xref: /openbmc/linux/drivers/misc/cxl/pci.c (revision 4c9ca2fd)
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 8/9 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(PCI_VENDOR_ID_IBM, 0x0623), },
127 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0628), },
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 P8_CAPP_UNIT0_ID 0xBA
328 #define P8_CAPP_UNIT1_ID 0XBE
329 #define P9_CAPP_UNIT0_ID 0xC0
330 #define P9_CAPP_UNIT1_ID 0xE0
331 
332 static int get_phb_index(struct device_node *np, u32 *phb_index)
333 {
334 	if (of_property_read_u32(np, "ibm,phb-index", phb_index))
335 		return -ENODEV;
336 	return 0;
337 }
338 
339 static u64 get_capp_unit_id(struct device_node *np, u32 phb_index)
340 {
341 	/*
342 	 * POWER 8:
343 	 *  - For chips other than POWER8NVL, we only have CAPP 0,
344 	 *    irrespective of which PHB is used.
345 	 *  - For POWER8NVL, assume CAPP 0 is attached to PHB0 and
346 	 *    CAPP 1 is attached to PHB1.
347 	 */
348 	if (cxl_is_power8()) {
349 		if (!pvr_version_is(PVR_POWER8NVL))
350 			return P8_CAPP_UNIT0_ID;
351 
352 		if (phb_index == 0)
353 			return P8_CAPP_UNIT0_ID;
354 
355 		if (phb_index == 1)
356 			return P8_CAPP_UNIT1_ID;
357 	}
358 
359 	/*
360 	 * POWER 9:
361 	 *   PEC0 (PHB0). Capp ID = CAPP0 (0b1100_0000)
362 	 *   PEC1 (PHB1 - PHB2). No capi mode
363 	 *   PEC2 (PHB3 - PHB4 - PHB5): Capi mode on PHB3 only. Capp ID = CAPP1 (0b1110_0000)
364 	 */
365 	if (cxl_is_power9()) {
366 		if (phb_index == 0)
367 			return P9_CAPP_UNIT0_ID;
368 
369 		if (phb_index == 3)
370 			return P9_CAPP_UNIT1_ID;
371 	}
372 
373 	return 0;
374 }
375 
376 int cxl_calc_capp_routing(struct pci_dev *dev, u64 *chipid,
377 			     u32 *phb_index, u64 *capp_unit_id)
378 {
379 	int rc;
380 	struct device_node *np;
381 	const __be32 *prop;
382 
383 	if (!(np = pnv_pci_get_phb_node(dev)))
384 		return -ENODEV;
385 
386 	while (np && !(prop = of_get_property(np, "ibm,chip-id", NULL)))
387 		np = of_get_next_parent(np);
388 	if (!np)
389 		return -ENODEV;
390 
391 	*chipid = be32_to_cpup(prop);
392 
393 	rc = get_phb_index(np, phb_index);
394 	if (rc) {
395 		pr_err("cxl: invalid phb index\n");
396 		return rc;
397 	}
398 
399 	*capp_unit_id = get_capp_unit_id(np, *phb_index);
400 	of_node_put(np);
401 	if (!*capp_unit_id) {
402 		pr_err("cxl: invalid capp unit id (phb_index: %d)\n",
403 		       *phb_index);
404 		return -ENODEV;
405 	}
406 
407 	return 0;
408 }
409 
410 static DEFINE_MUTEX(indications_mutex);
411 
412 static int get_phb_indications(struct pci_dev *dev, u64 *capiind, u64 *asnind,
413 			       u64 *nbwind)
414 {
415 	static u64 nbw, asn, capi = 0;
416 	struct device_node *np;
417 	const __be32 *prop;
418 
419 	mutex_lock(&indications_mutex);
420 	if (!capi) {
421 		if (!(np = pnv_pci_get_phb_node(dev))) {
422 			mutex_unlock(&indications_mutex);
423 			return -ENODEV;
424 		}
425 
426 		prop = of_get_property(np, "ibm,phb-indications", NULL);
427 		if (!prop) {
428 			nbw = 0x0300UL; /* legacy values */
429 			asn = 0x0400UL;
430 			capi = 0x0200UL;
431 		} else {
432 			nbw = (u64)be32_to_cpu(prop[2]);
433 			asn = (u64)be32_to_cpu(prop[1]);
434 			capi = (u64)be32_to_cpu(prop[0]);
435 		}
436 		of_node_put(np);
437 	}
438 	*capiind = capi;
439 	*asnind = asn;
440 	*nbwind = nbw;
441 	mutex_unlock(&indications_mutex);
442 	return 0;
443 }
444 
445 int cxl_get_xsl9_dsnctl(struct pci_dev *dev, u64 capp_unit_id, u64 *reg)
446 {
447 	u64 xsl_dsnctl;
448 	u64 capiind, asnind, nbwind;
449 
450 	/*
451 	 * CAPI Identifier bits [0:7]
452 	 * bit 61:60 MSI bits --> 0
453 	 * bit 59 TVT selector --> 0
454 	 */
455 	if (get_phb_indications(dev, &capiind, &asnind, &nbwind))
456 		return -ENODEV;
457 
458 	/*
459 	 * Tell XSL where to route data to.
460 	 * The field chipid should match the PHB CAPI_CMPM register
461 	 */
462 	xsl_dsnctl = (capiind << (63-15)); /* Bit 57 */
463 	xsl_dsnctl |= (capp_unit_id << (63-15));
464 
465 	/* nMMU_ID Defaults to: b’000001001’*/
466 	xsl_dsnctl |= ((u64)0x09 << (63-28));
467 
468 	if (!(cxl_is_power9_dd1())) {
469 		/*
470 		 * Used to identify CAPI packets which should be sorted into
471 		 * the Non-Blocking queues by the PHB. This field should match
472 		 * the PHB PBL_NBW_CMPM register
473 		 * nbwind=0x03, bits [57:58], must include capi indicator.
474 		 * Not supported on P9 DD1.
475 		 */
476 		xsl_dsnctl |= (nbwind << (63-55));
477 
478 		/*
479 		 * Upper 16b address bits of ASB_Notify messages sent to the
480 		 * system. Need to match the PHB’s ASN Compare/Mask Register.
481 		 * Not supported on P9 DD1.
482 		 */
483 		xsl_dsnctl |= asnind;
484 	}
485 
486 	*reg = xsl_dsnctl;
487 	return 0;
488 }
489 
490 static int init_implementation_adapter_regs_psl9(struct cxl *adapter,
491 						 struct pci_dev *dev)
492 {
493 	u64 xsl_dsnctl, psl_fircntl;
494 	u64 chipid;
495 	u32 phb_index;
496 	u64 capp_unit_id;
497 	u64 psl_debug;
498 	int rc;
499 
500 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
501 	if (rc)
502 		return rc;
503 
504 	rc = cxl_get_xsl9_dsnctl(dev, capp_unit_id, &xsl_dsnctl);
505 	if (rc)
506 		return rc;
507 
508 	cxl_p1_write(adapter, CXL_XSL9_DSNCTL, xsl_dsnctl);
509 
510 	/* Set fir_cntl to recommended value for production env */
511 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
512 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
513 	psl_fircntl |= 0x1ULL; /* ce_thresh */
514 	cxl_p1_write(adapter, CXL_PSL9_FIR_CNTL, psl_fircntl);
515 
516 	/* Setup the PSL to transmit packets on the PCIe before the
517 	 * CAPP is enabled. Make sure that CAPP virtual machines are disabled
518 	 */
519 	cxl_p1_write(adapter, CXL_PSL9_DSNDCTL, 0x0001001000012A10ULL);
520 
521 	/*
522 	 * A response to an ASB_Notify request is returned by the
523 	 * system as an MMIO write to the address defined in
524 	 * the PSL_TNR_ADDR register.
525 	 * keep the Reset Value: 0x00020000E0000000
526 	 */
527 
528 	/* Enable XSL rty limit */
529 	cxl_p1_write(adapter, CXL_XSL9_DEF, 0x51F8000000000005ULL);
530 
531 	/* Change XSL_INV dummy read threshold */
532 	cxl_p1_write(adapter, CXL_XSL9_INV, 0x0000040007FFC200ULL);
533 
534 	if (phb_index == 3) {
535 		/* disable machines 31-47 and 20-27 for DMA */
536 		cxl_p1_write(adapter, CXL_PSL9_APCDEDTYPE, 0x40000FF3FFFF0000ULL);
537 	}
538 
539 	/* Snoop machines */
540 	cxl_p1_write(adapter, CXL_PSL9_APCDEDALLOC, 0x800F000200000000ULL);
541 
542 	if (cxl_is_power9_dd1()) {
543 		/* Disabling deadlock counter CAR */
544 		cxl_p1_write(adapter, CXL_PSL9_GP_CT, 0x0020000000000001ULL);
545 		/* Enable NORST */
546 		cxl_p1_write(adapter, CXL_PSL9_DEBUG, 0x8000000000000000ULL);
547 	} else {
548 		/* Enable NORST and DD2 features */
549 		cxl_p1_write(adapter, CXL_PSL9_DEBUG, 0xC000000000000000ULL);
550 	}
551 
552 	/*
553 	 * Check if PSL has data-cache. We need to flush adapter datacache
554 	 * when as its about to be removed.
555 	 */
556 	psl_debug = cxl_p1_read(adapter, CXL_PSL9_DEBUG);
557 	if (psl_debug & CXL_PSL_DEBUG_CDC) {
558 		dev_dbg(&dev->dev, "No data-cache present\n");
559 		adapter->native->no_data_cache = true;
560 	}
561 
562 	return 0;
563 }
564 
565 static int init_implementation_adapter_regs_psl8(struct cxl *adapter, struct pci_dev *dev)
566 {
567 	u64 psl_dsnctl, psl_fircntl;
568 	u64 chipid;
569 	u32 phb_index;
570 	u64 capp_unit_id;
571 	int rc;
572 
573 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
574 	if (rc)
575 		return rc;
576 
577 	psl_dsnctl = 0x0000900000000000ULL; /* pteupd ttype, scdone */
578 	psl_dsnctl |= (0x2ULL << (63-38)); /* MMIO hang pulse: 256 us */
579 	/* Tell PSL where to route data to */
580 	psl_dsnctl |= (chipid << (63-5));
581 	psl_dsnctl |= (capp_unit_id << (63-13));
582 
583 	cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
584 	cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
585 	/* snoop write mask */
586 	cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
587 	/* set fir_cntl to recommended value for production env */
588 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
589 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
590 	psl_fircntl |= 0x1ULL; /* ce_thresh */
591 	cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, psl_fircntl);
592 	/* for debugging with trace arrays */
593 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
594 
595 	return 0;
596 }
597 
598 static int init_implementation_adapter_regs_xsl(struct cxl *adapter, struct pci_dev *dev)
599 {
600 	u64 xsl_dsnctl;
601 	u64 chipid;
602 	u32 phb_index;
603 	u64 capp_unit_id;
604 	int rc;
605 
606 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
607 	if (rc)
608 		return rc;
609 
610 	/* Tell XSL where to route data to */
611 	xsl_dsnctl = 0x0000600000000000ULL | (chipid << (63-5));
612 	xsl_dsnctl |= (capp_unit_id << (63-13));
613 	cxl_p1_write(adapter, CXL_XSL_DSNCTL, xsl_dsnctl);
614 
615 	return 0;
616 }
617 
618 /* PSL & XSL */
619 #define TBSYNC_CAL(n) (((u64)n & 0x7) << (63-3))
620 #define TBSYNC_CNT(n) (((u64)n & 0x7) << (63-6))
621 /* For the PSL this is a multiple for 0 < n <= 7: */
622 #define PSL_2048_250MHZ_CYCLES 1
623 
624 static void write_timebase_ctrl_psl8(struct cxl *adapter)
625 {
626 	cxl_p1_write(adapter, CXL_PSL_TB_CTLSTAT,
627 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
628 }
629 
630 /* XSL */
631 #define TBSYNC_ENA (1ULL << 63)
632 /* For the XSL this is 2**n * 2000 clocks for 0 < n <= 6: */
633 #define XSL_2000_CLOCKS 1
634 #define XSL_4000_CLOCKS 2
635 #define XSL_8000_CLOCKS 3
636 
637 static void write_timebase_ctrl_xsl(struct cxl *adapter)
638 {
639 	cxl_p1_write(adapter, CXL_XSL_TB_CTLSTAT,
640 		     TBSYNC_ENA |
641 		     TBSYNC_CAL(3) |
642 		     TBSYNC_CNT(XSL_4000_CLOCKS));
643 }
644 
645 static u64 timebase_read_psl9(struct cxl *adapter)
646 {
647 	return cxl_p1_read(adapter, CXL_PSL9_Timebase);
648 }
649 
650 static u64 timebase_read_psl8(struct cxl *adapter)
651 {
652 	return cxl_p1_read(adapter, CXL_PSL_Timebase);
653 }
654 
655 static u64 timebase_read_xsl(struct cxl *adapter)
656 {
657 	return cxl_p1_read(adapter, CXL_XSL_Timebase);
658 }
659 
660 static void cxl_setup_psl_timebase(struct cxl *adapter, struct pci_dev *dev)
661 {
662 	struct device_node *np;
663 
664 	adapter->psl_timebase_synced = false;
665 
666 	if (!(np = pnv_pci_get_phb_node(dev)))
667 		return;
668 
669 	/* Do not fail when CAPP timebase sync is not supported by OPAL */
670 	of_node_get(np);
671 	if (! of_get_property(np, "ibm,capp-timebase-sync", NULL)) {
672 		of_node_put(np);
673 		dev_info(&dev->dev, "PSL timebase inactive: OPAL support missing\n");
674 		return;
675 	}
676 	of_node_put(np);
677 
678 	/*
679 	 * Setup PSL Timebase Control and Status register
680 	 * with the recommended Timebase Sync Count value
681 	 */
682 	if (adapter->native->sl_ops->write_timebase_ctrl)
683 		adapter->native->sl_ops->write_timebase_ctrl(adapter);
684 
685 	/* Enable PSL Timebase */
686 	cxl_p1_write(adapter, CXL_PSL_Control, 0x0000000000000000);
687 	cxl_p1_write(adapter, CXL_PSL_Control, CXL_PSL_Control_tb);
688 
689 	return;
690 }
691 
692 static int init_implementation_afu_regs_psl9(struct cxl_afu *afu)
693 {
694 	return 0;
695 }
696 
697 static int init_implementation_afu_regs_psl8(struct cxl_afu *afu)
698 {
699 	/* read/write masks for this slice */
700 	cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
701 	/* APC read/write masks for this slice */
702 	cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
703 	/* for debugging with trace arrays */
704 	cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
705 	cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
706 
707 	return 0;
708 }
709 
710 int cxl_pci_setup_irq(struct cxl *adapter, unsigned int hwirq,
711 		unsigned int virq)
712 {
713 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
714 
715 	return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
716 }
717 
718 int cxl_update_image_control(struct cxl *adapter)
719 {
720 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
721 	int rc;
722 	int vsec;
723 	u8 image_state;
724 
725 	if (!(vsec = find_cxl_vsec(dev))) {
726 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
727 		return -ENODEV;
728 	}
729 
730 	if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
731 		dev_err(&dev->dev, "failed to read image state: %i\n", rc);
732 		return rc;
733 	}
734 
735 	if (adapter->perst_loads_image)
736 		image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
737 	else
738 		image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
739 
740 	if (adapter->perst_select_user)
741 		image_state |= CXL_VSEC_PERST_SELECT_USER;
742 	else
743 		image_state &= ~CXL_VSEC_PERST_SELECT_USER;
744 
745 	if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
746 		dev_err(&dev->dev, "failed to update image control: %i\n", rc);
747 		return rc;
748 	}
749 
750 	return 0;
751 }
752 
753 int cxl_pci_alloc_one_irq(struct cxl *adapter)
754 {
755 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
756 
757 	return pnv_cxl_alloc_hwirqs(dev, 1);
758 }
759 
760 void cxl_pci_release_one_irq(struct cxl *adapter, int hwirq)
761 {
762 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
763 
764 	return pnv_cxl_release_hwirqs(dev, hwirq, 1);
765 }
766 
767 int cxl_pci_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
768 			struct cxl *adapter, unsigned int num)
769 {
770 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
771 
772 	return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
773 }
774 
775 void cxl_pci_release_irq_ranges(struct cxl_irq_ranges *irqs,
776 				struct cxl *adapter)
777 {
778 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
779 
780 	pnv_cxl_release_hwirq_ranges(irqs, dev);
781 }
782 
783 static int setup_cxl_bars(struct pci_dev *dev)
784 {
785 	/* Safety check in case we get backported to < 3.17 without M64 */
786 	if ((p1_base(dev) < 0x100000000ULL) ||
787 	    (p2_base(dev) < 0x100000000ULL)) {
788 		dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
789 		return -ENODEV;
790 	}
791 
792 	/*
793 	 * BAR 4/5 has a special meaning for CXL and must be programmed with a
794 	 * special value corresponding to the CXL protocol address range.
795 	 * For POWER 8/9 that means bits 48:49 must be set to 10
796 	 */
797 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
798 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
799 
800 	return 0;
801 }
802 
803 #ifdef CONFIG_CXL_BIMODAL
804 
805 struct cxl_switch_work {
806 	struct pci_dev *dev;
807 	struct work_struct work;
808 	int vsec;
809 	int mode;
810 };
811 
812 static void switch_card_to_cxl(struct work_struct *work)
813 {
814 	struct cxl_switch_work *switch_work =
815 		container_of(work, struct cxl_switch_work, work);
816 	struct pci_dev *dev = switch_work->dev;
817 	struct pci_bus *bus = dev->bus;
818 	struct pci_controller *hose = pci_bus_to_host(bus);
819 	struct pci_dev *bridge;
820 	struct pnv_php_slot *php_slot;
821 	unsigned int devfn;
822 	u8 val;
823 	int rc;
824 
825 	dev_info(&bus->dev, "cxl: Preparing for mode switch...\n");
826 	bridge = list_first_entry_or_null(&hose->bus->devices, struct pci_dev,
827 					  bus_list);
828 	if (!bridge) {
829 		dev_WARN(&bus->dev, "cxl: Couldn't find root port!\n");
830 		goto err_dev_put;
831 	}
832 
833 	php_slot = pnv_php_find_slot(pci_device_to_OF_node(bridge));
834 	if (!php_slot) {
835 		dev_err(&bus->dev, "cxl: Failed to find slot hotplug "
836 			           "information. You may need to upgrade "
837 			           "skiboot. Aborting.\n");
838 		goto err_dev_put;
839 	}
840 
841 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, switch_work->vsec, &val);
842 	if (rc) {
843 		dev_err(&bus->dev, "cxl: Failed to read CAPI mode control: %i\n", rc);
844 		goto err_dev_put;
845 	}
846 	devfn = dev->devfn;
847 
848 	/* Release the reference obtained in cxl_check_and_switch_mode() */
849 	pci_dev_put(dev);
850 
851 	dev_dbg(&bus->dev, "cxl: Removing PCI devices from kernel\n");
852 	pci_lock_rescan_remove();
853 	pci_hp_remove_devices(bridge->subordinate);
854 	pci_unlock_rescan_remove();
855 
856 	/* Switch the CXL protocol on the card */
857 	if (switch_work->mode == CXL_BIMODE_CXL) {
858 		dev_info(&bus->dev, "cxl: Switching card to CXL mode\n");
859 		val &= ~CXL_VSEC_PROTOCOL_MASK;
860 		val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
861 		rc = pnv_cxl_enable_phb_kernel_api(hose, true);
862 		if (rc) {
863 			dev_err(&bus->dev, "cxl: Failed to enable kernel API"
864 				           " on real PHB, aborting\n");
865 			goto err_free_work;
866 		}
867 	} else {
868 		dev_WARN(&bus->dev, "cxl: Switching card to PCI mode not supported!\n");
869 		goto err_free_work;
870 	}
871 
872 	rc = CXL_WRITE_VSEC_MODE_CONTROL_BUS(bus, devfn, switch_work->vsec, val);
873 	if (rc) {
874 		dev_err(&bus->dev, "cxl: Failed to configure CXL protocol: %i\n", rc);
875 		goto err_free_work;
876 	}
877 
878 	/*
879 	 * The CAIA spec (v1.1, Section 10.6 Bi-modal Device Support) states
880 	 * we must wait 100ms after this mode switch before touching PCIe config
881 	 * space.
882 	 */
883 	msleep(100);
884 
885 	/*
886 	 * Hot reset to cause the card to come back in cxl mode. A
887 	 * OPAL_RESET_PCI_LINK would be sufficient, but currently lacks support
888 	 * in skiboot, so we use a hot reset instead.
889 	 *
890 	 * We call pci_set_pcie_reset_state() on the bridge, as a CAPI card is
891 	 * guaranteed to sit directly under the root port, and setting the reset
892 	 * state on a device directly under the root port is equivalent to doing
893 	 * it on the root port iself.
894 	 */
895 	dev_info(&bus->dev, "cxl: Configuration write complete, resetting card\n");
896 	pci_set_pcie_reset_state(bridge, pcie_hot_reset);
897 	pci_set_pcie_reset_state(bridge, pcie_deassert_reset);
898 
899 	dev_dbg(&bus->dev, "cxl: Offlining slot\n");
900 	rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_OFFLINE);
901 	if (rc) {
902 		dev_err(&bus->dev, "cxl: OPAL offlining call failed: %i\n", rc);
903 		goto err_free_work;
904 	}
905 
906 	dev_dbg(&bus->dev, "cxl: Onlining and probing slot\n");
907 	rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_ONLINE);
908 	if (rc) {
909 		dev_err(&bus->dev, "cxl: OPAL onlining call failed: %i\n", rc);
910 		goto err_free_work;
911 	}
912 
913 	pci_lock_rescan_remove();
914 	pci_hp_add_devices(bridge->subordinate);
915 	pci_unlock_rescan_remove();
916 
917 	dev_info(&bus->dev, "cxl: CAPI mode switch completed\n");
918 	kfree(switch_work);
919 	return;
920 
921 err_dev_put:
922 	/* Release the reference obtained in cxl_check_and_switch_mode() */
923 	pci_dev_put(dev);
924 err_free_work:
925 	kfree(switch_work);
926 }
927 
928 int cxl_check_and_switch_mode(struct pci_dev *dev, int mode, int vsec)
929 {
930 	struct cxl_switch_work *work;
931 	u8 val;
932 	int rc;
933 
934 	if (!cpu_has_feature(CPU_FTR_HVMODE))
935 		return -ENODEV;
936 
937 	if (!vsec) {
938 		vsec = find_cxl_vsec(dev);
939 		if (!vsec) {
940 			dev_info(&dev->dev, "CXL VSEC not found\n");
941 			return -ENODEV;
942 		}
943 	}
944 
945 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
946 	if (rc) {
947 		dev_err(&dev->dev, "Failed to read current mode control: %i", rc);
948 		return rc;
949 	}
950 
951 	if (mode == CXL_BIMODE_PCI) {
952 		if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
953 			dev_info(&dev->dev, "Card is already in PCI mode\n");
954 			return 0;
955 		}
956 		/*
957 		 * TODO: Before it's safe to switch the card back to PCI mode
958 		 * we need to disable the CAPP and make sure any cachelines the
959 		 * card holds have been flushed out. Needs skiboot support.
960 		 */
961 		dev_WARN(&dev->dev, "CXL mode switch to PCI unsupported!\n");
962 		return -EIO;
963 	}
964 
965 	if (val & CXL_VSEC_PROTOCOL_ENABLE) {
966 		dev_info(&dev->dev, "Card is already in CXL mode\n");
967 		return 0;
968 	}
969 
970 	dev_info(&dev->dev, "Card is in PCI mode, scheduling kernel thread "
971 			    "to switch to CXL mode\n");
972 
973 	work = kmalloc(sizeof(struct cxl_switch_work), GFP_KERNEL);
974 	if (!work)
975 		return -ENOMEM;
976 
977 	pci_dev_get(dev);
978 	work->dev = dev;
979 	work->vsec = vsec;
980 	work->mode = mode;
981 	INIT_WORK(&work->work, switch_card_to_cxl);
982 
983 	schedule_work(&work->work);
984 
985 	/*
986 	 * We return a failure now to abort the driver init. Once the
987 	 * link has been cycled and the card is in cxl mode we will
988 	 * come back (possibly using the generic cxl driver), but
989 	 * return success as the card should then be in cxl mode.
990 	 *
991 	 * TODO: What if the card comes back in PCI mode even after
992 	 *       the switch?  Don't want to spin endlessly.
993 	 */
994 	return -EBUSY;
995 }
996 EXPORT_SYMBOL_GPL(cxl_check_and_switch_mode);
997 
998 #endif /* CONFIG_CXL_BIMODAL */
999 
1000 static int setup_cxl_protocol_area(struct pci_dev *dev)
1001 {
1002 	u8 val;
1003 	int rc;
1004 	int vsec = find_cxl_vsec(dev);
1005 
1006 	if (!vsec) {
1007 		dev_info(&dev->dev, "CXL VSEC not found\n");
1008 		return -ENODEV;
1009 	}
1010 
1011 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
1012 	if (rc) {
1013 		dev_err(&dev->dev, "Failed to read current mode control: %i\n", rc);
1014 		return rc;
1015 	}
1016 
1017 	if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
1018 		dev_err(&dev->dev, "Card not in CAPI mode!\n");
1019 		return -EIO;
1020 	}
1021 
1022 	if ((val & CXL_VSEC_PROTOCOL_MASK) != CXL_VSEC_PROTOCOL_256TB) {
1023 		val &= ~CXL_VSEC_PROTOCOL_MASK;
1024 		val |= CXL_VSEC_PROTOCOL_256TB;
1025 		rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val);
1026 		if (rc) {
1027 			dev_err(&dev->dev, "Failed to set CXL protocol area: %i\n", rc);
1028 			return rc;
1029 		}
1030 	}
1031 
1032 	return 0;
1033 }
1034 
1035 static int pci_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1036 {
1037 	u64 p1n_base, p2n_base, afu_desc;
1038 	const u64 p1n_size = 0x100;
1039 	const u64 p2n_size = 0x1000;
1040 
1041 	p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
1042 	p2n_base = p2_base(dev) + (afu->slice * p2n_size);
1043 	afu->psn_phys = p2_base(dev) + (adapter->native->ps_off + (afu->slice * adapter->ps_size));
1044 	afu_desc = p2_base(dev) + adapter->native->afu_desc_off + (afu->slice * adapter->native->afu_desc_size);
1045 
1046 	if (!(afu->native->p1n_mmio = ioremap(p1n_base, p1n_size)))
1047 		goto err;
1048 	if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
1049 		goto err1;
1050 	if (afu_desc) {
1051 		if (!(afu->native->afu_desc_mmio = ioremap(afu_desc, adapter->native->afu_desc_size)))
1052 			goto err2;
1053 	}
1054 
1055 	return 0;
1056 err2:
1057 	iounmap(afu->p2n_mmio);
1058 err1:
1059 	iounmap(afu->native->p1n_mmio);
1060 err:
1061 	dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
1062 	return -ENOMEM;
1063 }
1064 
1065 static void pci_unmap_slice_regs(struct cxl_afu *afu)
1066 {
1067 	if (afu->p2n_mmio) {
1068 		iounmap(afu->p2n_mmio);
1069 		afu->p2n_mmio = NULL;
1070 	}
1071 	if (afu->native->p1n_mmio) {
1072 		iounmap(afu->native->p1n_mmio);
1073 		afu->native->p1n_mmio = NULL;
1074 	}
1075 	if (afu->native->afu_desc_mmio) {
1076 		iounmap(afu->native->afu_desc_mmio);
1077 		afu->native->afu_desc_mmio = NULL;
1078 	}
1079 }
1080 
1081 void cxl_pci_release_afu(struct device *dev)
1082 {
1083 	struct cxl_afu *afu = to_cxl_afu(dev);
1084 
1085 	pr_devel("%s\n", __func__);
1086 
1087 	idr_destroy(&afu->contexts_idr);
1088 	cxl_release_spa(afu);
1089 
1090 	kfree(afu->native);
1091 	kfree(afu);
1092 }
1093 
1094 /* Expects AFU struct to have recently been zeroed out */
1095 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
1096 {
1097 	u64 val;
1098 
1099 	val = AFUD_READ_INFO(afu);
1100 	afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
1101 	afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
1102 	afu->crs_num = AFUD_NUM_CRS(val);
1103 
1104 	if (AFUD_AFU_DIRECTED(val))
1105 		afu->modes_supported |= CXL_MODE_DIRECTED;
1106 	if (AFUD_DEDICATED_PROCESS(val))
1107 		afu->modes_supported |= CXL_MODE_DEDICATED;
1108 	if (AFUD_TIME_SLICED(val))
1109 		afu->modes_supported |= CXL_MODE_TIME_SLICED;
1110 
1111 	val = AFUD_READ_PPPSA(afu);
1112 	afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
1113 	afu->psa = AFUD_PPPSA_PSA(val);
1114 	if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
1115 		afu->native->pp_offset = AFUD_READ_PPPSA_OFF(afu);
1116 
1117 	val = AFUD_READ_CR(afu);
1118 	afu->crs_len = AFUD_CR_LEN(val) * 256;
1119 	afu->crs_offset = AFUD_READ_CR_OFF(afu);
1120 
1121 
1122 	/* eb_len is in multiple of 4K */
1123 	afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
1124 	afu->eb_offset = AFUD_READ_EB_OFF(afu);
1125 
1126 	/* eb_off is 4K aligned so lower 12 bits are always zero */
1127 	if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
1128 		dev_warn(&afu->dev,
1129 			 "Invalid AFU error buffer offset %Lx\n",
1130 			 afu->eb_offset);
1131 		dev_info(&afu->dev,
1132 			 "Ignoring AFU error buffer in the descriptor\n");
1133 		/* indicate that no afu buffer exists */
1134 		afu->eb_len = 0;
1135 	}
1136 
1137 	return 0;
1138 }
1139 
1140 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
1141 {
1142 	int i, rc;
1143 	u32 val;
1144 
1145 	if (afu->psa && afu->adapter->ps_size <
1146 			(afu->native->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
1147 		dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
1148 		return -ENODEV;
1149 	}
1150 
1151 	if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
1152 		dev_warn(&afu->dev, "AFU uses pp_size(%#016llx) < PAGE_SIZE per-process PSA!\n", afu->pp_size);
1153 
1154 	for (i = 0; i < afu->crs_num; i++) {
1155 		rc = cxl_ops->afu_cr_read32(afu, i, 0, &val);
1156 		if (rc || val == 0) {
1157 			dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
1158 			return -EINVAL;
1159 		}
1160 	}
1161 
1162 	if ((afu->modes_supported & ~CXL_MODE_DEDICATED) && afu->max_procs_virtualised == 0) {
1163 		/*
1164 		 * We could also check this for the dedicated process model
1165 		 * since the architecture indicates it should be set to 1, but
1166 		 * in that case we ignore the value and I'd rather not risk
1167 		 * breaking any existing dedicated process AFUs that left it as
1168 		 * 0 (not that I'm aware of any). It is clearly an error for an
1169 		 * AFU directed AFU to set this to 0, and would have previously
1170 		 * triggered a bug resulting in the maximum not being enforced
1171 		 * at all since idr_alloc treats 0 as no maximum.
1172 		 */
1173 		dev_err(&afu->dev, "AFU does not support any processes\n");
1174 		return -EINVAL;
1175 	}
1176 
1177 	return 0;
1178 }
1179 
1180 static int sanitise_afu_regs_psl9(struct cxl_afu *afu)
1181 {
1182 	u64 reg;
1183 
1184 	/*
1185 	 * Clear out any regs that contain either an IVTE or address or may be
1186 	 * waiting on an acknowledgment to try to be a bit safer as we bring
1187 	 * it online
1188 	 */
1189 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
1190 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
1191 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
1192 		if (cxl_ops->afu_reset(afu))
1193 			return -EIO;
1194 		if (cxl_afu_disable(afu))
1195 			return -EIO;
1196 		if (cxl_psl_purge(afu))
1197 			return -EIO;
1198 	}
1199 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
1200 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1201 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1202 	if (reg) {
1203 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1204 		if (reg & CXL_PSL9_DSISR_An_TF)
1205 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1206 		else
1207 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1208 	}
1209 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1210 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1211 		if (reg) {
1212 			if (reg & ~0x000000007fffffff)
1213 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1214 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1215 		}
1216 	}
1217 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1218 	if (reg) {
1219 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1220 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1221 	}
1222 
1223 	return 0;
1224 }
1225 
1226 static int sanitise_afu_regs_psl8(struct cxl_afu *afu)
1227 {
1228 	u64 reg;
1229 
1230 	/*
1231 	 * Clear out any regs that contain either an IVTE or address or may be
1232 	 * waiting on an acknowledgement to try to be a bit safer as we bring
1233 	 * it online
1234 	 */
1235 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
1236 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
1237 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
1238 		if (cxl_ops->afu_reset(afu))
1239 			return -EIO;
1240 		if (cxl_afu_disable(afu))
1241 			return -EIO;
1242 		if (cxl_psl_purge(afu))
1243 			return -EIO;
1244 	}
1245 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
1246 	cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
1247 	cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
1248 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1249 	cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
1250 	cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
1251 	cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
1252 	cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
1253 	cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
1254 	cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
1255 	cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
1256 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1257 	if (reg) {
1258 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1259 		if (reg & CXL_PSL_DSISR_TRANS)
1260 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1261 		else
1262 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1263 	}
1264 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1265 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1266 		if (reg) {
1267 			if (reg & ~0xffff)
1268 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1269 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1270 		}
1271 	}
1272 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1273 	if (reg) {
1274 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1275 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1276 	}
1277 
1278 	return 0;
1279 }
1280 
1281 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
1282 /*
1283  * afu_eb_read:
1284  * Called from sysfs and reads the afu error info buffer. The h/w only supports
1285  * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
1286  * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
1287  */
1288 ssize_t cxl_pci_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
1289 				loff_t off, size_t count)
1290 {
1291 	loff_t aligned_start, aligned_end;
1292 	size_t aligned_length;
1293 	void *tbuf;
1294 	const void __iomem *ebuf = afu->native->afu_desc_mmio + afu->eb_offset;
1295 
1296 	if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
1297 		return 0;
1298 
1299 	/* calculate aligned read window */
1300 	count = min((size_t)(afu->eb_len - off), count);
1301 	aligned_start = round_down(off, 8);
1302 	aligned_end = round_up(off + count, 8);
1303 	aligned_length = aligned_end - aligned_start;
1304 
1305 	/* max we can copy in one read is PAGE_SIZE */
1306 	if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
1307 		aligned_length = ERR_BUFF_MAX_COPY_SIZE;
1308 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
1309 	}
1310 
1311 	/* use bounce buffer for copy */
1312 	tbuf = (void *)__get_free_page(GFP_KERNEL);
1313 	if (!tbuf)
1314 		return -ENOMEM;
1315 
1316 	/* perform aligned read from the mmio region */
1317 	memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
1318 	memcpy(buf, tbuf + (off & 0x7), count);
1319 
1320 	free_page((unsigned long)tbuf);
1321 
1322 	return count;
1323 }
1324 
1325 static int pci_configure_afu(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1326 {
1327 	int rc;
1328 
1329 	if ((rc = pci_map_slice_regs(afu, adapter, dev)))
1330 		return rc;
1331 
1332 	if (adapter->native->sl_ops->sanitise_afu_regs) {
1333 		rc = adapter->native->sl_ops->sanitise_afu_regs(afu);
1334 		if (rc)
1335 			goto err1;
1336 	}
1337 
1338 	/* We need to reset the AFU before we can read the AFU descriptor */
1339 	if ((rc = cxl_ops->afu_reset(afu)))
1340 		goto err1;
1341 
1342 	if (cxl_verbose)
1343 		dump_afu_descriptor(afu);
1344 
1345 	if ((rc = cxl_read_afu_descriptor(afu)))
1346 		goto err1;
1347 
1348 	if ((rc = cxl_afu_descriptor_looks_ok(afu)))
1349 		goto err1;
1350 
1351 	if (adapter->native->sl_ops->afu_regs_init)
1352 		if ((rc = adapter->native->sl_ops->afu_regs_init(afu)))
1353 			goto err1;
1354 
1355 	if (adapter->native->sl_ops->register_serr_irq)
1356 		if ((rc = adapter->native->sl_ops->register_serr_irq(afu)))
1357 			goto err1;
1358 
1359 	if ((rc = cxl_native_register_psl_irq(afu)))
1360 		goto err2;
1361 
1362 	atomic_set(&afu->configured_state, 0);
1363 	return 0;
1364 
1365 err2:
1366 	if (adapter->native->sl_ops->release_serr_irq)
1367 		adapter->native->sl_ops->release_serr_irq(afu);
1368 err1:
1369 	pci_unmap_slice_regs(afu);
1370 	return rc;
1371 }
1372 
1373 static void pci_deconfigure_afu(struct cxl_afu *afu)
1374 {
1375 	/*
1376 	 * It's okay to deconfigure when AFU is already locked, otherwise wait
1377 	 * until there are no readers
1378 	 */
1379 	if (atomic_read(&afu->configured_state) != -1) {
1380 		while (atomic_cmpxchg(&afu->configured_state, 0, -1) != -1)
1381 			schedule();
1382 	}
1383 	cxl_native_release_psl_irq(afu);
1384 	if (afu->adapter->native->sl_ops->release_serr_irq)
1385 		afu->adapter->native->sl_ops->release_serr_irq(afu);
1386 	pci_unmap_slice_regs(afu);
1387 }
1388 
1389 static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
1390 {
1391 	struct cxl_afu *afu;
1392 	int rc = -ENOMEM;
1393 
1394 	afu = cxl_alloc_afu(adapter, slice);
1395 	if (!afu)
1396 		return -ENOMEM;
1397 
1398 	afu->native = kzalloc(sizeof(struct cxl_afu_native), GFP_KERNEL);
1399 	if (!afu->native)
1400 		goto err_free_afu;
1401 
1402 	mutex_init(&afu->native->spa_mutex);
1403 
1404 	rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice);
1405 	if (rc)
1406 		goto err_free_native;
1407 
1408 	rc = pci_configure_afu(afu, adapter, dev);
1409 	if (rc)
1410 		goto err_free_native;
1411 
1412 	/* Don't care if this fails */
1413 	cxl_debugfs_afu_add(afu);
1414 
1415 	/*
1416 	 * After we call this function we must not free the afu directly, even
1417 	 * if it returns an error!
1418 	 */
1419 	if ((rc = cxl_register_afu(afu)))
1420 		goto err_put1;
1421 
1422 	if ((rc = cxl_sysfs_afu_add(afu)))
1423 		goto err_put1;
1424 
1425 	adapter->afu[afu->slice] = afu;
1426 
1427 	if ((rc = cxl_pci_vphb_add(afu)))
1428 		dev_info(&afu->dev, "Can't register vPHB\n");
1429 
1430 	return 0;
1431 
1432 err_put1:
1433 	pci_deconfigure_afu(afu);
1434 	cxl_debugfs_afu_remove(afu);
1435 	device_unregister(&afu->dev);
1436 	return rc;
1437 
1438 err_free_native:
1439 	kfree(afu->native);
1440 err_free_afu:
1441 	kfree(afu);
1442 	return rc;
1443 
1444 }
1445 
1446 static void cxl_pci_remove_afu(struct cxl_afu *afu)
1447 {
1448 	pr_devel("%s\n", __func__);
1449 
1450 	if (!afu)
1451 		return;
1452 
1453 	cxl_pci_vphb_remove(afu);
1454 	cxl_sysfs_afu_remove(afu);
1455 	cxl_debugfs_afu_remove(afu);
1456 
1457 	spin_lock(&afu->adapter->afu_list_lock);
1458 	afu->adapter->afu[afu->slice] = NULL;
1459 	spin_unlock(&afu->adapter->afu_list_lock);
1460 
1461 	cxl_context_detach_all(afu);
1462 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1463 
1464 	pci_deconfigure_afu(afu);
1465 	device_unregister(&afu->dev);
1466 }
1467 
1468 int cxl_pci_reset(struct cxl *adapter)
1469 {
1470 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1471 	int rc;
1472 
1473 	if (adapter->perst_same_image) {
1474 		dev_warn(&dev->dev,
1475 			 "cxl: refusing to reset/reflash when perst_reloads_same_image is set.\n");
1476 		return -EINVAL;
1477 	}
1478 
1479 	dev_info(&dev->dev, "CXL reset\n");
1480 
1481 	/*
1482 	 * The adapter is about to be reset, so ignore errors.
1483 	 */
1484 	cxl_data_cache_flush(adapter);
1485 
1486 	/* pcie_warm_reset requests a fundamental pci reset which includes a
1487 	 * PERST assert/deassert.  PERST triggers a loading of the image
1488 	 * if "user" or "factory" is selected in sysfs */
1489 	if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
1490 		dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
1491 		return rc;
1492 	}
1493 
1494 	return rc;
1495 }
1496 
1497 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
1498 {
1499 	if (pci_request_region(dev, 2, "priv 2 regs"))
1500 		goto err1;
1501 	if (pci_request_region(dev, 0, "priv 1 regs"))
1502 		goto err2;
1503 
1504 	pr_devel("cxl_map_adapter_regs: p1: %#016llx %#llx, p2: %#016llx %#llx",
1505 			p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
1506 
1507 	if (!(adapter->native->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
1508 		goto err3;
1509 
1510 	if (!(adapter->native->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
1511 		goto err4;
1512 
1513 	return 0;
1514 
1515 err4:
1516 	iounmap(adapter->native->p1_mmio);
1517 	adapter->native->p1_mmio = NULL;
1518 err3:
1519 	pci_release_region(dev, 0);
1520 err2:
1521 	pci_release_region(dev, 2);
1522 err1:
1523 	return -ENOMEM;
1524 }
1525 
1526 static void cxl_unmap_adapter_regs(struct cxl *adapter)
1527 {
1528 	if (adapter->native->p1_mmio) {
1529 		iounmap(adapter->native->p1_mmio);
1530 		adapter->native->p1_mmio = NULL;
1531 		pci_release_region(to_pci_dev(adapter->dev.parent), 2);
1532 	}
1533 	if (adapter->native->p2_mmio) {
1534 		iounmap(adapter->native->p2_mmio);
1535 		adapter->native->p2_mmio = NULL;
1536 		pci_release_region(to_pci_dev(adapter->dev.parent), 0);
1537 	}
1538 }
1539 
1540 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
1541 {
1542 	int vsec;
1543 	u32 afu_desc_off, afu_desc_size;
1544 	u32 ps_off, ps_size;
1545 	u16 vseclen;
1546 	u8 image_state;
1547 
1548 	if (!(vsec = find_cxl_vsec(dev))) {
1549 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
1550 		return -ENODEV;
1551 	}
1552 
1553 	CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
1554 	if (vseclen < CXL_VSEC_MIN_SIZE) {
1555 		dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
1556 		return -EINVAL;
1557 	}
1558 
1559 	CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
1560 	CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
1561 	CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
1562 	CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
1563 	CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
1564 	CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
1565 	adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1566 	adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1567 	adapter->perst_loads_image = !!(image_state & CXL_VSEC_PERST_LOADS_IMAGE);
1568 
1569 	CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
1570 	CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
1571 	CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
1572 	CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
1573 	CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
1574 
1575 	/* Convert everything to bytes, because there is NO WAY I'd look at the
1576 	 * code a month later and forget what units these are in ;-) */
1577 	adapter->native->ps_off = ps_off * 64 * 1024;
1578 	adapter->ps_size = ps_size * 64 * 1024;
1579 	adapter->native->afu_desc_off = afu_desc_off * 64 * 1024;
1580 	adapter->native->afu_desc_size = afu_desc_size * 64 * 1024;
1581 
1582 	/* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
1583 	adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
1584 
1585 	return 0;
1586 }
1587 
1588 /*
1589  * Workaround a PCIe Host Bridge defect on some cards, that can cause
1590  * malformed Transaction Layer Packet (TLP) errors to be erroneously
1591  * reported. Mask this error in the Uncorrectable Error Mask Register.
1592  *
1593  * The upper nibble of the PSL revision is used to distinguish between
1594  * different cards. The affected ones have it set to 0.
1595  */
1596 static void cxl_fixup_malformed_tlp(struct cxl *adapter, struct pci_dev *dev)
1597 {
1598 	int aer;
1599 	u32 data;
1600 
1601 	if (adapter->psl_rev & 0xf000)
1602 		return;
1603 	if (!(aer = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR)))
1604 		return;
1605 	pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &data);
1606 	if (data & PCI_ERR_UNC_MALF_TLP)
1607 		if (data & PCI_ERR_UNC_INTN)
1608 			return;
1609 	data |= PCI_ERR_UNC_MALF_TLP;
1610 	data |= PCI_ERR_UNC_INTN;
1611 	pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, data);
1612 }
1613 
1614 static bool cxl_compatible_caia_version(struct cxl *adapter)
1615 {
1616 	if (cxl_is_power8() && (adapter->caia_major == 1))
1617 		return true;
1618 
1619 	if (cxl_is_power9() && (adapter->caia_major == 2))
1620 		return true;
1621 
1622 	return false;
1623 }
1624 
1625 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
1626 {
1627 	if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
1628 		return -EBUSY;
1629 
1630 	if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
1631 		dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
1632 		return -EINVAL;
1633 	}
1634 
1635 	if (!cxl_compatible_caia_version(adapter)) {
1636 		dev_info(&dev->dev, "Ignoring card. PSL type is not supported (caia version: %d)\n",
1637 			 adapter->caia_major);
1638 		return -ENODEV;
1639 	}
1640 
1641 	if (!adapter->slices) {
1642 		/* Once we support dynamic reprogramming we can use the card if
1643 		 * it supports loadable AFUs */
1644 		dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
1645 		return -EINVAL;
1646 	}
1647 
1648 	if (!adapter->native->afu_desc_off || !adapter->native->afu_desc_size) {
1649 		dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
1650 		return -EINVAL;
1651 	}
1652 
1653 	if (adapter->ps_size > p2_size(dev) - adapter->native->ps_off) {
1654 		dev_err(&dev->dev, "ABORTING: Problem state size larger than "
1655 				   "available in BAR2: 0x%llx > 0x%llx\n",
1656 			 adapter->ps_size, p2_size(dev) - adapter->native->ps_off);
1657 		return -EINVAL;
1658 	}
1659 
1660 	return 0;
1661 }
1662 
1663 ssize_t cxl_pci_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1664 {
1665 	return pci_read_vpd(to_pci_dev(adapter->dev.parent), 0, len, buf);
1666 }
1667 
1668 static void cxl_release_adapter(struct device *dev)
1669 {
1670 	struct cxl *adapter = to_cxl_adapter(dev);
1671 
1672 	pr_devel("cxl_release_adapter\n");
1673 
1674 	cxl_remove_adapter_nr(adapter);
1675 
1676 	kfree(adapter->native);
1677 	kfree(adapter);
1678 }
1679 
1680 #define CXL_PSL_ErrIVTE_tberror (0x1ull << (63-31))
1681 
1682 static int sanitise_adapter_regs(struct cxl *adapter)
1683 {
1684 	int rc = 0;
1685 
1686 	/* Clear PSL tberror bit by writing 1 to it */
1687 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, CXL_PSL_ErrIVTE_tberror);
1688 
1689 	if (adapter->native->sl_ops->invalidate_all) {
1690 		/* do not invalidate ERAT entries when not reloading on PERST */
1691 		if (cxl_is_power9() && (adapter->perst_loads_image))
1692 			return 0;
1693 		rc = adapter->native->sl_ops->invalidate_all(adapter);
1694 	}
1695 
1696 	return rc;
1697 }
1698 
1699 /* This should contain *only* operations that can safely be done in
1700  * both creation and recovery.
1701  */
1702 static int cxl_configure_adapter(struct cxl *adapter, struct pci_dev *dev)
1703 {
1704 	int rc;
1705 
1706 	adapter->dev.parent = &dev->dev;
1707 	adapter->dev.release = cxl_release_adapter;
1708 	pci_set_drvdata(dev, adapter);
1709 
1710 	rc = pci_enable_device(dev);
1711 	if (rc) {
1712 		dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1713 		return rc;
1714 	}
1715 
1716 	if ((rc = cxl_read_vsec(adapter, dev)))
1717 		return rc;
1718 
1719 	if ((rc = cxl_vsec_looks_ok(adapter, dev)))
1720 	        return rc;
1721 
1722 	cxl_fixup_malformed_tlp(adapter, dev);
1723 
1724 	if ((rc = setup_cxl_bars(dev)))
1725 		return rc;
1726 
1727 	if ((rc = setup_cxl_protocol_area(dev)))
1728 		return rc;
1729 
1730 	if ((rc = cxl_update_image_control(adapter)))
1731 		return rc;
1732 
1733 	if ((rc = cxl_map_adapter_regs(adapter, dev)))
1734 		return rc;
1735 
1736 	if ((rc = sanitise_adapter_regs(adapter)))
1737 		goto err;
1738 
1739 	if ((rc = adapter->native->sl_ops->adapter_regs_init(adapter, dev)))
1740 		goto err;
1741 
1742 	/* Required for devices using CAPP DMA mode, harmless for others */
1743 	pci_set_master(dev);
1744 
1745 	adapter->tunneled_ops_supported = false;
1746 
1747 	if (cxl_is_power9()) {
1748 		if (pnv_pci_set_tunnel_bar(dev, 0x00020000E0000000ull, 1))
1749 			dev_info(&dev->dev, "Tunneled operations unsupported\n");
1750 		else
1751 			adapter->tunneled_ops_supported = true;
1752 	}
1753 
1754 	if ((rc = pnv_phb_to_cxl_mode(dev, adapter->native->sl_ops->capi_mode)))
1755 		goto err;
1756 
1757 	/* If recovery happened, the last step is to turn on snooping.
1758 	 * In the non-recovery case this has no effect */
1759 	if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON)))
1760 		goto err;
1761 
1762 	/* Ignore error, adapter init is not dependant on timebase sync */
1763 	cxl_setup_psl_timebase(adapter, dev);
1764 
1765 	if ((rc = cxl_native_register_psl_err_irq(adapter)))
1766 		goto err;
1767 
1768 	return 0;
1769 
1770 err:
1771 	cxl_unmap_adapter_regs(adapter);
1772 	return rc;
1773 
1774 }
1775 
1776 static void cxl_deconfigure_adapter(struct cxl *adapter)
1777 {
1778 	struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1779 
1780 	if (cxl_is_power9())
1781 		pnv_pci_set_tunnel_bar(pdev, 0x00020000E0000000ull, 0);
1782 
1783 	cxl_native_release_psl_err_irq(adapter);
1784 	cxl_unmap_adapter_regs(adapter);
1785 
1786 	pci_disable_device(pdev);
1787 }
1788 
1789 static void cxl_stop_trace_psl9(struct cxl *adapter)
1790 {
1791 	int traceid;
1792 	u64 trace_state, trace_mask;
1793 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1794 
1795 	/* read each tracearray state and issue mmio to stop them is needed */
1796 	for (traceid = 0; traceid <= CXL_PSL9_TRACEID_MAX; ++traceid) {
1797 		trace_state = cxl_p1_read(adapter, CXL_PSL9_CTCCFG);
1798 		trace_mask = (0x3ULL << (62 - traceid * 2));
1799 		trace_state = (trace_state & trace_mask) >> (62 - traceid * 2);
1800 		dev_dbg(&dev->dev, "cxl: Traceid-%d trace_state=0x%0llX\n",
1801 			traceid, trace_state);
1802 
1803 		/* issue mmio if the trace array isn't in FIN state */
1804 		if (trace_state != CXL_PSL9_TRACESTATE_FIN)
1805 			cxl_p1_write(adapter, CXL_PSL9_TRACECFG,
1806 				     0x8400000000000000ULL | traceid);
1807 	}
1808 }
1809 
1810 static void cxl_stop_trace_psl8(struct cxl *adapter)
1811 {
1812 	int slice;
1813 
1814 	/* Stop the trace */
1815 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x8000000000000017LL);
1816 
1817 	/* Stop the slice traces */
1818 	spin_lock(&adapter->afu_list_lock);
1819 	for (slice = 0; slice < adapter->slices; slice++) {
1820 		if (adapter->afu[slice])
1821 			cxl_p1n_write(adapter->afu[slice], CXL_PSL_SLICE_TRACE,
1822 				      0x8000000000000000LL);
1823 	}
1824 	spin_unlock(&adapter->afu_list_lock);
1825 }
1826 
1827 static const struct cxl_service_layer_ops psl9_ops = {
1828 	.adapter_regs_init = init_implementation_adapter_regs_psl9,
1829 	.invalidate_all = cxl_invalidate_all_psl9,
1830 	.afu_regs_init = init_implementation_afu_regs_psl9,
1831 	.sanitise_afu_regs = sanitise_afu_regs_psl9,
1832 	.register_serr_irq = cxl_native_register_serr_irq,
1833 	.release_serr_irq = cxl_native_release_serr_irq,
1834 	.handle_interrupt = cxl_irq_psl9,
1835 	.fail_irq = cxl_fail_irq_psl,
1836 	.activate_dedicated_process = cxl_activate_dedicated_process_psl9,
1837 	.attach_afu_directed = cxl_attach_afu_directed_psl9,
1838 	.attach_dedicated_process = cxl_attach_dedicated_process_psl9,
1839 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl9,
1840 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl9,
1841 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl9,
1842 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl9,
1843 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl9,
1844 	.debugfs_stop_trace = cxl_stop_trace_psl9,
1845 	.timebase_read = timebase_read_psl9,
1846 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1847 	.needs_reset_before_disable = true,
1848 };
1849 
1850 static const struct cxl_service_layer_ops psl8_ops = {
1851 	.adapter_regs_init = init_implementation_adapter_regs_psl8,
1852 	.invalidate_all = cxl_invalidate_all_psl8,
1853 	.afu_regs_init = init_implementation_afu_regs_psl8,
1854 	.sanitise_afu_regs = sanitise_afu_regs_psl8,
1855 	.register_serr_irq = cxl_native_register_serr_irq,
1856 	.release_serr_irq = cxl_native_release_serr_irq,
1857 	.handle_interrupt = cxl_irq_psl8,
1858 	.fail_irq = cxl_fail_irq_psl,
1859 	.activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1860 	.attach_afu_directed = cxl_attach_afu_directed_psl8,
1861 	.attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1862 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1863 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl8,
1864 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl8,
1865 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl8,
1866 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl8,
1867 	.debugfs_stop_trace = cxl_stop_trace_psl8,
1868 	.write_timebase_ctrl = write_timebase_ctrl_psl8,
1869 	.timebase_read = timebase_read_psl8,
1870 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1871 	.needs_reset_before_disable = true,
1872 };
1873 
1874 static const struct cxl_service_layer_ops xsl_ops = {
1875 	.adapter_regs_init = init_implementation_adapter_regs_xsl,
1876 	.invalidate_all = cxl_invalidate_all_psl8,
1877 	.sanitise_afu_regs = sanitise_afu_regs_psl8,
1878 	.handle_interrupt = cxl_irq_psl8,
1879 	.fail_irq = cxl_fail_irq_psl,
1880 	.activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1881 	.attach_afu_directed = cxl_attach_afu_directed_psl8,
1882 	.attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1883 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1884 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_xsl,
1885 	.write_timebase_ctrl = write_timebase_ctrl_xsl,
1886 	.timebase_read = timebase_read_xsl,
1887 	.capi_mode = OPAL_PHB_CAPI_MODE_DMA,
1888 };
1889 
1890 static void set_sl_ops(struct cxl *adapter, struct pci_dev *dev)
1891 {
1892 	if (dev->vendor == PCI_VENDOR_ID_MELLANOX && dev->device == 0x1013) {
1893 		/* Mellanox CX-4 */
1894 		dev_info(&dev->dev, "Device uses an XSL\n");
1895 		adapter->native->sl_ops = &xsl_ops;
1896 		adapter->min_pe = 1; /* Workaround for CX-4 hardware bug */
1897 	} else {
1898 		if (cxl_is_power8()) {
1899 			dev_info(&dev->dev, "Device uses a PSL8\n");
1900 			adapter->native->sl_ops = &psl8_ops;
1901 		} else {
1902 			dev_info(&dev->dev, "Device uses a PSL9\n");
1903 			adapter->native->sl_ops = &psl9_ops;
1904 		}
1905 	}
1906 }
1907 
1908 
1909 static struct cxl *cxl_pci_init_adapter(struct pci_dev *dev)
1910 {
1911 	struct cxl *adapter;
1912 	int rc;
1913 
1914 	adapter = cxl_alloc_adapter();
1915 	if (!adapter)
1916 		return ERR_PTR(-ENOMEM);
1917 
1918 	adapter->native = kzalloc(sizeof(struct cxl_native), GFP_KERNEL);
1919 	if (!adapter->native) {
1920 		rc = -ENOMEM;
1921 		goto err_release;
1922 	}
1923 
1924 	set_sl_ops(adapter, dev);
1925 
1926 	/* Set defaults for parameters which need to persist over
1927 	 * configure/reconfigure
1928 	 */
1929 	adapter->perst_loads_image = true;
1930 	adapter->perst_same_image = false;
1931 
1932 	rc = cxl_configure_adapter(adapter, dev);
1933 	if (rc) {
1934 		pci_disable_device(dev);
1935 		goto err_release;
1936 	}
1937 
1938 	/* Don't care if this one fails: */
1939 	cxl_debugfs_adapter_add(adapter);
1940 
1941 	/*
1942 	 * After we call this function we must not free the adapter directly,
1943 	 * even if it returns an error!
1944 	 */
1945 	if ((rc = cxl_register_adapter(adapter)))
1946 		goto err_put1;
1947 
1948 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1949 		goto err_put1;
1950 
1951 	/* Release the context lock as adapter is configured */
1952 	cxl_adapter_context_unlock(adapter);
1953 
1954 	return adapter;
1955 
1956 err_put1:
1957 	/* This should mirror cxl_remove_adapter, except without the
1958 	 * sysfs parts
1959 	 */
1960 	cxl_debugfs_adapter_remove(adapter);
1961 	cxl_deconfigure_adapter(adapter);
1962 	device_unregister(&adapter->dev);
1963 	return ERR_PTR(rc);
1964 
1965 err_release:
1966 	cxl_release_adapter(&adapter->dev);
1967 	return ERR_PTR(rc);
1968 }
1969 
1970 static void cxl_pci_remove_adapter(struct cxl *adapter)
1971 {
1972 	pr_devel("cxl_remove_adapter\n");
1973 
1974 	cxl_sysfs_adapter_remove(adapter);
1975 	cxl_debugfs_adapter_remove(adapter);
1976 
1977 	/*
1978 	 * Flush adapter datacache as its about to be removed.
1979 	 */
1980 	cxl_data_cache_flush(adapter);
1981 
1982 	cxl_deconfigure_adapter(adapter);
1983 
1984 	device_unregister(&adapter->dev);
1985 }
1986 
1987 #define CXL_MAX_PCIEX_PARENT 2
1988 
1989 int cxl_slot_is_switched(struct pci_dev *dev)
1990 {
1991 	struct device_node *np;
1992 	int depth = 0;
1993 	const __be32 *prop;
1994 
1995 	if (!(np = pci_device_to_OF_node(dev))) {
1996 		pr_err("cxl: np = NULL\n");
1997 		return -ENODEV;
1998 	}
1999 	of_node_get(np);
2000 	while (np) {
2001 		np = of_get_next_parent(np);
2002 		prop = of_get_property(np, "device_type", NULL);
2003 		if (!prop || strcmp((char *)prop, "pciex"))
2004 			break;
2005 		depth++;
2006 	}
2007 	of_node_put(np);
2008 	return (depth > CXL_MAX_PCIEX_PARENT);
2009 }
2010 
2011 bool cxl_slot_is_supported(struct pci_dev *dev, int flags)
2012 {
2013 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2014 		return false;
2015 
2016 	if ((flags & CXL_SLOT_FLAG_DMA) && (!pvr_version_is(PVR_POWER8NVL))) {
2017 		/*
2018 		 * CAPP DMA mode is technically supported on regular P8, but
2019 		 * will EEH if the card attempts to access memory < 4GB, which
2020 		 * we cannot realistically avoid. We might be able to work
2021 		 * around the issue, but until then return unsupported:
2022 		 */
2023 		return false;
2024 	}
2025 
2026 	if (cxl_slot_is_switched(dev))
2027 		return false;
2028 
2029 	/*
2030 	 * XXX: This gets a little tricky on regular P8 (not POWER8NVL) since
2031 	 * the CAPP can be connected to PHB 0, 1 or 2 on a first come first
2032 	 * served basis, which is racy to check from here. If we need to
2033 	 * support this in future we might need to consider having this
2034 	 * function effectively reserve it ahead of time.
2035 	 *
2036 	 * Currently, the only user of this API is the Mellanox CX4, which is
2037 	 * only supported on P8NVL due to the above mentioned limitation of
2038 	 * CAPP DMA mode and therefore does not need to worry about this. If the
2039 	 * issue with CAPP DMA mode is later worked around on P8 we might need
2040 	 * to revisit this.
2041 	 */
2042 
2043 	return true;
2044 }
2045 EXPORT_SYMBOL_GPL(cxl_slot_is_supported);
2046 
2047 
2048 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
2049 {
2050 	struct cxl *adapter;
2051 	int slice;
2052 	int rc;
2053 
2054 	if (cxl_pci_is_vphb_device(dev)) {
2055 		dev_dbg(&dev->dev, "cxl_init_adapter: Ignoring cxl vphb device\n");
2056 		return -ENODEV;
2057 	}
2058 
2059 	if (cxl_slot_is_switched(dev)) {
2060 		dev_info(&dev->dev, "Ignoring card on incompatible PCI slot\n");
2061 		return -ENODEV;
2062 	}
2063 
2064 	if (cxl_is_power9() && !radix_enabled()) {
2065 		dev_info(&dev->dev, "Only Radix mode supported\n");
2066 		return -ENODEV;
2067 	}
2068 
2069 	if (cxl_verbose)
2070 		dump_cxl_config_space(dev);
2071 
2072 	adapter = cxl_pci_init_adapter(dev);
2073 	if (IS_ERR(adapter)) {
2074 		dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
2075 		return PTR_ERR(adapter);
2076 	}
2077 
2078 	for (slice = 0; slice < adapter->slices; slice++) {
2079 		if ((rc = pci_init_afu(adapter, slice, dev))) {
2080 			dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
2081 			continue;
2082 		}
2083 
2084 		rc = cxl_afu_select_best_mode(adapter->afu[slice]);
2085 		if (rc)
2086 			dev_err(&dev->dev, "AFU %i failed to start: %i\n", slice, rc);
2087 	}
2088 
2089 	if (pnv_pci_on_cxl_phb(dev) && adapter->slices >= 1)
2090 		pnv_cxl_phb_set_peer_afu(dev, adapter->afu[0]);
2091 
2092 	return 0;
2093 }
2094 
2095 static void cxl_remove(struct pci_dev *dev)
2096 {
2097 	struct cxl *adapter = pci_get_drvdata(dev);
2098 	struct cxl_afu *afu;
2099 	int i;
2100 
2101 	/*
2102 	 * Lock to prevent someone grabbing a ref through the adapter list as
2103 	 * we are removing it
2104 	 */
2105 	for (i = 0; i < adapter->slices; i++) {
2106 		afu = adapter->afu[i];
2107 		cxl_pci_remove_afu(afu);
2108 	}
2109 	cxl_pci_remove_adapter(adapter);
2110 }
2111 
2112 static pci_ers_result_t cxl_vphb_error_detected(struct cxl_afu *afu,
2113 						pci_channel_state_t state)
2114 {
2115 	struct pci_dev *afu_dev;
2116 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
2117 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
2118 
2119 	/* There should only be one entry, but go through the list
2120 	 * anyway
2121 	 */
2122 	if (afu->phb == NULL)
2123 		return result;
2124 
2125 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2126 		if (!afu_dev->driver)
2127 			continue;
2128 
2129 		afu_dev->error_state = state;
2130 
2131 		if (afu_dev->driver->err_handler)
2132 			afu_result = afu_dev->driver->err_handler->error_detected(afu_dev,
2133 										  state);
2134 		/* Disconnect trumps all, NONE trumps NEED_RESET */
2135 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2136 			result = PCI_ERS_RESULT_DISCONNECT;
2137 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
2138 			 (result == PCI_ERS_RESULT_NEED_RESET))
2139 			result = PCI_ERS_RESULT_NONE;
2140 	}
2141 	return result;
2142 }
2143 
2144 static pci_ers_result_t cxl_pci_error_detected(struct pci_dev *pdev,
2145 					       pci_channel_state_t state)
2146 {
2147 	struct cxl *adapter = pci_get_drvdata(pdev);
2148 	struct cxl_afu *afu;
2149 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET, afu_result;
2150 	int i;
2151 
2152 	/* At this point, we could still have an interrupt pending.
2153 	 * Let's try to get them out of the way before they do
2154 	 * anything we don't like.
2155 	 */
2156 	schedule();
2157 
2158 	/* If we're permanently dead, give up. */
2159 	if (state == pci_channel_io_perm_failure) {
2160 		for (i = 0; i < adapter->slices; i++) {
2161 			afu = adapter->afu[i];
2162 			/*
2163 			 * Tell the AFU drivers; but we don't care what they
2164 			 * say, we're going away.
2165 			 */
2166 			cxl_vphb_error_detected(afu, state);
2167 		}
2168 		return PCI_ERS_RESULT_DISCONNECT;
2169 	}
2170 
2171 	/* Are we reflashing?
2172 	 *
2173 	 * If we reflash, we could come back as something entirely
2174 	 * different, including a non-CAPI card. As such, by default
2175 	 * we don't participate in the process. We'll be unbound and
2176 	 * the slot re-probed. (TODO: check EEH doesn't blindly rebind
2177 	 * us!)
2178 	 *
2179 	 * However, this isn't the entire story: for reliablity
2180 	 * reasons, we usually want to reflash the FPGA on PERST in
2181 	 * order to get back to a more reliable known-good state.
2182 	 *
2183 	 * This causes us a bit of a problem: if we reflash we can't
2184 	 * trust that we'll come back the same - we could have a new
2185 	 * image and been PERSTed in order to load that
2186 	 * image. However, most of the time we actually *will* come
2187 	 * back the same - for example a regular EEH event.
2188 	 *
2189 	 * Therefore, we allow the user to assert that the image is
2190 	 * indeed the same and that we should continue on into EEH
2191 	 * anyway.
2192 	 */
2193 	if (adapter->perst_loads_image && !adapter->perst_same_image) {
2194 		/* TODO take the PHB out of CXL mode */
2195 		dev_info(&pdev->dev, "reflashing, so opting out of EEH!\n");
2196 		return PCI_ERS_RESULT_NONE;
2197 	}
2198 
2199 	/*
2200 	 * At this point, we want to try to recover.  We'll always
2201 	 * need a complete slot reset: we don't trust any other reset.
2202 	 *
2203 	 * Now, we go through each AFU:
2204 	 *  - We send the driver, if bound, an error_detected callback.
2205 	 *    We expect it to clean up, but it can also tell us to give
2206 	 *    up and permanently detach the card. To simplify things, if
2207 	 *    any bound AFU driver doesn't support EEH, we give up on EEH.
2208 	 *
2209 	 *  - We detach all contexts associated with the AFU. This
2210 	 *    does not free them, but puts them into a CLOSED state
2211 	 *    which causes any the associated files to return useful
2212 	 *    errors to userland. It also unmaps, but does not free,
2213 	 *    any IRQs.
2214 	 *
2215 	 *  - We clean up our side: releasing and unmapping resources we hold
2216 	 *    so we can wire them up again when the hardware comes back up.
2217 	 *
2218 	 * Driver authors should note:
2219 	 *
2220 	 *  - Any contexts you create in your kernel driver (except
2221 	 *    those associated with anonymous file descriptors) are
2222 	 *    your responsibility to free and recreate. Likewise with
2223 	 *    any attached resources.
2224 	 *
2225 	 *  - We will take responsibility for re-initialising the
2226 	 *    device context (the one set up for you in
2227 	 *    cxl_pci_enable_device_hook and accessed through
2228 	 *    cxl_get_context). If you've attached IRQs or other
2229 	 *    resources to it, they remains yours to free.
2230 	 *
2231 	 * You can call the same functions to release resources as you
2232 	 * normally would: we make sure that these functions continue
2233 	 * to work when the hardware is down.
2234 	 *
2235 	 * Two examples:
2236 	 *
2237 	 * 1) If you normally free all your resources at the end of
2238 	 *    each request, or if you use anonymous FDs, your
2239 	 *    error_detected callback can simply set a flag to tell
2240 	 *    your driver not to start any new calls. You can then
2241 	 *    clear the flag in the resume callback.
2242 	 *
2243 	 * 2) If you normally allocate your resources on startup:
2244 	 *     * Set a flag in error_detected as above.
2245 	 *     * Let CXL detach your contexts.
2246 	 *     * In slot_reset, free the old resources and allocate new ones.
2247 	 *     * In resume, clear the flag to allow things to start.
2248 	 */
2249 	for (i = 0; i < adapter->slices; i++) {
2250 		afu = adapter->afu[i];
2251 
2252 		afu_result = cxl_vphb_error_detected(afu, state);
2253 
2254 		cxl_context_detach_all(afu);
2255 		cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
2256 		pci_deconfigure_afu(afu);
2257 
2258 		/* Disconnect trumps all, NONE trumps NEED_RESET */
2259 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2260 			result = PCI_ERS_RESULT_DISCONNECT;
2261 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
2262 			 (result == PCI_ERS_RESULT_NEED_RESET))
2263 			result = PCI_ERS_RESULT_NONE;
2264 	}
2265 
2266 	/* should take the context lock here */
2267 	if (cxl_adapter_context_lock(adapter) != 0)
2268 		dev_warn(&adapter->dev,
2269 			 "Couldn't take context lock with %d active-contexts\n",
2270 			 atomic_read(&adapter->contexts_num));
2271 
2272 	cxl_deconfigure_adapter(adapter);
2273 
2274 	return result;
2275 }
2276 
2277 static pci_ers_result_t cxl_pci_slot_reset(struct pci_dev *pdev)
2278 {
2279 	struct cxl *adapter = pci_get_drvdata(pdev);
2280 	struct cxl_afu *afu;
2281 	struct cxl_context *ctx;
2282 	struct pci_dev *afu_dev;
2283 	pci_ers_result_t afu_result = PCI_ERS_RESULT_RECOVERED;
2284 	pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
2285 	int i;
2286 
2287 	if (cxl_configure_adapter(adapter, pdev))
2288 		goto err;
2289 
2290 	/*
2291 	 * Unlock context activation for the adapter. Ideally this should be
2292 	 * done in cxl_pci_resume but cxlflash module tries to activate the
2293 	 * master context as part of slot_reset callback.
2294 	 */
2295 	cxl_adapter_context_unlock(adapter);
2296 
2297 	for (i = 0; i < adapter->slices; i++) {
2298 		afu = adapter->afu[i];
2299 
2300 		if (pci_configure_afu(afu, adapter, pdev))
2301 			goto err;
2302 
2303 		if (cxl_afu_select_best_mode(afu))
2304 			goto err;
2305 
2306 		if (afu->phb == NULL)
2307 			continue;
2308 
2309 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2310 			/* Reset the device context.
2311 			 * TODO: make this less disruptive
2312 			 */
2313 			ctx = cxl_get_context(afu_dev);
2314 
2315 			if (ctx && cxl_release_context(ctx))
2316 				goto err;
2317 
2318 			ctx = cxl_dev_context_init(afu_dev);
2319 			if (IS_ERR(ctx))
2320 				goto err;
2321 
2322 			afu_dev->dev.archdata.cxl_ctx = ctx;
2323 
2324 			if (cxl_ops->afu_check_and_enable(afu))
2325 				goto err;
2326 
2327 			afu_dev->error_state = pci_channel_io_normal;
2328 
2329 			/* If there's a driver attached, allow it to
2330 			 * chime in on recovery. Drivers should check
2331 			 * if everything has come back OK, but
2332 			 * shouldn't start new work until we call
2333 			 * their resume function.
2334 			 */
2335 			if (!afu_dev->driver)
2336 				continue;
2337 
2338 			if (afu_dev->driver->err_handler &&
2339 			    afu_dev->driver->err_handler->slot_reset)
2340 				afu_result = afu_dev->driver->err_handler->slot_reset(afu_dev);
2341 
2342 			if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2343 				result = PCI_ERS_RESULT_DISCONNECT;
2344 		}
2345 	}
2346 	return result;
2347 
2348 err:
2349 	/* All the bits that happen in both error_detected and cxl_remove
2350 	 * should be idempotent, so we don't need to worry about leaving a mix
2351 	 * of unconfigured and reconfigured resources.
2352 	 */
2353 	dev_err(&pdev->dev, "EEH recovery failed. Asking to be disconnected.\n");
2354 	return PCI_ERS_RESULT_DISCONNECT;
2355 }
2356 
2357 static void cxl_pci_resume(struct pci_dev *pdev)
2358 {
2359 	struct cxl *adapter = pci_get_drvdata(pdev);
2360 	struct cxl_afu *afu;
2361 	struct pci_dev *afu_dev;
2362 	int i;
2363 
2364 	/* Everything is back now. Drivers should restart work now.
2365 	 * This is not the place to be checking if everything came back up
2366 	 * properly, because there's no return value: do that in slot_reset.
2367 	 */
2368 	for (i = 0; i < adapter->slices; i++) {
2369 		afu = adapter->afu[i];
2370 
2371 		if (afu->phb == NULL)
2372 			continue;
2373 
2374 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2375 			if (afu_dev->driver && afu_dev->driver->err_handler &&
2376 			    afu_dev->driver->err_handler->resume)
2377 				afu_dev->driver->err_handler->resume(afu_dev);
2378 		}
2379 	}
2380 }
2381 
2382 static const struct pci_error_handlers cxl_err_handler = {
2383 	.error_detected = cxl_pci_error_detected,
2384 	.slot_reset = cxl_pci_slot_reset,
2385 	.resume = cxl_pci_resume,
2386 };
2387 
2388 struct pci_driver cxl_pci_driver = {
2389 	.name = "cxl-pci",
2390 	.id_table = cxl_pci_tbl,
2391 	.probe = cxl_probe,
2392 	.remove = cxl_remove,
2393 	.shutdown = cxl_remove,
2394 	.err_handler = &cxl_err_handler,
2395 };
2396