xref: /openbmc/linux/drivers/misc/cxl/pci.c (revision 3b27d139)
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/pci-bridge.h> /* for struct pci_controller */
23 #include <asm/pnv-pci.h>
24 #include <asm/io.h>
25 
26 #include "cxl.h"
27 
28 
29 #define CXL_PCI_VSEC_ID	0x1280
30 #define CXL_VSEC_MIN_SIZE 0x80
31 
32 #define CXL_READ_VSEC_LENGTH(dev, vsec, dest)			\
33 	{							\
34 		pci_read_config_word(dev, vsec + 0x6, dest);	\
35 		*dest >>= 4;					\
36 	}
37 #define CXL_READ_VSEC_NAFUS(dev, vsec, dest) \
38 	pci_read_config_byte(dev, vsec + 0x8, dest)
39 
40 #define CXL_READ_VSEC_STATUS(dev, vsec, dest) \
41 	pci_read_config_byte(dev, vsec + 0x9, dest)
42 #define CXL_STATUS_SECOND_PORT  0x80
43 #define CXL_STATUS_MSI_X_FULL   0x40
44 #define CXL_STATUS_MSI_X_SINGLE 0x20
45 #define CXL_STATUS_FLASH_RW     0x08
46 #define CXL_STATUS_FLASH_RO     0x04
47 #define CXL_STATUS_LOADABLE_AFU 0x02
48 #define CXL_STATUS_LOADABLE_PSL 0x01
49 /* If we see these features we won't try to use the card */
50 #define CXL_UNSUPPORTED_FEATURES \
51 	(CXL_STATUS_MSI_X_FULL | CXL_STATUS_MSI_X_SINGLE)
52 
53 #define CXL_READ_VSEC_MODE_CONTROL(dev, vsec, dest) \
54 	pci_read_config_byte(dev, vsec + 0xa, dest)
55 #define CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val) \
56 	pci_write_config_byte(dev, vsec + 0xa, val)
57 #define CXL_VSEC_PROTOCOL_MASK   0xe0
58 #define CXL_VSEC_PROTOCOL_1024TB 0x80
59 #define CXL_VSEC_PROTOCOL_512TB  0x40
60 #define CXL_VSEC_PROTOCOL_256TB  0x20 /* Power 8 uses this */
61 #define CXL_VSEC_PROTOCOL_ENABLE 0x01
62 
63 #define CXL_READ_VSEC_PSL_REVISION(dev, vsec, dest) \
64 	pci_read_config_word(dev, vsec + 0xc, dest)
65 #define CXL_READ_VSEC_CAIA_MINOR(dev, vsec, dest) \
66 	pci_read_config_byte(dev, vsec + 0xe, dest)
67 #define CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, dest) \
68 	pci_read_config_byte(dev, vsec + 0xf, dest)
69 #define CXL_READ_VSEC_BASE_IMAGE(dev, vsec, dest) \
70 	pci_read_config_word(dev, vsec + 0x10, dest)
71 
72 #define CXL_READ_VSEC_IMAGE_STATE(dev, vsec, dest) \
73 	pci_read_config_byte(dev, vsec + 0x13, dest)
74 #define CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, val) \
75 	pci_write_config_byte(dev, vsec + 0x13, val)
76 #define CXL_VSEC_USER_IMAGE_LOADED 0x80 /* RO */
77 #define CXL_VSEC_PERST_LOADS_IMAGE 0x20 /* RW */
78 #define CXL_VSEC_PERST_SELECT_USER 0x10 /* RW */
79 
80 #define CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, dest) \
81 	pci_read_config_dword(dev, vsec + 0x20, dest)
82 #define CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, dest) \
83 	pci_read_config_dword(dev, vsec + 0x24, dest)
84 #define CXL_READ_VSEC_PS_OFF(dev, vsec, dest) \
85 	pci_read_config_dword(dev, vsec + 0x28, dest)
86 #define CXL_READ_VSEC_PS_SIZE(dev, vsec, dest) \
87 	pci_read_config_dword(dev, vsec + 0x2c, dest)
88 
89 
90 /* This works a little different than the p1/p2 register accesses to make it
91  * easier to pull out individual fields */
92 #define AFUD_READ(afu, off)		in_be64(afu->afu_desc_mmio + off)
93 #define AFUD_READ_LE(afu, off)		in_le64(afu->afu_desc_mmio + off)
94 #define EXTRACT_PPC_BIT(val, bit)	(!!(val & PPC_BIT(bit)))
95 #define EXTRACT_PPC_BITS(val, bs, be)	((val & PPC_BITMASK(bs, be)) >> PPC_BITLSHIFT(be))
96 
97 #define AFUD_READ_INFO(afu)		AFUD_READ(afu, 0x0)
98 #define   AFUD_NUM_INTS_PER_PROC(val)	EXTRACT_PPC_BITS(val,  0, 15)
99 #define   AFUD_NUM_PROCS(val)		EXTRACT_PPC_BITS(val, 16, 31)
100 #define   AFUD_NUM_CRS(val)		EXTRACT_PPC_BITS(val, 32, 47)
101 #define   AFUD_MULTIMODE(val)		EXTRACT_PPC_BIT(val, 48)
102 #define   AFUD_PUSH_BLOCK_TRANSFER(val)	EXTRACT_PPC_BIT(val, 55)
103 #define   AFUD_DEDICATED_PROCESS(val)	EXTRACT_PPC_BIT(val, 59)
104 #define   AFUD_AFU_DIRECTED(val)	EXTRACT_PPC_BIT(val, 61)
105 #define   AFUD_TIME_SLICED(val)		EXTRACT_PPC_BIT(val, 63)
106 #define AFUD_READ_CR(afu)		AFUD_READ(afu, 0x20)
107 #define   AFUD_CR_LEN(val)		EXTRACT_PPC_BITS(val, 8, 63)
108 #define AFUD_READ_CR_OFF(afu)		AFUD_READ(afu, 0x28)
109 #define AFUD_READ_PPPSA(afu)		AFUD_READ(afu, 0x30)
110 #define   AFUD_PPPSA_PP(val)		EXTRACT_PPC_BIT(val, 6)
111 #define   AFUD_PPPSA_PSA(val)		EXTRACT_PPC_BIT(val, 7)
112 #define   AFUD_PPPSA_LEN(val)		EXTRACT_PPC_BITS(val, 8, 63)
113 #define AFUD_READ_PPPSA_OFF(afu)	AFUD_READ(afu, 0x38)
114 #define AFUD_READ_EB(afu)		AFUD_READ(afu, 0x40)
115 #define   AFUD_EB_LEN(val)		EXTRACT_PPC_BITS(val, 8, 63)
116 #define AFUD_READ_EB_OFF(afu)		AFUD_READ(afu, 0x48)
117 
118 u16 cxl_afu_cr_read16(struct cxl_afu *afu, int cr, u64 off)
119 {
120 	u64 aligned_off = off & ~0x3L;
121 	u32 val;
122 
123 	val = cxl_afu_cr_read32(afu, cr, aligned_off);
124 	return (val >> ((off & 0x2) * 8)) & 0xffff;
125 }
126 
127 u8 cxl_afu_cr_read8(struct cxl_afu *afu, int cr, u64 off)
128 {
129 	u64 aligned_off = off & ~0x3L;
130 	u32 val;
131 
132 	val = cxl_afu_cr_read32(afu, cr, aligned_off);
133 	return (val >> ((off & 0x3) * 8)) & 0xff;
134 }
135 
136 static DEFINE_PCI_DEVICE_TABLE(cxl_pci_tbl) = {
137 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0477), },
138 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x044b), },
139 	{ PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x04cf), },
140 	{ PCI_DEVICE_CLASS(0x120000, ~0), },
141 
142 	{ }
143 };
144 MODULE_DEVICE_TABLE(pci, cxl_pci_tbl);
145 
146 
147 /*
148  * Mostly using these wrappers to avoid confusion:
149  * priv 1 is BAR2, while priv 2 is BAR0
150  */
151 static inline resource_size_t p1_base(struct pci_dev *dev)
152 {
153 	return pci_resource_start(dev, 2);
154 }
155 
156 static inline resource_size_t p1_size(struct pci_dev *dev)
157 {
158 	return pci_resource_len(dev, 2);
159 }
160 
161 static inline resource_size_t p2_base(struct pci_dev *dev)
162 {
163 	return pci_resource_start(dev, 0);
164 }
165 
166 static inline resource_size_t p2_size(struct pci_dev *dev)
167 {
168 	return pci_resource_len(dev, 0);
169 }
170 
171 static int find_cxl_vsec(struct pci_dev *dev)
172 {
173 	int vsec = 0;
174 	u16 val;
175 
176 	while ((vsec = pci_find_next_ext_capability(dev, vsec, PCI_EXT_CAP_ID_VNDR))) {
177 		pci_read_config_word(dev, vsec + 0x4, &val);
178 		if (val == CXL_PCI_VSEC_ID)
179 			return vsec;
180 	}
181 	return 0;
182 
183 }
184 
185 static void dump_cxl_config_space(struct pci_dev *dev)
186 {
187 	int vsec;
188 	u32 val;
189 
190 	dev_info(&dev->dev, "dump_cxl_config_space\n");
191 
192 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_0, &val);
193 	dev_info(&dev->dev, "BAR0: %#.8x\n", val);
194 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_1, &val);
195 	dev_info(&dev->dev, "BAR1: %#.8x\n", val);
196 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_2, &val);
197 	dev_info(&dev->dev, "BAR2: %#.8x\n", val);
198 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_3, &val);
199 	dev_info(&dev->dev, "BAR3: %#.8x\n", val);
200 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_4, &val);
201 	dev_info(&dev->dev, "BAR4: %#.8x\n", val);
202 	pci_read_config_dword(dev, PCI_BASE_ADDRESS_5, &val);
203 	dev_info(&dev->dev, "BAR5: %#.8x\n", val);
204 
205 	dev_info(&dev->dev, "p1 regs: %#llx, len: %#llx\n",
206 		p1_base(dev), p1_size(dev));
207 	dev_info(&dev->dev, "p2 regs: %#llx, len: %#llx\n",
208 		p2_base(dev), p2_size(dev));
209 	dev_info(&dev->dev, "BAR 4/5: %#llx, len: %#llx\n",
210 		pci_resource_start(dev, 4), pci_resource_len(dev, 4));
211 
212 	if (!(vsec = find_cxl_vsec(dev)))
213 		return;
214 
215 #define show_reg(name, what) \
216 	dev_info(&dev->dev, "cxl vsec: %30s: %#x\n", name, what)
217 
218 	pci_read_config_dword(dev, vsec + 0x0, &val);
219 	show_reg("Cap ID", (val >> 0) & 0xffff);
220 	show_reg("Cap Ver", (val >> 16) & 0xf);
221 	show_reg("Next Cap Ptr", (val >> 20) & 0xfff);
222 	pci_read_config_dword(dev, vsec + 0x4, &val);
223 	show_reg("VSEC ID", (val >> 0) & 0xffff);
224 	show_reg("VSEC Rev", (val >> 16) & 0xf);
225 	show_reg("VSEC Length",	(val >> 20) & 0xfff);
226 	pci_read_config_dword(dev, vsec + 0x8, &val);
227 	show_reg("Num AFUs", (val >> 0) & 0xff);
228 	show_reg("Status", (val >> 8) & 0xff);
229 	show_reg("Mode Control", (val >> 16) & 0xff);
230 	show_reg("Reserved", (val >> 24) & 0xff);
231 	pci_read_config_dword(dev, vsec + 0xc, &val);
232 	show_reg("PSL Rev", (val >> 0) & 0xffff);
233 	show_reg("CAIA Ver", (val >> 16) & 0xffff);
234 	pci_read_config_dword(dev, vsec + 0x10, &val);
235 	show_reg("Base Image Rev", (val >> 0) & 0xffff);
236 	show_reg("Reserved", (val >> 16) & 0x0fff);
237 	show_reg("Image Control", (val >> 28) & 0x3);
238 	show_reg("Reserved", (val >> 30) & 0x1);
239 	show_reg("Image Loaded", (val >> 31) & 0x1);
240 
241 	pci_read_config_dword(dev, vsec + 0x14, &val);
242 	show_reg("Reserved", val);
243 	pci_read_config_dword(dev, vsec + 0x18, &val);
244 	show_reg("Reserved", val);
245 	pci_read_config_dword(dev, vsec + 0x1c, &val);
246 	show_reg("Reserved", val);
247 
248 	pci_read_config_dword(dev, vsec + 0x20, &val);
249 	show_reg("AFU Descriptor Offset", val);
250 	pci_read_config_dword(dev, vsec + 0x24, &val);
251 	show_reg("AFU Descriptor Size", val);
252 	pci_read_config_dword(dev, vsec + 0x28, &val);
253 	show_reg("Problem State Offset", val);
254 	pci_read_config_dword(dev, vsec + 0x2c, &val);
255 	show_reg("Problem State Size", val);
256 
257 	pci_read_config_dword(dev, vsec + 0x30, &val);
258 	show_reg("Reserved", val);
259 	pci_read_config_dword(dev, vsec + 0x34, &val);
260 	show_reg("Reserved", val);
261 	pci_read_config_dword(dev, vsec + 0x38, &val);
262 	show_reg("Reserved", val);
263 	pci_read_config_dword(dev, vsec + 0x3c, &val);
264 	show_reg("Reserved", val);
265 
266 	pci_read_config_dword(dev, vsec + 0x40, &val);
267 	show_reg("PSL Programming Port", val);
268 	pci_read_config_dword(dev, vsec + 0x44, &val);
269 	show_reg("PSL Programming Control", val);
270 
271 	pci_read_config_dword(dev, vsec + 0x48, &val);
272 	show_reg("Reserved", val);
273 	pci_read_config_dword(dev, vsec + 0x4c, &val);
274 	show_reg("Reserved", val);
275 
276 	pci_read_config_dword(dev, vsec + 0x50, &val);
277 	show_reg("Flash Address Register", val);
278 	pci_read_config_dword(dev, vsec + 0x54, &val);
279 	show_reg("Flash Size Register", val);
280 	pci_read_config_dword(dev, vsec + 0x58, &val);
281 	show_reg("Flash Status/Control Register", val);
282 	pci_read_config_dword(dev, vsec + 0x58, &val);
283 	show_reg("Flash Data Port", val);
284 
285 #undef show_reg
286 }
287 
288 static void dump_afu_descriptor(struct cxl_afu *afu)
289 {
290 	u64 val, afu_cr_num, afu_cr_off, afu_cr_len;
291 	int i;
292 
293 #define show_reg(name, what) \
294 	dev_info(&afu->dev, "afu desc: %30s: %#llx\n", name, what)
295 
296 	val = AFUD_READ_INFO(afu);
297 	show_reg("num_ints_per_process", AFUD_NUM_INTS_PER_PROC(val));
298 	show_reg("num_of_processes", AFUD_NUM_PROCS(val));
299 	show_reg("num_of_afu_CRs", AFUD_NUM_CRS(val));
300 	show_reg("req_prog_mode", val & 0xffffULL);
301 	afu_cr_num = AFUD_NUM_CRS(val);
302 
303 	val = AFUD_READ(afu, 0x8);
304 	show_reg("Reserved", val);
305 	val = AFUD_READ(afu, 0x10);
306 	show_reg("Reserved", val);
307 	val = AFUD_READ(afu, 0x18);
308 	show_reg("Reserved", val);
309 
310 	val = AFUD_READ_CR(afu);
311 	show_reg("Reserved", (val >> (63-7)) & 0xff);
312 	show_reg("AFU_CR_len", AFUD_CR_LEN(val));
313 	afu_cr_len = AFUD_CR_LEN(val) * 256;
314 
315 	val = AFUD_READ_CR_OFF(afu);
316 	afu_cr_off = val;
317 	show_reg("AFU_CR_offset", val);
318 
319 	val = AFUD_READ_PPPSA(afu);
320 	show_reg("PerProcessPSA_control", (val >> (63-7)) & 0xff);
321 	show_reg("PerProcessPSA Length", AFUD_PPPSA_LEN(val));
322 
323 	val = AFUD_READ_PPPSA_OFF(afu);
324 	show_reg("PerProcessPSA_offset", val);
325 
326 	val = AFUD_READ_EB(afu);
327 	show_reg("Reserved", (val >> (63-7)) & 0xff);
328 	show_reg("AFU_EB_len", AFUD_EB_LEN(val));
329 
330 	val = AFUD_READ_EB_OFF(afu);
331 	show_reg("AFU_EB_offset", val);
332 
333 	for (i = 0; i < afu_cr_num; i++) {
334 		val = AFUD_READ_LE(afu, afu_cr_off + i * afu_cr_len);
335 		show_reg("CR Vendor", val & 0xffff);
336 		show_reg("CR Device", (val >> 16) & 0xffff);
337 	}
338 #undef show_reg
339 }
340 
341 static int init_implementation_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
342 {
343 	struct device_node *np;
344 	const __be32 *prop;
345 	u64 psl_dsnctl;
346 	u64 chipid;
347 
348 	if (!(np = pnv_pci_get_phb_node(dev)))
349 		return -ENODEV;
350 
351 	while (np && !(prop = of_get_property(np, "ibm,chip-id", NULL)))
352 		np = of_get_next_parent(np);
353 	if (!np)
354 		return -ENODEV;
355 	chipid = be32_to_cpup(prop);
356 	of_node_put(np);
357 
358 	/* Tell PSL where to route data to */
359 	psl_dsnctl = 0x02E8900002000000ULL | (chipid << (63-5));
360 	cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
361 	cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
362 	/* snoop write mask */
363 	cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
364 	/* set fir_accum */
365 	cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, 0x0800000000000000ULL);
366 	/* for debugging with trace arrays */
367 	cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
368 
369 	return 0;
370 }
371 
372 static int init_implementation_afu_regs(struct cxl_afu *afu)
373 {
374 	/* read/write masks for this slice */
375 	cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
376 	/* APC read/write masks for this slice */
377 	cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
378 	/* for debugging with trace arrays */
379 	cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
380 	cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
381 
382 	return 0;
383 }
384 
385 int cxl_setup_irq(struct cxl *adapter, unsigned int hwirq,
386 			 unsigned int virq)
387 {
388 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
389 
390 	return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
391 }
392 
393 int cxl_update_image_control(struct cxl *adapter)
394 {
395 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
396 	int rc;
397 	int vsec;
398 	u8 image_state;
399 
400 	if (!(vsec = find_cxl_vsec(dev))) {
401 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
402 		return -ENODEV;
403 	}
404 
405 	if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
406 		dev_err(&dev->dev, "failed to read image state: %i\n", rc);
407 		return rc;
408 	}
409 
410 	if (adapter->perst_loads_image)
411 		image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
412 	else
413 		image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
414 
415 	if (adapter->perst_select_user)
416 		image_state |= CXL_VSEC_PERST_SELECT_USER;
417 	else
418 		image_state &= ~CXL_VSEC_PERST_SELECT_USER;
419 
420 	if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
421 		dev_err(&dev->dev, "failed to update image control: %i\n", rc);
422 		return rc;
423 	}
424 
425 	return 0;
426 }
427 
428 int cxl_alloc_one_irq(struct cxl *adapter)
429 {
430 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
431 
432 	return pnv_cxl_alloc_hwirqs(dev, 1);
433 }
434 
435 void cxl_release_one_irq(struct cxl *adapter, int hwirq)
436 {
437 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
438 
439 	return pnv_cxl_release_hwirqs(dev, hwirq, 1);
440 }
441 
442 int cxl_alloc_irq_ranges(struct cxl_irq_ranges *irqs, struct cxl *adapter, unsigned int num)
443 {
444 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
445 
446 	return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
447 }
448 
449 void cxl_release_irq_ranges(struct cxl_irq_ranges *irqs, struct cxl *adapter)
450 {
451 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
452 
453 	pnv_cxl_release_hwirq_ranges(irqs, dev);
454 }
455 
456 static int setup_cxl_bars(struct pci_dev *dev)
457 {
458 	/* Safety check in case we get backported to < 3.17 without M64 */
459 	if ((p1_base(dev) < 0x100000000ULL) ||
460 	    (p2_base(dev) < 0x100000000ULL)) {
461 		dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
462 		return -ENODEV;
463 	}
464 
465 	/*
466 	 * BAR 4/5 has a special meaning for CXL and must be programmed with a
467 	 * special value corresponding to the CXL protocol address range.
468 	 * For POWER 8 that means bits 48:49 must be set to 10
469 	 */
470 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
471 	pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
472 
473 	return 0;
474 }
475 
476 /* pciex node: ibm,opal-m64-window = <0x3d058 0x0 0x3d058 0x0 0x8 0x0>; */
477 static int switch_card_to_cxl(struct pci_dev *dev)
478 {
479 	int vsec;
480 	u8 val;
481 	int rc;
482 
483 	dev_info(&dev->dev, "switch card to CXL\n");
484 
485 	if (!(vsec = find_cxl_vsec(dev))) {
486 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
487 		return -ENODEV;
488 	}
489 
490 	if ((rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val))) {
491 		dev_err(&dev->dev, "failed to read current mode control: %i", rc);
492 		return rc;
493 	}
494 	val &= ~CXL_VSEC_PROTOCOL_MASK;
495 	val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
496 	if ((rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val))) {
497 		dev_err(&dev->dev, "failed to enable CXL protocol: %i", rc);
498 		return rc;
499 	}
500 	/*
501 	 * The CAIA spec (v0.12 11.6 Bi-modal Device Support) states
502 	 * we must wait 100ms after this mode switch before touching
503 	 * PCIe config space.
504 	 */
505 	msleep(100);
506 
507 	return 0;
508 }
509 
510 static int cxl_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
511 {
512 	u64 p1n_base, p2n_base, afu_desc;
513 	const u64 p1n_size = 0x100;
514 	const u64 p2n_size = 0x1000;
515 
516 	p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
517 	p2n_base = p2_base(dev) + (afu->slice * p2n_size);
518 	afu->psn_phys = p2_base(dev) + (adapter->ps_off + (afu->slice * adapter->ps_size));
519 	afu_desc = p2_base(dev) + adapter->afu_desc_off + (afu->slice * adapter->afu_desc_size);
520 
521 	if (!(afu->p1n_mmio = ioremap(p1n_base, p1n_size)))
522 		goto err;
523 	if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
524 		goto err1;
525 	if (afu_desc) {
526 		if (!(afu->afu_desc_mmio = ioremap(afu_desc, adapter->afu_desc_size)))
527 			goto err2;
528 	}
529 
530 	return 0;
531 err2:
532 	iounmap(afu->p2n_mmio);
533 err1:
534 	iounmap(afu->p1n_mmio);
535 err:
536 	dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
537 	return -ENOMEM;
538 }
539 
540 static void cxl_unmap_slice_regs(struct cxl_afu *afu)
541 {
542 	if (afu->p2n_mmio)
543 		iounmap(afu->p2n_mmio);
544 	if (afu->p1n_mmio)
545 		iounmap(afu->p1n_mmio);
546 }
547 
548 static void cxl_release_afu(struct device *dev)
549 {
550 	struct cxl_afu *afu = to_cxl_afu(dev);
551 
552 	pr_devel("cxl_release_afu\n");
553 
554 	kfree(afu);
555 }
556 
557 static struct cxl_afu *cxl_alloc_afu(struct cxl *adapter, int slice)
558 {
559 	struct cxl_afu *afu;
560 
561 	if (!(afu = kzalloc(sizeof(struct cxl_afu), GFP_KERNEL)))
562 		return NULL;
563 
564 	afu->adapter = adapter;
565 	afu->dev.parent = &adapter->dev;
566 	afu->dev.release = cxl_release_afu;
567 	afu->slice = slice;
568 	idr_init(&afu->contexts_idr);
569 	mutex_init(&afu->contexts_lock);
570 	spin_lock_init(&afu->afu_cntl_lock);
571 	mutex_init(&afu->spa_mutex);
572 
573 	afu->prefault_mode = CXL_PREFAULT_NONE;
574 	afu->irqs_max = afu->adapter->user_irqs;
575 
576 	return afu;
577 }
578 
579 /* Expects AFU struct to have recently been zeroed out */
580 static int cxl_read_afu_descriptor(struct cxl_afu *afu)
581 {
582 	u64 val;
583 
584 	val = AFUD_READ_INFO(afu);
585 	afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
586 	afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
587 	afu->crs_num = AFUD_NUM_CRS(val);
588 
589 	if (AFUD_AFU_DIRECTED(val))
590 		afu->modes_supported |= CXL_MODE_DIRECTED;
591 	if (AFUD_DEDICATED_PROCESS(val))
592 		afu->modes_supported |= CXL_MODE_DEDICATED;
593 	if (AFUD_TIME_SLICED(val))
594 		afu->modes_supported |= CXL_MODE_TIME_SLICED;
595 
596 	val = AFUD_READ_PPPSA(afu);
597 	afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
598 	afu->psa = AFUD_PPPSA_PSA(val);
599 	if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
600 		afu->pp_offset = AFUD_READ_PPPSA_OFF(afu);
601 
602 	val = AFUD_READ_CR(afu);
603 	afu->crs_len = AFUD_CR_LEN(val) * 256;
604 	afu->crs_offset = AFUD_READ_CR_OFF(afu);
605 
606 
607 	/* eb_len is in multiple of 4K */
608 	afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
609 	afu->eb_offset = AFUD_READ_EB_OFF(afu);
610 
611 	/* eb_off is 4K aligned so lower 12 bits are always zero */
612 	if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
613 		dev_warn(&afu->dev,
614 			 "Invalid AFU error buffer offset %Lx\n",
615 			 afu->eb_offset);
616 		dev_info(&afu->dev,
617 			 "Ignoring AFU error buffer in the descriptor\n");
618 		/* indicate that no afu buffer exists */
619 		afu->eb_len = 0;
620 	}
621 
622 	return 0;
623 }
624 
625 static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
626 {
627 	int i;
628 
629 	if (afu->psa && afu->adapter->ps_size <
630 			(afu->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
631 		dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
632 		return -ENODEV;
633 	}
634 
635 	if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
636 		dev_warn(&afu->dev, "AFU uses < PAGE_SIZE per-process PSA!");
637 
638 	for (i = 0; i < afu->crs_num; i++) {
639 		if ((cxl_afu_cr_read32(afu, i, 0) == 0)) {
640 			dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
641 			return -EINVAL;
642 		}
643 	}
644 
645 	return 0;
646 }
647 
648 static int sanitise_afu_regs(struct cxl_afu *afu)
649 {
650 	u64 reg;
651 
652 	/*
653 	 * Clear out any regs that contain either an IVTE or address or may be
654 	 * waiting on an acknowledgement to try to be a bit safer as we bring
655 	 * it online
656 	 */
657 	reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
658 	if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
659 		dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#.16llx\n", reg);
660 		if (__cxl_afu_reset(afu))
661 			return -EIO;
662 		if (cxl_afu_disable(afu))
663 			return -EIO;
664 		if (cxl_psl_purge(afu))
665 			return -EIO;
666 	}
667 	cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
668 	cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
669 	cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
670 	cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
671 	cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
672 	cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
673 	cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
674 	cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
675 	cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
676 	cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
677 	cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
678 	reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
679 	if (reg) {
680 		dev_warn(&afu->dev, "AFU had pending DSISR: %#.16llx\n", reg);
681 		if (reg & CXL_PSL_DSISR_TRANS)
682 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
683 		else
684 			cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
685 	}
686 	reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
687 	if (reg) {
688 		if (reg & ~0xffff)
689 			dev_warn(&afu->dev, "AFU had pending SERR: %#.16llx\n", reg);
690 		cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
691 	}
692 	reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
693 	if (reg) {
694 		dev_warn(&afu->dev, "AFU had pending error status: %#.16llx\n", reg);
695 		cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
696 	}
697 
698 	return 0;
699 }
700 
701 #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
702 /*
703  * afu_eb_read:
704  * Called from sysfs and reads the afu error info buffer. The h/w only supports
705  * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
706  * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
707  */
708 ssize_t cxl_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
709 				loff_t off, size_t count)
710 {
711 	loff_t aligned_start, aligned_end;
712 	size_t aligned_length;
713 	void *tbuf;
714 	const void __iomem *ebuf = afu->afu_desc_mmio + afu->eb_offset;
715 
716 	if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
717 		return 0;
718 
719 	/* calculate aligned read window */
720 	count = min((size_t)(afu->eb_len - off), count);
721 	aligned_start = round_down(off, 8);
722 	aligned_end = round_up(off + count, 8);
723 	aligned_length = aligned_end - aligned_start;
724 
725 	/* max we can copy in one read is PAGE_SIZE */
726 	if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
727 		aligned_length = ERR_BUFF_MAX_COPY_SIZE;
728 		count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
729 	}
730 
731 	/* use bounce buffer for copy */
732 	tbuf = (void *)__get_free_page(GFP_TEMPORARY);
733 	if (!tbuf)
734 		return -ENOMEM;
735 
736 	/* perform aligned read from the mmio region */
737 	memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
738 	memcpy(buf, tbuf + (off & 0x7), count);
739 
740 	free_page((unsigned long)tbuf);
741 
742 	return count;
743 }
744 
745 static int cxl_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
746 {
747 	struct cxl_afu *afu;
748 	bool free = true;
749 	int rc;
750 
751 	if (!(afu = cxl_alloc_afu(adapter, slice)))
752 		return -ENOMEM;
753 
754 	if ((rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice)))
755 		goto err1;
756 
757 	if ((rc = cxl_map_slice_regs(afu, adapter, dev)))
758 		goto err1;
759 
760 	if ((rc = sanitise_afu_regs(afu)))
761 		goto err2;
762 
763 	/* We need to reset the AFU before we can read the AFU descriptor */
764 	if ((rc = __cxl_afu_reset(afu)))
765 		goto err2;
766 
767 	if (cxl_verbose)
768 		dump_afu_descriptor(afu);
769 
770 	if ((rc = cxl_read_afu_descriptor(afu)))
771 		goto err2;
772 
773 	if ((rc = cxl_afu_descriptor_looks_ok(afu)))
774 		goto err2;
775 
776 	if ((rc = init_implementation_afu_regs(afu)))
777 		goto err2;
778 
779 	if ((rc = cxl_register_serr_irq(afu)))
780 		goto err2;
781 
782 	if ((rc = cxl_register_psl_irq(afu)))
783 		goto err3;
784 
785 	/* Don't care if this fails */
786 	cxl_debugfs_afu_add(afu);
787 
788 	/*
789 	 * After we call this function we must not free the afu directly, even
790 	 * if it returns an error!
791 	 */
792 	if ((rc = cxl_register_afu(afu)))
793 		goto err_put1;
794 
795 	if ((rc = cxl_sysfs_afu_add(afu)))
796 		goto err_put1;
797 
798 
799 	if ((rc = cxl_afu_select_best_mode(afu)))
800 		goto err_put2;
801 
802 	adapter->afu[afu->slice] = afu;
803 
804 	if ((rc = cxl_pci_vphb_add(afu)))
805 		dev_info(&afu->dev, "Can't register vPHB\n");
806 
807 	return 0;
808 
809 err_put2:
810 	cxl_sysfs_afu_remove(afu);
811 err_put1:
812 	device_unregister(&afu->dev);
813 	free = false;
814 	cxl_debugfs_afu_remove(afu);
815 	cxl_release_psl_irq(afu);
816 err3:
817 	cxl_release_serr_irq(afu);
818 err2:
819 	cxl_unmap_slice_regs(afu);
820 err1:
821 	if (free)
822 		kfree(afu);
823 	return rc;
824 }
825 
826 static void cxl_remove_afu(struct cxl_afu *afu)
827 {
828 	pr_devel("cxl_remove_afu\n");
829 
830 	if (!afu)
831 		return;
832 
833 	cxl_sysfs_afu_remove(afu);
834 	cxl_debugfs_afu_remove(afu);
835 
836 	spin_lock(&afu->adapter->afu_list_lock);
837 	afu->adapter->afu[afu->slice] = NULL;
838 	spin_unlock(&afu->adapter->afu_list_lock);
839 
840 	cxl_context_detach_all(afu);
841 	cxl_afu_deactivate_mode(afu);
842 
843 	cxl_release_psl_irq(afu);
844 	cxl_release_serr_irq(afu);
845 	cxl_unmap_slice_regs(afu);
846 
847 	device_unregister(&afu->dev);
848 }
849 
850 int cxl_reset(struct cxl *adapter)
851 {
852 	struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
853 	int rc;
854 	int i;
855 	u32 val;
856 
857 	dev_info(&dev->dev, "CXL reset\n");
858 
859 	for (i = 0; i < adapter->slices; i++) {
860 		cxl_pci_vphb_remove(adapter->afu[i]);
861 		cxl_remove_afu(adapter->afu[i]);
862 	}
863 
864 	/* pcie_warm_reset requests a fundamental pci reset which includes a
865 	 * PERST assert/deassert.  PERST triggers a loading of the image
866 	 * if "user" or "factory" is selected in sysfs */
867 	if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
868 		dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
869 		return rc;
870 	}
871 
872 	/* the PERST done above fences the PHB.  So, reset depends on EEH
873 	 * to unbind the driver, tell Sapphire to reinit the PHB, and rebind
874 	 * the driver.  Do an mmio read explictly to ensure EEH notices the
875 	 * fenced PHB.  Retry for a few seconds before giving up. */
876 	i = 0;
877 	while (((val = mmio_read32be(adapter->p1_mmio)) != 0xffffffff) &&
878 		(i < 5)) {
879 		msleep(500);
880 		i++;
881 	}
882 
883 	if (val != 0xffffffff)
884 		dev_err(&dev->dev, "cxl: PERST failed to trigger EEH\n");
885 
886 	return rc;
887 }
888 
889 static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
890 {
891 	if (pci_request_region(dev, 2, "priv 2 regs"))
892 		goto err1;
893 	if (pci_request_region(dev, 0, "priv 1 regs"))
894 		goto err2;
895 
896 	pr_devel("cxl_map_adapter_regs: p1: %#.16llx %#llx, p2: %#.16llx %#llx",
897 			p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
898 
899 	if (!(adapter->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
900 		goto err3;
901 
902 	if (!(adapter->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
903 		goto err4;
904 
905 	return 0;
906 
907 err4:
908 	iounmap(adapter->p1_mmio);
909 	adapter->p1_mmio = NULL;
910 err3:
911 	pci_release_region(dev, 0);
912 err2:
913 	pci_release_region(dev, 2);
914 err1:
915 	return -ENOMEM;
916 }
917 
918 static void cxl_unmap_adapter_regs(struct cxl *adapter)
919 {
920 	if (adapter->p1_mmio)
921 		iounmap(adapter->p1_mmio);
922 	if (adapter->p2_mmio)
923 		iounmap(adapter->p2_mmio);
924 }
925 
926 static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
927 {
928 	int vsec;
929 	u32 afu_desc_off, afu_desc_size;
930 	u32 ps_off, ps_size;
931 	u16 vseclen;
932 	u8 image_state;
933 
934 	if (!(vsec = find_cxl_vsec(dev))) {
935 		dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
936 		return -ENODEV;
937 	}
938 
939 	CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
940 	if (vseclen < CXL_VSEC_MIN_SIZE) {
941 		dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
942 		return -EINVAL;
943 	}
944 
945 	CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
946 	CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
947 	CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
948 	CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
949 	CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
950 	CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
951 	adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
952 	adapter->perst_loads_image = true;
953 	adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
954 
955 	CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
956 	CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
957 	CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
958 	CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
959 	CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
960 
961 	/* Convert everything to bytes, because there is NO WAY I'd look at the
962 	 * code a month later and forget what units these are in ;-) */
963 	adapter->ps_off = ps_off * 64 * 1024;
964 	adapter->ps_size = ps_size * 64 * 1024;
965 	adapter->afu_desc_off = afu_desc_off * 64 * 1024;
966 	adapter->afu_desc_size = afu_desc_size *64 * 1024;
967 
968 	/* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
969 	adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
970 
971 	return 0;
972 }
973 
974 static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
975 {
976 	if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
977 		return -EBUSY;
978 
979 	if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
980 		dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
981 		return -EINVAL;
982 	}
983 
984 	if (!adapter->slices) {
985 		/* Once we support dynamic reprogramming we can use the card if
986 		 * it supports loadable AFUs */
987 		dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
988 		return -EINVAL;
989 	}
990 
991 	if (!adapter->afu_desc_off || !adapter->afu_desc_size) {
992 		dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
993 		return -EINVAL;
994 	}
995 
996 	if (adapter->ps_size > p2_size(dev) - adapter->ps_off) {
997 		dev_err(&dev->dev, "ABORTING: Problem state size larger than "
998 				   "available in BAR2: 0x%llx > 0x%llx\n",
999 			 adapter->ps_size, p2_size(dev) - adapter->ps_off);
1000 		return -EINVAL;
1001 	}
1002 
1003 	return 0;
1004 }
1005 
1006 static void cxl_release_adapter(struct device *dev)
1007 {
1008 	struct cxl *adapter = to_cxl_adapter(dev);
1009 
1010 	pr_devel("cxl_release_adapter\n");
1011 
1012 	kfree(adapter);
1013 }
1014 
1015 static struct cxl *cxl_alloc_adapter(struct pci_dev *dev)
1016 {
1017 	struct cxl *adapter;
1018 
1019 	if (!(adapter = kzalloc(sizeof(struct cxl), GFP_KERNEL)))
1020 		return NULL;
1021 
1022 	adapter->dev.parent = &dev->dev;
1023 	adapter->dev.release = cxl_release_adapter;
1024 	pci_set_drvdata(dev, adapter);
1025 	spin_lock_init(&adapter->afu_list_lock);
1026 
1027 	return adapter;
1028 }
1029 
1030 static int sanitise_adapter_regs(struct cxl *adapter)
1031 {
1032 	cxl_p1_write(adapter, CXL_PSL_ErrIVTE, 0x0000000000000000);
1033 	return cxl_tlb_slb_invalidate(adapter);
1034 }
1035 
1036 static struct cxl *cxl_init_adapter(struct pci_dev *dev)
1037 {
1038 	struct cxl *adapter;
1039 	bool free = true;
1040 	int rc;
1041 
1042 
1043 	if (!(adapter = cxl_alloc_adapter(dev)))
1044 		return ERR_PTR(-ENOMEM);
1045 
1046 	if ((rc = cxl_read_vsec(adapter, dev)))
1047 		goto err1;
1048 
1049 	if ((rc = cxl_vsec_looks_ok(adapter, dev)))
1050 		goto err1;
1051 
1052 	if ((rc = setup_cxl_bars(dev)))
1053 		goto err1;
1054 
1055 	if ((rc = switch_card_to_cxl(dev)))
1056 		goto err1;
1057 
1058 	if ((rc = cxl_alloc_adapter_nr(adapter)))
1059 		goto err1;
1060 
1061 	if ((rc = dev_set_name(&adapter->dev, "card%i", adapter->adapter_num)))
1062 		goto err2;
1063 
1064 	if ((rc = cxl_update_image_control(adapter)))
1065 		goto err2;
1066 
1067 	if ((rc = cxl_map_adapter_regs(adapter, dev)))
1068 		goto err2;
1069 
1070 	if ((rc = sanitise_adapter_regs(adapter)))
1071 		goto err2;
1072 
1073 	if ((rc = init_implementation_adapter_regs(adapter, dev)))
1074 		goto err3;
1075 
1076 	if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_CAPI)))
1077 		goto err3;
1078 
1079 	/* If recovery happened, the last step is to turn on snooping.
1080 	 * In the non-recovery case this has no effect */
1081 	if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON))) {
1082 		goto err3;
1083 	}
1084 
1085 	if ((rc = cxl_register_psl_err_irq(adapter)))
1086 		goto err3;
1087 
1088 	/* Don't care if this one fails: */
1089 	cxl_debugfs_adapter_add(adapter);
1090 
1091 	/*
1092 	 * After we call this function we must not free the adapter directly,
1093 	 * even if it returns an error!
1094 	 */
1095 	if ((rc = cxl_register_adapter(adapter)))
1096 		goto err_put1;
1097 
1098 	if ((rc = cxl_sysfs_adapter_add(adapter)))
1099 		goto err_put1;
1100 
1101 	return adapter;
1102 
1103 err_put1:
1104 	device_unregister(&adapter->dev);
1105 	free = false;
1106 	cxl_debugfs_adapter_remove(adapter);
1107 	cxl_release_psl_err_irq(adapter);
1108 err3:
1109 	cxl_unmap_adapter_regs(adapter);
1110 err2:
1111 	cxl_remove_adapter_nr(adapter);
1112 err1:
1113 	if (free)
1114 		kfree(adapter);
1115 	return ERR_PTR(rc);
1116 }
1117 
1118 static void cxl_remove_adapter(struct cxl *adapter)
1119 {
1120 	struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
1121 
1122 	pr_devel("cxl_release_adapter\n");
1123 
1124 	cxl_sysfs_adapter_remove(adapter);
1125 	cxl_debugfs_adapter_remove(adapter);
1126 	cxl_release_psl_err_irq(adapter);
1127 	cxl_unmap_adapter_regs(adapter);
1128 	cxl_remove_adapter_nr(adapter);
1129 
1130 	device_unregister(&adapter->dev);
1131 
1132 	pci_release_region(pdev, 0);
1133 	pci_release_region(pdev, 2);
1134 	pci_disable_device(pdev);
1135 }
1136 
1137 static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
1138 {
1139 	struct cxl *adapter;
1140 	int slice;
1141 	int rc;
1142 
1143 	pci_dev_get(dev);
1144 
1145 	if (cxl_verbose)
1146 		dump_cxl_config_space(dev);
1147 
1148 	if ((rc = pci_enable_device(dev))) {
1149 		dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
1150 		return rc;
1151 	}
1152 
1153 	adapter = cxl_init_adapter(dev);
1154 	if (IS_ERR(adapter)) {
1155 		dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
1156 		pci_disable_device(dev);
1157 		return PTR_ERR(adapter);
1158 	}
1159 
1160 	for (slice = 0; slice < adapter->slices; slice++) {
1161 		if ((rc = cxl_init_afu(adapter, slice, dev)))
1162 			dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
1163 	}
1164 
1165 	return 0;
1166 }
1167 
1168 static void cxl_remove(struct pci_dev *dev)
1169 {
1170 	struct cxl *adapter = pci_get_drvdata(dev);
1171 	struct cxl_afu *afu;
1172 	int i;
1173 
1174 	/*
1175 	 * Lock to prevent someone grabbing a ref through the adapter list as
1176 	 * we are removing it
1177 	 */
1178 	for (i = 0; i < adapter->slices; i++) {
1179 		afu = adapter->afu[i];
1180 		cxl_pci_vphb_remove(afu);
1181 		cxl_remove_afu(afu);
1182 	}
1183 	cxl_remove_adapter(adapter);
1184 }
1185 
1186 struct pci_driver cxl_pci_driver = {
1187 	.name = "cxl-pci",
1188 	.id_table = cxl_pci_tbl,
1189 	.probe = cxl_probe,
1190 	.remove = cxl_remove,
1191 	.shutdown = cxl_remove,
1192 };
1193