| 1 |
/* |
| 2 |
* This file is subject to the terms and conditions of the GNU General Public |
| 3 |
* License. See the file "COPYING" in the main directory of this archive |
| 4 |
* for more details. |
| 5 |
* |
| 6 |
* Copyright (C) 2005-2009 Cavium Networks |
| 7 |
*/ |
| 8 |
#include <linux/kernel.h> |
| 9 |
#include <linux/init.h> |
| 10 |
#include <linux/pci.h> |
| 11 |
#include <linux/interrupt.h> |
| 12 |
#include <linux/time.h> |
| 13 |
#include <linux/delay.h> |
| 14 |
#include <linux/platform_device.h> |
| 15 |
#include <linux/swiotlb.h> |
| 16 |
|
| 17 |
#include <asm/time.h> |
| 18 |
|
| 19 |
#include <asm/octeon/octeon.h> |
| 20 |
#include <asm/octeon/cvmx-npi-defs.h> |
| 21 |
#include <asm/octeon/cvmx-pci-defs.h> |
| 22 |
#include <asm/octeon/pci-octeon.h> |
| 23 |
|
| 24 |
#define USE_OCTEON_INTERNAL_ARBITER |
| 25 |
|
| 26 |
/* |
| 27 |
* Octeon's PCI controller uses did=3, subdid=2 for PCI IO |
| 28 |
* addresses. Use PCI endian swapping 1 so no address swapping is |
| 29 |
* necessary. The Linux io routines will endian swap the data. |
| 30 |
*/ |
| 31 |
#define OCTEON_PCI_IOSPACE_BASE 0x80011a0400000000ull |
| 32 |
#define OCTEON_PCI_IOSPACE_SIZE (1ull<<32) |
| 33 |
|
| 34 |
/* Octeon't PCI controller uses did=3, subdid=3 for PCI memory. */ |
| 35 |
#define OCTEON_PCI_MEMSPACE_OFFSET (0x00011b0000000000ull) |
| 36 |
|
| 37 |
u64 octeon_bar1_pci_phys; |
| 38 |
|
| 39 |
/** |
| 40 |
* This is the bit decoding used for the Octeon PCI controller addresses |
| 41 |
*/ |
| 42 |
union octeon_pci_address { |
| 43 |
uint64_t u64; |
| 44 |
struct { |
| 45 |
uint64_t upper:2; |
| 46 |
uint64_t reserved:13; |
| 47 |
uint64_t io:1; |
| 48 |
uint64_t did:5; |
| 49 |
uint64_t subdid:3; |
| 50 |
uint64_t reserved2:4; |
| 51 |
uint64_t endian_swap:2; |
| 52 |
uint64_t reserved3:10; |
| 53 |
uint64_t bus:8; |
| 54 |
uint64_t dev:5; |
| 55 |
uint64_t func:3; |
| 56 |
uint64_t reg:8; |
| 57 |
} s; |
| 58 |
}; |
| 59 |
|
| 60 |
int (*octeon_pcibios_map_irq)(const struct pci_dev *dev, u8 slot, u8 pin); |
| 61 |
enum octeon_dma_bar_type octeon_dma_bar_type = OCTEON_DMA_BAR_TYPE_INVALID; |
| 62 |
|
| 63 |
/** |
| 64 |
* Map a PCI device to the appropriate interrupt line |
| 65 |
* |
| 66 |
* @dev: The Linux PCI device structure for the device to map |
| 67 |
* @slot: The slot number for this device on __BUS 0__. Linux |
| 68 |
* enumerates through all the bridges and figures out the |
| 69 |
* slot on Bus 0 where this device eventually hooks to. |
| 70 |
* @pin: The PCI interrupt pin read from the device, then swizzled |
| 71 |
* as it goes through each bridge. |
| 72 |
* Returns Interrupt number for the device |
| 73 |
*/ |
| 74 |
int pcibios_map_irq(const struct pci_dev *dev, u8 slot, u8 pin) |
| 75 |
{ |
| 76 |
if (octeon_pcibios_map_irq) |
| 77 |
return octeon_pcibios_map_irq(dev, slot, pin); |
| 78 |
else |
| 79 |
panic("octeon_pcibios_map_irq not set."); |
| 80 |
} |
| 81 |
|
| 82 |
|
| 83 |
/* |
| 84 |
* Called to perform platform specific PCI setup |
| 85 |
*/ |
| 86 |
int pcibios_plat_dev_init(struct pci_dev *dev) |
| 87 |
{ |
| 88 |
uint16_t config; |
| 89 |
uint32_t dconfig; |
| 90 |
int pos; |
| 91 |
/* |
| 92 |
* Force the Cache line setting to 64 bytes. The standard |
| 93 |
* Linux bus scan doesn't seem to set it. Octeon really has |
| 94 |
* 128 byte lines, but Intel bridges get really upset if you |
| 95 |
* try and set values above 64 bytes. Value is specified in |
| 96 |
* 32bit words. |
| 97 |
*/ |
| 98 |
pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, 64 / 4); |
| 99 |
/* Set latency timers for all devices */ |
| 100 |
pci_write_config_byte(dev, PCI_LATENCY_TIMER, 64); |
| 101 |
|
| 102 |
/* Enable reporting System errors and parity errors on all devices */ |
| 103 |
/* Enable parity checking and error reporting */ |
| 104 |
pci_read_config_word(dev, PCI_COMMAND, &config); |
| 105 |
config |= PCI_COMMAND_PARITY | PCI_COMMAND_SERR; |
| 106 |
pci_write_config_word(dev, PCI_COMMAND, config); |
| 107 |
|
| 108 |
if (dev->subordinate) { |
| 109 |
/* Set latency timers on sub bridges */ |
| 110 |
pci_write_config_byte(dev, PCI_SEC_LATENCY_TIMER, 64); |
| 111 |
/* More bridge error detection */ |
| 112 |
pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &config); |
| 113 |
config |= PCI_BRIDGE_CTL_PARITY | PCI_BRIDGE_CTL_SERR; |
| 114 |
pci_write_config_word(dev, PCI_BRIDGE_CONTROL, config); |
| 115 |
} |
| 116 |
|
| 117 |
/* Enable the PCIe normal error reporting */ |
| 118 |
config = PCI_EXP_DEVCTL_CERE; /* Correctable Error Reporting */ |
| 119 |
config |= PCI_EXP_DEVCTL_NFERE; /* Non-Fatal Error Reporting */ |
| 120 |
config |= PCI_EXP_DEVCTL_FERE; /* Fatal Error Reporting */ |
| 121 |
config |= PCI_EXP_DEVCTL_URRE; /* Unsupported Request */ |
| 122 |
pcie_capability_set_word(dev, PCI_EXP_DEVCTL, config); |
| 123 |
|
| 124 |
/* Find the Advanced Error Reporting capability */ |
| 125 |
pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR); |
| 126 |
if (pos) { |
| 127 |
/* Clear Uncorrectable Error Status */ |
| 128 |
pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, |
| 129 |
&dconfig); |
| 130 |
pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, |
| 131 |
dconfig); |
| 132 |
/* Enable reporting of all uncorrectable errors */ |
| 133 |
/* Uncorrectable Error Mask - turned on bits disable errors */ |
| 134 |
pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, 0); |
| 135 |
/* |
| 136 |
* Leave severity at HW default. This only controls if |
| 137 |
* errors are reported as uncorrectable or |
| 138 |
* correctable, not if the error is reported. |
| 139 |
*/ |
| 140 |
/* PCI_ERR_UNCOR_SEVER - Uncorrectable Error Severity */ |
| 141 |
/* Clear Correctable Error Status */ |
| 142 |
pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &dconfig); |
| 143 |
pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, dconfig); |
| 144 |
/* Enable reporting of all correctable errors */ |
| 145 |
/* Correctable Error Mask - turned on bits disable errors */ |
| 146 |
pci_write_config_dword(dev, pos + PCI_ERR_COR_MASK, 0); |
| 147 |
/* Advanced Error Capabilities */ |
| 148 |
pci_read_config_dword(dev, pos + PCI_ERR_CAP, &dconfig); |
| 149 |
/* ECRC Generation Enable */ |
| 150 |
if (config & PCI_ERR_CAP_ECRC_GENC) |
| 151 |
config |= PCI_ERR_CAP_ECRC_GENE; |
| 152 |
/* ECRC Check Enable */ |
| 153 |
if (config & PCI_ERR_CAP_ECRC_CHKC) |
| 154 |
config |= PCI_ERR_CAP_ECRC_CHKE; |
| 155 |
pci_write_config_dword(dev, pos + PCI_ERR_CAP, dconfig); |
| 156 |
/* PCI_ERR_HEADER_LOG - Header Log Register (16 bytes) */ |
| 157 |
/* Report all errors to the root complex */ |
| 158 |
pci_write_config_dword(dev, pos + PCI_ERR_ROOT_COMMAND, |
| 159 |
PCI_ERR_ROOT_CMD_COR_EN | |
| 160 |
PCI_ERR_ROOT_CMD_NONFATAL_EN | |
| 161 |
PCI_ERR_ROOT_CMD_FATAL_EN); |
| 162 |
/* Clear the Root status register */ |
| 163 |
pci_read_config_dword(dev, pos + PCI_ERR_ROOT_STATUS, &dconfig); |
| 164 |
pci_write_config_dword(dev, pos + PCI_ERR_ROOT_STATUS, dconfig); |
| 165 |
} |
| 166 |
|
| 167 |
return 0; |
| 168 |
} |
| 169 |
|
| 170 |
/** |
| 171 |
* Return the mapping of PCI device number to IRQ line. Each |
| 172 |
* character in the return string represents the interrupt |
| 173 |
* line for the device at that position. Device 1 maps to the |
| 174 |
* first character, etc. The characters A-D are used for PCI |
| 175 |
* interrupts. |
| 176 |
* |
| 177 |
* Returns PCI interrupt mapping |
| 178 |
*/ |
| 179 |
const char *octeon_get_pci_interrupts(void) |
| 180 |
{ |
| 181 |
/* |
| 182 |
* Returning an empty string causes the interrupts to be |
| 183 |
* routed based on the PCI specification. From the PCI spec: |
| 184 |
* |
| 185 |
* INTA# of Device Number 0 is connected to IRQW on the system |
| 186 |
* board. (Device Number has no significance regarding being |
| 187 |
* located on the system board or in a connector.) INTA# of |
| 188 |
* Device Number 1 is connected to IRQX on the system |
| 189 |
* board. INTA# of Device Number 2 is connected to IRQY on the |
| 190 |
* system board. INTA# of Device Number 3 is connected to IRQZ |
| 191 |
* on the system board. The table below describes how each |
| 192 |
* agent's INTx# lines are connected to the system board |
| 193 |
* interrupt lines. The following equation can be used to |
| 194 |
* determine to which INTx# signal on the system board a given |
| 195 |
* device's INTx# line(s) is connected. |
| 196 |
* |
| 197 |
* MB = (D + I) MOD 4 MB = System board Interrupt (IRQW = 0, |
| 198 |
* IRQX = 1, IRQY = 2, and IRQZ = 3) D = Device Number I = |
| 199 |
* Interrupt Number (INTA# = 0, INTB# = 1, INTC# = 2, and |
| 200 |
* INTD# = 3) |
| 201 |
*/ |
| 202 |
if (of_machine_is_compatible("dlink,dsr-500n")) |
| 203 |
return "CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC"; |
| 204 |
switch (octeon_bootinfo->board_type) { |
| 205 |
case CVMX_BOARD_TYPE_NAO38: |
| 206 |
/* This is really the NAC38 */ |
| 207 |
return "AAAAADABAAAAAAAAAAAAAAAAAAAAAAAA"; |
| 208 |
case CVMX_BOARD_TYPE_EBH3100: |
| 209 |
case CVMX_BOARD_TYPE_CN3010_EVB_HS5: |
| 210 |
case CVMX_BOARD_TYPE_CN3005_EVB_HS5: |
| 211 |
return "AAABAAAAAAAAAAAAAAAAAAAAAAAAAAAA"; |
| 212 |
case CVMX_BOARD_TYPE_BBGW_REF: |
| 213 |
return "AABCD"; |
| 214 |
case CVMX_BOARD_TYPE_CUST_DSR1000N: |
| 215 |
return "CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC"; |
| 216 |
case CVMX_BOARD_TYPE_THUNDER: |
| 217 |
case CVMX_BOARD_TYPE_EBH3000: |
| 218 |
default: |
| 219 |
return ""; |
| 220 |
} |
| 221 |
} |
| 222 |
|
| 223 |
/** |
| 224 |
* Map a PCI device to the appropriate interrupt line |
| 225 |
* |
| 226 |
* @dev: The Linux PCI device structure for the device to map |
| 227 |
* @slot: The slot number for this device on __BUS 0__. Linux |
| 228 |
* enumerates through all the bridges and figures out the |
| 229 |
* slot on Bus 0 where this device eventually hooks to. |
| 230 |
* @pin: The PCI interrupt pin read from the device, then swizzled |
| 231 |
* as it goes through each bridge. |
| 232 |
* Returns Interrupt number for the device |
| 233 |
*/ |
| 234 |
int __init octeon_pci_pcibios_map_irq(const struct pci_dev *dev, |
| 235 |
u8 slot, u8 pin) |
| 236 |
{ |
| 237 |
int irq_num; |
| 238 |
const char *interrupts; |
| 239 |
int dev_num; |
| 240 |
|
| 241 |
/* Get the board specific interrupt mapping */ |
| 242 |
interrupts = octeon_get_pci_interrupts(); |
| 243 |
|
| 244 |
dev_num = dev->devfn >> 3; |
| 245 |
if (dev_num < strlen(interrupts)) |
| 246 |
irq_num = ((interrupts[dev_num] - 'A' + pin - 1) & 3) + |
| 247 |
OCTEON_IRQ_PCI_INT0; |
| 248 |
else |
| 249 |
irq_num = ((slot + pin - 3) & 3) + OCTEON_IRQ_PCI_INT0; |
| 250 |
return irq_num; |
| 251 |
} |
| 252 |
|
| 253 |
|
| 254 |
/* |
| 255 |
* Read a value from configuration space |
| 256 |
*/ |
| 257 |
static int octeon_read_config(struct pci_bus *bus, unsigned int devfn, |
| 258 |
int reg, int size, u32 *val) |
| 259 |
{ |
| 260 |
union octeon_pci_address pci_addr; |
| 261 |
|
| 262 |
pci_addr.u64 = 0; |
| 263 |
pci_addr.s.upper = 2; |
| 264 |
pci_addr.s.io = 1; |
| 265 |
pci_addr.s.did = 3; |
| 266 |
pci_addr.s.subdid = 1; |
| 267 |
pci_addr.s.endian_swap = 1; |
| 268 |
pci_addr.s.bus = bus->number; |
| 269 |
pci_addr.s.dev = devfn >> 3; |
| 270 |
pci_addr.s.func = devfn & 0x7; |
| 271 |
pci_addr.s.reg = reg; |
| 272 |
|
| 273 |
switch (size) { |
| 274 |
case 4: |
| 275 |
*val = le32_to_cpu(cvmx_read64_uint32(pci_addr.u64)); |
| 276 |
return PCIBIOS_SUCCESSFUL; |
| 277 |
case 2: |
| 278 |
*val = le16_to_cpu(cvmx_read64_uint16(pci_addr.u64)); |
| 279 |
return PCIBIOS_SUCCESSFUL; |
| 280 |
case 1: |
| 281 |
*val = cvmx_read64_uint8(pci_addr.u64); |
| 282 |
return PCIBIOS_SUCCESSFUL; |
| 283 |
} |
| 284 |
return PCIBIOS_FUNC_NOT_SUPPORTED; |
| 285 |
} |
| 286 |
|
| 287 |
|
| 288 |
/* |
| 289 |
* Write a value to PCI configuration space |
| 290 |
*/ |
| 291 |
static int octeon_write_config(struct pci_bus *bus, unsigned int devfn, |
| 292 |
int reg, int size, u32 val) |
| 293 |
{ |
| 294 |
union octeon_pci_address pci_addr; |
| 295 |
|
| 296 |
pci_addr.u64 = 0; |
| 297 |
pci_addr.s.upper = 2; |
| 298 |
pci_addr.s.io = 1; |
| 299 |
pci_addr.s.did = 3; |
| 300 |
pci_addr.s.subdid = 1; |
| 301 |
pci_addr.s.endian_swap = 1; |
| 302 |
pci_addr.s.bus = bus->number; |
| 303 |
pci_addr.s.dev = devfn >> 3; |
| 304 |
pci_addr.s.func = devfn & 0x7; |
| 305 |
pci_addr.s.reg = reg; |
| 306 |
|
| 307 |
switch (size) { |
| 308 |
case 4: |
| 309 |
cvmx_write64_uint32(pci_addr.u64, cpu_to_le32(val)); |
| 310 |
return PCIBIOS_SUCCESSFUL; |
| 311 |
case 2: |
| 312 |
cvmx_write64_uint16(pci_addr.u64, cpu_to_le16(val)); |
| 313 |
return PCIBIOS_SUCCESSFUL; |
| 314 |
case 1: |
| 315 |
cvmx_write64_uint8(pci_addr.u64, val); |
| 316 |
return PCIBIOS_SUCCESSFUL; |
| 317 |
} |
| 318 |
return PCIBIOS_FUNC_NOT_SUPPORTED; |
| 319 |
} |
| 320 |
|
| 321 |
|
| 322 |
static struct pci_ops octeon_pci_ops = { |
| 323 |
.read = octeon_read_config, |
| 324 |
.write = octeon_write_config, |
| 325 |
}; |
| 326 |
|
| 327 |
static struct resource octeon_pci_mem_resource = { |
| 328 |
.start = 0, |
| 329 |
.end = 0, |
| 330 |
.name = "Octeon PCI MEM", |
| 331 |
.flags = IORESOURCE_MEM, |
| 332 |
}; |
| 333 |
|
| 334 |
/* |
| 335 |
* PCI ports must be above 16KB so the ISA bus filtering in the PCI-X to PCI |
| 336 |
* bridge |
| 337 |
*/ |
| 338 |
static struct resource octeon_pci_io_resource = { |
| 339 |
.start = 0x4000, |
| 340 |
.end = OCTEON_PCI_IOSPACE_SIZE - 1, |
| 341 |
.name = "Octeon PCI IO", |
| 342 |
.flags = IORESOURCE_IO, |
| 343 |
}; |
| 344 |
|
| 345 |
static struct pci_controller octeon_pci_controller = { |
| 346 |
.pci_ops = &octeon_pci_ops, |
| 347 |
.mem_resource = &octeon_pci_mem_resource, |
| 348 |
.mem_offset = OCTEON_PCI_MEMSPACE_OFFSET, |
| 349 |
.io_resource = &octeon_pci_io_resource, |
| 350 |
.io_offset = 0, |
| 351 |
.io_map_base = OCTEON_PCI_IOSPACE_BASE, |
| 352 |
}; |
| 353 |
|
| 354 |
|
| 355 |
/* |
| 356 |
* Low level initialize the Octeon PCI controller |
| 357 |
*/ |
| 358 |
static void octeon_pci_initialize(void) |
| 359 |
{ |
| 360 |
union cvmx_pci_cfg01 cfg01; |
| 361 |
union cvmx_npi_ctl_status ctl_status; |
| 362 |
union cvmx_pci_ctl_status_2 ctl_status_2; |
| 363 |
union cvmx_pci_cfg19 cfg19; |
| 364 |
union cvmx_pci_cfg16 cfg16; |
| 365 |
union cvmx_pci_cfg22 cfg22; |
| 366 |
union cvmx_pci_cfg56 cfg56; |
| 367 |
|
| 368 |
/* Reset the PCI Bus */ |
| 369 |
cvmx_write_csr(CVMX_CIU_SOFT_PRST, 0x1); |
| 370 |
cvmx_read_csr(CVMX_CIU_SOFT_PRST); |
| 371 |
|
| 372 |
udelay(2000); /* Hold PCI reset for 2 ms */ |
| 373 |
|
| 374 |
ctl_status.u64 = 0; /* cvmx_read_csr(CVMX_NPI_CTL_STATUS); */ |
| 375 |
ctl_status.s.max_word = 1; |
| 376 |
ctl_status.s.timer = 1; |
| 377 |
cvmx_write_csr(CVMX_NPI_CTL_STATUS, ctl_status.u64); |
| 378 |
|
| 379 |
/* Deassert PCI reset and advertise PCX Host Mode Device Capability |
| 380 |
(64b) */ |
| 381 |
cvmx_write_csr(CVMX_CIU_SOFT_PRST, 0x4); |
| 382 |
cvmx_read_csr(CVMX_CIU_SOFT_PRST); |
| 383 |
|
| 384 |
udelay(2000); /* Wait 2 ms after deasserting PCI reset */ |
| 385 |
|
| 386 |
ctl_status_2.u32 = 0; |
| 387 |
ctl_status_2.s.tsr_hwm = 1; /* Initializes to 0. Must be set |
| 388 |
before any PCI reads. */ |
| 389 |
ctl_status_2.s.bar2pres = 1; /* Enable BAR2 */ |
| 390 |
ctl_status_2.s.bar2_enb = 1; |
| 391 |
ctl_status_2.s.bar2_cax = 1; /* Don't use L2 */ |
| 392 |
ctl_status_2.s.bar2_esx = 1; |
| 393 |
ctl_status_2.s.pmo_amod = 1; /* Round robin priority */ |
| 394 |
if (octeon_dma_bar_type == OCTEON_DMA_BAR_TYPE_BIG) { |
| 395 |
/* BAR1 hole */ |
| 396 |
ctl_status_2.s.bb1_hole = OCTEON_PCI_BAR1_HOLE_BITS; |
| 397 |
ctl_status_2.s.bb1_siz = 1; /* BAR1 is 2GB */ |
| 398 |
ctl_status_2.s.bb_ca = 1; /* Don't use L2 with big bars */ |
| 399 |
ctl_status_2.s.bb_es = 1; /* Big bar in byte swap mode */ |
| 400 |
ctl_status_2.s.bb1 = 1; /* BAR1 is big */ |
| 401 |
ctl_status_2.s.bb0 = 1; /* BAR0 is big */ |
| 402 |
} |
| 403 |
|
| 404 |
octeon_npi_write32(CVMX_NPI_PCI_CTL_STATUS_2, ctl_status_2.u32); |
| 405 |
udelay(2000); /* Wait 2 ms before doing PCI reads */ |
| 406 |
|
| 407 |
ctl_status_2.u32 = octeon_npi_read32(CVMX_NPI_PCI_CTL_STATUS_2); |
| 408 |
pr_notice("PCI Status: %s %s-bit\n", |
| 409 |
ctl_status_2.s.ap_pcix ? "PCI-X" : "PCI", |
| 410 |
ctl_status_2.s.ap_64ad ? "64" : "32"); |
| 411 |
|
| 412 |
if (OCTEON_IS_MODEL(OCTEON_CN58XX) || OCTEON_IS_MODEL(OCTEON_CN50XX)) { |
| 413 |
union cvmx_pci_cnt_reg cnt_reg_start; |
| 414 |
union cvmx_pci_cnt_reg cnt_reg_end; |
| 415 |
unsigned long cycles, pci_clock; |
| 416 |
|
| 417 |
cnt_reg_start.u64 = cvmx_read_csr(CVMX_NPI_PCI_CNT_REG); |
| 418 |
cycles = read_c0_cvmcount(); |
| 419 |
udelay(1000); |
| 420 |
cnt_reg_end.u64 = cvmx_read_csr(CVMX_NPI_PCI_CNT_REG); |
| 421 |
cycles = read_c0_cvmcount() - cycles; |
| 422 |
pci_clock = (cnt_reg_end.s.pcicnt - cnt_reg_start.s.pcicnt) / |
| 423 |
(cycles / (mips_hpt_frequency / 1000000)); |
| 424 |
pr_notice("PCI Clock: %lu MHz\n", pci_clock); |
| 425 |
} |
| 426 |
|
| 427 |
/* |
| 428 |
* TDOMC must be set to one in PCI mode. TDOMC should be set to 4 |
| 429 |
* in PCI-X mode to allow four outstanding splits. Otherwise, |
| 430 |
* should not change from its reset value. Don't write PCI_CFG19 |
| 431 |
* in PCI mode (0x82000001 reset value), write it to 0x82000004 |
| 432 |
* after PCI-X mode is known. MRBCI,MDWE,MDRE -> must be zero. |
| 433 |
* MRBCM -> must be one. |
| 434 |
*/ |
| 435 |
if (ctl_status_2.s.ap_pcix) { |
| 436 |
cfg19.u32 = 0; |
| 437 |
/* |
| 438 |
* Target Delayed/Split request outstanding maximum |
| 439 |
* count. [1..31] and 0=32. NOTE: If the user |
| 440 |
* programs these bits beyond the Designed Maximum |
| 441 |
* outstanding count, then the designed maximum table |
| 442 |
* depth will be used instead. No additional |
| 443 |
* Deferred/Split transactions will be accepted if |
| 444 |
* this outstanding maximum count is |
| 445 |
* reached. Furthermore, no additional deferred/split |
| 446 |
* transactions will be accepted if the I/O delay/ I/O |
| 447 |
* Split Request outstanding maximum is reached. |
| 448 |
*/ |
| 449 |
cfg19.s.tdomc = 4; |
| 450 |
/* |
| 451 |
* Master Deferred Read Request Outstanding Max Count |
| 452 |
* (PCI only). CR4C[26:24] Max SAC cycles MAX DAC |
| 453 |
* cycles 000 8 4 001 1 0 010 2 1 011 3 1 100 4 2 101 |
| 454 |
* 5 2 110 6 3 111 7 3 For example, if these bits are |
| 455 |
* programmed to 100, the core can support 2 DAC |
| 456 |
* cycles, 4 SAC cycles or a combination of 1 DAC and |
| 457 |
* 2 SAC cycles. NOTE: For the PCI-X maximum |
| 458 |
* outstanding split transactions, refer to |
| 459 |
* CRE0[22:20]. |
| 460 |
*/ |
| 461 |
cfg19.s.mdrrmc = 2; |
| 462 |
/* |
| 463 |
* Master Request (Memory Read) Byte Count/Byte Enable |
| 464 |
* select. 0 = Byte Enables valid. In PCI mode, a |
| 465 |
* burst transaction cannot be performed using Memory |
| 466 |
* Read command=4?h6. 1 = DWORD Byte Count valid |
| 467 |
* (default). In PCI Mode, the memory read byte |
| 468 |
* enables are automatically generated by the |
| 469 |
* core. Note: N3 Master Request transaction sizes are |
| 470 |
* always determined through the |
| 471 |
* am_attr[<35:32>|<7:0>] field. |
| 472 |
*/ |
| 473 |
cfg19.s.mrbcm = 1; |
| 474 |
octeon_npi_write32(CVMX_NPI_PCI_CFG19, cfg19.u32); |
| 475 |
} |
| 476 |
|
| 477 |
|
| 478 |
cfg01.u32 = 0; |
| 479 |
cfg01.s.msae = 1; /* Memory Space Access Enable */ |
| 480 |
cfg01.s.me = 1; /* Master Enable */ |
| 481 |
cfg01.s.pee = 1; /* PERR# Enable */ |
| 482 |
cfg01.s.see = 1; /* System Error Enable */ |
| 483 |
cfg01.s.fbbe = 1; /* Fast Back to Back Transaction Enable */ |
| 484 |
|
| 485 |
octeon_npi_write32(CVMX_NPI_PCI_CFG01, cfg01.u32); |
| 486 |
|
| 487 |
#ifdef USE_OCTEON_INTERNAL_ARBITER |
| 488 |
/* |
| 489 |
* When OCTEON is a PCI host, most systems will use OCTEON's |
| 490 |
* internal arbiter, so must enable it before any PCI/PCI-X |
| 491 |
* traffic can occur. |
| 492 |
*/ |
| 493 |
{ |
| 494 |
union cvmx_npi_pci_int_arb_cfg pci_int_arb_cfg; |
| 495 |
|
| 496 |
pci_int_arb_cfg.u64 = 0; |
| 497 |
pci_int_arb_cfg.s.en = 1; /* Internal arbiter enable */ |
| 498 |
cvmx_write_csr(CVMX_NPI_PCI_INT_ARB_CFG, pci_int_arb_cfg.u64); |
| 499 |
} |
| 500 |
#endif /* USE_OCTEON_INTERNAL_ARBITER */ |
| 501 |
|
| 502 |
/* |
| 503 |
* Preferably written to 1 to set MLTD. [RDSATI,TRTAE, |
| 504 |
* TWTAE,TMAE,DPPMR -> must be zero. TILT -> must not be set to |
| 505 |
* 1..7. |
| 506 |
*/ |
| 507 |
cfg16.u32 = 0; |
| 508 |
cfg16.s.mltd = 1; /* Master Latency Timer Disable */ |
| 509 |
octeon_npi_write32(CVMX_NPI_PCI_CFG16, cfg16.u32); |
| 510 |
|
| 511 |
/* |
| 512 |
* Should be written to 0x4ff00. MTTV -> must be zero. |
| 513 |
* FLUSH -> must be 1. MRV -> should be 0xFF. |
| 514 |
*/ |
| 515 |
cfg22.u32 = 0; |
| 516 |
/* Master Retry Value [1..255] and 0=infinite */ |
| 517 |
cfg22.s.mrv = 0xff; |
| 518 |
/* |
| 519 |
* AM_DO_FLUSH_I control NOTE: This bit MUST BE ONE for proper |
| 520 |
* N3K operation. |
| 521 |
*/ |
| 522 |
cfg22.s.flush = 1; |
| 523 |
octeon_npi_write32(CVMX_NPI_PCI_CFG22, cfg22.u32); |
| 524 |
|
| 525 |
/* |
| 526 |
* MOST Indicates the maximum number of outstanding splits (in -1 |
| 527 |
* notation) when OCTEON is in PCI-X mode. PCI-X performance is |
| 528 |
* affected by the MOST selection. Should generally be written |
| 529 |
* with one of 0x3be807, 0x2be807, 0x1be807, or 0x0be807, |
| 530 |
* depending on the desired MOST of 3, 2, 1, or 0, respectively. |
| 531 |
*/ |
| 532 |
cfg56.u32 = 0; |
| 533 |
cfg56.s.pxcid = 7; /* RO - PCI-X Capability ID */ |
| 534 |
cfg56.s.ncp = 0xe8; /* RO - Next Capability Pointer */ |
| 535 |
cfg56.s.dpere = 1; /* Data Parity Error Recovery Enable */ |
| 536 |
cfg56.s.roe = 1; /* Relaxed Ordering Enable */ |
| 537 |
cfg56.s.mmbc = 1; /* Maximum Memory Byte Count |
| 538 |
[0=512B,1=1024B,2=2048B,3=4096B] */ |
| 539 |
cfg56.s.most = 3; /* Maximum outstanding Split transactions [0=1 |
| 540 |
.. 7=32] */ |
| 541 |
|
| 542 |
octeon_npi_write32(CVMX_NPI_PCI_CFG56, cfg56.u32); |
| 543 |
|
| 544 |
/* |
| 545 |
* Affects PCI performance when OCTEON services reads to its |
| 546 |
* BAR1/BAR2. Refer to Section 10.6.1. The recommended values are |
| 547 |
* 0x22, 0x33, and 0x33 for PCI_READ_CMD_6, PCI_READ_CMD_C, and |
| 548 |
* PCI_READ_CMD_E, respectively. Unfortunately due to errata DDR-700, |
| 549 |
* these values need to be changed so they won't possibly prefetch off |
| 550 |
* of the end of memory if PCI is DMAing a buffer at the end of |
| 551 |
* memory. Note that these values differ from their reset values. |
| 552 |
*/ |
| 553 |
octeon_npi_write32(CVMX_NPI_PCI_READ_CMD_6, 0x21); |
| 554 |
octeon_npi_write32(CVMX_NPI_PCI_READ_CMD_C, 0x31); |
| 555 |
octeon_npi_write32(CVMX_NPI_PCI_READ_CMD_E, 0x31); |
| 556 |
} |
| 557 |
|
| 558 |
|
| 559 |
/* |
| 560 |
* Initialize the Octeon PCI controller |
| 561 |
*/ |
| 562 |
static int __init octeon_pci_setup(void) |
| 563 |
{ |
| 564 |
union cvmx_npi_mem_access_subidx mem_access; |
| 565 |
int index; |
| 566 |
|
| 567 |
/* Only these chips have PCI */ |
| 568 |
if (octeon_has_feature(OCTEON_FEATURE_PCIE)) |
| 569 |
return 0; |
| 570 |
|
| 571 |
if (!octeon_is_pci_host()) { |
| 572 |
pr_notice("Not in host mode, PCI Controller not initialized\n"); |
| 573 |
return 0; |
| 574 |
} |
| 575 |
|
| 576 |
/* Point pcibios_map_irq() to the PCI version of it */ |
| 577 |
octeon_pcibios_map_irq = octeon_pci_pcibios_map_irq; |
| 578 |
|
| 579 |
/* Only use the big bars on chips that support it */ |
| 580 |
if (OCTEON_IS_MODEL(OCTEON_CN31XX) || |
| 581 |
OCTEON_IS_MODEL(OCTEON_CN38XX_PASS2) || |
| 582 |
OCTEON_IS_MODEL(OCTEON_CN38XX_PASS1)) |
| 583 |
octeon_dma_bar_type = OCTEON_DMA_BAR_TYPE_SMALL; |
| 584 |
else |
| 585 |
octeon_dma_bar_type = OCTEON_DMA_BAR_TYPE_BIG; |
| 586 |
|
| 587 |
/* PCI I/O and PCI MEM values */ |
| 588 |
set_io_port_base(OCTEON_PCI_IOSPACE_BASE); |
| 589 |
ioport_resource.start = 0; |
| 590 |
ioport_resource.end = OCTEON_PCI_IOSPACE_SIZE - 1; |
| 591 |
|
| 592 |
pr_notice("%s Octeon big bar support\n", |
| 593 |
(octeon_dma_bar_type == |
| 594 |
OCTEON_DMA_BAR_TYPE_BIG) ? "Enabling" : "Disabling"); |
| 595 |
|
| 596 |
octeon_pci_initialize(); |
| 597 |
|
| 598 |
mem_access.u64 = 0; |
| 599 |
mem_access.s.esr = 1; /* Endian-Swap on read. */ |
| 600 |
mem_access.s.esw = 1; /* Endian-Swap on write. */ |
| 601 |
mem_access.s.nsr = 0; /* No-Snoop on read. */ |
| 602 |
mem_access.s.nsw = 0; /* No-Snoop on write. */ |
| 603 |
mem_access.s.ror = 0; /* Relax Read on read. */ |
| 604 |
mem_access.s.row = 0; /* Relax Order on write. */ |
| 605 |
mem_access.s.ba = 0; /* PCI Address bits [63:36]. */ |
| 606 |
cvmx_write_csr(CVMX_NPI_MEM_ACCESS_SUBID3, mem_access.u64); |
| 607 |
|
| 608 |
/* |
| 609 |
* Remap the Octeon BAR 2 above all 32 bit devices |
| 610 |
* (0x8000000000ul). This is done here so it is remapped |
| 611 |
* before the readl()'s below. We don't want BAR2 overlapping |
| 612 |
* with BAR0/BAR1 during these reads. |
| 613 |
*/ |
| 614 |
octeon_npi_write32(CVMX_NPI_PCI_CFG08, |
| 615 |
(u32)(OCTEON_BAR2_PCI_ADDRESS & 0xffffffffull)); |
| 616 |
octeon_npi_write32(CVMX_NPI_PCI_CFG09, |
| 617 |
(u32)(OCTEON_BAR2_PCI_ADDRESS >> 32)); |
| 618 |
|
| 619 |
if (octeon_dma_bar_type == OCTEON_DMA_BAR_TYPE_BIG) { |
| 620 |
/* Remap the Octeon BAR 0 to 0-2GB */ |
| 621 |
octeon_npi_write32(CVMX_NPI_PCI_CFG04, 0); |
| 622 |
octeon_npi_write32(CVMX_NPI_PCI_CFG05, 0); |
| 623 |
|
| 624 |
/* |
| 625 |
* Remap the Octeon BAR 1 to map 2GB-4GB (minus the |
| 626 |
* BAR 1 hole). |
| 627 |
*/ |
| 628 |
octeon_npi_write32(CVMX_NPI_PCI_CFG06, 2ul << 30); |
| 629 |
octeon_npi_write32(CVMX_NPI_PCI_CFG07, 0); |
| 630 |
|
| 631 |
/* BAR1 movable mappings set for identity mapping */ |
| 632 |
octeon_bar1_pci_phys = 0x80000000ull; |
| 633 |
for (index = 0; index < 32; index++) { |
| 634 |
union cvmx_pci_bar1_indexx bar1_index; |
| 635 |
|
| 636 |
bar1_index.u32 = 0; |
| 637 |
/* Address bits[35:22] sent to L2C */ |
| 638 |
bar1_index.s.addr_idx = |
| 639 |
(octeon_bar1_pci_phys >> 22) + index; |
| 640 |
/* Don't put PCI accesses in L2. */ |
| 641 |
bar1_index.s.ca = 1; |
| 642 |
/* Endian Swap Mode */ |
| 643 |
bar1_index.s.end_swp = 1; |
| 644 |
/* Set '1' when the selected address range is valid. */ |
| 645 |
bar1_index.s.addr_v = 1; |
| 646 |
octeon_npi_write32(CVMX_NPI_PCI_BAR1_INDEXX(index), |
| 647 |
bar1_index.u32); |
| 648 |
} |
| 649 |
|
| 650 |
/* Devices go after BAR1 */ |
| 651 |
octeon_pci_mem_resource.start = |
| 652 |
OCTEON_PCI_MEMSPACE_OFFSET + (4ul << 30) - |
| 653 |
(OCTEON_PCI_BAR1_HOLE_SIZE << 20); |
| 654 |
octeon_pci_mem_resource.end = |
| 655 |
octeon_pci_mem_resource.start + (1ul << 30); |
| 656 |
} else { |
| 657 |
/* Remap the Octeon BAR 0 to map 128MB-(128MB+4KB) */ |
| 658 |
octeon_npi_write32(CVMX_NPI_PCI_CFG04, 128ul << 20); |
| 659 |
octeon_npi_write32(CVMX_NPI_PCI_CFG05, 0); |
| 660 |
|
| 661 |
/* Remap the Octeon BAR 1 to map 0-128MB */ |
| 662 |
octeon_npi_write32(CVMX_NPI_PCI_CFG06, 0); |
| 663 |
octeon_npi_write32(CVMX_NPI_PCI_CFG07, 0); |
| 664 |
|
| 665 |
/* BAR1 movable regions contiguous to cover the swiotlb */ |
| 666 |
octeon_bar1_pci_phys = |
| 667 |
default_swiotlb_base() & ~((1ull << 22) - 1); |
| 668 |
|
| 669 |
for (index = 0; index < 32; index++) { |
| 670 |
union cvmx_pci_bar1_indexx bar1_index; |
| 671 |
|
| 672 |
bar1_index.u32 = 0; |
| 673 |
/* Address bits[35:22] sent to L2C */ |
| 674 |
bar1_index.s.addr_idx = |
| 675 |
(octeon_bar1_pci_phys >> 22) + index; |
| 676 |
/* Don't put PCI accesses in L2. */ |
| 677 |
bar1_index.s.ca = 1; |
| 678 |
/* Endian Swap Mode */ |
| 679 |
bar1_index.s.end_swp = 1; |
| 680 |
/* Set '1' when the selected address range is valid. */ |
| 681 |
bar1_index.s.addr_v = 1; |
| 682 |
octeon_npi_write32(CVMX_NPI_PCI_BAR1_INDEXX(index), |
| 683 |
bar1_index.u32); |
| 684 |
} |
| 685 |
|
| 686 |
/* Devices go after BAR0 */ |
| 687 |
octeon_pci_mem_resource.start = |
| 688 |
OCTEON_PCI_MEMSPACE_OFFSET + (128ul << 20) + |
| 689 |
(4ul << 10); |
| 690 |
octeon_pci_mem_resource.end = |
| 691 |
octeon_pci_mem_resource.start + (1ul << 30); |
| 692 |
} |
| 693 |
|
| 694 |
register_pci_controller(&octeon_pci_controller); |
| 695 |
|
| 696 |
/* |
| 697 |
* Clear any errors that might be pending from before the bus |
| 698 |
* was setup properly. |
| 699 |
*/ |
| 700 |
cvmx_write_csr(CVMX_NPI_PCI_INT_SUM2, -1); |
| 701 |
|
| 702 |
if (IS_ERR(platform_device_register_simple("octeon_pci_edac", |
| 703 |
-1, NULL, 0))) |
| 704 |
pr_err("Registration of co_pci_edac failed!\n"); |
| 705 |
|
| 706 |
octeon_pci_dma_init(); |
| 707 |
|
| 708 |
return 0; |
| 709 |
} |
| 710 |
|
| 711 |
arch_initcall(octeon_pci_setup); |