 	/* Safety check in case we get backported to < 3.17 without M64 */
602 	if ((p1_base(dev) < 0x100000000ULL) ||
603 	    (p2_base(dev) < 0x100000000ULL)) {
604 		dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
605 		return -ENODEV;
606 	}
607 
608 	/*
609 	 * BAR 4/5 has a special meaning for CXL and must be programmed with a
610 	 * special value corresponding to the CXL protocol address range.
611 	 * For POWER 8 that means bits 48:49 must be set to 10
612 	 */
613 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
614 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
615 
616 	return 0;
617 }
618 
619 #ifdef CONFIG_CXL_BIMODAL
620 
621 struct cxl_switch_work {
622 	struct pci_dev *dev;
623 	struct work_struct work;
624 	int vsec;
625 	int mode;
626 };
627 
628 static void switch_card_to_cxl(struct work_struct *work)
629 {
630 	struct cxl_switch_work *switch_work =
631 		container_of(work, struct cxl_switch_work, work);
632 	struct pci_dev *dev = switch_work->dev;
633 	struct pci_bus *bus = dev->bus;
634 	struct pci_controller *hose = pci_bus_to_host(bus);
635 	struct pci_dev *bridge;
636 	struct pnv_php_slot *php_slot;
637 	unsigned int devfn;
638 	u8 val;
639 	int rc;
640 
641 	dev_info(&bus->dev, "cxl: Preparing for mode switch...\n");
642 	bridge = list_first_entry_or_null(&hose->bus->devices, struct pci_dev,
643 					  bus_list);
644 	if (!bridge) {
645 		dev_WARN(&bus->dev, "cxl: Couldn't find root port!\n");
646 		goto err_dev_put;
647 	}
648 
649 	php_slot = pnv_php_find_slot(pci_device_to_OF_node(bridge));
650 	if (!php_slot) {
651 		dev_err(&bus->dev, "cxl: Failed to find slot hotplug "
652 			           "information. You may need to upgrade "
653 			           "skiboot. Aborting.\n");
654 		goto err_dev_put;
655 	}
656 
657 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, switch_work->vsec, &val);
658 	if (rc) {
659 		dev_err(&bus->dev, "cxl: Failed to read CAPI mode control: %i\n", rc);
660 		goto err_dev_put;
661 	}
662 	devfn = dev->devfn;
663 
664 	/* Release the reference obtained in cxl_check_and_switch_mode() */
665 	pci_dev_put(dev);
666 
667 	dev_dbg(&bus->dev, "cxl: Removing PCI devices from kernel\n");
668 	pci_lock_rescan_remove();
669 	pci_hp_remove_devices(bridge->subordinate);
670 	pci_unlock_rescan_remove();
671 
672 	/* Switch the CXL protocol on the card */
673 	if (switch_work->mode == CXL_BIMODE_CXL) {
674 		dev_info(&bus->dev, "cxl: Switching card to CXL mode\n");
675 		val &= ~CXL_VSEC_PROTOCOL_MASK;
676 		val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
677 		rc = pnv_cxl_enable_phb_kernel_api(hose, true);
678 		if (rc) {
679 			dev_err(&bus->dev, "cxl: Failed to enable kernel API"
680 				           " on real PHB, aborting\n");
681 			goto err_free_work;
682 		}
683 	} else {
684 		dev_WARN(&bus->dev, "cxl: Switching card to PCI mode not supported!\n");
685 		goto err_free_work;
686 	}
687 
688 	rc = CXL_WRITE_VSEC_MODE_CONTROL_BUS(bus, devfn, switch_work->vsec, val);
689 	if (rc) {
690 		dev_err(&bus->dev, "cxl: Failed to configure CXL protocol: %i\n", rc);
691 		goto err_free_work;
692 	}
693 
694 	/*
695 	 * The CAIA spec (v1.1, Section 10.6 Bi-modal Device Support) states
696 	 * we must wait 100ms after this mode switch before touching PCIe config
697 	 * space.
698 	 */
699 	msleep(100);
700 
701 	/*
702 	 * Hot reset to cause the card to come back in cxl mode. A
703 	 * OPAL_RESET_PCI_LINK would be sufficient, but currently lacks support
704 	 * in skiboot, so we use a hot reset instead.
705 	 *
706 	 * We call pci_set_pcie_reset_state() on the bridge, as a CAPI card is
707 	 * guaranteed to sit directly under the root port, and setting the reset
708 	 * state on a device directly under the root port is equivalent to doing
709 	 * it on the root port iself.
710 	 */
711 	dev_info(&bus->dev, "cxl: Configuration write complete, resetting card\n");
712 	pci_set_pcie_reset_state(bridge, pcie_hot_reset);
713 	pci_set_pcie_reset_state(bridge, pcie_deassert_reset);
714 
715 	dev_dbg(&bus->dev, "cxl: Offlining slot\n");
716 	rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_OFFLINE);
717 	if (rc) {
718 		dev_err(&bus->dev, "cxl: OPAL offlining call failed: %i\n", rc);
719 		goto err_free_work;
720 	}
721 
722 	dev_dbg(&bus->dev, "cxl: Onlining and probing slot\n");
723 	rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_ONLINE);
724 	if (rc) {
725 		dev_err(&bus->dev, "cxl: OPAL onlining call failed: %i\n", rc);
726 		goto err_free_work;
727 	}
728 
729 	pci_lock_rescan_remove();
730 	pci_hp_add_devices(bridge->subordinate);
731 	pci_unlock_rescan_remove();
732 
733 	dev_info(&bus->dev, "cxl: CAPI mode switch completed\n");
734 	kfree(switch_work);
735 	return;
736 
737 err_dev_put:
738 	/* Release the reference obtained in cxl_check_and_switch_mode() */
739 	pci_dev_put(dev);
740 err_free_work:
741 	kfree(switch_work);
742 }
743 
744 int cxl_check_and_switch_mode(struct pci_dev *dev, int mode, int vsec)
745 {
746 	struct cxl_switch_work *work;
747 	u8 val;
748 	int rc;
749 
750 	if (!cpu_has_feature(CPU_FTR_HVMODE))
751 		return -ENODEV;
752 
753 	if (!vsec) {
754 		vsec = find_cxl_vsec(dev);
755 		if (!vsec) {
756 			dev_info(&dev->dev, "CXL VSEC not found\n");
757 			return -ENODEV;
758 		}
759 	}
760 
761 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
762 	if (rc) {
763 		dev_err(&dev->dev, "Failed to read current mode control: %i", rc);
764 		return rc;
765 	}
766 
767 	if (mode == CXL_BIMODE_PCI) {
768 		if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
769 			dev_info(&dev->dev, "Card is already in PCI mode\n");
770 			return 0;
771 		}
772 		/*
773 		 * TODO: Before it's safe to switch the card back to PCI mode
774 		 * we need to disable the CAPP and make sure any cachelines the
775 		 * card holds have been flushed out. Needs skiboot support.
776 		 */
777 		dev_WARN(&dev->dev, "CXL mode switch to PCI unsupported!\n");
778 		return -EIO;
779 	}
780 
781 	if (val & CXL_VSEC_PROTOCOL_ENABLE) {
782 		dev_info(&dev->dev, "Card is already in CXL mode\n");
783 		return 0;
784 	}
785 
786 	dev_info(&dev->dev, "Card is in PCI mode, scheduling kernel thread "
787 			    "to switch to CXL mode\n");
788 
789 	work = kmalloc(sizeof(struct cxl_switch_work), GFP_KERNEL);
790 	if (!work)
791 		return -ENOMEM;
792 
793 	pci_dev_get(dev);
794 	work->dev = dev;
795 	work->vsec = vsec;
796 	work->mode = mode;
797 	INIT_WORK(&work->work, switch_card_to_cxl);
798 
799 	schedule_work(&work->work);
800 
801 	/*
802 	 * We return a failure now to abort the driver init. Once the
803 	 * link has been cycled and the card is in cxl mode we will
804 	 * come back (possibly using the generic cxl driver), but
805 	 * return success as the card should then be in cxl mode.
806 	 *
807 	 * TODO: What if the card comes back in PCI mode even after
808 	 *       the switch?  Don't want to spin endlessly.
809 	 */
810 	return -EBUSY;
811 }
812 EXPORT_SYMBOL_GPL(cxl_check_and_switch_mode);
813 
814 #endif /* CONFIG_CXL_BIMODAL */
815 
816 static int setup_cxl_protocol_area(struct pci_dev *dev)
817 {
818 	u8 val;
819 	int rc;
820 	int vsec = find_cxl_vsec(dev);
821 
822 	if (!vsec) {
823 		dev_info(&dev->dev, "CXL VSEC not found\n");
824 		return -ENODEV;
825 	}
826 
827 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
828 	if (rc) {
829 		dev_err(&dev->dev, "Failed to read current mode control: %i\n", rc);
830 		return rc;
831 	}
832 
833 	if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
834 		dev_err(&dev->dev, "Card not in CAPI mode!\n");
835 		return -EIO;
836 	}
837 
838 	if ((val & CXL_VSEC_PROTOCOL_MASK) != CXL_VSEC_PROTOCOL_256TB) {
839 		val &= ~CXL_VSEC_PROTOCOL_MASK;
840 		val |= CXL_VSEC_PROTOCOL_256TB;
841 		rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val);
842 		if (rc) {
843 			dev_err(&dev->dev, "Failed to set CXL protocol area: %i\n", rc);
844 			return rc;
845 		}
846 	}
847 
848 	return 0;
849 }
850 
851 static int pci_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
852 {
853 	u64 p1n_base, p2n_base, afu_desc;
854 	const u64 p1n_size = 0x100;
855 	const u64 p2n_size = 0x1000;
856 
857 	p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
858 	p2n_base = p2_base(dev) + (afu->slice * p2n_size);
859 	afu->psn_phys = p2_base(dev) + (adapter->native->ps_off + (afu->slice * adapter->ps_size));
860 	afu_desc = p2_base(dev) + adapter->native->afu_desc_off + (afu->slice * adapter->native->afu_desc_size);
861 
862 	if (!(afu->native->p1n_mmio = ioremap(p1n_base, p1n_size)))
863 		goto err;
864 	if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
865 		goto err1;
866 	if (afu_desc) {
867 		if (!(afu->native->afu_desc_mmio = ioremap(afu_desc, adapter->native->afu_desc_size)))
868 			goto err2;
869 	}
870 
871 	return 0;
872 err2:
873 	iounmap(afu->p2n_mmio);
874 err1:
875 	iounmap(afu->native->p1n_mmio);
876 err:
877 	dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
878 	return -ENOMEM;
879 }
880 
881 static void pci_unmap_slice_regs(struct cxl_afu *afu)
882 {
883 	if (afu->p2n_mmio) {
884 		iounmap(afu->p2n_mmio);
885 		afu->p2n_mmio = NULL;
886 	}
887 	if (afu->native->p1n_mmio) {
888 		iounmap(afu->native->p1n_mmio);
889 		afu->native->p1n_mmio = NULL;
890 	}
891 	if (afu->native->afu_desc_mmio) {
892 		iounmap(afu->native->afu_desc_mmio);
893 		afu->native->afu_desc_mmio = NULL;
894 	}
895 }
896 
897 void cxl_pci_release_afu(struct device *dev)
898 {
899 	struct cxl_afu *afu = to_cxl_afu(dev);
900 
901 	pr_devel("%s\n", __func__);
902 
903 	idr_destroy(&afu->contexts_idr);
904 	cxl_release_spa(afu);
905 
906 	kfree(afu->native);
907 	kfree(afu);
908 }
909 
910 /* Expects AFU struct to have recently been zeroed out */
911 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
912 {
913 	u64 val;
914 
915 	val = AFUD_READ_INFO(afu);
916 	afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
917 	afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
918 	afu->crs_num = AFUD_NUM_CRS(val);
919 
920 	if (AFUD_AFU_DIRECTED(val))
921 		afu->modes_supported |= CXL_MODE_DIRECTED;
922 	if (AFUD_DEDICATED_PROCESS(val))
923 		afu->modes_supported |= CXL_MODE_DEDICATED;
924 	if (AFUD_TIME_SLICED(val))
925 		afu->modes_supported |= CXL_MODE_TIME_SLICED;
926 
927 	val = AFUD_READ_PPPSA(afu);
928 	afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
929 	afu->psa = AFUD_PPPSA_PSA(val);
930 	if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
931 		afu->native->pp_offset = AFUD_READ_PPPSA_OFF(afu);
932 
933 	val = AFUD_READ_CR(afu);
934 	afu->crs_len = AFUD_CR_LEN(val) * 256;
935 	afu->crs_offset = AFUD_READ_CR_OFF(afu);
936 
937 
938 	/* eb_len is in multiple of 4K */
939 	afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
940 	afu->eb_offset = AFUD_READ_EB_OFF(afu);
941 
942 	/* eb_off is 4K aligned so lower 12 bits are always zero */
943 	if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
944 		dev_warn(&afu->dev,
945 			 "Invalid AFU error buffer offset %Lx\n",
946 			 afu->eb_offset);
947 		dev_info(&afu->dev,
948 			 "Ignoring AFU error buffer in the descriptor\n");
949 		/* indicate that no afu buffer exists */
950 		afu->eb_len = 0;
951 	}
952 
953 	return 0;
954 }
955 
956 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
957 {
958 	int i, rc;
959 	u32 val;
960 
961 	if (afu->psa && afu->adapter->ps_size <
962 			(afu->native->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
963 		dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
964 		return -ENODEV;
965 	}
966 
967 	if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
968 		dev_warn(&afu->dev, "AFU uses < PAGE_SIZE per-process PSA!");
969 
970 	for (i = 0; i < afu->crs_num; i++) {
971 		rc = cxl_ops->afu_cr_read32(afu, i, 0, &val);
972 		if (rc || val == 0) {
973 			dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
974 			return -EINVAL;
975 		}
976 	}
977 
978 	if ((afu->modes_supported & ~CXL_MODE_DEDICATED) && afu->max_procs_virtualised == 0) {
979 		/*
980 		 * We could also check this for the dedicated process model
981 		 * since the architecture indicates it should be set to 1, but
982 		 * in that case we ignore the value and I'd rather not risk
983 		 * breaking any existing dedicated process AFUs that left it as
984 		 * 0 (not that I'm aware of any). It is clearly an error for an
985 		 * AFU directed AFU to set this to 0, and would have previously
986 		 * triggered a bug resulting in the maximum not being enforced
987 		 * at all since idr_alloc treats 0 as no maximum.
988 		 */
989 		dev_err(&afu->dev, "AFU does not support any processes\n");
990 		return -EINVAL;
991 	}
992 
993 	return 0;
994 }
995 
996 static int sanitise_afu_regs(struct cxl_afu *afu)
997 {
998 	u64 reg;
999 
1000 	/*
1001 	 * Clear out any regs that contain either an IVTE or address or may be
1002 	 * waiting on an acknowledgement to try to be a bit safer as we bring
1003 	 * it online
1004 	 */
1005 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
1006 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
1007 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
1008 		if (cxl_ops->afu_reset(afu))
1009 			return -EIO;
1010 		if (cxl_afu_disable(afu))
1011 			return -EIO;
1012 		if (cxl_psl_purge(afu))
1013 			return -EIO;
1014 	}
1015 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
1016 	cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
1017 	cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
1018 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1019 	cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
1020 	cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
1021 	cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
1022 	cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
1023 	cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
1024 	cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
1025 	cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
1026 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1027 	if (reg) {
1028 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1029 		if (reg & CXL_PSL_DSISR_TRANS)
1030 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1031 		else
1032 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1033 	}
1034 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1035 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1036 		if (reg) {
1037 			if (reg & ~0xffff)
1038 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1039 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1040 		}
1041 	}
1042 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1043 	if (reg) {
1044 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1045 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1046 	}
1047 
1048 	return 0;
1049 }
1050 
1051 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
1052 /*
1053  * afu_eb_read:
1054  * Called from sysfs and reads the afu error info buffer. The h/w only supports
1055  * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
1056  * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
1057  */
1058 ssize_t cxl_pci_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
1059 				loff_t off, size_t count)
1060 {
1061 	loff_t aligned_start, aligned_end;
1062 	size_t aligned_length;
1063 	void *tbuf;
1064 	const void __iomem *ebuf = afu->native->afu_desc_mmio + afu->eb_offset;
1065 
1066 	if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
1067 		return 0;
1068 
1069 	/* calculate aligned read window */
1070 	count = min((size_t)(afu->eb_len - off), count);
1071 	aligned_start = round_down(off, 8);
1072 	aligned_end = round_up(off + count, 8);
1073 	aligned_length = aligned_end - aligned_start;
1074 
1075 	/* max we can copy in one read is PAGE_SIZE */
1076 	if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
1077 		aligned_length = ERR_BUFF_MAX_COPY_SIZE;
1078 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
1079 	}
1080 
1081 	/* use bounce buffer for copy */
1082 	tbuf = (void *)__get_free_page(GFP_TEMPORARY);
1083 	if (!tbuf)
1084 		return -ENOMEM;
1085 
1086 	/* perform aligned read from the mmio region */
1087 	memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
1088 	memcpy(buf, tbuf + (off & 0x7), count);
1089 
1090 	free_page((unsigned long)tbuf);
1091 
1092 	return count;
1093 }
1094 
1095 static int pci_configure_afu(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1096 {
1097 	int rc;
1098 
1099 	if ((rc = pci_map_slice_regs(afu, adapter, dev)))
1100 		return rc;
1101 
1102 	if ((rc = sanitise_afu_regs(afu)))
1103 		goto err1;
1104 
1105 	/* We need to reset the AFU before we can read the AFU descriptor */
1106 	if ((rc = cxl_ops->afu_reset(afu)))
1107 		goto err1;
1108 
1109 	if (cxl_verbose)
1110 		dump_afu_descriptor(afu);
1111 
1112 	if ((rc = cxl_read_afu_descriptor(afu)))
1113 		goto err1;
1114 
1115 	if ((rc = cxl_afu_descriptor_looks_ok(afu)))
1116 		goto err1;
1117 
1118 	if (adapter->native->sl_ops->afu_regs_init)
1119 		if ((rc = adapter->native->sl_ops->afu_regs_init(afu)))
1120 			goto err1;
1121 
1122 	if (adapter->native->sl_ops->register_serr_irq)
1123 		if ((rc = adapter->native->sl_ops->register_serr_irq(afu)))
1124 			goto err1;
1125 
1126 	if ((rc = cxl_native_register_psl_irq(afu)))
1127 		goto err2;
1128 
1129 	return 0;
1130 
1131 err2:
1132 	if (adapter->native->sl_ops->release_serr_irq)
1133 		adapter->native->sl_ops->release_serr_irq(afu);
1134 err1:
1135 	pci_unmap_slice_regs(afu);
1136 	return rc;
1137 }
1138 
1139 static void pci_deconfigure_afu(struct cxl_afu *afu)
1140 {
1141 	cxl_native_release_psl_irq(afu);
1142 	if (afu->adapter->native->sl_ops->release_serr_irq)
1143 		afu->adapter->native->sl_ops->release_serr_irq(afu);
1144 	pci_unmap_slice_regs(afu);
1145 }
1146 
1147 static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
1148 {
1149 	struct cxl_afu *afu;
1150 	int rc = -ENOMEM;
1151 
1152 	afu = cxl_alloc_afu(adapter, slice);
1153 	if (!afu)
1154 		return -ENOMEM;
1155 
1156 	afu->native = kzalloc(sizeof(struct cxl_afu_native), GFP_KERNEL);
1157 	if (!afu->native)
1158 		goto err_free_afu;
1159 
1160 	mutex_init(&afu->native->spa_mutex);
1161 
1162 	rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice);
1163 	if (rc)
1164 		goto err_free_native;
1165 
1166 	rc = pci_configure_afu(afu, adapter, dev);
1167 	if (rc)
1168 		goto err_free_native;
1169 
1170 	/* Don't care if this fails */
1171 	cxl_debugfs_afu_add(afu);
1172 
1173 	/*
1174 	 * After we call this function we must not free the afu directly, even
1175 	 * if it returns an error!
1176 	 */
1177 	if ((rc = cxl_register_afu(afu)))
1178 		goto err_put1;
1179 
1180 	if ((rc = cxl_sysfs_afu_add(afu)))
1181 		goto err_put1;
1182 
1183 	adapter->afu[afu->slice] = afu;
1184 
1185 	if ((rc = cxl_pci_vphb_add(afu)))
1186 		dev_info(&afu->dev, "Can't register vPHB\n");
1187 
1188 	return 0;
1189 
1190 err_put1:
1191 	pci_deconfigure_afu(afu);
1192 	cxl_debugfs_afu_remove(afu);
1193 	device_unregister(&afu->dev);
1194 	return rc;
1195 
1196 err_free_native:
1197 	kfree(afu->native);
1198 err_free_afu:
1199 	kfree(afu);
1200 	return rc;
1201 
1202 }
1203 
1204 static void cxl_pci_remove_afu(struct cxl_afu *afu)
1205 {
1206 	pr_devel("%s\n", __func__);
1207 
1208 	if (!afu)
1209 		return;
1210 
1211 	cxl_pci_vphb_remove(afu);
1212 	cxl_sysfs_afu_remove(afu);
1213 	cxl_debugfs_afu_remove(afu);
1214 
1215 	spin_lock(&afu->adapter->afu_list_lock);
1216 	afu->adapter->afu[afu->slice] = NULL;
1217 	spin_unlock(&afu->adapter->afu_list_lock);
1218 
1219 	cxl_context_detach_all(afu);
1220 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1221 
1222 	pci_deconfigure_afu(afu);
1223 	device_unregister(&afu->dev);
1224 }
1225 
1226 int cxl_pci_reset(struct cxl *adapter)
1227 {
1228 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1229 	int rc;
1230 
1231 	if (adapter->perst_same_image) {
1232 		dev_warn(&dev->dev,
1233 			 "cxl: refusing to reset/reflash when perst_reloads_same_image is set.\n");
1234 		return -EINVAL;
1235 	}
1236 
1237 	dev_info(&dev->dev, "CXL reset\n");
1238 
1239 	/* pcie_warm_reset requests a fundamental pci reset which includes a
1240 	 * PERST assert/deassert.  PERST triggers a loading of the image
1241 	 * if "user" or "factory" is selected in sysfs */
1242 	if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
1243 		dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
1244 		return rc;
1245 	}
1246 
1247 	return rc;
1248 }
1249 
1250 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
1251 {
1252 	if (pci_request_region(dev, 2, "priv 2 regs"))
1253 		goto err1;
1254 	if (pci_request_region(dev, 0, "priv 1 regs"))
1255 		goto err2;
1256 
1257 	pr_devel("cxl_map_adapter_regs: p1: %#016llx %#llx, p2: %#016llx %#llx",
1258 			p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
1259 
1260 	if (!(adapter->native->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
1261 		goto err3;
1262 
1263 	if (!(adapter->native->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
1264 		goto err4;
1265 
1266 	return 0;
1267 
1268 err4:
1269 	iounmap(adapter->native->p1_mmio);
1270 	adapter->native->p1_mmio = NULL;
1271 err3:
1272 	pci_release_region(dev, 0);
1273 err2:
1274 	pci_release_region(dev, 2);
1275 err1:
1276 	return -ENOMEM;
1277 }
1278 
1279 static void cxl_unmap_adapter_regs(struct cxl *adapter)
1280 {
1281 	if (adapter->native->p1_mmio) {
1282 		iounmap(adapter->native->p1_mmio);
1283 		adapter->native->p1_mmio = NULL;
1284 		pci_release_region(to_pci_dev(adapter->dev.parent), 2);
1285 	}
1286 	if (adapter->native->p2_mmio) {
1287 		iounmap(adapter->native->p2_mmio);
1288 		adapter->native->p2_mmio = NULL;
1289 		pci_release_region(to_pci_dev(adapter->dev.parent), 0);
1290 	}
1291 }
1292 
1293 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
1294 {
1295 	int vsec;
1296 	u32 afu_desc_off, afu_desc_size;
1297 	u32 ps_off, ps_size;
1298 	u16 vseclen;
1299 	u8 image_state;
1300 
1301 	if (!(vsec = find_cxl_vsec(dev))) {
1302 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
1303 		return -ENODEV;
1304 	}
1305 
1306 	CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
1307 	if (vseclen < CXL_VSEC_MIN_SIZE) {
1308 		dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
1309 		return -EINVAL;
1310 	}
1311 
1312 	CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
1313 	CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
1314 	CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
1315 	CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
1316 	CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
1317 	CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
1318 	adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1319 	adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1320 
1321 	CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
1322 	CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
1323 	CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
1324 	CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
1325 	CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
1326 
1327 	/* Convert everything to bytes, because there is NO WAY I'd look at the
1328 	 * code a month later and forget what units these are in ;-) */
1329 	adapter->native->ps_off = ps_off * 64 * 1024;
1330 	adapter->ps_size = ps_size * 64 * 1024;
1331 	adapter->native->afu_desc_off = afu_desc_off * 64 * 1024;
1332 	adapter->native->afu_desc_size = afu_desc_size * 64 * 1024;
1333 
1334 	/* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
1335 	adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
1336 
1337 	return 0;
1338 }
1339 
1340 /*
1341  * Workaround a PCIe Host Bridge defect on some cards, that can cause
1342  * malformed Transaction Layer Packet (TLP) errors to be erroneously
1343  * reported. Mask this error in the Uncorrectable Error Mask Register.
1344  *
1345  * The upper nibble of the PSL revision is used to distinguish between
1346  * different cards. The affected ones have it set to 0.
1347  */
1348 static void cxl_fixup_malformed_tlp(struct cxl *adapter, struct pci_dev *dev)
1349 {
1350 	int aer;
1351 	u32 data;
1352 
1353 	if (adapter->psl_rev & 0xf000)
1354 		return;
1355 	if (!(aer = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR)))
1356 		return;
1357 	pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &data);
1358 	if (data & PCI_ERR_UNC_MALF_TLP)
1359 		if (data & PCI_ERR_UNC_INTN)
1360 			return;
1361 	data |= PCI_ERR_UNC_MALF_TLP;
1362 	data |= PCI_ERR_UNC_INTN;
1363 	pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, data);
1364 }
1365 
1366 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
1367 {
1368 	if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
1369 		return -EBUSY;
1370 
1371 	if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
1372 		dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
1373 		return -EINVAL;
1374 	}
1375 
1376 	if (!adapter->slices) {
1377 		/* Once we support dynamic reprogramming we can use the card if
1378 		 * it supports loadable AFUs */
1379 		dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
1380 		return -EINVAL;
1381 	}
1382 
1383 	if (!adapter->native->afu_desc_off || !adapter->native->afu_desc_size) {
1384 		dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
1385 		return -EINVAL;
1386 	}
1387 
1388 	if (adapter->ps_size > p2_size(dev) - adapter->native->ps_off) {
1389 		dev_err(&dev->dev, "ABORTING: Problem state size larger than "
1390 				   "available in BAR2: 0x%llx > 0x%llx\n",
1391 			 adapter->ps_size, p2_size(dev) - adapter->native->ps_off);
1392 		return -EINVAL;
1393 	}
1394 
1395 	return 0;
1396 }
1397 
1398 ssize_t cxl_pci_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1399 {
1400 	return pci_read_vpd(to_pci_dev(adapter->dev.parent), 0, len, buf);
1401 }
1402 
1403 static void cxl_release_adapter(struct device *dev)
1404 {
1405 	struct cxl *adapter = to_cxl_adapter(dev);
1406 
1407 	pr_devel("cxl_release_adapter\n");
1408 
1409 	cxl_remove_adapter_nr(adapter);
1410 
1411 	kfree(adapter->native);
1412 	kfree(adapter);
1413 }
1414 
1415 #define CXL_PSL_ErrIVTE_tberror (0x1ull << (63-31))
1416 
1417 static int sanitise_adapter_regs(struct cxl *adapter)
1418 {
1419 	/* Clear PSL tberror bit by writing 1 to it */
1420 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, CXL_PSL_ErrIVTE_tberror);
1421 	return cxl_tlb_slb_invalidate(adapter);
1422 }
1423 
1424 /* This should contain *only* operations that can safely be done in
1425  * both creation and recovery.
1426  */
1427 static int cxl_configure_adapter(struct cxl *adapter, struct pci_dev *dev)
1428 {
1429 	int rc;
1430 
1431 	adapter->dev.parent = &dev->dev;
1432 	adapter->dev.release = cxl_release_adapter;
1433 	pci_set_drvdata(dev, adapter);
1434 
1435 	rc = pci_enable_device(dev);
1436 	if (rc) {
1437 		dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1438 		return rc;
1439 	}
1440 
1441 	if ((rc = cxl_read_vsec(adapter, dev)))
1442 		return rc;
1443 
1444 	if ((rc = cxl_vsec_looks_ok(adapter, dev)))
1445 	        return rc;
1446 
1447 	cxl_fixup_malformed_tlp(adapter, dev);
1448 
1449 	if ((rc = setup_cxl_bars(dev)))
1450 		return rc;
1451 
1452 	if ((rc = setup_cxl_protocol_area(dev)))
1453 		return rc;
1454 
1455 	if ((rc = cxl_update_image_control(adapter)))
1456 		return rc;
1457 
1458 	if ((rc = cxl_map_adapter_regs(adapter, dev)))
1459 		return rc;
1460 
1461 	if ((rc = sanitise_adapter_regs(adapter)))
1462 		goto err;
1463 
1464 	if ((rc = adapter->native->sl_ops->adapter_regs_init(adapter, dev)))
1465 		goto err;
1466 
1467 	/* Required for devices using CAPP DMA mode, harmless for others */
1468 	pci_set_master(dev);
1469 
1470 	if ((rc = pnv_phb_to_cxl_mode(dev, adapter->native->sl_ops->capi_mode)))
1471 		goto err;
1472 
1473 	/* If recovery happened, the last step is to turn on snooping.
1474 	 * In the non-recovery case this has no effect */
1475 	if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON)))
1476 		goto err;
1477 
1478 	/* Ignore error, adapter init is not dependant on timebase sync */
1479 	cxl_setup_psl_timebase(adapter, dev);
1480 
1481 	if ((rc = cxl_native_register_psl_err_irq(adapter)))
1482 		goto err;
1483 
1484 	return 0;
1485 
1486 err:
1487 	cxl_unmap_adapter_regs(adapter);
1488 	return rc;
1489 
1490 }
1491 
1492 static void cxl_deconfigure_adapter(struct cxl *adapter)
1493 {
1494 	struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1495 
1496 	cxl_native_release_psl_err_irq(adapter);
1497 	cxl_unmap_adapter_regs(adapter);
1498 
1499 	pci_disable_device(pdev);
1500 }
1501 
1502 static const struct cxl_service_layer_ops psl_ops = {
1503 	.adapter_regs_init = init_implementation_adapter_psl_regs,
1504 	.afu_regs_init = init_implementation_afu_psl_regs,
1505 	.register_serr_irq = cxl_native_register_serr_irq,
1506 	.release_serr_irq = cxl_native_release_serr_irq,
1507 	.debugfs_add_adapter_sl_regs = cxl_debugfs_add_adapter_psl_regs,
1508 	.debugfs_add_afu_sl_regs = cxl_debugfs_add_afu_psl_regs,
1509 	.psl_irq_dump_registers = cxl_native_psl_irq_dump_regs,
1510 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs,
1511 	.debugfs_stop_trace = cxl_stop_trace,
1512 	.write_timebase_ctrl = write_timebase_ctrl_psl,
1513 	.timebase_read = timebase_read_psl,
1514 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1515 	.needs_reset_before_disable = true,
1516 };
1517 
1518 static const struct cxl_service_layer_ops xsl_ops = {
1519 	.adapter_regs_init = init_implementation_adapter_xsl_regs,
1520 	.debugfs_add_adapter_sl_regs = cxl_debugfs_add_adapter_xsl_regs,
1521 	.write_timebase_ctrl = write_timebase_ctrl_xsl,
1522 	.timebase_read = timebase_read_xsl,
1523 	.capi_mode = OPAL_PHB_CAPI_MODE_DMA,
1524 	.min_pe = 1, /* Workaround for Mellanox CX4 HW bug */
1525 };
1526 
1527 static void set_sl_ops(struct cxl *adapter, struct pci_dev *dev)
1528 {
1529 	if (dev->vendor == PCI_VENDOR_ID_MELLANOX && dev->device == 0x1013) {
1530 		dev_info(&adapter->dev, "Device uses an XSL\n");
1531 		adapter->native->sl_ops = &xsl_ops;
1532 	} else {
1533 		dev_info(&adapter->dev, "Device uses a PSL\n");
1534 		adapter->native->sl_ops = &psl_ops;
1535 	}
1536 }
1537 
1538 
1539 static struct cxl *cxl_pci_init_adapter(struct pci_dev *dev)
1540 {
1541 	struct cxl *adapter;
1542 	int rc;
1543 
1544 	adapter = cxl_alloc_adapter();
1545 	if (!adapter)
1546 		return ERR_PTR(-ENOMEM);
1547 
1548 	adapter->native = kzalloc(sizeof(struct cxl_native), GFP_KERNEL);
1549 	if (!adapter->native) {
1550 		rc = -ENOMEM;
1551 		goto err_release;
1552 	}
1553 
1554 	set_sl_ops(adapter, dev);
1555 
1556 	/* Set defaults for parameters which need to persist over
1557 	 * configure/reconfigure
1558 	 */
1559 	adapter->perst_loads_image = true;
1560 	adapter->perst_same_image = false;
1561 
1562 	rc = cxl_configure_adapter(adapter, dev);
1563 	if (rc) {
1564 		pci_disable_device(dev);
1565 		goto err_release;
1566 	}
1567 
1568 	/* Don't care if this one fails: */
1569 	cxl_debugfs_adapter_add(adapter);
1570 
1571 	/*
1572 	 * After we call this function we must not free the adapter directly,
1573 	 * even if it returns an error!
1574 	 */
1575 	if ((rc = cxl_register_adapter(adapter)))
1576 		goto err_put1;
1577 
1578 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1579 		goto err_put1;
1580 
1581 	return adapter;
1582 
1583 err_put1:
1584 	/* This should mirror cxl_remove_adapter, except without the
1585 	 * sysfs parts
1586 	 */
1587 	cxl_debugfs_adapter_remove(adapter);
1588 	cxl_deconfigure_adapter(adapter);
1589 	device_unregister(&adapter->dev);
1590 	return ERR_PTR(rc);
1591 
1592 err_release:
1593 	cxl_release_adapter(&adapter->dev);
1594 	return ERR_PTR(rc);
1595 }
1596 
1597 static void cxl_pci_remove_adapter(struct cxl *adapter)
1598 {
1599 	pr_devel("cxl_remove_adapter\n");
1600 
1601 	cxl_sysfs_adapter_remove(adapter);
1602 	cxl_debugfs_adapter_remove(adapter);
1603 
1604 	cxl_deconfigure_adapter(adapter);
1605 
1606 	device_unregister(&adapter->dev);
1607 }
1608 
1609 #define CXL_MAX_PCIEX_PARENT 2
1610 
1611 static int cxl_slot_is_switched(struct pci_dev *dev)
1612 {
1613 	struct device_node *np;
1614 	int depth = 0;
1615 	const __be32 *prop;
1616 
1617 	if (!(np = pci_device_to_OF_node(dev))) {
1618 		pr_err("cxl: np = NULL\n");
1619 		return -ENODEV;
1620 	}
1621 	of_node_get(np);
1622 	while (np) {
1623 		np = of_get_next_parent(np);
1624 		prop = of_get_property(np, "device_type", NULL);
1625 		if (!prop || strcmp((char *)prop, "pciex"))
1626 			break;
1627 		depth++;
1628 	}
1629 	of_node_put(np);
1630 	return (depth > CXL_MAX_PCIEX_PARENT);
1631 }
1632 
1633 bool cxl_slot_is_supported(struct pci_dev *dev, int flags)
1634 {
1635 	if (!cpu_has_feature(CPU_FTR_HVMODE))
1636 		return false;
1637 
1638 	if ((flags & CXL_SLOT_FLAG_DMA) && (!pvr_version_is(PVR_POWER8NVL))) {
1639 		/*
1640 		 * CAPP DMA mode is technically supported on regular P8, but
1641 		 * will EEH if the card attempts to access memory < 4GB, which
1642 		 * we cannot realistically avoid. We might be able to work
1643 		 * around the issue, but until then return unsupported:
1644 		 */
1645 		return false;
1646 	}
1647 
1648 	if (cxl_slot_is_switched(dev))
1649 		return false;
1650 
1651 	/*
1652 	 * XXX: This gets a little tricky on regular P8 (not POWER8NVL) since
1653 	 * the CAPP can be connected to PHB 0, 1 or 2 on a first come first
1654 	 * served basis, which is racy to check from here. If we need to
1655 	 * support this in future we might need to consider having this
1656 	 * function effectively reserve it ahead of time.
1657 	 *
1658 	 * Currently, the only user of this API is the Mellanox CX4, which is
1659 	 * only supported on P8NVL due to the above mentioned limitation of
1660 	 * CAPP DMA mode and therefore does not need to worry about this. If the
1661 	 * issue with CAPP DMA mode is later worked around on P8 we might need
1662 	 * to revisit this.
1663 	 */
1664 
1665 	return true;
1666 }
1667 EXPORT_SYMBOL_GPL(cxl_slot_is_supported);
1668 
1669 
1670 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
1671 {
1672 	struct cxl *adapter;
1673 	int slice;
1674 	int rc;
1675 
1676 	if (cxl_pci_is_vphb_device(dev)) {
1677 		dev_dbg(&dev->dev, "cxl_init_adapter: Ignoring cxl vphb device\n");
1678 		return -ENODEV;
1679 	}
1680 
1681 	if (cxl_slot_is_switched(dev)) {
1682 		dev_info(&dev->dev, "Ignoring card on incompatible PCI slot\n");
1683 		return -ENODEV;
1684 	}
1685 
1686 	if (cxl_verbose)
1687 		dump_cxl_config_space(dev);
1688 
1689 	adapter = cxl_pci_init_adapter(dev);
1690 	if (IS_ERR(adapter)) {
1691 		dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
1692 		return PTR_ERR(adapter);
1693 	}
1694 
1695 	for (slice = 0; slice < adapter->slices; slice++) {
1696 		if ((rc = pci_init_afu(adapter, slice, dev))) {
1697 			dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
1698 			continue;
1699 		}
1700 
1701 		rc = cxl_afu_select_best_mode(adapter->afu[slice]);
1702 		if (rc)
1703 			dev_err(&dev->dev, "AFU %i failed to start: %i\n", slice, rc);
1704 	}
1705 
1706 	if (pnv_pci_on_cxl_phb(dev) && adapter->slices >= 1)
1707 		pnv_cxl_phb_set_peer_afu(dev, adapter->afu[0]);
1708 
1709 	return 0;
1710 }
1711 
1712 static void cxl_remove(struct pci_dev *dev)
1713 {
1714 	struct cxl *adapter = pci_get_drvdata(dev);
1715 	struct cxl_afu *afu;
1716 	int i;
1717 
1718 	/*
1719 	 * Lock to prevent someone grabbing a ref through the adapter list as
1720 	 * we are removing it
1721 	 */
1722 	for (i = 0; i < adapter->slices; i++) {
1723 		afu = adapter->afu[i];
1724 		cxl_pci_remove_afu(afu);
1725 	}
1726 	cxl_pci_remove_adapter(adapter);
1727 }
1728 
1729 static pci_ers_result_t cxl_vphb_error_detected(struct cxl_afu *afu,
1730 						pci_channel_state_t state)
1731 {
1732 	struct pci_dev *afu_dev;
1733 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
1734 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
1735 
1736 	/* There should only be one entry, but go through the list
1737 	 * anyway
1738 	 */
1739 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
1740 		if (!afu_dev->driver)
1741 			continue;
1742 
1743 		afu_dev->error_state = state;
1744 
1745 		if (afu_dev->driver->err_handler)
1746 			afu_result = afu_dev->driver->err_handler->error_detected(afu_dev,
1747 										  state);
1748 		/* Disconnect trumps all, NONE trumps NEED_RESET */
1749 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
1750 			result = PCI_ERS_RESULT_DISCONNECT;
1751 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
1752 			 (result == PCI_ERS_RESULT_NEED_RESET))
1753 			result = PCI_ERS_RESULT_NONE;
1754 	}
1755 	return result;
1756 }
1757 
1758 static pci_ers_result_t cxl_pci_error_detected(struct pci_dev *pdev,
1759 					       pci_channel_state_t state)
1760 {
1761 	struct cxl *adapter = pci_get_drvdata(pdev);
1762 	struct cxl_afu *afu;
1763 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
1764 	int i;
1765 
1766 	/* At this point, we could still have an interrupt pending.
1767 	 * Let's try to get them out of the way before they do
1768 	 * anything we don't like.
1769 	 */
1770 	schedule();
1771 
1772 	/* If we're permanently dead, give up. */
1773 	if (state == pci_channel_io_perm_failure) {
1774 		/* Tell the AFU drivers; but we don't care what they
1775 		 * say, we're going away.
1776 		 */
1777 		for (i = 0; i < adapter->slices; i++) {
1778 			afu = adapter->afu[i];
1779 			/* Only participate in EEH if we are on a virtual PHB */
1780 			if (afu->phb == NULL)
1781 				return PCI_ERS_RESULT_NONE;
1782 			cxl_vphb_error_detected(afu, state);
1783 		}
1784 		return PCI_ERS_RESULT_DISCONNECT;
1785 	}
1786 
1787 	/* Are we reflashing?
1788 	 *
1789 	 * If we reflash, we could come back as something entirely
1790 	 * different, including a non-CAPI card. As such, by default
1791 	 * we don't participate in the process. We'll be unbound and
1792 	 * the slot re-probed. (TODO: check EEH doesn't blindly rebind
1793 	 * us!)
1794 	 *
1795 	 * However, this isn't the entire story: for reliablity
1796 	 * reasons, we usually want to reflash the FPGA on PERST in
1797 	 * order to get back to a more reliable known-good state.
1798 	 *
1799 	 * This causes us a bit of a problem: if we reflash we can't
1800 	 * trust that we'll come back the same - we could have a new
1801 	 * image and been PERSTed in order to load that
1802 	 * image. However, most of the time we actually *will* come
1803 	 * back the same - for example a regular EEH event.
1804 	 *
1805 	 * Therefore, we allow the user to assert that the image is
1806 	 * indeed the same and that we should continue on into EEH
1807 	 * anyway.
1808 	 */
1809 	if (adapter->perst_loads_image && !adapter->perst_same_image) {
1810 		/* TODO take the PHB out of CXL mode */
1811 		dev_info(&pdev->dev, "reflashing, so opting out of EEH!\n");
1812 		return PCI_ERS_RESULT_NONE;
1813 	}
1814 
1815 	/*
1816 	 * At this point, we want to try to recover.  We'll always
1817 	 * need a complete slot reset: we don't trust any other reset.
1818 	 *
1819 	 * Now, we go through each AFU:
1820 	 *  - We send the driver, if bound, an error_detected callback.
1821 	 *    We expect it to clean up, but it can also tell us to give
1822 	 *    up and permanently detach the card. To simplify things, if
1823 	 *    any bound AFU driver doesn't support EEH, we give up on EEH.
1824 	 *
1825 	 *  - We detach all contexts associated with the AFU. This
1826 	 *    does not free them, but puts them into a CLOSED state
1827 	 *    which causes any the associated files to return useful
1828 	 *    errors to userland. It also unmaps, but does not free,
1829 	 *    any IRQs.
1830 	 *
1831 	 *  - We clean up our side: releasing and unmapping resources we hold
1832 	 *    so we can wire them up again when the hardware comes back up.
1833 	 *
1834 	 * Driver authors should note:
1835 	 *
1836 	 *  - Any contexts you create in your kernel driver (except
1837 	 *    those associated with anonymous file descriptors) are
1838 	 *    your responsibility to free and recreate. Likewise with
1839 	 *    any attached resources.
1840 	 *
1841 	 *  - We will take responsibility for re-initialising the
1842 	 *    device context (the one set up for you in
1843 	 *    cxl_pci_enable_device_hook and accessed through
1844 	 *    cxl_get_context). If you've attached IRQs or other
1845 	 *    resources to it, they remains yours to free.
1846 	 *
1847 	 * You can call the same functions to release resources as you
1848 	 * normally would: we make sure that these functions continue
1849 	 * to work when the hardware is down.
1850 	 *
1851 	 * Two examples:
1852 	 *
1853 	 * 1) If you normally free all your resources at the end of
1854 	 *    each request, or if you use anonymous FDs, your
1855 	 *    error_detected callback can simply set a flag to tell
1856 	 *    your driver not to start any new calls. You can then
1857 	 *    clear the flag in the resume callback.
1858 	 *
1859 	 * 2) If you normally allocate your resources on startup:
1860 	 *     * Set a flag in error_detected as above.
1861 	 *     * Let CXL detach your contexts.
1862 	 *     * In slot_reset, free the old resources and allocate new ones.
1863 	 *     * In resume, clear the flag to allow things to start.
1864 	 */
1865 	for (i = 0; i < adapter->slices; i++) {
1866 		afu = adapter->afu[i];
1867 
1868 		result = cxl_vphb_error_detected(afu, state);
1869 
1870 		/* Only continue if everyone agrees on NEED_RESET */
1871 		if (result != PCI_ERS_RESULT_NEED_RESET)
1872 			return result;
1873 
1874 		cxl_context_detach_all(afu);
1875 		cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1876 		pci_deconfigure_afu(afu);
1877 	}
1878 	cxl_deconfigure_adapter(adapter);
1879 
1880 	return result;
1881 }
1882 
1883 static pci_ers_result_t cxl_pci_slot_reset(struct pci_dev *pdev)
1884 {
1885 	struct cxl *adapter = pci_get_drvdata(pdev);
1886 	struct cxl_afu *afu;
1887 	struct cxl_context *ctx;
1888 	struct pci_dev *afu_dev;
1889 	pci_ers_result_t afu_result = PCI_ERS_RESULT_RECOVERED;
1890 	pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
1891 	int i;
1892 
1893 	if (cxl_configure_adapter(adapter, pdev))
1894 		goto err;
1895 
1896 	for (i = 0; i < adapter->slices; i++) {
1897 		afu = adapter->afu[i];
1898 
1899 		if (pci_configure_afu(afu, adapter, pdev))
1900 			goto err;
1901 
1902 		if (cxl_afu_select_best_mode(afu))
1903 			goto err;
1904 
1905 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
1906 			/* Reset the device context.
1907 			 * TODO: make this less disruptive
1908 			 */
1909 			ctx = cxl_get_context(afu_dev);
1910 
1911 			if (ctx && cxl_release_context(ctx))
1912 				goto err;
1913 
1914 			ctx = cxl_dev_context_init(afu_dev);
1915 			if (!ctx)
1916 				goto err;
1917 
1918 			afu_dev->dev.archdata.cxl_ctx = ctx;
1919 
1920 			if (cxl_ops->afu_check_and_enable(afu))
1921 				goto err;
1922 
1923 			afu_dev->error_state = pci_channel_io_normal;
1924 
1925 			/* If there's a driver attached, allow it to
1926 			 * chime in on recovery. Drivers should check
1927 			 * if everything has come back OK, but
1928 			 * shouldn't start new work until we call
1929 			 * their resume function.
1930 			 */
1931 			if (!afu_dev->driver)
1932 				continue;
1933 
1934 			if (afu_dev->driver->err_handler &&
1935 			    afu_dev->driver->err_handler->slot_reset)
1936 				afu_result = afu_dev->driver->err_handler->slot_reset(afu_dev);
1937 
1938 			if (afu_result == PCI_ERS_RESULT_DISCONNECT)
1939 				result = PCI_ERS_RESULT_DISCONNECT;
1940 		}
1941 	}
1942 	return result;
1943 
1944 err:
1945 	/* All the bits that happen in both error_detected and cxl_remove
1946 	 * should be idempotent, so we don't need to worry about leaving a mix
1947 	 * of unconfigured and reconfigured resources.
1948 	 */
1949 	dev_err(&pdev->dev, "EEH recovery failed. Asking to be disconnected.\n");
1950 	return PCI_ERS_RESULT_DISCONNECT;
1951 }
1952 
1953 static void cxl_pci_resume(struct pci_dev *pdev)
1954 {
1955 	struct cxl *adapter = pci_get_drvdata(pdev);
1956 	struct cxl_afu *afu;
1957 	struct pci_dev *afu_dev;
1958 	int i;
1959 
1960 	/* Everything is back now. Drivers should restart work now.
1961 	 * This is not the place to be checking if everything came back up
1962 	 * properly, because there's no return value: do that in slot_reset.
1963 	 */
1964 	for (i = 0; i < adapter->slices; i++) {
1965 		afu = adapter->afu[i];
1966 
1967 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
1968 			if (afu_dev->driver && afu_dev->driver->err_handler &&
1969 			    afu_dev->driver->err_handler->resume)
1970 				afu_dev->driver->err_handler->resume(afu_dev);
1971 		}
1972 	}
1973 }
1974 
1975 static const struct pci_error_handlers cxl_err_handler = {
1976 	.error_detected = cxl_pci_error_detected,
1977 	.slot_reset = cxl_pci_slot_reset,
1978 	.resume = cxl_pci_resume,
1979 };
1980 
1981 struct pci_driver cxl_pci_driver = {
1982 	.name = "cxl-pci",
1983 	.id_table = cxl_pci_tbl,
1984 	.probe = cxl_probe,
1985 	.remove = cxl_remove,
1986 	.shutdown = cxl_remove,
1987 	.err_handler = &cxl_err_handler,
1988 };
1989