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