xref: /openbmc/linux/drivers/misc/cxl/pci.c (revision e40573a4)
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 		return rc;
391 	}
392 
393 	*capp_unit_id = get_capp_unit_id(np, *phb_index);
394 	of_node_put(np);
395 	if (!*capp_unit_id) {
396 		pr_err("cxl: No capp unit found for PHB[%lld,%d]. Make sure the adapter is on a capi-compatible slot\n",
397 		       *chipid, *phb_index);
398 		return -ENODEV;
399 	}
400 
401 	return 0;
402 }
403 
404 static DEFINE_MUTEX(indications_mutex);
405 
406 static int get_phb_indications(struct pci_dev *dev, u64 *capiind, u64 *asnind,
407 			       u64 *nbwind)
408 {
409 	static u64 nbw, asn, capi = 0;
410 	struct device_node *np;
411 	const __be32 *prop;
412 
413 	mutex_lock(&indications_mutex);
414 	if (!capi) {
415 		if (!(np = pnv_pci_get_phb_node(dev))) {
416 			mutex_unlock(&indications_mutex);
417 			return -ENODEV;
418 		}
419 
420 		prop = of_get_property(np, "ibm,phb-indications", NULL);
421 		if (!prop) {
422 			nbw = 0x0300UL; /* legacy values */
423 			asn = 0x0400UL;
424 			capi = 0x0200UL;
425 		} else {
426 			nbw = (u64)be32_to_cpu(prop[2]);
427 			asn = (u64)be32_to_cpu(prop[1]);
428 			capi = (u64)be32_to_cpu(prop[0]);
429 		}
430 		of_node_put(np);
431 	}
432 	*capiind = capi;
433 	*asnind = asn;
434 	*nbwind = nbw;
435 	mutex_unlock(&indications_mutex);
436 	return 0;
437 }
438 
439 int cxl_get_xsl9_dsnctl(struct pci_dev *dev, u64 capp_unit_id, u64 *reg)
440 {
441 	u64 xsl_dsnctl;
442 	u64 capiind, asnind, nbwind;
443 
444 	/*
445 	 * CAPI Identifier bits [0:7]
446 	 * bit 61:60 MSI bits --> 0
447 	 * bit 59 TVT selector --> 0
448 	 */
449 	if (get_phb_indications(dev, &capiind, &asnind, &nbwind))
450 		return -ENODEV;
451 
452 	/*
453 	 * Tell XSL where to route data to.
454 	 * The field chipid should match the PHB CAPI_CMPM register
455 	 */
456 	xsl_dsnctl = (capiind << (63-15)); /* Bit 57 */
457 	xsl_dsnctl |= (capp_unit_id << (63-15));
458 
459 	/* nMMU_ID Defaults to: b’000001001’*/
460 	xsl_dsnctl |= ((u64)0x09 << (63-28));
461 
462 	/*
463 	 * Used to identify CAPI packets which should be sorted into
464 	 * the Non-Blocking queues by the PHB. This field should match
465 	 * the PHB PBL_NBW_CMPM register
466 	 * nbwind=0x03, bits [57:58], must include capi indicator.
467 	 * Not supported on P9 DD1.
468 	 */
469 	xsl_dsnctl |= (nbwind << (63-55));
470 
471 	/*
472 	 * Upper 16b address bits of ASB_Notify messages sent to the
473 	 * system. Need to match the PHB’s ASN Compare/Mask Register.
474 	 * Not supported on P9 DD1.
475 	 */
476 	xsl_dsnctl |= asnind;
477 
478 	*reg = xsl_dsnctl;
479 	return 0;
480 }
481 
482 static int init_implementation_adapter_regs_psl9(struct cxl *adapter,
483 						 struct pci_dev *dev)
484 {
485 	u64 xsl_dsnctl, psl_fircntl;
486 	u64 chipid;
487 	u32 phb_index;
488 	u64 capp_unit_id;
489 	u64 psl_debug;
490 	int rc;
491 
492 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
493 	if (rc)
494 		return rc;
495 
496 	rc = cxl_get_xsl9_dsnctl(dev, capp_unit_id, &xsl_dsnctl);
497 	if (rc)
498 		return rc;
499 
500 	cxl_p1_write(adapter, CXL_XSL9_DSNCTL, xsl_dsnctl);
501 
502 	/* Set fir_cntl to recommended value for production env */
503 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
504 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
505 	psl_fircntl |= 0x1ULL; /* ce_thresh */
506 	cxl_p1_write(adapter, CXL_PSL9_FIR_CNTL, psl_fircntl);
507 
508 	/* Setup the PSL to transmit packets on the PCIe before the
509 	 * CAPP is enabled. Make sure that CAPP virtual machines are disabled
510 	 */
511 	cxl_p1_write(adapter, CXL_PSL9_DSNDCTL, 0x0001001000012A10ULL);
512 
513 	/*
514 	 * A response to an ASB_Notify request is returned by the
515 	 * system as an MMIO write to the address defined in
516 	 * the PSL_TNR_ADDR register.
517 	 * keep the Reset Value: 0x00020000E0000000
518 	 */
519 
520 	/* Enable XSL rty limit */
521 	cxl_p1_write(adapter, CXL_XSL9_DEF, 0x51F8000000000005ULL);
522 
523 	/* Change XSL_INV dummy read threshold */
524 	cxl_p1_write(adapter, CXL_XSL9_INV, 0x0000040007FFC200ULL);
525 
526 	if (phb_index == 3) {
527 		/* disable machines 31-47 and 20-27 for DMA */
528 		cxl_p1_write(adapter, CXL_PSL9_APCDEDTYPE, 0x40000FF3FFFF0000ULL);
529 	}
530 
531 	/* Snoop machines */
532 	cxl_p1_write(adapter, CXL_PSL9_APCDEDALLOC, 0x800F000200000000ULL);
533 
534 	/* Enable NORST and DD2 features */
535 	cxl_p1_write(adapter, CXL_PSL9_DEBUG, 0xC000000000000000ULL);
536 
537 	/*
538 	 * Check if PSL has data-cache. We need to flush adapter datacache
539 	 * when as its about to be removed.
540 	 */
541 	psl_debug = cxl_p1_read(adapter, CXL_PSL9_DEBUG);
542 	if (psl_debug & CXL_PSL_DEBUG_CDC) {
543 		dev_dbg(&dev->dev, "No data-cache present\n");
544 		adapter->native->no_data_cache = true;
545 	}
546 
547 	return 0;
548 }
549 
550 static int init_implementation_adapter_regs_psl8(struct cxl *adapter, struct pci_dev *dev)
551 {
552 	u64 psl_dsnctl, psl_fircntl;
553 	u64 chipid;
554 	u32 phb_index;
555 	u64 capp_unit_id;
556 	int rc;
557 
558 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
559 	if (rc)
560 		return rc;
561 
562 	psl_dsnctl = 0x0000900000000000ULL; /* pteupd ttype, scdone */
563 	psl_dsnctl |= (0x2ULL << (63-38)); /* MMIO hang pulse: 256 us */
564 	/* Tell PSL where to route data to */
565 	psl_dsnctl |= (chipid << (63-5));
566 	psl_dsnctl |= (capp_unit_id << (63-13));
567 
568 	cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
569 	cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
570 	/* snoop write mask */
571 	cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
572 	/* set fir_cntl to recommended value for production env */
573 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
574 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
575 	psl_fircntl |= 0x1ULL; /* ce_thresh */
576 	cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, psl_fircntl);
577 	/* for debugging with trace arrays */
578 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
579 
580 	return 0;
581 }
582 
583 /* PSL */
584 #define TBSYNC_CAL(n) (((u64)n & 0x7) << (63-3))
585 #define TBSYNC_CNT(n) (((u64)n & 0x7) << (63-6))
586 /* For the PSL this is a multiple for 0 < n <= 7: */
587 #define PSL_2048_250MHZ_CYCLES 1
588 
589 static void write_timebase_ctrl_psl8(struct cxl *adapter)
590 {
591 	cxl_p1_write(adapter, CXL_PSL_TB_CTLSTAT,
592 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
593 }
594 
595 static u64 timebase_read_psl9(struct cxl *adapter)
596 {
597 	return cxl_p1_read(adapter, CXL_PSL9_Timebase);
598 }
599 
600 static u64 timebase_read_psl8(struct cxl *adapter)
601 {
602 	return cxl_p1_read(adapter, CXL_PSL_Timebase);
603 }
604 
605 static void cxl_setup_psl_timebase(struct cxl *adapter, struct pci_dev *dev)
606 {
607 	struct device_node *np;
608 
609 	adapter->psl_timebase_synced = false;
610 
611 	if (!(np = pnv_pci_get_phb_node(dev)))
612 		return;
613 
614 	/* Do not fail when CAPP timebase sync is not supported by OPAL */
615 	of_node_get(np);
616 	if (! of_get_property(np, "ibm,capp-timebase-sync", NULL)) {
617 		of_node_put(np);
618 		dev_info(&dev->dev, "PSL timebase inactive: OPAL support missing\n");
619 		return;
620 	}
621 	of_node_put(np);
622 
623 	/*
624 	 * Setup PSL Timebase Control and Status register
625 	 * with the recommended Timebase Sync Count value
626 	 */
627 	if (adapter->native->sl_ops->write_timebase_ctrl)
628 		adapter->native->sl_ops->write_timebase_ctrl(adapter);
629 
630 	/* Enable PSL Timebase */
631 	cxl_p1_write(adapter, CXL_PSL_Control, 0x0000000000000000);
632 	cxl_p1_write(adapter, CXL_PSL_Control, CXL_PSL_Control_tb);
633 
634 	return;
635 }
636 
637 static int init_implementation_afu_regs_psl9(struct cxl_afu *afu)
638 {
639 	return 0;
640 }
641 
642 static int init_implementation_afu_regs_psl8(struct cxl_afu *afu)
643 {
644 	/* read/write masks for this slice */
645 	cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
646 	/* APC read/write masks for this slice */
647 	cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
648 	/* for debugging with trace arrays */
649 	cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
650 	cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
651 
652 	return 0;
653 }
654 
655 int cxl_pci_setup_irq(struct cxl *adapter, unsigned int hwirq,
656 		unsigned int virq)
657 {
658 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
659 
660 	return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
661 }
662 
663 int cxl_update_image_control(struct cxl *adapter)
664 {
665 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
666 	int rc;
667 	int vsec;
668 	u8 image_state;
669 
670 	if (!(vsec = find_cxl_vsec(dev))) {
671 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
672 		return -ENODEV;
673 	}
674 
675 	if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
676 		dev_err(&dev->dev, "failed to read image state: %i\n", rc);
677 		return rc;
678 	}
679 
680 	if (adapter->perst_loads_image)
681 		image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
682 	else
683 		image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
684 
685 	if (adapter->perst_select_user)
686 		image_state |= CXL_VSEC_PERST_SELECT_USER;
687 	else
688 		image_state &= ~CXL_VSEC_PERST_SELECT_USER;
689 
690 	if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
691 		dev_err(&dev->dev, "failed to update image control: %i\n", rc);
692 		return rc;
693 	}
694 
695 	return 0;
696 }
697 
698 int cxl_pci_alloc_one_irq(struct cxl *adapter)
699 {
700 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
701 
702 	return pnv_cxl_alloc_hwirqs(dev, 1);
703 }
704 
705 void cxl_pci_release_one_irq(struct cxl *adapter, int hwirq)
706 {
707 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
708 
709 	return pnv_cxl_release_hwirqs(dev, hwirq, 1);
710 }
711 
712 int cxl_pci_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
713 			struct cxl *adapter, unsigned int num)
714 {
715 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
716 
717 	return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
718 }
719 
720 void cxl_pci_release_irq_ranges(struct cxl_irq_ranges *irqs,
721 				struct cxl *adapter)
722 {
723 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
724 
725 	pnv_cxl_release_hwirq_ranges(irqs, dev);
726 }
727 
728 static int setup_cxl_bars(struct pci_dev *dev)
729 {
730 	/* Safety check in case we get backported to < 3.17 without M64 */
731 	if ((p1_base(dev) < 0x100000000ULL) ||
732 	    (p2_base(dev) < 0x100000000ULL)) {
733 		dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
734 		return -ENODEV;
735 	}
736 
737 	/*
738 	 * BAR 4/5 has a special meaning for CXL and must be programmed with a
739 	 * special value corresponding to the CXL protocol address range.
740 	 * For POWER 8/9 that means bits 48:49 must be set to 10
741 	 */
742 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
743 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
744 
745 	return 0;
746 }
747 
748 /* pciex node: ibm,opal-m64-window = <0x3d058 0x0 0x3d058 0x0 0x8 0x0>; */
749 static int switch_card_to_cxl(struct pci_dev *dev)
750 {
751 	int vsec;
752 	u8 val;
753 	int rc;
754 
755 	dev_info(&dev->dev, "switch card to CXL\n");
756 
757 	if (!(vsec = find_cxl_vsec(dev))) {
758 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
759 		return -ENODEV;
760 	}
761 
762 	if ((rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val))) {
763 		dev_err(&dev->dev, "failed to read current mode control: %i", rc);
764 		return rc;
765 	}
766 	val &= ~CXL_VSEC_PROTOCOL_MASK;
767 	val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
768 	if ((rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val))) {
769 		dev_err(&dev->dev, "failed to enable CXL protocol: %i", rc);
770 		return rc;
771 	}
772 	/*
773 	 * The CAIA spec (v0.12 11.6 Bi-modal Device Support) states
774 	 * we must wait 100ms after this mode switch before touching
775 	 * PCIe config space.
776 	 */
777 	msleep(100);
778 
779 	return 0;
780 }
781 
782 static int pci_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
783 {
784 	u64 p1n_base, p2n_base, afu_desc;
785 	const u64 p1n_size = 0x100;
786 	const u64 p2n_size = 0x1000;
787 
788 	p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
789 	p2n_base = p2_base(dev) + (afu->slice * p2n_size);
790 	afu->psn_phys = p2_base(dev) + (adapter->native->ps_off + (afu->slice * adapter->ps_size));
791 	afu_desc = p2_base(dev) + adapter->native->afu_desc_off + (afu->slice * adapter->native->afu_desc_size);
792 
793 	if (!(afu->native->p1n_mmio = ioremap(p1n_base, p1n_size)))
794 		goto err;
795 	if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
796 		goto err1;
797 	if (afu_desc) {
798 		if (!(afu->native->afu_desc_mmio = ioremap(afu_desc, adapter->native->afu_desc_size)))
799 			goto err2;
800 	}
801 
802 	return 0;
803 err2:
804 	iounmap(afu->p2n_mmio);
805 err1:
806 	iounmap(afu->native->p1n_mmio);
807 err:
808 	dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
809 	return -ENOMEM;
810 }
811 
812 static void pci_unmap_slice_regs(struct cxl_afu *afu)
813 {
814 	if (afu->p2n_mmio) {
815 		iounmap(afu->p2n_mmio);
816 		afu->p2n_mmio = NULL;
817 	}
818 	if (afu->native->p1n_mmio) {
819 		iounmap(afu->native->p1n_mmio);
820 		afu->native->p1n_mmio = NULL;
821 	}
822 	if (afu->native->afu_desc_mmio) {
823 		iounmap(afu->native->afu_desc_mmio);
824 		afu->native->afu_desc_mmio = NULL;
825 	}
826 }
827 
828 void cxl_pci_release_afu(struct device *dev)
829 {
830 	struct cxl_afu *afu = to_cxl_afu(dev);
831 
832 	pr_devel("%s\n", __func__);
833 
834 	idr_destroy(&afu->contexts_idr);
835 	cxl_release_spa(afu);
836 
837 	kfree(afu->native);
838 	kfree(afu);
839 }
840 
841 /* Expects AFU struct to have recently been zeroed out */
842 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
843 {
844 	u64 val;
845 
846 	val = AFUD_READ_INFO(afu);
847 	afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
848 	afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
849 	afu->crs_num = AFUD_NUM_CRS(val);
850 
851 	if (AFUD_AFU_DIRECTED(val))
852 		afu->modes_supported |= CXL_MODE_DIRECTED;
853 	if (AFUD_DEDICATED_PROCESS(val))
854 		afu->modes_supported |= CXL_MODE_DEDICATED;
855 	if (AFUD_TIME_SLICED(val))
856 		afu->modes_supported |= CXL_MODE_TIME_SLICED;
857 
858 	val = AFUD_READ_PPPSA(afu);
859 	afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
860 	afu->psa = AFUD_PPPSA_PSA(val);
861 	if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
862 		afu->native->pp_offset = AFUD_READ_PPPSA_OFF(afu);
863 
864 	val = AFUD_READ_CR(afu);
865 	afu->crs_len = AFUD_CR_LEN(val) * 256;
866 	afu->crs_offset = AFUD_READ_CR_OFF(afu);
867 
868 
869 	/* eb_len is in multiple of 4K */
870 	afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
871 	afu->eb_offset = AFUD_READ_EB_OFF(afu);
872 
873 	/* eb_off is 4K aligned so lower 12 bits are always zero */
874 	if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
875 		dev_warn(&afu->dev,
876 			 "Invalid AFU error buffer offset %Lx\n",
877 			 afu->eb_offset);
878 		dev_info(&afu->dev,
879 			 "Ignoring AFU error buffer in the descriptor\n");
880 		/* indicate that no afu buffer exists */
881 		afu->eb_len = 0;
882 	}
883 
884 	return 0;
885 }
886 
887 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
888 {
889 	int i, rc;
890 	u32 val;
891 
892 	if (afu->psa && afu->adapter->ps_size <
893 			(afu->native->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
894 		dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
895 		return -ENODEV;
896 	}
897 
898 	if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
899 		dev_warn(&afu->dev, "AFU uses pp_size(%#016llx) < PAGE_SIZE per-process PSA!\n", afu->pp_size);
900 
901 	for (i = 0; i < afu->crs_num; i++) {
902 		rc = cxl_ops->afu_cr_read32(afu, i, 0, &val);
903 		if (rc || val == 0) {
904 			dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
905 			return -EINVAL;
906 		}
907 	}
908 
909 	if ((afu->modes_supported & ~CXL_MODE_DEDICATED) && afu->max_procs_virtualised == 0) {
910 		/*
911 		 * We could also check this for the dedicated process model
912 		 * since the architecture indicates it should be set to 1, but
913 		 * in that case we ignore the value and I'd rather not risk
914 		 * breaking any existing dedicated process AFUs that left it as
915 		 * 0 (not that I'm aware of any). It is clearly an error for an
916 		 * AFU directed AFU to set this to 0, and would have previously
917 		 * triggered a bug resulting in the maximum not being enforced
918 		 * at all since idr_alloc treats 0 as no maximum.
919 		 */
920 		dev_err(&afu->dev, "AFU does not support any processes\n");
921 		return -EINVAL;
922 	}
923 
924 	return 0;
925 }
926 
927 static int sanitise_afu_regs_psl9(struct cxl_afu *afu)
928 {
929 	u64 reg;
930 
931 	/*
932 	 * Clear out any regs that contain either an IVTE or address or may be
933 	 * waiting on an acknowledgment to try to be a bit safer as we bring
934 	 * it online
935 	 */
936 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
937 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
938 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
939 		if (cxl_ops->afu_reset(afu))
940 			return -EIO;
941 		if (cxl_afu_disable(afu))
942 			return -EIO;
943 		if (cxl_psl_purge(afu))
944 			return -EIO;
945 	}
946 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
947 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
948 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
949 	if (reg) {
950 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
951 		if (reg & CXL_PSL9_DSISR_An_TF)
952 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
953 		else
954 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
955 	}
956 	if (afu->adapter->native->sl_ops->register_serr_irq) {
957 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
958 		if (reg) {
959 			if (reg & ~0x000000007fffffff)
960 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
961 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
962 		}
963 	}
964 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
965 	if (reg) {
966 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
967 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
968 	}
969 
970 	return 0;
971 }
972 
973 static int sanitise_afu_regs_psl8(struct cxl_afu *afu)
974 {
975 	u64 reg;
976 
977 	/*
978 	 * Clear out any regs that contain either an IVTE or address or may be
979 	 * waiting on an acknowledgement to try to be a bit safer as we bring
980 	 * it online
981 	 */
982 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
983 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
984 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
985 		if (cxl_ops->afu_reset(afu))
986 			return -EIO;
987 		if (cxl_afu_disable(afu))
988 			return -EIO;
989 		if (cxl_psl_purge(afu))
990 			return -EIO;
991 	}
992 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
993 	cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
994 	cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
995 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
996 	cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
997 	cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
998 	cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
999 	cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
1000 	cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
1001 	cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
1002 	cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
1003 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1004 	if (reg) {
1005 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1006 		if (reg & CXL_PSL_DSISR_TRANS)
1007 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1008 		else
1009 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1010 	}
1011 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1012 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1013 		if (reg) {
1014 			if (reg & ~0xffff)
1015 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1016 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1017 		}
1018 	}
1019 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1020 	if (reg) {
1021 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1022 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1023 	}
1024 
1025 	return 0;
1026 }
1027 
1028 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
1029 /*
1030  * afu_eb_read:
1031  * Called from sysfs and reads the afu error info buffer. The h/w only supports
1032  * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
1033  * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
1034  */
1035 ssize_t cxl_pci_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
1036 				loff_t off, size_t count)
1037 {
1038 	loff_t aligned_start, aligned_end;
1039 	size_t aligned_length;
1040 	void *tbuf;
1041 	const void __iomem *ebuf = afu->native->afu_desc_mmio + afu->eb_offset;
1042 
1043 	if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
1044 		return 0;
1045 
1046 	/* calculate aligned read window */
1047 	count = min((size_t)(afu->eb_len - off), count);
1048 	aligned_start = round_down(off, 8);
1049 	aligned_end = round_up(off + count, 8);
1050 	aligned_length = aligned_end - aligned_start;
1051 
1052 	/* max we can copy in one read is PAGE_SIZE */
1053 	if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
1054 		aligned_length = ERR_BUFF_MAX_COPY_SIZE;
1055 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
1056 	}
1057 
1058 	/* use bounce buffer for copy */
1059 	tbuf = (void *)__get_free_page(GFP_KERNEL);
1060 	if (!tbuf)
1061 		return -ENOMEM;
1062 
1063 	/* perform aligned read from the mmio region */
1064 	memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
1065 	memcpy(buf, tbuf + (off & 0x7), count);
1066 
1067 	free_page((unsigned long)tbuf);
1068 
1069 	return count;
1070 }
1071 
1072 static int pci_configure_afu(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1073 {
1074 	int rc;
1075 
1076 	if ((rc = pci_map_slice_regs(afu, adapter, dev)))
1077 		return rc;
1078 
1079 	if (adapter->native->sl_ops->sanitise_afu_regs) {
1080 		rc = adapter->native->sl_ops->sanitise_afu_regs(afu);
1081 		if (rc)
1082 			goto err1;
1083 	}
1084 
1085 	/* We need to reset the AFU before we can read the AFU descriptor */
1086 	if ((rc = cxl_ops->afu_reset(afu)))
1087 		goto err1;
1088 
1089 	if (cxl_verbose)
1090 		dump_afu_descriptor(afu);
1091 
1092 	if ((rc = cxl_read_afu_descriptor(afu)))
1093 		goto err1;
1094 
1095 	if ((rc = cxl_afu_descriptor_looks_ok(afu)))
1096 		goto err1;
1097 
1098 	if (adapter->native->sl_ops->afu_regs_init)
1099 		if ((rc = adapter->native->sl_ops->afu_regs_init(afu)))
1100 			goto err1;
1101 
1102 	if (adapter->native->sl_ops->register_serr_irq)
1103 		if ((rc = adapter->native->sl_ops->register_serr_irq(afu)))
1104 			goto err1;
1105 
1106 	if ((rc = cxl_native_register_psl_irq(afu)))
1107 		goto err2;
1108 
1109 	atomic_set(&afu->configured_state, 0);
1110 	return 0;
1111 
1112 err2:
1113 	if (adapter->native->sl_ops->release_serr_irq)
1114 		adapter->native->sl_ops->release_serr_irq(afu);
1115 err1:
1116 	pci_unmap_slice_regs(afu);
1117 	return rc;
1118 }
1119 
1120 static void pci_deconfigure_afu(struct cxl_afu *afu)
1121 {
1122 	/*
1123 	 * It's okay to deconfigure when AFU is already locked, otherwise wait
1124 	 * until there are no readers
1125 	 */
1126 	if (atomic_read(&afu->configured_state) != -1) {
1127 		while (atomic_cmpxchg(&afu->configured_state, 0, -1) != -1)
1128 			schedule();
1129 	}
1130 	cxl_native_release_psl_irq(afu);
1131 	if (afu->adapter->native->sl_ops->release_serr_irq)
1132 		afu->adapter->native->sl_ops->release_serr_irq(afu);
1133 	pci_unmap_slice_regs(afu);
1134 }
1135 
1136 static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
1137 {
1138 	struct cxl_afu *afu;
1139 	int rc = -ENOMEM;
1140 
1141 	afu = cxl_alloc_afu(adapter, slice);
1142 	if (!afu)
1143 		return -ENOMEM;
1144 
1145 	afu->native = kzalloc(sizeof(struct cxl_afu_native), GFP_KERNEL);
1146 	if (!afu->native)
1147 		goto err_free_afu;
1148 
1149 	mutex_init(&afu->native->spa_mutex);
1150 
1151 	rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice);
1152 	if (rc)
1153 		goto err_free_native;
1154 
1155 	rc = pci_configure_afu(afu, adapter, dev);
1156 	if (rc)
1157 		goto err_free_native;
1158 
1159 	/* Don't care if this fails */
1160 	cxl_debugfs_afu_add(afu);
1161 
1162 	/*
1163 	 * After we call this function we must not free the afu directly, even
1164 	 * if it returns an error!
1165 	 */
1166 	if ((rc = cxl_register_afu(afu)))
1167 		goto err_put1;
1168 
1169 	if ((rc = cxl_sysfs_afu_add(afu)))
1170 		goto err_put1;
1171 
1172 	adapter->afu[afu->slice] = afu;
1173 
1174 	if ((rc = cxl_pci_vphb_add(afu)))
1175 		dev_info(&afu->dev, "Can't register vPHB\n");
1176 
1177 	return 0;
1178 
1179 err_put1:
1180 	pci_deconfigure_afu(afu);
1181 	cxl_debugfs_afu_remove(afu);
1182 	device_unregister(&afu->dev);
1183 	return rc;
1184 
1185 err_free_native:
1186 	kfree(afu->native);
1187 err_free_afu:
1188 	kfree(afu);
1189 	return rc;
1190 
1191 }
1192 
1193 static void cxl_pci_remove_afu(struct cxl_afu *afu)
1194 {
1195 	pr_devel("%s\n", __func__);
1196 
1197 	if (!afu)
1198 		return;
1199 
1200 	cxl_pci_vphb_remove(afu);
1201 	cxl_sysfs_afu_remove(afu);
1202 	cxl_debugfs_afu_remove(afu);
1203 
1204 	spin_lock(&afu->adapter->afu_list_lock);
1205 	afu->adapter->afu[afu->slice] = NULL;
1206 	spin_unlock(&afu->adapter->afu_list_lock);
1207 
1208 	cxl_context_detach_all(afu);
1209 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1210 
1211 	pci_deconfigure_afu(afu);
1212 	device_unregister(&afu->dev);
1213 }
1214 
1215 int cxl_pci_reset(struct cxl *adapter)
1216 {
1217 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1218 	int rc;
1219 
1220 	if (adapter->perst_same_image) {
1221 		dev_warn(&dev->dev,
1222 			 "cxl: refusing to reset/reflash when perst_reloads_same_image is set.\n");
1223 		return -EINVAL;
1224 	}
1225 
1226 	dev_info(&dev->dev, "CXL reset\n");
1227 
1228 	/*
1229 	 * The adapter is about to be reset, so ignore errors.
1230 	 */
1231 	cxl_data_cache_flush(adapter);
1232 
1233 	/* pcie_warm_reset requests a fundamental pci reset which includes a
1234 	 * PERST assert/deassert.  PERST triggers a loading of the image
1235 	 * if "user" or "factory" is selected in sysfs */
1236 	if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
1237 		dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
1238 		return rc;
1239 	}
1240 
1241 	return rc;
1242 }
1243 
1244 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
1245 {
1246 	if (pci_request_region(dev, 2, "priv 2 regs"))
1247 		goto err1;
1248 	if (pci_request_region(dev, 0, "priv 1 regs"))
1249 		goto err2;
1250 
1251 	pr_devel("cxl_map_adapter_regs: p1: %#016llx %#llx, p2: %#016llx %#llx",
1252 			p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
1253 
1254 	if (!(adapter->native->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
1255 		goto err3;
1256 
1257 	if (!(adapter->native->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
1258 		goto err4;
1259 
1260 	return 0;
1261 
1262 err4:
1263 	iounmap(adapter->native->p1_mmio);
1264 	adapter->native->p1_mmio = NULL;
1265 err3:
1266 	pci_release_region(dev, 0);
1267 err2:
1268 	pci_release_region(dev, 2);
1269 err1:
1270 	return -ENOMEM;
1271 }
1272 
1273 static void cxl_unmap_adapter_regs(struct cxl *adapter)
1274 {
1275 	if (adapter->native->p1_mmio) {
1276 		iounmap(adapter->native->p1_mmio);
1277 		adapter->native->p1_mmio = NULL;
1278 		pci_release_region(to_pci_dev(adapter->dev.parent), 2);
1279 	}
1280 	if (adapter->native->p2_mmio) {
1281 		iounmap(adapter->native->p2_mmio);
1282 		adapter->native->p2_mmio = NULL;
1283 		pci_release_region(to_pci_dev(adapter->dev.parent), 0);
1284 	}
1285 }
1286 
1287 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
1288 {
1289 	int vsec;
1290 	u32 afu_desc_off, afu_desc_size;
1291 	u32 ps_off, ps_size;
1292 	u16 vseclen;
1293 	u8 image_state;
1294 
1295 	if (!(vsec = find_cxl_vsec(dev))) {
1296 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
1297 		return -ENODEV;
1298 	}
1299 
1300 	CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
1301 	if (vseclen < CXL_VSEC_MIN_SIZE) {
1302 		dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
1303 		return -EINVAL;
1304 	}
1305 
1306 	CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
1307 	CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
1308 	CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
1309 	CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
1310 	CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
1311 	CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
1312 	adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1313 	adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1314 	adapter->perst_loads_image = !!(image_state & CXL_VSEC_PERST_LOADS_IMAGE);
1315 
1316 	CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
1317 	CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
1318 	CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
1319 	CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
1320 	CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
1321 
1322 	/* Convert everything to bytes, because there is NO WAY I'd look at the
1323 	 * code a month later and forget what units these are in ;-) */
1324 	adapter->native->ps_off = ps_off * 64 * 1024;
1325 	adapter->ps_size = ps_size * 64 * 1024;
1326 	adapter->native->afu_desc_off = afu_desc_off * 64 * 1024;
1327 	adapter->native->afu_desc_size = afu_desc_size * 64 * 1024;
1328 
1329 	/* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
1330 	adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
1331 
1332 	return 0;
1333 }
1334 
1335 /*
1336  * Workaround a PCIe Host Bridge defect on some cards, that can cause
1337  * malformed Transaction Layer Packet (TLP) errors to be erroneously
1338  * reported. Mask this error in the Uncorrectable Error Mask Register.
1339  *
1340  * The upper nibble of the PSL revision is used to distinguish between
1341  * different cards. The affected ones have it set to 0.
1342  */
1343 static void cxl_fixup_malformed_tlp(struct cxl *adapter, struct pci_dev *dev)
1344 {
1345 	int aer;
1346 	u32 data;
1347 
1348 	if (adapter->psl_rev & 0xf000)
1349 		return;
1350 	if (!(aer = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR)))
1351 		return;
1352 	pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &data);
1353 	if (data & PCI_ERR_UNC_MALF_TLP)
1354 		if (data & PCI_ERR_UNC_INTN)
1355 			return;
1356 	data |= PCI_ERR_UNC_MALF_TLP;
1357 	data |= PCI_ERR_UNC_INTN;
1358 	pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, data);
1359 }
1360 
1361 static bool cxl_compatible_caia_version(struct cxl *adapter)
1362 {
1363 	if (cxl_is_power8() && (adapter->caia_major == 1))
1364 		return true;
1365 
1366 	if (cxl_is_power9() && (adapter->caia_major == 2))
1367 		return true;
1368 
1369 	return false;
1370 }
1371 
1372 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
1373 {
1374 	if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
1375 		return -EBUSY;
1376 
1377 	if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
1378 		dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
1379 		return -EINVAL;
1380 	}
1381 
1382 	if (!cxl_compatible_caia_version(adapter)) {
1383 		dev_info(&dev->dev, "Ignoring card. PSL type is not supported (caia version: %d)\n",
1384 			 adapter->caia_major);
1385 		return -ENODEV;
1386 	}
1387 
1388 	if (!adapter->slices) {
1389 		/* Once we support dynamic reprogramming we can use the card if
1390 		 * it supports loadable AFUs */
1391 		dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
1392 		return -EINVAL;
1393 	}
1394 
1395 	if (!adapter->native->afu_desc_off || !adapter->native->afu_desc_size) {
1396 		dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
1397 		return -EINVAL;
1398 	}
1399 
1400 	if (adapter->ps_size > p2_size(dev) - adapter->native->ps_off) {
1401 		dev_err(&dev->dev, "ABORTING: Problem state size larger than "
1402 				   "available in BAR2: 0x%llx > 0x%llx\n",
1403 			 adapter->ps_size, p2_size(dev) - adapter->native->ps_off);
1404 		return -EINVAL;
1405 	}
1406 
1407 	return 0;
1408 }
1409 
1410 ssize_t cxl_pci_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1411 {
1412 	return pci_read_vpd(to_pci_dev(adapter->dev.parent), 0, len, buf);
1413 }
1414 
1415 static void cxl_release_adapter(struct device *dev)
1416 {
1417 	struct cxl *adapter = to_cxl_adapter(dev);
1418 
1419 	pr_devel("cxl_release_adapter\n");
1420 
1421 	cxl_remove_adapter_nr(adapter);
1422 
1423 	kfree(adapter->native);
1424 	kfree(adapter);
1425 }
1426 
1427 #define CXL_PSL_ErrIVTE_tberror (0x1ull << (63-31))
1428 
1429 static int sanitise_adapter_regs(struct cxl *adapter)
1430 {
1431 	int rc = 0;
1432 
1433 	/* Clear PSL tberror bit by writing 1 to it */
1434 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, CXL_PSL_ErrIVTE_tberror);
1435 
1436 	if (adapter->native->sl_ops->invalidate_all) {
1437 		/* do not invalidate ERAT entries when not reloading on PERST */
1438 		if (cxl_is_power9() && (adapter->perst_loads_image))
1439 			return 0;
1440 		rc = adapter->native->sl_ops->invalidate_all(adapter);
1441 	}
1442 
1443 	return rc;
1444 }
1445 
1446 /* This should contain *only* operations that can safely be done in
1447  * both creation and recovery.
1448  */
1449 static int cxl_configure_adapter(struct cxl *adapter, struct pci_dev *dev)
1450 {
1451 	int rc;
1452 
1453 	adapter->dev.parent = &dev->dev;
1454 	adapter->dev.release = cxl_release_adapter;
1455 	pci_set_drvdata(dev, adapter);
1456 
1457 	rc = pci_enable_device(dev);
1458 	if (rc) {
1459 		dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1460 		return rc;
1461 	}
1462 
1463 	if ((rc = cxl_read_vsec(adapter, dev)))
1464 		return rc;
1465 
1466 	if ((rc = cxl_vsec_looks_ok(adapter, dev)))
1467 	        return rc;
1468 
1469 	cxl_fixup_malformed_tlp(adapter, dev);
1470 
1471 	if ((rc = setup_cxl_bars(dev)))
1472 		return rc;
1473 
1474 	if ((rc = switch_card_to_cxl(dev)))
1475 		return rc;
1476 
1477 	if ((rc = cxl_update_image_control(adapter)))
1478 		return rc;
1479 
1480 	if ((rc = cxl_map_adapter_regs(adapter, dev)))
1481 		return rc;
1482 
1483 	if ((rc = sanitise_adapter_regs(adapter)))
1484 		goto err;
1485 
1486 	if ((rc = adapter->native->sl_ops->adapter_regs_init(adapter, dev)))
1487 		goto err;
1488 
1489 	/* Required for devices using CAPP DMA mode, harmless for others */
1490 	pci_set_master(dev);
1491 
1492 	adapter->tunneled_ops_supported = false;
1493 
1494 	if (cxl_is_power9()) {
1495 		if (pnv_pci_set_tunnel_bar(dev, 0x00020000E0000000ull, 1))
1496 			dev_info(&dev->dev, "Tunneled operations unsupported\n");
1497 		else
1498 			adapter->tunneled_ops_supported = true;
1499 	}
1500 
1501 	if ((rc = pnv_phb_to_cxl_mode(dev, adapter->native->sl_ops->capi_mode)))
1502 		goto err;
1503 
1504 	/* If recovery happened, the last step is to turn on snooping.
1505 	 * In the non-recovery case this has no effect */
1506 	if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON)))
1507 		goto err;
1508 
1509 	/* Ignore error, adapter init is not dependant on timebase sync */
1510 	cxl_setup_psl_timebase(adapter, dev);
1511 
1512 	if ((rc = cxl_native_register_psl_err_irq(adapter)))
1513 		goto err;
1514 
1515 	return 0;
1516 
1517 err:
1518 	cxl_unmap_adapter_regs(adapter);
1519 	return rc;
1520 
1521 }
1522 
1523 static void cxl_deconfigure_adapter(struct cxl *adapter)
1524 {
1525 	struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1526 
1527 	if (cxl_is_power9())
1528 		pnv_pci_set_tunnel_bar(pdev, 0x00020000E0000000ull, 0);
1529 
1530 	cxl_native_release_psl_err_irq(adapter);
1531 	cxl_unmap_adapter_regs(adapter);
1532 
1533 	pci_disable_device(pdev);
1534 }
1535 
1536 static void cxl_stop_trace_psl9(struct cxl *adapter)
1537 {
1538 	int traceid;
1539 	u64 trace_state, trace_mask;
1540 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1541 
1542 	/* read each tracearray state and issue mmio to stop them is needed */
1543 	for (traceid = 0; traceid <= CXL_PSL9_TRACEID_MAX; ++traceid) {
1544 		trace_state = cxl_p1_read(adapter, CXL_PSL9_CTCCFG);
1545 		trace_mask = (0x3ULL << (62 - traceid * 2));
1546 		trace_state = (trace_state & trace_mask) >> (62 - traceid * 2);
1547 		dev_dbg(&dev->dev, "cxl: Traceid-%d trace_state=0x%0llX\n",
1548 			traceid, trace_state);
1549 
1550 		/* issue mmio if the trace array isn't in FIN state */
1551 		if (trace_state != CXL_PSL9_TRACESTATE_FIN)
1552 			cxl_p1_write(adapter, CXL_PSL9_TRACECFG,
1553 				     0x8400000000000000ULL | traceid);
1554 	}
1555 }
1556 
1557 static void cxl_stop_trace_psl8(struct cxl *adapter)
1558 {
1559 	int slice;
1560 
1561 	/* Stop the trace */
1562 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x8000000000000017LL);
1563 
1564 	/* Stop the slice traces */
1565 	spin_lock(&adapter->afu_list_lock);
1566 	for (slice = 0; slice < adapter->slices; slice++) {
1567 		if (adapter->afu[slice])
1568 			cxl_p1n_write(adapter->afu[slice], CXL_PSL_SLICE_TRACE,
1569 				      0x8000000000000000LL);
1570 	}
1571 	spin_unlock(&adapter->afu_list_lock);
1572 }
1573 
1574 static const struct cxl_service_layer_ops psl9_ops = {
1575 	.adapter_regs_init = init_implementation_adapter_regs_psl9,
1576 	.invalidate_all = cxl_invalidate_all_psl9,
1577 	.afu_regs_init = init_implementation_afu_regs_psl9,
1578 	.sanitise_afu_regs = sanitise_afu_regs_psl9,
1579 	.register_serr_irq = cxl_native_register_serr_irq,
1580 	.release_serr_irq = cxl_native_release_serr_irq,
1581 	.handle_interrupt = cxl_irq_psl9,
1582 	.fail_irq = cxl_fail_irq_psl,
1583 	.activate_dedicated_process = cxl_activate_dedicated_process_psl9,
1584 	.attach_afu_directed = cxl_attach_afu_directed_psl9,
1585 	.attach_dedicated_process = cxl_attach_dedicated_process_psl9,
1586 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl9,
1587 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl9,
1588 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl9,
1589 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl9,
1590 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl9,
1591 	.debugfs_stop_trace = cxl_stop_trace_psl9,
1592 	.timebase_read = timebase_read_psl9,
1593 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1594 	.needs_reset_before_disable = true,
1595 };
1596 
1597 static const struct cxl_service_layer_ops psl8_ops = {
1598 	.adapter_regs_init = init_implementation_adapter_regs_psl8,
1599 	.invalidate_all = cxl_invalidate_all_psl8,
1600 	.afu_regs_init = init_implementation_afu_regs_psl8,
1601 	.sanitise_afu_regs = sanitise_afu_regs_psl8,
1602 	.register_serr_irq = cxl_native_register_serr_irq,
1603 	.release_serr_irq = cxl_native_release_serr_irq,
1604 	.handle_interrupt = cxl_irq_psl8,
1605 	.fail_irq = cxl_fail_irq_psl,
1606 	.activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1607 	.attach_afu_directed = cxl_attach_afu_directed_psl8,
1608 	.attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1609 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1610 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl8,
1611 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl8,
1612 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl8,
1613 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl8,
1614 	.debugfs_stop_trace = cxl_stop_trace_psl8,
1615 	.write_timebase_ctrl = write_timebase_ctrl_psl8,
1616 	.timebase_read = timebase_read_psl8,
1617 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1618 	.needs_reset_before_disable = true,
1619 };
1620 
1621 static void set_sl_ops(struct cxl *adapter, struct pci_dev *dev)
1622 {
1623 	if (cxl_is_power8()) {
1624 		dev_info(&dev->dev, "Device uses a PSL8\n");
1625 		adapter->native->sl_ops = &psl8_ops;
1626 	} else {
1627 		dev_info(&dev->dev, "Device uses a PSL9\n");
1628 		adapter->native->sl_ops = &psl9_ops;
1629 	}
1630 }
1631 
1632 
1633 static struct cxl *cxl_pci_init_adapter(struct pci_dev *dev)
1634 {
1635 	struct cxl *adapter;
1636 	int rc;
1637 
1638 	adapter = cxl_alloc_adapter();
1639 	if (!adapter)
1640 		return ERR_PTR(-ENOMEM);
1641 
1642 	adapter->native = kzalloc(sizeof(struct cxl_native), GFP_KERNEL);
1643 	if (!adapter->native) {
1644 		rc = -ENOMEM;
1645 		goto err_release;
1646 	}
1647 
1648 	set_sl_ops(adapter, dev);
1649 
1650 	/* Set defaults for parameters which need to persist over
1651 	 * configure/reconfigure
1652 	 */
1653 	adapter->perst_loads_image = true;
1654 	adapter->perst_same_image = false;
1655 
1656 	rc = cxl_configure_adapter(adapter, dev);
1657 	if (rc) {
1658 		pci_disable_device(dev);
1659 		goto err_release;
1660 	}
1661 
1662 	/* Don't care if this one fails: */
1663 	cxl_debugfs_adapter_add(adapter);
1664 
1665 	/*
1666 	 * After we call this function we must not free the adapter directly,
1667 	 * even if it returns an error!
1668 	 */
1669 	if ((rc = cxl_register_adapter(adapter)))
1670 		goto err_put1;
1671 
1672 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1673 		goto err_put1;
1674 
1675 	/* Release the context lock as adapter is configured */
1676 	cxl_adapter_context_unlock(adapter);
1677 
1678 	return adapter;
1679 
1680 err_put1:
1681 	/* This should mirror cxl_remove_adapter, except without the
1682 	 * sysfs parts
1683 	 */
1684 	cxl_debugfs_adapter_remove(adapter);
1685 	cxl_deconfigure_adapter(adapter);
1686 	device_unregister(&adapter->dev);
1687 	return ERR_PTR(rc);
1688 
1689 err_release:
1690 	cxl_release_adapter(&adapter->dev);
1691 	return ERR_PTR(rc);
1692 }
1693 
1694 static void cxl_pci_remove_adapter(struct cxl *adapter)
1695 {
1696 	pr_devel("cxl_remove_adapter\n");
1697 
1698 	cxl_sysfs_adapter_remove(adapter);
1699 	cxl_debugfs_adapter_remove(adapter);
1700 
1701 	/*
1702 	 * Flush adapter datacache as its about to be removed.
1703 	 */
1704 	cxl_data_cache_flush(adapter);
1705 
1706 	cxl_deconfigure_adapter(adapter);
1707 
1708 	device_unregister(&adapter->dev);
1709 }
1710 
1711 #define CXL_MAX_PCIEX_PARENT 2
1712 
1713 int cxl_slot_is_switched(struct pci_dev *dev)
1714 {
1715 	struct device_node *np;
1716 	int depth = 0;
1717 
1718 	if (!(np = pci_device_to_OF_node(dev))) {
1719 		pr_err("cxl: np = NULL\n");
1720 		return -ENODEV;
1721 	}
1722 	of_node_get(np);
1723 	while (np) {
1724 		np = of_get_next_parent(np);
1725 		if (!of_node_is_type(np, "pciex"))
1726 			break;
1727 		depth++;
1728 	}
1729 	of_node_put(np);
1730 	return (depth > CXL_MAX_PCIEX_PARENT);
1731 }
1732 
1733 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
1734 {
1735 	struct cxl *adapter;
1736 	int slice;
1737 	int rc;
1738 
1739 	if (cxl_pci_is_vphb_device(dev)) {
1740 		dev_dbg(&dev->dev, "cxl_init_adapter: Ignoring cxl vphb device\n");
1741 		return -ENODEV;
1742 	}
1743 
1744 	if (cxl_slot_is_switched(dev)) {
1745 		dev_info(&dev->dev, "Ignoring card on incompatible PCI slot\n");
1746 		return -ENODEV;
1747 	}
1748 
1749 	if (cxl_is_power9() && !radix_enabled()) {
1750 		dev_info(&dev->dev, "Only Radix mode supported\n");
1751 		return -ENODEV;
1752 	}
1753 
1754 	if (cxl_verbose)
1755 		dump_cxl_config_space(dev);
1756 
1757 	adapter = cxl_pci_init_adapter(dev);
1758 	if (IS_ERR(adapter)) {
1759 		dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
1760 		return PTR_ERR(adapter);
1761 	}
1762 
1763 	for (slice = 0; slice < adapter->slices; slice++) {
1764 		if ((rc = pci_init_afu(adapter, slice, dev))) {
1765 			dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
1766 			continue;
1767 		}
1768 
1769 		rc = cxl_afu_select_best_mode(adapter->afu[slice]);
1770 		if (rc)
1771 			dev_err(&dev->dev, "AFU %i failed to start: %i\n", slice, rc);
1772 	}
1773 
1774 	return 0;
1775 }
1776 
1777 static void cxl_remove(struct pci_dev *dev)
1778 {
1779 	struct cxl *adapter = pci_get_drvdata(dev);
1780 	struct cxl_afu *afu;
1781 	int i;
1782 
1783 	/*
1784 	 * Lock to prevent someone grabbing a ref through the adapter list as
1785 	 * we are removing it
1786 	 */
1787 	for (i = 0; i < adapter->slices; i++) {
1788 		afu = adapter->afu[i];
1789 		cxl_pci_remove_afu(afu);
1790 	}
1791 	cxl_pci_remove_adapter(adapter);
1792 }
1793 
1794 static pci_ers_result_t cxl_vphb_error_detected(struct cxl_afu *afu,
1795 						pci_channel_state_t state)
1796 {
1797 	struct pci_dev *afu_dev;
1798 	struct pci_driver *afu_drv;
1799 	const struct pci_error_handlers *err_handler;
1800 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
1801 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
1802 
1803 	/* There should only be one entry, but go through the list
1804 	 * anyway
1805 	 */
1806 	if (afu == NULL || afu->phb == NULL)
1807 		return result;
1808 
1809 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
1810 		afu_drv = to_pci_driver(afu_dev->dev.driver);
1811 		if (!afu_drv)
1812 			continue;
1813 
1814 		afu_dev->error_state = state;
1815 
1816 		err_handler = afu_drv->err_handler;
1817 		if (err_handler)
1818 			afu_result = err_handler->error_detected(afu_dev,
1819 								 state);
1820 		/* Disconnect trumps all, NONE trumps NEED_RESET */
1821 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
1822 			result = PCI_ERS_RESULT_DISCONNECT;
1823 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
1824 			 (result == PCI_ERS_RESULT_NEED_RESET))
1825 			result = PCI_ERS_RESULT_NONE;
1826 	}
1827 	return result;
1828 }
1829 
1830 static pci_ers_result_t cxl_pci_error_detected(struct pci_dev *pdev,
1831 					       pci_channel_state_t state)
1832 {
1833 	struct cxl *adapter = pci_get_drvdata(pdev);
1834 	struct cxl_afu *afu;
1835 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
1836 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
1837 	int i;
1838 
1839 	/* At this point, we could still have an interrupt pending.
1840 	 * Let's try to get them out of the way before they do
1841 	 * anything we don't like.
1842 	 */
1843 	schedule();
1844 
1845 	/* If we're permanently dead, give up. */
1846 	if (state == pci_channel_io_perm_failure) {
1847 		spin_lock(&adapter->afu_list_lock);
1848 		for (i = 0; i < adapter->slices; i++) {
1849 			afu = adapter->afu[i];
1850 			/*
1851 			 * Tell the AFU drivers; but we don't care what they
1852 			 * say, we're going away.
1853 			 */
1854 			cxl_vphb_error_detected(afu, state);
1855 		}
1856 		spin_unlock(&adapter->afu_list_lock);
1857 		return PCI_ERS_RESULT_DISCONNECT;
1858 	}
1859 
1860 	/* Are we reflashing?
1861 	 *
1862 	 * If we reflash, we could come back as something entirely
1863 	 * different, including a non-CAPI card. As such, by default
1864 	 * we don't participate in the process. We'll be unbound and
1865 	 * the slot re-probed. (TODO: check EEH doesn't blindly rebind
1866 	 * us!)
1867 	 *
1868 	 * However, this isn't the entire story: for reliablity
1869 	 * reasons, we usually want to reflash the FPGA on PERST in
1870 	 * order to get back to a more reliable known-good state.
1871 	 *
1872 	 * This causes us a bit of a problem: if we reflash we can't
1873 	 * trust that we'll come back the same - we could have a new
1874 	 * image and been PERSTed in order to load that
1875 	 * image. However, most of the time we actually *will* come
1876 	 * back the same - for example a regular EEH event.
1877 	 *
1878 	 * Therefore, we allow the user to assert that the image is
1879 	 * indeed the same and that we should continue on into EEH
1880 	 * anyway.
1881 	 */
1882 	if (adapter->perst_loads_image && !adapter->perst_same_image) {
1883 		/* TODO take the PHB out of CXL mode */
1884 		dev_info(&pdev->dev, "reflashing, so opting out of EEH!\n");
1885 		return PCI_ERS_RESULT_NONE;
1886 	}
1887 
1888 	/*
1889 	 * At this point, we want to try to recover.  We'll always
1890 	 * need a complete slot reset: we don't trust any other reset.
1891 	 *
1892 	 * Now, we go through each AFU:
1893 	 *  - We send the driver, if bound, an error_detected callback.
1894 	 *    We expect it to clean up, but it can also tell us to give
1895 	 *    up and permanently detach the card. To simplify things, if
1896 	 *    any bound AFU driver doesn't support EEH, we give up on EEH.
1897 	 *
1898 	 *  - We detach all contexts associated with the AFU. This
1899 	 *    does not free them, but puts them into a CLOSED state
1900 	 *    which causes any the associated files to return useful
1901 	 *    errors to userland. It also unmaps, but does not free,
1902 	 *    any IRQs.
1903 	 *
1904 	 *  - We clean up our side: releasing and unmapping resources we hold
1905 	 *    so we can wire them up again when the hardware comes back up.
1906 	 *
1907 	 * Driver authors should note:
1908 	 *
1909 	 *  - Any contexts you create in your kernel driver (except
1910 	 *    those associated with anonymous file descriptors) are
1911 	 *    your responsibility to free and recreate. Likewise with
1912 	 *    any attached resources.
1913 	 *
1914 	 *  - We will take responsibility for re-initialising the
1915 	 *    device context (the one set up for you in
1916 	 *    cxl_pci_enable_device_hook and accessed through
1917 	 *    cxl_get_context). If you've attached IRQs or other
1918 	 *    resources to it, they remains yours to free.
1919 	 *
1920 	 * You can call the same functions to release resources as you
1921 	 * normally would: we make sure that these functions continue
1922 	 * to work when the hardware is down.
1923 	 *
1924 	 * Two examples:
1925 	 *
1926 	 * 1) If you normally free all your resources at the end of
1927 	 *    each request, or if you use anonymous FDs, your
1928 	 *    error_detected callback can simply set a flag to tell
1929 	 *    your driver not to start any new calls. You can then
1930 	 *    clear the flag in the resume callback.
1931 	 *
1932 	 * 2) If you normally allocate your resources on startup:
1933 	 *     * Set a flag in error_detected as above.
1934 	 *     * Let CXL detach your contexts.
1935 	 *     * In slot_reset, free the old resources and allocate new ones.
1936 	 *     * In resume, clear the flag to allow things to start.
1937 	 */
1938 
1939 	/* Make sure no one else changes the afu list */
1940 	spin_lock(&adapter->afu_list_lock);
1941 
1942 	for (i = 0; i < adapter->slices; i++) {
1943 		afu = adapter->afu[i];
1944 
1945 		if (afu == NULL)
1946 			continue;
1947 
1948 		afu_result = cxl_vphb_error_detected(afu, state);
1949 		cxl_context_detach_all(afu);
1950 		cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1951 		pci_deconfigure_afu(afu);
1952 
1953 		/* Disconnect trumps all, NONE trumps NEED_RESET */
1954 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
1955 			result = PCI_ERS_RESULT_DISCONNECT;
1956 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
1957 			 (result == PCI_ERS_RESULT_NEED_RESET))
1958 			result = PCI_ERS_RESULT_NONE;
1959 	}
1960 	spin_unlock(&adapter->afu_list_lock);
1961 
1962 	/* should take the context lock here */
1963 	if (cxl_adapter_context_lock(adapter) != 0)
1964 		dev_warn(&adapter->dev,
1965 			 "Couldn't take context lock with %d active-contexts\n",
1966 			 atomic_read(&adapter->contexts_num));
1967 
1968 	cxl_deconfigure_adapter(adapter);
1969 
1970 	return result;
1971 }
1972 
1973 static pci_ers_result_t cxl_pci_slot_reset(struct pci_dev *pdev)
1974 {
1975 	struct cxl *adapter = pci_get_drvdata(pdev);
1976 	struct cxl_afu *afu;
1977 	struct cxl_context *ctx;
1978 	struct pci_dev *afu_dev;
1979 	struct pci_driver *afu_drv;
1980 	const struct pci_error_handlers *err_handler;
1981 	pci_ers_result_t afu_result = PCI_ERS_RESULT_RECOVERED;
1982 	pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
1983 	int i;
1984 
1985 	if (cxl_configure_adapter(adapter, pdev))
1986 		goto err;
1987 
1988 	/*
1989 	 * Unlock context activation for the adapter. Ideally this should be
1990 	 * done in cxl_pci_resume but cxlflash module tries to activate the
1991 	 * master context as part of slot_reset callback.
1992 	 */
1993 	cxl_adapter_context_unlock(adapter);
1994 
1995 	spin_lock(&adapter->afu_list_lock);
1996 	for (i = 0; i < adapter->slices; i++) {
1997 		afu = adapter->afu[i];
1998 
1999 		if (afu == NULL)
2000 			continue;
2001 
2002 		if (pci_configure_afu(afu, adapter, pdev))
2003 			goto err_unlock;
2004 
2005 		if (cxl_afu_select_best_mode(afu))
2006 			goto err_unlock;
2007 
2008 		if (afu->phb == NULL)
2009 			continue;
2010 
2011 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2012 			/* Reset the device context.
2013 			 * TODO: make this less disruptive
2014 			 */
2015 			ctx = cxl_get_context(afu_dev);
2016 
2017 			if (ctx && cxl_release_context(ctx))
2018 				goto err_unlock;
2019 
2020 			ctx = cxl_dev_context_init(afu_dev);
2021 			if (IS_ERR(ctx))
2022 				goto err_unlock;
2023 
2024 			afu_dev->dev.archdata.cxl_ctx = ctx;
2025 
2026 			if (cxl_ops->afu_check_and_enable(afu))
2027 				goto err_unlock;
2028 
2029 			afu_dev->error_state = pci_channel_io_normal;
2030 
2031 			/* If there's a driver attached, allow it to
2032 			 * chime in on recovery. Drivers should check
2033 			 * if everything has come back OK, but
2034 			 * shouldn't start new work until we call
2035 			 * their resume function.
2036 			 */
2037 			afu_drv = to_pci_driver(afu_dev->dev.driver);
2038 			if (!afu_drv)
2039 				continue;
2040 
2041 			err_handler = afu_drv->err_handler;
2042 			if (err_handler && err_handler->slot_reset)
2043 				afu_result = err_handler->slot_reset(afu_dev);
2044 
2045 			if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2046 				result = PCI_ERS_RESULT_DISCONNECT;
2047 		}
2048 	}
2049 
2050 	spin_unlock(&adapter->afu_list_lock);
2051 	return result;
2052 
2053 err_unlock:
2054 	spin_unlock(&adapter->afu_list_lock);
2055 
2056 err:
2057 	/* All the bits that happen in both error_detected and cxl_remove
2058 	 * should be idempotent, so we don't need to worry about leaving a mix
2059 	 * of unconfigured and reconfigured resources.
2060 	 */
2061 	dev_err(&pdev->dev, "EEH recovery failed. Asking to be disconnected.\n");
2062 	return PCI_ERS_RESULT_DISCONNECT;
2063 }
2064 
2065 static void cxl_pci_resume(struct pci_dev *pdev)
2066 {
2067 	struct cxl *adapter = pci_get_drvdata(pdev);
2068 	struct cxl_afu *afu;
2069 	struct pci_dev *afu_dev;
2070 	struct pci_driver *afu_drv;
2071 	const struct pci_error_handlers *err_handler;
2072 	int i;
2073 
2074 	/* Everything is back now. Drivers should restart work now.
2075 	 * This is not the place to be checking if everything came back up
2076 	 * properly, because there's no return value: do that in slot_reset.
2077 	 */
2078 	spin_lock(&adapter->afu_list_lock);
2079 	for (i = 0; i < adapter->slices; i++) {
2080 		afu = adapter->afu[i];
2081 
2082 		if (afu == NULL || afu->phb == NULL)
2083 			continue;
2084 
2085 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2086 			afu_drv = to_pci_driver(afu_dev->dev.driver);
2087 			if (!afu_drv)
2088 				continue;
2089 
2090 			err_handler = afu_drv->err_handler;
2091 			if (err_handler && err_handler->resume)
2092 				err_handler->resume(afu_dev);
2093 		}
2094 	}
2095 	spin_unlock(&adapter->afu_list_lock);
2096 }
2097 
2098 static const struct pci_error_handlers cxl_err_handler = {
2099 	.error_detected = cxl_pci_error_detected,
2100 	.slot_reset = cxl_pci_slot_reset,
2101 	.resume = cxl_pci_resume,
2102 };
2103 
2104 struct pci_driver cxl_pci_driver = {
2105 	.name = "cxl-pci",
2106 	.id_table = cxl_pci_tbl,
2107 	.probe = cxl_probe,
2108 	.remove = cxl_remove,
2109 	.shutdown = cxl_remove,
2110 	.err_handler = &cxl_err_handler,
2111 };
2112