xref: /openbmc/linux/drivers/misc/cxl/pci.c (revision 020c5260)
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 	{ PCI_DEVICE_CLASS(0x120000, ~0), },
129 
130 	{ }
131 };
132 MODULE_DEVICE_TABLE(pci, cxl_pci_tbl);
133 
134 
135 /*
136  * Mostly using these wrappers to avoid confusion:
137  * priv 1 is BAR2, while priv 2 is BAR0
138  */
139 static inline resource_size_t p1_base(struct pci_dev *dev)
140 {
141 	return pci_resource_start(dev, 2);
142 }
143 
144 static inline resource_size_t p1_size(struct pci_dev *dev)
145 {
146 	return pci_resource_len(dev, 2);
147 }
148 
149 static inline resource_size_t p2_base(struct pci_dev *dev)
150 {
151 	return pci_resource_start(dev, 0);
152 }
153 
154 static inline resource_size_t p2_size(struct pci_dev *dev)
155 {
156 	return pci_resource_len(dev, 0);
157 }
158 
159 static int find_cxl_vsec(struct pci_dev *dev)
160 {
161 	int vsec = 0;
162 	u16 val;
163 
164 	while ((vsec = pci_find_next_ext_capability(dev, vsec, PCI_EXT_CAP_ID_VNDR))) {
165 		pci_read_config_word(dev, vsec + 0x4, &val);
166 		if (val == CXL_PCI_VSEC_ID)
167 			return vsec;
168 	}
169 	return 0;
170 
171 }
172 
173 static void dump_cxl_config_space(struct pci_dev *dev)
174 {
175 	int vsec;
176 	u32 val;
177 
178 	dev_info(&dev->dev, "dump_cxl_config_space\n");
179 
180 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_0, &val);
181 	dev_info(&dev->dev, "BAR0: %#.8x\n", val);
182 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_1, &val);
183 	dev_info(&dev->dev, "BAR1: %#.8x\n", val);
184 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_2, &val);
185 	dev_info(&dev->dev, "BAR2: %#.8x\n", val);
186 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_3, &val);
187 	dev_info(&dev->dev, "BAR3: %#.8x\n", val);
188 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_4, &val);
189 	dev_info(&dev->dev, "BAR4: %#.8x\n", val);
190 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_5, &val);
191 	dev_info(&dev->dev, "BAR5: %#.8x\n", val);
192 
193 	dev_info(&dev->dev, "p1 regs: %#llx, len: %#llx\n",
194 		p1_base(dev), p1_size(dev));
195 	dev_info(&dev->dev, "p2 regs: %#llx, len: %#llx\n",
196 		p2_base(dev), p2_size(dev));
197 	dev_info(&dev->dev, "BAR 4/5: %#llx, len: %#llx\n",
198 		pci_resource_start(dev, 4), pci_resource_len(dev, 4));
199 
200 	if (!(vsec = find_cxl_vsec(dev)))
201 		return;
202 
203 #define show_reg(name, what) \
204 	dev_info(&dev->dev, "cxl vsec: %30s: %#x\n", name, what)
205 
206 	pci_read_config_dword(dev, vsec + 0x0, &val);
207 	show_reg("Cap ID", (val >> 0) & 0xffff);
208 	show_reg("Cap Ver", (val >> 16) & 0xf);
209 	show_reg("Next Cap Ptr", (val >> 20) & 0xfff);
210 	pci_read_config_dword(dev, vsec + 0x4, &val);
211 	show_reg("VSEC ID", (val >> 0) & 0xffff);
212 	show_reg("VSEC Rev", (val >> 16) & 0xf);
213 	show_reg("VSEC Length",	(val >> 20) & 0xfff);
214 	pci_read_config_dword(dev, vsec + 0x8, &val);
215 	show_reg("Num AFUs", (val >> 0) & 0xff);
216 	show_reg("Status", (val >> 8) & 0xff);
217 	show_reg("Mode Control", (val >> 16) & 0xff);
218 	show_reg("Reserved", (val >> 24) & 0xff);
219 	pci_read_config_dword(dev, vsec + 0xc, &val);
220 	show_reg("PSL Rev", (val >> 0) & 0xffff);
221 	show_reg("CAIA Ver", (val >> 16) & 0xffff);
222 	pci_read_config_dword(dev, vsec + 0x10, &val);
223 	show_reg("Base Image Rev", (val >> 0) & 0xffff);
224 	show_reg("Reserved", (val >> 16) & 0x0fff);
225 	show_reg("Image Control", (val >> 28) & 0x3);
226 	show_reg("Reserved", (val >> 30) & 0x1);
227 	show_reg("Image Loaded", (val >> 31) & 0x1);
228 
229 	pci_read_config_dword(dev, vsec + 0x14, &val);
230 	show_reg("Reserved", val);
231 	pci_read_config_dword(dev, vsec + 0x18, &val);
232 	show_reg("Reserved", val);
233 	pci_read_config_dword(dev, vsec + 0x1c, &val);
234 	show_reg("Reserved", val);
235 
236 	pci_read_config_dword(dev, vsec + 0x20, &val);
237 	show_reg("AFU Descriptor Offset", val);
238 	pci_read_config_dword(dev, vsec + 0x24, &val);
239 	show_reg("AFU Descriptor Size", val);
240 	pci_read_config_dword(dev, vsec + 0x28, &val);
241 	show_reg("Problem State Offset", val);
242 	pci_read_config_dword(dev, vsec + 0x2c, &val);
243 	show_reg("Problem State Size", val);
244 
245 	pci_read_config_dword(dev, vsec + 0x30, &val);
246 	show_reg("Reserved", val);
247 	pci_read_config_dword(dev, vsec + 0x34, &val);
248 	show_reg("Reserved", val);
249 	pci_read_config_dword(dev, vsec + 0x38, &val);
250 	show_reg("Reserved", val);
251 	pci_read_config_dword(dev, vsec + 0x3c, &val);
252 	show_reg("Reserved", val);
253 
254 	pci_read_config_dword(dev, vsec + 0x40, &val);
255 	show_reg("PSL Programming Port", val);
256 	pci_read_config_dword(dev, vsec + 0x44, &val);
257 	show_reg("PSL Programming Control", val);
258 
259 	pci_read_config_dword(dev, vsec + 0x48, &val);
260 	show_reg("Reserved", val);
261 	pci_read_config_dword(dev, vsec + 0x4c, &val);
262 	show_reg("Reserved", val);
263 
264 	pci_read_config_dword(dev, vsec + 0x50, &val);
265 	show_reg("Flash Address Register", val);
266 	pci_read_config_dword(dev, vsec + 0x54, &val);
267 	show_reg("Flash Size Register", val);
268 	pci_read_config_dword(dev, vsec + 0x58, &val);
269 	show_reg("Flash Status/Control Register", val);
270 	pci_read_config_dword(dev, vsec + 0x58, &val);
271 	show_reg("Flash Data Port", val);
272 
273 #undef show_reg
274 }
275 
276 static void dump_afu_descriptor(struct cxl_afu *afu)
277 {
278 	u64 val, afu_cr_num, afu_cr_off, afu_cr_len;
279 	int i;
280 
281 #define show_reg(name, what) \
282 	dev_info(&afu->dev, "afu desc: %30s: %#llx\n", name, what)
283 
284 	val = AFUD_READ_INFO(afu);
285 	show_reg("num_ints_per_process", AFUD_NUM_INTS_PER_PROC(val));
286 	show_reg("num_of_processes", AFUD_NUM_PROCS(val));
287 	show_reg("num_of_afu_CRs", AFUD_NUM_CRS(val));
288 	show_reg("req_prog_mode", val & 0xffffULL);
289 	afu_cr_num = AFUD_NUM_CRS(val);
290 
291 	val = AFUD_READ(afu, 0x8);
292 	show_reg("Reserved", val);
293 	val = AFUD_READ(afu, 0x10);
294 	show_reg("Reserved", val);
295 	val = AFUD_READ(afu, 0x18);
296 	show_reg("Reserved", val);
297 
298 	val = AFUD_READ_CR(afu);
299 	show_reg("Reserved", (val >> (63-7)) & 0xff);
300 	show_reg("AFU_CR_len", AFUD_CR_LEN(val));
301 	afu_cr_len = AFUD_CR_LEN(val) * 256;
302 
303 	val = AFUD_READ_CR_OFF(afu);
304 	afu_cr_off = val;
305 	show_reg("AFU_CR_offset", val);
306 
307 	val = AFUD_READ_PPPSA(afu);
308 	show_reg("PerProcessPSA_control", (val >> (63-7)) & 0xff);
309 	show_reg("PerProcessPSA Length", AFUD_PPPSA_LEN(val));
310 
311 	val = AFUD_READ_PPPSA_OFF(afu);
312 	show_reg("PerProcessPSA_offset", val);
313 
314 	val = AFUD_READ_EB(afu);
315 	show_reg("Reserved", (val >> (63-7)) & 0xff);
316 	show_reg("AFU_EB_len", AFUD_EB_LEN(val));
317 
318 	val = AFUD_READ_EB_OFF(afu);
319 	show_reg("AFU_EB_offset", val);
320 
321 	for (i = 0; i < afu_cr_num; i++) {
322 		val = AFUD_READ_LE(afu, afu_cr_off + i * afu_cr_len);
323 		show_reg("CR Vendor", val & 0xffff);
324 		show_reg("CR Device", (val >> 16) & 0xffff);
325 	}
326 #undef show_reg
327 }
328 
329 #define P8_CAPP_UNIT0_ID 0xBA
330 #define P8_CAPP_UNIT1_ID 0XBE
331 #define P9_CAPP_UNIT0_ID 0xC0
332 #define P9_CAPP_UNIT1_ID 0xE0
333 
334 static int get_phb_index(struct device_node *np, u32 *phb_index)
335 {
336 	if (of_property_read_u32(np, "ibm,phb-index", phb_index))
337 		return -ENODEV;
338 	return 0;
339 }
340 
341 static u64 get_capp_unit_id(struct device_node *np, u32 phb_index)
342 {
343 	/*
344 	 * POWER 8:
345 	 *  - For chips other than POWER8NVL, we only have CAPP 0,
346 	 *    irrespective of which PHB is used.
347 	 *  - For POWER8NVL, assume CAPP 0 is attached to PHB0 and
348 	 *    CAPP 1 is attached to PHB1.
349 	 */
350 	if (cxl_is_power8()) {
351 		if (!pvr_version_is(PVR_POWER8NVL))
352 			return P8_CAPP_UNIT0_ID;
353 
354 		if (phb_index == 0)
355 			return P8_CAPP_UNIT0_ID;
356 
357 		if (phb_index == 1)
358 			return P8_CAPP_UNIT1_ID;
359 	}
360 
361 	/*
362 	 * POWER 9:
363 	 *   PEC0 (PHB0). Capp ID = CAPP0 (0b1100_0000)
364 	 *   PEC1 (PHB1 - PHB2). No capi mode
365 	 *   PEC2 (PHB3 - PHB4 - PHB5): Capi mode on PHB3 only. Capp ID = CAPP1 (0b1110_0000)
366 	 */
367 	if (cxl_is_power9()) {
368 		if (phb_index == 0)
369 			return P9_CAPP_UNIT0_ID;
370 
371 		if (phb_index == 3)
372 			return P9_CAPP_UNIT1_ID;
373 	}
374 
375 	return 0;
376 }
377 
378 static int calc_capp_routing(struct pci_dev *dev, u64 *chipid,
379 			     u32 *phb_index, u64 *capp_unit_id)
380 {
381 	int rc;
382 	struct device_node *np;
383 	const __be32 *prop;
384 
385 	if (!(np = pnv_pci_get_phb_node(dev)))
386 		return -ENODEV;
387 
388 	while (np && !(prop = of_get_property(np, "ibm,chip-id", NULL)))
389 		np = of_get_next_parent(np);
390 	if (!np)
391 		return -ENODEV;
392 
393 	*chipid = be32_to_cpup(prop);
394 
395 	rc = get_phb_index(np, phb_index);
396 	if (rc) {
397 		pr_err("cxl: invalid phb index\n");
398 		return rc;
399 	}
400 
401 	*capp_unit_id = get_capp_unit_id(np, *phb_index);
402 	of_node_put(np);
403 	if (!*capp_unit_id) {
404 		pr_err("cxl: invalid capp unit id\n");
405 		return -ENODEV;
406 	}
407 
408 	return 0;
409 }
410 
411 static int init_implementation_adapter_regs_psl9(struct cxl *adapter, struct pci_dev *dev)
412 {
413 	u64 xsl_dsnctl, psl_fircntl;
414 	u64 chipid;
415 	u32 phb_index;
416 	u64 capp_unit_id;
417 	int rc;
418 
419 	rc = calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
420 	if (rc)
421 		return rc;
422 
423 	/*
424 	 * CAPI Identifier bits [0:7]
425 	 * bit 61:60 MSI bits --> 0
426 	 * bit 59 TVT selector --> 0
427 	 */
428 
429 	/*
430 	 * Tell XSL where to route data to.
431 	 * The field chipid should match the PHB CAPI_CMPM register
432 	 */
433 	xsl_dsnctl = ((u64)0x2 << (63-7)); /* Bit 57 */
434 	xsl_dsnctl |= (capp_unit_id << (63-15));
435 
436 	/* nMMU_ID Defaults to: b’000001001’*/
437 	xsl_dsnctl |= ((u64)0x09 << (63-28));
438 
439 	if (cxl_is_power9() && !cpu_has_feature(CPU_FTR_POWER9_DD1)) {
440 		/*
441 		 * Used to identify CAPI packets which should be sorted into
442 		 * the Non-Blocking queues by the PHB. This field should match
443 		 * the PHB PBL_NBW_CMPM register
444 		 * nbwind=0x03, bits [57:58], must include capi indicator.
445 		 * Not supported on P9 DD1.
446 		 */
447 		xsl_dsnctl |= ((u64)0x03 << (63-47));
448 
449 		/*
450 		 * Upper 16b address bits of ASB_Notify messages sent to the
451 		 * system. Need to match the PHB’s ASN Compare/Mask Register.
452 		 * Not supported on P9 DD1.
453 		 */
454 		xsl_dsnctl |= ((u64)0x04 << (63-55));
455 	}
456 
457 	cxl_p1_write(adapter, CXL_XSL9_DSNCTL, xsl_dsnctl);
458 
459 	/* Set fir_cntl to recommended value for production env */
460 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
461 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
462 	psl_fircntl |= 0x1ULL; /* ce_thresh */
463 	cxl_p1_write(adapter, CXL_PSL9_FIR_CNTL, psl_fircntl);
464 
465 	/* vccredits=0x1  pcklat=0x4 */
466 	cxl_p1_write(adapter, CXL_PSL9_DSNDCTL, 0x0000000000001810ULL);
467 
468 	/*
469 	 * For debugging with trace arrays.
470 	 * Configure RX trace 0 segmented mode.
471 	 * Configure CT trace 0 segmented mode.
472 	 * Configure LA0 trace 0 segmented mode.
473 	 * Configure LA1 trace 0 segmented mode.
474 	 */
475 	cxl_p1_write(adapter, CXL_PSL9_TRACECFG, 0x8040800080000000ULL);
476 	cxl_p1_write(adapter, CXL_PSL9_TRACECFG, 0x8040800080000003ULL);
477 	cxl_p1_write(adapter, CXL_PSL9_TRACECFG, 0x8040800080000005ULL);
478 	cxl_p1_write(adapter, CXL_PSL9_TRACECFG, 0x8040800080000006ULL);
479 
480 	/*
481 	 * A response to an ASB_Notify request is returned by the
482 	 * system as an MMIO write to the address defined in
483 	 * the PSL_TNR_ADDR register
484 	 */
485 	/* PSL_TNR_ADDR */
486 
487 	/* NORST */
488 	cxl_p1_write(adapter, CXL_PSL9_DEBUG, 0x8000000000000000ULL);
489 
490 	/* allocate the apc machines */
491 	cxl_p1_write(adapter, CXL_PSL9_APCDEDTYPE, 0x40000003FFFF0000ULL);
492 
493 	/* Disable vc dd1 fix */
494 	if ((cxl_is_power9() && cpu_has_feature(CPU_FTR_POWER9_DD1)))
495 		cxl_p1_write(adapter, CXL_PSL9_GP_CT, 0x0400000000000001ULL);
496 
497 	return 0;
498 }
499 
500 static int init_implementation_adapter_regs_psl8(struct cxl *adapter, struct pci_dev *dev)
501 {
502 	u64 psl_dsnctl, psl_fircntl;
503 	u64 chipid;
504 	u32 phb_index;
505 	u64 capp_unit_id;
506 	int rc;
507 
508 	rc = calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
509 	if (rc)
510 		return rc;
511 
512 	psl_dsnctl = 0x0000900000000000ULL; /* pteupd ttype, scdone */
513 	psl_dsnctl |= (0x2ULL << (63-38)); /* MMIO hang pulse: 256 us */
514 	/* Tell PSL where to route data to */
515 	psl_dsnctl |= (chipid << (63-5));
516 	psl_dsnctl |= (capp_unit_id << (63-13));
517 
518 	cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
519 	cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
520 	/* snoop write mask */
521 	cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
522 	/* set fir_cntl to recommended value for production env */
523 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
524 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
525 	psl_fircntl |= 0x1ULL; /* ce_thresh */
526 	cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, psl_fircntl);
527 	/* for debugging with trace arrays */
528 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
529 
530 	return 0;
531 }
532 
533 static int init_implementation_adapter_regs_xsl(struct cxl *adapter, struct pci_dev *dev)
534 {
535 	u64 xsl_dsnctl;
536 	u64 chipid;
537 	u32 phb_index;
538 	u64 capp_unit_id;
539 	int rc;
540 
541 	rc = calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
542 	if (rc)
543 		return rc;
544 
545 	/* Tell XSL where to route data to */
546 	xsl_dsnctl = 0x0000600000000000ULL | (chipid << (63-5));
547 	xsl_dsnctl |= (capp_unit_id << (63-13));
548 	cxl_p1_write(adapter, CXL_XSL_DSNCTL, xsl_dsnctl);
549 
550 	return 0;
551 }
552 
553 /* PSL & XSL */
554 #define TBSYNC_CAL(n) (((u64)n & 0x7) << (63-3))
555 #define TBSYNC_CNT(n) (((u64)n & 0x7) << (63-6))
556 /* For the PSL this is a multiple for 0 < n <= 7: */
557 #define PSL_2048_250MHZ_CYCLES 1
558 
559 static void write_timebase_ctrl_psl9(struct cxl *adapter)
560 {
561 	cxl_p1_write(adapter, CXL_PSL9_TB_CTLSTAT,
562 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
563 }
564 
565 static void write_timebase_ctrl_psl8(struct cxl *adapter)
566 {
567 	cxl_p1_write(adapter, CXL_PSL_TB_CTLSTAT,
568 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
569 }
570 
571 /* XSL */
572 #define TBSYNC_ENA (1ULL << 63)
573 /* For the XSL this is 2**n * 2000 clocks for 0 < n <= 6: */
574 #define XSL_2000_CLOCKS 1
575 #define XSL_4000_CLOCKS 2
576 #define XSL_8000_CLOCKS 3
577 
578 static void write_timebase_ctrl_xsl(struct cxl *adapter)
579 {
580 	cxl_p1_write(adapter, CXL_XSL_TB_CTLSTAT,
581 		     TBSYNC_ENA |
582 		     TBSYNC_CAL(3) |
583 		     TBSYNC_CNT(XSL_4000_CLOCKS));
584 }
585 
586 static u64 timebase_read_psl9(struct cxl *adapter)
587 {
588 	return cxl_p1_read(adapter, CXL_PSL9_Timebase);
589 }
590 
591 static u64 timebase_read_psl8(struct cxl *adapter)
592 {
593 	return cxl_p1_read(adapter, CXL_PSL_Timebase);
594 }
595 
596 static u64 timebase_read_xsl(struct cxl *adapter)
597 {
598 	return cxl_p1_read(adapter, CXL_XSL_Timebase);
599 }
600 
601 static void cxl_setup_psl_timebase(struct cxl *adapter, struct pci_dev *dev)
602 {
603 	u64 psl_tb;
604 	int delta;
605 	unsigned int retry = 0;
606 	struct device_node *np;
607 
608 	adapter->psl_timebase_synced = false;
609 
610 	if (!(np = pnv_pci_get_phb_node(dev)))
611 		return;
612 
613 	/* Do not fail when CAPP timebase sync is not supported by OPAL */
614 	of_node_get(np);
615 	if (! of_get_property(np, "ibm,capp-timebase-sync", NULL)) {
616 		of_node_put(np);
617 		dev_info(&dev->dev, "PSL timebase inactive: OPAL support missing\n");
618 		return;
619 	}
620 	of_node_put(np);
621 
622 	/*
623 	 * Setup PSL Timebase Control and Status register
624 	 * with the recommended Timebase Sync Count value
625 	 */
626 	adapter->native->sl_ops->write_timebase_ctrl(adapter);
627 
628 	/* Enable PSL Timebase */
629 	cxl_p1_write(adapter, CXL_PSL_Control, 0x0000000000000000);
630 	cxl_p1_write(adapter, CXL_PSL_Control, CXL_PSL_Control_tb);
631 
632 	/* Wait until CORE TB and PSL TB difference <= 16usecs */
633 	do {
634 		msleep(1);
635 		if (retry++ > 5) {
636 			dev_info(&dev->dev, "PSL timebase can't synchronize\n");
637 			return;
638 		}
639 		psl_tb = adapter->native->sl_ops->timebase_read(adapter);
640 		delta = mftb() - psl_tb;
641 		if (delta < 0)
642 			delta = -delta;
643 	} while (tb_to_ns(delta) > 16000);
644 
645 	adapter->psl_timebase_synced = true;
646 	return;
647 }
648 
649 static int init_implementation_afu_regs_psl9(struct cxl_afu *afu)
650 {
651 	return 0;
652 }
653 
654 static int init_implementation_afu_regs_psl8(struct cxl_afu *afu)
655 {
656 	/* read/write masks for this slice */
657 	cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
658 	/* APC read/write masks for this slice */
659 	cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
660 	/* for debugging with trace arrays */
661 	cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
662 	cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
663 
664 	return 0;
665 }
666 
667 int cxl_pci_setup_irq(struct cxl *adapter, unsigned int hwirq,
668 		unsigned int virq)
669 {
670 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
671 
672 	return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
673 }
674 
675 int cxl_update_image_control(struct cxl *adapter)
676 {
677 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
678 	int rc;
679 	int vsec;
680 	u8 image_state;
681 
682 	if (!(vsec = find_cxl_vsec(dev))) {
683 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
684 		return -ENODEV;
685 	}
686 
687 	if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
688 		dev_err(&dev->dev, "failed to read image state: %i\n", rc);
689 		return rc;
690 	}
691 
692 	if (adapter->perst_loads_image)
693 		image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
694 	else
695 		image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
696 
697 	if (adapter->perst_select_user)
698 		image_state |= CXL_VSEC_PERST_SELECT_USER;
699 	else
700 		image_state &= ~CXL_VSEC_PERST_SELECT_USER;
701 
702 	if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
703 		dev_err(&dev->dev, "failed to update image control: %i\n", rc);
704 		return rc;
705 	}
706 
707 	return 0;
708 }
709 
710 int cxl_pci_alloc_one_irq(struct cxl *adapter)
711 {
712 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
713 
714 	return pnv_cxl_alloc_hwirqs(dev, 1);
715 }
716 
717 void cxl_pci_release_one_irq(struct cxl *adapter, int hwirq)
718 {
719 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
720 
721 	return pnv_cxl_release_hwirqs(dev, hwirq, 1);
722 }
723 
724 int cxl_pci_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
725 			struct cxl *adapter, unsigned int num)
726 {
727 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
728 
729 	return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
730 }
731 
732 void cxl_pci_release_irq_ranges(struct cxl_irq_ranges *irqs,
733 				struct cxl *adapter)
734 {
735 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
736 
737 	pnv_cxl_release_hwirq_ranges(irqs, dev);
738 }
739 
740 static int setup_cxl_bars(struct pci_dev *dev)
741 {
742 	/* Safety check in case we get backported to < 3.17 without M64 */
743 	if ((p1_base(dev) < 0x100000000ULL) ||
744 	    (p2_base(dev) < 0x100000000ULL)) {
745 		dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
746 		return -ENODEV;
747 	}
748 
749 	/*
750 	 * BAR 4/5 has a special meaning for CXL and must be programmed with a
751 	 * special value corresponding to the CXL protocol address range.
752 	 * For POWER 8/9 that means bits 48:49 must be set to 10
753 	 */
754 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
755 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
756 
757 	return 0;
758 }
759 
760 #ifdef CONFIG_CXL_BIMODAL
761 
762 struct cxl_switch_work {
763 	struct pci_dev *dev;
764 	struct work_struct work;
765 	int vsec;
766 	int mode;
767 };
768 
769 static void switch_card_to_cxl(struct work_struct *work)
770 {
771 	struct cxl_switch_work *switch_work =
772 		container_of(work, struct cxl_switch_work, work);
773 	struct pci_dev *dev = switch_work->dev;
774 	struct pci_bus *bus = dev->bus;
775 	struct pci_controller *hose = pci_bus_to_host(bus);
776 	struct pci_dev *bridge;
777 	struct pnv_php_slot *php_slot;
778 	unsigned int devfn;
779 	u8 val;
780 	int rc;
781 
782 	dev_info(&bus->dev, "cxl: Preparing for mode switch...\n");
783 	bridge = list_first_entry_or_null(&hose->bus->devices, struct pci_dev,
784 					  bus_list);
785 	if (!bridge) {
786 		dev_WARN(&bus->dev, "cxl: Couldn't find root port!\n");
787 		goto err_dev_put;
788 	}
789 
790 	php_slot = pnv_php_find_slot(pci_device_to_OF_node(bridge));
791 	if (!php_slot) {
792 		dev_err(&bus->dev, "cxl: Failed to find slot hotplug "
793 			           "information. You may need to upgrade "
794 			           "skiboot. Aborting.\n");
795 		goto err_dev_put;
796 	}
797 
798 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, switch_work->vsec, &val);
799 	if (rc) {
800 		dev_err(&bus->dev, "cxl: Failed to read CAPI mode control: %i\n", rc);
801 		goto err_dev_put;
802 	}
803 	devfn = dev->devfn;
804 
805 	/* Release the reference obtained in cxl_check_and_switch_mode() */
806 	pci_dev_put(dev);
807 
808 	dev_dbg(&bus->dev, "cxl: Removing PCI devices from kernel\n");
809 	pci_lock_rescan_remove();
810 	pci_hp_remove_devices(bridge->subordinate);
811 	pci_unlock_rescan_remove();
812 
813 	/* Switch the CXL protocol on the card */
814 	if (switch_work->mode == CXL_BIMODE_CXL) {
815 		dev_info(&bus->dev, "cxl: Switching card to CXL mode\n");
816 		val &= ~CXL_VSEC_PROTOCOL_MASK;
817 		val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
818 		rc = pnv_cxl_enable_phb_kernel_api(hose, true);
819 		if (rc) {
820 			dev_err(&bus->dev, "cxl: Failed to enable kernel API"
821 				           " on real PHB, aborting\n");
822 			goto err_free_work;
823 		}
824 	} else {
825 		dev_WARN(&bus->dev, "cxl: Switching card to PCI mode not supported!\n");
826 		goto err_free_work;
827 	}
828 
829 	rc = CXL_WRITE_VSEC_MODE_CONTROL_BUS(bus, devfn, switch_work->vsec, val);
830 	if (rc) {
831 		dev_err(&bus->dev, "cxl: Failed to configure CXL protocol: %i\n", rc);
832 		goto err_free_work;
833 	}
834 
835 	/*
836 	 * The CAIA spec (v1.1, Section 10.6 Bi-modal Device Support) states
837 	 * we must wait 100ms after this mode switch before touching PCIe config
838 	 * space.
839 	 */
840 	msleep(100);
841 
842 	/*
843 	 * Hot reset to cause the card to come back in cxl mode. A
844 	 * OPAL_RESET_PCI_LINK would be sufficient, but currently lacks support
845 	 * in skiboot, so we use a hot reset instead.
846 	 *
847 	 * We call pci_set_pcie_reset_state() on the bridge, as a CAPI card is
848 	 * guaranteed to sit directly under the root port, and setting the reset
849 	 * state on a device directly under the root port is equivalent to doing
850 	 * it on the root port iself.
851 	 */
852 	dev_info(&bus->dev, "cxl: Configuration write complete, resetting card\n");
853 	pci_set_pcie_reset_state(bridge, pcie_hot_reset);
854 	pci_set_pcie_reset_state(bridge, pcie_deassert_reset);
855 
856 	dev_dbg(&bus->dev, "cxl: Offlining slot\n");
857 	rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_OFFLINE);
858 	if (rc) {
859 		dev_err(&bus->dev, "cxl: OPAL offlining call failed: %i\n", rc);
860 		goto err_free_work;
861 	}
862 
863 	dev_dbg(&bus->dev, "cxl: Onlining and probing slot\n");
864 	rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_ONLINE);
865 	if (rc) {
866 		dev_err(&bus->dev, "cxl: OPAL onlining call failed: %i\n", rc);
867 		goto err_free_work;
868 	}
869 
870 	pci_lock_rescan_remove();
871 	pci_hp_add_devices(bridge->subordinate);
872 	pci_unlock_rescan_remove();
873 
874 	dev_info(&bus->dev, "cxl: CAPI mode switch completed\n");
875 	kfree(switch_work);
876 	return;
877 
878 err_dev_put:
879 	/* Release the reference obtained in cxl_check_and_switch_mode() */
880 	pci_dev_put(dev);
881 err_free_work:
882 	kfree(switch_work);
883 }
884 
885 int cxl_check_and_switch_mode(struct pci_dev *dev, int mode, int vsec)
886 {
887 	struct cxl_switch_work *work;
888 	u8 val;
889 	int rc;
890 
891 	if (!cpu_has_feature(CPU_FTR_HVMODE))
892 		return -ENODEV;
893 
894 	if (!vsec) {
895 		vsec = find_cxl_vsec(dev);
896 		if (!vsec) {
897 			dev_info(&dev->dev, "CXL VSEC not found\n");
898 			return -ENODEV;
899 		}
900 	}
901 
902 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
903 	if (rc) {
904 		dev_err(&dev->dev, "Failed to read current mode control: %i", rc);
905 		return rc;
906 	}
907 
908 	if (mode == CXL_BIMODE_PCI) {
909 		if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
910 			dev_info(&dev->dev, "Card is already in PCI mode\n");
911 			return 0;
912 		}
913 		/*
914 		 * TODO: Before it's safe to switch the card back to PCI mode
915 		 * we need to disable the CAPP and make sure any cachelines the
916 		 * card holds have been flushed out. Needs skiboot support.
917 		 */
918 		dev_WARN(&dev->dev, "CXL mode switch to PCI unsupported!\n");
919 		return -EIO;
920 	}
921 
922 	if (val & CXL_VSEC_PROTOCOL_ENABLE) {
923 		dev_info(&dev->dev, "Card is already in CXL mode\n");
924 		return 0;
925 	}
926 
927 	dev_info(&dev->dev, "Card is in PCI mode, scheduling kernel thread "
928 			    "to switch to CXL mode\n");
929 
930 	work = kmalloc(sizeof(struct cxl_switch_work), GFP_KERNEL);
931 	if (!work)
932 		return -ENOMEM;
933 
934 	pci_dev_get(dev);
935 	work->dev = dev;
936 	work->vsec = vsec;
937 	work->mode = mode;
938 	INIT_WORK(&work->work, switch_card_to_cxl);
939 
940 	schedule_work(&work->work);
941 
942 	/*
943 	 * We return a failure now to abort the driver init. Once the
944 	 * link has been cycled and the card is in cxl mode we will
945 	 * come back (possibly using the generic cxl driver), but
946 	 * return success as the card should then be in cxl mode.
947 	 *
948 	 * TODO: What if the card comes back in PCI mode even after
949 	 *       the switch?  Don't want to spin endlessly.
950 	 */
951 	return -EBUSY;
952 }
953 EXPORT_SYMBOL_GPL(cxl_check_and_switch_mode);
954 
955 #endif /* CONFIG_CXL_BIMODAL */
956 
957 static int setup_cxl_protocol_area(struct pci_dev *dev)
958 {
959 	u8 val;
960 	int rc;
961 	int vsec = find_cxl_vsec(dev);
962 
963 	if (!vsec) {
964 		dev_info(&dev->dev, "CXL VSEC not found\n");
965 		return -ENODEV;
966 	}
967 
968 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
969 	if (rc) {
970 		dev_err(&dev->dev, "Failed to read current mode control: %i\n", rc);
971 		return rc;
972 	}
973 
974 	if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
975 		dev_err(&dev->dev, "Card not in CAPI mode!\n");
976 		return -EIO;
977 	}
978 
979 	if ((val & CXL_VSEC_PROTOCOL_MASK) != CXL_VSEC_PROTOCOL_256TB) {
980 		val &= ~CXL_VSEC_PROTOCOL_MASK;
981 		val |= CXL_VSEC_PROTOCOL_256TB;
982 		rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val);
983 		if (rc) {
984 			dev_err(&dev->dev, "Failed to set CXL protocol area: %i\n", rc);
985 			return rc;
986 		}
987 	}
988 
989 	return 0;
990 }
991 
992 static int pci_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
993 {
994 	u64 p1n_base, p2n_base, afu_desc;
995 	const u64 p1n_size = 0x100;
996 	const u64 p2n_size = 0x1000;
997 
998 	p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
999 	p2n_base = p2_base(dev) + (afu->slice * p2n_size);
1000 	afu->psn_phys = p2_base(dev) + (adapter->native->ps_off + (afu->slice * adapter->ps_size));
1001 	afu_desc = p2_base(dev) + adapter->native->afu_desc_off + (afu->slice * adapter->native->afu_desc_size);
1002 
1003 	if (!(afu->native->p1n_mmio = ioremap(p1n_base, p1n_size)))
1004 		goto err;
1005 	if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
1006 		goto err1;
1007 	if (afu_desc) {
1008 		if (!(afu->native->afu_desc_mmio = ioremap(afu_desc, adapter->native->afu_desc_size)))
1009 			goto err2;
1010 	}
1011 
1012 	return 0;
1013 err2:
1014 	iounmap(afu->p2n_mmio);
1015 err1:
1016 	iounmap(afu->native->p1n_mmio);
1017 err:
1018 	dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
1019 	return -ENOMEM;
1020 }
1021 
1022 static void pci_unmap_slice_regs(struct cxl_afu *afu)
1023 {
1024 	if (afu->p2n_mmio) {
1025 		iounmap(afu->p2n_mmio);
1026 		afu->p2n_mmio = NULL;
1027 	}
1028 	if (afu->native->p1n_mmio) {
1029 		iounmap(afu->native->p1n_mmio);
1030 		afu->native->p1n_mmio = NULL;
1031 	}
1032 	if (afu->native->afu_desc_mmio) {
1033 		iounmap(afu->native->afu_desc_mmio);
1034 		afu->native->afu_desc_mmio = NULL;
1035 	}
1036 }
1037 
1038 void cxl_pci_release_afu(struct device *dev)
1039 {
1040 	struct cxl_afu *afu = to_cxl_afu(dev);
1041 
1042 	pr_devel("%s\n", __func__);
1043 
1044 	idr_destroy(&afu->contexts_idr);
1045 	cxl_release_spa(afu);
1046 
1047 	kfree(afu->native);
1048 	kfree(afu);
1049 }
1050 
1051 /* Expects AFU struct to have recently been zeroed out */
1052 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
1053 {
1054 	u64 val;
1055 
1056 	val = AFUD_READ_INFO(afu);
1057 	afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
1058 	afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
1059 	afu->crs_num = AFUD_NUM_CRS(val);
1060 
1061 	if (AFUD_AFU_DIRECTED(val))
1062 		afu->modes_supported |= CXL_MODE_DIRECTED;
1063 	if (AFUD_DEDICATED_PROCESS(val))
1064 		afu->modes_supported |= CXL_MODE_DEDICATED;
1065 	if (AFUD_TIME_SLICED(val))
1066 		afu->modes_supported |= CXL_MODE_TIME_SLICED;
1067 
1068 	val = AFUD_READ_PPPSA(afu);
1069 	afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
1070 	afu->psa = AFUD_PPPSA_PSA(val);
1071 	if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
1072 		afu->native->pp_offset = AFUD_READ_PPPSA_OFF(afu);
1073 
1074 	val = AFUD_READ_CR(afu);
1075 	afu->crs_len = AFUD_CR_LEN(val) * 256;
1076 	afu->crs_offset = AFUD_READ_CR_OFF(afu);
1077 
1078 
1079 	/* eb_len is in multiple of 4K */
1080 	afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
1081 	afu->eb_offset = AFUD_READ_EB_OFF(afu);
1082 
1083 	/* eb_off is 4K aligned so lower 12 bits are always zero */
1084 	if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
1085 		dev_warn(&afu->dev,
1086 			 "Invalid AFU error buffer offset %Lx\n",
1087 			 afu->eb_offset);
1088 		dev_info(&afu->dev,
1089 			 "Ignoring AFU error buffer in the descriptor\n");
1090 		/* indicate that no afu buffer exists */
1091 		afu->eb_len = 0;
1092 	}
1093 
1094 	return 0;
1095 }
1096 
1097 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
1098 {
1099 	int i, rc;
1100 	u32 val;
1101 
1102 	if (afu->psa && afu->adapter->ps_size <
1103 			(afu->native->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
1104 		dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
1105 		return -ENODEV;
1106 	}
1107 
1108 	if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
1109 		dev_warn(&afu->dev, "AFU uses pp_size(%#016llx) < PAGE_SIZE per-process PSA!\n", afu->pp_size);
1110 
1111 	for (i = 0; i < afu->crs_num; i++) {
1112 		rc = cxl_ops->afu_cr_read32(afu, i, 0, &val);
1113 		if (rc || val == 0) {
1114 			dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
1115 			return -EINVAL;
1116 		}
1117 	}
1118 
1119 	if ((afu->modes_supported & ~CXL_MODE_DEDICATED) && afu->max_procs_virtualised == 0) {
1120 		/*
1121 		 * We could also check this for the dedicated process model
1122 		 * since the architecture indicates it should be set to 1, but
1123 		 * in that case we ignore the value and I'd rather not risk
1124 		 * breaking any existing dedicated process AFUs that left it as
1125 		 * 0 (not that I'm aware of any). It is clearly an error for an
1126 		 * AFU directed AFU to set this to 0, and would have previously
1127 		 * triggered a bug resulting in the maximum not being enforced
1128 		 * at all since idr_alloc treats 0 as no maximum.
1129 		 */
1130 		dev_err(&afu->dev, "AFU does not support any processes\n");
1131 		return -EINVAL;
1132 	}
1133 
1134 	return 0;
1135 }
1136 
1137 static int sanitise_afu_regs_psl9(struct cxl_afu *afu)
1138 {
1139 	u64 reg;
1140 
1141 	/*
1142 	 * Clear out any regs that contain either an IVTE or address or may be
1143 	 * waiting on an acknowledgment to try to be a bit safer as we bring
1144 	 * it online
1145 	 */
1146 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
1147 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
1148 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
1149 		if (cxl_ops->afu_reset(afu))
1150 			return -EIO;
1151 		if (cxl_afu_disable(afu))
1152 			return -EIO;
1153 		if (cxl_psl_purge(afu))
1154 			return -EIO;
1155 	}
1156 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
1157 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1158 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1159 	if (reg) {
1160 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1161 		if (reg & CXL_PSL9_DSISR_An_TF)
1162 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1163 		else
1164 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1165 	}
1166 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1167 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1168 		if (reg) {
1169 			if (reg & ~0x000000007fffffff)
1170 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1171 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1172 		}
1173 	}
1174 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1175 	if (reg) {
1176 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1177 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1178 	}
1179 
1180 	return 0;
1181 }
1182 
1183 static int sanitise_afu_regs_psl8(struct cxl_afu *afu)
1184 {
1185 	u64 reg;
1186 
1187 	/*
1188 	 * Clear out any regs that contain either an IVTE or address or may be
1189 	 * waiting on an acknowledgement to try to be a bit safer as we bring
1190 	 * it online
1191 	 */
1192 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
1193 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
1194 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
1195 		if (cxl_ops->afu_reset(afu))
1196 			return -EIO;
1197 		if (cxl_afu_disable(afu))
1198 			return -EIO;
1199 		if (cxl_psl_purge(afu))
1200 			return -EIO;
1201 	}
1202 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
1203 	cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
1204 	cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
1205 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1206 	cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
1207 	cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
1208 	cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
1209 	cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
1210 	cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
1211 	cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
1212 	cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
1213 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1214 	if (reg) {
1215 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1216 		if (reg & CXL_PSL_DSISR_TRANS)
1217 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1218 		else
1219 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1220 	}
1221 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1222 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1223 		if (reg) {
1224 			if (reg & ~0xffff)
1225 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1226 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1227 		}
1228 	}
1229 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1230 	if (reg) {
1231 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1232 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1233 	}
1234 
1235 	return 0;
1236 }
1237 
1238 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
1239 /*
1240  * afu_eb_read:
1241  * Called from sysfs and reads the afu error info buffer. The h/w only supports
1242  * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
1243  * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
1244  */
1245 ssize_t cxl_pci_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
1246 				loff_t off, size_t count)
1247 {
1248 	loff_t aligned_start, aligned_end;
1249 	size_t aligned_length;
1250 	void *tbuf;
1251 	const void __iomem *ebuf = afu->native->afu_desc_mmio + afu->eb_offset;
1252 
1253 	if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
1254 		return 0;
1255 
1256 	/* calculate aligned read window */
1257 	count = min((size_t)(afu->eb_len - off), count);
1258 	aligned_start = round_down(off, 8);
1259 	aligned_end = round_up(off + count, 8);
1260 	aligned_length = aligned_end - aligned_start;
1261 
1262 	/* max we can copy in one read is PAGE_SIZE */
1263 	if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
1264 		aligned_length = ERR_BUFF_MAX_COPY_SIZE;
1265 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
1266 	}
1267 
1268 	/* use bounce buffer for copy */
1269 	tbuf = (void *)__get_free_page(GFP_TEMPORARY);
1270 	if (!tbuf)
1271 		return -ENOMEM;
1272 
1273 	/* perform aligned read from the mmio region */
1274 	memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
1275 	memcpy(buf, tbuf + (off & 0x7), count);
1276 
1277 	free_page((unsigned long)tbuf);
1278 
1279 	return count;
1280 }
1281 
1282 static int pci_configure_afu(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1283 {
1284 	int rc;
1285 
1286 	if ((rc = pci_map_slice_regs(afu, adapter, dev)))
1287 		return rc;
1288 
1289 	if (adapter->native->sl_ops->sanitise_afu_regs) {
1290 		rc = adapter->native->sl_ops->sanitise_afu_regs(afu);
1291 		if (rc)
1292 			goto err1;
1293 	}
1294 
1295 	/* We need to reset the AFU before we can read the AFU descriptor */
1296 	if ((rc = cxl_ops->afu_reset(afu)))
1297 		goto err1;
1298 
1299 	if (cxl_verbose)
1300 		dump_afu_descriptor(afu);
1301 
1302 	if ((rc = cxl_read_afu_descriptor(afu)))
1303 		goto err1;
1304 
1305 	if ((rc = cxl_afu_descriptor_looks_ok(afu)))
1306 		goto err1;
1307 
1308 	if (adapter->native->sl_ops->afu_regs_init)
1309 		if ((rc = adapter->native->sl_ops->afu_regs_init(afu)))
1310 			goto err1;
1311 
1312 	if (adapter->native->sl_ops->register_serr_irq)
1313 		if ((rc = adapter->native->sl_ops->register_serr_irq(afu)))
1314 			goto err1;
1315 
1316 	if ((rc = cxl_native_register_psl_irq(afu)))
1317 		goto err2;
1318 
1319 	atomic_set(&afu->configured_state, 0);
1320 	return 0;
1321 
1322 err2:
1323 	if (adapter->native->sl_ops->release_serr_irq)
1324 		adapter->native->sl_ops->release_serr_irq(afu);
1325 err1:
1326 	pci_unmap_slice_regs(afu);
1327 	return rc;
1328 }
1329 
1330 static void pci_deconfigure_afu(struct cxl_afu *afu)
1331 {
1332 	/*
1333 	 * It's okay to deconfigure when AFU is already locked, otherwise wait
1334 	 * until there are no readers
1335 	 */
1336 	if (atomic_read(&afu->configured_state) != -1) {
1337 		while (atomic_cmpxchg(&afu->configured_state, 0, -1) != -1)
1338 			schedule();
1339 	}
1340 	cxl_native_release_psl_irq(afu);
1341 	if (afu->adapter->native->sl_ops->release_serr_irq)
1342 		afu->adapter->native->sl_ops->release_serr_irq(afu);
1343 	pci_unmap_slice_regs(afu);
1344 }
1345 
1346 static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
1347 {
1348 	struct cxl_afu *afu;
1349 	int rc = -ENOMEM;
1350 
1351 	afu = cxl_alloc_afu(adapter, slice);
1352 	if (!afu)
1353 		return -ENOMEM;
1354 
1355 	afu->native = kzalloc(sizeof(struct cxl_afu_native), GFP_KERNEL);
1356 	if (!afu->native)
1357 		goto err_free_afu;
1358 
1359 	mutex_init(&afu->native->spa_mutex);
1360 
1361 	rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice);
1362 	if (rc)
1363 		goto err_free_native;
1364 
1365 	rc = pci_configure_afu(afu, adapter, dev);
1366 	if (rc)
1367 		goto err_free_native;
1368 
1369 	/* Don't care if this fails */
1370 	cxl_debugfs_afu_add(afu);
1371 
1372 	/*
1373 	 * After we call this function we must not free the afu directly, even
1374 	 * if it returns an error!
1375 	 */
1376 	if ((rc = cxl_register_afu(afu)))
1377 		goto err_put1;
1378 
1379 	if ((rc = cxl_sysfs_afu_add(afu)))
1380 		goto err_put1;
1381 
1382 	adapter->afu[afu->slice] = afu;
1383 
1384 	if ((rc = cxl_pci_vphb_add(afu)))
1385 		dev_info(&afu->dev, "Can't register vPHB\n");
1386 
1387 	return 0;
1388 
1389 err_put1:
1390 	pci_deconfigure_afu(afu);
1391 	cxl_debugfs_afu_remove(afu);
1392 	device_unregister(&afu->dev);
1393 	return rc;
1394 
1395 err_free_native:
1396 	kfree(afu->native);
1397 err_free_afu:
1398 	kfree(afu);
1399 	return rc;
1400 
1401 }
1402 
1403 static void cxl_pci_remove_afu(struct cxl_afu *afu)
1404 {
1405 	pr_devel("%s\n", __func__);
1406 
1407 	if (!afu)
1408 		return;
1409 
1410 	cxl_pci_vphb_remove(afu);
1411 	cxl_sysfs_afu_remove(afu);
1412 	cxl_debugfs_afu_remove(afu);
1413 
1414 	spin_lock(&afu->adapter->afu_list_lock);
1415 	afu->adapter->afu[afu->slice] = NULL;
1416 	spin_unlock(&afu->adapter->afu_list_lock);
1417 
1418 	cxl_context_detach_all(afu);
1419 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1420 
1421 	pci_deconfigure_afu(afu);
1422 	device_unregister(&afu->dev);
1423 }
1424 
1425 int cxl_pci_reset(struct cxl *adapter)
1426 {
1427 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1428 	int rc;
1429 
1430 	if (adapter->perst_same_image) {
1431 		dev_warn(&dev->dev,
1432 			 "cxl: refusing to reset/reflash when perst_reloads_same_image is set.\n");
1433 		return -EINVAL;
1434 	}
1435 
1436 	dev_info(&dev->dev, "CXL reset\n");
1437 
1438 	/*
1439 	 * The adapter is about to be reset, so ignore errors.
1440 	 * Not supported on P9 DD1
1441 	 */
1442 	if ((cxl_is_power8()) ||
1443 	    ((cxl_is_power9() && !cpu_has_feature(CPU_FTR_POWER9_DD1))))
1444 		cxl_data_cache_flush(adapter);
1445 
1446 	/* pcie_warm_reset requests a fundamental pci reset which includes a
1447 	 * PERST assert/deassert.  PERST triggers a loading of the image
1448 	 * if "user" or "factory" is selected in sysfs */
1449 	if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
1450 		dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
1451 		return rc;
1452 	}
1453 
1454 	return rc;
1455 }
1456 
1457 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
1458 {
1459 	if (pci_request_region(dev, 2, "priv 2 regs"))
1460 		goto err1;
1461 	if (pci_request_region(dev, 0, "priv 1 regs"))
1462 		goto err2;
1463 
1464 	pr_devel("cxl_map_adapter_regs: p1: %#016llx %#llx, p2: %#016llx %#llx",
1465 			p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
1466 
1467 	if (!(adapter->native->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
1468 		goto err3;
1469 
1470 	if (!(adapter->native->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
1471 		goto err4;
1472 
1473 	return 0;
1474 
1475 err4:
1476 	iounmap(adapter->native->p1_mmio);
1477 	adapter->native->p1_mmio = NULL;
1478 err3:
1479 	pci_release_region(dev, 0);
1480 err2:
1481 	pci_release_region(dev, 2);
1482 err1:
1483 	return -ENOMEM;
1484 }
1485 
1486 static void cxl_unmap_adapter_regs(struct cxl *adapter)
1487 {
1488 	if (adapter->native->p1_mmio) {
1489 		iounmap(adapter->native->p1_mmio);
1490 		adapter->native->p1_mmio = NULL;
1491 		pci_release_region(to_pci_dev(adapter->dev.parent), 2);
1492 	}
1493 	if (adapter->native->p2_mmio) {
1494 		iounmap(adapter->native->p2_mmio);
1495 		adapter->native->p2_mmio = NULL;
1496 		pci_release_region(to_pci_dev(adapter->dev.parent), 0);
1497 	}
1498 }
1499 
1500 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
1501 {
1502 	int vsec;
1503 	u32 afu_desc_off, afu_desc_size;
1504 	u32 ps_off, ps_size;
1505 	u16 vseclen;
1506 	u8 image_state;
1507 
1508 	if (!(vsec = find_cxl_vsec(dev))) {
1509 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
1510 		return -ENODEV;
1511 	}
1512 
1513 	CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
1514 	if (vseclen < CXL_VSEC_MIN_SIZE) {
1515 		dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
1516 		return -EINVAL;
1517 	}
1518 
1519 	CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
1520 	CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
1521 	CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
1522 	CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
1523 	CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
1524 	CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
1525 	adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1526 	adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1527 	adapter->perst_loads_image = !!(image_state & CXL_VSEC_PERST_LOADS_IMAGE);
1528 
1529 	CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
1530 	CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
1531 	CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
1532 	CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
1533 	CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
1534 
1535 	/* Convert everything to bytes, because there is NO WAY I'd look at the
1536 	 * code a month later and forget what units these are in ;-) */
1537 	adapter->native->ps_off = ps_off * 64 * 1024;
1538 	adapter->ps_size = ps_size * 64 * 1024;
1539 	adapter->native->afu_desc_off = afu_desc_off * 64 * 1024;
1540 	adapter->native->afu_desc_size = afu_desc_size * 64 * 1024;
1541 
1542 	/* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
1543 	adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
1544 
1545 	return 0;
1546 }
1547 
1548 /*
1549  * Workaround a PCIe Host Bridge defect on some cards, that can cause
1550  * malformed Transaction Layer Packet (TLP) errors to be erroneously
1551  * reported. Mask this error in the Uncorrectable Error Mask Register.
1552  *
1553  * The upper nibble of the PSL revision is used to distinguish between
1554  * different cards. The affected ones have it set to 0.
1555  */
1556 static void cxl_fixup_malformed_tlp(struct cxl *adapter, struct pci_dev *dev)
1557 {
1558 	int aer;
1559 	u32 data;
1560 
1561 	if (adapter->psl_rev & 0xf000)
1562 		return;
1563 	if (!(aer = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR)))
1564 		return;
1565 	pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &data);
1566 	if (data & PCI_ERR_UNC_MALF_TLP)
1567 		if (data & PCI_ERR_UNC_INTN)
1568 			return;
1569 	data |= PCI_ERR_UNC_MALF_TLP;
1570 	data |= PCI_ERR_UNC_INTN;
1571 	pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, data);
1572 }
1573 
1574 static bool cxl_compatible_caia_version(struct cxl *adapter)
1575 {
1576 	if (cxl_is_power8() && (adapter->caia_major == 1))
1577 		return true;
1578 
1579 	if (cxl_is_power9() && (adapter->caia_major == 2))
1580 		return true;
1581 
1582 	return false;
1583 }
1584 
1585 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
1586 {
1587 	if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
1588 		return -EBUSY;
1589 
1590 	if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
1591 		dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
1592 		return -EINVAL;
1593 	}
1594 
1595 	if (!cxl_compatible_caia_version(adapter)) {
1596 		dev_info(&dev->dev, "Ignoring card. PSL type is not supported (caia version: %d)\n",
1597 			 adapter->caia_major);
1598 		return -ENODEV;
1599 	}
1600 
1601 	if (!adapter->slices) {
1602 		/* Once we support dynamic reprogramming we can use the card if
1603 		 * it supports loadable AFUs */
1604 		dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
1605 		return -EINVAL;
1606 	}
1607 
1608 	if (!adapter->native->afu_desc_off || !adapter->native->afu_desc_size) {
1609 		dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
1610 		return -EINVAL;
1611 	}
1612 
1613 	if (adapter->ps_size > p2_size(dev) - adapter->native->ps_off) {
1614 		dev_err(&dev->dev, "ABORTING: Problem state size larger than "
1615 				   "available in BAR2: 0x%llx > 0x%llx\n",
1616 			 adapter->ps_size, p2_size(dev) - adapter->native->ps_off);
1617 		return -EINVAL;
1618 	}
1619 
1620 	return 0;
1621 }
1622 
1623 ssize_t cxl_pci_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1624 {
1625 	return pci_read_vpd(to_pci_dev(adapter->dev.parent), 0, len, buf);
1626 }
1627 
1628 static void cxl_release_adapter(struct device *dev)
1629 {
1630 	struct cxl *adapter = to_cxl_adapter(dev);
1631 
1632 	pr_devel("cxl_release_adapter\n");
1633 
1634 	cxl_remove_adapter_nr(adapter);
1635 
1636 	kfree(adapter->native);
1637 	kfree(adapter);
1638 }
1639 
1640 #define CXL_PSL_ErrIVTE_tberror (0x1ull << (63-31))
1641 
1642 static int sanitise_adapter_regs(struct cxl *adapter)
1643 {
1644 	int rc = 0;
1645 
1646 	/* Clear PSL tberror bit by writing 1 to it */
1647 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, CXL_PSL_ErrIVTE_tberror);
1648 
1649 	if (adapter->native->sl_ops->invalidate_all) {
1650 		/* do not invalidate ERAT entries when not reloading on PERST */
1651 		if (cxl_is_power9() && (adapter->perst_loads_image))
1652 			return 0;
1653 		rc = adapter->native->sl_ops->invalidate_all(adapter);
1654 	}
1655 
1656 	return rc;
1657 }
1658 
1659 /* This should contain *only* operations that can safely be done in
1660  * both creation and recovery.
1661  */
1662 static int cxl_configure_adapter(struct cxl *adapter, struct pci_dev *dev)
1663 {
1664 	int rc;
1665 
1666 	adapter->dev.parent = &dev->dev;
1667 	adapter->dev.release = cxl_release_adapter;
1668 	pci_set_drvdata(dev, adapter);
1669 
1670 	rc = pci_enable_device(dev);
1671 	if (rc) {
1672 		dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1673 		return rc;
1674 	}
1675 
1676 	if ((rc = cxl_read_vsec(adapter, dev)))
1677 		return rc;
1678 
1679 	if ((rc = cxl_vsec_looks_ok(adapter, dev)))
1680 	        return rc;
1681 
1682 	cxl_fixup_malformed_tlp(adapter, dev);
1683 
1684 	if ((rc = setup_cxl_bars(dev)))
1685 		return rc;
1686 
1687 	if ((rc = setup_cxl_protocol_area(dev)))
1688 		return rc;
1689 
1690 	if ((rc = cxl_update_image_control(adapter)))
1691 		return rc;
1692 
1693 	if ((rc = cxl_map_adapter_regs(adapter, dev)))
1694 		return rc;
1695 
1696 	if ((rc = sanitise_adapter_regs(adapter)))
1697 		goto err;
1698 
1699 	if ((rc = adapter->native->sl_ops->adapter_regs_init(adapter, dev)))
1700 		goto err;
1701 
1702 	/* Required for devices using CAPP DMA mode, harmless for others */
1703 	pci_set_master(dev);
1704 
1705 	if ((rc = pnv_phb_to_cxl_mode(dev, adapter->native->sl_ops->capi_mode)))
1706 		goto err;
1707 
1708 	/* If recovery happened, the last step is to turn on snooping.
1709 	 * In the non-recovery case this has no effect */
1710 	if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON)))
1711 		goto err;
1712 
1713 	/* Ignore error, adapter init is not dependant on timebase sync */
1714 	cxl_setup_psl_timebase(adapter, dev);
1715 
1716 	if ((rc = cxl_native_register_psl_err_irq(adapter)))
1717 		goto err;
1718 
1719 	return 0;
1720 
1721 err:
1722 	cxl_unmap_adapter_regs(adapter);
1723 	return rc;
1724 
1725 }
1726 
1727 static void cxl_deconfigure_adapter(struct cxl *adapter)
1728 {
1729 	struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1730 
1731 	cxl_native_release_psl_err_irq(adapter);
1732 	cxl_unmap_adapter_regs(adapter);
1733 
1734 	pci_disable_device(pdev);
1735 }
1736 
1737 static const struct cxl_service_layer_ops psl9_ops = {
1738 	.adapter_regs_init = init_implementation_adapter_regs_psl9,
1739 	.invalidate_all = cxl_invalidate_all_psl9,
1740 	.afu_regs_init = init_implementation_afu_regs_psl9,
1741 	.sanitise_afu_regs = sanitise_afu_regs_psl9,
1742 	.register_serr_irq = cxl_native_register_serr_irq,
1743 	.release_serr_irq = cxl_native_release_serr_irq,
1744 	.handle_interrupt = cxl_irq_psl9,
1745 	.fail_irq = cxl_fail_irq_psl,
1746 	.activate_dedicated_process = cxl_activate_dedicated_process_psl9,
1747 	.attach_afu_directed = cxl_attach_afu_directed_psl9,
1748 	.attach_dedicated_process = cxl_attach_dedicated_process_psl9,
1749 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl9,
1750 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl9,
1751 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl9,
1752 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl9,
1753 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs,
1754 	.debugfs_stop_trace = cxl_stop_trace_psl9,
1755 	.write_timebase_ctrl = write_timebase_ctrl_psl9,
1756 	.timebase_read = timebase_read_psl9,
1757 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1758 	.needs_reset_before_disable = true,
1759 };
1760 
1761 static const struct cxl_service_layer_ops psl8_ops = {
1762 	.adapter_regs_init = init_implementation_adapter_regs_psl8,
1763 	.invalidate_all = cxl_invalidate_all_psl8,
1764 	.afu_regs_init = init_implementation_afu_regs_psl8,
1765 	.sanitise_afu_regs = sanitise_afu_regs_psl8,
1766 	.register_serr_irq = cxl_native_register_serr_irq,
1767 	.release_serr_irq = cxl_native_release_serr_irq,
1768 	.handle_interrupt = cxl_irq_psl8,
1769 	.fail_irq = cxl_fail_irq_psl,
1770 	.activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1771 	.attach_afu_directed = cxl_attach_afu_directed_psl8,
1772 	.attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1773 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1774 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl8,
1775 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl8,
1776 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl8,
1777 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs,
1778 	.debugfs_stop_trace = cxl_stop_trace_psl8,
1779 	.write_timebase_ctrl = write_timebase_ctrl_psl8,
1780 	.timebase_read = timebase_read_psl8,
1781 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1782 	.needs_reset_before_disable = true,
1783 };
1784 
1785 static const struct cxl_service_layer_ops xsl_ops = {
1786 	.adapter_regs_init = init_implementation_adapter_regs_xsl,
1787 	.invalidate_all = cxl_invalidate_all_psl8,
1788 	.sanitise_afu_regs = sanitise_afu_regs_psl8,
1789 	.handle_interrupt = cxl_irq_psl8,
1790 	.fail_irq = cxl_fail_irq_psl,
1791 	.activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1792 	.attach_afu_directed = cxl_attach_afu_directed_psl8,
1793 	.attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1794 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1795 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_xsl,
1796 	.write_timebase_ctrl = write_timebase_ctrl_xsl,
1797 	.timebase_read = timebase_read_xsl,
1798 	.capi_mode = OPAL_PHB_CAPI_MODE_DMA,
1799 };
1800 
1801 static void set_sl_ops(struct cxl *adapter, struct pci_dev *dev)
1802 {
1803 	if (dev->vendor == PCI_VENDOR_ID_MELLANOX && dev->device == 0x1013) {
1804 		/* Mellanox CX-4 */
1805 		dev_info(&dev->dev, "Device uses an XSL\n");
1806 		adapter->native->sl_ops = &xsl_ops;
1807 		adapter->min_pe = 1; /* Workaround for CX-4 hardware bug */
1808 	} else {
1809 		if (cxl_is_power8()) {
1810 			dev_info(&dev->dev, "Device uses a PSL8\n");
1811 			adapter->native->sl_ops = &psl8_ops;
1812 		} else {
1813 			dev_info(&dev->dev, "Device uses a PSL9\n");
1814 			adapter->native->sl_ops = &psl9_ops;
1815 		}
1816 	}
1817 }
1818 
1819 
1820 static struct cxl *cxl_pci_init_adapter(struct pci_dev *dev)
1821 {
1822 	struct cxl *adapter;
1823 	int rc;
1824 
1825 	adapter = cxl_alloc_adapter();
1826 	if (!adapter)
1827 		return ERR_PTR(-ENOMEM);
1828 
1829 	adapter->native = kzalloc(sizeof(struct cxl_native), GFP_KERNEL);
1830 	if (!adapter->native) {
1831 		rc = -ENOMEM;
1832 		goto err_release;
1833 	}
1834 
1835 	set_sl_ops(adapter, dev);
1836 
1837 	/* Set defaults for parameters which need to persist over
1838 	 * configure/reconfigure
1839 	 */
1840 	adapter->perst_loads_image = true;
1841 	adapter->perst_same_image = false;
1842 
1843 	rc = cxl_configure_adapter(adapter, dev);
1844 	if (rc) {
1845 		pci_disable_device(dev);
1846 		goto err_release;
1847 	}
1848 
1849 	/* Don't care if this one fails: */
1850 	cxl_debugfs_adapter_add(adapter);
1851 
1852 	/*
1853 	 * After we call this function we must not free the adapter directly,
1854 	 * even if it returns an error!
1855 	 */
1856 	if ((rc = cxl_register_adapter(adapter)))
1857 		goto err_put1;
1858 
1859 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1860 		goto err_put1;
1861 
1862 	/* Release the context lock as adapter is configured */
1863 	cxl_adapter_context_unlock(adapter);
1864 
1865 	return adapter;
1866 
1867 err_put1:
1868 	/* This should mirror cxl_remove_adapter, except without the
1869 	 * sysfs parts
1870 	 */
1871 	cxl_debugfs_adapter_remove(adapter);
1872 	cxl_deconfigure_adapter(adapter);
1873 	device_unregister(&adapter->dev);
1874 	return ERR_PTR(rc);
1875 
1876 err_release:
1877 	cxl_release_adapter(&adapter->dev);
1878 	return ERR_PTR(rc);
1879 }
1880 
1881 static void cxl_pci_remove_adapter(struct cxl *adapter)
1882 {
1883 	pr_devel("cxl_remove_adapter\n");
1884 
1885 	cxl_sysfs_adapter_remove(adapter);
1886 	cxl_debugfs_adapter_remove(adapter);
1887 
1888 	/*
1889 	 * Flush adapter datacache as its about to be removed.
1890 	 * Not supported on P9 DD1.
1891 	 */
1892 	if ((cxl_is_power8()) ||
1893 	    ((cxl_is_power9() && !cpu_has_feature(CPU_FTR_POWER9_DD1))))
1894 		cxl_data_cache_flush(adapter);
1895 
1896 	cxl_deconfigure_adapter(adapter);
1897 
1898 	device_unregister(&adapter->dev);
1899 }
1900 
1901 #define CXL_MAX_PCIEX_PARENT 2
1902 
1903 static int cxl_slot_is_switched(struct pci_dev *dev)
1904 {
1905 	struct device_node *np;
1906 	int depth = 0;
1907 	const __be32 *prop;
1908 
1909 	if (!(np = pci_device_to_OF_node(dev))) {
1910 		pr_err("cxl: np = NULL\n");
1911 		return -ENODEV;
1912 	}
1913 	of_node_get(np);
1914 	while (np) {
1915 		np = of_get_next_parent(np);
1916 		prop = of_get_property(np, "device_type", NULL);
1917 		if (!prop || strcmp((char *)prop, "pciex"))
1918 			break;
1919 		depth++;
1920 	}
1921 	of_node_put(np);
1922 	return (depth > CXL_MAX_PCIEX_PARENT);
1923 }
1924 
1925 bool cxl_slot_is_supported(struct pci_dev *dev, int flags)
1926 {
1927 	if (!cpu_has_feature(CPU_FTR_HVMODE))
1928 		return false;
1929 
1930 	if ((flags & CXL_SLOT_FLAG_DMA) && (!pvr_version_is(PVR_POWER8NVL))) {
1931 		/*
1932 		 * CAPP DMA mode is technically supported on regular P8, but
1933 		 * will EEH if the card attempts to access memory < 4GB, which
1934 		 * we cannot realistically avoid. We might be able to work
1935 		 * around the issue, but until then return unsupported:
1936 		 */
1937 		return false;
1938 	}
1939 
1940 	if (cxl_slot_is_switched(dev))
1941 		return false;
1942 
1943 	/*
1944 	 * XXX: This gets a little tricky on regular P8 (not POWER8NVL) since
1945 	 * the CAPP can be connected to PHB 0, 1 or 2 on a first come first
1946 	 * served basis, which is racy to check from here. If we need to
1947 	 * support this in future we might need to consider having this
1948 	 * function effectively reserve it ahead of time.
1949 	 *
1950 	 * Currently, the only user of this API is the Mellanox CX4, which is
1951 	 * only supported on P8NVL due to the above mentioned limitation of
1952 	 * CAPP DMA mode and therefore does not need to worry about this. If the
1953 	 * issue with CAPP DMA mode is later worked around on P8 we might need
1954 	 * to revisit this.
1955 	 */
1956 
1957 	return true;
1958 }
1959 EXPORT_SYMBOL_GPL(cxl_slot_is_supported);
1960 
1961 
1962 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
1963 {
1964 	struct cxl *adapter;
1965 	int slice;
1966 	int rc;
1967 
1968 	if (cxl_pci_is_vphb_device(dev)) {
1969 		dev_dbg(&dev->dev, "cxl_init_adapter: Ignoring cxl vphb device\n");
1970 		return -ENODEV;
1971 	}
1972 
1973 	if (cxl_slot_is_switched(dev)) {
1974 		dev_info(&dev->dev, "Ignoring card on incompatible PCI slot\n");
1975 		return -ENODEV;
1976 	}
1977 
1978 	if (cxl_is_power9() && !radix_enabled()) {
1979 		dev_info(&dev->dev, "Only Radix mode supported\n");
1980 		return -ENODEV;
1981 	}
1982 
1983 	if (cxl_verbose)
1984 		dump_cxl_config_space(dev);
1985 
1986 	adapter = cxl_pci_init_adapter(dev);
1987 	if (IS_ERR(adapter)) {
1988 		dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
1989 		return PTR_ERR(adapter);
1990 	}
1991 
1992 	for (slice = 0; slice < adapter->slices; slice++) {
1993 		if ((rc = pci_init_afu(adapter, slice, dev))) {
1994 			dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
1995 			continue;
1996 		}
1997 
1998 		rc = cxl_afu_select_best_mode(adapter->afu[slice]);
1999 		if (rc)
2000 			dev_err(&dev->dev, "AFU %i failed to start: %i\n", slice, rc);
2001 	}
2002 
2003 	if (pnv_pci_on_cxl_phb(dev) && adapter->slices >= 1)
2004 		pnv_cxl_phb_set_peer_afu(dev, adapter->afu[0]);
2005 
2006 	return 0;
2007 }
2008 
2009 static void cxl_remove(struct pci_dev *dev)
2010 {
2011 	struct cxl *adapter = pci_get_drvdata(dev);
2012 	struct cxl_afu *afu;
2013 	int i;
2014 
2015 	/*
2016 	 * Lock to prevent someone grabbing a ref through the adapter list as
2017 	 * we are removing it
2018 	 */
2019 	for (i = 0; i < adapter->slices; i++) {
2020 		afu = adapter->afu[i];
2021 		cxl_pci_remove_afu(afu);
2022 	}
2023 	cxl_pci_remove_adapter(adapter);
2024 }
2025 
2026 static pci_ers_result_t cxl_vphb_error_detected(struct cxl_afu *afu,
2027 						pci_channel_state_t state)
2028 {
2029 	struct pci_dev *afu_dev;
2030 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
2031 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
2032 
2033 	/* There should only be one entry, but go through the list
2034 	 * anyway
2035 	 */
2036 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2037 		if (!afu_dev->driver)
2038 			continue;
2039 
2040 		afu_dev->error_state = state;
2041 
2042 		if (afu_dev->driver->err_handler)
2043 			afu_result = afu_dev->driver->err_handler->error_detected(afu_dev,
2044 										  state);
2045 		/* Disconnect trumps all, NONE trumps NEED_RESET */
2046 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2047 			result = PCI_ERS_RESULT_DISCONNECT;
2048 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
2049 			 (result == PCI_ERS_RESULT_NEED_RESET))
2050 			result = PCI_ERS_RESULT_NONE;
2051 	}
2052 	return result;
2053 }
2054 
2055 static pci_ers_result_t cxl_pci_error_detected(struct pci_dev *pdev,
2056 					       pci_channel_state_t state)
2057 {
2058 	struct cxl *adapter = pci_get_drvdata(pdev);
2059 	struct cxl_afu *afu;
2060 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET, afu_result;
2061 	int i;
2062 
2063 	/* At this point, we could still have an interrupt pending.
2064 	 * Let's try to get them out of the way before they do
2065 	 * anything we don't like.
2066 	 */
2067 	schedule();
2068 
2069 	/* If we're permanently dead, give up. */
2070 	if (state == pci_channel_io_perm_failure) {
2071 		for (i = 0; i < adapter->slices; i++) {
2072 			afu = adapter->afu[i];
2073 			/*
2074 			 * Tell the AFU drivers; but we don't care what they
2075 			 * say, we're going away.
2076 			 */
2077 			if (afu->phb != NULL)
2078 				cxl_vphb_error_detected(afu, state);
2079 		}
2080 		return PCI_ERS_RESULT_DISCONNECT;
2081 	}
2082 
2083 	/* Are we reflashing?
2084 	 *
2085 	 * If we reflash, we could come back as something entirely
2086 	 * different, including a non-CAPI card. As such, by default
2087 	 * we don't participate in the process. We'll be unbound and
2088 	 * the slot re-probed. (TODO: check EEH doesn't blindly rebind
2089 	 * us!)
2090 	 *
2091 	 * However, this isn't the entire story: for reliablity
2092 	 * reasons, we usually want to reflash the FPGA on PERST in
2093 	 * order to get back to a more reliable known-good state.
2094 	 *
2095 	 * This causes us a bit of a problem: if we reflash we can't
2096 	 * trust that we'll come back the same - we could have a new
2097 	 * image and been PERSTed in order to load that
2098 	 * image. However, most of the time we actually *will* come
2099 	 * back the same - for example a regular EEH event.
2100 	 *
2101 	 * Therefore, we allow the user to assert that the image is
2102 	 * indeed the same and that we should continue on into EEH
2103 	 * anyway.
2104 	 */
2105 	if (adapter->perst_loads_image && !adapter->perst_same_image) {
2106 		/* TODO take the PHB out of CXL mode */
2107 		dev_info(&pdev->dev, "reflashing, so opting out of EEH!\n");
2108 		return PCI_ERS_RESULT_NONE;
2109 	}
2110 
2111 	/*
2112 	 * At this point, we want to try to recover.  We'll always
2113 	 * need a complete slot reset: we don't trust any other reset.
2114 	 *
2115 	 * Now, we go through each AFU:
2116 	 *  - We send the driver, if bound, an error_detected callback.
2117 	 *    We expect it to clean up, but it can also tell us to give
2118 	 *    up and permanently detach the card. To simplify things, if
2119 	 *    any bound AFU driver doesn't support EEH, we give up on EEH.
2120 	 *
2121 	 *  - We detach all contexts associated with the AFU. This
2122 	 *    does not free them, but puts them into a CLOSED state
2123 	 *    which causes any the associated files to return useful
2124 	 *    errors to userland. It also unmaps, but does not free,
2125 	 *    any IRQs.
2126 	 *
2127 	 *  - We clean up our side: releasing and unmapping resources we hold
2128 	 *    so we can wire them up again when the hardware comes back up.
2129 	 *
2130 	 * Driver authors should note:
2131 	 *
2132 	 *  - Any contexts you create in your kernel driver (except
2133 	 *    those associated with anonymous file descriptors) are
2134 	 *    your responsibility to free and recreate. Likewise with
2135 	 *    any attached resources.
2136 	 *
2137 	 *  - We will take responsibility for re-initialising the
2138 	 *    device context (the one set up for you in
2139 	 *    cxl_pci_enable_device_hook and accessed through
2140 	 *    cxl_get_context). If you've attached IRQs or other
2141 	 *    resources to it, they remains yours to free.
2142 	 *
2143 	 * You can call the same functions to release resources as you
2144 	 * normally would: we make sure that these functions continue
2145 	 * to work when the hardware is down.
2146 	 *
2147 	 * Two examples:
2148 	 *
2149 	 * 1) If you normally free all your resources at the end of
2150 	 *    each request, or if you use anonymous FDs, your
2151 	 *    error_detected callback can simply set a flag to tell
2152 	 *    your driver not to start any new calls. You can then
2153 	 *    clear the flag in the resume callback.
2154 	 *
2155 	 * 2) If you normally allocate your resources on startup:
2156 	 *     * Set a flag in error_detected as above.
2157 	 *     * Let CXL detach your contexts.
2158 	 *     * In slot_reset, free the old resources and allocate new ones.
2159 	 *     * In resume, clear the flag to allow things to start.
2160 	 */
2161 	for (i = 0; i < adapter->slices; i++) {
2162 		afu = adapter->afu[i];
2163 
2164 		afu_result = cxl_vphb_error_detected(afu, state);
2165 
2166 		cxl_context_detach_all(afu);
2167 		cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
2168 		pci_deconfigure_afu(afu);
2169 
2170 		/* Disconnect trumps all, NONE trumps NEED_RESET */
2171 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2172 			result = PCI_ERS_RESULT_DISCONNECT;
2173 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
2174 			 (result == PCI_ERS_RESULT_NEED_RESET))
2175 			result = PCI_ERS_RESULT_NONE;
2176 	}
2177 
2178 	/* should take the context lock here */
2179 	if (cxl_adapter_context_lock(adapter) != 0)
2180 		dev_warn(&adapter->dev,
2181 			 "Couldn't take context lock with %d active-contexts\n",
2182 			 atomic_read(&adapter->contexts_num));
2183 
2184 	cxl_deconfigure_adapter(adapter);
2185 
2186 	return result;
2187 }
2188 
2189 static pci_ers_result_t cxl_pci_slot_reset(struct pci_dev *pdev)
2190 {
2191 	struct cxl *adapter = pci_get_drvdata(pdev);
2192 	struct cxl_afu *afu;
2193 	struct cxl_context *ctx;
2194 	struct pci_dev *afu_dev;
2195 	pci_ers_result_t afu_result = PCI_ERS_RESULT_RECOVERED;
2196 	pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
2197 	int i;
2198 
2199 	if (cxl_configure_adapter(adapter, pdev))
2200 		goto err;
2201 
2202 	/*
2203 	 * Unlock context activation for the adapter. Ideally this should be
2204 	 * done in cxl_pci_resume but cxlflash module tries to activate the
2205 	 * master context as part of slot_reset callback.
2206 	 */
2207 	cxl_adapter_context_unlock(adapter);
2208 
2209 	for (i = 0; i < adapter->slices; i++) {
2210 		afu = adapter->afu[i];
2211 
2212 		if (pci_configure_afu(afu, adapter, pdev))
2213 			goto err;
2214 
2215 		if (cxl_afu_select_best_mode(afu))
2216 			goto err;
2217 
2218 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2219 			/* Reset the device context.
2220 			 * TODO: make this less disruptive
2221 			 */
2222 			ctx = cxl_get_context(afu_dev);
2223 
2224 			if (ctx && cxl_release_context(ctx))
2225 				goto err;
2226 
2227 			ctx = cxl_dev_context_init(afu_dev);
2228 			if (IS_ERR(ctx))
2229 				goto err;
2230 
2231 			afu_dev->dev.archdata.cxl_ctx = ctx;
2232 
2233 			if (cxl_ops->afu_check_and_enable(afu))
2234 				goto err;
2235 
2236 			afu_dev->error_state = pci_channel_io_normal;
2237 
2238 			/* If there's a driver attached, allow it to
2239 			 * chime in on recovery. Drivers should check
2240 			 * if everything has come back OK, but
2241 			 * shouldn't start new work until we call
2242 			 * their resume function.
2243 			 */
2244 			if (!afu_dev->driver)
2245 				continue;
2246 
2247 			if (afu_dev->driver->err_handler &&
2248 			    afu_dev->driver->err_handler->slot_reset)
2249 				afu_result = afu_dev->driver->err_handler->slot_reset(afu_dev);
2250 
2251 			if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2252 				result = PCI_ERS_RESULT_DISCONNECT;
2253 		}
2254 	}
2255 	return result;
2256 
2257 err:
2258 	/* All the bits that happen in both error_detected and cxl_remove
2259 	 * should be idempotent, so we don't need to worry about leaving a mix
2260 	 * of unconfigured and reconfigured resources.
2261 	 */
2262 	dev_err(&pdev->dev, "EEH recovery failed. Asking to be disconnected.\n");
2263 	return PCI_ERS_RESULT_DISCONNECT;
2264 }
2265 
2266 static void cxl_pci_resume(struct pci_dev *pdev)
2267 {
2268 	struct cxl *adapter = pci_get_drvdata(pdev);
2269 	struct cxl_afu *afu;
2270 	struct pci_dev *afu_dev;
2271 	int i;
2272 
2273 	/* Everything is back now. Drivers should restart work now.
2274 	 * This is not the place to be checking if everything came back up
2275 	 * properly, because there's no return value: do that in slot_reset.
2276 	 */
2277 	for (i = 0; i < adapter->slices; i++) {
2278 		afu = adapter->afu[i];
2279 
2280 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2281 			if (afu_dev->driver && afu_dev->driver->err_handler &&
2282 			    afu_dev->driver->err_handler->resume)
2283 				afu_dev->driver->err_handler->resume(afu_dev);
2284 		}
2285 	}
2286 }
2287 
2288 static const struct pci_error_handlers cxl_err_handler = {
2289 	.error_detected = cxl_pci_error_detected,
2290 	.slot_reset = cxl_pci_slot_reset,
2291 	.resume = cxl_pci_resume,
2292 };
2293 
2294 struct pci_driver cxl_pci_driver = {
2295 	.name = "cxl-pci",
2296 	.id_table = cxl_pci_tbl,
2297 	.probe = cxl_probe,
2298 	.remove = cxl_remove,
2299 	.shutdown = cxl_remove,
2300 	.err_handler = &cxl_err_handler,
2301 };
2302