xref: /openbmc/linux/drivers/misc/cxl/pci.c (revision feac8c8b)
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 int cxl_get_xsl9_dsnctl(u64 capp_unit_id, u64 *reg)
411 {
412 	u64 xsl_dsnctl;
413 
414 	/*
415 	 * CAPI Identifier bits [0:7]
416 	 * bit 61:60 MSI bits --> 0
417 	 * bit 59 TVT selector --> 0
418 	 */
419 
420 	/*
421 	 * Tell XSL where to route data to.
422 	 * The field chipid should match the PHB CAPI_CMPM register
423 	 */
424 	xsl_dsnctl = ((u64)0x2 << (63-7)); /* Bit 57 */
425 	xsl_dsnctl |= (capp_unit_id << (63-15));
426 
427 	/* nMMU_ID Defaults to: b’000001001’*/
428 	xsl_dsnctl |= ((u64)0x09 << (63-28));
429 
430 	if (!(cxl_is_power9_dd1())) {
431 		/*
432 		 * Used to identify CAPI packets which should be sorted into
433 		 * the Non-Blocking queues by the PHB. This field should match
434 		 * the PHB PBL_NBW_CMPM register
435 		 * nbwind=0x03, bits [57:58], must include capi indicator.
436 		 * Not supported on P9 DD1.
437 		 */
438 		xsl_dsnctl |= ((u64)0x03 << (63-47));
439 
440 		/*
441 		 * Upper 16b address bits of ASB_Notify messages sent to the
442 		 * system. Need to match the PHB’s ASN Compare/Mask Register.
443 		 * Not supported on P9 DD1.
444 		 */
445 		xsl_dsnctl |= ((u64)0x04 << (63-55));
446 	}
447 
448 	*reg = xsl_dsnctl;
449 	return 0;
450 }
451 
452 static int init_implementation_adapter_regs_psl9(struct cxl *adapter,
453 						 struct pci_dev *dev)
454 {
455 	u64 xsl_dsnctl, psl_fircntl;
456 	u64 chipid;
457 	u32 phb_index;
458 	u64 capp_unit_id;
459 	int rc;
460 
461 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
462 	if (rc)
463 		return rc;
464 
465 	rc = cxl_get_xsl9_dsnctl(capp_unit_id, &xsl_dsnctl);
466 	if (rc)
467 		return rc;
468 
469 	cxl_p1_write(adapter, CXL_XSL9_DSNCTL, xsl_dsnctl);
470 
471 	/* Set fir_cntl to recommended value for production env */
472 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
473 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
474 	psl_fircntl |= 0x1ULL; /* ce_thresh */
475 	cxl_p1_write(adapter, CXL_PSL9_FIR_CNTL, psl_fircntl);
476 
477 	/* Setup the PSL to transmit packets on the PCIe before the
478 	 * CAPP is enabled
479 	 */
480 	cxl_p1_write(adapter, CXL_PSL9_DSNDCTL, 0x0001001000002A10ULL);
481 
482 	/*
483 	 * A response to an ASB_Notify request is returned by the
484 	 * system as an MMIO write to the address defined in
485 	 * the PSL_TNR_ADDR register.
486 	 * keep the Reset Value: 0x00020000E0000000
487 	 */
488 
489 	/* Enable XSL rty limit */
490 	cxl_p1_write(adapter, CXL_XSL9_DEF, 0x51F8000000000005ULL);
491 
492 	/* Change XSL_INV dummy read threshold */
493 	cxl_p1_write(adapter, CXL_XSL9_INV, 0x0000040007FFC200ULL);
494 
495 	if (phb_index == 3) {
496 		/* disable machines 31-47 and 20-27 for DMA */
497 		cxl_p1_write(adapter, CXL_PSL9_APCDEDTYPE, 0x40000FF3FFFF0000ULL);
498 	}
499 
500 	/* Snoop machines */
501 	cxl_p1_write(adapter, CXL_PSL9_APCDEDALLOC, 0x800F000200000000ULL);
502 
503 	if (cxl_is_power9_dd1()) {
504 		/* Disabling deadlock counter CAR */
505 		cxl_p1_write(adapter, CXL_PSL9_GP_CT, 0x0020000000000001ULL);
506 	} else
507 		cxl_p1_write(adapter, CXL_PSL9_DEBUG, 0x4000000000000000ULL);
508 
509 	return 0;
510 }
511 
512 static int init_implementation_adapter_regs_psl8(struct cxl *adapter, struct pci_dev *dev)
513 {
514 	u64 psl_dsnctl, psl_fircntl;
515 	u64 chipid;
516 	u32 phb_index;
517 	u64 capp_unit_id;
518 	int rc;
519 
520 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
521 	if (rc)
522 		return rc;
523 
524 	psl_dsnctl = 0x0000900000000000ULL; /* pteupd ttype, scdone */
525 	psl_dsnctl |= (0x2ULL << (63-38)); /* MMIO hang pulse: 256 us */
526 	/* Tell PSL where to route data to */
527 	psl_dsnctl |= (chipid << (63-5));
528 	psl_dsnctl |= (capp_unit_id << (63-13));
529 
530 	cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
531 	cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
532 	/* snoop write mask */
533 	cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
534 	/* set fir_cntl to recommended value for production env */
535 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
536 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
537 	psl_fircntl |= 0x1ULL; /* ce_thresh */
538 	cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, psl_fircntl);
539 	/* for debugging with trace arrays */
540 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
541 
542 	return 0;
543 }
544 
545 static int init_implementation_adapter_regs_xsl(struct cxl *adapter, struct pci_dev *dev)
546 {
547 	u64 xsl_dsnctl;
548 	u64 chipid;
549 	u32 phb_index;
550 	u64 capp_unit_id;
551 	int rc;
552 
553 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
554 	if (rc)
555 		return rc;
556 
557 	/* Tell XSL where to route data to */
558 	xsl_dsnctl = 0x0000600000000000ULL | (chipid << (63-5));
559 	xsl_dsnctl |= (capp_unit_id << (63-13));
560 	cxl_p1_write(adapter, CXL_XSL_DSNCTL, xsl_dsnctl);
561 
562 	return 0;
563 }
564 
565 /* PSL & XSL */
566 #define TBSYNC_CAL(n) (((u64)n & 0x7) << (63-3))
567 #define TBSYNC_CNT(n) (((u64)n & 0x7) << (63-6))
568 /* For the PSL this is a multiple for 0 < n <= 7: */
569 #define PSL_2048_250MHZ_CYCLES 1
570 
571 static void write_timebase_ctrl_psl9(struct cxl *adapter)
572 {
573 	cxl_p1_write(adapter, CXL_PSL9_TB_CTLSTAT,
574 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
575 }
576 
577 static void write_timebase_ctrl_psl8(struct cxl *adapter)
578 {
579 	cxl_p1_write(adapter, CXL_PSL_TB_CTLSTAT,
580 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
581 }
582 
583 /* XSL */
584 #define TBSYNC_ENA (1ULL << 63)
585 /* For the XSL this is 2**n * 2000 clocks for 0 < n <= 6: */
586 #define XSL_2000_CLOCKS 1
587 #define XSL_4000_CLOCKS 2
588 #define XSL_8000_CLOCKS 3
589 
590 static void write_timebase_ctrl_xsl(struct cxl *adapter)
591 {
592 	cxl_p1_write(adapter, CXL_XSL_TB_CTLSTAT,
593 		     TBSYNC_ENA |
594 		     TBSYNC_CAL(3) |
595 		     TBSYNC_CNT(XSL_4000_CLOCKS));
596 }
597 
598 static u64 timebase_read_psl9(struct cxl *adapter)
599 {
600 	return cxl_p1_read(adapter, CXL_PSL9_Timebase);
601 }
602 
603 static u64 timebase_read_psl8(struct cxl *adapter)
604 {
605 	return cxl_p1_read(adapter, CXL_PSL_Timebase);
606 }
607 
608 static u64 timebase_read_xsl(struct cxl *adapter)
609 {
610 	return cxl_p1_read(adapter, CXL_XSL_Timebase);
611 }
612 
613 static void cxl_setup_psl_timebase(struct cxl *adapter, struct pci_dev *dev)
614 {
615 	u64 psl_tb;
616 	int delta;
617 	unsigned int retry = 0;
618 	struct device_node *np;
619 
620 	adapter->psl_timebase_synced = false;
621 
622 	if (!(np = pnv_pci_get_phb_node(dev)))
623 		return;
624 
625 	/* Do not fail when CAPP timebase sync is not supported by OPAL */
626 	of_node_get(np);
627 	if (! of_get_property(np, "ibm,capp-timebase-sync", NULL)) {
628 		of_node_put(np);
629 		dev_info(&dev->dev, "PSL timebase inactive: OPAL support missing\n");
630 		return;
631 	}
632 	of_node_put(np);
633 
634 	/*
635 	 * Setup PSL Timebase Control and Status register
636 	 * with the recommended Timebase Sync Count value
637 	 */
638 	adapter->native->sl_ops->write_timebase_ctrl(adapter);
639 
640 	/* Enable PSL Timebase */
641 	cxl_p1_write(adapter, CXL_PSL_Control, 0x0000000000000000);
642 	cxl_p1_write(adapter, CXL_PSL_Control, CXL_PSL_Control_tb);
643 
644 	/* Wait until CORE TB and PSL TB difference <= 16usecs */
645 	do {
646 		msleep(1);
647 		if (retry++ > 5) {
648 			dev_info(&dev->dev, "PSL timebase can't synchronize\n");
649 			return;
650 		}
651 		psl_tb = adapter->native->sl_ops->timebase_read(adapter);
652 		delta = mftb() - psl_tb;
653 		if (delta < 0)
654 			delta = -delta;
655 	} while (tb_to_ns(delta) > 16000);
656 
657 	adapter->psl_timebase_synced = true;
658 	return;
659 }
660 
661 static int init_implementation_afu_regs_psl9(struct cxl_afu *afu)
662 {
663 	return 0;
664 }
665 
666 static int init_implementation_afu_regs_psl8(struct cxl_afu *afu)
667 {
668 	/* read/write masks for this slice */
669 	cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
670 	/* APC read/write masks for this slice */
671 	cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
672 	/* for debugging with trace arrays */
673 	cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
674 	cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
675 
676 	return 0;
677 }
678 
679 int cxl_pci_setup_irq(struct cxl *adapter, unsigned int hwirq,
680 		unsigned int virq)
681 {
682 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
683 
684 	return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
685 }
686 
687 int cxl_update_image_control(struct cxl *adapter)
688 {
689 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
690 	int rc;
691 	int vsec;
692 	u8 image_state;
693 
694 	if (!(vsec = find_cxl_vsec(dev))) {
695 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
696 		return -ENODEV;
697 	}
698 
699 	if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
700 		dev_err(&dev->dev, "failed to read image state: %i\n", rc);
701 		return rc;
702 	}
703 
704 	if (adapter->perst_loads_image)
705 		image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
706 	else
707 		image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
708 
709 	if (adapter->perst_select_user)
710 		image_state |= CXL_VSEC_PERST_SELECT_USER;
711 	else
712 		image_state &= ~CXL_VSEC_PERST_SELECT_USER;
713 
714 	if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
715 		dev_err(&dev->dev, "failed to update image control: %i\n", rc);
716 		return rc;
717 	}
718 
719 	return 0;
720 }
721 
722 int cxl_pci_alloc_one_irq(struct cxl *adapter)
723 {
724 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
725 
726 	return pnv_cxl_alloc_hwirqs(dev, 1);
727 }
728 
729 void cxl_pci_release_one_irq(struct cxl *adapter, int hwirq)
730 {
731 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
732 
733 	return pnv_cxl_release_hwirqs(dev, hwirq, 1);
734 }
735 
736 int cxl_pci_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
737 			struct cxl *adapter, unsigned int num)
738 {
739 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
740 
741 	return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
742 }
743 
744 void cxl_pci_release_irq_ranges(struct cxl_irq_ranges *irqs,
745 				struct cxl *adapter)
746 {
747 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
748 
749 	pnv_cxl_release_hwirq_ranges(irqs, dev);
750 }
751 
752 static int setup_cxl_bars(struct pci_dev *dev)
753 {
754 	/* Safety check in case we get backported to < 3.17 without M64 */
755 	if ((p1_base(dev) < 0x100000000ULL) ||
756 	    (p2_base(dev) < 0x100000000ULL)) {
757 		dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
758 		return -ENODEV;
759 	}
760 
761 	/*
762 	 * BAR 4/5 has a special meaning for CXL and must be programmed with a
763 	 * special value corresponding to the CXL protocol address range.
764 	 * For POWER 8/9 that means bits 48:49 must be set to 10
765 	 */
766 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
767 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
768 
769 	return 0;
770 }
771 
772 #ifdef CONFIG_CXL_BIMODAL
773 
774 struct cxl_switch_work {
775 	struct pci_dev *dev;
776 	struct work_struct work;
777 	int vsec;
778 	int mode;
779 };
780 
781 static void switch_card_to_cxl(struct work_struct *work)
782 {
783 	struct cxl_switch_work *switch_work =
784 		container_of(work, struct cxl_switch_work, work);
785 	struct pci_dev *dev = switch_work->dev;
786 	struct pci_bus *bus = dev->bus;
787 	struct pci_controller *hose = pci_bus_to_host(bus);
788 	struct pci_dev *bridge;
789 	struct pnv_php_slot *php_slot;
790 	unsigned int devfn;
791 	u8 val;
792 	int rc;
793 
794 	dev_info(&bus->dev, "cxl: Preparing for mode switch...\n");
795 	bridge = list_first_entry_or_null(&hose->bus->devices, struct pci_dev,
796 					  bus_list);
797 	if (!bridge) {
798 		dev_WARN(&bus->dev, "cxl: Couldn't find root port!\n");
799 		goto err_dev_put;
800 	}
801 
802 	php_slot = pnv_php_find_slot(pci_device_to_OF_node(bridge));
803 	if (!php_slot) {
804 		dev_err(&bus->dev, "cxl: Failed to find slot hotplug "
805 			           "information. You may need to upgrade "
806 			           "skiboot. Aborting.\n");
807 		goto err_dev_put;
808 	}
809 
810 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, switch_work->vsec, &val);
811 	if (rc) {
812 		dev_err(&bus->dev, "cxl: Failed to read CAPI mode control: %i\n", rc);
813 		goto err_dev_put;
814 	}
815 	devfn = dev->devfn;
816 
817 	/* Release the reference obtained in cxl_check_and_switch_mode() */
818 	pci_dev_put(dev);
819 
820 	dev_dbg(&bus->dev, "cxl: Removing PCI devices from kernel\n");
821 	pci_lock_rescan_remove();
822 	pci_hp_remove_devices(bridge->subordinate);
823 	pci_unlock_rescan_remove();
824 
825 	/* Switch the CXL protocol on the card */
826 	if (switch_work->mode == CXL_BIMODE_CXL) {
827 		dev_info(&bus->dev, "cxl: Switching card to CXL mode\n");
828 		val &= ~CXL_VSEC_PROTOCOL_MASK;
829 		val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
830 		rc = pnv_cxl_enable_phb_kernel_api(hose, true);
831 		if (rc) {
832 			dev_err(&bus->dev, "cxl: Failed to enable kernel API"
833 				           " on real PHB, aborting\n");
834 			goto err_free_work;
835 		}
836 	} else {
837 		dev_WARN(&bus->dev, "cxl: Switching card to PCI mode not supported!\n");
838 		goto err_free_work;
839 	}
840 
841 	rc = CXL_WRITE_VSEC_MODE_CONTROL_BUS(bus, devfn, switch_work->vsec, val);
842 	if (rc) {
843 		dev_err(&bus->dev, "cxl: Failed to configure CXL protocol: %i\n", rc);
844 		goto err_free_work;
845 	}
846 
847 	/*
848 	 * The CAIA spec (v1.1, Section 10.6 Bi-modal Device Support) states
849 	 * we must wait 100ms after this mode switch before touching PCIe config
850 	 * space.
851 	 */
852 	msleep(100);
853 
854 	/*
855 	 * Hot reset to cause the card to come back in cxl mode. A
856 	 * OPAL_RESET_PCI_LINK would be sufficient, but currently lacks support
857 	 * in skiboot, so we use a hot reset instead.
858 	 *
859 	 * We call pci_set_pcie_reset_state() on the bridge, as a CAPI card is
860 	 * guaranteed to sit directly under the root port, and setting the reset
861 	 * state on a device directly under the root port is equivalent to doing
862 	 * it on the root port iself.
863 	 */
864 	dev_info(&bus->dev, "cxl: Configuration write complete, resetting card\n");
865 	pci_set_pcie_reset_state(bridge, pcie_hot_reset);
866 	pci_set_pcie_reset_state(bridge, pcie_deassert_reset);
867 
868 	dev_dbg(&bus->dev, "cxl: Offlining slot\n");
869 	rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_OFFLINE);
870 	if (rc) {
871 		dev_err(&bus->dev, "cxl: OPAL offlining call failed: %i\n", rc);
872 		goto err_free_work;
873 	}
874 
875 	dev_dbg(&bus->dev, "cxl: Onlining and probing slot\n");
876 	rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_ONLINE);
877 	if (rc) {
878 		dev_err(&bus->dev, "cxl: OPAL onlining call failed: %i\n", rc);
879 		goto err_free_work;
880 	}
881 
882 	pci_lock_rescan_remove();
883 	pci_hp_add_devices(bridge->subordinate);
884 	pci_unlock_rescan_remove();
885 
886 	dev_info(&bus->dev, "cxl: CAPI mode switch completed\n");
887 	kfree(switch_work);
888 	return;
889 
890 err_dev_put:
891 	/* Release the reference obtained in cxl_check_and_switch_mode() */
892 	pci_dev_put(dev);
893 err_free_work:
894 	kfree(switch_work);
895 }
896 
897 int cxl_check_and_switch_mode(struct pci_dev *dev, int mode, int vsec)
898 {
899 	struct cxl_switch_work *work;
900 	u8 val;
901 	int rc;
902 
903 	if (!cpu_has_feature(CPU_FTR_HVMODE))
904 		return -ENODEV;
905 
906 	if (!vsec) {
907 		vsec = find_cxl_vsec(dev);
908 		if (!vsec) {
909 			dev_info(&dev->dev, "CXL VSEC not found\n");
910 			return -ENODEV;
911 		}
912 	}
913 
914 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
915 	if (rc) {
916 		dev_err(&dev->dev, "Failed to read current mode control: %i", rc);
917 		return rc;
918 	}
919 
920 	if (mode == CXL_BIMODE_PCI) {
921 		if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
922 			dev_info(&dev->dev, "Card is already in PCI mode\n");
923 			return 0;
924 		}
925 		/*
926 		 * TODO: Before it's safe to switch the card back to PCI mode
927 		 * we need to disable the CAPP and make sure any cachelines the
928 		 * card holds have been flushed out. Needs skiboot support.
929 		 */
930 		dev_WARN(&dev->dev, "CXL mode switch to PCI unsupported!\n");
931 		return -EIO;
932 	}
933 
934 	if (val & CXL_VSEC_PROTOCOL_ENABLE) {
935 		dev_info(&dev->dev, "Card is already in CXL mode\n");
936 		return 0;
937 	}
938 
939 	dev_info(&dev->dev, "Card is in PCI mode, scheduling kernel thread "
940 			    "to switch to CXL mode\n");
941 
942 	work = kmalloc(sizeof(struct cxl_switch_work), GFP_KERNEL);
943 	if (!work)
944 		return -ENOMEM;
945 
946 	pci_dev_get(dev);
947 	work->dev = dev;
948 	work->vsec = vsec;
949 	work->mode = mode;
950 	INIT_WORK(&work->work, switch_card_to_cxl);
951 
952 	schedule_work(&work->work);
953 
954 	/*
955 	 * We return a failure now to abort the driver init. Once the
956 	 * link has been cycled and the card is in cxl mode we will
957 	 * come back (possibly using the generic cxl driver), but
958 	 * return success as the card should then be in cxl mode.
959 	 *
960 	 * TODO: What if the card comes back in PCI mode even after
961 	 *       the switch?  Don't want to spin endlessly.
962 	 */
963 	return -EBUSY;
964 }
965 EXPORT_SYMBOL_GPL(cxl_check_and_switch_mode);
966 
967 #endif /* CONFIG_CXL_BIMODAL */
968 
969 static int setup_cxl_protocol_area(struct pci_dev *dev)
970 {
971 	u8 val;
972 	int rc;
973 	int vsec = find_cxl_vsec(dev);
974 
975 	if (!vsec) {
976 		dev_info(&dev->dev, "CXL VSEC not found\n");
977 		return -ENODEV;
978 	}
979 
980 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
981 	if (rc) {
982 		dev_err(&dev->dev, "Failed to read current mode control: %i\n", rc);
983 		return rc;
984 	}
985 
986 	if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
987 		dev_err(&dev->dev, "Card not in CAPI mode!\n");
988 		return -EIO;
989 	}
990 
991 	if ((val & CXL_VSEC_PROTOCOL_MASK) != CXL_VSEC_PROTOCOL_256TB) {
992 		val &= ~CXL_VSEC_PROTOCOL_MASK;
993 		val |= CXL_VSEC_PROTOCOL_256TB;
994 		rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val);
995 		if (rc) {
996 			dev_err(&dev->dev, "Failed to set CXL protocol area: %i\n", rc);
997 			return rc;
998 		}
999 	}
1000 
1001 	return 0;
1002 }
1003 
1004 static int pci_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1005 {
1006 	u64 p1n_base, p2n_base, afu_desc;
1007 	const u64 p1n_size = 0x100;
1008 	const u64 p2n_size = 0x1000;
1009 
1010 	p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
1011 	p2n_base = p2_base(dev) + (afu->slice * p2n_size);
1012 	afu->psn_phys = p2_base(dev) + (adapter->native->ps_off + (afu->slice * adapter->ps_size));
1013 	afu_desc = p2_base(dev) + adapter->native->afu_desc_off + (afu->slice * adapter->native->afu_desc_size);
1014 
1015 	if (!(afu->native->p1n_mmio = ioremap(p1n_base, p1n_size)))
1016 		goto err;
1017 	if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
1018 		goto err1;
1019 	if (afu_desc) {
1020 		if (!(afu->native->afu_desc_mmio = ioremap(afu_desc, adapter->native->afu_desc_size)))
1021 			goto err2;
1022 	}
1023 
1024 	return 0;
1025 err2:
1026 	iounmap(afu->p2n_mmio);
1027 err1:
1028 	iounmap(afu->native->p1n_mmio);
1029 err:
1030 	dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
1031 	return -ENOMEM;
1032 }
1033 
1034 static void pci_unmap_slice_regs(struct cxl_afu *afu)
1035 {
1036 	if (afu->p2n_mmio) {
1037 		iounmap(afu->p2n_mmio);
1038 		afu->p2n_mmio = NULL;
1039 	}
1040 	if (afu->native->p1n_mmio) {
1041 		iounmap(afu->native->p1n_mmio);
1042 		afu->native->p1n_mmio = NULL;
1043 	}
1044 	if (afu->native->afu_desc_mmio) {
1045 		iounmap(afu->native->afu_desc_mmio);
1046 		afu->native->afu_desc_mmio = NULL;
1047 	}
1048 }
1049 
1050 void cxl_pci_release_afu(struct device *dev)
1051 {
1052 	struct cxl_afu *afu = to_cxl_afu(dev);
1053 
1054 	pr_devel("%s\n", __func__);
1055 
1056 	idr_destroy(&afu->contexts_idr);
1057 	cxl_release_spa(afu);
1058 
1059 	kfree(afu->native);
1060 	kfree(afu);
1061 }
1062 
1063 /* Expects AFU struct to have recently been zeroed out */
1064 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
1065 {
1066 	u64 val;
1067 
1068 	val = AFUD_READ_INFO(afu);
1069 	afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
1070 	afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
1071 	afu->crs_num = AFUD_NUM_CRS(val);
1072 
1073 	if (AFUD_AFU_DIRECTED(val))
1074 		afu->modes_supported |= CXL_MODE_DIRECTED;
1075 	if (AFUD_DEDICATED_PROCESS(val))
1076 		afu->modes_supported |= CXL_MODE_DEDICATED;
1077 	if (AFUD_TIME_SLICED(val))
1078 		afu->modes_supported |= CXL_MODE_TIME_SLICED;
1079 
1080 	val = AFUD_READ_PPPSA(afu);
1081 	afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
1082 	afu->psa = AFUD_PPPSA_PSA(val);
1083 	if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
1084 		afu->native->pp_offset = AFUD_READ_PPPSA_OFF(afu);
1085 
1086 	val = AFUD_READ_CR(afu);
1087 	afu->crs_len = AFUD_CR_LEN(val) * 256;
1088 	afu->crs_offset = AFUD_READ_CR_OFF(afu);
1089 
1090 
1091 	/* eb_len is in multiple of 4K */
1092 	afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
1093 	afu->eb_offset = AFUD_READ_EB_OFF(afu);
1094 
1095 	/* eb_off is 4K aligned so lower 12 bits are always zero */
1096 	if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
1097 		dev_warn(&afu->dev,
1098 			 "Invalid AFU error buffer offset %Lx\n",
1099 			 afu->eb_offset);
1100 		dev_info(&afu->dev,
1101 			 "Ignoring AFU error buffer in the descriptor\n");
1102 		/* indicate that no afu buffer exists */
1103 		afu->eb_len = 0;
1104 	}
1105 
1106 	return 0;
1107 }
1108 
1109 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
1110 {
1111 	int i, rc;
1112 	u32 val;
1113 
1114 	if (afu->psa && afu->adapter->ps_size <
1115 			(afu->native->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
1116 		dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
1117 		return -ENODEV;
1118 	}
1119 
1120 	if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
1121 		dev_warn(&afu->dev, "AFU uses pp_size(%#016llx) < PAGE_SIZE per-process PSA!\n", afu->pp_size);
1122 
1123 	for (i = 0; i < afu->crs_num; i++) {
1124 		rc = cxl_ops->afu_cr_read32(afu, i, 0, &val);
1125 		if (rc || val == 0) {
1126 			dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
1127 			return -EINVAL;
1128 		}
1129 	}
1130 
1131 	if ((afu->modes_supported & ~CXL_MODE_DEDICATED) && afu->max_procs_virtualised == 0) {
1132 		/*
1133 		 * We could also check this for the dedicated process model
1134 		 * since the architecture indicates it should be set to 1, but
1135 		 * in that case we ignore the value and I'd rather not risk
1136 		 * breaking any existing dedicated process AFUs that left it as
1137 		 * 0 (not that I'm aware of any). It is clearly an error for an
1138 		 * AFU directed AFU to set this to 0, and would have previously
1139 		 * triggered a bug resulting in the maximum not being enforced
1140 		 * at all since idr_alloc treats 0 as no maximum.
1141 		 */
1142 		dev_err(&afu->dev, "AFU does not support any processes\n");
1143 		return -EINVAL;
1144 	}
1145 
1146 	return 0;
1147 }
1148 
1149 static int sanitise_afu_regs_psl9(struct cxl_afu *afu)
1150 {
1151 	u64 reg;
1152 
1153 	/*
1154 	 * Clear out any regs that contain either an IVTE or address or may be
1155 	 * waiting on an acknowledgment to try to be a bit safer as we bring
1156 	 * it online
1157 	 */
1158 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
1159 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
1160 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
1161 		if (cxl_ops->afu_reset(afu))
1162 			return -EIO;
1163 		if (cxl_afu_disable(afu))
1164 			return -EIO;
1165 		if (cxl_psl_purge(afu))
1166 			return -EIO;
1167 	}
1168 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
1169 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1170 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1171 	if (reg) {
1172 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1173 		if (reg & CXL_PSL9_DSISR_An_TF)
1174 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1175 		else
1176 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1177 	}
1178 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1179 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1180 		if (reg) {
1181 			if (reg & ~0x000000007fffffff)
1182 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1183 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1184 		}
1185 	}
1186 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1187 	if (reg) {
1188 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1189 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1190 	}
1191 
1192 	return 0;
1193 }
1194 
1195 static int sanitise_afu_regs_psl8(struct cxl_afu *afu)
1196 {
1197 	u64 reg;
1198 
1199 	/*
1200 	 * Clear out any regs that contain either an IVTE or address or may be
1201 	 * waiting on an acknowledgement to try to be a bit safer as we bring
1202 	 * it online
1203 	 */
1204 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
1205 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
1206 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
1207 		if (cxl_ops->afu_reset(afu))
1208 			return -EIO;
1209 		if (cxl_afu_disable(afu))
1210 			return -EIO;
1211 		if (cxl_psl_purge(afu))
1212 			return -EIO;
1213 	}
1214 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
1215 	cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
1216 	cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
1217 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1218 	cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
1219 	cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
1220 	cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
1221 	cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
1222 	cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
1223 	cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
1224 	cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
1225 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1226 	if (reg) {
1227 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1228 		if (reg & CXL_PSL_DSISR_TRANS)
1229 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1230 		else
1231 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1232 	}
1233 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1234 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1235 		if (reg) {
1236 			if (reg & ~0xffff)
1237 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1238 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1239 		}
1240 	}
1241 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1242 	if (reg) {
1243 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1244 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1245 	}
1246 
1247 	return 0;
1248 }
1249 
1250 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
1251 /*
1252  * afu_eb_read:
1253  * Called from sysfs and reads the afu error info buffer. The h/w only supports
1254  * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
1255  * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
1256  */
1257 ssize_t cxl_pci_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
1258 				loff_t off, size_t count)
1259 {
1260 	loff_t aligned_start, aligned_end;
1261 	size_t aligned_length;
1262 	void *tbuf;
1263 	const void __iomem *ebuf = afu->native->afu_desc_mmio + afu->eb_offset;
1264 
1265 	if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
1266 		return 0;
1267 
1268 	/* calculate aligned read window */
1269 	count = min((size_t)(afu->eb_len - off), count);
1270 	aligned_start = round_down(off, 8);
1271 	aligned_end = round_up(off + count, 8);
1272 	aligned_length = aligned_end - aligned_start;
1273 
1274 	/* max we can copy in one read is PAGE_SIZE */
1275 	if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
1276 		aligned_length = ERR_BUFF_MAX_COPY_SIZE;
1277 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
1278 	}
1279 
1280 	/* use bounce buffer for copy */
1281 	tbuf = (void *)__get_free_page(GFP_KERNEL);
1282 	if (!tbuf)
1283 		return -ENOMEM;
1284 
1285 	/* perform aligned read from the mmio region */
1286 	memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
1287 	memcpy(buf, tbuf + (off & 0x7), count);
1288 
1289 	free_page((unsigned long)tbuf);
1290 
1291 	return count;
1292 }
1293 
1294 static int pci_configure_afu(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1295 {
1296 	int rc;
1297 
1298 	if ((rc = pci_map_slice_regs(afu, adapter, dev)))
1299 		return rc;
1300 
1301 	if (adapter->native->sl_ops->sanitise_afu_regs) {
1302 		rc = adapter->native->sl_ops->sanitise_afu_regs(afu);
1303 		if (rc)
1304 			goto err1;
1305 	}
1306 
1307 	/* We need to reset the AFU before we can read the AFU descriptor */
1308 	if ((rc = cxl_ops->afu_reset(afu)))
1309 		goto err1;
1310 
1311 	if (cxl_verbose)
1312 		dump_afu_descriptor(afu);
1313 
1314 	if ((rc = cxl_read_afu_descriptor(afu)))
1315 		goto err1;
1316 
1317 	if ((rc = cxl_afu_descriptor_looks_ok(afu)))
1318 		goto err1;
1319 
1320 	if (adapter->native->sl_ops->afu_regs_init)
1321 		if ((rc = adapter->native->sl_ops->afu_regs_init(afu)))
1322 			goto err1;
1323 
1324 	if (adapter->native->sl_ops->register_serr_irq)
1325 		if ((rc = adapter->native->sl_ops->register_serr_irq(afu)))
1326 			goto err1;
1327 
1328 	if ((rc = cxl_native_register_psl_irq(afu)))
1329 		goto err2;
1330 
1331 	atomic_set(&afu->configured_state, 0);
1332 	return 0;
1333 
1334 err2:
1335 	if (adapter->native->sl_ops->release_serr_irq)
1336 		adapter->native->sl_ops->release_serr_irq(afu);
1337 err1:
1338 	pci_unmap_slice_regs(afu);
1339 	return rc;
1340 }
1341 
1342 static void pci_deconfigure_afu(struct cxl_afu *afu)
1343 {
1344 	/*
1345 	 * It's okay to deconfigure when AFU is already locked, otherwise wait
1346 	 * until there are no readers
1347 	 */
1348 	if (atomic_read(&afu->configured_state) != -1) {
1349 		while (atomic_cmpxchg(&afu->configured_state, 0, -1) != -1)
1350 			schedule();
1351 	}
1352 	cxl_native_release_psl_irq(afu);
1353 	if (afu->adapter->native->sl_ops->release_serr_irq)
1354 		afu->adapter->native->sl_ops->release_serr_irq(afu);
1355 	pci_unmap_slice_regs(afu);
1356 }
1357 
1358 static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
1359 {
1360 	struct cxl_afu *afu;
1361 	int rc = -ENOMEM;
1362 
1363 	afu = cxl_alloc_afu(adapter, slice);
1364 	if (!afu)
1365 		return -ENOMEM;
1366 
1367 	afu->native = kzalloc(sizeof(struct cxl_afu_native), GFP_KERNEL);
1368 	if (!afu->native)
1369 		goto err_free_afu;
1370 
1371 	mutex_init(&afu->native->spa_mutex);
1372 
1373 	rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice);
1374 	if (rc)
1375 		goto err_free_native;
1376 
1377 	rc = pci_configure_afu(afu, adapter, dev);
1378 	if (rc)
1379 		goto err_free_native;
1380 
1381 	/* Don't care if this fails */
1382 	cxl_debugfs_afu_add(afu);
1383 
1384 	/*
1385 	 * After we call this function we must not free the afu directly, even
1386 	 * if it returns an error!
1387 	 */
1388 	if ((rc = cxl_register_afu(afu)))
1389 		goto err_put1;
1390 
1391 	if ((rc = cxl_sysfs_afu_add(afu)))
1392 		goto err_put1;
1393 
1394 	adapter->afu[afu->slice] = afu;
1395 
1396 	if ((rc = cxl_pci_vphb_add(afu)))
1397 		dev_info(&afu->dev, "Can't register vPHB\n");
1398 
1399 	return 0;
1400 
1401 err_put1:
1402 	pci_deconfigure_afu(afu);
1403 	cxl_debugfs_afu_remove(afu);
1404 	device_unregister(&afu->dev);
1405 	return rc;
1406 
1407 err_free_native:
1408 	kfree(afu->native);
1409 err_free_afu:
1410 	kfree(afu);
1411 	return rc;
1412 
1413 }
1414 
1415 static void cxl_pci_remove_afu(struct cxl_afu *afu)
1416 {
1417 	pr_devel("%s\n", __func__);
1418 
1419 	if (!afu)
1420 		return;
1421 
1422 	cxl_pci_vphb_remove(afu);
1423 	cxl_sysfs_afu_remove(afu);
1424 	cxl_debugfs_afu_remove(afu);
1425 
1426 	spin_lock(&afu->adapter->afu_list_lock);
1427 	afu->adapter->afu[afu->slice] = NULL;
1428 	spin_unlock(&afu->adapter->afu_list_lock);
1429 
1430 	cxl_context_detach_all(afu);
1431 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1432 
1433 	pci_deconfigure_afu(afu);
1434 	device_unregister(&afu->dev);
1435 }
1436 
1437 int cxl_pci_reset(struct cxl *adapter)
1438 {
1439 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1440 	int rc;
1441 
1442 	if (adapter->perst_same_image) {
1443 		dev_warn(&dev->dev,
1444 			 "cxl: refusing to reset/reflash when perst_reloads_same_image is set.\n");
1445 		return -EINVAL;
1446 	}
1447 
1448 	dev_info(&dev->dev, "CXL reset\n");
1449 
1450 	/*
1451 	 * The adapter is about to be reset, so ignore errors.
1452 	 * Not supported on P9 DD1
1453 	 */
1454 	if ((cxl_is_power8()) || (!(cxl_is_power9_dd1())))
1455 		cxl_data_cache_flush(adapter);
1456 
1457 	/* pcie_warm_reset requests a fundamental pci reset which includes a
1458 	 * PERST assert/deassert.  PERST triggers a loading of the image
1459 	 * if "user" or "factory" is selected in sysfs */
1460 	if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
1461 		dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
1462 		return rc;
1463 	}
1464 
1465 	return rc;
1466 }
1467 
1468 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
1469 {
1470 	if (pci_request_region(dev, 2, "priv 2 regs"))
1471 		goto err1;
1472 	if (pci_request_region(dev, 0, "priv 1 regs"))
1473 		goto err2;
1474 
1475 	pr_devel("cxl_map_adapter_regs: p1: %#016llx %#llx, p2: %#016llx %#llx",
1476 			p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
1477 
1478 	if (!(adapter->native->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
1479 		goto err3;
1480 
1481 	if (!(adapter->native->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
1482 		goto err4;
1483 
1484 	return 0;
1485 
1486 err4:
1487 	iounmap(adapter->native->p1_mmio);
1488 	adapter->native->p1_mmio = NULL;
1489 err3:
1490 	pci_release_region(dev, 0);
1491 err2:
1492 	pci_release_region(dev, 2);
1493 err1:
1494 	return -ENOMEM;
1495 }
1496 
1497 static void cxl_unmap_adapter_regs(struct cxl *adapter)
1498 {
1499 	if (adapter->native->p1_mmio) {
1500 		iounmap(adapter->native->p1_mmio);
1501 		adapter->native->p1_mmio = NULL;
1502 		pci_release_region(to_pci_dev(adapter->dev.parent), 2);
1503 	}
1504 	if (adapter->native->p2_mmio) {
1505 		iounmap(adapter->native->p2_mmio);
1506 		adapter->native->p2_mmio = NULL;
1507 		pci_release_region(to_pci_dev(adapter->dev.parent), 0);
1508 	}
1509 }
1510 
1511 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
1512 {
1513 	int vsec;
1514 	u32 afu_desc_off, afu_desc_size;
1515 	u32 ps_off, ps_size;
1516 	u16 vseclen;
1517 	u8 image_state;
1518 
1519 	if (!(vsec = find_cxl_vsec(dev))) {
1520 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
1521 		return -ENODEV;
1522 	}
1523 
1524 	CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
1525 	if (vseclen < CXL_VSEC_MIN_SIZE) {
1526 		dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
1527 		return -EINVAL;
1528 	}
1529 
1530 	CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
1531 	CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
1532 	CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
1533 	CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
1534 	CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
1535 	CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
1536 	adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1537 	adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1538 	adapter->perst_loads_image = !!(image_state & CXL_VSEC_PERST_LOADS_IMAGE);
1539 
1540 	CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
1541 	CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
1542 	CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
1543 	CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
1544 	CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
1545 
1546 	/* Convert everything to bytes, because there is NO WAY I'd look at the
1547 	 * code a month later and forget what units these are in ;-) */
1548 	adapter->native->ps_off = ps_off * 64 * 1024;
1549 	adapter->ps_size = ps_size * 64 * 1024;
1550 	adapter->native->afu_desc_off = afu_desc_off * 64 * 1024;
1551 	adapter->native->afu_desc_size = afu_desc_size * 64 * 1024;
1552 
1553 	/* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
1554 	adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
1555 
1556 	return 0;
1557 }
1558 
1559 /*
1560  * Workaround a PCIe Host Bridge defect on some cards, that can cause
1561  * malformed Transaction Layer Packet (TLP) errors to be erroneously
1562  * reported. Mask this error in the Uncorrectable Error Mask Register.
1563  *
1564  * The upper nibble of the PSL revision is used to distinguish between
1565  * different cards. The affected ones have it set to 0.
1566  */
1567 static void cxl_fixup_malformed_tlp(struct cxl *adapter, struct pci_dev *dev)
1568 {
1569 	int aer;
1570 	u32 data;
1571 
1572 	if (adapter->psl_rev & 0xf000)
1573 		return;
1574 	if (!(aer = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR)))
1575 		return;
1576 	pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &data);
1577 	if (data & PCI_ERR_UNC_MALF_TLP)
1578 		if (data & PCI_ERR_UNC_INTN)
1579 			return;
1580 	data |= PCI_ERR_UNC_MALF_TLP;
1581 	data |= PCI_ERR_UNC_INTN;
1582 	pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, data);
1583 }
1584 
1585 static bool cxl_compatible_caia_version(struct cxl *adapter)
1586 {
1587 	if (cxl_is_power8() && (adapter->caia_major == 1))
1588 		return true;
1589 
1590 	if (cxl_is_power9() && (adapter->caia_major == 2))
1591 		return true;
1592 
1593 	return false;
1594 }
1595 
1596 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
1597 {
1598 	if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
1599 		return -EBUSY;
1600 
1601 	if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
1602 		dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
1603 		return -EINVAL;
1604 	}
1605 
1606 	if (!cxl_compatible_caia_version(adapter)) {
1607 		dev_info(&dev->dev, "Ignoring card. PSL type is not supported (caia version: %d)\n",
1608 			 adapter->caia_major);
1609 		return -ENODEV;
1610 	}
1611 
1612 	if (!adapter->slices) {
1613 		/* Once we support dynamic reprogramming we can use the card if
1614 		 * it supports loadable AFUs */
1615 		dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
1616 		return -EINVAL;
1617 	}
1618 
1619 	if (!adapter->native->afu_desc_off || !adapter->native->afu_desc_size) {
1620 		dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
1621 		return -EINVAL;
1622 	}
1623 
1624 	if (adapter->ps_size > p2_size(dev) - adapter->native->ps_off) {
1625 		dev_err(&dev->dev, "ABORTING: Problem state size larger than "
1626 				   "available in BAR2: 0x%llx > 0x%llx\n",
1627 			 adapter->ps_size, p2_size(dev) - adapter->native->ps_off);
1628 		return -EINVAL;
1629 	}
1630 
1631 	return 0;
1632 }
1633 
1634 ssize_t cxl_pci_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1635 {
1636 	return pci_read_vpd(to_pci_dev(adapter->dev.parent), 0, len, buf);
1637 }
1638 
1639 static void cxl_release_adapter(struct device *dev)
1640 {
1641 	struct cxl *adapter = to_cxl_adapter(dev);
1642 
1643 	pr_devel("cxl_release_adapter\n");
1644 
1645 	cxl_remove_adapter_nr(adapter);
1646 
1647 	kfree(adapter->native);
1648 	kfree(adapter);
1649 }
1650 
1651 #define CXL_PSL_ErrIVTE_tberror (0x1ull << (63-31))
1652 
1653 static int sanitise_adapter_regs(struct cxl *adapter)
1654 {
1655 	int rc = 0;
1656 
1657 	/* Clear PSL tberror bit by writing 1 to it */
1658 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, CXL_PSL_ErrIVTE_tberror);
1659 
1660 	if (adapter->native->sl_ops->invalidate_all) {
1661 		/* do not invalidate ERAT entries when not reloading on PERST */
1662 		if (cxl_is_power9() && (adapter->perst_loads_image))
1663 			return 0;
1664 		rc = adapter->native->sl_ops->invalidate_all(adapter);
1665 	}
1666 
1667 	return rc;
1668 }
1669 
1670 /* This should contain *only* operations that can safely be done in
1671  * both creation and recovery.
1672  */
1673 static int cxl_configure_adapter(struct cxl *adapter, struct pci_dev *dev)
1674 {
1675 	int rc;
1676 
1677 	adapter->dev.parent = &dev->dev;
1678 	adapter->dev.release = cxl_release_adapter;
1679 	pci_set_drvdata(dev, adapter);
1680 
1681 	rc = pci_enable_device(dev);
1682 	if (rc) {
1683 		dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1684 		return rc;
1685 	}
1686 
1687 	if ((rc = cxl_read_vsec(adapter, dev)))
1688 		return rc;
1689 
1690 	if ((rc = cxl_vsec_looks_ok(adapter, dev)))
1691 	        return rc;
1692 
1693 	cxl_fixup_malformed_tlp(adapter, dev);
1694 
1695 	if ((rc = setup_cxl_bars(dev)))
1696 		return rc;
1697 
1698 	if ((rc = setup_cxl_protocol_area(dev)))
1699 		return rc;
1700 
1701 	if ((rc = cxl_update_image_control(adapter)))
1702 		return rc;
1703 
1704 	if ((rc = cxl_map_adapter_regs(adapter, dev)))
1705 		return rc;
1706 
1707 	if ((rc = sanitise_adapter_regs(adapter)))
1708 		goto err;
1709 
1710 	if ((rc = adapter->native->sl_ops->adapter_regs_init(adapter, dev)))
1711 		goto err;
1712 
1713 	/* Required for devices using CAPP DMA mode, harmless for others */
1714 	pci_set_master(dev);
1715 
1716 	if ((rc = pnv_phb_to_cxl_mode(dev, adapter->native->sl_ops->capi_mode)))
1717 		goto err;
1718 
1719 	/* If recovery happened, the last step is to turn on snooping.
1720 	 * In the non-recovery case this has no effect */
1721 	if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON)))
1722 		goto err;
1723 
1724 	/* Ignore error, adapter init is not dependant on timebase sync */
1725 	cxl_setup_psl_timebase(adapter, dev);
1726 
1727 	if ((rc = cxl_native_register_psl_err_irq(adapter)))
1728 		goto err;
1729 
1730 	return 0;
1731 
1732 err:
1733 	cxl_unmap_adapter_regs(adapter);
1734 	return rc;
1735 
1736 }
1737 
1738 static void cxl_deconfigure_adapter(struct cxl *adapter)
1739 {
1740 	struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1741 
1742 	cxl_native_release_psl_err_irq(adapter);
1743 	cxl_unmap_adapter_regs(adapter);
1744 
1745 	pci_disable_device(pdev);
1746 }
1747 
1748 static void cxl_stop_trace_psl9(struct cxl *adapter)
1749 {
1750 	int traceid;
1751 	u64 trace_state, trace_mask;
1752 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1753 
1754 	/* read each tracearray state and issue mmio to stop them is needed */
1755 	for (traceid = 0; traceid <= CXL_PSL9_TRACEID_MAX; ++traceid) {
1756 		trace_state = cxl_p1_read(adapter, CXL_PSL9_CTCCFG);
1757 		trace_mask = (0x3ULL << (62 - traceid * 2));
1758 		trace_state = (trace_state & trace_mask) >> (62 - traceid * 2);
1759 		dev_dbg(&dev->dev, "cxl: Traceid-%d trace_state=0x%0llX\n",
1760 			traceid, trace_state);
1761 
1762 		/* issue mmio if the trace array isn't in FIN state */
1763 		if (trace_state != CXL_PSL9_TRACESTATE_FIN)
1764 			cxl_p1_write(adapter, CXL_PSL9_TRACECFG,
1765 				     0x8400000000000000ULL | traceid);
1766 	}
1767 }
1768 
1769 static void cxl_stop_trace_psl8(struct cxl *adapter)
1770 {
1771 	int slice;
1772 
1773 	/* Stop the trace */
1774 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x8000000000000017LL);
1775 
1776 	/* Stop the slice traces */
1777 	spin_lock(&adapter->afu_list_lock);
1778 	for (slice = 0; slice < adapter->slices; slice++) {
1779 		if (adapter->afu[slice])
1780 			cxl_p1n_write(adapter->afu[slice], CXL_PSL_SLICE_TRACE,
1781 				      0x8000000000000000LL);
1782 	}
1783 	spin_unlock(&adapter->afu_list_lock);
1784 }
1785 
1786 static const struct cxl_service_layer_ops psl9_ops = {
1787 	.adapter_regs_init = init_implementation_adapter_regs_psl9,
1788 	.invalidate_all = cxl_invalidate_all_psl9,
1789 	.afu_regs_init = init_implementation_afu_regs_psl9,
1790 	.sanitise_afu_regs = sanitise_afu_regs_psl9,
1791 	.register_serr_irq = cxl_native_register_serr_irq,
1792 	.release_serr_irq = cxl_native_release_serr_irq,
1793 	.handle_interrupt = cxl_irq_psl9,
1794 	.fail_irq = cxl_fail_irq_psl,
1795 	.activate_dedicated_process = cxl_activate_dedicated_process_psl9,
1796 	.attach_afu_directed = cxl_attach_afu_directed_psl9,
1797 	.attach_dedicated_process = cxl_attach_dedicated_process_psl9,
1798 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl9,
1799 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl9,
1800 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl9,
1801 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl9,
1802 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl9,
1803 	.debugfs_stop_trace = cxl_stop_trace_psl9,
1804 	.write_timebase_ctrl = write_timebase_ctrl_psl9,
1805 	.timebase_read = timebase_read_psl9,
1806 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1807 	.needs_reset_before_disable = true,
1808 };
1809 
1810 static const struct cxl_service_layer_ops psl8_ops = {
1811 	.adapter_regs_init = init_implementation_adapter_regs_psl8,
1812 	.invalidate_all = cxl_invalidate_all_psl8,
1813 	.afu_regs_init = init_implementation_afu_regs_psl8,
1814 	.sanitise_afu_regs = sanitise_afu_regs_psl8,
1815 	.register_serr_irq = cxl_native_register_serr_irq,
1816 	.release_serr_irq = cxl_native_release_serr_irq,
1817 	.handle_interrupt = cxl_irq_psl8,
1818 	.fail_irq = cxl_fail_irq_psl,
1819 	.activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1820 	.attach_afu_directed = cxl_attach_afu_directed_psl8,
1821 	.attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1822 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1823 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl8,
1824 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl8,
1825 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl8,
1826 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl8,
1827 	.debugfs_stop_trace = cxl_stop_trace_psl8,
1828 	.write_timebase_ctrl = write_timebase_ctrl_psl8,
1829 	.timebase_read = timebase_read_psl8,
1830 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1831 	.needs_reset_before_disable = true,
1832 };
1833 
1834 static const struct cxl_service_layer_ops xsl_ops = {
1835 	.adapter_regs_init = init_implementation_adapter_regs_xsl,
1836 	.invalidate_all = cxl_invalidate_all_psl8,
1837 	.sanitise_afu_regs = sanitise_afu_regs_psl8,
1838 	.handle_interrupt = cxl_irq_psl8,
1839 	.fail_irq = cxl_fail_irq_psl,
1840 	.activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1841 	.attach_afu_directed = cxl_attach_afu_directed_psl8,
1842 	.attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1843 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1844 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_xsl,
1845 	.write_timebase_ctrl = write_timebase_ctrl_xsl,
1846 	.timebase_read = timebase_read_xsl,
1847 	.capi_mode = OPAL_PHB_CAPI_MODE_DMA,
1848 };
1849 
1850 static void set_sl_ops(struct cxl *adapter, struct pci_dev *dev)
1851 {
1852 	if (dev->vendor == PCI_VENDOR_ID_MELLANOX && dev->device == 0x1013) {
1853 		/* Mellanox CX-4 */
1854 		dev_info(&dev->dev, "Device uses an XSL\n");
1855 		adapter->native->sl_ops = &xsl_ops;
1856 		adapter->min_pe = 1; /* Workaround for CX-4 hardware bug */
1857 	} else {
1858 		if (cxl_is_power8()) {
1859 			dev_info(&dev->dev, "Device uses a PSL8\n");
1860 			adapter->native->sl_ops = &psl8_ops;
1861 		} else {
1862 			dev_info(&dev->dev, "Device uses a PSL9\n");
1863 			adapter->native->sl_ops = &psl9_ops;
1864 		}
1865 	}
1866 }
1867 
1868 
1869 static struct cxl *cxl_pci_init_adapter(struct pci_dev *dev)
1870 {
1871 	struct cxl *adapter;
1872 	int rc;
1873 
1874 	adapter = cxl_alloc_adapter();
1875 	if (!adapter)
1876 		return ERR_PTR(-ENOMEM);
1877 
1878 	adapter->native = kzalloc(sizeof(struct cxl_native), GFP_KERNEL);
1879 	if (!adapter->native) {
1880 		rc = -ENOMEM;
1881 		goto err_release;
1882 	}
1883 
1884 	set_sl_ops(adapter, dev);
1885 
1886 	/* Set defaults for parameters which need to persist over
1887 	 * configure/reconfigure
1888 	 */
1889 	adapter->perst_loads_image = true;
1890 	adapter->perst_same_image = false;
1891 
1892 	rc = cxl_configure_adapter(adapter, dev);
1893 	if (rc) {
1894 		pci_disable_device(dev);
1895 		goto err_release;
1896 	}
1897 
1898 	/* Don't care if this one fails: */
1899 	cxl_debugfs_adapter_add(adapter);
1900 
1901 	/*
1902 	 * After we call this function we must not free the adapter directly,
1903 	 * even if it returns an error!
1904 	 */
1905 	if ((rc = cxl_register_adapter(adapter)))
1906 		goto err_put1;
1907 
1908 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1909 		goto err_put1;
1910 
1911 	/* Release the context lock as adapter is configured */
1912 	cxl_adapter_context_unlock(adapter);
1913 
1914 	return adapter;
1915 
1916 err_put1:
1917 	/* This should mirror cxl_remove_adapter, except without the
1918 	 * sysfs parts
1919 	 */
1920 	cxl_debugfs_adapter_remove(adapter);
1921 	cxl_deconfigure_adapter(adapter);
1922 	device_unregister(&adapter->dev);
1923 	return ERR_PTR(rc);
1924 
1925 err_release:
1926 	cxl_release_adapter(&adapter->dev);
1927 	return ERR_PTR(rc);
1928 }
1929 
1930 static void cxl_pci_remove_adapter(struct cxl *adapter)
1931 {
1932 	pr_devel("cxl_remove_adapter\n");
1933 
1934 	cxl_sysfs_adapter_remove(adapter);
1935 	cxl_debugfs_adapter_remove(adapter);
1936 
1937 	/*
1938 	 * Flush adapter datacache as its about to be removed.
1939 	 * Not supported on P9 DD1.
1940 	 */
1941 	if ((cxl_is_power8()) || (!(cxl_is_power9_dd1())))
1942 		cxl_data_cache_flush(adapter);
1943 
1944 	cxl_deconfigure_adapter(adapter);
1945 
1946 	device_unregister(&adapter->dev);
1947 }
1948 
1949 #define CXL_MAX_PCIEX_PARENT 2
1950 
1951 int cxl_slot_is_switched(struct pci_dev *dev)
1952 {
1953 	struct device_node *np;
1954 	int depth = 0;
1955 	const __be32 *prop;
1956 
1957 	if (!(np = pci_device_to_OF_node(dev))) {
1958 		pr_err("cxl: np = NULL\n");
1959 		return -ENODEV;
1960 	}
1961 	of_node_get(np);
1962 	while (np) {
1963 		np = of_get_next_parent(np);
1964 		prop = of_get_property(np, "device_type", NULL);
1965 		if (!prop || strcmp((char *)prop, "pciex"))
1966 			break;
1967 		depth++;
1968 	}
1969 	of_node_put(np);
1970 	return (depth > CXL_MAX_PCIEX_PARENT);
1971 }
1972 
1973 bool cxl_slot_is_supported(struct pci_dev *dev, int flags)
1974 {
1975 	if (!cpu_has_feature(CPU_FTR_HVMODE))
1976 		return false;
1977 
1978 	if ((flags & CXL_SLOT_FLAG_DMA) && (!pvr_version_is(PVR_POWER8NVL))) {
1979 		/*
1980 		 * CAPP DMA mode is technically supported on regular P8, but
1981 		 * will EEH if the card attempts to access memory < 4GB, which
1982 		 * we cannot realistically avoid. We might be able to work
1983 		 * around the issue, but until then return unsupported:
1984 		 */
1985 		return false;
1986 	}
1987 
1988 	if (cxl_slot_is_switched(dev))
1989 		return false;
1990 
1991 	/*
1992 	 * XXX: This gets a little tricky on regular P8 (not POWER8NVL) since
1993 	 * the CAPP can be connected to PHB 0, 1 or 2 on a first come first
1994 	 * served basis, which is racy to check from here. If we need to
1995 	 * support this in future we might need to consider having this
1996 	 * function effectively reserve it ahead of time.
1997 	 *
1998 	 * Currently, the only user of this API is the Mellanox CX4, which is
1999 	 * only supported on P8NVL due to the above mentioned limitation of
2000 	 * CAPP DMA mode and therefore does not need to worry about this. If the
2001 	 * issue with CAPP DMA mode is later worked around on P8 we might need
2002 	 * to revisit this.
2003 	 */
2004 
2005 	return true;
2006 }
2007 EXPORT_SYMBOL_GPL(cxl_slot_is_supported);
2008 
2009 
2010 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
2011 {
2012 	struct cxl *adapter;
2013 	int slice;
2014 	int rc;
2015 
2016 	if (cxl_pci_is_vphb_device(dev)) {
2017 		dev_dbg(&dev->dev, "cxl_init_adapter: Ignoring cxl vphb device\n");
2018 		return -ENODEV;
2019 	}
2020 
2021 	if (cxl_slot_is_switched(dev)) {
2022 		dev_info(&dev->dev, "Ignoring card on incompatible PCI slot\n");
2023 		return -ENODEV;
2024 	}
2025 
2026 	if (cxl_is_power9() && !radix_enabled()) {
2027 		dev_info(&dev->dev, "Only Radix mode supported\n");
2028 		return -ENODEV;
2029 	}
2030 
2031 	if (cxl_verbose)
2032 		dump_cxl_config_space(dev);
2033 
2034 	adapter = cxl_pci_init_adapter(dev);
2035 	if (IS_ERR(adapter)) {
2036 		dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
2037 		return PTR_ERR(adapter);
2038 	}
2039 
2040 	for (slice = 0; slice < adapter->slices; slice++) {
2041 		if ((rc = pci_init_afu(adapter, slice, dev))) {
2042 			dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
2043 			continue;
2044 		}
2045 
2046 		rc = cxl_afu_select_best_mode(adapter->afu[slice]);
2047 		if (rc)
2048 			dev_err(&dev->dev, "AFU %i failed to start: %i\n", slice, rc);
2049 	}
2050 
2051 	if (pnv_pci_on_cxl_phb(dev) && adapter->slices >= 1)
2052 		pnv_cxl_phb_set_peer_afu(dev, adapter->afu[0]);
2053 
2054 	return 0;
2055 }
2056 
2057 static void cxl_remove(struct pci_dev *dev)
2058 {
2059 	struct cxl *adapter = pci_get_drvdata(dev);
2060 	struct cxl_afu *afu;
2061 	int i;
2062 
2063 	/*
2064 	 * Lock to prevent someone grabbing a ref through the adapter list as
2065 	 * we are removing it
2066 	 */
2067 	for (i = 0; i < adapter->slices; i++) {
2068 		afu = adapter->afu[i];
2069 		cxl_pci_remove_afu(afu);
2070 	}
2071 	cxl_pci_remove_adapter(adapter);
2072 }
2073 
2074 static pci_ers_result_t cxl_vphb_error_detected(struct cxl_afu *afu,
2075 						pci_channel_state_t state)
2076 {
2077 	struct pci_dev *afu_dev;
2078 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
2079 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
2080 
2081 	/* There should only be one entry, but go through the list
2082 	 * anyway
2083 	 */
2084 	if (afu->phb == NULL)
2085 		return result;
2086 
2087 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2088 		if (!afu_dev->driver)
2089 			continue;
2090 
2091 		afu_dev->error_state = state;
2092 
2093 		if (afu_dev->driver->err_handler)
2094 			afu_result = afu_dev->driver->err_handler->error_detected(afu_dev,
2095 										  state);
2096 		/* Disconnect trumps all, NONE trumps NEED_RESET */
2097 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2098 			result = PCI_ERS_RESULT_DISCONNECT;
2099 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
2100 			 (result == PCI_ERS_RESULT_NEED_RESET))
2101 			result = PCI_ERS_RESULT_NONE;
2102 	}
2103 	return result;
2104 }
2105 
2106 static pci_ers_result_t cxl_pci_error_detected(struct pci_dev *pdev,
2107 					       pci_channel_state_t state)
2108 {
2109 	struct cxl *adapter = pci_get_drvdata(pdev);
2110 	struct cxl_afu *afu;
2111 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET, afu_result;
2112 	int i;
2113 
2114 	/* At this point, we could still have an interrupt pending.
2115 	 * Let's try to get them out of the way before they do
2116 	 * anything we don't like.
2117 	 */
2118 	schedule();
2119 
2120 	/* If we're permanently dead, give up. */
2121 	if (state == pci_channel_io_perm_failure) {
2122 		for (i = 0; i < adapter->slices; i++) {
2123 			afu = adapter->afu[i];
2124 			/*
2125 			 * Tell the AFU drivers; but we don't care what they
2126 			 * say, we're going away.
2127 			 */
2128 			cxl_vphb_error_detected(afu, state);
2129 		}
2130 		return PCI_ERS_RESULT_DISCONNECT;
2131 	}
2132 
2133 	/* Are we reflashing?
2134 	 *
2135 	 * If we reflash, we could come back as something entirely
2136 	 * different, including a non-CAPI card. As such, by default
2137 	 * we don't participate in the process. We'll be unbound and
2138 	 * the slot re-probed. (TODO: check EEH doesn't blindly rebind
2139 	 * us!)
2140 	 *
2141 	 * However, this isn't the entire story: for reliablity
2142 	 * reasons, we usually want to reflash the FPGA on PERST in
2143 	 * order to get back to a more reliable known-good state.
2144 	 *
2145 	 * This causes us a bit of a problem: if we reflash we can't
2146 	 * trust that we'll come back the same - we could have a new
2147 	 * image and been PERSTed in order to load that
2148 	 * image. However, most of the time we actually *will* come
2149 	 * back the same - for example a regular EEH event.
2150 	 *
2151 	 * Therefore, we allow the user to assert that the image is
2152 	 * indeed the same and that we should continue on into EEH
2153 	 * anyway.
2154 	 */
2155 	if (adapter->perst_loads_image && !adapter->perst_same_image) {
2156 		/* TODO take the PHB out of CXL mode */
2157 		dev_info(&pdev->dev, "reflashing, so opting out of EEH!\n");
2158 		return PCI_ERS_RESULT_NONE;
2159 	}
2160 
2161 	/*
2162 	 * At this point, we want to try to recover.  We'll always
2163 	 * need a complete slot reset: we don't trust any other reset.
2164 	 *
2165 	 * Now, we go through each AFU:
2166 	 *  - We send the driver, if bound, an error_detected callback.
2167 	 *    We expect it to clean up, but it can also tell us to give
2168 	 *    up and permanently detach the card. To simplify things, if
2169 	 *    any bound AFU driver doesn't support EEH, we give up on EEH.
2170 	 *
2171 	 *  - We detach all contexts associated with the AFU. This
2172 	 *    does not free them, but puts them into a CLOSED state
2173 	 *    which causes any the associated files to return useful
2174 	 *    errors to userland. It also unmaps, but does not free,
2175 	 *    any IRQs.
2176 	 *
2177 	 *  - We clean up our side: releasing and unmapping resources we hold
2178 	 *    so we can wire them up again when the hardware comes back up.
2179 	 *
2180 	 * Driver authors should note:
2181 	 *
2182 	 *  - Any contexts you create in your kernel driver (except
2183 	 *    those associated with anonymous file descriptors) are
2184 	 *    your responsibility to free and recreate. Likewise with
2185 	 *    any attached resources.
2186 	 *
2187 	 *  - We will take responsibility for re-initialising the
2188 	 *    device context (the one set up for you in
2189 	 *    cxl_pci_enable_device_hook and accessed through
2190 	 *    cxl_get_context). If you've attached IRQs or other
2191 	 *    resources to it, they remains yours to free.
2192 	 *
2193 	 * You can call the same functions to release resources as you
2194 	 * normally would: we make sure that these functions continue
2195 	 * to work when the hardware is down.
2196 	 *
2197 	 * Two examples:
2198 	 *
2199 	 * 1) If you normally free all your resources at the end of
2200 	 *    each request, or if you use anonymous FDs, your
2201 	 *    error_detected callback can simply set a flag to tell
2202 	 *    your driver not to start any new calls. You can then
2203 	 *    clear the flag in the resume callback.
2204 	 *
2205 	 * 2) If you normally allocate your resources on startup:
2206 	 *     * Set a flag in error_detected as above.
2207 	 *     * Let CXL detach your contexts.
2208 	 *     * In slot_reset, free the old resources and allocate new ones.
2209 	 *     * In resume, clear the flag to allow things to start.
2210 	 */
2211 	for (i = 0; i < adapter->slices; i++) {
2212 		afu = adapter->afu[i];
2213 
2214 		afu_result = cxl_vphb_error_detected(afu, state);
2215 
2216 		cxl_context_detach_all(afu);
2217 		cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
2218 		pci_deconfigure_afu(afu);
2219 
2220 		/* Disconnect trumps all, NONE trumps NEED_RESET */
2221 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2222 			result = PCI_ERS_RESULT_DISCONNECT;
2223 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
2224 			 (result == PCI_ERS_RESULT_NEED_RESET))
2225 			result = PCI_ERS_RESULT_NONE;
2226 	}
2227 
2228 	/* should take the context lock here */
2229 	if (cxl_adapter_context_lock(adapter) != 0)
2230 		dev_warn(&adapter->dev,
2231 			 "Couldn't take context lock with %d active-contexts\n",
2232 			 atomic_read(&adapter->contexts_num));
2233 
2234 	cxl_deconfigure_adapter(adapter);
2235 
2236 	return result;
2237 }
2238 
2239 static pci_ers_result_t cxl_pci_slot_reset(struct pci_dev *pdev)
2240 {
2241 	struct cxl *adapter = pci_get_drvdata(pdev);
2242 	struct cxl_afu *afu;
2243 	struct cxl_context *ctx;
2244 	struct pci_dev *afu_dev;
2245 	pci_ers_result_t afu_result = PCI_ERS_RESULT_RECOVERED;
2246 	pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
2247 	int i;
2248 
2249 	if (cxl_configure_adapter(adapter, pdev))
2250 		goto err;
2251 
2252 	/*
2253 	 * Unlock context activation for the adapter. Ideally this should be
2254 	 * done in cxl_pci_resume but cxlflash module tries to activate the
2255 	 * master context as part of slot_reset callback.
2256 	 */
2257 	cxl_adapter_context_unlock(adapter);
2258 
2259 	for (i = 0; i < adapter->slices; i++) {
2260 		afu = adapter->afu[i];
2261 
2262 		if (pci_configure_afu(afu, adapter, pdev))
2263 			goto err;
2264 
2265 		if (cxl_afu_select_best_mode(afu))
2266 			goto err;
2267 
2268 		if (afu->phb == NULL)
2269 			continue;
2270 
2271 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2272 			/* Reset the device context.
2273 			 * TODO: make this less disruptive
2274 			 */
2275 			ctx = cxl_get_context(afu_dev);
2276 
2277 			if (ctx && cxl_release_context(ctx))
2278 				goto err;
2279 
2280 			ctx = cxl_dev_context_init(afu_dev);
2281 			if (IS_ERR(ctx))
2282 				goto err;
2283 
2284 			afu_dev->dev.archdata.cxl_ctx = ctx;
2285 
2286 			if (cxl_ops->afu_check_and_enable(afu))
2287 				goto err;
2288 
2289 			afu_dev->error_state = pci_channel_io_normal;
2290 
2291 			/* If there's a driver attached, allow it to
2292 			 * chime in on recovery. Drivers should check
2293 			 * if everything has come back OK, but
2294 			 * shouldn't start new work until we call
2295 			 * their resume function.
2296 			 */
2297 			if (!afu_dev->driver)
2298 				continue;
2299 
2300 			if (afu_dev->driver->err_handler &&
2301 			    afu_dev->driver->err_handler->slot_reset)
2302 				afu_result = afu_dev->driver->err_handler->slot_reset(afu_dev);
2303 
2304 			if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2305 				result = PCI_ERS_RESULT_DISCONNECT;
2306 		}
2307 	}
2308 	return result;
2309 
2310 err:
2311 	/* All the bits that happen in both error_detected and cxl_remove
2312 	 * should be idempotent, so we don't need to worry about leaving a mix
2313 	 * of unconfigured and reconfigured resources.
2314 	 */
2315 	dev_err(&pdev->dev, "EEH recovery failed. Asking to be disconnected.\n");
2316 	return PCI_ERS_RESULT_DISCONNECT;
2317 }
2318 
2319 static void cxl_pci_resume(struct pci_dev *pdev)
2320 {
2321 	struct cxl *adapter = pci_get_drvdata(pdev);
2322 	struct cxl_afu *afu;
2323 	struct pci_dev *afu_dev;
2324 	int i;
2325 
2326 	/* Everything is back now. Drivers should restart work now.
2327 	 * This is not the place to be checking if everything came back up
2328 	 * properly, because there's no return value: do that in slot_reset.
2329 	 */
2330 	for (i = 0; i < adapter->slices; i++) {
2331 		afu = adapter->afu[i];
2332 
2333 		if (afu->phb == NULL)
2334 			continue;
2335 
2336 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2337 			if (afu_dev->driver && afu_dev->driver->err_handler &&
2338 			    afu_dev->driver->err_handler->resume)
2339 				afu_dev->driver->err_handler->resume(afu_dev);
2340 		}
2341 	}
2342 }
2343 
2344 static const struct pci_error_handlers cxl_err_handler = {
2345 	.error_detected = cxl_pci_error_detected,
2346 	.slot_reset = cxl_pci_slot_reset,
2347 	.resume = cxl_pci_resume,
2348 };
2349 
2350 struct pci_driver cxl_pci_driver = {
2351 	.name = "cxl-pci",
2352 	.id_table = cxl_pci_tbl,
2353 	.probe = cxl_probe,
2354 	.remove = cxl_remove,
2355 	.shutdown = cxl_remove,
2356 	.err_handler = &cxl_err_handler,
2357 };
2358