xref: /openbmc/linux/drivers/misc/cxl/pci.c (revision 5b4cb650)
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_VSEC_PROTOCOL_MASK   0xe0
59 #define CXL_VSEC_PROTOCOL_1024TB 0x80
60 #define CXL_VSEC_PROTOCOL_512TB  0x40
61 #define CXL_VSEC_PROTOCOL_256TB  0x20 /* Power 8/9 uses this */
62 #define CXL_VSEC_PROTOCOL_ENABLE 0x01
63 
64 #define CXL_READ_VSEC_PSL_REVISION(dev, vsec, dest) \
65 	pci_read_config_word(dev, vsec + 0xc, dest)
66 #define CXL_READ_VSEC_CAIA_MINOR(dev, vsec, dest) \
67 	pci_read_config_byte(dev, vsec + 0xe, dest)
68 #define CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, dest) \
69 	pci_read_config_byte(dev, vsec + 0xf, dest)
70 #define CXL_READ_VSEC_BASE_IMAGE(dev, vsec, dest) \
71 	pci_read_config_word(dev, vsec + 0x10, dest)
72 
73 #define CXL_READ_VSEC_IMAGE_STATE(dev, vsec, dest) \
74 	pci_read_config_byte(dev, vsec + 0x13, dest)
75 #define CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, val) \
76 	pci_write_config_byte(dev, vsec + 0x13, val)
77 #define CXL_VSEC_USER_IMAGE_LOADED 0x80 /* RO */
78 #define CXL_VSEC_PERST_LOADS_IMAGE 0x20 /* RW */
79 #define CXL_VSEC_PERST_SELECT_USER 0x10 /* RW */
80 
81 #define CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, dest) \
82 	pci_read_config_dword(dev, vsec + 0x20, dest)
83 #define CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, dest) \
84 	pci_read_config_dword(dev, vsec + 0x24, dest)
85 #define CXL_READ_VSEC_PS_OFF(dev, vsec, dest) \
86 	pci_read_config_dword(dev, vsec + 0x28, dest)
87 #define CXL_READ_VSEC_PS_SIZE(dev, vsec, dest) \
88 	pci_read_config_dword(dev, vsec + 0x2c, dest)
89 
90 
91 /* This works a little different than the p1/p2 register accesses to make it
92  * easier to pull out individual fields */
93 #define AFUD_READ(afu, off)		in_be64(afu->native->afu_desc_mmio + off)
94 #define AFUD_READ_LE(afu, off)		in_le64(afu->native->afu_desc_mmio + off)
95 #define EXTRACT_PPC_BIT(val, bit)	(!!(val & PPC_BIT(bit)))
96 #define EXTRACT_PPC_BITS(val, bs, be)	((val & PPC_BITMASK(bs, be)) >> PPC_BITLSHIFT(be))
97 
98 #define AFUD_READ_INFO(afu)		AFUD_READ(afu, 0x0)
99 #define   AFUD_NUM_INTS_PER_PROC(val)	EXTRACT_PPC_BITS(val,  0, 15)
100 #define   AFUD_NUM_PROCS(val)		EXTRACT_PPC_BITS(val, 16, 31)
101 #define   AFUD_NUM_CRS(val)		EXTRACT_PPC_BITS(val, 32, 47)
102 #define   AFUD_MULTIMODE(val)		EXTRACT_PPC_BIT(val, 48)
103 #define   AFUD_PUSH_BLOCK_TRANSFER(val)	EXTRACT_PPC_BIT(val, 55)
104 #define   AFUD_DEDICATED_PROCESS(val)	EXTRACT_PPC_BIT(val, 59)
105 #define   AFUD_AFU_DIRECTED(val)	EXTRACT_PPC_BIT(val, 61)
106 #define   AFUD_TIME_SLICED(val)		EXTRACT_PPC_BIT(val, 63)
107 #define AFUD_READ_CR(afu)		AFUD_READ(afu, 0x20)
108 #define   AFUD_CR_LEN(val)		EXTRACT_PPC_BITS(val, 8, 63)
109 #define AFUD_READ_CR_OFF(afu)		AFUD_READ(afu, 0x28)
110 #define AFUD_READ_PPPSA(afu)		AFUD_READ(afu, 0x30)
111 #define   AFUD_PPPSA_PP(val)		EXTRACT_PPC_BIT(val, 6)
112 #define   AFUD_PPPSA_PSA(val)		EXTRACT_PPC_BIT(val, 7)
113 #define   AFUD_PPPSA_LEN(val)		EXTRACT_PPC_BITS(val, 8, 63)
114 #define AFUD_READ_PPPSA_OFF(afu)	AFUD_READ(afu, 0x38)
115 #define AFUD_READ_EB(afu)		AFUD_READ(afu, 0x40)
116 #define   AFUD_EB_LEN(val)		EXTRACT_PPC_BITS(val, 8, 63)
117 #define AFUD_READ_EB_OFF(afu)		AFUD_READ(afu, 0x48)
118 
119 static const struct pci_device_id cxl_pci_tbl[] = {
120 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0477), },
121 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x044b), },
122 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x04cf), },
123 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0601), },
124 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0623), },
125 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0628), },
126 	{ }
127 };
128 MODULE_DEVICE_TABLE(pci, cxl_pci_tbl);
129 
130 
131 /*
132  * Mostly using these wrappers to avoid confusion:
133  * priv 1 is BAR2, while priv 2 is BAR0
134  */
135 static inline resource_size_t p1_base(struct pci_dev *dev)
136 {
137 	return pci_resource_start(dev, 2);
138 }
139 
140 static inline resource_size_t p1_size(struct pci_dev *dev)
141 {
142 	return pci_resource_len(dev, 2);
143 }
144 
145 static inline resource_size_t p2_base(struct pci_dev *dev)
146 {
147 	return pci_resource_start(dev, 0);
148 }
149 
150 static inline resource_size_t p2_size(struct pci_dev *dev)
151 {
152 	return pci_resource_len(dev, 0);
153 }
154 
155 static int find_cxl_vsec(struct pci_dev *dev)
156 {
157 	int vsec = 0;
158 	u16 val;
159 
160 	while ((vsec = pci_find_next_ext_capability(dev, vsec, PCI_EXT_CAP_ID_VNDR))) {
161 		pci_read_config_word(dev, vsec + 0x4, &val);
162 		if (val == CXL_PCI_VSEC_ID)
163 			return vsec;
164 	}
165 	return 0;
166 
167 }
168 
169 static void dump_cxl_config_space(struct pci_dev *dev)
170 {
171 	int vsec;
172 	u32 val;
173 
174 	dev_info(&dev->dev, "dump_cxl_config_space\n");
175 
176 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_0, &val);
177 	dev_info(&dev->dev, "BAR0: %#.8x\n", val);
178 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_1, &val);
179 	dev_info(&dev->dev, "BAR1: %#.8x\n", val);
180 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_2, &val);
181 	dev_info(&dev->dev, "BAR2: %#.8x\n", val);
182 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_3, &val);
183 	dev_info(&dev->dev, "BAR3: %#.8x\n", val);
184 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_4, &val);
185 	dev_info(&dev->dev, "BAR4: %#.8x\n", val);
186 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_5, &val);
187 	dev_info(&dev->dev, "BAR5: %#.8x\n", val);
188 
189 	dev_info(&dev->dev, "p1 regs: %#llx, len: %#llx\n",
190 		p1_base(dev), p1_size(dev));
191 	dev_info(&dev->dev, "p2 regs: %#llx, len: %#llx\n",
192 		p2_base(dev), p2_size(dev));
193 	dev_info(&dev->dev, "BAR 4/5: %#llx, len: %#llx\n",
194 		pci_resource_start(dev, 4), pci_resource_len(dev, 4));
195 
196 	if (!(vsec = find_cxl_vsec(dev)))
197 		return;
198 
199 #define show_reg(name, what) \
200 	dev_info(&dev->dev, "cxl vsec: %30s: %#x\n", name, what)
201 
202 	pci_read_config_dword(dev, vsec + 0x0, &val);
203 	show_reg("Cap ID", (val >> 0) & 0xffff);
204 	show_reg("Cap Ver", (val >> 16) & 0xf);
205 	show_reg("Next Cap Ptr", (val >> 20) & 0xfff);
206 	pci_read_config_dword(dev, vsec + 0x4, &val);
207 	show_reg("VSEC ID", (val >> 0) & 0xffff);
208 	show_reg("VSEC Rev", (val >> 16) & 0xf);
209 	show_reg("VSEC Length",	(val >> 20) & 0xfff);
210 	pci_read_config_dword(dev, vsec + 0x8, &val);
211 	show_reg("Num AFUs", (val >> 0) & 0xff);
212 	show_reg("Status", (val >> 8) & 0xff);
213 	show_reg("Mode Control", (val >> 16) & 0xff);
214 	show_reg("Reserved", (val >> 24) & 0xff);
215 	pci_read_config_dword(dev, vsec + 0xc, &val);
216 	show_reg("PSL Rev", (val >> 0) & 0xffff);
217 	show_reg("CAIA Ver", (val >> 16) & 0xffff);
218 	pci_read_config_dword(dev, vsec + 0x10, &val);
219 	show_reg("Base Image Rev", (val >> 0) & 0xffff);
220 	show_reg("Reserved", (val >> 16) & 0x0fff);
221 	show_reg("Image Control", (val >> 28) & 0x3);
222 	show_reg("Reserved", (val >> 30) & 0x1);
223 	show_reg("Image Loaded", (val >> 31) & 0x1);
224 
225 	pci_read_config_dword(dev, vsec + 0x14, &val);
226 	show_reg("Reserved", val);
227 	pci_read_config_dword(dev, vsec + 0x18, &val);
228 	show_reg("Reserved", val);
229 	pci_read_config_dword(dev, vsec + 0x1c, &val);
230 	show_reg("Reserved", val);
231 
232 	pci_read_config_dword(dev, vsec + 0x20, &val);
233 	show_reg("AFU Descriptor Offset", val);
234 	pci_read_config_dword(dev, vsec + 0x24, &val);
235 	show_reg("AFU Descriptor Size", val);
236 	pci_read_config_dword(dev, vsec + 0x28, &val);
237 	show_reg("Problem State Offset", val);
238 	pci_read_config_dword(dev, vsec + 0x2c, &val);
239 	show_reg("Problem State Size", val);
240 
241 	pci_read_config_dword(dev, vsec + 0x30, &val);
242 	show_reg("Reserved", val);
243 	pci_read_config_dword(dev, vsec + 0x34, &val);
244 	show_reg("Reserved", val);
245 	pci_read_config_dword(dev, vsec + 0x38, &val);
246 	show_reg("Reserved", val);
247 	pci_read_config_dword(dev, vsec + 0x3c, &val);
248 	show_reg("Reserved", val);
249 
250 	pci_read_config_dword(dev, vsec + 0x40, &val);
251 	show_reg("PSL Programming Port", val);
252 	pci_read_config_dword(dev, vsec + 0x44, &val);
253 	show_reg("PSL Programming Control", val);
254 
255 	pci_read_config_dword(dev, vsec + 0x48, &val);
256 	show_reg("Reserved", val);
257 	pci_read_config_dword(dev, vsec + 0x4c, &val);
258 	show_reg("Reserved", val);
259 
260 	pci_read_config_dword(dev, vsec + 0x50, &val);
261 	show_reg("Flash Address Register", val);
262 	pci_read_config_dword(dev, vsec + 0x54, &val);
263 	show_reg("Flash Size Register", val);
264 	pci_read_config_dword(dev, vsec + 0x58, &val);
265 	show_reg("Flash Status/Control Register", val);
266 	pci_read_config_dword(dev, vsec + 0x58, &val);
267 	show_reg("Flash Data Port", val);
268 
269 #undef show_reg
270 }
271 
272 static void dump_afu_descriptor(struct cxl_afu *afu)
273 {
274 	u64 val, afu_cr_num, afu_cr_off, afu_cr_len;
275 	int i;
276 
277 #define show_reg(name, what) \
278 	dev_info(&afu->dev, "afu desc: %30s: %#llx\n", name, what)
279 
280 	val = AFUD_READ_INFO(afu);
281 	show_reg("num_ints_per_process", AFUD_NUM_INTS_PER_PROC(val));
282 	show_reg("num_of_processes", AFUD_NUM_PROCS(val));
283 	show_reg("num_of_afu_CRs", AFUD_NUM_CRS(val));
284 	show_reg("req_prog_mode", val & 0xffffULL);
285 	afu_cr_num = AFUD_NUM_CRS(val);
286 
287 	val = AFUD_READ(afu, 0x8);
288 	show_reg("Reserved", val);
289 	val = AFUD_READ(afu, 0x10);
290 	show_reg("Reserved", val);
291 	val = AFUD_READ(afu, 0x18);
292 	show_reg("Reserved", val);
293 
294 	val = AFUD_READ_CR(afu);
295 	show_reg("Reserved", (val >> (63-7)) & 0xff);
296 	show_reg("AFU_CR_len", AFUD_CR_LEN(val));
297 	afu_cr_len = AFUD_CR_LEN(val) * 256;
298 
299 	val = AFUD_READ_CR_OFF(afu);
300 	afu_cr_off = val;
301 	show_reg("AFU_CR_offset", val);
302 
303 	val = AFUD_READ_PPPSA(afu);
304 	show_reg("PerProcessPSA_control", (val >> (63-7)) & 0xff);
305 	show_reg("PerProcessPSA Length", AFUD_PPPSA_LEN(val));
306 
307 	val = AFUD_READ_PPPSA_OFF(afu);
308 	show_reg("PerProcessPSA_offset", val);
309 
310 	val = AFUD_READ_EB(afu);
311 	show_reg("Reserved", (val >> (63-7)) & 0xff);
312 	show_reg("AFU_EB_len", AFUD_EB_LEN(val));
313 
314 	val = AFUD_READ_EB_OFF(afu);
315 	show_reg("AFU_EB_offset", val);
316 
317 	for (i = 0; i < afu_cr_num; i++) {
318 		val = AFUD_READ_LE(afu, afu_cr_off + i * afu_cr_len);
319 		show_reg("CR Vendor", val & 0xffff);
320 		show_reg("CR Device", (val >> 16) & 0xffff);
321 	}
322 #undef show_reg
323 }
324 
325 #define P8_CAPP_UNIT0_ID 0xBA
326 #define P8_CAPP_UNIT1_ID 0XBE
327 #define P9_CAPP_UNIT0_ID 0xC0
328 #define P9_CAPP_UNIT1_ID 0xE0
329 
330 static int get_phb_index(struct device_node *np, u32 *phb_index)
331 {
332 	if (of_property_read_u32(np, "ibm,phb-index", phb_index))
333 		return -ENODEV;
334 	return 0;
335 }
336 
337 static u64 get_capp_unit_id(struct device_node *np, u32 phb_index)
338 {
339 	/*
340 	 * POWER 8:
341 	 *  - For chips other than POWER8NVL, we only have CAPP 0,
342 	 *    irrespective of which PHB is used.
343 	 *  - For POWER8NVL, assume CAPP 0 is attached to PHB0 and
344 	 *    CAPP 1 is attached to PHB1.
345 	 */
346 	if (cxl_is_power8()) {
347 		if (!pvr_version_is(PVR_POWER8NVL))
348 			return P8_CAPP_UNIT0_ID;
349 
350 		if (phb_index == 0)
351 			return P8_CAPP_UNIT0_ID;
352 
353 		if (phb_index == 1)
354 			return P8_CAPP_UNIT1_ID;
355 	}
356 
357 	/*
358 	 * POWER 9:
359 	 *   PEC0 (PHB0). Capp ID = CAPP0 (0b1100_0000)
360 	 *   PEC1 (PHB1 - PHB2). No capi mode
361 	 *   PEC2 (PHB3 - PHB4 - PHB5): Capi mode on PHB3 only. Capp ID = CAPP1 (0b1110_0000)
362 	 */
363 	if (cxl_is_power9()) {
364 		if (phb_index == 0)
365 			return P9_CAPP_UNIT0_ID;
366 
367 		if (phb_index == 3)
368 			return P9_CAPP_UNIT1_ID;
369 	}
370 
371 	return 0;
372 }
373 
374 int cxl_calc_capp_routing(struct pci_dev *dev, u64 *chipid,
375 			     u32 *phb_index, u64 *capp_unit_id)
376 {
377 	int rc;
378 	struct device_node *np;
379 	const __be32 *prop;
380 
381 	if (!(np = pnv_pci_get_phb_node(dev)))
382 		return -ENODEV;
383 
384 	while (np && !(prop = of_get_property(np, "ibm,chip-id", NULL)))
385 		np = of_get_next_parent(np);
386 	if (!np)
387 		return -ENODEV;
388 
389 	*chipid = be32_to_cpup(prop);
390 
391 	rc = get_phb_index(np, phb_index);
392 	if (rc) {
393 		pr_err("cxl: invalid phb index\n");
394 		return rc;
395 	}
396 
397 	*capp_unit_id = get_capp_unit_id(np, *phb_index);
398 	of_node_put(np);
399 	if (!*capp_unit_id) {
400 		pr_err("cxl: invalid capp unit id (phb_index: %d)\n",
401 		       *phb_index);
402 		return -ENODEV;
403 	}
404 
405 	return 0;
406 }
407 
408 static DEFINE_MUTEX(indications_mutex);
409 
410 static int get_phb_indications(struct pci_dev *dev, u64 *capiind, u64 *asnind,
411 			       u64 *nbwind)
412 {
413 	static u64 nbw, asn, capi = 0;
414 	struct device_node *np;
415 	const __be32 *prop;
416 
417 	mutex_lock(&indications_mutex);
418 	if (!capi) {
419 		if (!(np = pnv_pci_get_phb_node(dev))) {
420 			mutex_unlock(&indications_mutex);
421 			return -ENODEV;
422 		}
423 
424 		prop = of_get_property(np, "ibm,phb-indications", NULL);
425 		if (!prop) {
426 			nbw = 0x0300UL; /* legacy values */
427 			asn = 0x0400UL;
428 			capi = 0x0200UL;
429 		} else {
430 			nbw = (u64)be32_to_cpu(prop[2]);
431 			asn = (u64)be32_to_cpu(prop[1]);
432 			capi = (u64)be32_to_cpu(prop[0]);
433 		}
434 		of_node_put(np);
435 	}
436 	*capiind = capi;
437 	*asnind = asn;
438 	*nbwind = nbw;
439 	mutex_unlock(&indications_mutex);
440 	return 0;
441 }
442 
443 int cxl_get_xsl9_dsnctl(struct pci_dev *dev, u64 capp_unit_id, u64 *reg)
444 {
445 	u64 xsl_dsnctl;
446 	u64 capiind, asnind, nbwind;
447 
448 	/*
449 	 * CAPI Identifier bits [0:7]
450 	 * bit 61:60 MSI bits --> 0
451 	 * bit 59 TVT selector --> 0
452 	 */
453 	if (get_phb_indications(dev, &capiind, &asnind, &nbwind))
454 		return -ENODEV;
455 
456 	/*
457 	 * Tell XSL where to route data to.
458 	 * The field chipid should match the PHB CAPI_CMPM register
459 	 */
460 	xsl_dsnctl = (capiind << (63-15)); /* Bit 57 */
461 	xsl_dsnctl |= (capp_unit_id << (63-15));
462 
463 	/* nMMU_ID Defaults to: b’000001001’*/
464 	xsl_dsnctl |= ((u64)0x09 << (63-28));
465 
466 	/*
467 	 * Used to identify CAPI packets which should be sorted into
468 	 * the Non-Blocking queues by the PHB. This field should match
469 	 * the PHB PBL_NBW_CMPM register
470 	 * nbwind=0x03, bits [57:58], must include capi indicator.
471 	 * Not supported on P9 DD1.
472 	 */
473 	xsl_dsnctl |= (nbwind << (63-55));
474 
475 	/*
476 	 * Upper 16b address bits of ASB_Notify messages sent to the
477 	 * system. Need to match the PHB’s ASN Compare/Mask Register.
478 	 * Not supported on P9 DD1.
479 	 */
480 	xsl_dsnctl |= asnind;
481 
482 	*reg = xsl_dsnctl;
483 	return 0;
484 }
485 
486 static int init_implementation_adapter_regs_psl9(struct cxl *adapter,
487 						 struct pci_dev *dev)
488 {
489 	u64 xsl_dsnctl, psl_fircntl;
490 	u64 chipid;
491 	u32 phb_index;
492 	u64 capp_unit_id;
493 	u64 psl_debug;
494 	int rc;
495 
496 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
497 	if (rc)
498 		return rc;
499 
500 	rc = cxl_get_xsl9_dsnctl(dev, capp_unit_id, &xsl_dsnctl);
501 	if (rc)
502 		return rc;
503 
504 	cxl_p1_write(adapter, CXL_XSL9_DSNCTL, xsl_dsnctl);
505 
506 	/* Set fir_cntl to recommended value for production env */
507 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
508 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
509 	psl_fircntl |= 0x1ULL; /* ce_thresh */
510 	cxl_p1_write(adapter, CXL_PSL9_FIR_CNTL, psl_fircntl);
511 
512 	/* Setup the PSL to transmit packets on the PCIe before the
513 	 * CAPP is enabled. Make sure that CAPP virtual machines are disabled
514 	 */
515 	cxl_p1_write(adapter, CXL_PSL9_DSNDCTL, 0x0001001000012A10ULL);
516 
517 	/*
518 	 * A response to an ASB_Notify request is returned by the
519 	 * system as an MMIO write to the address defined in
520 	 * the PSL_TNR_ADDR register.
521 	 * keep the Reset Value: 0x00020000E0000000
522 	 */
523 
524 	/* Enable XSL rty limit */
525 	cxl_p1_write(adapter, CXL_XSL9_DEF, 0x51F8000000000005ULL);
526 
527 	/* Change XSL_INV dummy read threshold */
528 	cxl_p1_write(adapter, CXL_XSL9_INV, 0x0000040007FFC200ULL);
529 
530 	if (phb_index == 3) {
531 		/* disable machines 31-47 and 20-27 for DMA */
532 		cxl_p1_write(adapter, CXL_PSL9_APCDEDTYPE, 0x40000FF3FFFF0000ULL);
533 	}
534 
535 	/* Snoop machines */
536 	cxl_p1_write(adapter, CXL_PSL9_APCDEDALLOC, 0x800F000200000000ULL);
537 
538 	/* Enable NORST and DD2 features */
539 	cxl_p1_write(adapter, CXL_PSL9_DEBUG, 0xC000000000000000ULL);
540 
541 	/*
542 	 * Check if PSL has data-cache. We need to flush adapter datacache
543 	 * when as its about to be removed.
544 	 */
545 	psl_debug = cxl_p1_read(adapter, CXL_PSL9_DEBUG);
546 	if (psl_debug & CXL_PSL_DEBUG_CDC) {
547 		dev_dbg(&dev->dev, "No data-cache present\n");
548 		adapter->native->no_data_cache = true;
549 	}
550 
551 	return 0;
552 }
553 
554 static int init_implementation_adapter_regs_psl8(struct cxl *adapter, struct pci_dev *dev)
555 {
556 	u64 psl_dsnctl, psl_fircntl;
557 	u64 chipid;
558 	u32 phb_index;
559 	u64 capp_unit_id;
560 	int rc;
561 
562 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
563 	if (rc)
564 		return rc;
565 
566 	psl_dsnctl = 0x0000900000000000ULL; /* pteupd ttype, scdone */
567 	psl_dsnctl |= (0x2ULL << (63-38)); /* MMIO hang pulse: 256 us */
568 	/* Tell PSL where to route data to */
569 	psl_dsnctl |= (chipid << (63-5));
570 	psl_dsnctl |= (capp_unit_id << (63-13));
571 
572 	cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
573 	cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
574 	/* snoop write mask */
575 	cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
576 	/* set fir_cntl to recommended value for production env */
577 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
578 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
579 	psl_fircntl |= 0x1ULL; /* ce_thresh */
580 	cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, psl_fircntl);
581 	/* for debugging with trace arrays */
582 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
583 
584 	return 0;
585 }
586 
587 /* PSL */
588 #define TBSYNC_CAL(n) (((u64)n & 0x7) << (63-3))
589 #define TBSYNC_CNT(n) (((u64)n & 0x7) << (63-6))
590 /* For the PSL this is a multiple for 0 < n <= 7: */
591 #define PSL_2048_250MHZ_CYCLES 1
592 
593 static void write_timebase_ctrl_psl8(struct cxl *adapter)
594 {
595 	cxl_p1_write(adapter, CXL_PSL_TB_CTLSTAT,
596 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
597 }
598 
599 static u64 timebase_read_psl9(struct cxl *adapter)
600 {
601 	return cxl_p1_read(adapter, CXL_PSL9_Timebase);
602 }
603 
604 static u64 timebase_read_psl8(struct cxl *adapter)
605 {
606 	return cxl_p1_read(adapter, CXL_PSL_Timebase);
607 }
608 
609 static void cxl_setup_psl_timebase(struct cxl *adapter, struct pci_dev *dev)
610 {
611 	struct device_node *np;
612 
613 	adapter->psl_timebase_synced = false;
614 
615 	if (!(np = pnv_pci_get_phb_node(dev)))
616 		return;
617 
618 	/* Do not fail when CAPP timebase sync is not supported by OPAL */
619 	of_node_get(np);
620 	if (! of_get_property(np, "ibm,capp-timebase-sync", NULL)) {
621 		of_node_put(np);
622 		dev_info(&dev->dev, "PSL timebase inactive: OPAL support missing\n");
623 		return;
624 	}
625 	of_node_put(np);
626 
627 	/*
628 	 * Setup PSL Timebase Control and Status register
629 	 * with the recommended Timebase Sync Count value
630 	 */
631 	if (adapter->native->sl_ops->write_timebase_ctrl)
632 		adapter->native->sl_ops->write_timebase_ctrl(adapter);
633 
634 	/* Enable PSL Timebase */
635 	cxl_p1_write(adapter, CXL_PSL_Control, 0x0000000000000000);
636 	cxl_p1_write(adapter, CXL_PSL_Control, CXL_PSL_Control_tb);
637 
638 	return;
639 }
640 
641 static int init_implementation_afu_regs_psl9(struct cxl_afu *afu)
642 {
643 	return 0;
644 }
645 
646 static int init_implementation_afu_regs_psl8(struct cxl_afu *afu)
647 {
648 	/* read/write masks for this slice */
649 	cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
650 	/* APC read/write masks for this slice */
651 	cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
652 	/* for debugging with trace arrays */
653 	cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
654 	cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
655 
656 	return 0;
657 }
658 
659 int cxl_pci_setup_irq(struct cxl *adapter, unsigned int hwirq,
660 		unsigned int virq)
661 {
662 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
663 
664 	return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
665 }
666 
667 int cxl_update_image_control(struct cxl *adapter)
668 {
669 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
670 	int rc;
671 	int vsec;
672 	u8 image_state;
673 
674 	if (!(vsec = find_cxl_vsec(dev))) {
675 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
676 		return -ENODEV;
677 	}
678 
679 	if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
680 		dev_err(&dev->dev, "failed to read image state: %i\n", rc);
681 		return rc;
682 	}
683 
684 	if (adapter->perst_loads_image)
685 		image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
686 	else
687 		image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
688 
689 	if (adapter->perst_select_user)
690 		image_state |= CXL_VSEC_PERST_SELECT_USER;
691 	else
692 		image_state &= ~CXL_VSEC_PERST_SELECT_USER;
693 
694 	if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
695 		dev_err(&dev->dev, "failed to update image control: %i\n", rc);
696 		return rc;
697 	}
698 
699 	return 0;
700 }
701 
702 int cxl_pci_alloc_one_irq(struct cxl *adapter)
703 {
704 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
705 
706 	return pnv_cxl_alloc_hwirqs(dev, 1);
707 }
708 
709 void cxl_pci_release_one_irq(struct cxl *adapter, int hwirq)
710 {
711 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
712 
713 	return pnv_cxl_release_hwirqs(dev, hwirq, 1);
714 }
715 
716 int cxl_pci_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
717 			struct cxl *adapter, unsigned int num)
718 {
719 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
720 
721 	return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
722 }
723 
724 void cxl_pci_release_irq_ranges(struct cxl_irq_ranges *irqs,
725 				struct cxl *adapter)
726 {
727 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
728 
729 	pnv_cxl_release_hwirq_ranges(irqs, dev);
730 }
731 
732 static int setup_cxl_bars(struct pci_dev *dev)
733 {
734 	/* Safety check in case we get backported to < 3.17 without M64 */
735 	if ((p1_base(dev) < 0x100000000ULL) ||
736 	    (p2_base(dev) < 0x100000000ULL)) {
737 		dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
738 		return -ENODEV;
739 	}
740 
741 	/*
742 	 * BAR 4/5 has a special meaning for CXL and must be programmed with a
743 	 * special value corresponding to the CXL protocol address range.
744 	 * For POWER 8/9 that means bits 48:49 must be set to 10
745 	 */
746 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
747 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
748 
749 	return 0;
750 }
751 
752 /* pciex node: ibm,opal-m64-window = <0x3d058 0x0 0x3d058 0x0 0x8 0x0>; */
753 static int switch_card_to_cxl(struct pci_dev *dev)
754 {
755 	int vsec;
756 	u8 val;
757 	int rc;
758 
759 	dev_info(&dev->dev, "switch card to CXL\n");
760 
761 	if (!(vsec = find_cxl_vsec(dev))) {
762 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
763 		return -ENODEV;
764 	}
765 
766 	if ((rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val))) {
767 		dev_err(&dev->dev, "failed to read current mode control: %i", rc);
768 		return rc;
769 	}
770 	val &= ~CXL_VSEC_PROTOCOL_MASK;
771 	val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
772 	if ((rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val))) {
773 		dev_err(&dev->dev, "failed to enable CXL protocol: %i", rc);
774 		return rc;
775 	}
776 	/*
777 	 * The CAIA spec (v0.12 11.6 Bi-modal Device Support) states
778 	 * we must wait 100ms after this mode switch before touching
779 	 * PCIe config space.
780 	 */
781 	msleep(100);
782 
783 	return 0;
784 }
785 
786 static int pci_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
787 {
788 	u64 p1n_base, p2n_base, afu_desc;
789 	const u64 p1n_size = 0x100;
790 	const u64 p2n_size = 0x1000;
791 
792 	p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
793 	p2n_base = p2_base(dev) + (afu->slice * p2n_size);
794 	afu->psn_phys = p2_base(dev) + (adapter->native->ps_off + (afu->slice * adapter->ps_size));
795 	afu_desc = p2_base(dev) + adapter->native->afu_desc_off + (afu->slice * adapter->native->afu_desc_size);
796 
797 	if (!(afu->native->p1n_mmio = ioremap(p1n_base, p1n_size)))
798 		goto err;
799 	if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
800 		goto err1;
801 	if (afu_desc) {
802 		if (!(afu->native->afu_desc_mmio = ioremap(afu_desc, adapter->native->afu_desc_size)))
803 			goto err2;
804 	}
805 
806 	return 0;
807 err2:
808 	iounmap(afu->p2n_mmio);
809 err1:
810 	iounmap(afu->native->p1n_mmio);
811 err:
812 	dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
813 	return -ENOMEM;
814 }
815 
816 static void pci_unmap_slice_regs(struct cxl_afu *afu)
817 {
818 	if (afu->p2n_mmio) {
819 		iounmap(afu->p2n_mmio);
820 		afu->p2n_mmio = NULL;
821 	}
822 	if (afu->native->p1n_mmio) {
823 		iounmap(afu->native->p1n_mmio);
824 		afu->native->p1n_mmio = NULL;
825 	}
826 	if (afu->native->afu_desc_mmio) {
827 		iounmap(afu->native->afu_desc_mmio);
828 		afu->native->afu_desc_mmio = NULL;
829 	}
830 }
831 
832 void cxl_pci_release_afu(struct device *dev)
833 {
834 	struct cxl_afu *afu = to_cxl_afu(dev);
835 
836 	pr_devel("%s\n", __func__);
837 
838 	idr_destroy(&afu->contexts_idr);
839 	cxl_release_spa(afu);
840 
841 	kfree(afu->native);
842 	kfree(afu);
843 }
844 
845 /* Expects AFU struct to have recently been zeroed out */
846 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
847 {
848 	u64 val;
849 
850 	val = AFUD_READ_INFO(afu);
851 	afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
852 	afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
853 	afu->crs_num = AFUD_NUM_CRS(val);
854 
855 	if (AFUD_AFU_DIRECTED(val))
856 		afu->modes_supported |= CXL_MODE_DIRECTED;
857 	if (AFUD_DEDICATED_PROCESS(val))
858 		afu->modes_supported |= CXL_MODE_DEDICATED;
859 	if (AFUD_TIME_SLICED(val))
860 		afu->modes_supported |= CXL_MODE_TIME_SLICED;
861 
862 	val = AFUD_READ_PPPSA(afu);
863 	afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
864 	afu->psa = AFUD_PPPSA_PSA(val);
865 	if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
866 		afu->native->pp_offset = AFUD_READ_PPPSA_OFF(afu);
867 
868 	val = AFUD_READ_CR(afu);
869 	afu->crs_len = AFUD_CR_LEN(val) * 256;
870 	afu->crs_offset = AFUD_READ_CR_OFF(afu);
871 
872 
873 	/* eb_len is in multiple of 4K */
874 	afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
875 	afu->eb_offset = AFUD_READ_EB_OFF(afu);
876 
877 	/* eb_off is 4K aligned so lower 12 bits are always zero */
878 	if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
879 		dev_warn(&afu->dev,
880 			 "Invalid AFU error buffer offset %Lx\n",
881 			 afu->eb_offset);
882 		dev_info(&afu->dev,
883 			 "Ignoring AFU error buffer in the descriptor\n");
884 		/* indicate that no afu buffer exists */
885 		afu->eb_len = 0;
886 	}
887 
888 	return 0;
889 }
890 
891 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
892 {
893 	int i, rc;
894 	u32 val;
895 
896 	if (afu->psa && afu->adapter->ps_size <
897 			(afu->native->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
898 		dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
899 		return -ENODEV;
900 	}
901 
902 	if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
903 		dev_warn(&afu->dev, "AFU uses pp_size(%#016llx) < PAGE_SIZE per-process PSA!\n", afu->pp_size);
904 
905 	for (i = 0; i < afu->crs_num; i++) {
906 		rc = cxl_ops->afu_cr_read32(afu, i, 0, &val);
907 		if (rc || val == 0) {
908 			dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
909 			return -EINVAL;
910 		}
911 	}
912 
913 	if ((afu->modes_supported & ~CXL_MODE_DEDICATED) && afu->max_procs_virtualised == 0) {
914 		/*
915 		 * We could also check this for the dedicated process model
916 		 * since the architecture indicates it should be set to 1, but
917 		 * in that case we ignore the value and I'd rather not risk
918 		 * breaking any existing dedicated process AFUs that left it as
919 		 * 0 (not that I'm aware of any). It is clearly an error for an
920 		 * AFU directed AFU to set this to 0, and would have previously
921 		 * triggered a bug resulting in the maximum not being enforced
922 		 * at all since idr_alloc treats 0 as no maximum.
923 		 */
924 		dev_err(&afu->dev, "AFU does not support any processes\n");
925 		return -EINVAL;
926 	}
927 
928 	return 0;
929 }
930 
931 static int sanitise_afu_regs_psl9(struct cxl_afu *afu)
932 {
933 	u64 reg;
934 
935 	/*
936 	 * Clear out any regs that contain either an IVTE or address or may be
937 	 * waiting on an acknowledgment to try to be a bit safer as we bring
938 	 * it online
939 	 */
940 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
941 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
942 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
943 		if (cxl_ops->afu_reset(afu))
944 			return -EIO;
945 		if (cxl_afu_disable(afu))
946 			return -EIO;
947 		if (cxl_psl_purge(afu))
948 			return -EIO;
949 	}
950 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
951 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
952 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
953 	if (reg) {
954 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
955 		if (reg & CXL_PSL9_DSISR_An_TF)
956 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
957 		else
958 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
959 	}
960 	if (afu->adapter->native->sl_ops->register_serr_irq) {
961 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
962 		if (reg) {
963 			if (reg & ~0x000000007fffffff)
964 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
965 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
966 		}
967 	}
968 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
969 	if (reg) {
970 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
971 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
972 	}
973 
974 	return 0;
975 }
976 
977 static int sanitise_afu_regs_psl8(struct cxl_afu *afu)
978 {
979 	u64 reg;
980 
981 	/*
982 	 * Clear out any regs that contain either an IVTE or address or may be
983 	 * waiting on an acknowledgement to try to be a bit safer as we bring
984 	 * it online
985 	 */
986 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
987 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
988 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
989 		if (cxl_ops->afu_reset(afu))
990 			return -EIO;
991 		if (cxl_afu_disable(afu))
992 			return -EIO;
993 		if (cxl_psl_purge(afu))
994 			return -EIO;
995 	}
996 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
997 	cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
998 	cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
999 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1000 	cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
1001 	cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
1002 	cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
1003 	cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
1004 	cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
1005 	cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
1006 	cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
1007 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1008 	if (reg) {
1009 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1010 		if (reg & CXL_PSL_DSISR_TRANS)
1011 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1012 		else
1013 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1014 	}
1015 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1016 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1017 		if (reg) {
1018 			if (reg & ~0xffff)
1019 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1020 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1021 		}
1022 	}
1023 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1024 	if (reg) {
1025 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1026 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1027 	}
1028 
1029 	return 0;
1030 }
1031 
1032 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
1033 /*
1034  * afu_eb_read:
1035  * Called from sysfs and reads the afu error info buffer. The h/w only supports
1036  * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
1037  * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
1038  */
1039 ssize_t cxl_pci_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
1040 				loff_t off, size_t count)
1041 {
1042 	loff_t aligned_start, aligned_end;
1043 	size_t aligned_length;
1044 	void *tbuf;
1045 	const void __iomem *ebuf = afu->native->afu_desc_mmio + afu->eb_offset;
1046 
1047 	if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
1048 		return 0;
1049 
1050 	/* calculate aligned read window */
1051 	count = min((size_t)(afu->eb_len - off), count);
1052 	aligned_start = round_down(off, 8);
1053 	aligned_end = round_up(off + count, 8);
1054 	aligned_length = aligned_end - aligned_start;
1055 
1056 	/* max we can copy in one read is PAGE_SIZE */
1057 	if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
1058 		aligned_length = ERR_BUFF_MAX_COPY_SIZE;
1059 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
1060 	}
1061 
1062 	/* use bounce buffer for copy */
1063 	tbuf = (void *)__get_free_page(GFP_KERNEL);
1064 	if (!tbuf)
1065 		return -ENOMEM;
1066 
1067 	/* perform aligned read from the mmio region */
1068 	memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
1069 	memcpy(buf, tbuf + (off & 0x7), count);
1070 
1071 	free_page((unsigned long)tbuf);
1072 
1073 	return count;
1074 }
1075 
1076 static int pci_configure_afu(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1077 {
1078 	int rc;
1079 
1080 	if ((rc = pci_map_slice_regs(afu, adapter, dev)))
1081 		return rc;
1082 
1083 	if (adapter->native->sl_ops->sanitise_afu_regs) {
1084 		rc = adapter->native->sl_ops->sanitise_afu_regs(afu);
1085 		if (rc)
1086 			goto err1;
1087 	}
1088 
1089 	/* We need to reset the AFU before we can read the AFU descriptor */
1090 	if ((rc = cxl_ops->afu_reset(afu)))
1091 		goto err1;
1092 
1093 	if (cxl_verbose)
1094 		dump_afu_descriptor(afu);
1095 
1096 	if ((rc = cxl_read_afu_descriptor(afu)))
1097 		goto err1;
1098 
1099 	if ((rc = cxl_afu_descriptor_looks_ok(afu)))
1100 		goto err1;
1101 
1102 	if (adapter->native->sl_ops->afu_regs_init)
1103 		if ((rc = adapter->native->sl_ops->afu_regs_init(afu)))
1104 			goto err1;
1105 
1106 	if (adapter->native->sl_ops->register_serr_irq)
1107 		if ((rc = adapter->native->sl_ops->register_serr_irq(afu)))
1108 			goto err1;
1109 
1110 	if ((rc = cxl_native_register_psl_irq(afu)))
1111 		goto err2;
1112 
1113 	atomic_set(&afu->configured_state, 0);
1114 	return 0;
1115 
1116 err2:
1117 	if (adapter->native->sl_ops->release_serr_irq)
1118 		adapter->native->sl_ops->release_serr_irq(afu);
1119 err1:
1120 	pci_unmap_slice_regs(afu);
1121 	return rc;
1122 }
1123 
1124 static void pci_deconfigure_afu(struct cxl_afu *afu)
1125 {
1126 	/*
1127 	 * It's okay to deconfigure when AFU is already locked, otherwise wait
1128 	 * until there are no readers
1129 	 */
1130 	if (atomic_read(&afu->configured_state) != -1) {
1131 		while (atomic_cmpxchg(&afu->configured_state, 0, -1) != -1)
1132 			schedule();
1133 	}
1134 	cxl_native_release_psl_irq(afu);
1135 	if (afu->adapter->native->sl_ops->release_serr_irq)
1136 		afu->adapter->native->sl_ops->release_serr_irq(afu);
1137 	pci_unmap_slice_regs(afu);
1138 }
1139 
1140 static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
1141 {
1142 	struct cxl_afu *afu;
1143 	int rc = -ENOMEM;
1144 
1145 	afu = cxl_alloc_afu(adapter, slice);
1146 	if (!afu)
1147 		return -ENOMEM;
1148 
1149 	afu->native = kzalloc(sizeof(struct cxl_afu_native), GFP_KERNEL);
1150 	if (!afu->native)
1151 		goto err_free_afu;
1152 
1153 	mutex_init(&afu->native->spa_mutex);
1154 
1155 	rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice);
1156 	if (rc)
1157 		goto err_free_native;
1158 
1159 	rc = pci_configure_afu(afu, adapter, dev);
1160 	if (rc)
1161 		goto err_free_native;
1162 
1163 	/* Don't care if this fails */
1164 	cxl_debugfs_afu_add(afu);
1165 
1166 	/*
1167 	 * After we call this function we must not free the afu directly, even
1168 	 * if it returns an error!
1169 	 */
1170 	if ((rc = cxl_register_afu(afu)))
1171 		goto err_put1;
1172 
1173 	if ((rc = cxl_sysfs_afu_add(afu)))
1174 		goto err_put1;
1175 
1176 	adapter->afu[afu->slice] = afu;
1177 
1178 	if ((rc = cxl_pci_vphb_add(afu)))
1179 		dev_info(&afu->dev, "Can't register vPHB\n");
1180 
1181 	return 0;
1182 
1183 err_put1:
1184 	pci_deconfigure_afu(afu);
1185 	cxl_debugfs_afu_remove(afu);
1186 	device_unregister(&afu->dev);
1187 	return rc;
1188 
1189 err_free_native:
1190 	kfree(afu->native);
1191 err_free_afu:
1192 	kfree(afu);
1193 	return rc;
1194 
1195 }
1196 
1197 static void cxl_pci_remove_afu(struct cxl_afu *afu)
1198 {
1199 	pr_devel("%s\n", __func__);
1200 
1201 	if (!afu)
1202 		return;
1203 
1204 	cxl_pci_vphb_remove(afu);
1205 	cxl_sysfs_afu_remove(afu);
1206 	cxl_debugfs_afu_remove(afu);
1207 
1208 	spin_lock(&afu->adapter->afu_list_lock);
1209 	afu->adapter->afu[afu->slice] = NULL;
1210 	spin_unlock(&afu->adapter->afu_list_lock);
1211 
1212 	cxl_context_detach_all(afu);
1213 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1214 
1215 	pci_deconfigure_afu(afu);
1216 	device_unregister(&afu->dev);
1217 }
1218 
1219 int cxl_pci_reset(struct cxl *adapter)
1220 {
1221 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1222 	int rc;
1223 
1224 	if (adapter->perst_same_image) {
1225 		dev_warn(&dev->dev,
1226 			 "cxl: refusing to reset/reflash when perst_reloads_same_image is set.\n");
1227 		return -EINVAL;
1228 	}
1229 
1230 	dev_info(&dev->dev, "CXL reset\n");
1231 
1232 	/*
1233 	 * The adapter is about to be reset, so ignore errors.
1234 	 */
1235 	cxl_data_cache_flush(adapter);
1236 
1237 	/* pcie_warm_reset requests a fundamental pci reset which includes a
1238 	 * PERST assert/deassert.  PERST triggers a loading of the image
1239 	 * if "user" or "factory" is selected in sysfs */
1240 	if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
1241 		dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
1242 		return rc;
1243 	}
1244 
1245 	return rc;
1246 }
1247 
1248 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
1249 {
1250 	if (pci_request_region(dev, 2, "priv 2 regs"))
1251 		goto err1;
1252 	if (pci_request_region(dev, 0, "priv 1 regs"))
1253 		goto err2;
1254 
1255 	pr_devel("cxl_map_adapter_regs: p1: %#016llx %#llx, p2: %#016llx %#llx",
1256 			p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
1257 
1258 	if (!(adapter->native->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
1259 		goto err3;
1260 
1261 	if (!(adapter->native->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
1262 		goto err4;
1263 
1264 	return 0;
1265 
1266 err4:
1267 	iounmap(adapter->native->p1_mmio);
1268 	adapter->native->p1_mmio = NULL;
1269 err3:
1270 	pci_release_region(dev, 0);
1271 err2:
1272 	pci_release_region(dev, 2);
1273 err1:
1274 	return -ENOMEM;
1275 }
1276 
1277 static void cxl_unmap_adapter_regs(struct cxl *adapter)
1278 {
1279 	if (adapter->native->p1_mmio) {
1280 		iounmap(adapter->native->p1_mmio);
1281 		adapter->native->p1_mmio = NULL;
1282 		pci_release_region(to_pci_dev(adapter->dev.parent), 2);
1283 	}
1284 	if (adapter->native->p2_mmio) {
1285 		iounmap(adapter->native->p2_mmio);
1286 		adapter->native->p2_mmio = NULL;
1287 		pci_release_region(to_pci_dev(adapter->dev.parent), 0);
1288 	}
1289 }
1290 
1291 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
1292 {
1293 	int vsec;
1294 	u32 afu_desc_off, afu_desc_size;
1295 	u32 ps_off, ps_size;
1296 	u16 vseclen;
1297 	u8 image_state;
1298 
1299 	if (!(vsec = find_cxl_vsec(dev))) {
1300 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
1301 		return -ENODEV;
1302 	}
1303 
1304 	CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
1305 	if (vseclen < CXL_VSEC_MIN_SIZE) {
1306 		dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
1307 		return -EINVAL;
1308 	}
1309 
1310 	CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
1311 	CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
1312 	CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
1313 	CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
1314 	CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
1315 	CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
1316 	adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1317 	adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1318 	adapter->perst_loads_image = !!(image_state & CXL_VSEC_PERST_LOADS_IMAGE);
1319 
1320 	CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
1321 	CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
1322 	CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
1323 	CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
1324 	CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
1325 
1326 	/* Convert everything to bytes, because there is NO WAY I'd look at the
1327 	 * code a month later and forget what units these are in ;-) */
1328 	adapter->native->ps_off = ps_off * 64 * 1024;
1329 	adapter->ps_size = ps_size * 64 * 1024;
1330 	adapter->native->afu_desc_off = afu_desc_off * 64 * 1024;
1331 	adapter->native->afu_desc_size = afu_desc_size * 64 * 1024;
1332 
1333 	/* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
1334 	adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
1335 
1336 	return 0;
1337 }
1338 
1339 /*
1340  * Workaround a PCIe Host Bridge defect on some cards, that can cause
1341  * malformed Transaction Layer Packet (TLP) errors to be erroneously
1342  * reported. Mask this error in the Uncorrectable Error Mask Register.
1343  *
1344  * The upper nibble of the PSL revision is used to distinguish between
1345  * different cards. The affected ones have it set to 0.
1346  */
1347 static void cxl_fixup_malformed_tlp(struct cxl *adapter, struct pci_dev *dev)
1348 {
1349 	int aer;
1350 	u32 data;
1351 
1352 	if (adapter->psl_rev & 0xf000)
1353 		return;
1354 	if (!(aer = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR)))
1355 		return;
1356 	pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &data);
1357 	if (data & PCI_ERR_UNC_MALF_TLP)
1358 		if (data & PCI_ERR_UNC_INTN)
1359 			return;
1360 	data |= PCI_ERR_UNC_MALF_TLP;
1361 	data |= PCI_ERR_UNC_INTN;
1362 	pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, data);
1363 }
1364 
1365 static bool cxl_compatible_caia_version(struct cxl *adapter)
1366 {
1367 	if (cxl_is_power8() && (adapter->caia_major == 1))
1368 		return true;
1369 
1370 	if (cxl_is_power9() && (adapter->caia_major == 2))
1371 		return true;
1372 
1373 	return false;
1374 }
1375 
1376 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
1377 {
1378 	if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
1379 		return -EBUSY;
1380 
1381 	if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
1382 		dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
1383 		return -EINVAL;
1384 	}
1385 
1386 	if (!cxl_compatible_caia_version(adapter)) {
1387 		dev_info(&dev->dev, "Ignoring card. PSL type is not supported (caia version: %d)\n",
1388 			 adapter->caia_major);
1389 		return -ENODEV;
1390 	}
1391 
1392 	if (!adapter->slices) {
1393 		/* Once we support dynamic reprogramming we can use the card if
1394 		 * it supports loadable AFUs */
1395 		dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
1396 		return -EINVAL;
1397 	}
1398 
1399 	if (!adapter->native->afu_desc_off || !adapter->native->afu_desc_size) {
1400 		dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
1401 		return -EINVAL;
1402 	}
1403 
1404 	if (adapter->ps_size > p2_size(dev) - adapter->native->ps_off) {
1405 		dev_err(&dev->dev, "ABORTING: Problem state size larger than "
1406 				   "available in BAR2: 0x%llx > 0x%llx\n",
1407 			 adapter->ps_size, p2_size(dev) - adapter->native->ps_off);
1408 		return -EINVAL;
1409 	}
1410 
1411 	return 0;
1412 }
1413 
1414 ssize_t cxl_pci_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1415 {
1416 	return pci_read_vpd(to_pci_dev(adapter->dev.parent), 0, len, buf);
1417 }
1418 
1419 static void cxl_release_adapter(struct device *dev)
1420 {
1421 	struct cxl *adapter = to_cxl_adapter(dev);
1422 
1423 	pr_devel("cxl_release_adapter\n");
1424 
1425 	cxl_remove_adapter_nr(adapter);
1426 
1427 	kfree(adapter->native);
1428 	kfree(adapter);
1429 }
1430 
1431 #define CXL_PSL_ErrIVTE_tberror (0x1ull << (63-31))
1432 
1433 static int sanitise_adapter_regs(struct cxl *adapter)
1434 {
1435 	int rc = 0;
1436 
1437 	/* Clear PSL tberror bit by writing 1 to it */
1438 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, CXL_PSL_ErrIVTE_tberror);
1439 
1440 	if (adapter->native->sl_ops->invalidate_all) {
1441 		/* do not invalidate ERAT entries when not reloading on PERST */
1442 		if (cxl_is_power9() && (adapter->perst_loads_image))
1443 			return 0;
1444 		rc = adapter->native->sl_ops->invalidate_all(adapter);
1445 	}
1446 
1447 	return rc;
1448 }
1449 
1450 /* This should contain *only* operations that can safely be done in
1451  * both creation and recovery.
1452  */
1453 static int cxl_configure_adapter(struct cxl *adapter, struct pci_dev *dev)
1454 {
1455 	int rc;
1456 
1457 	adapter->dev.parent = &dev->dev;
1458 	adapter->dev.release = cxl_release_adapter;
1459 	pci_set_drvdata(dev, adapter);
1460 
1461 	rc = pci_enable_device(dev);
1462 	if (rc) {
1463 		dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1464 		return rc;
1465 	}
1466 
1467 	if ((rc = cxl_read_vsec(adapter, dev)))
1468 		return rc;
1469 
1470 	if ((rc = cxl_vsec_looks_ok(adapter, dev)))
1471 	        return rc;
1472 
1473 	cxl_fixup_malformed_tlp(adapter, dev);
1474 
1475 	if ((rc = setup_cxl_bars(dev)))
1476 		return rc;
1477 
1478 	if ((rc = switch_card_to_cxl(dev)))
1479 		return rc;
1480 
1481 	if ((rc = cxl_update_image_control(adapter)))
1482 		return rc;
1483 
1484 	if ((rc = cxl_map_adapter_regs(adapter, dev)))
1485 		return rc;
1486 
1487 	if ((rc = sanitise_adapter_regs(adapter)))
1488 		goto err;
1489 
1490 	if ((rc = adapter->native->sl_ops->adapter_regs_init(adapter, dev)))
1491 		goto err;
1492 
1493 	/* Required for devices using CAPP DMA mode, harmless for others */
1494 	pci_set_master(dev);
1495 
1496 	adapter->tunneled_ops_supported = false;
1497 
1498 	if (cxl_is_power9()) {
1499 		if (pnv_pci_set_tunnel_bar(dev, 0x00020000E0000000ull, 1))
1500 			dev_info(&dev->dev, "Tunneled operations unsupported\n");
1501 		else
1502 			adapter->tunneled_ops_supported = true;
1503 	}
1504 
1505 	if ((rc = pnv_phb_to_cxl_mode(dev, adapter->native->sl_ops->capi_mode)))
1506 		goto err;
1507 
1508 	/* If recovery happened, the last step is to turn on snooping.
1509 	 * In the non-recovery case this has no effect */
1510 	if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON)))
1511 		goto err;
1512 
1513 	/* Ignore error, adapter init is not dependant on timebase sync */
1514 	cxl_setup_psl_timebase(adapter, dev);
1515 
1516 	if ((rc = cxl_native_register_psl_err_irq(adapter)))
1517 		goto err;
1518 
1519 	return 0;
1520 
1521 err:
1522 	cxl_unmap_adapter_regs(adapter);
1523 	return rc;
1524 
1525 }
1526 
1527 static void cxl_deconfigure_adapter(struct cxl *adapter)
1528 {
1529 	struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1530 
1531 	if (cxl_is_power9())
1532 		pnv_pci_set_tunnel_bar(pdev, 0x00020000E0000000ull, 0);
1533 
1534 	cxl_native_release_psl_err_irq(adapter);
1535 	cxl_unmap_adapter_regs(adapter);
1536 
1537 	pci_disable_device(pdev);
1538 }
1539 
1540 static void cxl_stop_trace_psl9(struct cxl *adapter)
1541 {
1542 	int traceid;
1543 	u64 trace_state, trace_mask;
1544 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1545 
1546 	/* read each tracearray state and issue mmio to stop them is needed */
1547 	for (traceid = 0; traceid <= CXL_PSL9_TRACEID_MAX; ++traceid) {
1548 		trace_state = cxl_p1_read(adapter, CXL_PSL9_CTCCFG);
1549 		trace_mask = (0x3ULL << (62 - traceid * 2));
1550 		trace_state = (trace_state & trace_mask) >> (62 - traceid * 2);
1551 		dev_dbg(&dev->dev, "cxl: Traceid-%d trace_state=0x%0llX\n",
1552 			traceid, trace_state);
1553 
1554 		/* issue mmio if the trace array isn't in FIN state */
1555 		if (trace_state != CXL_PSL9_TRACESTATE_FIN)
1556 			cxl_p1_write(adapter, CXL_PSL9_TRACECFG,
1557 				     0x8400000000000000ULL | traceid);
1558 	}
1559 }
1560 
1561 static void cxl_stop_trace_psl8(struct cxl *adapter)
1562 {
1563 	int slice;
1564 
1565 	/* Stop the trace */
1566 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x8000000000000017LL);
1567 
1568 	/* Stop the slice traces */
1569 	spin_lock(&adapter->afu_list_lock);
1570 	for (slice = 0; slice < adapter->slices; slice++) {
1571 		if (adapter->afu[slice])
1572 			cxl_p1n_write(adapter->afu[slice], CXL_PSL_SLICE_TRACE,
1573 				      0x8000000000000000LL);
1574 	}
1575 	spin_unlock(&adapter->afu_list_lock);
1576 }
1577 
1578 static const struct cxl_service_layer_ops psl9_ops = {
1579 	.adapter_regs_init = init_implementation_adapter_regs_psl9,
1580 	.invalidate_all = cxl_invalidate_all_psl9,
1581 	.afu_regs_init = init_implementation_afu_regs_psl9,
1582 	.sanitise_afu_regs = sanitise_afu_regs_psl9,
1583 	.register_serr_irq = cxl_native_register_serr_irq,
1584 	.release_serr_irq = cxl_native_release_serr_irq,
1585 	.handle_interrupt = cxl_irq_psl9,
1586 	.fail_irq = cxl_fail_irq_psl,
1587 	.activate_dedicated_process = cxl_activate_dedicated_process_psl9,
1588 	.attach_afu_directed = cxl_attach_afu_directed_psl9,
1589 	.attach_dedicated_process = cxl_attach_dedicated_process_psl9,
1590 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl9,
1591 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl9,
1592 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl9,
1593 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl9,
1594 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl9,
1595 	.debugfs_stop_trace = cxl_stop_trace_psl9,
1596 	.timebase_read = timebase_read_psl9,
1597 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1598 	.needs_reset_before_disable = true,
1599 };
1600 
1601 static const struct cxl_service_layer_ops psl8_ops = {
1602 	.adapter_regs_init = init_implementation_adapter_regs_psl8,
1603 	.invalidate_all = cxl_invalidate_all_psl8,
1604 	.afu_regs_init = init_implementation_afu_regs_psl8,
1605 	.sanitise_afu_regs = sanitise_afu_regs_psl8,
1606 	.register_serr_irq = cxl_native_register_serr_irq,
1607 	.release_serr_irq = cxl_native_release_serr_irq,
1608 	.handle_interrupt = cxl_irq_psl8,
1609 	.fail_irq = cxl_fail_irq_psl,
1610 	.activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1611 	.attach_afu_directed = cxl_attach_afu_directed_psl8,
1612 	.attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1613 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1614 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl8,
1615 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl8,
1616 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl8,
1617 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl8,
1618 	.debugfs_stop_trace = cxl_stop_trace_psl8,
1619 	.write_timebase_ctrl = write_timebase_ctrl_psl8,
1620 	.timebase_read = timebase_read_psl8,
1621 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1622 	.needs_reset_before_disable = true,
1623 };
1624 
1625 static void set_sl_ops(struct cxl *adapter, struct pci_dev *dev)
1626 {
1627 	if (cxl_is_power8()) {
1628 		dev_info(&dev->dev, "Device uses a PSL8\n");
1629 		adapter->native->sl_ops = &psl8_ops;
1630 	} else {
1631 		dev_info(&dev->dev, "Device uses a PSL9\n");
1632 		adapter->native->sl_ops = &psl9_ops;
1633 	}
1634 }
1635 
1636 
1637 static struct cxl *cxl_pci_init_adapter(struct pci_dev *dev)
1638 {
1639 	struct cxl *adapter;
1640 	int rc;
1641 
1642 	adapter = cxl_alloc_adapter();
1643 	if (!adapter)
1644 		return ERR_PTR(-ENOMEM);
1645 
1646 	adapter->native = kzalloc(sizeof(struct cxl_native), GFP_KERNEL);
1647 	if (!adapter->native) {
1648 		rc = -ENOMEM;
1649 		goto err_release;
1650 	}
1651 
1652 	set_sl_ops(adapter, dev);
1653 
1654 	/* Set defaults for parameters which need to persist over
1655 	 * configure/reconfigure
1656 	 */
1657 	adapter->perst_loads_image = true;
1658 	adapter->perst_same_image = false;
1659 
1660 	rc = cxl_configure_adapter(adapter, dev);
1661 	if (rc) {
1662 		pci_disable_device(dev);
1663 		goto err_release;
1664 	}
1665 
1666 	/* Don't care if this one fails: */
1667 	cxl_debugfs_adapter_add(adapter);
1668 
1669 	/*
1670 	 * After we call this function we must not free the adapter directly,
1671 	 * even if it returns an error!
1672 	 */
1673 	if ((rc = cxl_register_adapter(adapter)))
1674 		goto err_put1;
1675 
1676 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1677 		goto err_put1;
1678 
1679 	/* Release the context lock as adapter is configured */
1680 	cxl_adapter_context_unlock(adapter);
1681 
1682 	return adapter;
1683 
1684 err_put1:
1685 	/* This should mirror cxl_remove_adapter, except without the
1686 	 * sysfs parts
1687 	 */
1688 	cxl_debugfs_adapter_remove(adapter);
1689 	cxl_deconfigure_adapter(adapter);
1690 	device_unregister(&adapter->dev);
1691 	return ERR_PTR(rc);
1692 
1693 err_release:
1694 	cxl_release_adapter(&adapter->dev);
1695 	return ERR_PTR(rc);
1696 }
1697 
1698 static void cxl_pci_remove_adapter(struct cxl *adapter)
1699 {
1700 	pr_devel("cxl_remove_adapter\n");
1701 
1702 	cxl_sysfs_adapter_remove(adapter);
1703 	cxl_debugfs_adapter_remove(adapter);
1704 
1705 	/*
1706 	 * Flush adapter datacache as its about to be removed.
1707 	 */
1708 	cxl_data_cache_flush(adapter);
1709 
1710 	cxl_deconfigure_adapter(adapter);
1711 
1712 	device_unregister(&adapter->dev);
1713 }
1714 
1715 #define CXL_MAX_PCIEX_PARENT 2
1716 
1717 int cxl_slot_is_switched(struct pci_dev *dev)
1718 {
1719 	struct device_node *np;
1720 	int depth = 0;
1721 
1722 	if (!(np = pci_device_to_OF_node(dev))) {
1723 		pr_err("cxl: np = NULL\n");
1724 		return -ENODEV;
1725 	}
1726 	of_node_get(np);
1727 	while (np) {
1728 		np = of_get_next_parent(np);
1729 		if (!of_node_is_type(np, "pciex"))
1730 			break;
1731 		depth++;
1732 	}
1733 	of_node_put(np);
1734 	return (depth > CXL_MAX_PCIEX_PARENT);
1735 }
1736 
1737 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
1738 {
1739 	struct cxl *adapter;
1740 	int slice;
1741 	int rc;
1742 
1743 	if (cxl_pci_is_vphb_device(dev)) {
1744 		dev_dbg(&dev->dev, "cxl_init_adapter: Ignoring cxl vphb device\n");
1745 		return -ENODEV;
1746 	}
1747 
1748 	if (cxl_slot_is_switched(dev)) {
1749 		dev_info(&dev->dev, "Ignoring card on incompatible PCI slot\n");
1750 		return -ENODEV;
1751 	}
1752 
1753 	if (cxl_is_power9() && !radix_enabled()) {
1754 		dev_info(&dev->dev, "Only Radix mode supported\n");
1755 		return -ENODEV;
1756 	}
1757 
1758 	if (cxl_verbose)
1759 		dump_cxl_config_space(dev);
1760 
1761 	adapter = cxl_pci_init_adapter(dev);
1762 	if (IS_ERR(adapter)) {
1763 		dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
1764 		return PTR_ERR(adapter);
1765 	}
1766 
1767 	for (slice = 0; slice < adapter->slices; slice++) {
1768 		if ((rc = pci_init_afu(adapter, slice, dev))) {
1769 			dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
1770 			continue;
1771 		}
1772 
1773 		rc = cxl_afu_select_best_mode(adapter->afu[slice]);
1774 		if (rc)
1775 			dev_err(&dev->dev, "AFU %i failed to start: %i\n", slice, rc);
1776 	}
1777 
1778 	return 0;
1779 }
1780 
1781 static void cxl_remove(struct pci_dev *dev)
1782 {
1783 	struct cxl *adapter = pci_get_drvdata(dev);
1784 	struct cxl_afu *afu;
1785 	int i;
1786 
1787 	/*
1788 	 * Lock to prevent someone grabbing a ref through the adapter list as
1789 	 * we are removing it
1790 	 */
1791 	for (i = 0; i < adapter->slices; i++) {
1792 		afu = adapter->afu[i];
1793 		cxl_pci_remove_afu(afu);
1794 	}
1795 	cxl_pci_remove_adapter(adapter);
1796 }
1797 
1798 static pci_ers_result_t cxl_vphb_error_detected(struct cxl_afu *afu,
1799 						pci_channel_state_t state)
1800 {
1801 	struct pci_dev *afu_dev;
1802 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
1803 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
1804 
1805 	/* There should only be one entry, but go through the list
1806 	 * anyway
1807 	 */
1808 	if (afu->phb == NULL)
1809 		return result;
1810 
1811 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
1812 		if (!afu_dev->driver)
1813 			continue;
1814 
1815 		afu_dev->error_state = state;
1816 
1817 		if (afu_dev->driver->err_handler)
1818 			afu_result = afu_dev->driver->err_handler->error_detected(afu_dev,
1819 										  state);
1820 		/* Disconnect trumps all, NONE trumps NEED_RESET */
1821 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
1822 			result = PCI_ERS_RESULT_DISCONNECT;
1823 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
1824 			 (result == PCI_ERS_RESULT_NEED_RESET))
1825 			result = PCI_ERS_RESULT_NONE;
1826 	}
1827 	return result;
1828 }
1829 
1830 static pci_ers_result_t cxl_pci_error_detected(struct pci_dev *pdev,
1831 					       pci_channel_state_t state)
1832 {
1833 	struct cxl *adapter = pci_get_drvdata(pdev);
1834 	struct cxl_afu *afu;
1835 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET, afu_result;
1836 	int i;
1837 
1838 	/* At this point, we could still have an interrupt pending.
1839 	 * Let's try to get them out of the way before they do
1840 	 * anything we don't like.
1841 	 */
1842 	schedule();
1843 
1844 	/* If we're permanently dead, give up. */
1845 	if (state == pci_channel_io_perm_failure) {
1846 		for (i = 0; i < adapter->slices; i++) {
1847 			afu = adapter->afu[i];
1848 			/*
1849 			 * Tell the AFU drivers; but we don't care what they
1850 			 * say, we're going away.
1851 			 */
1852 			cxl_vphb_error_detected(afu, state);
1853 		}
1854 		return PCI_ERS_RESULT_DISCONNECT;
1855 	}
1856 
1857 	/* Are we reflashing?
1858 	 *
1859 	 * If we reflash, we could come back as something entirely
1860 	 * different, including a non-CAPI card. As such, by default
1861 	 * we don't participate in the process. We'll be unbound and
1862 	 * the slot re-probed. (TODO: check EEH doesn't blindly rebind
1863 	 * us!)
1864 	 *
1865 	 * However, this isn't the entire story: for reliablity
1866 	 * reasons, we usually want to reflash the FPGA on PERST in
1867 	 * order to get back to a more reliable known-good state.
1868 	 *
1869 	 * This causes us a bit of a problem: if we reflash we can't
1870 	 * trust that we'll come back the same - we could have a new
1871 	 * image and been PERSTed in order to load that
1872 	 * image. However, most of the time we actually *will* come
1873 	 * back the same - for example a regular EEH event.
1874 	 *
1875 	 * Therefore, we allow the user to assert that the image is
1876 	 * indeed the same and that we should continue on into EEH
1877 	 * anyway.
1878 	 */
1879 	if (adapter->perst_loads_image && !adapter->perst_same_image) {
1880 		/* TODO take the PHB out of CXL mode */
1881 		dev_info(&pdev->dev, "reflashing, so opting out of EEH!\n");
1882 		return PCI_ERS_RESULT_NONE;
1883 	}
1884 
1885 	/*
1886 	 * At this point, we want to try to recover.  We'll always
1887 	 * need a complete slot reset: we don't trust any other reset.
1888 	 *
1889 	 * Now, we go through each AFU:
1890 	 *  - We send the driver, if bound, an error_detected callback.
1891 	 *    We expect it to clean up, but it can also tell us to give
1892 	 *    up and permanently detach the card. To simplify things, if
1893 	 *    any bound AFU driver doesn't support EEH, we give up on EEH.
1894 	 *
1895 	 *  - We detach all contexts associated with the AFU. This
1896 	 *    does not free them, but puts them into a CLOSED state
1897 	 *    which causes any the associated files to return useful
1898 	 *    errors to userland. It also unmaps, but does not free,
1899 	 *    any IRQs.
1900 	 *
1901 	 *  - We clean up our side: releasing and unmapping resources we hold
1902 	 *    so we can wire them up again when the hardware comes back up.
1903 	 *
1904 	 * Driver authors should note:
1905 	 *
1906 	 *  - Any contexts you create in your kernel driver (except
1907 	 *    those associated with anonymous file descriptors) are
1908 	 *    your responsibility to free and recreate. Likewise with
1909 	 *    any attached resources.
1910 	 *
1911 	 *  - We will take responsibility for re-initialising the
1912 	 *    device context (the one set up for you in
1913 	 *    cxl_pci_enable_device_hook and accessed through
1914 	 *    cxl_get_context). If you've attached IRQs or other
1915 	 *    resources to it, they remains yours to free.
1916 	 *
1917 	 * You can call the same functions to release resources as you
1918 	 * normally would: we make sure that these functions continue
1919 	 * to work when the hardware is down.
1920 	 *
1921 	 * Two examples:
1922 	 *
1923 	 * 1) If you normally free all your resources at the end of
1924 	 *    each request, or if you use anonymous FDs, your
1925 	 *    error_detected callback can simply set a flag to tell
1926 	 *    your driver not to start any new calls. You can then
1927 	 *    clear the flag in the resume callback.
1928 	 *
1929 	 * 2) If you normally allocate your resources on startup:
1930 	 *     * Set a flag in error_detected as above.
1931 	 *     * Let CXL detach your contexts.
1932 	 *     * In slot_reset, free the old resources and allocate new ones.
1933 	 *     * In resume, clear the flag to allow things to start.
1934 	 */
1935 	for (i = 0; i < adapter->slices; i++) {
1936 		afu = adapter->afu[i];
1937 
1938 		afu_result = cxl_vphb_error_detected(afu, state);
1939 
1940 		cxl_context_detach_all(afu);
1941 		cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1942 		pci_deconfigure_afu(afu);
1943 
1944 		/* Disconnect trumps all, NONE trumps NEED_RESET */
1945 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
1946 			result = PCI_ERS_RESULT_DISCONNECT;
1947 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
1948 			 (result == PCI_ERS_RESULT_NEED_RESET))
1949 			result = PCI_ERS_RESULT_NONE;
1950 	}
1951 
1952 	/* should take the context lock here */
1953 	if (cxl_adapter_context_lock(adapter) != 0)
1954 		dev_warn(&adapter->dev,
1955 			 "Couldn't take context lock with %d active-contexts\n",
1956 			 atomic_read(&adapter->contexts_num));
1957 
1958 	cxl_deconfigure_adapter(adapter);
1959 
1960 	return result;
1961 }
1962 
1963 static pci_ers_result_t cxl_pci_slot_reset(struct pci_dev *pdev)
1964 {
1965 	struct cxl *adapter = pci_get_drvdata(pdev);
1966 	struct cxl_afu *afu;
1967 	struct cxl_context *ctx;
1968 	struct pci_dev *afu_dev;
1969 	pci_ers_result_t afu_result = PCI_ERS_RESULT_RECOVERED;
1970 	pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
1971 	int i;
1972 
1973 	if (cxl_configure_adapter(adapter, pdev))
1974 		goto err;
1975 
1976 	/*
1977 	 * Unlock context activation for the adapter. Ideally this should be
1978 	 * done in cxl_pci_resume but cxlflash module tries to activate the
1979 	 * master context as part of slot_reset callback.
1980 	 */
1981 	cxl_adapter_context_unlock(adapter);
1982 
1983 	for (i = 0; i < adapter->slices; i++) {
1984 		afu = adapter->afu[i];
1985 
1986 		if (pci_configure_afu(afu, adapter, pdev))
1987 			goto err;
1988 
1989 		if (cxl_afu_select_best_mode(afu))
1990 			goto err;
1991 
1992 		if (afu->phb == NULL)
1993 			continue;
1994 
1995 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
1996 			/* Reset the device context.
1997 			 * TODO: make this less disruptive
1998 			 */
1999 			ctx = cxl_get_context(afu_dev);
2000 
2001 			if (ctx && cxl_release_context(ctx))
2002 				goto err;
2003 
2004 			ctx = cxl_dev_context_init(afu_dev);
2005 			if (IS_ERR(ctx))
2006 				goto err;
2007 
2008 			afu_dev->dev.archdata.cxl_ctx = ctx;
2009 
2010 			if (cxl_ops->afu_check_and_enable(afu))
2011 				goto err;
2012 
2013 			afu_dev->error_state = pci_channel_io_normal;
2014 
2015 			/* If there's a driver attached, allow it to
2016 			 * chime in on recovery. Drivers should check
2017 			 * if everything has come back OK, but
2018 			 * shouldn't start new work until we call
2019 			 * their resume function.
2020 			 */
2021 			if (!afu_dev->driver)
2022 				continue;
2023 
2024 			if (afu_dev->driver->err_handler &&
2025 			    afu_dev->driver->err_handler->slot_reset)
2026 				afu_result = afu_dev->driver->err_handler->slot_reset(afu_dev);
2027 
2028 			if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2029 				result = PCI_ERS_RESULT_DISCONNECT;
2030 		}
2031 	}
2032 	return result;
2033 
2034 err:
2035 	/* All the bits that happen in both error_detected and cxl_remove
2036 	 * should be idempotent, so we don't need to worry about leaving a mix
2037 	 * of unconfigured and reconfigured resources.
2038 	 */
2039 	dev_err(&pdev->dev, "EEH recovery failed. Asking to be disconnected.\n");
2040 	return PCI_ERS_RESULT_DISCONNECT;
2041 }
2042 
2043 static void cxl_pci_resume(struct pci_dev *pdev)
2044 {
2045 	struct cxl *adapter = pci_get_drvdata(pdev);
2046 	struct cxl_afu *afu;
2047 	struct pci_dev *afu_dev;
2048 	int i;
2049 
2050 	/* Everything is back now. Drivers should restart work now.
2051 	 * This is not the place to be checking if everything came back up
2052 	 * properly, because there's no return value: do that in slot_reset.
2053 	 */
2054 	for (i = 0; i < adapter->slices; i++) {
2055 		afu = adapter->afu[i];
2056 
2057 		if (afu->phb == NULL)
2058 			continue;
2059 
2060 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2061 			if (afu_dev->driver && afu_dev->driver->err_handler &&
2062 			    afu_dev->driver->err_handler->resume)
2063 				afu_dev->driver->err_handler->resume(afu_dev);
2064 		}
2065 	}
2066 }
2067 
2068 static const struct pci_error_handlers cxl_err_handler = {
2069 	.error_detected = cxl_pci_error_detected,
2070 	.slot_reset = cxl_pci_slot_reset,
2071 	.resume = cxl_pci_resume,
2072 };
2073 
2074 struct pci_driver cxl_pci_driver = {
2075 	.name = "cxl-pci",
2076 	.id_table = cxl_pci_tbl,
2077 	.probe = cxl_probe,
2078 	.remove = cxl_remove,
2079 	.shutdown = cxl_remove,
2080 	.err_handler = &cxl_err_handler,
2081 };
2082