xref: /linux/drivers/misc/cxl/pci.c (revision d99ff463ecf651437e9e4abe68f331dfb6b5bd9d)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright 2014 IBM Corp.
4  */
5 
6 #include <linux/pci_regs.h>
7 #include <linux/pci_ids.h>
8 #include <linux/device.h>
9 #include <linux/module.h>
10 #include <linux/kernel.h>
11 #include <linux/slab.h>
12 #include <linux/sort.h>
13 #include <linux/pci.h>
14 #include <linux/of.h>
15 #include <linux/delay.h>
16 #include <asm/opal.h>
17 #include <asm/msi_bitmap.h>
18 #include <asm/pnv-pci.h>
19 #include <asm/io.h>
20 #include <asm/reg.h>
21 
22 #include "cxl.h"
23 #include <misc/cxl.h>
24 
25 
26 #define CXL_PCI_VSEC_ID	0x1280
27 #define CXL_VSEC_MIN_SIZE 0x80
28 
29 #define CXL_READ_VSEC_LENGTH(dev, vsec, dest)			\
30 	{							\
31 		pci_read_config_word(dev, vsec + 0x6, dest);	\
32 		*dest >>= 4;					\
33 	}
34 #define CXL_READ_VSEC_NAFUS(dev, vsec, dest) \
35 	pci_read_config_byte(dev, vsec + 0x8, dest)
36 
37 #define CXL_READ_VSEC_STATUS(dev, vsec, dest) \
38 	pci_read_config_byte(dev, vsec + 0x9, dest)
39 #define CXL_STATUS_SECOND_PORT  0x80
40 #define CXL_STATUS_MSI_X_FULL   0x40
41 #define CXL_STATUS_MSI_X_SINGLE 0x20
42 #define CXL_STATUS_FLASH_RW     0x08
43 #define CXL_STATUS_FLASH_RO     0x04
44 #define CXL_STATUS_LOADABLE_AFU 0x02
45 #define CXL_STATUS_LOADABLE_PSL 0x01
46 /* If we see these features we won't try to use the card */
47 #define CXL_UNSUPPORTED_FEATURES \
48 	(CXL_STATUS_MSI_X_FULL | CXL_STATUS_MSI_X_SINGLE)
49 
50 #define CXL_READ_VSEC_MODE_CONTROL(dev, vsec, dest) \
51 	pci_read_config_byte(dev, vsec + 0xa, dest)
52 #define CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val) \
53 	pci_write_config_byte(dev, vsec + 0xa, val)
54 #define CXL_VSEC_PROTOCOL_MASK   0xe0
55 #define CXL_VSEC_PROTOCOL_1024TB 0x80
56 #define CXL_VSEC_PROTOCOL_512TB  0x40
57 #define CXL_VSEC_PROTOCOL_256TB  0x20 /* Power 8/9 uses this */
58 #define CXL_VSEC_PROTOCOL_ENABLE 0x01
59 
60 #define CXL_READ_VSEC_PSL_REVISION(dev, vsec, dest) \
61 	pci_read_config_word(dev, vsec + 0xc, dest)
62 #define CXL_READ_VSEC_CAIA_MINOR(dev, vsec, dest) \
63 	pci_read_config_byte(dev, vsec + 0xe, dest)
64 #define CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, dest) \
65 	pci_read_config_byte(dev, vsec + 0xf, dest)
66 #define CXL_READ_VSEC_BASE_IMAGE(dev, vsec, dest) \
67 	pci_read_config_word(dev, vsec + 0x10, dest)
68 
69 #define CXL_READ_VSEC_IMAGE_STATE(dev, vsec, dest) \
70 	pci_read_config_byte(dev, vsec + 0x13, dest)
71 #define CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, val) \
72 	pci_write_config_byte(dev, vsec + 0x13, val)
73 #define CXL_VSEC_USER_IMAGE_LOADED 0x80 /* RO */
74 #define CXL_VSEC_PERST_LOADS_IMAGE 0x20 /* RW */
75 #define CXL_VSEC_PERST_SELECT_USER 0x10 /* RW */
76 
77 #define CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, dest) \
78 	pci_read_config_dword(dev, vsec + 0x20, dest)
79 #define CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, dest) \
80 	pci_read_config_dword(dev, vsec + 0x24, dest)
81 #define CXL_READ_VSEC_PS_OFF(dev, vsec, dest) \
82 	pci_read_config_dword(dev, vsec + 0x28, dest)
83 #define CXL_READ_VSEC_PS_SIZE(dev, vsec, dest) \
84 	pci_read_config_dword(dev, vsec + 0x2c, dest)
85 
86 
87 /* This works a little different than the p1/p2 register accesses to make it
88  * easier to pull out individual fields */
89 #define AFUD_READ(afu, off)		in_be64(afu->native->afu_desc_mmio + off)
90 #define AFUD_READ_LE(afu, off)		in_le64(afu->native->afu_desc_mmio + off)
91 #define EXTRACT_PPC_BIT(val, bit)	(!!(val & PPC_BIT(bit)))
92 #define EXTRACT_PPC_BITS(val, bs, be)	((val & PPC_BITMASK(bs, be)) >> PPC_BITLSHIFT(be))
93 
94 #define AFUD_READ_INFO(afu)		AFUD_READ(afu, 0x0)
95 #define   AFUD_NUM_INTS_PER_PROC(val)	EXTRACT_PPC_BITS(val,  0, 15)
96 #define   AFUD_NUM_PROCS(val)		EXTRACT_PPC_BITS(val, 16, 31)
97 #define   AFUD_NUM_CRS(val)		EXTRACT_PPC_BITS(val, 32, 47)
98 #define   AFUD_MULTIMODE(val)		EXTRACT_PPC_BIT(val, 48)
99 #define   AFUD_PUSH_BLOCK_TRANSFER(val)	EXTRACT_PPC_BIT(val, 55)
100 #define   AFUD_DEDICATED_PROCESS(val)	EXTRACT_PPC_BIT(val, 59)
101 #define   AFUD_AFU_DIRECTED(val)	EXTRACT_PPC_BIT(val, 61)
102 #define   AFUD_TIME_SLICED(val)		EXTRACT_PPC_BIT(val, 63)
103 #define AFUD_READ_CR(afu)		AFUD_READ(afu, 0x20)
104 #define   AFUD_CR_LEN(val)		EXTRACT_PPC_BITS(val, 8, 63)
105 #define AFUD_READ_CR_OFF(afu)		AFUD_READ(afu, 0x28)
106 #define AFUD_READ_PPPSA(afu)		AFUD_READ(afu, 0x30)
107 #define   AFUD_PPPSA_PP(val)		EXTRACT_PPC_BIT(val, 6)
108 #define   AFUD_PPPSA_PSA(val)		EXTRACT_PPC_BIT(val, 7)
109 #define   AFUD_PPPSA_LEN(val)		EXTRACT_PPC_BITS(val, 8, 63)
110 #define AFUD_READ_PPPSA_OFF(afu)	AFUD_READ(afu, 0x38)
111 #define AFUD_READ_EB(afu)		AFUD_READ(afu, 0x40)
112 #define   AFUD_EB_LEN(val)		EXTRACT_PPC_BITS(val, 8, 63)
113 #define AFUD_READ_EB_OFF(afu)		AFUD_READ(afu, 0x48)
114 
115 static const struct pci_device_id cxl_pci_tbl[] = {
116 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0477), },
117 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x044b), },
118 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x04cf), },
119 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0601), },
120 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0623), },
121 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0628), },
122 	{ }
123 };
124 MODULE_DEVICE_TABLE(pci, cxl_pci_tbl);
125 
126 
127 /*
128  * Mostly using these wrappers to avoid confusion:
129  * priv 1 is BAR2, while priv 2 is BAR0
130  */
131 static inline resource_size_t p1_base(struct pci_dev *dev)
132 {
133 	return pci_resource_start(dev, 2);
134 }
135 
136 static inline resource_size_t p1_size(struct pci_dev *dev)
137 {
138 	return pci_resource_len(dev, 2);
139 }
140 
141 static inline resource_size_t p2_base(struct pci_dev *dev)
142 {
143 	return pci_resource_start(dev, 0);
144 }
145 
146 static inline resource_size_t p2_size(struct pci_dev *dev)
147 {
148 	return pci_resource_len(dev, 0);
149 }
150 
151 static int find_cxl_vsec(struct pci_dev *dev)
152 {
153 	int vsec = 0;
154 	u16 val;
155 
156 	while ((vsec = pci_find_next_ext_capability(dev, vsec, PCI_EXT_CAP_ID_VNDR))) {
157 		pci_read_config_word(dev, vsec + 0x4, &val);
158 		if (val == CXL_PCI_VSEC_ID)
159 			return vsec;
160 	}
161 	return 0;
162 
163 }
164 
165 static void dump_cxl_config_space(struct pci_dev *dev)
166 {
167 	int vsec;
168 	u32 val;
169 
170 	dev_info(&dev->dev, "dump_cxl_config_space\n");
171 
172 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_0, &val);
173 	dev_info(&dev->dev, "BAR0: %#.8x\n", val);
174 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_1, &val);
175 	dev_info(&dev->dev, "BAR1: %#.8x\n", val);
176 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_2, &val);
177 	dev_info(&dev->dev, "BAR2: %#.8x\n", val);
178 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_3, &val);
179 	dev_info(&dev->dev, "BAR3: %#.8x\n", val);
180 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_4, &val);
181 	dev_info(&dev->dev, "BAR4: %#.8x\n", val);
182 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_5, &val);
183 	dev_info(&dev->dev, "BAR5: %#.8x\n", val);
184 
185 	dev_info(&dev->dev, "p1 regs: %#llx, len: %#llx\n",
186 		p1_base(dev), p1_size(dev));
187 	dev_info(&dev->dev, "p2 regs: %#llx, len: %#llx\n",
188 		p2_base(dev), p2_size(dev));
189 	dev_info(&dev->dev, "BAR 4/5: %#llx, len: %#llx\n",
190 		pci_resource_start(dev, 4), pci_resource_len(dev, 4));
191 
192 	if (!(vsec = find_cxl_vsec(dev)))
193 		return;
194 
195 #define show_reg(name, what) \
196 	dev_info(&dev->dev, "cxl vsec: %30s: %#x\n", name, what)
197 
198 	pci_read_config_dword(dev, vsec + 0x0, &val);
199 	show_reg("Cap ID", (val >> 0) & 0xffff);
200 	show_reg("Cap Ver", (val >> 16) & 0xf);
201 	show_reg("Next Cap Ptr", (val >> 20) & 0xfff);
202 	pci_read_config_dword(dev, vsec + 0x4, &val);
203 	show_reg("VSEC ID", (val >> 0) & 0xffff);
204 	show_reg("VSEC Rev", (val >> 16) & 0xf);
205 	show_reg("VSEC Length",	(val >> 20) & 0xfff);
206 	pci_read_config_dword(dev, vsec + 0x8, &val);
207 	show_reg("Num AFUs", (val >> 0) & 0xff);
208 	show_reg("Status", (val >> 8) & 0xff);
209 	show_reg("Mode Control", (val >> 16) & 0xff);
210 	show_reg("Reserved", (val >> 24) & 0xff);
211 	pci_read_config_dword(dev, vsec + 0xc, &val);
212 	show_reg("PSL Rev", (val >> 0) & 0xffff);
213 	show_reg("CAIA Ver", (val >> 16) & 0xffff);
214 	pci_read_config_dword(dev, vsec + 0x10, &val);
215 	show_reg("Base Image Rev", (val >> 0) & 0xffff);
216 	show_reg("Reserved", (val >> 16) & 0x0fff);
217 	show_reg("Image Control", (val >> 28) & 0x3);
218 	show_reg("Reserved", (val >> 30) & 0x1);
219 	show_reg("Image Loaded", (val >> 31) & 0x1);
220 
221 	pci_read_config_dword(dev, vsec + 0x14, &val);
222 	show_reg("Reserved", val);
223 	pci_read_config_dword(dev, vsec + 0x18, &val);
224 	show_reg("Reserved", val);
225 	pci_read_config_dword(dev, vsec + 0x1c, &val);
226 	show_reg("Reserved", val);
227 
228 	pci_read_config_dword(dev, vsec + 0x20, &val);
229 	show_reg("AFU Descriptor Offset", val);
230 	pci_read_config_dword(dev, vsec + 0x24, &val);
231 	show_reg("AFU Descriptor Size", val);
232 	pci_read_config_dword(dev, vsec + 0x28, &val);
233 	show_reg("Problem State Offset", val);
234 	pci_read_config_dword(dev, vsec + 0x2c, &val);
235 	show_reg("Problem State Size", val);
236 
237 	pci_read_config_dword(dev, vsec + 0x30, &val);
238 	show_reg("Reserved", val);
239 	pci_read_config_dword(dev, vsec + 0x34, &val);
240 	show_reg("Reserved", val);
241 	pci_read_config_dword(dev, vsec + 0x38, &val);
242 	show_reg("Reserved", val);
243 	pci_read_config_dword(dev, vsec + 0x3c, &val);
244 	show_reg("Reserved", val);
245 
246 	pci_read_config_dword(dev, vsec + 0x40, &val);
247 	show_reg("PSL Programming Port", val);
248 	pci_read_config_dword(dev, vsec + 0x44, &val);
249 	show_reg("PSL Programming Control", val);
250 
251 	pci_read_config_dword(dev, vsec + 0x48, &val);
252 	show_reg("Reserved", val);
253 	pci_read_config_dword(dev, vsec + 0x4c, &val);
254 	show_reg("Reserved", val);
255 
256 	pci_read_config_dword(dev, vsec + 0x50, &val);
257 	show_reg("Flash Address Register", val);
258 	pci_read_config_dword(dev, vsec + 0x54, &val);
259 	show_reg("Flash Size Register", val);
260 	pci_read_config_dword(dev, vsec + 0x58, &val);
261 	show_reg("Flash Status/Control Register", val);
262 	pci_read_config_dword(dev, vsec + 0x58, &val);
263 	show_reg("Flash Data Port", val);
264 
265 #undef show_reg
266 }
267 
268 static void dump_afu_descriptor(struct cxl_afu *afu)
269 {
270 	u64 val, afu_cr_num, afu_cr_off, afu_cr_len;
271 	int i;
272 
273 #define show_reg(name, what) \
274 	dev_info(&afu->dev, "afu desc: %30s: %#llx\n", name, what)
275 
276 	val = AFUD_READ_INFO(afu);
277 	show_reg("num_ints_per_process", AFUD_NUM_INTS_PER_PROC(val));
278 	show_reg("num_of_processes", AFUD_NUM_PROCS(val));
279 	show_reg("num_of_afu_CRs", AFUD_NUM_CRS(val));
280 	show_reg("req_prog_mode", val & 0xffffULL);
281 	afu_cr_num = AFUD_NUM_CRS(val);
282 
283 	val = AFUD_READ(afu, 0x8);
284 	show_reg("Reserved", val);
285 	val = AFUD_READ(afu, 0x10);
286 	show_reg("Reserved", val);
287 	val = AFUD_READ(afu, 0x18);
288 	show_reg("Reserved", val);
289 
290 	val = AFUD_READ_CR(afu);
291 	show_reg("Reserved", (val >> (63-7)) & 0xff);
292 	show_reg("AFU_CR_len", AFUD_CR_LEN(val));
293 	afu_cr_len = AFUD_CR_LEN(val) * 256;
294 
295 	val = AFUD_READ_CR_OFF(afu);
296 	afu_cr_off = val;
297 	show_reg("AFU_CR_offset", val);
298 
299 	val = AFUD_READ_PPPSA(afu);
300 	show_reg("PerProcessPSA_control", (val >> (63-7)) & 0xff);
301 	show_reg("PerProcessPSA Length", AFUD_PPPSA_LEN(val));
302 
303 	val = AFUD_READ_PPPSA_OFF(afu);
304 	show_reg("PerProcessPSA_offset", val);
305 
306 	val = AFUD_READ_EB(afu);
307 	show_reg("Reserved", (val >> (63-7)) & 0xff);
308 	show_reg("AFU_EB_len", AFUD_EB_LEN(val));
309 
310 	val = AFUD_READ_EB_OFF(afu);
311 	show_reg("AFU_EB_offset", val);
312 
313 	for (i = 0; i < afu_cr_num; i++) {
314 		val = AFUD_READ_LE(afu, afu_cr_off + i * afu_cr_len);
315 		show_reg("CR Vendor", val & 0xffff);
316 		show_reg("CR Device", (val >> 16) & 0xffff);
317 	}
318 #undef show_reg
319 }
320 
321 #define P8_CAPP_UNIT0_ID 0xBA
322 #define P8_CAPP_UNIT1_ID 0XBE
323 #define P9_CAPP_UNIT0_ID 0xC0
324 #define P9_CAPP_UNIT1_ID 0xE0
325 
326 static int get_phb_index(struct device_node *np, u32 *phb_index)
327 {
328 	if (of_property_read_u32(np, "ibm,phb-index", phb_index))
329 		return -ENODEV;
330 	return 0;
331 }
332 
333 static u64 get_capp_unit_id(struct device_node *np, u32 phb_index)
334 {
335 	/*
336 	 * POWER 8:
337 	 *  - For chips other than POWER8NVL, we only have CAPP 0,
338 	 *    irrespective of which PHB is used.
339 	 *  - For POWER8NVL, assume CAPP 0 is attached to PHB0 and
340 	 *    CAPP 1 is attached to PHB1.
341 	 */
342 	if (cxl_is_power8()) {
343 		if (!pvr_version_is(PVR_POWER8NVL))
344 			return P8_CAPP_UNIT0_ID;
345 
346 		if (phb_index == 0)
347 			return P8_CAPP_UNIT0_ID;
348 
349 		if (phb_index == 1)
350 			return P8_CAPP_UNIT1_ID;
351 	}
352 
353 	/*
354 	 * POWER 9:
355 	 *   PEC0 (PHB0). Capp ID = CAPP0 (0b1100_0000)
356 	 *   PEC1 (PHB1 - PHB2). No capi mode
357 	 *   PEC2 (PHB3 - PHB4 - PHB5): Capi mode on PHB3 only. Capp ID = CAPP1 (0b1110_0000)
358 	 */
359 	if (cxl_is_power9()) {
360 		if (phb_index == 0)
361 			return P9_CAPP_UNIT0_ID;
362 
363 		if (phb_index == 3)
364 			return P9_CAPP_UNIT1_ID;
365 	}
366 
367 	return 0;
368 }
369 
370 int cxl_calc_capp_routing(struct pci_dev *dev, u64 *chipid,
371 			     u32 *phb_index, u64 *capp_unit_id)
372 {
373 	int rc;
374 	struct device_node *np;
375 	const __be32 *prop;
376 
377 	if (!(np = pnv_pci_get_phb_node(dev)))
378 		return -ENODEV;
379 
380 	while (np && !(prop = of_get_property(np, "ibm,chip-id", NULL)))
381 		np = of_get_next_parent(np);
382 	if (!np)
383 		return -ENODEV;
384 
385 	*chipid = be32_to_cpup(prop);
386 
387 	rc = get_phb_index(np, phb_index);
388 	if (rc) {
389 		pr_err("cxl: invalid phb index\n");
390 		of_node_put(np);
391 		return rc;
392 	}
393 
394 	*capp_unit_id = get_capp_unit_id(np, *phb_index);
395 	of_node_put(np);
396 	if (!*capp_unit_id) {
397 		pr_err("cxl: No capp unit found for PHB[%lld,%d]. Make sure the adapter is on a capi-compatible slot\n",
398 		       *chipid, *phb_index);
399 		return -ENODEV;
400 	}
401 
402 	return 0;
403 }
404 
405 static DEFINE_MUTEX(indications_mutex);
406 
407 static int get_phb_indications(struct pci_dev *dev, u64 *capiind, u64 *asnind,
408 			       u64 *nbwind)
409 {
410 	static u64 nbw, asn, capi = 0;
411 	struct device_node *np;
412 	const __be32 *prop;
413 
414 	mutex_lock(&indications_mutex);
415 	if (!capi) {
416 		if (!(np = pnv_pci_get_phb_node(dev))) {
417 			mutex_unlock(&indications_mutex);
418 			return -ENODEV;
419 		}
420 
421 		prop = of_get_property(np, "ibm,phb-indications", NULL);
422 		if (!prop) {
423 			nbw = 0x0300UL; /* legacy values */
424 			asn = 0x0400UL;
425 			capi = 0x0200UL;
426 		} else {
427 			nbw = (u64)be32_to_cpu(prop[2]);
428 			asn = (u64)be32_to_cpu(prop[1]);
429 			capi = (u64)be32_to_cpu(prop[0]);
430 		}
431 		of_node_put(np);
432 	}
433 	*capiind = capi;
434 	*asnind = asn;
435 	*nbwind = nbw;
436 	mutex_unlock(&indications_mutex);
437 	return 0;
438 }
439 
440 int cxl_get_xsl9_dsnctl(struct pci_dev *dev, u64 capp_unit_id, u64 *reg)
441 {
442 	u64 xsl_dsnctl;
443 	u64 capiind, asnind, nbwind;
444 
445 	/*
446 	 * CAPI Identifier bits [0:7]
447 	 * bit 61:60 MSI bits --> 0
448 	 * bit 59 TVT selector --> 0
449 	 */
450 	if (get_phb_indications(dev, &capiind, &asnind, &nbwind))
451 		return -ENODEV;
452 
453 	/*
454 	 * Tell XSL where to route data to.
455 	 * The field chipid should match the PHB CAPI_CMPM register
456 	 */
457 	xsl_dsnctl = (capiind << (63-15)); /* Bit 57 */
458 	xsl_dsnctl |= (capp_unit_id << (63-15));
459 
460 	/* nMMU_ID Defaults to: b’000001001’*/
461 	xsl_dsnctl |= ((u64)0x09 << (63-28));
462 
463 	/*
464 	 * Used to identify CAPI packets which should be sorted into
465 	 * the Non-Blocking queues by the PHB. This field should match
466 	 * the PHB PBL_NBW_CMPM register
467 	 * nbwind=0x03, bits [57:58], must include capi indicator.
468 	 * Not supported on P9 DD1.
469 	 */
470 	xsl_dsnctl |= (nbwind << (63-55));
471 
472 	/*
473 	 * Upper 16b address bits of ASB_Notify messages sent to the
474 	 * system. Need to match the PHB’s ASN Compare/Mask Register.
475 	 * Not supported on P9 DD1.
476 	 */
477 	xsl_dsnctl |= asnind;
478 
479 	*reg = xsl_dsnctl;
480 	return 0;
481 }
482 
483 static int init_implementation_adapter_regs_psl9(struct cxl *adapter,
484 						 struct pci_dev *dev)
485 {
486 	u64 xsl_dsnctl, psl_fircntl;
487 	u64 chipid;
488 	u32 phb_index;
489 	u64 capp_unit_id;
490 	u64 psl_debug;
491 	int rc;
492 
493 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
494 	if (rc)
495 		return rc;
496 
497 	rc = cxl_get_xsl9_dsnctl(dev, capp_unit_id, &xsl_dsnctl);
498 	if (rc)
499 		return rc;
500 
501 	cxl_p1_write(adapter, CXL_XSL9_DSNCTL, xsl_dsnctl);
502 
503 	/* Set fir_cntl to recommended value for production env */
504 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
505 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
506 	psl_fircntl |= 0x1ULL; /* ce_thresh */
507 	cxl_p1_write(adapter, CXL_PSL9_FIR_CNTL, psl_fircntl);
508 
509 	/* Setup the PSL to transmit packets on the PCIe before the
510 	 * CAPP is enabled. Make sure that CAPP virtual machines are disabled
511 	 */
512 	cxl_p1_write(adapter, CXL_PSL9_DSNDCTL, 0x0001001000012A10ULL);
513 
514 	/*
515 	 * A response to an ASB_Notify request is returned by the
516 	 * system as an MMIO write to the address defined in
517 	 * the PSL_TNR_ADDR register.
518 	 * keep the Reset Value: 0x00020000E0000000
519 	 */
520 
521 	/* Enable XSL rty limit */
522 	cxl_p1_write(adapter, CXL_XSL9_DEF, 0x51F8000000000005ULL);
523 
524 	/* Change XSL_INV dummy read threshold */
525 	cxl_p1_write(adapter, CXL_XSL9_INV, 0x0000040007FFC200ULL);
526 
527 	if (phb_index == 3) {
528 		/* disable machines 31-47 and 20-27 for DMA */
529 		cxl_p1_write(adapter, CXL_PSL9_APCDEDTYPE, 0x40000FF3FFFF0000ULL);
530 	}
531 
532 	/* Snoop machines */
533 	cxl_p1_write(adapter, CXL_PSL9_APCDEDALLOC, 0x800F000200000000ULL);
534 
535 	/* Enable NORST and DD2 features */
536 	cxl_p1_write(adapter, CXL_PSL9_DEBUG, 0xC000000000000000ULL);
537 
538 	/*
539 	 * Check if PSL has data-cache. We need to flush adapter datacache
540 	 * when as its about to be removed.
541 	 */
542 	psl_debug = cxl_p1_read(adapter, CXL_PSL9_DEBUG);
543 	if (psl_debug & CXL_PSL_DEBUG_CDC) {
544 		dev_dbg(&dev->dev, "No data-cache present\n");
545 		adapter->native->no_data_cache = true;
546 	}
547 
548 	return 0;
549 }
550 
551 static int init_implementation_adapter_regs_psl8(struct cxl *adapter, struct pci_dev *dev)
552 {
553 	u64 psl_dsnctl, psl_fircntl;
554 	u64 chipid;
555 	u32 phb_index;
556 	u64 capp_unit_id;
557 	int rc;
558 
559 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
560 	if (rc)
561 		return rc;
562 
563 	psl_dsnctl = 0x0000900000000000ULL; /* pteupd ttype, scdone */
564 	psl_dsnctl |= (0x2ULL << (63-38)); /* MMIO hang pulse: 256 us */
565 	/* Tell PSL where to route data to */
566 	psl_dsnctl |= (chipid << (63-5));
567 	psl_dsnctl |= (capp_unit_id << (63-13));
568 
569 	cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
570 	cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
571 	/* snoop write mask */
572 	cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
573 	/* set fir_cntl to recommended value for production env */
574 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
575 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
576 	psl_fircntl |= 0x1ULL; /* ce_thresh */
577 	cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, psl_fircntl);
578 	/* for debugging with trace arrays */
579 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
580 
581 	return 0;
582 }
583 
584 /* PSL */
585 #define TBSYNC_CAL(n) (((u64)n & 0x7) << (63-3))
586 #define TBSYNC_CNT(n) (((u64)n & 0x7) << (63-6))
587 /* For the PSL this is a multiple for 0 < n <= 7: */
588 #define PSL_2048_250MHZ_CYCLES 1
589 
590 static void write_timebase_ctrl_psl8(struct cxl *adapter)
591 {
592 	cxl_p1_write(adapter, CXL_PSL_TB_CTLSTAT,
593 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
594 }
595 
596 static u64 timebase_read_psl9(struct cxl *adapter)
597 {
598 	return cxl_p1_read(adapter, CXL_PSL9_Timebase);
599 }
600 
601 static u64 timebase_read_psl8(struct cxl *adapter)
602 {
603 	return cxl_p1_read(adapter, CXL_PSL_Timebase);
604 }
605 
606 static void cxl_setup_psl_timebase(struct cxl *adapter, struct pci_dev *dev)
607 {
608 	struct device_node *np;
609 
610 	adapter->psl_timebase_synced = false;
611 
612 	if (!(np = pnv_pci_get_phb_node(dev)))
613 		return;
614 
615 	/* Do not fail when CAPP timebase sync is not supported by OPAL */
616 	of_node_get(np);
617 	if (! of_get_property(np, "ibm,capp-timebase-sync", NULL)) {
618 		of_node_put(np);
619 		dev_info(&dev->dev, "PSL timebase inactive: OPAL support missing\n");
620 		return;
621 	}
622 	of_node_put(np);
623 
624 	/*
625 	 * Setup PSL Timebase Control and Status register
626 	 * with the recommended Timebase Sync Count value
627 	 */
628 	if (adapter->native->sl_ops->write_timebase_ctrl)
629 		adapter->native->sl_ops->write_timebase_ctrl(adapter);
630 
631 	/* Enable PSL Timebase */
632 	cxl_p1_write(adapter, CXL_PSL_Control, 0x0000000000000000);
633 	cxl_p1_write(adapter, CXL_PSL_Control, CXL_PSL_Control_tb);
634 
635 	return;
636 }
637 
638 static int init_implementation_afu_regs_psl9(struct cxl_afu *afu)
639 {
640 	return 0;
641 }
642 
643 static int init_implementation_afu_regs_psl8(struct cxl_afu *afu)
644 {
645 	/* read/write masks for this slice */
646 	cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
647 	/* APC read/write masks for this slice */
648 	cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
649 	/* for debugging with trace arrays */
650 	cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
651 	cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
652 
653 	return 0;
654 }
655 
656 int cxl_pci_setup_irq(struct cxl *adapter, unsigned int hwirq,
657 		unsigned int virq)
658 {
659 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
660 
661 	return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
662 }
663 
664 int cxl_update_image_control(struct cxl *adapter)
665 {
666 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
667 	int rc;
668 	int vsec;
669 	u8 image_state;
670 
671 	if (!(vsec = find_cxl_vsec(dev))) {
672 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
673 		return -ENODEV;
674 	}
675 
676 	if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
677 		dev_err(&dev->dev, "failed to read image state: %i\n", rc);
678 		return rc;
679 	}
680 
681 	if (adapter->perst_loads_image)
682 		image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
683 	else
684 		image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
685 
686 	if (adapter->perst_select_user)
687 		image_state |= CXL_VSEC_PERST_SELECT_USER;
688 	else
689 		image_state &= ~CXL_VSEC_PERST_SELECT_USER;
690 
691 	if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
692 		dev_err(&dev->dev, "failed to update image control: %i\n", rc);
693 		return rc;
694 	}
695 
696 	return 0;
697 }
698 
699 int cxl_pci_alloc_one_irq(struct cxl *adapter)
700 {
701 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
702 
703 	return pnv_cxl_alloc_hwirqs(dev, 1);
704 }
705 
706 void cxl_pci_release_one_irq(struct cxl *adapter, int hwirq)
707 {
708 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
709 
710 	return pnv_cxl_release_hwirqs(dev, hwirq, 1);
711 }
712 
713 int cxl_pci_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
714 			struct cxl *adapter, unsigned int num)
715 {
716 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
717 
718 	return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
719 }
720 
721 void cxl_pci_release_irq_ranges(struct cxl_irq_ranges *irqs,
722 				struct cxl *adapter)
723 {
724 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
725 
726 	pnv_cxl_release_hwirq_ranges(irqs, dev);
727 }
728 
729 static int setup_cxl_bars(struct pci_dev *dev)
730 {
731 	/* Safety check in case we get backported to < 3.17 without M64 */
732 	if ((p1_base(dev) < 0x100000000ULL) ||
733 	    (p2_base(dev) < 0x100000000ULL)) {
734 		dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
735 		return -ENODEV;
736 	}
737 
738 	/*
739 	 * BAR 4/5 has a special meaning for CXL and must be programmed with a
740 	 * special value corresponding to the CXL protocol address range.
741 	 * For POWER 8/9 that means bits 48:49 must be set to 10
742 	 */
743 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
744 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
745 
746 	return 0;
747 }
748 
749 /* pciex node: ibm,opal-m64-window = <0x3d058 0x0 0x3d058 0x0 0x8 0x0>; */
750 static int switch_card_to_cxl(struct pci_dev *dev)
751 {
752 	int vsec;
753 	u8 val;
754 	int rc;
755 
756 	dev_info(&dev->dev, "switch card to CXL\n");
757 
758 	if (!(vsec = find_cxl_vsec(dev))) {
759 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
760 		return -ENODEV;
761 	}
762 
763 	if ((rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val))) {
764 		dev_err(&dev->dev, "failed to read current mode control: %i", rc);
765 		return rc;
766 	}
767 	val &= ~CXL_VSEC_PROTOCOL_MASK;
768 	val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
769 	if ((rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val))) {
770 		dev_err(&dev->dev, "failed to enable CXL protocol: %i", rc);
771 		return rc;
772 	}
773 	/*
774 	 * The CAIA spec (v0.12 11.6 Bi-modal Device Support) states
775 	 * we must wait 100ms after this mode switch before touching
776 	 * PCIe config space.
777 	 */
778 	msleep(100);
779 
780 	return 0;
781 }
782 
783 static int pci_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
784 {
785 	u64 p1n_base, p2n_base, afu_desc;
786 	const u64 p1n_size = 0x100;
787 	const u64 p2n_size = 0x1000;
788 
789 	p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
790 	p2n_base = p2_base(dev) + (afu->slice * p2n_size);
791 	afu->psn_phys = p2_base(dev) + (adapter->native->ps_off + (afu->slice * adapter->ps_size));
792 	afu_desc = p2_base(dev) + adapter->native->afu_desc_off + (afu->slice * adapter->native->afu_desc_size);
793 
794 	if (!(afu->native->p1n_mmio = ioremap(p1n_base, p1n_size)))
795 		goto err;
796 	if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
797 		goto err1;
798 	if (afu_desc) {
799 		if (!(afu->native->afu_desc_mmio = ioremap(afu_desc, adapter->native->afu_desc_size)))
800 			goto err2;
801 	}
802 
803 	return 0;
804 err2:
805 	iounmap(afu->p2n_mmio);
806 err1:
807 	iounmap(afu->native->p1n_mmio);
808 err:
809 	dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
810 	return -ENOMEM;
811 }
812 
813 static void pci_unmap_slice_regs(struct cxl_afu *afu)
814 {
815 	if (afu->p2n_mmio) {
816 		iounmap(afu->p2n_mmio);
817 		afu->p2n_mmio = NULL;
818 	}
819 	if (afu->native->p1n_mmio) {
820 		iounmap(afu->native->p1n_mmio);
821 		afu->native->p1n_mmio = NULL;
822 	}
823 	if (afu->native->afu_desc_mmio) {
824 		iounmap(afu->native->afu_desc_mmio);
825 		afu->native->afu_desc_mmio = NULL;
826 	}
827 }
828 
829 void cxl_pci_release_afu(struct device *dev)
830 {
831 	struct cxl_afu *afu = to_cxl_afu(dev);
832 
833 	pr_devel("%s\n", __func__);
834 
835 	idr_destroy(&afu->contexts_idr);
836 	cxl_release_spa(afu);
837 
838 	kfree(afu->native);
839 	kfree(afu);
840 }
841 
842 /* Expects AFU struct to have recently been zeroed out */
843 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
844 {
845 	u64 val;
846 
847 	val = AFUD_READ_INFO(afu);
848 	afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
849 	afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
850 	afu->crs_num = AFUD_NUM_CRS(val);
851 
852 	if (AFUD_AFU_DIRECTED(val))
853 		afu->modes_supported |= CXL_MODE_DIRECTED;
854 	if (AFUD_DEDICATED_PROCESS(val))
855 		afu->modes_supported |= CXL_MODE_DEDICATED;
856 	if (AFUD_TIME_SLICED(val))
857 		afu->modes_supported |= CXL_MODE_TIME_SLICED;
858 
859 	val = AFUD_READ_PPPSA(afu);
860 	afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
861 	afu->psa = AFUD_PPPSA_PSA(val);
862 	if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
863 		afu->native->pp_offset = AFUD_READ_PPPSA_OFF(afu);
864 
865 	val = AFUD_READ_CR(afu);
866 	afu->crs_len = AFUD_CR_LEN(val) * 256;
867 	afu->crs_offset = AFUD_READ_CR_OFF(afu);
868 
869 
870 	/* eb_len is in multiple of 4K */
871 	afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
872 	afu->eb_offset = AFUD_READ_EB_OFF(afu);
873 
874 	/* eb_off is 4K aligned so lower 12 bits are always zero */
875 	if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
876 		dev_warn(&afu->dev,
877 			 "Invalid AFU error buffer offset %Lx\n",
878 			 afu->eb_offset);
879 		dev_info(&afu->dev,
880 			 "Ignoring AFU error buffer in the descriptor\n");
881 		/* indicate that no afu buffer exists */
882 		afu->eb_len = 0;
883 	}
884 
885 	return 0;
886 }
887 
888 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
889 {
890 	int i, rc;
891 	u32 val;
892 
893 	if (afu->psa && afu->adapter->ps_size <
894 			(afu->native->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
895 		dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
896 		return -ENODEV;
897 	}
898 
899 	if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
900 		dev_warn(&afu->dev, "AFU uses pp_size(%#016llx) < PAGE_SIZE per-process PSA!\n", afu->pp_size);
901 
902 	for (i = 0; i < afu->crs_num; i++) {
903 		rc = cxl_ops->afu_cr_read32(afu, i, 0, &val);
904 		if (rc || val == 0) {
905 			dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
906 			return -EINVAL;
907 		}
908 	}
909 
910 	if ((afu->modes_supported & ~CXL_MODE_DEDICATED) && afu->max_procs_virtualised == 0) {
911 		/*
912 		 * We could also check this for the dedicated process model
913 		 * since the architecture indicates it should be set to 1, but
914 		 * in that case we ignore the value and I'd rather not risk
915 		 * breaking any existing dedicated process AFUs that left it as
916 		 * 0 (not that I'm aware of any). It is clearly an error for an
917 		 * AFU directed AFU to set this to 0, and would have previously
918 		 * triggered a bug resulting in the maximum not being enforced
919 		 * at all since idr_alloc treats 0 as no maximum.
920 		 */
921 		dev_err(&afu->dev, "AFU does not support any processes\n");
922 		return -EINVAL;
923 	}
924 
925 	return 0;
926 }
927 
928 static int sanitise_afu_regs_psl9(struct cxl_afu *afu)
929 {
930 	u64 reg;
931 
932 	/*
933 	 * Clear out any regs that contain either an IVTE or address or may be
934 	 * waiting on an acknowledgment to try to be a bit safer as we bring
935 	 * it online
936 	 */
937 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
938 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
939 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
940 		if (cxl_ops->afu_reset(afu))
941 			return -EIO;
942 		if (cxl_afu_disable(afu))
943 			return -EIO;
944 		if (cxl_psl_purge(afu))
945 			return -EIO;
946 	}
947 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
948 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
949 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
950 	if (reg) {
951 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
952 		if (reg & CXL_PSL9_DSISR_An_TF)
953 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
954 		else
955 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
956 	}
957 	if (afu->adapter->native->sl_ops->register_serr_irq) {
958 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
959 		if (reg) {
960 			if (reg & ~0x000000007fffffff)
961 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
962 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
963 		}
964 	}
965 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
966 	if (reg) {
967 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
968 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
969 	}
970 
971 	return 0;
972 }
973 
974 static int sanitise_afu_regs_psl8(struct cxl_afu *afu)
975 {
976 	u64 reg;
977 
978 	/*
979 	 * Clear out any regs that contain either an IVTE or address or may be
980 	 * waiting on an acknowledgement to try to be a bit safer as we bring
981 	 * it online
982 	 */
983 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
984 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
985 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
986 		if (cxl_ops->afu_reset(afu))
987 			return -EIO;
988 		if (cxl_afu_disable(afu))
989 			return -EIO;
990 		if (cxl_psl_purge(afu))
991 			return -EIO;
992 	}
993 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
994 	cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
995 	cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
996 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
997 	cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
998 	cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
999 	cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
1000 	cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
1001 	cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
1002 	cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
1003 	cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
1004 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1005 	if (reg) {
1006 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1007 		if (reg & CXL_PSL_DSISR_TRANS)
1008 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1009 		else
1010 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1011 	}
1012 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1013 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1014 		if (reg) {
1015 			if (reg & ~0xffff)
1016 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1017 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1018 		}
1019 	}
1020 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1021 	if (reg) {
1022 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1023 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1024 	}
1025 
1026 	return 0;
1027 }
1028 
1029 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
1030 /*
1031  * afu_eb_read:
1032  * Called from sysfs and reads the afu error info buffer. The h/w only supports
1033  * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
1034  * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
1035  */
1036 ssize_t cxl_pci_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
1037 				loff_t off, size_t count)
1038 {
1039 	loff_t aligned_start, aligned_end;
1040 	size_t aligned_length;
1041 	void *tbuf;
1042 	const void __iomem *ebuf = afu->native->afu_desc_mmio + afu->eb_offset;
1043 
1044 	if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
1045 		return 0;
1046 
1047 	/* calculate aligned read window */
1048 	count = min((size_t)(afu->eb_len - off), count);
1049 	aligned_start = round_down(off, 8);
1050 	aligned_end = round_up(off + count, 8);
1051 	aligned_length = aligned_end - aligned_start;
1052 
1053 	/* max we can copy in one read is PAGE_SIZE */
1054 	if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
1055 		aligned_length = ERR_BUFF_MAX_COPY_SIZE;
1056 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
1057 	}
1058 
1059 	/* use bounce buffer for copy */
1060 	tbuf = (void *)__get_free_page(GFP_KERNEL);
1061 	if (!tbuf)
1062 		return -ENOMEM;
1063 
1064 	/* perform aligned read from the mmio region */
1065 	memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
1066 	memcpy(buf, tbuf + (off & 0x7), count);
1067 
1068 	free_page((unsigned long)tbuf);
1069 
1070 	return count;
1071 }
1072 
1073 static int pci_configure_afu(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1074 {
1075 	int rc;
1076 
1077 	if ((rc = pci_map_slice_regs(afu, adapter, dev)))
1078 		return rc;
1079 
1080 	if (adapter->native->sl_ops->sanitise_afu_regs) {
1081 		rc = adapter->native->sl_ops->sanitise_afu_regs(afu);
1082 		if (rc)
1083 			goto err1;
1084 	}
1085 
1086 	/* We need to reset the AFU before we can read the AFU descriptor */
1087 	if ((rc = cxl_ops->afu_reset(afu)))
1088 		goto err1;
1089 
1090 	if (cxl_verbose)
1091 		dump_afu_descriptor(afu);
1092 
1093 	if ((rc = cxl_read_afu_descriptor(afu)))
1094 		goto err1;
1095 
1096 	if ((rc = cxl_afu_descriptor_looks_ok(afu)))
1097 		goto err1;
1098 
1099 	if (adapter->native->sl_ops->afu_regs_init)
1100 		if ((rc = adapter->native->sl_ops->afu_regs_init(afu)))
1101 			goto err1;
1102 
1103 	if (adapter->native->sl_ops->register_serr_irq)
1104 		if ((rc = adapter->native->sl_ops->register_serr_irq(afu)))
1105 			goto err1;
1106 
1107 	if ((rc = cxl_native_register_psl_irq(afu)))
1108 		goto err2;
1109 
1110 	atomic_set(&afu->configured_state, 0);
1111 	return 0;
1112 
1113 err2:
1114 	if (adapter->native->sl_ops->release_serr_irq)
1115 		adapter->native->sl_ops->release_serr_irq(afu);
1116 err1:
1117 	pci_unmap_slice_regs(afu);
1118 	return rc;
1119 }
1120 
1121 static void pci_deconfigure_afu(struct cxl_afu *afu)
1122 {
1123 	/*
1124 	 * It's okay to deconfigure when AFU is already locked, otherwise wait
1125 	 * until there are no readers
1126 	 */
1127 	if (atomic_read(&afu->configured_state) != -1) {
1128 		while (atomic_cmpxchg(&afu->configured_state, 0, -1) != -1)
1129 			schedule();
1130 	}
1131 	cxl_native_release_psl_irq(afu);
1132 	if (afu->adapter->native->sl_ops->release_serr_irq)
1133 		afu->adapter->native->sl_ops->release_serr_irq(afu);
1134 	pci_unmap_slice_regs(afu);
1135 }
1136 
1137 static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
1138 {
1139 	struct cxl_afu *afu;
1140 	int rc = -ENOMEM;
1141 
1142 	afu = cxl_alloc_afu(adapter, slice);
1143 	if (!afu)
1144 		return -ENOMEM;
1145 
1146 	afu->native = kzalloc(sizeof(struct cxl_afu_native), GFP_KERNEL);
1147 	if (!afu->native)
1148 		goto err_free_afu;
1149 
1150 	mutex_init(&afu->native->spa_mutex);
1151 
1152 	rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice);
1153 	if (rc)
1154 		goto err_free_native;
1155 
1156 	rc = pci_configure_afu(afu, adapter, dev);
1157 	if (rc)
1158 		goto err_free_native;
1159 
1160 	/* Don't care if this fails */
1161 	cxl_debugfs_afu_add(afu);
1162 
1163 	/*
1164 	 * After we call this function we must not free the afu directly, even
1165 	 * if it returns an error!
1166 	 */
1167 	if ((rc = cxl_register_afu(afu)))
1168 		goto err_put_dev;
1169 
1170 	if ((rc = cxl_sysfs_afu_add(afu)))
1171 		goto err_del_dev;
1172 
1173 	adapter->afu[afu->slice] = afu;
1174 
1175 	if ((rc = cxl_pci_vphb_add(afu)))
1176 		dev_info(&afu->dev, "Can't register vPHB\n");
1177 
1178 	return 0;
1179 
1180 err_del_dev:
1181 	device_del(&afu->dev);
1182 err_put_dev:
1183 	pci_deconfigure_afu(afu);
1184 	cxl_debugfs_afu_remove(afu);
1185 	put_device(&afu->dev);
1186 	return rc;
1187 
1188 err_free_native:
1189 	kfree(afu->native);
1190 err_free_afu:
1191 	kfree(afu);
1192 	return rc;
1193 
1194 }
1195 
1196 static void cxl_pci_remove_afu(struct cxl_afu *afu)
1197 {
1198 	pr_devel("%s\n", __func__);
1199 
1200 	if (!afu)
1201 		return;
1202 
1203 	cxl_pci_vphb_remove(afu);
1204 	cxl_sysfs_afu_remove(afu);
1205 	cxl_debugfs_afu_remove(afu);
1206 
1207 	spin_lock(&afu->adapter->afu_list_lock);
1208 	afu->adapter->afu[afu->slice] = NULL;
1209 	spin_unlock(&afu->adapter->afu_list_lock);
1210 
1211 	cxl_context_detach_all(afu);
1212 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1213 
1214 	pci_deconfigure_afu(afu);
1215 	device_unregister(&afu->dev);
1216 }
1217 
1218 int cxl_pci_reset(struct cxl *adapter)
1219 {
1220 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1221 	int rc;
1222 
1223 	if (adapter->perst_same_image) {
1224 		dev_warn(&dev->dev,
1225 			 "cxl: refusing to reset/reflash when perst_reloads_same_image is set.\n");
1226 		return -EINVAL;
1227 	}
1228 
1229 	dev_info(&dev->dev, "CXL reset\n");
1230 
1231 	/*
1232 	 * The adapter is about to be reset, so ignore errors.
1233 	 */
1234 	cxl_data_cache_flush(adapter);
1235 
1236 	/* pcie_warm_reset requests a fundamental pci reset which includes a
1237 	 * PERST assert/deassert.  PERST triggers a loading of the image
1238 	 * if "user" or "factory" is selected in sysfs */
1239 	if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
1240 		dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
1241 		return rc;
1242 	}
1243 
1244 	return rc;
1245 }
1246 
1247 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
1248 {
1249 	if (pci_request_region(dev, 2, "priv 2 regs"))
1250 		goto err1;
1251 	if (pci_request_region(dev, 0, "priv 1 regs"))
1252 		goto err2;
1253 
1254 	pr_devel("cxl_map_adapter_regs: p1: %#016llx %#llx, p2: %#016llx %#llx",
1255 			p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
1256 
1257 	if (!(adapter->native->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
1258 		goto err3;
1259 
1260 	if (!(adapter->native->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
1261 		goto err4;
1262 
1263 	return 0;
1264 
1265 err4:
1266 	iounmap(adapter->native->p1_mmio);
1267 	adapter->native->p1_mmio = NULL;
1268 err3:
1269 	pci_release_region(dev, 0);
1270 err2:
1271 	pci_release_region(dev, 2);
1272 err1:
1273 	return -ENOMEM;
1274 }
1275 
1276 static void cxl_unmap_adapter_regs(struct cxl *adapter)
1277 {
1278 	if (adapter->native->p1_mmio) {
1279 		iounmap(adapter->native->p1_mmio);
1280 		adapter->native->p1_mmio = NULL;
1281 		pci_release_region(to_pci_dev(adapter->dev.parent), 2);
1282 	}
1283 	if (adapter->native->p2_mmio) {
1284 		iounmap(adapter->native->p2_mmio);
1285 		adapter->native->p2_mmio = NULL;
1286 		pci_release_region(to_pci_dev(adapter->dev.parent), 0);
1287 	}
1288 }
1289 
1290 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
1291 {
1292 	int vsec;
1293 	u32 afu_desc_off, afu_desc_size;
1294 	u32 ps_off, ps_size;
1295 	u16 vseclen;
1296 	u8 image_state;
1297 
1298 	if (!(vsec = find_cxl_vsec(dev))) {
1299 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
1300 		return -ENODEV;
1301 	}
1302 
1303 	CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
1304 	if (vseclen < CXL_VSEC_MIN_SIZE) {
1305 		dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
1306 		return -EINVAL;
1307 	}
1308 
1309 	CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
1310 	CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
1311 	CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
1312 	CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
1313 	CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
1314 	CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
1315 	adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1316 	adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1317 	adapter->perst_loads_image = !!(image_state & CXL_VSEC_PERST_LOADS_IMAGE);
1318 
1319 	CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
1320 	CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
1321 	CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
1322 	CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
1323 	CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
1324 
1325 	/* Convert everything to bytes, because there is NO WAY I'd look at the
1326 	 * code a month later and forget what units these are in ;-) */
1327 	adapter->native->ps_off = ps_off * 64 * 1024;
1328 	adapter->ps_size = ps_size * 64 * 1024;
1329 	adapter->native->afu_desc_off = afu_desc_off * 64 * 1024;
1330 	adapter->native->afu_desc_size = afu_desc_size * 64 * 1024;
1331 
1332 	/* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
1333 	adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
1334 
1335 	return 0;
1336 }
1337 
1338 /*
1339  * Workaround a PCIe Host Bridge defect on some cards, that can cause
1340  * malformed Transaction Layer Packet (TLP) errors to be erroneously
1341  * reported. Mask this error in the Uncorrectable Error Mask Register.
1342  *
1343  * The upper nibble of the PSL revision is used to distinguish between
1344  * different cards. The affected ones have it set to 0.
1345  */
1346 static void cxl_fixup_malformed_tlp(struct cxl *adapter, struct pci_dev *dev)
1347 {
1348 	int aer;
1349 	u32 data;
1350 
1351 	if (adapter->psl_rev & 0xf000)
1352 		return;
1353 	if (!(aer = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR)))
1354 		return;
1355 	pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &data);
1356 	if (data & PCI_ERR_UNC_MALF_TLP)
1357 		if (data & PCI_ERR_UNC_INTN)
1358 			return;
1359 	data |= PCI_ERR_UNC_MALF_TLP;
1360 	data |= PCI_ERR_UNC_INTN;
1361 	pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, data);
1362 }
1363 
1364 static bool cxl_compatible_caia_version(struct cxl *adapter)
1365 {
1366 	if (cxl_is_power8() && (adapter->caia_major == 1))
1367 		return true;
1368 
1369 	if (cxl_is_power9() && (adapter->caia_major == 2))
1370 		return true;
1371 
1372 	return false;
1373 }
1374 
1375 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
1376 {
1377 	if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
1378 		return -EBUSY;
1379 
1380 	if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
1381 		dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
1382 		return -EINVAL;
1383 	}
1384 
1385 	if (!cxl_compatible_caia_version(adapter)) {
1386 		dev_info(&dev->dev, "Ignoring card. PSL type is not supported (caia version: %d)\n",
1387 			 adapter->caia_major);
1388 		return -ENODEV;
1389 	}
1390 
1391 	if (!adapter->slices) {
1392 		/* Once we support dynamic reprogramming we can use the card if
1393 		 * it supports loadable AFUs */
1394 		dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
1395 		return -EINVAL;
1396 	}
1397 
1398 	if (!adapter->native->afu_desc_off || !adapter->native->afu_desc_size) {
1399 		dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
1400 		return -EINVAL;
1401 	}
1402 
1403 	if (adapter->ps_size > p2_size(dev) - adapter->native->ps_off) {
1404 		dev_err(&dev->dev, "ABORTING: Problem state size larger than "
1405 				   "available in BAR2: 0x%llx > 0x%llx\n",
1406 			 adapter->ps_size, p2_size(dev) - adapter->native->ps_off);
1407 		return -EINVAL;
1408 	}
1409 
1410 	return 0;
1411 }
1412 
1413 ssize_t cxl_pci_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1414 {
1415 	return pci_read_vpd(to_pci_dev(adapter->dev.parent), 0, len, buf);
1416 }
1417 
1418 static void cxl_release_adapter(struct device *dev)
1419 {
1420 	struct cxl *adapter = to_cxl_adapter(dev);
1421 
1422 	pr_devel("cxl_release_adapter\n");
1423 
1424 	cxl_remove_adapter_nr(adapter);
1425 
1426 	kfree(adapter->native);
1427 	kfree(adapter);
1428 }
1429 
1430 #define CXL_PSL_ErrIVTE_tberror (0x1ull << (63-31))
1431 
1432 static int sanitise_adapter_regs(struct cxl *adapter)
1433 {
1434 	int rc = 0;
1435 
1436 	/* Clear PSL tberror bit by writing 1 to it */
1437 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, CXL_PSL_ErrIVTE_tberror);
1438 
1439 	if (adapter->native->sl_ops->invalidate_all) {
1440 		/* do not invalidate ERAT entries when not reloading on PERST */
1441 		if (cxl_is_power9() && (adapter->perst_loads_image))
1442 			return 0;
1443 		rc = adapter->native->sl_ops->invalidate_all(adapter);
1444 	}
1445 
1446 	return rc;
1447 }
1448 
1449 /* This should contain *only* operations that can safely be done in
1450  * both creation and recovery.
1451  */
1452 static int cxl_configure_adapter(struct cxl *adapter, struct pci_dev *dev)
1453 {
1454 	int rc;
1455 
1456 	adapter->dev.parent = &dev->dev;
1457 	adapter->dev.release = cxl_release_adapter;
1458 	pci_set_drvdata(dev, adapter);
1459 
1460 	rc = pci_enable_device(dev);
1461 	if (rc) {
1462 		dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1463 		return rc;
1464 	}
1465 
1466 	if ((rc = cxl_read_vsec(adapter, dev)))
1467 		return rc;
1468 
1469 	if ((rc = cxl_vsec_looks_ok(adapter, dev)))
1470 	        return rc;
1471 
1472 	cxl_fixup_malformed_tlp(adapter, dev);
1473 
1474 	if ((rc = setup_cxl_bars(dev)))
1475 		return rc;
1476 
1477 	if ((rc = switch_card_to_cxl(dev)))
1478 		return rc;
1479 
1480 	if ((rc = cxl_update_image_control(adapter)))
1481 		return rc;
1482 
1483 	if ((rc = cxl_map_adapter_regs(adapter, dev)))
1484 		return rc;
1485 
1486 	if ((rc = sanitise_adapter_regs(adapter)))
1487 		goto err;
1488 
1489 	if ((rc = adapter->native->sl_ops->adapter_regs_init(adapter, dev)))
1490 		goto err;
1491 
1492 	/* Required for devices using CAPP DMA mode, harmless for others */
1493 	pci_set_master(dev);
1494 
1495 	adapter->tunneled_ops_supported = false;
1496 
1497 	if (cxl_is_power9()) {
1498 		if (pnv_pci_set_tunnel_bar(dev, 0x00020000E0000000ull, 1))
1499 			dev_info(&dev->dev, "Tunneled operations unsupported\n");
1500 		else
1501 			adapter->tunneled_ops_supported = true;
1502 	}
1503 
1504 	if ((rc = pnv_phb_to_cxl_mode(dev, adapter->native->sl_ops->capi_mode)))
1505 		goto err;
1506 
1507 	/* If recovery happened, the last step is to turn on snooping.
1508 	 * In the non-recovery case this has no effect */
1509 	if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON)))
1510 		goto err;
1511 
1512 	/* Ignore error, adapter init is not dependant on timebase sync */
1513 	cxl_setup_psl_timebase(adapter, dev);
1514 
1515 	if ((rc = cxl_native_register_psl_err_irq(adapter)))
1516 		goto err;
1517 
1518 	return 0;
1519 
1520 err:
1521 	cxl_unmap_adapter_regs(adapter);
1522 	return rc;
1523 
1524 }
1525 
1526 static void cxl_deconfigure_adapter(struct cxl *adapter)
1527 {
1528 	struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1529 
1530 	if (cxl_is_power9())
1531 		pnv_pci_set_tunnel_bar(pdev, 0x00020000E0000000ull, 0);
1532 
1533 	cxl_native_release_psl_err_irq(adapter);
1534 	cxl_unmap_adapter_regs(adapter);
1535 
1536 	pci_disable_device(pdev);
1537 }
1538 
1539 static void cxl_stop_trace_psl9(struct cxl *adapter)
1540 {
1541 	int traceid;
1542 	u64 trace_state, trace_mask;
1543 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1544 
1545 	/* read each tracearray state and issue mmio to stop them is needed */
1546 	for (traceid = 0; traceid <= CXL_PSL9_TRACEID_MAX; ++traceid) {
1547 		trace_state = cxl_p1_read(adapter, CXL_PSL9_CTCCFG);
1548 		trace_mask = (0x3ULL << (62 - traceid * 2));
1549 		trace_state = (trace_state & trace_mask) >> (62 - traceid * 2);
1550 		dev_dbg(&dev->dev, "cxl: Traceid-%d trace_state=0x%0llX\n",
1551 			traceid, trace_state);
1552 
1553 		/* issue mmio if the trace array isn't in FIN state */
1554 		if (trace_state != CXL_PSL9_TRACESTATE_FIN)
1555 			cxl_p1_write(adapter, CXL_PSL9_TRACECFG,
1556 				     0x8400000000000000ULL | traceid);
1557 	}
1558 }
1559 
1560 static void cxl_stop_trace_psl8(struct cxl *adapter)
1561 {
1562 	int slice;
1563 
1564 	/* Stop the trace */
1565 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x8000000000000017LL);
1566 
1567 	/* Stop the slice traces */
1568 	spin_lock(&adapter->afu_list_lock);
1569 	for (slice = 0; slice < adapter->slices; slice++) {
1570 		if (adapter->afu[slice])
1571 			cxl_p1n_write(adapter->afu[slice], CXL_PSL_SLICE_TRACE,
1572 				      0x8000000000000000LL);
1573 	}
1574 	spin_unlock(&adapter->afu_list_lock);
1575 }
1576 
1577 static const struct cxl_service_layer_ops psl9_ops = {
1578 	.adapter_regs_init = init_implementation_adapter_regs_psl9,
1579 	.invalidate_all = cxl_invalidate_all_psl9,
1580 	.afu_regs_init = init_implementation_afu_regs_psl9,
1581 	.sanitise_afu_regs = sanitise_afu_regs_psl9,
1582 	.register_serr_irq = cxl_native_register_serr_irq,
1583 	.release_serr_irq = cxl_native_release_serr_irq,
1584 	.handle_interrupt = cxl_irq_psl9,
1585 	.fail_irq = cxl_fail_irq_psl,
1586 	.activate_dedicated_process = cxl_activate_dedicated_process_psl9,
1587 	.attach_afu_directed = cxl_attach_afu_directed_psl9,
1588 	.attach_dedicated_process = cxl_attach_dedicated_process_psl9,
1589 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl9,
1590 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl9,
1591 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl9,
1592 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl9,
1593 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl9,
1594 	.debugfs_stop_trace = cxl_stop_trace_psl9,
1595 	.timebase_read = timebase_read_psl9,
1596 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1597 	.needs_reset_before_disable = true,
1598 };
1599 
1600 static const struct cxl_service_layer_ops psl8_ops = {
1601 	.adapter_regs_init = init_implementation_adapter_regs_psl8,
1602 	.invalidate_all = cxl_invalidate_all_psl8,
1603 	.afu_regs_init = init_implementation_afu_regs_psl8,
1604 	.sanitise_afu_regs = sanitise_afu_regs_psl8,
1605 	.register_serr_irq = cxl_native_register_serr_irq,
1606 	.release_serr_irq = cxl_native_release_serr_irq,
1607 	.handle_interrupt = cxl_irq_psl8,
1608 	.fail_irq = cxl_fail_irq_psl,
1609 	.activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1610 	.attach_afu_directed = cxl_attach_afu_directed_psl8,
1611 	.attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1612 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1613 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl8,
1614 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl8,
1615 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl8,
1616 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl8,
1617 	.debugfs_stop_trace = cxl_stop_trace_psl8,
1618 	.write_timebase_ctrl = write_timebase_ctrl_psl8,
1619 	.timebase_read = timebase_read_psl8,
1620 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1621 	.needs_reset_before_disable = true,
1622 };
1623 
1624 static void set_sl_ops(struct cxl *adapter, struct pci_dev *dev)
1625 {
1626 	if (cxl_is_power8()) {
1627 		dev_info(&dev->dev, "Device uses a PSL8\n");
1628 		adapter->native->sl_ops = &psl8_ops;
1629 	} else {
1630 		dev_info(&dev->dev, "Device uses a PSL9\n");
1631 		adapter->native->sl_ops = &psl9_ops;
1632 	}
1633 }
1634 
1635 
1636 static struct cxl *cxl_pci_init_adapter(struct pci_dev *dev)
1637 {
1638 	struct cxl *adapter;
1639 	int rc;
1640 
1641 	adapter = cxl_alloc_adapter();
1642 	if (!adapter)
1643 		return ERR_PTR(-ENOMEM);
1644 
1645 	adapter->native = kzalloc(sizeof(struct cxl_native), GFP_KERNEL);
1646 	if (!adapter->native) {
1647 		rc = -ENOMEM;
1648 		goto err_release;
1649 	}
1650 
1651 	set_sl_ops(adapter, dev);
1652 
1653 	/* Set defaults for parameters which need to persist over
1654 	 * configure/reconfigure
1655 	 */
1656 	adapter->perst_loads_image = true;
1657 	adapter->perst_same_image = false;
1658 
1659 	rc = cxl_configure_adapter(adapter, dev);
1660 	if (rc) {
1661 		pci_disable_device(dev);
1662 		goto err_release;
1663 	}
1664 
1665 	/* Don't care if this one fails: */
1666 	cxl_debugfs_adapter_add(adapter);
1667 
1668 	/*
1669 	 * After we call this function we must not free the adapter directly,
1670 	 * even if it returns an error!
1671 	 */
1672 	if ((rc = cxl_register_adapter(adapter)))
1673 		goto err_put_dev;
1674 
1675 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1676 		goto err_del_dev;
1677 
1678 	/* Release the context lock as adapter is configured */
1679 	cxl_adapter_context_unlock(adapter);
1680 
1681 	return adapter;
1682 
1683 err_del_dev:
1684 	device_del(&adapter->dev);
1685 err_put_dev:
1686 	/* This should mirror cxl_remove_adapter, except without the
1687 	 * sysfs parts
1688 	 */
1689 	cxl_debugfs_adapter_remove(adapter);
1690 	cxl_deconfigure_adapter(adapter);
1691 	put_device(&adapter->dev);
1692 	return ERR_PTR(rc);
1693 
1694 err_release:
1695 	cxl_release_adapter(&adapter->dev);
1696 	return ERR_PTR(rc);
1697 }
1698 
1699 static void cxl_pci_remove_adapter(struct cxl *adapter)
1700 {
1701 	pr_devel("cxl_remove_adapter\n");
1702 
1703 	cxl_sysfs_adapter_remove(adapter);
1704 	cxl_debugfs_adapter_remove(adapter);
1705 
1706 	/*
1707 	 * Flush adapter datacache as its about to be removed.
1708 	 */
1709 	cxl_data_cache_flush(adapter);
1710 
1711 	cxl_deconfigure_adapter(adapter);
1712 
1713 	device_unregister(&adapter->dev);
1714 }
1715 
1716 #define CXL_MAX_PCIEX_PARENT 2
1717 
1718 int cxl_slot_is_switched(struct pci_dev *dev)
1719 {
1720 	struct device_node *np;
1721 	int depth = 0;
1722 
1723 	if (!(np = pci_device_to_OF_node(dev))) {
1724 		pr_err("cxl: np = NULL\n");
1725 		return -ENODEV;
1726 	}
1727 	of_node_get(np);
1728 	while (np) {
1729 		np = of_get_next_parent(np);
1730 		if (!of_node_is_type(np, "pciex"))
1731 			break;
1732 		depth++;
1733 	}
1734 	of_node_put(np);
1735 	return (depth > CXL_MAX_PCIEX_PARENT);
1736 }
1737 
1738 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
1739 {
1740 	struct cxl *adapter;
1741 	int slice;
1742 	int rc;
1743 
1744 	if (cxl_pci_is_vphb_device(dev)) {
1745 		dev_dbg(&dev->dev, "cxl_init_adapter: Ignoring cxl vphb device\n");
1746 		return -ENODEV;
1747 	}
1748 
1749 	if (cxl_slot_is_switched(dev)) {
1750 		dev_info(&dev->dev, "Ignoring card on incompatible PCI slot\n");
1751 		return -ENODEV;
1752 	}
1753 
1754 	if (cxl_is_power9() && !radix_enabled()) {
1755 		dev_info(&dev->dev, "Only Radix mode supported\n");
1756 		return -ENODEV;
1757 	}
1758 
1759 	if (cxl_verbose)
1760 		dump_cxl_config_space(dev);
1761 
1762 	adapter = cxl_pci_init_adapter(dev);
1763 	if (IS_ERR(adapter)) {
1764 		dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
1765 		return PTR_ERR(adapter);
1766 	}
1767 
1768 	for (slice = 0; slice < adapter->slices; slice++) {
1769 		if ((rc = pci_init_afu(adapter, slice, dev))) {
1770 			dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
1771 			continue;
1772 		}
1773 
1774 		rc = cxl_afu_select_best_mode(adapter->afu[slice]);
1775 		if (rc)
1776 			dev_err(&dev->dev, "AFU %i failed to start: %i\n", slice, rc);
1777 	}
1778 
1779 	return 0;
1780 }
1781 
1782 static void cxl_remove(struct pci_dev *dev)
1783 {
1784 	struct cxl *adapter = pci_get_drvdata(dev);
1785 	struct cxl_afu *afu;
1786 	int i;
1787 
1788 	/*
1789 	 * Lock to prevent someone grabbing a ref through the adapter list as
1790 	 * we are removing it
1791 	 */
1792 	for (i = 0; i < adapter->slices; i++) {
1793 		afu = adapter->afu[i];
1794 		cxl_pci_remove_afu(afu);
1795 	}
1796 	cxl_pci_remove_adapter(adapter);
1797 }
1798 
1799 static pci_ers_result_t cxl_vphb_error_detected(struct cxl_afu *afu,
1800 						pci_channel_state_t state)
1801 {
1802 	struct pci_dev *afu_dev;
1803 	struct pci_driver *afu_drv;
1804 	const struct pci_error_handlers *err_handler;
1805 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
1806 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
1807 
1808 	/* There should only be one entry, but go through the list
1809 	 * anyway
1810 	 */
1811 	if (afu == NULL || afu->phb == NULL)
1812 		return result;
1813 
1814 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
1815 		afu_drv = to_pci_driver(afu_dev->dev.driver);
1816 		if (!afu_drv)
1817 			continue;
1818 
1819 		afu_dev->error_state = state;
1820 
1821 		err_handler = afu_drv->err_handler;
1822 		if (err_handler)
1823 			afu_result = err_handler->error_detected(afu_dev,
1824 								 state);
1825 		/* Disconnect trumps all, NONE trumps NEED_RESET */
1826 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
1827 			result = PCI_ERS_RESULT_DISCONNECT;
1828 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
1829 			 (result == PCI_ERS_RESULT_NEED_RESET))
1830 			result = PCI_ERS_RESULT_NONE;
1831 	}
1832 	return result;
1833 }
1834 
1835 static pci_ers_result_t cxl_pci_error_detected(struct pci_dev *pdev,
1836 					       pci_channel_state_t state)
1837 {
1838 	struct cxl *adapter = pci_get_drvdata(pdev);
1839 	struct cxl_afu *afu;
1840 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
1841 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
1842 	int i;
1843 
1844 	/* At this point, we could still have an interrupt pending.
1845 	 * Let's try to get them out of the way before they do
1846 	 * anything we don't like.
1847 	 */
1848 	schedule();
1849 
1850 	/* If we're permanently dead, give up. */
1851 	if (state == pci_channel_io_perm_failure) {
1852 		spin_lock(&adapter->afu_list_lock);
1853 		for (i = 0; i < adapter->slices; i++) {
1854 			afu = adapter->afu[i];
1855 			/*
1856 			 * Tell the AFU drivers; but we don't care what they
1857 			 * say, we're going away.
1858 			 */
1859 			cxl_vphb_error_detected(afu, state);
1860 		}
1861 		spin_unlock(&adapter->afu_list_lock);
1862 		return PCI_ERS_RESULT_DISCONNECT;
1863 	}
1864 
1865 	/* Are we reflashing?
1866 	 *
1867 	 * If we reflash, we could come back as something entirely
1868 	 * different, including a non-CAPI card. As such, by default
1869 	 * we don't participate in the process. We'll be unbound and
1870 	 * the slot re-probed. (TODO: check EEH doesn't blindly rebind
1871 	 * us!)
1872 	 *
1873 	 * However, this isn't the entire story: for reliablity
1874 	 * reasons, we usually want to reflash the FPGA on PERST in
1875 	 * order to get back to a more reliable known-good state.
1876 	 *
1877 	 * This causes us a bit of a problem: if we reflash we can't
1878 	 * trust that we'll come back the same - we could have a new
1879 	 * image and been PERSTed in order to load that
1880 	 * image. However, most of the time we actually *will* come
1881 	 * back the same - for example a regular EEH event.
1882 	 *
1883 	 * Therefore, we allow the user to assert that the image is
1884 	 * indeed the same and that we should continue on into EEH
1885 	 * anyway.
1886 	 */
1887 	if (adapter->perst_loads_image && !adapter->perst_same_image) {
1888 		/* TODO take the PHB out of CXL mode */
1889 		dev_info(&pdev->dev, "reflashing, so opting out of EEH!\n");
1890 		return PCI_ERS_RESULT_NONE;
1891 	}
1892 
1893 	/*
1894 	 * At this point, we want to try to recover.  We'll always
1895 	 * need a complete slot reset: we don't trust any other reset.
1896 	 *
1897 	 * Now, we go through each AFU:
1898 	 *  - We send the driver, if bound, an error_detected callback.
1899 	 *    We expect it to clean up, but it can also tell us to give
1900 	 *    up and permanently detach the card. To simplify things, if
1901 	 *    any bound AFU driver doesn't support EEH, we give up on EEH.
1902 	 *
1903 	 *  - We detach all contexts associated with the AFU. This
1904 	 *    does not free them, but puts them into a CLOSED state
1905 	 *    which causes any the associated files to return useful
1906 	 *    errors to userland. It also unmaps, but does not free,
1907 	 *    any IRQs.
1908 	 *
1909 	 *  - We clean up our side: releasing and unmapping resources we hold
1910 	 *    so we can wire them up again when the hardware comes back up.
1911 	 *
1912 	 * Driver authors should note:
1913 	 *
1914 	 *  - Any contexts you create in your kernel driver (except
1915 	 *    those associated with anonymous file descriptors) are
1916 	 *    your responsibility to free and recreate. Likewise with
1917 	 *    any attached resources.
1918 	 *
1919 	 *  - We will take responsibility for re-initialising the
1920 	 *    device context (the one set up for you in
1921 	 *    cxl_pci_enable_device_hook and accessed through
1922 	 *    cxl_get_context). If you've attached IRQs or other
1923 	 *    resources to it, they remains yours to free.
1924 	 *
1925 	 * You can call the same functions to release resources as you
1926 	 * normally would: we make sure that these functions continue
1927 	 * to work when the hardware is down.
1928 	 *
1929 	 * Two examples:
1930 	 *
1931 	 * 1) If you normally free all your resources at the end of
1932 	 *    each request, or if you use anonymous FDs, your
1933 	 *    error_detected callback can simply set a flag to tell
1934 	 *    your driver not to start any new calls. You can then
1935 	 *    clear the flag in the resume callback.
1936 	 *
1937 	 * 2) If you normally allocate your resources on startup:
1938 	 *     * Set a flag in error_detected as above.
1939 	 *     * Let CXL detach your contexts.
1940 	 *     * In slot_reset, free the old resources and allocate new ones.
1941 	 *     * In resume, clear the flag to allow things to start.
1942 	 */
1943 
1944 	/* Make sure no one else changes the afu list */
1945 	spin_lock(&adapter->afu_list_lock);
1946 
1947 	for (i = 0; i < adapter->slices; i++) {
1948 		afu = adapter->afu[i];
1949 
1950 		if (afu == NULL)
1951 			continue;
1952 
1953 		afu_result = cxl_vphb_error_detected(afu, state);
1954 		cxl_context_detach_all(afu);
1955 		cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1956 		pci_deconfigure_afu(afu);
1957 
1958 		/* Disconnect trumps all, NONE trumps NEED_RESET */
1959 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
1960 			result = PCI_ERS_RESULT_DISCONNECT;
1961 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
1962 			 (result == PCI_ERS_RESULT_NEED_RESET))
1963 			result = PCI_ERS_RESULT_NONE;
1964 	}
1965 	spin_unlock(&adapter->afu_list_lock);
1966 
1967 	/* should take the context lock here */
1968 	if (cxl_adapter_context_lock(adapter) != 0)
1969 		dev_warn(&adapter->dev,
1970 			 "Couldn't take context lock with %d active-contexts\n",
1971 			 atomic_read(&adapter->contexts_num));
1972 
1973 	cxl_deconfigure_adapter(adapter);
1974 
1975 	return result;
1976 }
1977 
1978 static pci_ers_result_t cxl_pci_slot_reset(struct pci_dev *pdev)
1979 {
1980 	struct cxl *adapter = pci_get_drvdata(pdev);
1981 	struct cxl_afu *afu;
1982 	struct cxl_context *ctx;
1983 	struct pci_dev *afu_dev;
1984 	struct pci_driver *afu_drv;
1985 	const struct pci_error_handlers *err_handler;
1986 	pci_ers_result_t afu_result = PCI_ERS_RESULT_RECOVERED;
1987 	pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
1988 	int i;
1989 
1990 	if (cxl_configure_adapter(adapter, pdev))
1991 		goto err;
1992 
1993 	/*
1994 	 * Unlock context activation for the adapter. Ideally this should be
1995 	 * done in cxl_pci_resume but cxlflash module tries to activate the
1996 	 * master context as part of slot_reset callback.
1997 	 */
1998 	cxl_adapter_context_unlock(adapter);
1999 
2000 	spin_lock(&adapter->afu_list_lock);
2001 	for (i = 0; i < adapter->slices; i++) {
2002 		afu = adapter->afu[i];
2003 
2004 		if (afu == NULL)
2005 			continue;
2006 
2007 		if (pci_configure_afu(afu, adapter, pdev))
2008 			goto err_unlock;
2009 
2010 		if (cxl_afu_select_best_mode(afu))
2011 			goto err_unlock;
2012 
2013 		if (afu->phb == NULL)
2014 			continue;
2015 
2016 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2017 			/* Reset the device context.
2018 			 * TODO: make this less disruptive
2019 			 */
2020 			ctx = cxl_get_context(afu_dev);
2021 
2022 			if (ctx && cxl_release_context(ctx))
2023 				goto err_unlock;
2024 
2025 			ctx = cxl_dev_context_init(afu_dev);
2026 			if (IS_ERR(ctx))
2027 				goto err_unlock;
2028 
2029 			afu_dev->dev.archdata.cxl_ctx = ctx;
2030 
2031 			if (cxl_ops->afu_check_and_enable(afu))
2032 				goto err_unlock;
2033 
2034 			afu_dev->error_state = pci_channel_io_normal;
2035 
2036 			/* If there's a driver attached, allow it to
2037 			 * chime in on recovery. Drivers should check
2038 			 * if everything has come back OK, but
2039 			 * shouldn't start new work until we call
2040 			 * their resume function.
2041 			 */
2042 			afu_drv = to_pci_driver(afu_dev->dev.driver);
2043 			if (!afu_drv)
2044 				continue;
2045 
2046 			err_handler = afu_drv->err_handler;
2047 			if (err_handler && err_handler->slot_reset)
2048 				afu_result = err_handler->slot_reset(afu_dev);
2049 
2050 			if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2051 				result = PCI_ERS_RESULT_DISCONNECT;
2052 		}
2053 	}
2054 
2055 	spin_unlock(&adapter->afu_list_lock);
2056 	return result;
2057 
2058 err_unlock:
2059 	spin_unlock(&adapter->afu_list_lock);
2060 
2061 err:
2062 	/* All the bits that happen in both error_detected and cxl_remove
2063 	 * should be idempotent, so we don't need to worry about leaving a mix
2064 	 * of unconfigured and reconfigured resources.
2065 	 */
2066 	dev_err(&pdev->dev, "EEH recovery failed. Asking to be disconnected.\n");
2067 	return PCI_ERS_RESULT_DISCONNECT;
2068 }
2069 
2070 static void cxl_pci_resume(struct pci_dev *pdev)
2071 {
2072 	struct cxl *adapter = pci_get_drvdata(pdev);
2073 	struct cxl_afu *afu;
2074 	struct pci_dev *afu_dev;
2075 	struct pci_driver *afu_drv;
2076 	const struct pci_error_handlers *err_handler;
2077 	int i;
2078 
2079 	/* Everything is back now. Drivers should restart work now.
2080 	 * This is not the place to be checking if everything came back up
2081 	 * properly, because there's no return value: do that in slot_reset.
2082 	 */
2083 	spin_lock(&adapter->afu_list_lock);
2084 	for (i = 0; i < adapter->slices; i++) {
2085 		afu = adapter->afu[i];
2086 
2087 		if (afu == NULL || afu->phb == NULL)
2088 			continue;
2089 
2090 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2091 			afu_drv = to_pci_driver(afu_dev->dev.driver);
2092 			if (!afu_drv)
2093 				continue;
2094 
2095 			err_handler = afu_drv->err_handler;
2096 			if (err_handler && err_handler->resume)
2097 				err_handler->resume(afu_dev);
2098 		}
2099 	}
2100 	spin_unlock(&adapter->afu_list_lock);
2101 }
2102 
2103 static const struct pci_error_handlers cxl_err_handler = {
2104 	.error_detected = cxl_pci_error_detected,
2105 	.slot_reset = cxl_pci_slot_reset,
2106 	.resume = cxl_pci_resume,
2107 };
2108 
2109 struct pci_driver cxl_pci_driver = {
2110 	.name = "cxl-pci",
2111 	.id_table = cxl_pci_tbl,
2112 	.probe = cxl_probe,
2113 	.remove = cxl_remove,
2114 	.shutdown = cxl_remove,
2115 	.err_handler = &cxl_err_handler,
2116 };
2117