xref: /linux/drivers/misc/cxl/pci.c (revision c7546e2c3cb739a3c1a2f5acaf9bb629d401afe5)
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 	return pci_find_vsec_capability(dev, PCI_VENDOR_ID_IBM, CXL_PCI_VSEC_ID);
154 }
155 
156 static void dump_cxl_config_space(struct pci_dev *dev)
157 {
158 	int vsec;
159 	u32 val;
160 
161 	dev_info(&dev->dev, "dump_cxl_config_space\n");
162 
163 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_0, &val);
164 	dev_info(&dev->dev, "BAR0: %#.8x\n", val);
165 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_1, &val);
166 	dev_info(&dev->dev, "BAR1: %#.8x\n", val);
167 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_2, &val);
168 	dev_info(&dev->dev, "BAR2: %#.8x\n", val);
169 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_3, &val);
170 	dev_info(&dev->dev, "BAR3: %#.8x\n", val);
171 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_4, &val);
172 	dev_info(&dev->dev, "BAR4: %#.8x\n", val);
173 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_5, &val);
174 	dev_info(&dev->dev, "BAR5: %#.8x\n", val);
175 
176 	dev_info(&dev->dev, "p1 regs: %#llx, len: %#llx\n",
177 		p1_base(dev), p1_size(dev));
178 	dev_info(&dev->dev, "p2 regs: %#llx, len: %#llx\n",
179 		p2_base(dev), p2_size(dev));
180 	dev_info(&dev->dev, "BAR 4/5: %#llx, len: %#llx\n",
181 		pci_resource_start(dev, 4), pci_resource_len(dev, 4));
182 
183 	if (!(vsec = find_cxl_vsec(dev)))
184 		return;
185 
186 #define show_reg(name, what) \
187 	dev_info(&dev->dev, "cxl vsec: %30s: %#x\n", name, what)
188 
189 	pci_read_config_dword(dev, vsec + 0x0, &val);
190 	show_reg("Cap ID", (val >> 0) & 0xffff);
191 	show_reg("Cap Ver", (val >> 16) & 0xf);
192 	show_reg("Next Cap Ptr", (val >> 20) & 0xfff);
193 	pci_read_config_dword(dev, vsec + 0x4, &val);
194 	show_reg("VSEC ID", (val >> 0) & 0xffff);
195 	show_reg("VSEC Rev", (val >> 16) & 0xf);
196 	show_reg("VSEC Length",	(val >> 20) & 0xfff);
197 	pci_read_config_dword(dev, vsec + 0x8, &val);
198 	show_reg("Num AFUs", (val >> 0) & 0xff);
199 	show_reg("Status", (val >> 8) & 0xff);
200 	show_reg("Mode Control", (val >> 16) & 0xff);
201 	show_reg("Reserved", (val >> 24) & 0xff);
202 	pci_read_config_dword(dev, vsec + 0xc, &val);
203 	show_reg("PSL Rev", (val >> 0) & 0xffff);
204 	show_reg("CAIA Ver", (val >> 16) & 0xffff);
205 	pci_read_config_dword(dev, vsec + 0x10, &val);
206 	show_reg("Base Image Rev", (val >> 0) & 0xffff);
207 	show_reg("Reserved", (val >> 16) & 0x0fff);
208 	show_reg("Image Control", (val >> 28) & 0x3);
209 	show_reg("Reserved", (val >> 30) & 0x1);
210 	show_reg("Image Loaded", (val >> 31) & 0x1);
211 
212 	pci_read_config_dword(dev, vsec + 0x14, &val);
213 	show_reg("Reserved", val);
214 	pci_read_config_dword(dev, vsec + 0x18, &val);
215 	show_reg("Reserved", val);
216 	pci_read_config_dword(dev, vsec + 0x1c, &val);
217 	show_reg("Reserved", val);
218 
219 	pci_read_config_dword(dev, vsec + 0x20, &val);
220 	show_reg("AFU Descriptor Offset", val);
221 	pci_read_config_dword(dev, vsec + 0x24, &val);
222 	show_reg("AFU Descriptor Size", val);
223 	pci_read_config_dword(dev, vsec + 0x28, &val);
224 	show_reg("Problem State Offset", val);
225 	pci_read_config_dword(dev, vsec + 0x2c, &val);
226 	show_reg("Problem State Size", val);
227 
228 	pci_read_config_dword(dev, vsec + 0x30, &val);
229 	show_reg("Reserved", val);
230 	pci_read_config_dword(dev, vsec + 0x34, &val);
231 	show_reg("Reserved", val);
232 	pci_read_config_dword(dev, vsec + 0x38, &val);
233 	show_reg("Reserved", val);
234 	pci_read_config_dword(dev, vsec + 0x3c, &val);
235 	show_reg("Reserved", val);
236 
237 	pci_read_config_dword(dev, vsec + 0x40, &val);
238 	show_reg("PSL Programming Port", val);
239 	pci_read_config_dword(dev, vsec + 0x44, &val);
240 	show_reg("PSL Programming Control", val);
241 
242 	pci_read_config_dword(dev, vsec + 0x48, &val);
243 	show_reg("Reserved", val);
244 	pci_read_config_dword(dev, vsec + 0x4c, &val);
245 	show_reg("Reserved", val);
246 
247 	pci_read_config_dword(dev, vsec + 0x50, &val);
248 	show_reg("Flash Address Register", val);
249 	pci_read_config_dword(dev, vsec + 0x54, &val);
250 	show_reg("Flash Size Register", val);
251 	pci_read_config_dword(dev, vsec + 0x58, &val);
252 	show_reg("Flash Status/Control Register", val);
253 	pci_read_config_dword(dev, vsec + 0x58, &val);
254 	show_reg("Flash Data Port", val);
255 
256 #undef show_reg
257 }
258 
259 static void dump_afu_descriptor(struct cxl_afu *afu)
260 {
261 	u64 val, afu_cr_num, afu_cr_off, afu_cr_len;
262 	int i;
263 
264 #define show_reg(name, what) \
265 	dev_info(&afu->dev, "afu desc: %30s: %#llx\n", name, what)
266 
267 	val = AFUD_READ_INFO(afu);
268 	show_reg("num_ints_per_process", AFUD_NUM_INTS_PER_PROC(val));
269 	show_reg("num_of_processes", AFUD_NUM_PROCS(val));
270 	show_reg("num_of_afu_CRs", AFUD_NUM_CRS(val));
271 	show_reg("req_prog_mode", val & 0xffffULL);
272 	afu_cr_num = AFUD_NUM_CRS(val);
273 
274 	val = AFUD_READ(afu, 0x8);
275 	show_reg("Reserved", val);
276 	val = AFUD_READ(afu, 0x10);
277 	show_reg("Reserved", val);
278 	val = AFUD_READ(afu, 0x18);
279 	show_reg("Reserved", val);
280 
281 	val = AFUD_READ_CR(afu);
282 	show_reg("Reserved", (val >> (63-7)) & 0xff);
283 	show_reg("AFU_CR_len", AFUD_CR_LEN(val));
284 	afu_cr_len = AFUD_CR_LEN(val) * 256;
285 
286 	val = AFUD_READ_CR_OFF(afu);
287 	afu_cr_off = val;
288 	show_reg("AFU_CR_offset", val);
289 
290 	val = AFUD_READ_PPPSA(afu);
291 	show_reg("PerProcessPSA_control", (val >> (63-7)) & 0xff);
292 	show_reg("PerProcessPSA Length", AFUD_PPPSA_LEN(val));
293 
294 	val = AFUD_READ_PPPSA_OFF(afu);
295 	show_reg("PerProcessPSA_offset", val);
296 
297 	val = AFUD_READ_EB(afu);
298 	show_reg("Reserved", (val >> (63-7)) & 0xff);
299 	show_reg("AFU_EB_len", AFUD_EB_LEN(val));
300 
301 	val = AFUD_READ_EB_OFF(afu);
302 	show_reg("AFU_EB_offset", val);
303 
304 	for (i = 0; i < afu_cr_num; i++) {
305 		val = AFUD_READ_LE(afu, afu_cr_off + i * afu_cr_len);
306 		show_reg("CR Vendor", val & 0xffff);
307 		show_reg("CR Device", (val >> 16) & 0xffff);
308 	}
309 #undef show_reg
310 }
311 
312 #define P8_CAPP_UNIT0_ID 0xBA
313 #define P8_CAPP_UNIT1_ID 0XBE
314 #define P9_CAPP_UNIT0_ID 0xC0
315 #define P9_CAPP_UNIT1_ID 0xE0
316 
317 static int get_phb_index(struct device_node *np, u32 *phb_index)
318 {
319 	if (of_property_read_u32(np, "ibm,phb-index", phb_index))
320 		return -ENODEV;
321 	return 0;
322 }
323 
324 static u64 get_capp_unit_id(struct device_node *np, u32 phb_index)
325 {
326 	/*
327 	 * POWER 8:
328 	 *  - For chips other than POWER8NVL, we only have CAPP 0,
329 	 *    irrespective of which PHB is used.
330 	 *  - For POWER8NVL, assume CAPP 0 is attached to PHB0 and
331 	 *    CAPP 1 is attached to PHB1.
332 	 */
333 	if (cxl_is_power8()) {
334 		if (!pvr_version_is(PVR_POWER8NVL))
335 			return P8_CAPP_UNIT0_ID;
336 
337 		if (phb_index == 0)
338 			return P8_CAPP_UNIT0_ID;
339 
340 		if (phb_index == 1)
341 			return P8_CAPP_UNIT1_ID;
342 	}
343 
344 	/*
345 	 * POWER 9:
346 	 *   PEC0 (PHB0). Capp ID = CAPP0 (0b1100_0000)
347 	 *   PEC1 (PHB1 - PHB2). No capi mode
348 	 *   PEC2 (PHB3 - PHB4 - PHB5): Capi mode on PHB3 only. Capp ID = CAPP1 (0b1110_0000)
349 	 */
350 	if (cxl_is_power9()) {
351 		if (phb_index == 0)
352 			return P9_CAPP_UNIT0_ID;
353 
354 		if (phb_index == 3)
355 			return P9_CAPP_UNIT1_ID;
356 	}
357 
358 	return 0;
359 }
360 
361 int cxl_calc_capp_routing(struct pci_dev *dev, u64 *chipid,
362 			     u32 *phb_index, u64 *capp_unit_id)
363 {
364 	int rc;
365 	struct device_node *np;
366 	u32 id;
367 
368 	if (!(np = pnv_pci_get_phb_node(dev)))
369 		return -ENODEV;
370 
371 	while (np && of_property_read_u32(np, "ibm,chip-id", &id))
372 		np = of_get_next_parent(np);
373 	if (!np)
374 		return -ENODEV;
375 
376 	*chipid = id;
377 
378 	rc = get_phb_index(np, phb_index);
379 	if (rc) {
380 		pr_err("cxl: invalid phb index\n");
381 		of_node_put(np);
382 		return rc;
383 	}
384 
385 	*capp_unit_id = get_capp_unit_id(np, *phb_index);
386 	of_node_put(np);
387 	if (!*capp_unit_id) {
388 		pr_err("cxl: No capp unit found for PHB[%lld,%d]. Make sure the adapter is on a capi-compatible slot\n",
389 		       *chipid, *phb_index);
390 		return -ENODEV;
391 	}
392 
393 	return 0;
394 }
395 
396 static DEFINE_MUTEX(indications_mutex);
397 
398 static int get_phb_indications(struct pci_dev *dev, u64 *capiind, u64 *asnind,
399 			       u64 *nbwind)
400 {
401 	static u32 val[3];
402 	struct device_node *np;
403 
404 	mutex_lock(&indications_mutex);
405 	if (!val[0]) {
406 		if (!(np = pnv_pci_get_phb_node(dev))) {
407 			mutex_unlock(&indications_mutex);
408 			return -ENODEV;
409 		}
410 
411 		if (of_property_read_u32_array(np, "ibm,phb-indications", val, 3)) {
412 			val[2] = 0x0300UL; /* legacy values */
413 			val[1] = 0x0400UL;
414 			val[0] = 0x0200UL;
415 		}
416 		of_node_put(np);
417 	}
418 	*capiind = val[0];
419 	*asnind = val[1];
420 	*nbwind = val[2];
421 	mutex_unlock(&indications_mutex);
422 	return 0;
423 }
424 
425 int cxl_get_xsl9_dsnctl(struct pci_dev *dev, u64 capp_unit_id, u64 *reg)
426 {
427 	u64 xsl_dsnctl;
428 	u64 capiind, asnind, nbwind;
429 
430 	/*
431 	 * CAPI Identifier bits [0:7]
432 	 * bit 61:60 MSI bits --> 0
433 	 * bit 59 TVT selector --> 0
434 	 */
435 	if (get_phb_indications(dev, &capiind, &asnind, &nbwind))
436 		return -ENODEV;
437 
438 	/*
439 	 * Tell XSL where to route data to.
440 	 * The field chipid should match the PHB CAPI_CMPM register
441 	 */
442 	xsl_dsnctl = (capiind << (63-15)); /* Bit 57 */
443 	xsl_dsnctl |= (capp_unit_id << (63-15));
444 
445 	/* nMMU_ID Defaults to: b’000001001’*/
446 	xsl_dsnctl |= ((u64)0x09 << (63-28));
447 
448 	/*
449 	 * Used to identify CAPI packets which should be sorted into
450 	 * the Non-Blocking queues by the PHB. This field should match
451 	 * the PHB PBL_NBW_CMPM register
452 	 * nbwind=0x03, bits [57:58], must include capi indicator.
453 	 * Not supported on P9 DD1.
454 	 */
455 	xsl_dsnctl |= (nbwind << (63-55));
456 
457 	/*
458 	 * Upper 16b address bits of ASB_Notify messages sent to the
459 	 * system. Need to match the PHB’s ASN Compare/Mask Register.
460 	 * Not supported on P9 DD1.
461 	 */
462 	xsl_dsnctl |= asnind;
463 
464 	*reg = xsl_dsnctl;
465 	return 0;
466 }
467 
468 static int init_implementation_adapter_regs_psl9(struct cxl *adapter,
469 						 struct pci_dev *dev)
470 {
471 	u64 xsl_dsnctl, psl_fircntl;
472 	u64 chipid;
473 	u32 phb_index;
474 	u64 capp_unit_id;
475 	u64 psl_debug;
476 	int rc;
477 
478 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
479 	if (rc)
480 		return rc;
481 
482 	rc = cxl_get_xsl9_dsnctl(dev, capp_unit_id, &xsl_dsnctl);
483 	if (rc)
484 		return rc;
485 
486 	cxl_p1_write(adapter, CXL_XSL9_DSNCTL, xsl_dsnctl);
487 
488 	/* Set fir_cntl to recommended value for production env */
489 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
490 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
491 	psl_fircntl |= 0x1ULL; /* ce_thresh */
492 	cxl_p1_write(adapter, CXL_PSL9_FIR_CNTL, psl_fircntl);
493 
494 	/* Setup the PSL to transmit packets on the PCIe before the
495 	 * CAPP is enabled. Make sure that CAPP virtual machines are disabled
496 	 */
497 	cxl_p1_write(adapter, CXL_PSL9_DSNDCTL, 0x0001001000012A10ULL);
498 
499 	/*
500 	 * A response to an ASB_Notify request is returned by the
501 	 * system as an MMIO write to the address defined in
502 	 * the PSL_TNR_ADDR register.
503 	 * keep the Reset Value: 0x00020000E0000000
504 	 */
505 
506 	/* Enable XSL rty limit */
507 	cxl_p1_write(adapter, CXL_XSL9_DEF, 0x51F8000000000005ULL);
508 
509 	/* Change XSL_INV dummy read threshold */
510 	cxl_p1_write(adapter, CXL_XSL9_INV, 0x0000040007FFC200ULL);
511 
512 	if (phb_index == 3) {
513 		/* disable machines 31-47 and 20-27 for DMA */
514 		cxl_p1_write(adapter, CXL_PSL9_APCDEDTYPE, 0x40000FF3FFFF0000ULL);
515 	}
516 
517 	/* Snoop machines */
518 	cxl_p1_write(adapter, CXL_PSL9_APCDEDALLOC, 0x800F000200000000ULL);
519 
520 	/* Enable NORST and DD2 features */
521 	cxl_p1_write(adapter, CXL_PSL9_DEBUG, 0xC000000000000000ULL);
522 
523 	/*
524 	 * Check if PSL has data-cache. We need to flush adapter datacache
525 	 * when as its about to be removed.
526 	 */
527 	psl_debug = cxl_p1_read(adapter, CXL_PSL9_DEBUG);
528 	if (psl_debug & CXL_PSL_DEBUG_CDC) {
529 		dev_dbg(&dev->dev, "No data-cache present\n");
530 		adapter->native->no_data_cache = true;
531 	}
532 
533 	return 0;
534 }
535 
536 static int init_implementation_adapter_regs_psl8(struct cxl *adapter, struct pci_dev *dev)
537 {
538 	u64 psl_dsnctl, psl_fircntl;
539 	u64 chipid;
540 	u32 phb_index;
541 	u64 capp_unit_id;
542 	int rc;
543 
544 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
545 	if (rc)
546 		return rc;
547 
548 	psl_dsnctl = 0x0000900000000000ULL; /* pteupd ttype, scdone */
549 	psl_dsnctl |= (0x2ULL << (63-38)); /* MMIO hang pulse: 256 us */
550 	/* Tell PSL where to route data to */
551 	psl_dsnctl |= (chipid << (63-5));
552 	psl_dsnctl |= (capp_unit_id << (63-13));
553 
554 	cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
555 	cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
556 	/* snoop write mask */
557 	cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
558 	/* set fir_cntl to recommended value for production env */
559 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
560 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
561 	psl_fircntl |= 0x1ULL; /* ce_thresh */
562 	cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, psl_fircntl);
563 	/* for debugging with trace arrays */
564 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
565 
566 	return 0;
567 }
568 
569 /* PSL */
570 #define TBSYNC_CAL(n) (((u64)n & 0x7) << (63-3))
571 #define TBSYNC_CNT(n) (((u64)n & 0x7) << (63-6))
572 /* For the PSL this is a multiple for 0 < n <= 7: */
573 #define PSL_2048_250MHZ_CYCLES 1
574 
575 static void write_timebase_ctrl_psl8(struct cxl *adapter)
576 {
577 	cxl_p1_write(adapter, CXL_PSL_TB_CTLSTAT,
578 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
579 }
580 
581 static u64 timebase_read_psl9(struct cxl *adapter)
582 {
583 	return cxl_p1_read(adapter, CXL_PSL9_Timebase);
584 }
585 
586 static u64 timebase_read_psl8(struct cxl *adapter)
587 {
588 	return cxl_p1_read(adapter, CXL_PSL_Timebase);
589 }
590 
591 static void cxl_setup_psl_timebase(struct cxl *adapter, struct pci_dev *dev)
592 {
593 	struct device_node *np;
594 
595 	adapter->psl_timebase_synced = false;
596 
597 	if (!(np = pnv_pci_get_phb_node(dev)))
598 		return;
599 
600 	/* Do not fail when CAPP timebase sync is not supported by OPAL */
601 	of_node_get(np);
602 	if (!of_property_present(np, "ibm,capp-timebase-sync")) {
603 		of_node_put(np);
604 		dev_info(&dev->dev, "PSL timebase inactive: OPAL support missing\n");
605 		return;
606 	}
607 	of_node_put(np);
608 
609 	/*
610 	 * Setup PSL Timebase Control and Status register
611 	 * with the recommended Timebase Sync Count value
612 	 */
613 	if (adapter->native->sl_ops->write_timebase_ctrl)
614 		adapter->native->sl_ops->write_timebase_ctrl(adapter);
615 
616 	/* Enable PSL Timebase */
617 	cxl_p1_write(adapter, CXL_PSL_Control, 0x0000000000000000);
618 	cxl_p1_write(adapter, CXL_PSL_Control, CXL_PSL_Control_tb);
619 
620 	return;
621 }
622 
623 static int init_implementation_afu_regs_psl9(struct cxl_afu *afu)
624 {
625 	return 0;
626 }
627 
628 static int init_implementation_afu_regs_psl8(struct cxl_afu *afu)
629 {
630 	/* read/write masks for this slice */
631 	cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
632 	/* APC read/write masks for this slice */
633 	cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
634 	/* for debugging with trace arrays */
635 	cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
636 	cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
637 
638 	return 0;
639 }
640 
641 int cxl_pci_setup_irq(struct cxl *adapter, unsigned int hwirq,
642 		unsigned int virq)
643 {
644 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
645 
646 	return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
647 }
648 
649 int cxl_update_image_control(struct cxl *adapter)
650 {
651 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
652 	int rc;
653 	int vsec;
654 	u8 image_state;
655 
656 	if (!(vsec = find_cxl_vsec(dev))) {
657 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
658 		return -ENODEV;
659 	}
660 
661 	if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
662 		dev_err(&dev->dev, "failed to read image state: %i\n", rc);
663 		return rc;
664 	}
665 
666 	if (adapter->perst_loads_image)
667 		image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
668 	else
669 		image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
670 
671 	if (adapter->perst_select_user)
672 		image_state |= CXL_VSEC_PERST_SELECT_USER;
673 	else
674 		image_state &= ~CXL_VSEC_PERST_SELECT_USER;
675 
676 	if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
677 		dev_err(&dev->dev, "failed to update image control: %i\n", rc);
678 		return rc;
679 	}
680 
681 	return 0;
682 }
683 
684 int cxl_pci_alloc_one_irq(struct cxl *adapter)
685 {
686 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
687 
688 	return pnv_cxl_alloc_hwirqs(dev, 1);
689 }
690 
691 void cxl_pci_release_one_irq(struct cxl *adapter, int hwirq)
692 {
693 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
694 
695 	return pnv_cxl_release_hwirqs(dev, hwirq, 1);
696 }
697 
698 int cxl_pci_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
699 			struct cxl *adapter, unsigned int num)
700 {
701 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
702 
703 	return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
704 }
705 
706 void cxl_pci_release_irq_ranges(struct cxl_irq_ranges *irqs,
707 				struct cxl *adapter)
708 {
709 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
710 
711 	pnv_cxl_release_hwirq_ranges(irqs, dev);
712 }
713 
714 static int setup_cxl_bars(struct pci_dev *dev)
715 {
716 	/* Safety check in case we get backported to < 3.17 without M64 */
717 	if ((p1_base(dev) < 0x100000000ULL) ||
718 	    (p2_base(dev) < 0x100000000ULL)) {
719 		dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
720 		return -ENODEV;
721 	}
722 
723 	/*
724 	 * BAR 4/5 has a special meaning for CXL and must be programmed with a
725 	 * special value corresponding to the CXL protocol address range.
726 	 * For POWER 8/9 that means bits 48:49 must be set to 10
727 	 */
728 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
729 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
730 
731 	return 0;
732 }
733 
734 /* pciex node: ibm,opal-m64-window = <0x3d058 0x0 0x3d058 0x0 0x8 0x0>; */
735 static int switch_card_to_cxl(struct pci_dev *dev)
736 {
737 	int vsec;
738 	u8 val;
739 	int rc;
740 
741 	dev_info(&dev->dev, "switch card to CXL\n");
742 
743 	if (!(vsec = find_cxl_vsec(dev))) {
744 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
745 		return -ENODEV;
746 	}
747 
748 	if ((rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val))) {
749 		dev_err(&dev->dev, "failed to read current mode control: %i", rc);
750 		return rc;
751 	}
752 	val &= ~CXL_VSEC_PROTOCOL_MASK;
753 	val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
754 	if ((rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val))) {
755 		dev_err(&dev->dev, "failed to enable CXL protocol: %i", rc);
756 		return rc;
757 	}
758 	/*
759 	 * The CAIA spec (v0.12 11.6 Bi-modal Device Support) states
760 	 * we must wait 100ms after this mode switch before touching
761 	 * PCIe config space.
762 	 */
763 	msleep(100);
764 
765 	return 0;
766 }
767 
768 static int pci_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
769 {
770 	u64 p1n_base, p2n_base, afu_desc;
771 	const u64 p1n_size = 0x100;
772 	const u64 p2n_size = 0x1000;
773 
774 	p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
775 	p2n_base = p2_base(dev) + (afu->slice * p2n_size);
776 	afu->psn_phys = p2_base(dev) + (adapter->native->ps_off + (afu->slice * adapter->ps_size));
777 	afu_desc = p2_base(dev) + adapter->native->afu_desc_off + (afu->slice * adapter->native->afu_desc_size);
778 
779 	if (!(afu->native->p1n_mmio = ioremap(p1n_base, p1n_size)))
780 		goto err;
781 	if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
782 		goto err1;
783 	if (afu_desc) {
784 		if (!(afu->native->afu_desc_mmio = ioremap(afu_desc, adapter->native->afu_desc_size)))
785 			goto err2;
786 	}
787 
788 	return 0;
789 err2:
790 	iounmap(afu->p2n_mmio);
791 err1:
792 	iounmap(afu->native->p1n_mmio);
793 err:
794 	dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
795 	return -ENOMEM;
796 }
797 
798 static void pci_unmap_slice_regs(struct cxl_afu *afu)
799 {
800 	if (afu->p2n_mmio) {
801 		iounmap(afu->p2n_mmio);
802 		afu->p2n_mmio = NULL;
803 	}
804 	if (afu->native->p1n_mmio) {
805 		iounmap(afu->native->p1n_mmio);
806 		afu->native->p1n_mmio = NULL;
807 	}
808 	if (afu->native->afu_desc_mmio) {
809 		iounmap(afu->native->afu_desc_mmio);
810 		afu->native->afu_desc_mmio = NULL;
811 	}
812 }
813 
814 void cxl_pci_release_afu(struct device *dev)
815 {
816 	struct cxl_afu *afu = to_cxl_afu(dev);
817 
818 	pr_devel("%s\n", __func__);
819 
820 	idr_destroy(&afu->contexts_idr);
821 	cxl_release_spa(afu);
822 
823 	kfree(afu->native);
824 	kfree(afu);
825 }
826 
827 /* Expects AFU struct to have recently been zeroed out */
828 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
829 {
830 	u64 val;
831 
832 	val = AFUD_READ_INFO(afu);
833 	afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
834 	afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
835 	afu->crs_num = AFUD_NUM_CRS(val);
836 
837 	if (AFUD_AFU_DIRECTED(val))
838 		afu->modes_supported |= CXL_MODE_DIRECTED;
839 	if (AFUD_DEDICATED_PROCESS(val))
840 		afu->modes_supported |= CXL_MODE_DEDICATED;
841 	if (AFUD_TIME_SLICED(val))
842 		afu->modes_supported |= CXL_MODE_TIME_SLICED;
843 
844 	val = AFUD_READ_PPPSA(afu);
845 	afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
846 	afu->psa = AFUD_PPPSA_PSA(val);
847 	if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
848 		afu->native->pp_offset = AFUD_READ_PPPSA_OFF(afu);
849 
850 	val = AFUD_READ_CR(afu);
851 	afu->crs_len = AFUD_CR_LEN(val) * 256;
852 	afu->crs_offset = AFUD_READ_CR_OFF(afu);
853 
854 
855 	/* eb_len is in multiple of 4K */
856 	afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
857 	afu->eb_offset = AFUD_READ_EB_OFF(afu);
858 
859 	/* eb_off is 4K aligned so lower 12 bits are always zero */
860 	if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
861 		dev_warn(&afu->dev,
862 			 "Invalid AFU error buffer offset %Lx\n",
863 			 afu->eb_offset);
864 		dev_info(&afu->dev,
865 			 "Ignoring AFU error buffer in the descriptor\n");
866 		/* indicate that no afu buffer exists */
867 		afu->eb_len = 0;
868 	}
869 
870 	return 0;
871 }
872 
873 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
874 {
875 	int i, rc;
876 	u32 val;
877 
878 	if (afu->psa && afu->adapter->ps_size <
879 			(afu->native->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
880 		dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
881 		return -ENODEV;
882 	}
883 
884 	if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
885 		dev_warn(&afu->dev, "AFU uses pp_size(%#016llx) < PAGE_SIZE per-process PSA!\n", afu->pp_size);
886 
887 	for (i = 0; i < afu->crs_num; i++) {
888 		rc = cxl_ops->afu_cr_read32(afu, i, 0, &val);
889 		if (rc || val == 0) {
890 			dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
891 			return -EINVAL;
892 		}
893 	}
894 
895 	if ((afu->modes_supported & ~CXL_MODE_DEDICATED) && afu->max_procs_virtualised == 0) {
896 		/*
897 		 * We could also check this for the dedicated process model
898 		 * since the architecture indicates it should be set to 1, but
899 		 * in that case we ignore the value and I'd rather not risk
900 		 * breaking any existing dedicated process AFUs that left it as
901 		 * 0 (not that I'm aware of any). It is clearly an error for an
902 		 * AFU directed AFU to set this to 0, and would have previously
903 		 * triggered a bug resulting in the maximum not being enforced
904 		 * at all since idr_alloc treats 0 as no maximum.
905 		 */
906 		dev_err(&afu->dev, "AFU does not support any processes\n");
907 		return -EINVAL;
908 	}
909 
910 	return 0;
911 }
912 
913 static int sanitise_afu_regs_psl9(struct cxl_afu *afu)
914 {
915 	u64 reg;
916 
917 	/*
918 	 * Clear out any regs that contain either an IVTE or address or may be
919 	 * waiting on an acknowledgment to try to be a bit safer as we bring
920 	 * it online
921 	 */
922 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
923 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
924 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
925 		if (cxl_ops->afu_reset(afu))
926 			return -EIO;
927 		if (cxl_afu_disable(afu))
928 			return -EIO;
929 		if (cxl_psl_purge(afu))
930 			return -EIO;
931 	}
932 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
933 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
934 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
935 	if (reg) {
936 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
937 		if (reg & CXL_PSL9_DSISR_An_TF)
938 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
939 		else
940 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
941 	}
942 	if (afu->adapter->native->sl_ops->register_serr_irq) {
943 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
944 		if (reg) {
945 			if (reg & ~0x000000007fffffff)
946 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
947 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
948 		}
949 	}
950 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
951 	if (reg) {
952 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
953 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
954 	}
955 
956 	return 0;
957 }
958 
959 static int sanitise_afu_regs_psl8(struct cxl_afu *afu)
960 {
961 	u64 reg;
962 
963 	/*
964 	 * Clear out any regs that contain either an IVTE or address or may be
965 	 * waiting on an acknowledgement to try to be a bit safer as we bring
966 	 * it online
967 	 */
968 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
969 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
970 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
971 		if (cxl_ops->afu_reset(afu))
972 			return -EIO;
973 		if (cxl_afu_disable(afu))
974 			return -EIO;
975 		if (cxl_psl_purge(afu))
976 			return -EIO;
977 	}
978 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
979 	cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
980 	cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
981 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
982 	cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
983 	cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
984 	cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
985 	cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
986 	cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
987 	cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
988 	cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
989 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
990 	if (reg) {
991 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
992 		if (reg & CXL_PSL_DSISR_TRANS)
993 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
994 		else
995 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
996 	}
997 	if (afu->adapter->native->sl_ops->register_serr_irq) {
998 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
999 		if (reg) {
1000 			if (reg & ~0xffff)
1001 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1002 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1003 		}
1004 	}
1005 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1006 	if (reg) {
1007 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1008 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1009 	}
1010 
1011 	return 0;
1012 }
1013 
1014 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
1015 /*
1016  * afu_eb_read:
1017  * Called from sysfs and reads the afu error info buffer. The h/w only supports
1018  * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
1019  * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
1020  */
1021 ssize_t cxl_pci_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
1022 				loff_t off, size_t count)
1023 {
1024 	loff_t aligned_start, aligned_end;
1025 	size_t aligned_length;
1026 	void *tbuf;
1027 	const void __iomem *ebuf = afu->native->afu_desc_mmio + afu->eb_offset;
1028 
1029 	if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
1030 		return 0;
1031 
1032 	/* calculate aligned read window */
1033 	count = min((size_t)(afu->eb_len - off), count);
1034 	aligned_start = round_down(off, 8);
1035 	aligned_end = round_up(off + count, 8);
1036 	aligned_length = aligned_end - aligned_start;
1037 
1038 	/* max we can copy in one read is PAGE_SIZE */
1039 	if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
1040 		aligned_length = ERR_BUFF_MAX_COPY_SIZE;
1041 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
1042 	}
1043 
1044 	/* use bounce buffer for copy */
1045 	tbuf = (void *)__get_free_page(GFP_KERNEL);
1046 	if (!tbuf)
1047 		return -ENOMEM;
1048 
1049 	/* perform aligned read from the mmio region */
1050 	memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
1051 	memcpy(buf, tbuf + (off & 0x7), count);
1052 
1053 	free_page((unsigned long)tbuf);
1054 
1055 	return count;
1056 }
1057 
1058 static int pci_configure_afu(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1059 {
1060 	int rc;
1061 
1062 	if ((rc = pci_map_slice_regs(afu, adapter, dev)))
1063 		return rc;
1064 
1065 	if (adapter->native->sl_ops->sanitise_afu_regs) {
1066 		rc = adapter->native->sl_ops->sanitise_afu_regs(afu);
1067 		if (rc)
1068 			goto err1;
1069 	}
1070 
1071 	/* We need to reset the AFU before we can read the AFU descriptor */
1072 	if ((rc = cxl_ops->afu_reset(afu)))
1073 		goto err1;
1074 
1075 	if (cxl_verbose)
1076 		dump_afu_descriptor(afu);
1077 
1078 	if ((rc = cxl_read_afu_descriptor(afu)))
1079 		goto err1;
1080 
1081 	if ((rc = cxl_afu_descriptor_looks_ok(afu)))
1082 		goto err1;
1083 
1084 	if (adapter->native->sl_ops->afu_regs_init)
1085 		if ((rc = adapter->native->sl_ops->afu_regs_init(afu)))
1086 			goto err1;
1087 
1088 	if (adapter->native->sl_ops->register_serr_irq)
1089 		if ((rc = adapter->native->sl_ops->register_serr_irq(afu)))
1090 			goto err1;
1091 
1092 	if ((rc = cxl_native_register_psl_irq(afu)))
1093 		goto err2;
1094 
1095 	atomic_set(&afu->configured_state, 0);
1096 	return 0;
1097 
1098 err2:
1099 	if (adapter->native->sl_ops->release_serr_irq)
1100 		adapter->native->sl_ops->release_serr_irq(afu);
1101 err1:
1102 	pci_unmap_slice_regs(afu);
1103 	return rc;
1104 }
1105 
1106 static void pci_deconfigure_afu(struct cxl_afu *afu)
1107 {
1108 	/*
1109 	 * It's okay to deconfigure when AFU is already locked, otherwise wait
1110 	 * until there are no readers
1111 	 */
1112 	if (atomic_read(&afu->configured_state) != -1) {
1113 		while (atomic_cmpxchg(&afu->configured_state, 0, -1) != -1)
1114 			schedule();
1115 	}
1116 	cxl_native_release_psl_irq(afu);
1117 	if (afu->adapter->native->sl_ops->release_serr_irq)
1118 		afu->adapter->native->sl_ops->release_serr_irq(afu);
1119 	pci_unmap_slice_regs(afu);
1120 }
1121 
1122 static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
1123 {
1124 	struct cxl_afu *afu;
1125 	int rc = -ENOMEM;
1126 
1127 	afu = cxl_alloc_afu(adapter, slice);
1128 	if (!afu)
1129 		return -ENOMEM;
1130 
1131 	afu->native = kzalloc(sizeof(struct cxl_afu_native), GFP_KERNEL);
1132 	if (!afu->native)
1133 		goto err_free_afu;
1134 
1135 	mutex_init(&afu->native->spa_mutex);
1136 
1137 	rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice);
1138 	if (rc)
1139 		goto err_free_native;
1140 
1141 	rc = pci_configure_afu(afu, adapter, dev);
1142 	if (rc)
1143 		goto err_free_native;
1144 
1145 	/* Don't care if this fails */
1146 	cxl_debugfs_afu_add(afu);
1147 
1148 	/*
1149 	 * After we call this function we must not free the afu directly, even
1150 	 * if it returns an error!
1151 	 */
1152 	if ((rc = cxl_register_afu(afu)))
1153 		goto err_put_dev;
1154 
1155 	if ((rc = cxl_sysfs_afu_add(afu)))
1156 		goto err_del_dev;
1157 
1158 	adapter->afu[afu->slice] = afu;
1159 
1160 	if ((rc = cxl_pci_vphb_add(afu)))
1161 		dev_info(&afu->dev, "Can't register vPHB\n");
1162 
1163 	return 0;
1164 
1165 err_del_dev:
1166 	device_del(&afu->dev);
1167 err_put_dev:
1168 	pci_deconfigure_afu(afu);
1169 	cxl_debugfs_afu_remove(afu);
1170 	put_device(&afu->dev);
1171 	return rc;
1172 
1173 err_free_native:
1174 	kfree(afu->native);
1175 err_free_afu:
1176 	kfree(afu);
1177 	return rc;
1178 
1179 }
1180 
1181 static void cxl_pci_remove_afu(struct cxl_afu *afu)
1182 {
1183 	pr_devel("%s\n", __func__);
1184 
1185 	if (!afu)
1186 		return;
1187 
1188 	cxl_pci_vphb_remove(afu);
1189 	cxl_sysfs_afu_remove(afu);
1190 	cxl_debugfs_afu_remove(afu);
1191 
1192 	spin_lock(&afu->adapter->afu_list_lock);
1193 	afu->adapter->afu[afu->slice] = NULL;
1194 	spin_unlock(&afu->adapter->afu_list_lock);
1195 
1196 	cxl_context_detach_all(afu);
1197 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1198 
1199 	pci_deconfigure_afu(afu);
1200 	device_unregister(&afu->dev);
1201 }
1202 
1203 int cxl_pci_reset(struct cxl *adapter)
1204 {
1205 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1206 	int rc;
1207 
1208 	if (adapter->perst_same_image) {
1209 		dev_warn(&dev->dev,
1210 			 "cxl: refusing to reset/reflash when perst_reloads_same_image is set.\n");
1211 		return -EINVAL;
1212 	}
1213 
1214 	dev_info(&dev->dev, "CXL reset\n");
1215 
1216 	/*
1217 	 * The adapter is about to be reset, so ignore errors.
1218 	 */
1219 	cxl_data_cache_flush(adapter);
1220 
1221 	/* pcie_warm_reset requests a fundamental pci reset which includes a
1222 	 * PERST assert/deassert.  PERST triggers a loading of the image
1223 	 * if "user" or "factory" is selected in sysfs */
1224 	if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
1225 		dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
1226 		return rc;
1227 	}
1228 
1229 	return rc;
1230 }
1231 
1232 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
1233 {
1234 	if (pci_request_region(dev, 2, "priv 2 regs"))
1235 		goto err1;
1236 	if (pci_request_region(dev, 0, "priv 1 regs"))
1237 		goto err2;
1238 
1239 	pr_devel("cxl_map_adapter_regs: p1: %#016llx %#llx, p2: %#016llx %#llx",
1240 			p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
1241 
1242 	if (!(adapter->native->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
1243 		goto err3;
1244 
1245 	if (!(adapter->native->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
1246 		goto err4;
1247 
1248 	return 0;
1249 
1250 err4:
1251 	iounmap(adapter->native->p1_mmio);
1252 	adapter->native->p1_mmio = NULL;
1253 err3:
1254 	pci_release_region(dev, 0);
1255 err2:
1256 	pci_release_region(dev, 2);
1257 err1:
1258 	return -ENOMEM;
1259 }
1260 
1261 static void cxl_unmap_adapter_regs(struct cxl *adapter)
1262 {
1263 	if (adapter->native->p1_mmio) {
1264 		iounmap(adapter->native->p1_mmio);
1265 		adapter->native->p1_mmio = NULL;
1266 		pci_release_region(to_pci_dev(adapter->dev.parent), 2);
1267 	}
1268 	if (adapter->native->p2_mmio) {
1269 		iounmap(adapter->native->p2_mmio);
1270 		adapter->native->p2_mmio = NULL;
1271 		pci_release_region(to_pci_dev(adapter->dev.parent), 0);
1272 	}
1273 }
1274 
1275 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
1276 {
1277 	int vsec;
1278 	u32 afu_desc_off, afu_desc_size;
1279 	u32 ps_off, ps_size;
1280 	u16 vseclen;
1281 	u8 image_state;
1282 
1283 	if (!(vsec = find_cxl_vsec(dev))) {
1284 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
1285 		return -ENODEV;
1286 	}
1287 
1288 	CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
1289 	if (vseclen < CXL_VSEC_MIN_SIZE) {
1290 		dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
1291 		return -EINVAL;
1292 	}
1293 
1294 	CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
1295 	CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
1296 	CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
1297 	CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
1298 	CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
1299 	CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
1300 	adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1301 	adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1302 	adapter->perst_loads_image = !!(image_state & CXL_VSEC_PERST_LOADS_IMAGE);
1303 
1304 	CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
1305 	CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
1306 	CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
1307 	CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
1308 	CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
1309 
1310 	/* Convert everything to bytes, because there is NO WAY I'd look at the
1311 	 * code a month later and forget what units these are in ;-) */
1312 	adapter->native->ps_off = ps_off * 64 * 1024;
1313 	adapter->ps_size = ps_size * 64 * 1024;
1314 	adapter->native->afu_desc_off = afu_desc_off * 64 * 1024;
1315 	adapter->native->afu_desc_size = afu_desc_size * 64 * 1024;
1316 
1317 	/* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
1318 	adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
1319 
1320 	return 0;
1321 }
1322 
1323 /*
1324  * Workaround a PCIe Host Bridge defect on some cards, that can cause
1325  * malformed Transaction Layer Packet (TLP) errors to be erroneously
1326  * reported. Mask this error in the Uncorrectable Error Mask Register.
1327  *
1328  * The upper nibble of the PSL revision is used to distinguish between
1329  * different cards. The affected ones have it set to 0.
1330  */
1331 static void cxl_fixup_malformed_tlp(struct cxl *adapter, struct pci_dev *dev)
1332 {
1333 	int aer;
1334 	u32 data;
1335 
1336 	if (adapter->psl_rev & 0xf000)
1337 		return;
1338 	if (!(aer = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR)))
1339 		return;
1340 	pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &data);
1341 	if (data & PCI_ERR_UNC_MALF_TLP)
1342 		if (data & PCI_ERR_UNC_INTN)
1343 			return;
1344 	data |= PCI_ERR_UNC_MALF_TLP;
1345 	data |= PCI_ERR_UNC_INTN;
1346 	pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, data);
1347 }
1348 
1349 static bool cxl_compatible_caia_version(struct cxl *adapter)
1350 {
1351 	if (cxl_is_power8() && (adapter->caia_major == 1))
1352 		return true;
1353 
1354 	if (cxl_is_power9() && (adapter->caia_major == 2))
1355 		return true;
1356 
1357 	return false;
1358 }
1359 
1360 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
1361 {
1362 	if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
1363 		return -EBUSY;
1364 
1365 	if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
1366 		dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
1367 		return -EINVAL;
1368 	}
1369 
1370 	if (!cxl_compatible_caia_version(adapter)) {
1371 		dev_info(&dev->dev, "Ignoring card. PSL type is not supported (caia version: %d)\n",
1372 			 adapter->caia_major);
1373 		return -ENODEV;
1374 	}
1375 
1376 	if (!adapter->slices) {
1377 		/* Once we support dynamic reprogramming we can use the card if
1378 		 * it supports loadable AFUs */
1379 		dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
1380 		return -EINVAL;
1381 	}
1382 
1383 	if (!adapter->native->afu_desc_off || !adapter->native->afu_desc_size) {
1384 		dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
1385 		return -EINVAL;
1386 	}
1387 
1388 	if (adapter->ps_size > p2_size(dev) - adapter->native->ps_off) {
1389 		dev_err(&dev->dev, "ABORTING: Problem state size larger than "
1390 				   "available in BAR2: 0x%llx > 0x%llx\n",
1391 			 adapter->ps_size, p2_size(dev) - adapter->native->ps_off);
1392 		return -EINVAL;
1393 	}
1394 
1395 	return 0;
1396 }
1397 
1398 ssize_t cxl_pci_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1399 {
1400 	return pci_read_vpd(to_pci_dev(adapter->dev.parent), 0, len, buf);
1401 }
1402 
1403 static void cxl_release_adapter(struct device *dev)
1404 {
1405 	struct cxl *adapter = to_cxl_adapter(dev);
1406 
1407 	pr_devel("cxl_release_adapter\n");
1408 
1409 	cxl_remove_adapter_nr(adapter);
1410 
1411 	kfree(adapter->native);
1412 	kfree(adapter);
1413 }
1414 
1415 #define CXL_PSL_ErrIVTE_tberror (0x1ull << (63-31))
1416 
1417 static int sanitise_adapter_regs(struct cxl *adapter)
1418 {
1419 	int rc = 0;
1420 
1421 	/* Clear PSL tberror bit by writing 1 to it */
1422 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, CXL_PSL_ErrIVTE_tberror);
1423 
1424 	if (adapter->native->sl_ops->invalidate_all) {
1425 		/* do not invalidate ERAT entries when not reloading on PERST */
1426 		if (cxl_is_power9() && (adapter->perst_loads_image))
1427 			return 0;
1428 		rc = adapter->native->sl_ops->invalidate_all(adapter);
1429 	}
1430 
1431 	return rc;
1432 }
1433 
1434 /* This should contain *only* operations that can safely be done in
1435  * both creation and recovery.
1436  */
1437 static int cxl_configure_adapter(struct cxl *adapter, struct pci_dev *dev)
1438 {
1439 	int rc;
1440 
1441 	adapter->dev.parent = &dev->dev;
1442 	adapter->dev.release = cxl_release_adapter;
1443 	pci_set_drvdata(dev, adapter);
1444 
1445 	rc = pci_enable_device(dev);
1446 	if (rc) {
1447 		dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1448 		return rc;
1449 	}
1450 
1451 	if ((rc = cxl_read_vsec(adapter, dev)))
1452 		return rc;
1453 
1454 	if ((rc = cxl_vsec_looks_ok(adapter, dev)))
1455 	        return rc;
1456 
1457 	cxl_fixup_malformed_tlp(adapter, dev);
1458 
1459 	if ((rc = setup_cxl_bars(dev)))
1460 		return rc;
1461 
1462 	if ((rc = switch_card_to_cxl(dev)))
1463 		return rc;
1464 
1465 	if ((rc = cxl_update_image_control(adapter)))
1466 		return rc;
1467 
1468 	if ((rc = cxl_map_adapter_regs(adapter, dev)))
1469 		return rc;
1470 
1471 	if ((rc = sanitise_adapter_regs(adapter)))
1472 		goto err;
1473 
1474 	if ((rc = adapter->native->sl_ops->adapter_regs_init(adapter, dev)))
1475 		goto err;
1476 
1477 	/* Required for devices using CAPP DMA mode, harmless for others */
1478 	pci_set_master(dev);
1479 
1480 	adapter->tunneled_ops_supported = false;
1481 
1482 	if (cxl_is_power9()) {
1483 		if (pnv_pci_set_tunnel_bar(dev, 0x00020000E0000000ull, 1))
1484 			dev_info(&dev->dev, "Tunneled operations unsupported\n");
1485 		else
1486 			adapter->tunneled_ops_supported = true;
1487 	}
1488 
1489 	if ((rc = pnv_phb_to_cxl_mode(dev, adapter->native->sl_ops->capi_mode)))
1490 		goto err;
1491 
1492 	/* If recovery happened, the last step is to turn on snooping.
1493 	 * In the non-recovery case this has no effect */
1494 	if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON)))
1495 		goto err;
1496 
1497 	/* Ignore error, adapter init is not dependant on timebase sync */
1498 	cxl_setup_psl_timebase(adapter, dev);
1499 
1500 	if ((rc = cxl_native_register_psl_err_irq(adapter)))
1501 		goto err;
1502 
1503 	return 0;
1504 
1505 err:
1506 	cxl_unmap_adapter_regs(adapter);
1507 	return rc;
1508 
1509 }
1510 
1511 static void cxl_deconfigure_adapter(struct cxl *adapter)
1512 {
1513 	struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1514 
1515 	if (cxl_is_power9())
1516 		pnv_pci_set_tunnel_bar(pdev, 0x00020000E0000000ull, 0);
1517 
1518 	cxl_native_release_psl_err_irq(adapter);
1519 	cxl_unmap_adapter_regs(adapter);
1520 
1521 	pci_disable_device(pdev);
1522 }
1523 
1524 static void cxl_stop_trace_psl9(struct cxl *adapter)
1525 {
1526 	int traceid;
1527 	u64 trace_state, trace_mask;
1528 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1529 
1530 	/* read each tracearray state and issue mmio to stop them is needed */
1531 	for (traceid = 0; traceid <= CXL_PSL9_TRACEID_MAX; ++traceid) {
1532 		trace_state = cxl_p1_read(adapter, CXL_PSL9_CTCCFG);
1533 		trace_mask = (0x3ULL << (62 - traceid * 2));
1534 		trace_state = (trace_state & trace_mask) >> (62 - traceid * 2);
1535 		dev_dbg(&dev->dev, "cxl: Traceid-%d trace_state=0x%0llX\n",
1536 			traceid, trace_state);
1537 
1538 		/* issue mmio if the trace array isn't in FIN state */
1539 		if (trace_state != CXL_PSL9_TRACESTATE_FIN)
1540 			cxl_p1_write(adapter, CXL_PSL9_TRACECFG,
1541 				     0x8400000000000000ULL | traceid);
1542 	}
1543 }
1544 
1545 static void cxl_stop_trace_psl8(struct cxl *adapter)
1546 {
1547 	int slice;
1548 
1549 	/* Stop the trace */
1550 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x8000000000000017LL);
1551 
1552 	/* Stop the slice traces */
1553 	spin_lock(&adapter->afu_list_lock);
1554 	for (slice = 0; slice < adapter->slices; slice++) {
1555 		if (adapter->afu[slice])
1556 			cxl_p1n_write(adapter->afu[slice], CXL_PSL_SLICE_TRACE,
1557 				      0x8000000000000000LL);
1558 	}
1559 	spin_unlock(&adapter->afu_list_lock);
1560 }
1561 
1562 static const struct cxl_service_layer_ops psl9_ops = {
1563 	.adapter_regs_init = init_implementation_adapter_regs_psl9,
1564 	.invalidate_all = cxl_invalidate_all_psl9,
1565 	.afu_regs_init = init_implementation_afu_regs_psl9,
1566 	.sanitise_afu_regs = sanitise_afu_regs_psl9,
1567 	.register_serr_irq = cxl_native_register_serr_irq,
1568 	.release_serr_irq = cxl_native_release_serr_irq,
1569 	.handle_interrupt = cxl_irq_psl9,
1570 	.fail_irq = cxl_fail_irq_psl,
1571 	.activate_dedicated_process = cxl_activate_dedicated_process_psl9,
1572 	.attach_afu_directed = cxl_attach_afu_directed_psl9,
1573 	.attach_dedicated_process = cxl_attach_dedicated_process_psl9,
1574 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl9,
1575 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl9,
1576 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl9,
1577 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl9,
1578 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl9,
1579 	.debugfs_stop_trace = cxl_stop_trace_psl9,
1580 	.timebase_read = timebase_read_psl9,
1581 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1582 	.needs_reset_before_disable = true,
1583 };
1584 
1585 static const struct cxl_service_layer_ops psl8_ops = {
1586 	.adapter_regs_init = init_implementation_adapter_regs_psl8,
1587 	.invalidate_all = cxl_invalidate_all_psl8,
1588 	.afu_regs_init = init_implementation_afu_regs_psl8,
1589 	.sanitise_afu_regs = sanitise_afu_regs_psl8,
1590 	.register_serr_irq = cxl_native_register_serr_irq,
1591 	.release_serr_irq = cxl_native_release_serr_irq,
1592 	.handle_interrupt = cxl_irq_psl8,
1593 	.fail_irq = cxl_fail_irq_psl,
1594 	.activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1595 	.attach_afu_directed = cxl_attach_afu_directed_psl8,
1596 	.attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1597 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1598 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl8,
1599 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl8,
1600 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl8,
1601 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl8,
1602 	.debugfs_stop_trace = cxl_stop_trace_psl8,
1603 	.write_timebase_ctrl = write_timebase_ctrl_psl8,
1604 	.timebase_read = timebase_read_psl8,
1605 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1606 	.needs_reset_before_disable = true,
1607 };
1608 
1609 static void set_sl_ops(struct cxl *adapter, struct pci_dev *dev)
1610 {
1611 	if (cxl_is_power8()) {
1612 		dev_info(&dev->dev, "Device uses a PSL8\n");
1613 		adapter->native->sl_ops = &psl8_ops;
1614 	} else {
1615 		dev_info(&dev->dev, "Device uses a PSL9\n");
1616 		adapter->native->sl_ops = &psl9_ops;
1617 	}
1618 }
1619 
1620 
1621 static struct cxl *cxl_pci_init_adapter(struct pci_dev *dev)
1622 {
1623 	struct cxl *adapter;
1624 	int rc;
1625 
1626 	adapter = cxl_alloc_adapter();
1627 	if (!adapter)
1628 		return ERR_PTR(-ENOMEM);
1629 
1630 	adapter->native = kzalloc(sizeof(struct cxl_native), GFP_KERNEL);
1631 	if (!adapter->native) {
1632 		rc = -ENOMEM;
1633 		goto err_release;
1634 	}
1635 
1636 	set_sl_ops(adapter, dev);
1637 
1638 	/* Set defaults for parameters which need to persist over
1639 	 * configure/reconfigure
1640 	 */
1641 	adapter->perst_loads_image = true;
1642 	adapter->perst_same_image = false;
1643 
1644 	rc = cxl_configure_adapter(adapter, dev);
1645 	if (rc) {
1646 		pci_disable_device(dev);
1647 		goto err_release;
1648 	}
1649 
1650 	/* Don't care if this one fails: */
1651 	cxl_debugfs_adapter_add(adapter);
1652 
1653 	/*
1654 	 * After we call this function we must not free the adapter directly,
1655 	 * even if it returns an error!
1656 	 */
1657 	if ((rc = cxl_register_adapter(adapter)))
1658 		goto err_put_dev;
1659 
1660 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1661 		goto err_del_dev;
1662 
1663 	/* Release the context lock as adapter is configured */
1664 	cxl_adapter_context_unlock(adapter);
1665 
1666 	return adapter;
1667 
1668 err_del_dev:
1669 	device_del(&adapter->dev);
1670 err_put_dev:
1671 	/* This should mirror cxl_remove_adapter, except without the
1672 	 * sysfs parts
1673 	 */
1674 	cxl_debugfs_adapter_remove(adapter);
1675 	cxl_deconfigure_adapter(adapter);
1676 	put_device(&adapter->dev);
1677 	return ERR_PTR(rc);
1678 
1679 err_release:
1680 	cxl_release_adapter(&adapter->dev);
1681 	return ERR_PTR(rc);
1682 }
1683 
1684 static void cxl_pci_remove_adapter(struct cxl *adapter)
1685 {
1686 	pr_devel("cxl_remove_adapter\n");
1687 
1688 	cxl_sysfs_adapter_remove(adapter);
1689 	cxl_debugfs_adapter_remove(adapter);
1690 
1691 	/*
1692 	 * Flush adapter datacache as its about to be removed.
1693 	 */
1694 	cxl_data_cache_flush(adapter);
1695 
1696 	cxl_deconfigure_adapter(adapter);
1697 
1698 	device_unregister(&adapter->dev);
1699 }
1700 
1701 #define CXL_MAX_PCIEX_PARENT 2
1702 
1703 int cxl_slot_is_switched(struct pci_dev *dev)
1704 {
1705 	struct device_node *np;
1706 	int depth = 0;
1707 
1708 	if (!(np = pci_device_to_OF_node(dev))) {
1709 		pr_err("cxl: np = NULL\n");
1710 		return -ENODEV;
1711 	}
1712 	of_node_get(np);
1713 	while (np) {
1714 		np = of_get_next_parent(np);
1715 		if (!of_node_is_type(np, "pciex"))
1716 			break;
1717 		depth++;
1718 	}
1719 	of_node_put(np);
1720 	return (depth > CXL_MAX_PCIEX_PARENT);
1721 }
1722 
1723 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
1724 {
1725 	struct cxl *adapter;
1726 	int slice;
1727 	int rc;
1728 
1729 	if (cxl_pci_is_vphb_device(dev)) {
1730 		dev_dbg(&dev->dev, "cxl_init_adapter: Ignoring cxl vphb device\n");
1731 		return -ENODEV;
1732 	}
1733 
1734 	if (cxl_slot_is_switched(dev)) {
1735 		dev_info(&dev->dev, "Ignoring card on incompatible PCI slot\n");
1736 		return -ENODEV;
1737 	}
1738 
1739 	if (cxl_is_power9() && !radix_enabled()) {
1740 		dev_info(&dev->dev, "Only Radix mode supported\n");
1741 		return -ENODEV;
1742 	}
1743 
1744 	if (cxl_verbose)
1745 		dump_cxl_config_space(dev);
1746 
1747 	adapter = cxl_pci_init_adapter(dev);
1748 	if (IS_ERR(adapter)) {
1749 		dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
1750 		return PTR_ERR(adapter);
1751 	}
1752 
1753 	for (slice = 0; slice < adapter->slices; slice++) {
1754 		if ((rc = pci_init_afu(adapter, slice, dev))) {
1755 			dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
1756 			continue;
1757 		}
1758 
1759 		rc = cxl_afu_select_best_mode(adapter->afu[slice]);
1760 		if (rc)
1761 			dev_err(&dev->dev, "AFU %i failed to start: %i\n", slice, rc);
1762 	}
1763 
1764 	return 0;
1765 }
1766 
1767 static void cxl_remove(struct pci_dev *dev)
1768 {
1769 	struct cxl *adapter = pci_get_drvdata(dev);
1770 	struct cxl_afu *afu;
1771 	int i;
1772 
1773 	/*
1774 	 * Lock to prevent someone grabbing a ref through the adapter list as
1775 	 * we are removing it
1776 	 */
1777 	for (i = 0; i < adapter->slices; i++) {
1778 		afu = adapter->afu[i];
1779 		cxl_pci_remove_afu(afu);
1780 	}
1781 	cxl_pci_remove_adapter(adapter);
1782 }
1783 
1784 static pci_ers_result_t cxl_vphb_error_detected(struct cxl_afu *afu,
1785 						pci_channel_state_t state)
1786 {
1787 	struct pci_dev *afu_dev;
1788 	struct pci_driver *afu_drv;
1789 	const struct pci_error_handlers *err_handler;
1790 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
1791 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
1792 
1793 	/* There should only be one entry, but go through the list
1794 	 * anyway
1795 	 */
1796 	if (afu == NULL || afu->phb == NULL)
1797 		return result;
1798 
1799 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
1800 		afu_drv = to_pci_driver(afu_dev->dev.driver);
1801 		if (!afu_drv)
1802 			continue;
1803 
1804 		afu_dev->error_state = state;
1805 
1806 		err_handler = afu_drv->err_handler;
1807 		if (err_handler)
1808 			afu_result = err_handler->error_detected(afu_dev,
1809 								 state);
1810 		/* Disconnect trumps all, NONE trumps NEED_RESET */
1811 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
1812 			result = PCI_ERS_RESULT_DISCONNECT;
1813 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
1814 			 (result == PCI_ERS_RESULT_NEED_RESET))
1815 			result = PCI_ERS_RESULT_NONE;
1816 	}
1817 	return result;
1818 }
1819 
1820 static pci_ers_result_t cxl_pci_error_detected(struct pci_dev *pdev,
1821 					       pci_channel_state_t state)
1822 {
1823 	struct cxl *adapter = pci_get_drvdata(pdev);
1824 	struct cxl_afu *afu;
1825 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
1826 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
1827 	int i;
1828 
1829 	/* At this point, we could still have an interrupt pending.
1830 	 * Let's try to get them out of the way before they do
1831 	 * anything we don't like.
1832 	 */
1833 	schedule();
1834 
1835 	/* If we're permanently dead, give up. */
1836 	if (state == pci_channel_io_perm_failure) {
1837 		spin_lock(&adapter->afu_list_lock);
1838 		for (i = 0; i < adapter->slices; i++) {
1839 			afu = adapter->afu[i];
1840 			/*
1841 			 * Tell the AFU drivers; but we don't care what they
1842 			 * say, we're going away.
1843 			 */
1844 			cxl_vphb_error_detected(afu, state);
1845 		}
1846 		spin_unlock(&adapter->afu_list_lock);
1847 		return PCI_ERS_RESULT_DISCONNECT;
1848 	}
1849 
1850 	/* Are we reflashing?
1851 	 *
1852 	 * If we reflash, we could come back as something entirely
1853 	 * different, including a non-CAPI card. As such, by default
1854 	 * we don't participate in the process. We'll be unbound and
1855 	 * the slot re-probed. (TODO: check EEH doesn't blindly rebind
1856 	 * us!)
1857 	 *
1858 	 * However, this isn't the entire story: for reliablity
1859 	 * reasons, we usually want to reflash the FPGA on PERST in
1860 	 * order to get back to a more reliable known-good state.
1861 	 *
1862 	 * This causes us a bit of a problem: if we reflash we can't
1863 	 * trust that we'll come back the same - we could have a new
1864 	 * image and been PERSTed in order to load that
1865 	 * image. However, most of the time we actually *will* come
1866 	 * back the same - for example a regular EEH event.
1867 	 *
1868 	 * Therefore, we allow the user to assert that the image is
1869 	 * indeed the same and that we should continue on into EEH
1870 	 * anyway.
1871 	 */
1872 	if (adapter->perst_loads_image && !adapter->perst_same_image) {
1873 		/* TODO take the PHB out of CXL mode */
1874 		dev_info(&pdev->dev, "reflashing, so opting out of EEH!\n");
1875 		return PCI_ERS_RESULT_NONE;
1876 	}
1877 
1878 	/*
1879 	 * At this point, we want to try to recover.  We'll always
1880 	 * need a complete slot reset: we don't trust any other reset.
1881 	 *
1882 	 * Now, we go through each AFU:
1883 	 *  - We send the driver, if bound, an error_detected callback.
1884 	 *    We expect it to clean up, but it can also tell us to give
1885 	 *    up and permanently detach the card. To simplify things, if
1886 	 *    any bound AFU driver doesn't support EEH, we give up on EEH.
1887 	 *
1888 	 *  - We detach all contexts associated with the AFU. This
1889 	 *    does not free them, but puts them into a CLOSED state
1890 	 *    which causes any the associated files to return useful
1891 	 *    errors to userland. It also unmaps, but does not free,
1892 	 *    any IRQs.
1893 	 *
1894 	 *  - We clean up our side: releasing and unmapping resources we hold
1895 	 *    so we can wire them up again when the hardware comes back up.
1896 	 *
1897 	 * Driver authors should note:
1898 	 *
1899 	 *  - Any contexts you create in your kernel driver (except
1900 	 *    those associated with anonymous file descriptors) are
1901 	 *    your responsibility to free and recreate. Likewise with
1902 	 *    any attached resources.
1903 	 *
1904 	 *  - We will take responsibility for re-initialising the
1905 	 *    device context (the one set up for you in
1906 	 *    cxl_pci_enable_device_hook and accessed through
1907 	 *    cxl_get_context). If you've attached IRQs or other
1908 	 *    resources to it, they remains yours to free.
1909 	 *
1910 	 * You can call the same functions to release resources as you
1911 	 * normally would: we make sure that these functions continue
1912 	 * to work when the hardware is down.
1913 	 *
1914 	 * Two examples:
1915 	 *
1916 	 * 1) If you normally free all your resources at the end of
1917 	 *    each request, or if you use anonymous FDs, your
1918 	 *    error_detected callback can simply set a flag to tell
1919 	 *    your driver not to start any new calls. You can then
1920 	 *    clear the flag in the resume callback.
1921 	 *
1922 	 * 2) If you normally allocate your resources on startup:
1923 	 *     * Set a flag in error_detected as above.
1924 	 *     * Let CXL detach your contexts.
1925 	 *     * In slot_reset, free the old resources and allocate new ones.
1926 	 *     * In resume, clear the flag to allow things to start.
1927 	 */
1928 
1929 	/* Make sure no one else changes the afu list */
1930 	spin_lock(&adapter->afu_list_lock);
1931 
1932 	for (i = 0; i < adapter->slices; i++) {
1933 		afu = adapter->afu[i];
1934 
1935 		if (afu == NULL)
1936 			continue;
1937 
1938 		afu_result = cxl_vphb_error_detected(afu, state);
1939 		cxl_context_detach_all(afu);
1940 		cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1941 		pci_deconfigure_afu(afu);
1942 
1943 		/* Disconnect trumps all, NONE trumps NEED_RESET */
1944 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
1945 			result = PCI_ERS_RESULT_DISCONNECT;
1946 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
1947 			 (result == PCI_ERS_RESULT_NEED_RESET))
1948 			result = PCI_ERS_RESULT_NONE;
1949 	}
1950 	spin_unlock(&adapter->afu_list_lock);
1951 
1952 	/* should take the context lock here */
1953 	if (cxl_adapter_context_lock(adapter) != 0)
1954 		dev_warn(&adapter->dev,
1955 			 "Couldn't take context lock with %d active-contexts\n",
1956 			 atomic_read(&adapter->contexts_num));
1957 
1958 	cxl_deconfigure_adapter(adapter);
1959 
1960 	return result;
1961 }
1962 
1963 static pci_ers_result_t cxl_pci_slot_reset(struct pci_dev *pdev)
1964 {
1965 	struct cxl *adapter = pci_get_drvdata(pdev);
1966 	struct cxl_afu *afu;
1967 	struct cxl_context *ctx;
1968 	struct pci_dev *afu_dev;
1969 	struct pci_driver *afu_drv;
1970 	const struct pci_error_handlers *err_handler;
1971 	pci_ers_result_t afu_result = PCI_ERS_RESULT_RECOVERED;
1972 	pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
1973 	int i;
1974 
1975 	if (cxl_configure_adapter(adapter, pdev))
1976 		goto err;
1977 
1978 	/*
1979 	 * Unlock context activation for the adapter. Ideally this should be
1980 	 * done in cxl_pci_resume but cxlflash module tries to activate the
1981 	 * master context as part of slot_reset callback.
1982 	 */
1983 	cxl_adapter_context_unlock(adapter);
1984 
1985 	spin_lock(&adapter->afu_list_lock);
1986 	for (i = 0; i < adapter->slices; i++) {
1987 		afu = adapter->afu[i];
1988 
1989 		if (afu == NULL)
1990 			continue;
1991 
1992 		if (pci_configure_afu(afu, adapter, pdev))
1993 			goto err_unlock;
1994 
1995 		if (cxl_afu_select_best_mode(afu))
1996 			goto err_unlock;
1997 
1998 		if (afu->phb == NULL)
1999 			continue;
2000 
2001 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2002 			/* Reset the device context.
2003 			 * TODO: make this less disruptive
2004 			 */
2005 			ctx = cxl_get_context(afu_dev);
2006 
2007 			if (ctx && cxl_release_context(ctx))
2008 				goto err_unlock;
2009 
2010 			ctx = cxl_dev_context_init(afu_dev);
2011 			if (IS_ERR(ctx))
2012 				goto err_unlock;
2013 
2014 			afu_dev->dev.archdata.cxl_ctx = ctx;
2015 
2016 			if (cxl_ops->afu_check_and_enable(afu))
2017 				goto err_unlock;
2018 
2019 			afu_dev->error_state = pci_channel_io_normal;
2020 
2021 			/* If there's a driver attached, allow it to
2022 			 * chime in on recovery. Drivers should check
2023 			 * if everything has come back OK, but
2024 			 * shouldn't start new work until we call
2025 			 * their resume function.
2026 			 */
2027 			afu_drv = to_pci_driver(afu_dev->dev.driver);
2028 			if (!afu_drv)
2029 				continue;
2030 
2031 			err_handler = afu_drv->err_handler;
2032 			if (err_handler && err_handler->slot_reset)
2033 				afu_result = err_handler->slot_reset(afu_dev);
2034 
2035 			if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2036 				result = PCI_ERS_RESULT_DISCONNECT;
2037 		}
2038 	}
2039 
2040 	spin_unlock(&adapter->afu_list_lock);
2041 	return result;
2042 
2043 err_unlock:
2044 	spin_unlock(&adapter->afu_list_lock);
2045 
2046 err:
2047 	/* All the bits that happen in both error_detected and cxl_remove
2048 	 * should be idempotent, so we don't need to worry about leaving a mix
2049 	 * of unconfigured and reconfigured resources.
2050 	 */
2051 	dev_err(&pdev->dev, "EEH recovery failed. Asking to be disconnected.\n");
2052 	return PCI_ERS_RESULT_DISCONNECT;
2053 }
2054 
2055 static void cxl_pci_resume(struct pci_dev *pdev)
2056 {
2057 	struct cxl *adapter = pci_get_drvdata(pdev);
2058 	struct cxl_afu *afu;
2059 	struct pci_dev *afu_dev;
2060 	struct pci_driver *afu_drv;
2061 	const struct pci_error_handlers *err_handler;
2062 	int i;
2063 
2064 	/* Everything is back now. Drivers should restart work now.
2065 	 * This is not the place to be checking if everything came back up
2066 	 * properly, because there's no return value: do that in slot_reset.
2067 	 */
2068 	spin_lock(&adapter->afu_list_lock);
2069 	for (i = 0; i < adapter->slices; i++) {
2070 		afu = adapter->afu[i];
2071 
2072 		if (afu == NULL || afu->phb == NULL)
2073 			continue;
2074 
2075 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2076 			afu_drv = to_pci_driver(afu_dev->dev.driver);
2077 			if (!afu_drv)
2078 				continue;
2079 
2080 			err_handler = afu_drv->err_handler;
2081 			if (err_handler && err_handler->resume)
2082 				err_handler->resume(afu_dev);
2083 		}
2084 	}
2085 	spin_unlock(&adapter->afu_list_lock);
2086 }
2087 
2088 static const struct pci_error_handlers cxl_err_handler = {
2089 	.error_detected = cxl_pci_error_detected,
2090 	.slot_reset = cxl_pci_slot_reset,
2091 	.resume = cxl_pci_resume,
2092 };
2093 
2094 struct pci_driver cxl_pci_driver = {
2095 	.name = "cxl-pci",
2096 	.id_table = cxl_pci_tbl,
2097 	.probe = cxl_probe,
2098 	.remove = cxl_remove,
2099 	.shutdown = cxl_remove,
2100 	.err_handler = &cxl_err_handler,
2101 };
2102