xref: /linux/drivers/misc/cxl/pci.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
1 /*
2  * Copyright 2014 IBM Corp.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public License
6  * as published by the Free Software Foundation; either version
7  * 2 of the License, or (at your option) any later version.
8  */
9 
10 #include <linux/pci_regs.h>
11 #include <linux/pci_ids.h>
12 #include <linux/device.h>
13 #include <linux/module.h>
14 #include <linux/kernel.h>
15 #include <linux/slab.h>
16 #include <linux/sort.h>
17 #include <linux/pci.h>
18 #include <linux/of.h>
19 #include <linux/delay.h>
20 #include <asm/opal.h>
21 #include <asm/msi_bitmap.h>
22 #include <asm/pnv-pci.h>
23 #include <asm/io.h>
24 #include <asm/reg.h>
25 
26 #include "cxl.h"
27 #include <misc/cxl.h>
28 
29 
30 #define CXL_PCI_VSEC_ID	0x1280
31 #define CXL_VSEC_MIN_SIZE 0x80
32 
33 #define CXL_READ_VSEC_LENGTH(dev, vsec, dest)			\
34 	{							\
35 		pci_read_config_word(dev, vsec + 0x6, dest);	\
36 		*dest >>= 4;					\
37 	}
38 #define CXL_READ_VSEC_NAFUS(dev, vsec, dest) \
39 	pci_read_config_byte(dev, vsec + 0x8, dest)
40 
41 #define CXL_READ_VSEC_STATUS(dev, vsec, dest) \
42 	pci_read_config_byte(dev, vsec + 0x9, dest)
43 #define CXL_STATUS_SECOND_PORT  0x80
44 #define CXL_STATUS_MSI_X_FULL   0x40
45 #define CXL_STATUS_MSI_X_SINGLE 0x20
46 #define CXL_STATUS_FLASH_RW     0x08
47 #define CXL_STATUS_FLASH_RO     0x04
48 #define CXL_STATUS_LOADABLE_AFU 0x02
49 #define CXL_STATUS_LOADABLE_PSL 0x01
50 /* If we see these features we won't try to use the card */
51 #define CXL_UNSUPPORTED_FEATURES \
52 	(CXL_STATUS_MSI_X_FULL | CXL_STATUS_MSI_X_SINGLE)
53 
54 #define CXL_READ_VSEC_MODE_CONTROL(dev, vsec, dest) \
55 	pci_read_config_byte(dev, vsec + 0xa, dest)
56 #define CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val) \
57 	pci_write_config_byte(dev, vsec + 0xa, val)
58 #define CXL_WRITE_VSEC_MODE_CONTROL_BUS(bus, devfn, vsec, val) \
59 	pci_bus_write_config_byte(bus, devfn, vsec + 0xa, val)
60 #define CXL_VSEC_PROTOCOL_MASK   0xe0
61 #define CXL_VSEC_PROTOCOL_1024TB 0x80
62 #define CXL_VSEC_PROTOCOL_512TB  0x40
63 #define CXL_VSEC_PROTOCOL_256TB  0x20 /* Power 8/9 uses this */
64 #define CXL_VSEC_PROTOCOL_ENABLE 0x01
65 
66 #define CXL_READ_VSEC_PSL_REVISION(dev, vsec, dest) \
67 	pci_read_config_word(dev, vsec + 0xc, dest)
68 #define CXL_READ_VSEC_CAIA_MINOR(dev, vsec, dest) \
69 	pci_read_config_byte(dev, vsec + 0xe, dest)
70 #define CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, dest) \
71 	pci_read_config_byte(dev, vsec + 0xf, dest)
72 #define CXL_READ_VSEC_BASE_IMAGE(dev, vsec, dest) \
73 	pci_read_config_word(dev, vsec + 0x10, dest)
74 
75 #define CXL_READ_VSEC_IMAGE_STATE(dev, vsec, dest) \
76 	pci_read_config_byte(dev, vsec + 0x13, dest)
77 #define CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, val) \
78 	pci_write_config_byte(dev, vsec + 0x13, val)
79 #define CXL_VSEC_USER_IMAGE_LOADED 0x80 /* RO */
80 #define CXL_VSEC_PERST_LOADS_IMAGE 0x20 /* RW */
81 #define CXL_VSEC_PERST_SELECT_USER 0x10 /* RW */
82 
83 #define CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, dest) \
84 	pci_read_config_dword(dev, vsec + 0x20, dest)
85 #define CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, dest) \
86 	pci_read_config_dword(dev, vsec + 0x24, dest)
87 #define CXL_READ_VSEC_PS_OFF(dev, vsec, dest) \
88 	pci_read_config_dword(dev, vsec + 0x28, dest)
89 #define CXL_READ_VSEC_PS_SIZE(dev, vsec, dest) \
90 	pci_read_config_dword(dev, vsec + 0x2c, dest)
91 
92 
93 /* This works a little different than the p1/p2 register accesses to make it
94  * easier to pull out individual fields */
95 #define AFUD_READ(afu, off)		in_be64(afu->native->afu_desc_mmio + off)
96 #define AFUD_READ_LE(afu, off)		in_le64(afu->native->afu_desc_mmio + off)
97 #define EXTRACT_PPC_BIT(val, bit)	(!!(val & PPC_BIT(bit)))
98 #define EXTRACT_PPC_BITS(val, bs, be)	((val & PPC_BITMASK(bs, be)) >> PPC_BITLSHIFT(be))
99 
100 #define AFUD_READ_INFO(afu)		AFUD_READ(afu, 0x0)
101 #define   AFUD_NUM_INTS_PER_PROC(val)	EXTRACT_PPC_BITS(val,  0, 15)
102 #define   AFUD_NUM_PROCS(val)		EXTRACT_PPC_BITS(val, 16, 31)
103 #define   AFUD_NUM_CRS(val)		EXTRACT_PPC_BITS(val, 32, 47)
104 #define   AFUD_MULTIMODE(val)		EXTRACT_PPC_BIT(val, 48)
105 #define   AFUD_PUSH_BLOCK_TRANSFER(val)	EXTRACT_PPC_BIT(val, 55)
106 #define   AFUD_DEDICATED_PROCESS(val)	EXTRACT_PPC_BIT(val, 59)
107 #define   AFUD_AFU_DIRECTED(val)	EXTRACT_PPC_BIT(val, 61)
108 #define   AFUD_TIME_SLICED(val)		EXTRACT_PPC_BIT(val, 63)
109 #define AFUD_READ_CR(afu)		AFUD_READ(afu, 0x20)
110 #define   AFUD_CR_LEN(val)		EXTRACT_PPC_BITS(val, 8, 63)
111 #define AFUD_READ_CR_OFF(afu)		AFUD_READ(afu, 0x28)
112 #define AFUD_READ_PPPSA(afu)		AFUD_READ(afu, 0x30)
113 #define   AFUD_PPPSA_PP(val)		EXTRACT_PPC_BIT(val, 6)
114 #define   AFUD_PPPSA_PSA(val)		EXTRACT_PPC_BIT(val, 7)
115 #define   AFUD_PPPSA_LEN(val)		EXTRACT_PPC_BITS(val, 8, 63)
116 #define AFUD_READ_PPPSA_OFF(afu)	AFUD_READ(afu, 0x38)
117 #define AFUD_READ_EB(afu)		AFUD_READ(afu, 0x40)
118 #define   AFUD_EB_LEN(val)		EXTRACT_PPC_BITS(val, 8, 63)
119 #define AFUD_READ_EB_OFF(afu)		AFUD_READ(afu, 0x48)
120 
121 static const struct pci_device_id cxl_pci_tbl[] = {
122 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0477), },
123 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x044b), },
124 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x04cf), },
125 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0601), },
126 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0623), },
127 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0628), },
128 	{ PCI_DEVICE_CLASS(0x120000, ~0), },
129 
130 	{ }
131 };
132 MODULE_DEVICE_TABLE(pci, cxl_pci_tbl);
133 
134 
135 /*
136  * Mostly using these wrappers to avoid confusion:
137  * priv 1 is BAR2, while priv 2 is BAR0
138  */
139 static inline resource_size_t p1_base(struct pci_dev *dev)
140 {
141 	return pci_resource_start(dev, 2);
142 }
143 
144 static inline resource_size_t p1_size(struct pci_dev *dev)
145 {
146 	return pci_resource_len(dev, 2);
147 }
148 
149 static inline resource_size_t p2_base(struct pci_dev *dev)
150 {
151 	return pci_resource_start(dev, 0);
152 }
153 
154 static inline resource_size_t p2_size(struct pci_dev *dev)
155 {
156 	return pci_resource_len(dev, 0);
157 }
158 
159 static int find_cxl_vsec(struct pci_dev *dev)
160 {
161 	int vsec = 0;
162 	u16 val;
163 
164 	while ((vsec = pci_find_next_ext_capability(dev, vsec, PCI_EXT_CAP_ID_VNDR))) {
165 		pci_read_config_word(dev, vsec + 0x4, &val);
166 		if (val == CXL_PCI_VSEC_ID)
167 			return vsec;
168 	}
169 	return 0;
170 
171 }
172 
173 static void dump_cxl_config_space(struct pci_dev *dev)
174 {
175 	int vsec;
176 	u32 val;
177 
178 	dev_info(&dev->dev, "dump_cxl_config_space\n");
179 
180 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_0, &val);
181 	dev_info(&dev->dev, "BAR0: %#.8x\n", val);
182 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_1, &val);
183 	dev_info(&dev->dev, "BAR1: %#.8x\n", val);
184 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_2, &val);
185 	dev_info(&dev->dev, "BAR2: %#.8x\n", val);
186 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_3, &val);
187 	dev_info(&dev->dev, "BAR3: %#.8x\n", val);
188 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_4, &val);
189 	dev_info(&dev->dev, "BAR4: %#.8x\n", val);
190 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_5, &val);
191 	dev_info(&dev->dev, "BAR5: %#.8x\n", val);
192 
193 	dev_info(&dev->dev, "p1 regs: %#llx, len: %#llx\n",
194 		p1_base(dev), p1_size(dev));
195 	dev_info(&dev->dev, "p2 regs: %#llx, len: %#llx\n",
196 		p2_base(dev), p2_size(dev));
197 	dev_info(&dev->dev, "BAR 4/5: %#llx, len: %#llx\n",
198 		pci_resource_start(dev, 4), pci_resource_len(dev, 4));
199 
200 	if (!(vsec = find_cxl_vsec(dev)))
201 		return;
202 
203 #define show_reg(name, what) \
204 	dev_info(&dev->dev, "cxl vsec: %30s: %#x\n", name, what)
205 
206 	pci_read_config_dword(dev, vsec + 0x0, &val);
207 	show_reg("Cap ID", (val >> 0) & 0xffff);
208 	show_reg("Cap Ver", (val >> 16) & 0xf);
209 	show_reg("Next Cap Ptr", (val >> 20) & 0xfff);
210 	pci_read_config_dword(dev, vsec + 0x4, &val);
211 	show_reg("VSEC ID", (val >> 0) & 0xffff);
212 	show_reg("VSEC Rev", (val >> 16) & 0xf);
213 	show_reg("VSEC Length",	(val >> 20) & 0xfff);
214 	pci_read_config_dword(dev, vsec + 0x8, &val);
215 	show_reg("Num AFUs", (val >> 0) & 0xff);
216 	show_reg("Status", (val >> 8) & 0xff);
217 	show_reg("Mode Control", (val >> 16) & 0xff);
218 	show_reg("Reserved", (val >> 24) & 0xff);
219 	pci_read_config_dword(dev, vsec + 0xc, &val);
220 	show_reg("PSL Rev", (val >> 0) & 0xffff);
221 	show_reg("CAIA Ver", (val >> 16) & 0xffff);
222 	pci_read_config_dword(dev, vsec + 0x10, &val);
223 	show_reg("Base Image Rev", (val >> 0) & 0xffff);
224 	show_reg("Reserved", (val >> 16) & 0x0fff);
225 	show_reg("Image Control", (val >> 28) & 0x3);
226 	show_reg("Reserved", (val >> 30) & 0x1);
227 	show_reg("Image Loaded", (val >> 31) & 0x1);
228 
229 	pci_read_config_dword(dev, vsec + 0x14, &val);
230 	show_reg("Reserved", val);
231 	pci_read_config_dword(dev, vsec + 0x18, &val);
232 	show_reg("Reserved", val);
233 	pci_read_config_dword(dev, vsec + 0x1c, &val);
234 	show_reg("Reserved", val);
235 
236 	pci_read_config_dword(dev, vsec + 0x20, &val);
237 	show_reg("AFU Descriptor Offset", val);
238 	pci_read_config_dword(dev, vsec + 0x24, &val);
239 	show_reg("AFU Descriptor Size", val);
240 	pci_read_config_dword(dev, vsec + 0x28, &val);
241 	show_reg("Problem State Offset", val);
242 	pci_read_config_dword(dev, vsec + 0x2c, &val);
243 	show_reg("Problem State Size", val);
244 
245 	pci_read_config_dword(dev, vsec + 0x30, &val);
246 	show_reg("Reserved", val);
247 	pci_read_config_dword(dev, vsec + 0x34, &val);
248 	show_reg("Reserved", val);
249 	pci_read_config_dword(dev, vsec + 0x38, &val);
250 	show_reg("Reserved", val);
251 	pci_read_config_dword(dev, vsec + 0x3c, &val);
252 	show_reg("Reserved", val);
253 
254 	pci_read_config_dword(dev, vsec + 0x40, &val);
255 	show_reg("PSL Programming Port", val);
256 	pci_read_config_dword(dev, vsec + 0x44, &val);
257 	show_reg("PSL Programming Control", val);
258 
259 	pci_read_config_dword(dev, vsec + 0x48, &val);
260 	show_reg("Reserved", val);
261 	pci_read_config_dword(dev, vsec + 0x4c, &val);
262 	show_reg("Reserved", val);
263 
264 	pci_read_config_dword(dev, vsec + 0x50, &val);
265 	show_reg("Flash Address Register", val);
266 	pci_read_config_dword(dev, vsec + 0x54, &val);
267 	show_reg("Flash Size Register", val);
268 	pci_read_config_dword(dev, vsec + 0x58, &val);
269 	show_reg("Flash Status/Control Register", val);
270 	pci_read_config_dword(dev, vsec + 0x58, &val);
271 	show_reg("Flash Data Port", val);
272 
273 #undef show_reg
274 }
275 
276 static void dump_afu_descriptor(struct cxl_afu *afu)
277 {
278 	u64 val, afu_cr_num, afu_cr_off, afu_cr_len;
279 	int i;
280 
281 #define show_reg(name, what) \
282 	dev_info(&afu->dev, "afu desc: %30s: %#llx\n", name, what)
283 
284 	val = AFUD_READ_INFO(afu);
285 	show_reg("num_ints_per_process", AFUD_NUM_INTS_PER_PROC(val));
286 	show_reg("num_of_processes", AFUD_NUM_PROCS(val));
287 	show_reg("num_of_afu_CRs", AFUD_NUM_CRS(val));
288 	show_reg("req_prog_mode", val & 0xffffULL);
289 	afu_cr_num = AFUD_NUM_CRS(val);
290 
291 	val = AFUD_READ(afu, 0x8);
292 	show_reg("Reserved", val);
293 	val = AFUD_READ(afu, 0x10);
294 	show_reg("Reserved", val);
295 	val = AFUD_READ(afu, 0x18);
296 	show_reg("Reserved", val);
297 
298 	val = AFUD_READ_CR(afu);
299 	show_reg("Reserved", (val >> (63-7)) & 0xff);
300 	show_reg("AFU_CR_len", AFUD_CR_LEN(val));
301 	afu_cr_len = AFUD_CR_LEN(val) * 256;
302 
303 	val = AFUD_READ_CR_OFF(afu);
304 	afu_cr_off = val;
305 	show_reg("AFU_CR_offset", val);
306 
307 	val = AFUD_READ_PPPSA(afu);
308 	show_reg("PerProcessPSA_control", (val >> (63-7)) & 0xff);
309 	show_reg("PerProcessPSA Length", AFUD_PPPSA_LEN(val));
310 
311 	val = AFUD_READ_PPPSA_OFF(afu);
312 	show_reg("PerProcessPSA_offset", val);
313 
314 	val = AFUD_READ_EB(afu);
315 	show_reg("Reserved", (val >> (63-7)) & 0xff);
316 	show_reg("AFU_EB_len", AFUD_EB_LEN(val));
317 
318 	val = AFUD_READ_EB_OFF(afu);
319 	show_reg("AFU_EB_offset", val);
320 
321 	for (i = 0; i < afu_cr_num; i++) {
322 		val = AFUD_READ_LE(afu, afu_cr_off + i * afu_cr_len);
323 		show_reg("CR Vendor", val & 0xffff);
324 		show_reg("CR Device", (val >> 16) & 0xffff);
325 	}
326 #undef show_reg
327 }
328 
329 #define P8_CAPP_UNIT0_ID 0xBA
330 #define P8_CAPP_UNIT1_ID 0XBE
331 #define P9_CAPP_UNIT0_ID 0xC0
332 #define P9_CAPP_UNIT1_ID 0xE0
333 
334 static int get_phb_index(struct device_node *np, u32 *phb_index)
335 {
336 	if (of_property_read_u32(np, "ibm,phb-index", phb_index))
337 		return -ENODEV;
338 	return 0;
339 }
340 
341 static u64 get_capp_unit_id(struct device_node *np, u32 phb_index)
342 {
343 	/*
344 	 * POWER 8:
345 	 *  - For chips other than POWER8NVL, we only have CAPP 0,
346 	 *    irrespective of which PHB is used.
347 	 *  - For POWER8NVL, assume CAPP 0 is attached to PHB0 and
348 	 *    CAPP 1 is attached to PHB1.
349 	 */
350 	if (cxl_is_power8()) {
351 		if (!pvr_version_is(PVR_POWER8NVL))
352 			return P8_CAPP_UNIT0_ID;
353 
354 		if (phb_index == 0)
355 			return P8_CAPP_UNIT0_ID;
356 
357 		if (phb_index == 1)
358 			return P8_CAPP_UNIT1_ID;
359 	}
360 
361 	/*
362 	 * POWER 9:
363 	 *   PEC0 (PHB0). Capp ID = CAPP0 (0b1100_0000)
364 	 *   PEC1 (PHB1 - PHB2). No capi mode
365 	 *   PEC2 (PHB3 - PHB4 - PHB5): Capi mode on PHB3 only. Capp ID = CAPP1 (0b1110_0000)
366 	 */
367 	if (cxl_is_power9()) {
368 		if (phb_index == 0)
369 			return P9_CAPP_UNIT0_ID;
370 
371 		if (phb_index == 3)
372 			return P9_CAPP_UNIT1_ID;
373 	}
374 
375 	return 0;
376 }
377 
378 int cxl_calc_capp_routing(struct pci_dev *dev, u64 *chipid,
379 			     u32 *phb_index, u64 *capp_unit_id)
380 {
381 	int rc;
382 	struct device_node *np;
383 	const __be32 *prop;
384 
385 	if (!(np = pnv_pci_get_phb_node(dev)))
386 		return -ENODEV;
387 
388 	while (np && !(prop = of_get_property(np, "ibm,chip-id", NULL)))
389 		np = of_get_next_parent(np);
390 	if (!np)
391 		return -ENODEV;
392 
393 	*chipid = be32_to_cpup(prop);
394 
395 	rc = get_phb_index(np, phb_index);
396 	if (rc) {
397 		pr_err("cxl: invalid phb index\n");
398 		return rc;
399 	}
400 
401 	*capp_unit_id = get_capp_unit_id(np, *phb_index);
402 	of_node_put(np);
403 	if (!*capp_unit_id) {
404 		pr_err("cxl: invalid capp unit id (phb_index: %d)\n",
405 		       *phb_index);
406 		return -ENODEV;
407 	}
408 
409 	return 0;
410 }
411 
412 int cxl_get_xsl9_dsnctl(u64 capp_unit_id, u64 *reg)
413 {
414 	u64 xsl_dsnctl;
415 
416 	/*
417 	 * CAPI Identifier bits [0:7]
418 	 * bit 61:60 MSI bits --> 0
419 	 * bit 59 TVT selector --> 0
420 	 */
421 
422 	/*
423 	 * Tell XSL where to route data to.
424 	 * The field chipid should match the PHB CAPI_CMPM register
425 	 */
426 	xsl_dsnctl = ((u64)0x2 << (63-7)); /* Bit 57 */
427 	xsl_dsnctl |= (capp_unit_id << (63-15));
428 
429 	/* nMMU_ID Defaults to: b’000001001’*/
430 	xsl_dsnctl |= ((u64)0x09 << (63-28));
431 
432 	if (!(cxl_is_power9_dd1())) {
433 		/*
434 		 * Used to identify CAPI packets which should be sorted into
435 		 * the Non-Blocking queues by the PHB. This field should match
436 		 * the PHB PBL_NBW_CMPM register
437 		 * nbwind=0x03, bits [57:58], must include capi indicator.
438 		 * Not supported on P9 DD1.
439 		 */
440 		xsl_dsnctl |= ((u64)0x03 << (63-47));
441 
442 		/*
443 		 * Upper 16b address bits of ASB_Notify messages sent to the
444 		 * system. Need to match the PHB’s ASN Compare/Mask Register.
445 		 * Not supported on P9 DD1.
446 		 */
447 		xsl_dsnctl |= ((u64)0x04 << (63-55));
448 	}
449 
450 	*reg = xsl_dsnctl;
451 	return 0;
452 }
453 
454 static int init_implementation_adapter_regs_psl9(struct cxl *adapter,
455 						 struct pci_dev *dev)
456 {
457 	u64 xsl_dsnctl, psl_fircntl;
458 	u64 chipid;
459 	u32 phb_index;
460 	u64 capp_unit_id;
461 	int rc;
462 
463 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
464 	if (rc)
465 		return rc;
466 
467 	rc = cxl_get_xsl9_dsnctl(capp_unit_id, &xsl_dsnctl);
468 	if (rc)
469 		return rc;
470 
471 	cxl_p1_write(adapter, CXL_XSL9_DSNCTL, xsl_dsnctl);
472 
473 	/* Set fir_cntl to recommended value for production env */
474 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
475 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
476 	psl_fircntl |= 0x1ULL; /* ce_thresh */
477 	cxl_p1_write(adapter, CXL_PSL9_FIR_CNTL, psl_fircntl);
478 
479 	/* Setup the PSL to transmit packets on the PCIe before the
480 	 * CAPP is enabled
481 	 */
482 	cxl_p1_write(adapter, CXL_PSL9_DSNDCTL, 0x0001001000002A10ULL);
483 
484 	/*
485 	 * A response to an ASB_Notify request is returned by the
486 	 * system as an MMIO write to the address defined in
487 	 * the PSL_TNR_ADDR register.
488 	 * keep the Reset Value: 0x00020000E0000000
489 	 */
490 
491 	/* Enable XSL rty limit */
492 	cxl_p1_write(adapter, CXL_XSL9_DEF, 0x51F8000000000005ULL);
493 
494 	/* Change XSL_INV dummy read threshold */
495 	cxl_p1_write(adapter, CXL_XSL9_INV, 0x0000040007FFC200ULL);
496 
497 	if (phb_index == 3) {
498 		/* disable machines 31-47 and 20-27 for DMA */
499 		cxl_p1_write(adapter, CXL_PSL9_APCDEDTYPE, 0x40000FF3FFFF0000ULL);
500 	}
501 
502 	/* Snoop machines */
503 	cxl_p1_write(adapter, CXL_PSL9_APCDEDALLOC, 0x800F000200000000ULL);
504 
505 	if (cxl_is_power9_dd1()) {
506 		/* Disabling deadlock counter CAR */
507 		cxl_p1_write(adapter, CXL_PSL9_GP_CT, 0x0020000000000001ULL);
508 	} else
509 		cxl_p1_write(adapter, CXL_PSL9_DEBUG, 0x4000000000000000ULL);
510 
511 	return 0;
512 }
513 
514 static int init_implementation_adapter_regs_psl8(struct cxl *adapter, struct pci_dev *dev)
515 {
516 	u64 psl_dsnctl, psl_fircntl;
517 	u64 chipid;
518 	u32 phb_index;
519 	u64 capp_unit_id;
520 	int rc;
521 
522 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
523 	if (rc)
524 		return rc;
525 
526 	psl_dsnctl = 0x0000900000000000ULL; /* pteupd ttype, scdone */
527 	psl_dsnctl |= (0x2ULL << (63-38)); /* MMIO hang pulse: 256 us */
528 	/* Tell PSL where to route data to */
529 	psl_dsnctl |= (chipid << (63-5));
530 	psl_dsnctl |= (capp_unit_id << (63-13));
531 
532 	cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
533 	cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
534 	/* snoop write mask */
535 	cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
536 	/* set fir_cntl to recommended value for production env */
537 	psl_fircntl = (0x2ULL << (63-3)); /* ce_report */
538 	psl_fircntl |= (0x1ULL << (63-6)); /* FIR_report */
539 	psl_fircntl |= 0x1ULL; /* ce_thresh */
540 	cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, psl_fircntl);
541 	/* for debugging with trace arrays */
542 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
543 
544 	return 0;
545 }
546 
547 static int init_implementation_adapter_regs_xsl(struct cxl *adapter, struct pci_dev *dev)
548 {
549 	u64 xsl_dsnctl;
550 	u64 chipid;
551 	u32 phb_index;
552 	u64 capp_unit_id;
553 	int rc;
554 
555 	rc = cxl_calc_capp_routing(dev, &chipid, &phb_index, &capp_unit_id);
556 	if (rc)
557 		return rc;
558 
559 	/* Tell XSL where to route data to */
560 	xsl_dsnctl = 0x0000600000000000ULL | (chipid << (63-5));
561 	xsl_dsnctl |= (capp_unit_id << (63-13));
562 	cxl_p1_write(adapter, CXL_XSL_DSNCTL, xsl_dsnctl);
563 
564 	return 0;
565 }
566 
567 /* PSL & XSL */
568 #define TBSYNC_CAL(n) (((u64)n & 0x7) << (63-3))
569 #define TBSYNC_CNT(n) (((u64)n & 0x7) << (63-6))
570 /* For the PSL this is a multiple for 0 < n <= 7: */
571 #define PSL_2048_250MHZ_CYCLES 1
572 
573 static void write_timebase_ctrl_psl9(struct cxl *adapter)
574 {
575 	cxl_p1_write(adapter, CXL_PSL9_TB_CTLSTAT,
576 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
577 }
578 
579 static void write_timebase_ctrl_psl8(struct cxl *adapter)
580 {
581 	cxl_p1_write(adapter, CXL_PSL_TB_CTLSTAT,
582 		     TBSYNC_CNT(2 * PSL_2048_250MHZ_CYCLES));
583 }
584 
585 /* XSL */
586 #define TBSYNC_ENA (1ULL << 63)
587 /* For the XSL this is 2**n * 2000 clocks for 0 < n <= 6: */
588 #define XSL_2000_CLOCKS 1
589 #define XSL_4000_CLOCKS 2
590 #define XSL_8000_CLOCKS 3
591 
592 static void write_timebase_ctrl_xsl(struct cxl *adapter)
593 {
594 	cxl_p1_write(adapter, CXL_XSL_TB_CTLSTAT,
595 		     TBSYNC_ENA |
596 		     TBSYNC_CAL(3) |
597 		     TBSYNC_CNT(XSL_4000_CLOCKS));
598 }
599 
600 static u64 timebase_read_psl9(struct cxl *adapter)
601 {
602 	return cxl_p1_read(adapter, CXL_PSL9_Timebase);
603 }
604 
605 static u64 timebase_read_psl8(struct cxl *adapter)
606 {
607 	return cxl_p1_read(adapter, CXL_PSL_Timebase);
608 }
609 
610 static u64 timebase_read_xsl(struct cxl *adapter)
611 {
612 	return cxl_p1_read(adapter, CXL_XSL_Timebase);
613 }
614 
615 static void cxl_setup_psl_timebase(struct cxl *adapter, struct pci_dev *dev)
616 {
617 	u64 psl_tb;
618 	int delta;
619 	unsigned int retry = 0;
620 	struct device_node *np;
621 
622 	adapter->psl_timebase_synced = false;
623 
624 	if (!(np = pnv_pci_get_phb_node(dev)))
625 		return;
626 
627 	/* Do not fail when CAPP timebase sync is not supported by OPAL */
628 	of_node_get(np);
629 	if (! of_get_property(np, "ibm,capp-timebase-sync", NULL)) {
630 		of_node_put(np);
631 		dev_info(&dev->dev, "PSL timebase inactive: OPAL support missing\n");
632 		return;
633 	}
634 	of_node_put(np);
635 
636 	/*
637 	 * Setup PSL Timebase Control and Status register
638 	 * with the recommended Timebase Sync Count value
639 	 */
640 	adapter->native->sl_ops->write_timebase_ctrl(adapter);
641 
642 	/* Enable PSL Timebase */
643 	cxl_p1_write(adapter, CXL_PSL_Control, 0x0000000000000000);
644 	cxl_p1_write(adapter, CXL_PSL_Control, CXL_PSL_Control_tb);
645 
646 	/* Wait until CORE TB and PSL TB difference <= 16usecs */
647 	do {
648 		msleep(1);
649 		if (retry++ > 5) {
650 			dev_info(&dev->dev, "PSL timebase can't synchronize\n");
651 			return;
652 		}
653 		psl_tb = adapter->native->sl_ops->timebase_read(adapter);
654 		delta = mftb() - psl_tb;
655 		if (delta < 0)
656 			delta = -delta;
657 	} while (tb_to_ns(delta) > 16000);
658 
659 	adapter->psl_timebase_synced = true;
660 	return;
661 }
662 
663 static int init_implementation_afu_regs_psl9(struct cxl_afu *afu)
664 {
665 	return 0;
666 }
667 
668 static int init_implementation_afu_regs_psl8(struct cxl_afu *afu)
669 {
670 	/* read/write masks for this slice */
671 	cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
672 	/* APC read/write masks for this slice */
673 	cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
674 	/* for debugging with trace arrays */
675 	cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
676 	cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
677 
678 	return 0;
679 }
680 
681 int cxl_pci_setup_irq(struct cxl *adapter, unsigned int hwirq,
682 		unsigned int virq)
683 {
684 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
685 
686 	return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
687 }
688 
689 int cxl_update_image_control(struct cxl *adapter)
690 {
691 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
692 	int rc;
693 	int vsec;
694 	u8 image_state;
695 
696 	if (!(vsec = find_cxl_vsec(dev))) {
697 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
698 		return -ENODEV;
699 	}
700 
701 	if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
702 		dev_err(&dev->dev, "failed to read image state: %i\n", rc);
703 		return rc;
704 	}
705 
706 	if (adapter->perst_loads_image)
707 		image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
708 	else
709 		image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
710 
711 	if (adapter->perst_select_user)
712 		image_state |= CXL_VSEC_PERST_SELECT_USER;
713 	else
714 		image_state &= ~CXL_VSEC_PERST_SELECT_USER;
715 
716 	if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
717 		dev_err(&dev->dev, "failed to update image control: %i\n", rc);
718 		return rc;
719 	}
720 
721 	return 0;
722 }
723 
724 int cxl_pci_alloc_one_irq(struct cxl *adapter)
725 {
726 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
727 
728 	return pnv_cxl_alloc_hwirqs(dev, 1);
729 }
730 
731 void cxl_pci_release_one_irq(struct cxl *adapter, int hwirq)
732 {
733 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
734 
735 	return pnv_cxl_release_hwirqs(dev, hwirq, 1);
736 }
737 
738 int cxl_pci_alloc_irq_ranges(struct cxl_irq_ranges *irqs,
739 			struct cxl *adapter, unsigned int num)
740 {
741 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
742 
743 	return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
744 }
745 
746 void cxl_pci_release_irq_ranges(struct cxl_irq_ranges *irqs,
747 				struct cxl *adapter)
748 {
749 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
750 
751 	pnv_cxl_release_hwirq_ranges(irqs, dev);
752 }
753 
754 static int setup_cxl_bars(struct pci_dev *dev)
755 {
756 	/* Safety check in case we get backported to < 3.17 without M64 */
757 	if ((p1_base(dev) < 0x100000000ULL) ||
758 	    (p2_base(dev) < 0x100000000ULL)) {
759 		dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
760 		return -ENODEV;
761 	}
762 
763 	/*
764 	 * BAR 4/5 has a special meaning for CXL and must be programmed with a
765 	 * special value corresponding to the CXL protocol address range.
766 	 * For POWER 8/9 that means bits 48:49 must be set to 10
767 	 */
768 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
769 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
770 
771 	return 0;
772 }
773 
774 #ifdef CONFIG_CXL_BIMODAL
775 
776 struct cxl_switch_work {
777 	struct pci_dev *dev;
778 	struct work_struct work;
779 	int vsec;
780 	int mode;
781 };
782 
783 static void switch_card_to_cxl(struct work_struct *work)
784 {
785 	struct cxl_switch_work *switch_work =
786 		container_of(work, struct cxl_switch_work, work);
787 	struct pci_dev *dev = switch_work->dev;
788 	struct pci_bus *bus = dev->bus;
789 	struct pci_controller *hose = pci_bus_to_host(bus);
790 	struct pci_dev *bridge;
791 	struct pnv_php_slot *php_slot;
792 	unsigned int devfn;
793 	u8 val;
794 	int rc;
795 
796 	dev_info(&bus->dev, "cxl: Preparing for mode switch...\n");
797 	bridge = list_first_entry_or_null(&hose->bus->devices, struct pci_dev,
798 					  bus_list);
799 	if (!bridge) {
800 		dev_WARN(&bus->dev, "cxl: Couldn't find root port!\n");
801 		goto err_dev_put;
802 	}
803 
804 	php_slot = pnv_php_find_slot(pci_device_to_OF_node(bridge));
805 	if (!php_slot) {
806 		dev_err(&bus->dev, "cxl: Failed to find slot hotplug "
807 			           "information. You may need to upgrade "
808 			           "skiboot. Aborting.\n");
809 		goto err_dev_put;
810 	}
811 
812 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, switch_work->vsec, &val);
813 	if (rc) {
814 		dev_err(&bus->dev, "cxl: Failed to read CAPI mode control: %i\n", rc);
815 		goto err_dev_put;
816 	}
817 	devfn = dev->devfn;
818 
819 	/* Release the reference obtained in cxl_check_and_switch_mode() */
820 	pci_dev_put(dev);
821 
822 	dev_dbg(&bus->dev, "cxl: Removing PCI devices from kernel\n");
823 	pci_lock_rescan_remove();
824 	pci_hp_remove_devices(bridge->subordinate);
825 	pci_unlock_rescan_remove();
826 
827 	/* Switch the CXL protocol on the card */
828 	if (switch_work->mode == CXL_BIMODE_CXL) {
829 		dev_info(&bus->dev, "cxl: Switching card to CXL mode\n");
830 		val &= ~CXL_VSEC_PROTOCOL_MASK;
831 		val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
832 		rc = pnv_cxl_enable_phb_kernel_api(hose, true);
833 		if (rc) {
834 			dev_err(&bus->dev, "cxl: Failed to enable kernel API"
835 				           " on real PHB, aborting\n");
836 			goto err_free_work;
837 		}
838 	} else {
839 		dev_WARN(&bus->dev, "cxl: Switching card to PCI mode not supported!\n");
840 		goto err_free_work;
841 	}
842 
843 	rc = CXL_WRITE_VSEC_MODE_CONTROL_BUS(bus, devfn, switch_work->vsec, val);
844 	if (rc) {
845 		dev_err(&bus->dev, "cxl: Failed to configure CXL protocol: %i\n", rc);
846 		goto err_free_work;
847 	}
848 
849 	/*
850 	 * The CAIA spec (v1.1, Section 10.6 Bi-modal Device Support) states
851 	 * we must wait 100ms after this mode switch before touching PCIe config
852 	 * space.
853 	 */
854 	msleep(100);
855 
856 	/*
857 	 * Hot reset to cause the card to come back in cxl mode. A
858 	 * OPAL_RESET_PCI_LINK would be sufficient, but currently lacks support
859 	 * in skiboot, so we use a hot reset instead.
860 	 *
861 	 * We call pci_set_pcie_reset_state() on the bridge, as a CAPI card is
862 	 * guaranteed to sit directly under the root port, and setting the reset
863 	 * state on a device directly under the root port is equivalent to doing
864 	 * it on the root port iself.
865 	 */
866 	dev_info(&bus->dev, "cxl: Configuration write complete, resetting card\n");
867 	pci_set_pcie_reset_state(bridge, pcie_hot_reset);
868 	pci_set_pcie_reset_state(bridge, pcie_deassert_reset);
869 
870 	dev_dbg(&bus->dev, "cxl: Offlining slot\n");
871 	rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_OFFLINE);
872 	if (rc) {
873 		dev_err(&bus->dev, "cxl: OPAL offlining call failed: %i\n", rc);
874 		goto err_free_work;
875 	}
876 
877 	dev_dbg(&bus->dev, "cxl: Onlining and probing slot\n");
878 	rc = pnv_php_set_slot_power_state(&php_slot->slot, OPAL_PCI_SLOT_ONLINE);
879 	if (rc) {
880 		dev_err(&bus->dev, "cxl: OPAL onlining call failed: %i\n", rc);
881 		goto err_free_work;
882 	}
883 
884 	pci_lock_rescan_remove();
885 	pci_hp_add_devices(bridge->subordinate);
886 	pci_unlock_rescan_remove();
887 
888 	dev_info(&bus->dev, "cxl: CAPI mode switch completed\n");
889 	kfree(switch_work);
890 	return;
891 
892 err_dev_put:
893 	/* Release the reference obtained in cxl_check_and_switch_mode() */
894 	pci_dev_put(dev);
895 err_free_work:
896 	kfree(switch_work);
897 }
898 
899 int cxl_check_and_switch_mode(struct pci_dev *dev, int mode, int vsec)
900 {
901 	struct cxl_switch_work *work;
902 	u8 val;
903 	int rc;
904 
905 	if (!cpu_has_feature(CPU_FTR_HVMODE))
906 		return -ENODEV;
907 
908 	if (!vsec) {
909 		vsec = find_cxl_vsec(dev);
910 		if (!vsec) {
911 			dev_info(&dev->dev, "CXL VSEC not found\n");
912 			return -ENODEV;
913 		}
914 	}
915 
916 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
917 	if (rc) {
918 		dev_err(&dev->dev, "Failed to read current mode control: %i", rc);
919 		return rc;
920 	}
921 
922 	if (mode == CXL_BIMODE_PCI) {
923 		if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
924 			dev_info(&dev->dev, "Card is already in PCI mode\n");
925 			return 0;
926 		}
927 		/*
928 		 * TODO: Before it's safe to switch the card back to PCI mode
929 		 * we need to disable the CAPP and make sure any cachelines the
930 		 * card holds have been flushed out. Needs skiboot support.
931 		 */
932 		dev_WARN(&dev->dev, "CXL mode switch to PCI unsupported!\n");
933 		return -EIO;
934 	}
935 
936 	if (val & CXL_VSEC_PROTOCOL_ENABLE) {
937 		dev_info(&dev->dev, "Card is already in CXL mode\n");
938 		return 0;
939 	}
940 
941 	dev_info(&dev->dev, "Card is in PCI mode, scheduling kernel thread "
942 			    "to switch to CXL mode\n");
943 
944 	work = kmalloc(sizeof(struct cxl_switch_work), GFP_KERNEL);
945 	if (!work)
946 		return -ENOMEM;
947 
948 	pci_dev_get(dev);
949 	work->dev = dev;
950 	work->vsec = vsec;
951 	work->mode = mode;
952 	INIT_WORK(&work->work, switch_card_to_cxl);
953 
954 	schedule_work(&work->work);
955 
956 	/*
957 	 * We return a failure now to abort the driver init. Once the
958 	 * link has been cycled and the card is in cxl mode we will
959 	 * come back (possibly using the generic cxl driver), but
960 	 * return success as the card should then be in cxl mode.
961 	 *
962 	 * TODO: What if the card comes back in PCI mode even after
963 	 *       the switch?  Don't want to spin endlessly.
964 	 */
965 	return -EBUSY;
966 }
967 EXPORT_SYMBOL_GPL(cxl_check_and_switch_mode);
968 
969 #endif /* CONFIG_CXL_BIMODAL */
970 
971 static int setup_cxl_protocol_area(struct pci_dev *dev)
972 {
973 	u8 val;
974 	int rc;
975 	int vsec = find_cxl_vsec(dev);
976 
977 	if (!vsec) {
978 		dev_info(&dev->dev, "CXL VSEC not found\n");
979 		return -ENODEV;
980 	}
981 
982 	rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val);
983 	if (rc) {
984 		dev_err(&dev->dev, "Failed to read current mode control: %i\n", rc);
985 		return rc;
986 	}
987 
988 	if (!(val & CXL_VSEC_PROTOCOL_ENABLE)) {
989 		dev_err(&dev->dev, "Card not in CAPI mode!\n");
990 		return -EIO;
991 	}
992 
993 	if ((val & CXL_VSEC_PROTOCOL_MASK) != CXL_VSEC_PROTOCOL_256TB) {
994 		val &= ~CXL_VSEC_PROTOCOL_MASK;
995 		val |= CXL_VSEC_PROTOCOL_256TB;
996 		rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val);
997 		if (rc) {
998 			dev_err(&dev->dev, "Failed to set CXL protocol area: %i\n", rc);
999 			return rc;
1000 		}
1001 	}
1002 
1003 	return 0;
1004 }
1005 
1006 static int pci_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1007 {
1008 	u64 p1n_base, p2n_base, afu_desc;
1009 	const u64 p1n_size = 0x100;
1010 	const u64 p2n_size = 0x1000;
1011 
1012 	p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
1013 	p2n_base = p2_base(dev) + (afu->slice * p2n_size);
1014 	afu->psn_phys = p2_base(dev) + (adapter->native->ps_off + (afu->slice * adapter->ps_size));
1015 	afu_desc = p2_base(dev) + adapter->native->afu_desc_off + (afu->slice * adapter->native->afu_desc_size);
1016 
1017 	if (!(afu->native->p1n_mmio = ioremap(p1n_base, p1n_size)))
1018 		goto err;
1019 	if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
1020 		goto err1;
1021 	if (afu_desc) {
1022 		if (!(afu->native->afu_desc_mmio = ioremap(afu_desc, adapter->native->afu_desc_size)))
1023 			goto err2;
1024 	}
1025 
1026 	return 0;
1027 err2:
1028 	iounmap(afu->p2n_mmio);
1029 err1:
1030 	iounmap(afu->native->p1n_mmio);
1031 err:
1032 	dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
1033 	return -ENOMEM;
1034 }
1035 
1036 static void pci_unmap_slice_regs(struct cxl_afu *afu)
1037 {
1038 	if (afu->p2n_mmio) {
1039 		iounmap(afu->p2n_mmio);
1040 		afu->p2n_mmio = NULL;
1041 	}
1042 	if (afu->native->p1n_mmio) {
1043 		iounmap(afu->native->p1n_mmio);
1044 		afu->native->p1n_mmio = NULL;
1045 	}
1046 	if (afu->native->afu_desc_mmio) {
1047 		iounmap(afu->native->afu_desc_mmio);
1048 		afu->native->afu_desc_mmio = NULL;
1049 	}
1050 }
1051 
1052 void cxl_pci_release_afu(struct device *dev)
1053 {
1054 	struct cxl_afu *afu = to_cxl_afu(dev);
1055 
1056 	pr_devel("%s\n", __func__);
1057 
1058 	idr_destroy(&afu->contexts_idr);
1059 	cxl_release_spa(afu);
1060 
1061 	kfree(afu->native);
1062 	kfree(afu);
1063 }
1064 
1065 /* Expects AFU struct to have recently been zeroed out */
1066 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
1067 {
1068 	u64 val;
1069 
1070 	val = AFUD_READ_INFO(afu);
1071 	afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
1072 	afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
1073 	afu->crs_num = AFUD_NUM_CRS(val);
1074 
1075 	if (AFUD_AFU_DIRECTED(val))
1076 		afu->modes_supported |= CXL_MODE_DIRECTED;
1077 	if (AFUD_DEDICATED_PROCESS(val))
1078 		afu->modes_supported |= CXL_MODE_DEDICATED;
1079 	if (AFUD_TIME_SLICED(val))
1080 		afu->modes_supported |= CXL_MODE_TIME_SLICED;
1081 
1082 	val = AFUD_READ_PPPSA(afu);
1083 	afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
1084 	afu->psa = AFUD_PPPSA_PSA(val);
1085 	if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
1086 		afu->native->pp_offset = AFUD_READ_PPPSA_OFF(afu);
1087 
1088 	val = AFUD_READ_CR(afu);
1089 	afu->crs_len = AFUD_CR_LEN(val) * 256;
1090 	afu->crs_offset = AFUD_READ_CR_OFF(afu);
1091 
1092 
1093 	/* eb_len is in multiple of 4K */
1094 	afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
1095 	afu->eb_offset = AFUD_READ_EB_OFF(afu);
1096 
1097 	/* eb_off is 4K aligned so lower 12 bits are always zero */
1098 	if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
1099 		dev_warn(&afu->dev,
1100 			 "Invalid AFU error buffer offset %Lx\n",
1101 			 afu->eb_offset);
1102 		dev_info(&afu->dev,
1103 			 "Ignoring AFU error buffer in the descriptor\n");
1104 		/* indicate that no afu buffer exists */
1105 		afu->eb_len = 0;
1106 	}
1107 
1108 	return 0;
1109 }
1110 
1111 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
1112 {
1113 	int i, rc;
1114 	u32 val;
1115 
1116 	if (afu->psa && afu->adapter->ps_size <
1117 			(afu->native->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
1118 		dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
1119 		return -ENODEV;
1120 	}
1121 
1122 	if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
1123 		dev_warn(&afu->dev, "AFU uses pp_size(%#016llx) < PAGE_SIZE per-process PSA!\n", afu->pp_size);
1124 
1125 	for (i = 0; i < afu->crs_num; i++) {
1126 		rc = cxl_ops->afu_cr_read32(afu, i, 0, &val);
1127 		if (rc || val == 0) {
1128 			dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
1129 			return -EINVAL;
1130 		}
1131 	}
1132 
1133 	if ((afu->modes_supported & ~CXL_MODE_DEDICATED) && afu->max_procs_virtualised == 0) {
1134 		/*
1135 		 * We could also check this for the dedicated process model
1136 		 * since the architecture indicates it should be set to 1, but
1137 		 * in that case we ignore the value and I'd rather not risk
1138 		 * breaking any existing dedicated process AFUs that left it as
1139 		 * 0 (not that I'm aware of any). It is clearly an error for an
1140 		 * AFU directed AFU to set this to 0, and would have previously
1141 		 * triggered a bug resulting in the maximum not being enforced
1142 		 * at all since idr_alloc treats 0 as no maximum.
1143 		 */
1144 		dev_err(&afu->dev, "AFU does not support any processes\n");
1145 		return -EINVAL;
1146 	}
1147 
1148 	return 0;
1149 }
1150 
1151 static int sanitise_afu_regs_psl9(struct cxl_afu *afu)
1152 {
1153 	u64 reg;
1154 
1155 	/*
1156 	 * Clear out any regs that contain either an IVTE or address or may be
1157 	 * waiting on an acknowledgment to try to be a bit safer as we bring
1158 	 * it online
1159 	 */
1160 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
1161 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
1162 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
1163 		if (cxl_ops->afu_reset(afu))
1164 			return -EIO;
1165 		if (cxl_afu_disable(afu))
1166 			return -EIO;
1167 		if (cxl_psl_purge(afu))
1168 			return -EIO;
1169 	}
1170 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
1171 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1172 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1173 	if (reg) {
1174 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1175 		if (reg & CXL_PSL9_DSISR_An_TF)
1176 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1177 		else
1178 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1179 	}
1180 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1181 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1182 		if (reg) {
1183 			if (reg & ~0x000000007fffffff)
1184 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1185 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1186 		}
1187 	}
1188 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1189 	if (reg) {
1190 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1191 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1192 	}
1193 
1194 	return 0;
1195 }
1196 
1197 static int sanitise_afu_regs_psl8(struct cxl_afu *afu)
1198 {
1199 	u64 reg;
1200 
1201 	/*
1202 	 * Clear out any regs that contain either an IVTE or address or may be
1203 	 * waiting on an acknowledgement to try to be a bit safer as we bring
1204 	 * it online
1205 	 */
1206 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
1207 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
1208 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
1209 		if (cxl_ops->afu_reset(afu))
1210 			return -EIO;
1211 		if (cxl_afu_disable(afu))
1212 			return -EIO;
1213 		if (cxl_psl_purge(afu))
1214 			return -EIO;
1215 	}
1216 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
1217 	cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
1218 	cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
1219 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
1220 	cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
1221 	cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
1222 	cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
1223 	cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
1224 	cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
1225 	cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
1226 	cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
1227 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
1228 	if (reg) {
1229 		dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
1230 		if (reg & CXL_PSL_DSISR_TRANS)
1231 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
1232 		else
1233 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
1234 	}
1235 	if (afu->adapter->native->sl_ops->register_serr_irq) {
1236 		reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
1237 		if (reg) {
1238 			if (reg & ~0xffff)
1239 				dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
1240 			cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
1241 		}
1242 	}
1243 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
1244 	if (reg) {
1245 		dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
1246 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
1247 	}
1248 
1249 	return 0;
1250 }
1251 
1252 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
1253 /*
1254  * afu_eb_read:
1255  * Called from sysfs and reads the afu error info buffer. The h/w only supports
1256  * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
1257  * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
1258  */
1259 ssize_t cxl_pci_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
1260 				loff_t off, size_t count)
1261 {
1262 	loff_t aligned_start, aligned_end;
1263 	size_t aligned_length;
1264 	void *tbuf;
1265 	const void __iomem *ebuf = afu->native->afu_desc_mmio + afu->eb_offset;
1266 
1267 	if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
1268 		return 0;
1269 
1270 	/* calculate aligned read window */
1271 	count = min((size_t)(afu->eb_len - off), count);
1272 	aligned_start = round_down(off, 8);
1273 	aligned_end = round_up(off + count, 8);
1274 	aligned_length = aligned_end - aligned_start;
1275 
1276 	/* max we can copy in one read is PAGE_SIZE */
1277 	if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
1278 		aligned_length = ERR_BUFF_MAX_COPY_SIZE;
1279 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
1280 	}
1281 
1282 	/* use bounce buffer for copy */
1283 	tbuf = (void *)__get_free_page(GFP_KERNEL);
1284 	if (!tbuf)
1285 		return -ENOMEM;
1286 
1287 	/* perform aligned read from the mmio region */
1288 	memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
1289 	memcpy(buf, tbuf + (off & 0x7), count);
1290 
1291 	free_page((unsigned long)tbuf);
1292 
1293 	return count;
1294 }
1295 
1296 static int pci_configure_afu(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
1297 {
1298 	int rc;
1299 
1300 	if ((rc = pci_map_slice_regs(afu, adapter, dev)))
1301 		return rc;
1302 
1303 	if (adapter->native->sl_ops->sanitise_afu_regs) {
1304 		rc = adapter->native->sl_ops->sanitise_afu_regs(afu);
1305 		if (rc)
1306 			goto err1;
1307 	}
1308 
1309 	/* We need to reset the AFU before we can read the AFU descriptor */
1310 	if ((rc = cxl_ops->afu_reset(afu)))
1311 		goto err1;
1312 
1313 	if (cxl_verbose)
1314 		dump_afu_descriptor(afu);
1315 
1316 	if ((rc = cxl_read_afu_descriptor(afu)))
1317 		goto err1;
1318 
1319 	if ((rc = cxl_afu_descriptor_looks_ok(afu)))
1320 		goto err1;
1321 
1322 	if (adapter->native->sl_ops->afu_regs_init)
1323 		if ((rc = adapter->native->sl_ops->afu_regs_init(afu)))
1324 			goto err1;
1325 
1326 	if (adapter->native->sl_ops->register_serr_irq)
1327 		if ((rc = adapter->native->sl_ops->register_serr_irq(afu)))
1328 			goto err1;
1329 
1330 	if ((rc = cxl_native_register_psl_irq(afu)))
1331 		goto err2;
1332 
1333 	atomic_set(&afu->configured_state, 0);
1334 	return 0;
1335 
1336 err2:
1337 	if (adapter->native->sl_ops->release_serr_irq)
1338 		adapter->native->sl_ops->release_serr_irq(afu);
1339 err1:
1340 	pci_unmap_slice_regs(afu);
1341 	return rc;
1342 }
1343 
1344 static void pci_deconfigure_afu(struct cxl_afu *afu)
1345 {
1346 	/*
1347 	 * It's okay to deconfigure when AFU is already locked, otherwise wait
1348 	 * until there are no readers
1349 	 */
1350 	if (atomic_read(&afu->configured_state) != -1) {
1351 		while (atomic_cmpxchg(&afu->configured_state, 0, -1) != -1)
1352 			schedule();
1353 	}
1354 	cxl_native_release_psl_irq(afu);
1355 	if (afu->adapter->native->sl_ops->release_serr_irq)
1356 		afu->adapter->native->sl_ops->release_serr_irq(afu);
1357 	pci_unmap_slice_regs(afu);
1358 }
1359 
1360 static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
1361 {
1362 	struct cxl_afu *afu;
1363 	int rc = -ENOMEM;
1364 
1365 	afu = cxl_alloc_afu(adapter, slice);
1366 	if (!afu)
1367 		return -ENOMEM;
1368 
1369 	afu->native = kzalloc(sizeof(struct cxl_afu_native), GFP_KERNEL);
1370 	if (!afu->native)
1371 		goto err_free_afu;
1372 
1373 	mutex_init(&afu->native->spa_mutex);
1374 
1375 	rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice);
1376 	if (rc)
1377 		goto err_free_native;
1378 
1379 	rc = pci_configure_afu(afu, adapter, dev);
1380 	if (rc)
1381 		goto err_free_native;
1382 
1383 	/* Don't care if this fails */
1384 	cxl_debugfs_afu_add(afu);
1385 
1386 	/*
1387 	 * After we call this function we must not free the afu directly, even
1388 	 * if it returns an error!
1389 	 */
1390 	if ((rc = cxl_register_afu(afu)))
1391 		goto err_put1;
1392 
1393 	if ((rc = cxl_sysfs_afu_add(afu)))
1394 		goto err_put1;
1395 
1396 	adapter->afu[afu->slice] = afu;
1397 
1398 	if ((rc = cxl_pci_vphb_add(afu)))
1399 		dev_info(&afu->dev, "Can't register vPHB\n");
1400 
1401 	return 0;
1402 
1403 err_put1:
1404 	pci_deconfigure_afu(afu);
1405 	cxl_debugfs_afu_remove(afu);
1406 	device_unregister(&afu->dev);
1407 	return rc;
1408 
1409 err_free_native:
1410 	kfree(afu->native);
1411 err_free_afu:
1412 	kfree(afu);
1413 	return rc;
1414 
1415 }
1416 
1417 static void cxl_pci_remove_afu(struct cxl_afu *afu)
1418 {
1419 	pr_devel("%s\n", __func__);
1420 
1421 	if (!afu)
1422 		return;
1423 
1424 	cxl_pci_vphb_remove(afu);
1425 	cxl_sysfs_afu_remove(afu);
1426 	cxl_debugfs_afu_remove(afu);
1427 
1428 	spin_lock(&afu->adapter->afu_list_lock);
1429 	afu->adapter->afu[afu->slice] = NULL;
1430 	spin_unlock(&afu->adapter->afu_list_lock);
1431 
1432 	cxl_context_detach_all(afu);
1433 	cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
1434 
1435 	pci_deconfigure_afu(afu);
1436 	device_unregister(&afu->dev);
1437 }
1438 
1439 int cxl_pci_reset(struct cxl *adapter)
1440 {
1441 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1442 	int rc;
1443 
1444 	if (adapter->perst_same_image) {
1445 		dev_warn(&dev->dev,
1446 			 "cxl: refusing to reset/reflash when perst_reloads_same_image is set.\n");
1447 		return -EINVAL;
1448 	}
1449 
1450 	dev_info(&dev->dev, "CXL reset\n");
1451 
1452 	/*
1453 	 * The adapter is about to be reset, so ignore errors.
1454 	 * Not supported on P9 DD1
1455 	 */
1456 	if ((cxl_is_power8()) || (!(cxl_is_power9_dd1())))
1457 		cxl_data_cache_flush(adapter);
1458 
1459 	/* pcie_warm_reset requests a fundamental pci reset which includes a
1460 	 * PERST assert/deassert.  PERST triggers a loading of the image
1461 	 * if "user" or "factory" is selected in sysfs */
1462 	if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
1463 		dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
1464 		return rc;
1465 	}
1466 
1467 	return rc;
1468 }
1469 
1470 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
1471 {
1472 	if (pci_request_region(dev, 2, "priv 2 regs"))
1473 		goto err1;
1474 	if (pci_request_region(dev, 0, "priv 1 regs"))
1475 		goto err2;
1476 
1477 	pr_devel("cxl_map_adapter_regs: p1: %#016llx %#llx, p2: %#016llx %#llx",
1478 			p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
1479 
1480 	if (!(adapter->native->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
1481 		goto err3;
1482 
1483 	if (!(adapter->native->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
1484 		goto err4;
1485 
1486 	return 0;
1487 
1488 err4:
1489 	iounmap(adapter->native->p1_mmio);
1490 	adapter->native->p1_mmio = NULL;
1491 err3:
1492 	pci_release_region(dev, 0);
1493 err2:
1494 	pci_release_region(dev, 2);
1495 err1:
1496 	return -ENOMEM;
1497 }
1498 
1499 static void cxl_unmap_adapter_regs(struct cxl *adapter)
1500 {
1501 	if (adapter->native->p1_mmio) {
1502 		iounmap(adapter->native->p1_mmio);
1503 		adapter->native->p1_mmio = NULL;
1504 		pci_release_region(to_pci_dev(adapter->dev.parent), 2);
1505 	}
1506 	if (adapter->native->p2_mmio) {
1507 		iounmap(adapter->native->p2_mmio);
1508 		adapter->native->p2_mmio = NULL;
1509 		pci_release_region(to_pci_dev(adapter->dev.parent), 0);
1510 	}
1511 }
1512 
1513 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
1514 {
1515 	int vsec;
1516 	u32 afu_desc_off, afu_desc_size;
1517 	u32 ps_off, ps_size;
1518 	u16 vseclen;
1519 	u8 image_state;
1520 
1521 	if (!(vsec = find_cxl_vsec(dev))) {
1522 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
1523 		return -ENODEV;
1524 	}
1525 
1526 	CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
1527 	if (vseclen < CXL_VSEC_MIN_SIZE) {
1528 		dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
1529 		return -EINVAL;
1530 	}
1531 
1532 	CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
1533 	CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
1534 	CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
1535 	CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
1536 	CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
1537 	CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
1538 	adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1539 	adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
1540 	adapter->perst_loads_image = !!(image_state & CXL_VSEC_PERST_LOADS_IMAGE);
1541 
1542 	CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
1543 	CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
1544 	CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
1545 	CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
1546 	CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
1547 
1548 	/* Convert everything to bytes, because there is NO WAY I'd look at the
1549 	 * code a month later and forget what units these are in ;-) */
1550 	adapter->native->ps_off = ps_off * 64 * 1024;
1551 	adapter->ps_size = ps_size * 64 * 1024;
1552 	adapter->native->afu_desc_off = afu_desc_off * 64 * 1024;
1553 	adapter->native->afu_desc_size = afu_desc_size * 64 * 1024;
1554 
1555 	/* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
1556 	adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
1557 
1558 	return 0;
1559 }
1560 
1561 /*
1562  * Workaround a PCIe Host Bridge defect on some cards, that can cause
1563  * malformed Transaction Layer Packet (TLP) errors to be erroneously
1564  * reported. Mask this error in the Uncorrectable Error Mask Register.
1565  *
1566  * The upper nibble of the PSL revision is used to distinguish between
1567  * different cards. The affected ones have it set to 0.
1568  */
1569 static void cxl_fixup_malformed_tlp(struct cxl *adapter, struct pci_dev *dev)
1570 {
1571 	int aer;
1572 	u32 data;
1573 
1574 	if (adapter->psl_rev & 0xf000)
1575 		return;
1576 	if (!(aer = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR)))
1577 		return;
1578 	pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &data);
1579 	if (data & PCI_ERR_UNC_MALF_TLP)
1580 		if (data & PCI_ERR_UNC_INTN)
1581 			return;
1582 	data |= PCI_ERR_UNC_MALF_TLP;
1583 	data |= PCI_ERR_UNC_INTN;
1584 	pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, data);
1585 }
1586 
1587 static bool cxl_compatible_caia_version(struct cxl *adapter)
1588 {
1589 	if (cxl_is_power8() && (adapter->caia_major == 1))
1590 		return true;
1591 
1592 	if (cxl_is_power9() && (adapter->caia_major == 2))
1593 		return true;
1594 
1595 	return false;
1596 }
1597 
1598 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
1599 {
1600 	if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
1601 		return -EBUSY;
1602 
1603 	if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
1604 		dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
1605 		return -EINVAL;
1606 	}
1607 
1608 	if (!cxl_compatible_caia_version(adapter)) {
1609 		dev_info(&dev->dev, "Ignoring card. PSL type is not supported (caia version: %d)\n",
1610 			 adapter->caia_major);
1611 		return -ENODEV;
1612 	}
1613 
1614 	if (!adapter->slices) {
1615 		/* Once we support dynamic reprogramming we can use the card if
1616 		 * it supports loadable AFUs */
1617 		dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
1618 		return -EINVAL;
1619 	}
1620 
1621 	if (!adapter->native->afu_desc_off || !adapter->native->afu_desc_size) {
1622 		dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
1623 		return -EINVAL;
1624 	}
1625 
1626 	if (adapter->ps_size > p2_size(dev) - adapter->native->ps_off) {
1627 		dev_err(&dev->dev, "ABORTING: Problem state size larger than "
1628 				   "available in BAR2: 0x%llx > 0x%llx\n",
1629 			 adapter->ps_size, p2_size(dev) - adapter->native->ps_off);
1630 		return -EINVAL;
1631 	}
1632 
1633 	return 0;
1634 }
1635 
1636 ssize_t cxl_pci_read_adapter_vpd(struct cxl *adapter, void *buf, size_t len)
1637 {
1638 	return pci_read_vpd(to_pci_dev(adapter->dev.parent), 0, len, buf);
1639 }
1640 
1641 static void cxl_release_adapter(struct device *dev)
1642 {
1643 	struct cxl *adapter = to_cxl_adapter(dev);
1644 
1645 	pr_devel("cxl_release_adapter\n");
1646 
1647 	cxl_remove_adapter_nr(adapter);
1648 
1649 	kfree(adapter->native);
1650 	kfree(adapter);
1651 }
1652 
1653 #define CXL_PSL_ErrIVTE_tberror (0x1ull << (63-31))
1654 
1655 static int sanitise_adapter_regs(struct cxl *adapter)
1656 {
1657 	int rc = 0;
1658 
1659 	/* Clear PSL tberror bit by writing 1 to it */
1660 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, CXL_PSL_ErrIVTE_tberror);
1661 
1662 	if (adapter->native->sl_ops->invalidate_all) {
1663 		/* do not invalidate ERAT entries when not reloading on PERST */
1664 		if (cxl_is_power9() && (adapter->perst_loads_image))
1665 			return 0;
1666 		rc = adapter->native->sl_ops->invalidate_all(adapter);
1667 	}
1668 
1669 	return rc;
1670 }
1671 
1672 /* This should contain *only* operations that can safely be done in
1673  * both creation and recovery.
1674  */
1675 static int cxl_configure_adapter(struct cxl *adapter, struct pci_dev *dev)
1676 {
1677 	int rc;
1678 
1679 	adapter->dev.parent = &dev->dev;
1680 	adapter->dev.release = cxl_release_adapter;
1681 	pci_set_drvdata(dev, adapter);
1682 
1683 	rc = pci_enable_device(dev);
1684 	if (rc) {
1685 		dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1686 		return rc;
1687 	}
1688 
1689 	if ((rc = cxl_read_vsec(adapter, dev)))
1690 		return rc;
1691 
1692 	if ((rc = cxl_vsec_looks_ok(adapter, dev)))
1693 	        return rc;
1694 
1695 	cxl_fixup_malformed_tlp(adapter, dev);
1696 
1697 	if ((rc = setup_cxl_bars(dev)))
1698 		return rc;
1699 
1700 	if ((rc = setup_cxl_protocol_area(dev)))
1701 		return rc;
1702 
1703 	if ((rc = cxl_update_image_control(adapter)))
1704 		return rc;
1705 
1706 	if ((rc = cxl_map_adapter_regs(adapter, dev)))
1707 		return rc;
1708 
1709 	if ((rc = sanitise_adapter_regs(adapter)))
1710 		goto err;
1711 
1712 	if ((rc = adapter->native->sl_ops->adapter_regs_init(adapter, dev)))
1713 		goto err;
1714 
1715 	/* Required for devices using CAPP DMA mode, harmless for others */
1716 	pci_set_master(dev);
1717 
1718 	if ((rc = pnv_phb_to_cxl_mode(dev, adapter->native->sl_ops->capi_mode)))
1719 		goto err;
1720 
1721 	/* If recovery happened, the last step is to turn on snooping.
1722 	 * In the non-recovery case this has no effect */
1723 	if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON)))
1724 		goto err;
1725 
1726 	/* Ignore error, adapter init is not dependant on timebase sync */
1727 	cxl_setup_psl_timebase(adapter, dev);
1728 
1729 	if ((rc = cxl_native_register_psl_err_irq(adapter)))
1730 		goto err;
1731 
1732 	return 0;
1733 
1734 err:
1735 	cxl_unmap_adapter_regs(adapter);
1736 	return rc;
1737 
1738 }
1739 
1740 static void cxl_deconfigure_adapter(struct cxl *adapter)
1741 {
1742 	struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1743 
1744 	cxl_native_release_psl_err_irq(adapter);
1745 	cxl_unmap_adapter_regs(adapter);
1746 
1747 	pci_disable_device(pdev);
1748 }
1749 
1750 static void cxl_stop_trace_psl9(struct cxl *adapter)
1751 {
1752 	int traceid;
1753 	u64 trace_state, trace_mask;
1754 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
1755 
1756 	/* read each tracearray state and issue mmio to stop them is needed */
1757 	for (traceid = 0; traceid <= CXL_PSL9_TRACEID_MAX; ++traceid) {
1758 		trace_state = cxl_p1_read(adapter, CXL_PSL9_CTCCFG);
1759 		trace_mask = (0x3ULL << (62 - traceid * 2));
1760 		trace_state = (trace_state & trace_mask) >> (62 - traceid * 2);
1761 		dev_dbg(&dev->dev, "cxl: Traceid-%d trace_state=0x%0llX\n",
1762 			traceid, trace_state);
1763 
1764 		/* issue mmio if the trace array isn't in FIN state */
1765 		if (trace_state != CXL_PSL9_TRACESTATE_FIN)
1766 			cxl_p1_write(adapter, CXL_PSL9_TRACECFG,
1767 				     0x8400000000000000ULL | traceid);
1768 	}
1769 }
1770 
1771 static void cxl_stop_trace_psl8(struct cxl *adapter)
1772 {
1773 	int slice;
1774 
1775 	/* Stop the trace */
1776 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x8000000000000017LL);
1777 
1778 	/* Stop the slice traces */
1779 	spin_lock(&adapter->afu_list_lock);
1780 	for (slice = 0; slice < adapter->slices; slice++) {
1781 		if (adapter->afu[slice])
1782 			cxl_p1n_write(adapter->afu[slice], CXL_PSL_SLICE_TRACE,
1783 				      0x8000000000000000LL);
1784 	}
1785 	spin_unlock(&adapter->afu_list_lock);
1786 }
1787 
1788 static const struct cxl_service_layer_ops psl9_ops = {
1789 	.adapter_regs_init = init_implementation_adapter_regs_psl9,
1790 	.invalidate_all = cxl_invalidate_all_psl9,
1791 	.afu_regs_init = init_implementation_afu_regs_psl9,
1792 	.sanitise_afu_regs = sanitise_afu_regs_psl9,
1793 	.register_serr_irq = cxl_native_register_serr_irq,
1794 	.release_serr_irq = cxl_native_release_serr_irq,
1795 	.handle_interrupt = cxl_irq_psl9,
1796 	.fail_irq = cxl_fail_irq_psl,
1797 	.activate_dedicated_process = cxl_activate_dedicated_process_psl9,
1798 	.attach_afu_directed = cxl_attach_afu_directed_psl9,
1799 	.attach_dedicated_process = cxl_attach_dedicated_process_psl9,
1800 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl9,
1801 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl9,
1802 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl9,
1803 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl9,
1804 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl9,
1805 	.debugfs_stop_trace = cxl_stop_trace_psl9,
1806 	.write_timebase_ctrl = write_timebase_ctrl_psl9,
1807 	.timebase_read = timebase_read_psl9,
1808 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1809 	.needs_reset_before_disable = true,
1810 };
1811 
1812 static const struct cxl_service_layer_ops psl8_ops = {
1813 	.adapter_regs_init = init_implementation_adapter_regs_psl8,
1814 	.invalidate_all = cxl_invalidate_all_psl8,
1815 	.afu_regs_init = init_implementation_afu_regs_psl8,
1816 	.sanitise_afu_regs = sanitise_afu_regs_psl8,
1817 	.register_serr_irq = cxl_native_register_serr_irq,
1818 	.release_serr_irq = cxl_native_release_serr_irq,
1819 	.handle_interrupt = cxl_irq_psl8,
1820 	.fail_irq = cxl_fail_irq_psl,
1821 	.activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1822 	.attach_afu_directed = cxl_attach_afu_directed_psl8,
1823 	.attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1824 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1825 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_psl8,
1826 	.debugfs_add_afu_regs = cxl_debugfs_add_afu_regs_psl8,
1827 	.psl_irq_dump_registers = cxl_native_irq_dump_regs_psl8,
1828 	.err_irq_dump_registers = cxl_native_err_irq_dump_regs_psl8,
1829 	.debugfs_stop_trace = cxl_stop_trace_psl8,
1830 	.write_timebase_ctrl = write_timebase_ctrl_psl8,
1831 	.timebase_read = timebase_read_psl8,
1832 	.capi_mode = OPAL_PHB_CAPI_MODE_CAPI,
1833 	.needs_reset_before_disable = true,
1834 };
1835 
1836 static const struct cxl_service_layer_ops xsl_ops = {
1837 	.adapter_regs_init = init_implementation_adapter_regs_xsl,
1838 	.invalidate_all = cxl_invalidate_all_psl8,
1839 	.sanitise_afu_regs = sanitise_afu_regs_psl8,
1840 	.handle_interrupt = cxl_irq_psl8,
1841 	.fail_irq = cxl_fail_irq_psl,
1842 	.activate_dedicated_process = cxl_activate_dedicated_process_psl8,
1843 	.attach_afu_directed = cxl_attach_afu_directed_psl8,
1844 	.attach_dedicated_process = cxl_attach_dedicated_process_psl8,
1845 	.update_dedicated_ivtes = cxl_update_dedicated_ivtes_psl8,
1846 	.debugfs_add_adapter_regs = cxl_debugfs_add_adapter_regs_xsl,
1847 	.write_timebase_ctrl = write_timebase_ctrl_xsl,
1848 	.timebase_read = timebase_read_xsl,
1849 	.capi_mode = OPAL_PHB_CAPI_MODE_DMA,
1850 };
1851 
1852 static void set_sl_ops(struct cxl *adapter, struct pci_dev *dev)
1853 {
1854 	if (dev->vendor == PCI_VENDOR_ID_MELLANOX && dev->device == 0x1013) {
1855 		/* Mellanox CX-4 */
1856 		dev_info(&dev->dev, "Device uses an XSL\n");
1857 		adapter->native->sl_ops = &xsl_ops;
1858 		adapter->min_pe = 1; /* Workaround for CX-4 hardware bug */
1859 	} else {
1860 		if (cxl_is_power8()) {
1861 			dev_info(&dev->dev, "Device uses a PSL8\n");
1862 			adapter->native->sl_ops = &psl8_ops;
1863 		} else {
1864 			dev_info(&dev->dev, "Device uses a PSL9\n");
1865 			adapter->native->sl_ops = &psl9_ops;
1866 		}
1867 	}
1868 }
1869 
1870 
1871 static struct cxl *cxl_pci_init_adapter(struct pci_dev *dev)
1872 {
1873 	struct cxl *adapter;
1874 	int rc;
1875 
1876 	adapter = cxl_alloc_adapter();
1877 	if (!adapter)
1878 		return ERR_PTR(-ENOMEM);
1879 
1880 	adapter->native = kzalloc(sizeof(struct cxl_native), GFP_KERNEL);
1881 	if (!adapter->native) {
1882 		rc = -ENOMEM;
1883 		goto err_release;
1884 	}
1885 
1886 	set_sl_ops(adapter, dev);
1887 
1888 	/* Set defaults for parameters which need to persist over
1889 	 * configure/reconfigure
1890 	 */
1891 	adapter->perst_loads_image = true;
1892 	adapter->perst_same_image = false;
1893 
1894 	rc = cxl_configure_adapter(adapter, dev);
1895 	if (rc) {
1896 		pci_disable_device(dev);
1897 		goto err_release;
1898 	}
1899 
1900 	/* Don't care if this one fails: */
1901 	cxl_debugfs_adapter_add(adapter);
1902 
1903 	/*
1904 	 * After we call this function we must not free the adapter directly,
1905 	 * even if it returns an error!
1906 	 */
1907 	if ((rc = cxl_register_adapter(adapter)))
1908 		goto err_put1;
1909 
1910 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1911 		goto err_put1;
1912 
1913 	/* Release the context lock as adapter is configured */
1914 	cxl_adapter_context_unlock(adapter);
1915 
1916 	return adapter;
1917 
1918 err_put1:
1919 	/* This should mirror cxl_remove_adapter, except without the
1920 	 * sysfs parts
1921 	 */
1922 	cxl_debugfs_adapter_remove(adapter);
1923 	cxl_deconfigure_adapter(adapter);
1924 	device_unregister(&adapter->dev);
1925 	return ERR_PTR(rc);
1926 
1927 err_release:
1928 	cxl_release_adapter(&adapter->dev);
1929 	return ERR_PTR(rc);
1930 }
1931 
1932 static void cxl_pci_remove_adapter(struct cxl *adapter)
1933 {
1934 	pr_devel("cxl_remove_adapter\n");
1935 
1936 	cxl_sysfs_adapter_remove(adapter);
1937 	cxl_debugfs_adapter_remove(adapter);
1938 
1939 	/*
1940 	 * Flush adapter datacache as its about to be removed.
1941 	 * Not supported on P9 DD1.
1942 	 */
1943 	if ((cxl_is_power8()) || (!(cxl_is_power9_dd1())))
1944 		cxl_data_cache_flush(adapter);
1945 
1946 	cxl_deconfigure_adapter(adapter);
1947 
1948 	device_unregister(&adapter->dev);
1949 }
1950 
1951 #define CXL_MAX_PCIEX_PARENT 2
1952 
1953 int cxl_slot_is_switched(struct pci_dev *dev)
1954 {
1955 	struct device_node *np;
1956 	int depth = 0;
1957 	const __be32 *prop;
1958 
1959 	if (!(np = pci_device_to_OF_node(dev))) {
1960 		pr_err("cxl: np = NULL\n");
1961 		return -ENODEV;
1962 	}
1963 	of_node_get(np);
1964 	while (np) {
1965 		np = of_get_next_parent(np);
1966 		prop = of_get_property(np, "device_type", NULL);
1967 		if (!prop || strcmp((char *)prop, "pciex"))
1968 			break;
1969 		depth++;
1970 	}
1971 	of_node_put(np);
1972 	return (depth > CXL_MAX_PCIEX_PARENT);
1973 }
1974 
1975 bool cxl_slot_is_supported(struct pci_dev *dev, int flags)
1976 {
1977 	if (!cpu_has_feature(CPU_FTR_HVMODE))
1978 		return false;
1979 
1980 	if ((flags & CXL_SLOT_FLAG_DMA) && (!pvr_version_is(PVR_POWER8NVL))) {
1981 		/*
1982 		 * CAPP DMA mode is technically supported on regular P8, but
1983 		 * will EEH if the card attempts to access memory < 4GB, which
1984 		 * we cannot realistically avoid. We might be able to work
1985 		 * around the issue, but until then return unsupported:
1986 		 */
1987 		return false;
1988 	}
1989 
1990 	if (cxl_slot_is_switched(dev))
1991 		return false;
1992 
1993 	/*
1994 	 * XXX: This gets a little tricky on regular P8 (not POWER8NVL) since
1995 	 * the CAPP can be connected to PHB 0, 1 or 2 on a first come first
1996 	 * served basis, which is racy to check from here. If we need to
1997 	 * support this in future we might need to consider having this
1998 	 * function effectively reserve it ahead of time.
1999 	 *
2000 	 * Currently, the only user of this API is the Mellanox CX4, which is
2001 	 * only supported on P8NVL due to the above mentioned limitation of
2002 	 * CAPP DMA mode and therefore does not need to worry about this. If the
2003 	 * issue with CAPP DMA mode is later worked around on P8 we might need
2004 	 * to revisit this.
2005 	 */
2006 
2007 	return true;
2008 }
2009 EXPORT_SYMBOL_GPL(cxl_slot_is_supported);
2010 
2011 
2012 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
2013 {
2014 	struct cxl *adapter;
2015 	int slice;
2016 	int rc;
2017 
2018 	if (cxl_pci_is_vphb_device(dev)) {
2019 		dev_dbg(&dev->dev, "cxl_init_adapter: Ignoring cxl vphb device\n");
2020 		return -ENODEV;
2021 	}
2022 
2023 	if (cxl_slot_is_switched(dev)) {
2024 		dev_info(&dev->dev, "Ignoring card on incompatible PCI slot\n");
2025 		return -ENODEV;
2026 	}
2027 
2028 	if (cxl_is_power9() && !radix_enabled()) {
2029 		dev_info(&dev->dev, "Only Radix mode supported\n");
2030 		return -ENODEV;
2031 	}
2032 
2033 	if (cxl_verbose)
2034 		dump_cxl_config_space(dev);
2035 
2036 	adapter = cxl_pci_init_adapter(dev);
2037 	if (IS_ERR(adapter)) {
2038 		dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
2039 		return PTR_ERR(adapter);
2040 	}
2041 
2042 	for (slice = 0; slice < adapter->slices; slice++) {
2043 		if ((rc = pci_init_afu(adapter, slice, dev))) {
2044 			dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
2045 			continue;
2046 		}
2047 
2048 		rc = cxl_afu_select_best_mode(adapter->afu[slice]);
2049 		if (rc)
2050 			dev_err(&dev->dev, "AFU %i failed to start: %i\n", slice, rc);
2051 	}
2052 
2053 	if (pnv_pci_on_cxl_phb(dev) && adapter->slices >= 1)
2054 		pnv_cxl_phb_set_peer_afu(dev, adapter->afu[0]);
2055 
2056 	return 0;
2057 }
2058 
2059 static void cxl_remove(struct pci_dev *dev)
2060 {
2061 	struct cxl *adapter = pci_get_drvdata(dev);
2062 	struct cxl_afu *afu;
2063 	int i;
2064 
2065 	/*
2066 	 * Lock to prevent someone grabbing a ref through the adapter list as
2067 	 * we are removing it
2068 	 */
2069 	for (i = 0; i < adapter->slices; i++) {
2070 		afu = adapter->afu[i];
2071 		cxl_pci_remove_afu(afu);
2072 	}
2073 	cxl_pci_remove_adapter(adapter);
2074 }
2075 
2076 static pci_ers_result_t cxl_vphb_error_detected(struct cxl_afu *afu,
2077 						pci_channel_state_t state)
2078 {
2079 	struct pci_dev *afu_dev;
2080 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
2081 	pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
2082 
2083 	/* There should only be one entry, but go through the list
2084 	 * anyway
2085 	 */
2086 	if (afu->phb == NULL)
2087 		return result;
2088 
2089 	list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2090 		if (!afu_dev->driver)
2091 			continue;
2092 
2093 		afu_dev->error_state = state;
2094 
2095 		if (afu_dev->driver->err_handler)
2096 			afu_result = afu_dev->driver->err_handler->error_detected(afu_dev,
2097 										  state);
2098 		/* Disconnect trumps all, NONE trumps NEED_RESET */
2099 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2100 			result = PCI_ERS_RESULT_DISCONNECT;
2101 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
2102 			 (result == PCI_ERS_RESULT_NEED_RESET))
2103 			result = PCI_ERS_RESULT_NONE;
2104 	}
2105 	return result;
2106 }
2107 
2108 static pci_ers_result_t cxl_pci_error_detected(struct pci_dev *pdev,
2109 					       pci_channel_state_t state)
2110 {
2111 	struct cxl *adapter = pci_get_drvdata(pdev);
2112 	struct cxl_afu *afu;
2113 	pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET, afu_result;
2114 	int i;
2115 
2116 	/* At this point, we could still have an interrupt pending.
2117 	 * Let's try to get them out of the way before they do
2118 	 * anything we don't like.
2119 	 */
2120 	schedule();
2121 
2122 	/* If we're permanently dead, give up. */
2123 	if (state == pci_channel_io_perm_failure) {
2124 		for (i = 0; i < adapter->slices; i++) {
2125 			afu = adapter->afu[i];
2126 			/*
2127 			 * Tell the AFU drivers; but we don't care what they
2128 			 * say, we're going away.
2129 			 */
2130 			cxl_vphb_error_detected(afu, state);
2131 		}
2132 		return PCI_ERS_RESULT_DISCONNECT;
2133 	}
2134 
2135 	/* Are we reflashing?
2136 	 *
2137 	 * If we reflash, we could come back as something entirely
2138 	 * different, including a non-CAPI card. As such, by default
2139 	 * we don't participate in the process. We'll be unbound and
2140 	 * the slot re-probed. (TODO: check EEH doesn't blindly rebind
2141 	 * us!)
2142 	 *
2143 	 * However, this isn't the entire story: for reliablity
2144 	 * reasons, we usually want to reflash the FPGA on PERST in
2145 	 * order to get back to a more reliable known-good state.
2146 	 *
2147 	 * This causes us a bit of a problem: if we reflash we can't
2148 	 * trust that we'll come back the same - we could have a new
2149 	 * image and been PERSTed in order to load that
2150 	 * image. However, most of the time we actually *will* come
2151 	 * back the same - for example a regular EEH event.
2152 	 *
2153 	 * Therefore, we allow the user to assert that the image is
2154 	 * indeed the same and that we should continue on into EEH
2155 	 * anyway.
2156 	 */
2157 	if (adapter->perst_loads_image && !adapter->perst_same_image) {
2158 		/* TODO take the PHB out of CXL mode */
2159 		dev_info(&pdev->dev, "reflashing, so opting out of EEH!\n");
2160 		return PCI_ERS_RESULT_NONE;
2161 	}
2162 
2163 	/*
2164 	 * At this point, we want to try to recover.  We'll always
2165 	 * need a complete slot reset: we don't trust any other reset.
2166 	 *
2167 	 * Now, we go through each AFU:
2168 	 *  - We send the driver, if bound, an error_detected callback.
2169 	 *    We expect it to clean up, but it can also tell us to give
2170 	 *    up and permanently detach the card. To simplify things, if
2171 	 *    any bound AFU driver doesn't support EEH, we give up on EEH.
2172 	 *
2173 	 *  - We detach all contexts associated with the AFU. This
2174 	 *    does not free them, but puts them into a CLOSED state
2175 	 *    which causes any the associated files to return useful
2176 	 *    errors to userland. It also unmaps, but does not free,
2177 	 *    any IRQs.
2178 	 *
2179 	 *  - We clean up our side: releasing and unmapping resources we hold
2180 	 *    so we can wire them up again when the hardware comes back up.
2181 	 *
2182 	 * Driver authors should note:
2183 	 *
2184 	 *  - Any contexts you create in your kernel driver (except
2185 	 *    those associated with anonymous file descriptors) are
2186 	 *    your responsibility to free and recreate. Likewise with
2187 	 *    any attached resources.
2188 	 *
2189 	 *  - We will take responsibility for re-initialising the
2190 	 *    device context (the one set up for you in
2191 	 *    cxl_pci_enable_device_hook and accessed through
2192 	 *    cxl_get_context). If you've attached IRQs or other
2193 	 *    resources to it, they remains yours to free.
2194 	 *
2195 	 * You can call the same functions to release resources as you
2196 	 * normally would: we make sure that these functions continue
2197 	 * to work when the hardware is down.
2198 	 *
2199 	 * Two examples:
2200 	 *
2201 	 * 1) If you normally free all your resources at the end of
2202 	 *    each request, or if you use anonymous FDs, your
2203 	 *    error_detected callback can simply set a flag to tell
2204 	 *    your driver not to start any new calls. You can then
2205 	 *    clear the flag in the resume callback.
2206 	 *
2207 	 * 2) If you normally allocate your resources on startup:
2208 	 *     * Set a flag in error_detected as above.
2209 	 *     * Let CXL detach your contexts.
2210 	 *     * In slot_reset, free the old resources and allocate new ones.
2211 	 *     * In resume, clear the flag to allow things to start.
2212 	 */
2213 	for (i = 0; i < adapter->slices; i++) {
2214 		afu = adapter->afu[i];
2215 
2216 		afu_result = cxl_vphb_error_detected(afu, state);
2217 
2218 		cxl_context_detach_all(afu);
2219 		cxl_ops->afu_deactivate_mode(afu, afu->current_mode);
2220 		pci_deconfigure_afu(afu);
2221 
2222 		/* Disconnect trumps all, NONE trumps NEED_RESET */
2223 		if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2224 			result = PCI_ERS_RESULT_DISCONNECT;
2225 		else if ((afu_result == PCI_ERS_RESULT_NONE) &&
2226 			 (result == PCI_ERS_RESULT_NEED_RESET))
2227 			result = PCI_ERS_RESULT_NONE;
2228 	}
2229 
2230 	/* should take the context lock here */
2231 	if (cxl_adapter_context_lock(adapter) != 0)
2232 		dev_warn(&adapter->dev,
2233 			 "Couldn't take context lock with %d active-contexts\n",
2234 			 atomic_read(&adapter->contexts_num));
2235 
2236 	cxl_deconfigure_adapter(adapter);
2237 
2238 	return result;
2239 }
2240 
2241 static pci_ers_result_t cxl_pci_slot_reset(struct pci_dev *pdev)
2242 {
2243 	struct cxl *adapter = pci_get_drvdata(pdev);
2244 	struct cxl_afu *afu;
2245 	struct cxl_context *ctx;
2246 	struct pci_dev *afu_dev;
2247 	pci_ers_result_t afu_result = PCI_ERS_RESULT_RECOVERED;
2248 	pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
2249 	int i;
2250 
2251 	if (cxl_configure_adapter(adapter, pdev))
2252 		goto err;
2253 
2254 	/*
2255 	 * Unlock context activation for the adapter. Ideally this should be
2256 	 * done in cxl_pci_resume but cxlflash module tries to activate the
2257 	 * master context as part of slot_reset callback.
2258 	 */
2259 	cxl_adapter_context_unlock(adapter);
2260 
2261 	for (i = 0; i < adapter->slices; i++) {
2262 		afu = adapter->afu[i];
2263 
2264 		if (pci_configure_afu(afu, adapter, pdev))
2265 			goto err;
2266 
2267 		if (cxl_afu_select_best_mode(afu))
2268 			goto err;
2269 
2270 		if (afu->phb == NULL)
2271 			continue;
2272 
2273 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2274 			/* Reset the device context.
2275 			 * TODO: make this less disruptive
2276 			 */
2277 			ctx = cxl_get_context(afu_dev);
2278 
2279 			if (ctx && cxl_release_context(ctx))
2280 				goto err;
2281 
2282 			ctx = cxl_dev_context_init(afu_dev);
2283 			if (IS_ERR(ctx))
2284 				goto err;
2285 
2286 			afu_dev->dev.archdata.cxl_ctx = ctx;
2287 
2288 			if (cxl_ops->afu_check_and_enable(afu))
2289 				goto err;
2290 
2291 			afu_dev->error_state = pci_channel_io_normal;
2292 
2293 			/* If there's a driver attached, allow it to
2294 			 * chime in on recovery. Drivers should check
2295 			 * if everything has come back OK, but
2296 			 * shouldn't start new work until we call
2297 			 * their resume function.
2298 			 */
2299 			if (!afu_dev->driver)
2300 				continue;
2301 
2302 			if (afu_dev->driver->err_handler &&
2303 			    afu_dev->driver->err_handler->slot_reset)
2304 				afu_result = afu_dev->driver->err_handler->slot_reset(afu_dev);
2305 
2306 			if (afu_result == PCI_ERS_RESULT_DISCONNECT)
2307 				result = PCI_ERS_RESULT_DISCONNECT;
2308 		}
2309 	}
2310 	return result;
2311 
2312 err:
2313 	/* All the bits that happen in both error_detected and cxl_remove
2314 	 * should be idempotent, so we don't need to worry about leaving a mix
2315 	 * of unconfigured and reconfigured resources.
2316 	 */
2317 	dev_err(&pdev->dev, "EEH recovery failed. Asking to be disconnected.\n");
2318 	return PCI_ERS_RESULT_DISCONNECT;
2319 }
2320 
2321 static void cxl_pci_resume(struct pci_dev *pdev)
2322 {
2323 	struct cxl *adapter = pci_get_drvdata(pdev);
2324 	struct cxl_afu *afu;
2325 	struct pci_dev *afu_dev;
2326 	int i;
2327 
2328 	/* Everything is back now. Drivers should restart work now.
2329 	 * This is not the place to be checking if everything came back up
2330 	 * properly, because there's no return value: do that in slot_reset.
2331 	 */
2332 	for (i = 0; i < adapter->slices; i++) {
2333 		afu = adapter->afu[i];
2334 
2335 		if (afu->phb == NULL)
2336 			continue;
2337 
2338 		list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
2339 			if (afu_dev->driver && afu_dev->driver->err_handler &&
2340 			    afu_dev->driver->err_handler->resume)
2341 				afu_dev->driver->err_handler->resume(afu_dev);
2342 		}
2343 	}
2344 }
2345 
2346 static const struct pci_error_handlers cxl_err_handler = {
2347 	.error_detected = cxl_pci_error_detected,
2348 	.slot_reset = cxl_pci_slot_reset,
2349 	.resume = cxl_pci_resume,
2350 };
2351 
2352 struct pci_driver cxl_pci_driver = {
2353 	.name = "cxl-pci",
2354 	.id_table = cxl_pci_tbl,
2355 	.probe = cxl_probe,
2356 	.remove = cxl_remove,
2357 	.shutdown = cxl_remove,
2358 	.err_handler = &cxl_err_handler,
2359 };
2360