| 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, 2010 Cavium Networks |
| 7 |
*/ |
| 8 |
#include <linux/kernel.h> |
| 9 |
#include <linux/init.h> |
| 10 |
#include <linux/msi.h> |
| 11 |
#include <linux/spinlock.h> |
| 12 |
#include <linux/interrupt.h> |
| 13 |
|
| 14 |
#include <asm/octeon/octeon.h> |
| 15 |
#include <asm/octeon/cvmx-npi-defs.h> |
| 16 |
#include <asm/octeon/cvmx-pci-defs.h> |
| 17 |
#include <asm/octeon/cvmx-npei-defs.h> |
| 18 |
#include <asm/octeon/cvmx-sli-defs.h> |
| 19 |
#include <asm/octeon/cvmx-pexp-defs.h> |
| 20 |
#include <asm/octeon/pci-octeon.h> |
| 21 |
|
| 22 |
/* |
| 23 |
* Each bit in msi_free_irq_bitmask represents a MSI interrupt that is |
| 24 |
* in use. |
| 25 |
*/ |
| 26 |
static u64 msi_free_irq_bitmask[4]; |
| 27 |
|
| 28 |
/* |
| 29 |
* Each bit in msi_multiple_irq_bitmask tells that the device using |
| 30 |
* this bit in msi_free_irq_bitmask is also using the next bit. This |
| 31 |
* is used so we can disable all of the MSI interrupts when a device |
| 32 |
* uses multiple. |
| 33 |
*/ |
| 34 |
static u64 msi_multiple_irq_bitmask[4]; |
| 35 |
|
| 36 |
/* |
| 37 |
* This lock controls updates to msi_free_irq_bitmask and |
| 38 |
* msi_multiple_irq_bitmask. |
| 39 |
*/ |
| 40 |
static DEFINE_SPINLOCK(msi_free_irq_bitmask_lock); |
| 41 |
|
| 42 |
/* |
| 43 |
* Number of MSI IRQs used. This variable is set up in |
| 44 |
* the module init time. |
| 45 |
*/ |
| 46 |
static int msi_irq_size; |
| 47 |
|
| 48 |
/** |
| 49 |
* arch_setup_msi_irq() - setup MSI IRQs for a device |
| 50 |
* @dev: Device requesting MSI interrupts |
| 51 |
* @desc: MSI descriptor |
| 52 |
* |
| 53 |
* Called when a driver requests MSI interrupts instead of the |
| 54 |
* legacy INT A-D. This routine will allocate multiple interrupts |
| 55 |
* for MSI devices that support them. A device can override this by |
| 56 |
* programming the MSI control bits [6:4] before calling |
| 57 |
* pci_enable_msi(). |
| 58 |
* |
| 59 |
* Return: %0 on success, non-%0 on error. |
| 60 |
*/ |
| 61 |
int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc) |
| 62 |
{ |
| 63 |
struct msi_msg msg; |
| 64 |
u16 control; |
| 65 |
int configured_private_bits; |
| 66 |
int request_private_bits; |
| 67 |
int irq = 0; |
| 68 |
int irq_step; |
| 69 |
u64 search_mask; |
| 70 |
int index; |
| 71 |
|
| 72 |
if (desc->pci.msi_attrib.is_msix) |
| 73 |
return -EINVAL; |
| 74 |
|
| 75 |
/* |
| 76 |
* Read the MSI config to figure out how many IRQs this device |
| 77 |
* wants. Most devices only want 1, which will give |
| 78 |
* configured_private_bits and request_private_bits equal 0. |
| 79 |
*/ |
| 80 |
pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control); |
| 81 |
|
| 82 |
/* |
| 83 |
* If the number of private bits has been configured then use |
| 84 |
* that value instead of the requested number. This gives the |
| 85 |
* driver the chance to override the number of interrupts |
| 86 |
* before calling pci_enable_msi(). |
| 87 |
*/ |
| 88 |
configured_private_bits = (control & PCI_MSI_FLAGS_QSIZE) >> 4; |
| 89 |
if (configured_private_bits == 0) { |
| 90 |
/* Nothing is configured, so use the hardware requested size */ |
| 91 |
request_private_bits = (control & PCI_MSI_FLAGS_QMASK) >> 1; |
| 92 |
} else { |
| 93 |
/* |
| 94 |
* Use the number of configured bits, assuming the |
| 95 |
* driver wanted to override the hardware request |
| 96 |
* value. |
| 97 |
*/ |
| 98 |
request_private_bits = configured_private_bits; |
| 99 |
} |
| 100 |
|
| 101 |
/* |
| 102 |
* The PCI 2.3 spec mandates that there are at most 32 |
| 103 |
* interrupts. If this device asks for more, only give it one. |
| 104 |
*/ |
| 105 |
if (request_private_bits > 5) |
| 106 |
request_private_bits = 0; |
| 107 |
|
| 108 |
try_only_one: |
| 109 |
/* |
| 110 |
* The IRQs have to be aligned on a power of two based on the |
| 111 |
* number being requested. |
| 112 |
*/ |
| 113 |
irq_step = 1 << request_private_bits; |
| 114 |
|
| 115 |
/* Mask with one bit for each IRQ */ |
| 116 |
search_mask = (1 << irq_step) - 1; |
| 117 |
|
| 118 |
/* |
| 119 |
* We're going to search msi_free_irq_bitmask_lock for zero |
| 120 |
* bits. This represents an MSI interrupt number that isn't in |
| 121 |
* use. |
| 122 |
*/ |
| 123 |
spin_lock(&msi_free_irq_bitmask_lock); |
| 124 |
for (index = 0; index < msi_irq_size/64; index++) { |
| 125 |
for (irq = 0; irq < 64; irq += irq_step) { |
| 126 |
if ((msi_free_irq_bitmask[index] & (search_mask << irq)) == 0) { |
| 127 |
msi_free_irq_bitmask[index] |= search_mask << irq; |
| 128 |
msi_multiple_irq_bitmask[index] |= (search_mask >> 1) << irq; |
| 129 |
goto msi_irq_allocated; |
| 130 |
} |
| 131 |
} |
| 132 |
} |
| 133 |
msi_irq_allocated: |
| 134 |
spin_unlock(&msi_free_irq_bitmask_lock); |
| 135 |
|
| 136 |
/* Make sure the search for available interrupts didn't fail */ |
| 137 |
if (irq >= 64) { |
| 138 |
if (request_private_bits) { |
| 139 |
pr_err("arch_setup_msi_irq: Unable to find %d free interrupts, trying just one", |
| 140 |
1 << request_private_bits); |
| 141 |
request_private_bits = 0; |
| 142 |
goto try_only_one; |
| 143 |
} else |
| 144 |
panic("arch_setup_msi_irq: Unable to find a free MSI interrupt"); |
| 145 |
} |
| 146 |
|
| 147 |
/* MSI interrupts start at logical IRQ OCTEON_IRQ_MSI_BIT0 */ |
| 148 |
irq += index*64; |
| 149 |
irq += OCTEON_IRQ_MSI_BIT0; |
| 150 |
|
| 151 |
switch (octeon_dma_bar_type) { |
| 152 |
case OCTEON_DMA_BAR_TYPE_SMALL: |
| 153 |
/* When not using big bar, Bar 0 is based at 128MB */ |
| 154 |
msg.address_lo = |
| 155 |
((128ul << 20) + CVMX_PCI_MSI_RCV) & 0xffffffff; |
| 156 |
msg.address_hi = ((128ul << 20) + CVMX_PCI_MSI_RCV) >> 32; |
| 157 |
break; |
| 158 |
case OCTEON_DMA_BAR_TYPE_BIG: |
| 159 |
/* When using big bar, Bar 0 is based at 0 */ |
| 160 |
msg.address_lo = (0 + CVMX_PCI_MSI_RCV) & 0xffffffff; |
| 161 |
msg.address_hi = (0 + CVMX_PCI_MSI_RCV) >> 32; |
| 162 |
break; |
| 163 |
case OCTEON_DMA_BAR_TYPE_PCIE: |
| 164 |
/* When using PCIe, Bar 0 is based at 0 */ |
| 165 |
/* FIXME CVMX_NPEI_MSI_RCV* other than 0? */ |
| 166 |
msg.address_lo = (0 + CVMX_NPEI_PCIE_MSI_RCV) & 0xffffffff; |
| 167 |
msg.address_hi = (0 + CVMX_NPEI_PCIE_MSI_RCV) >> 32; |
| 168 |
break; |
| 169 |
case OCTEON_DMA_BAR_TYPE_PCIE2: |
| 170 |
/* When using PCIe2, Bar 0 is based at 0 */ |
| 171 |
msg.address_lo = (0 + CVMX_SLI_PCIE_MSI_RCV) & 0xffffffff; |
| 172 |
msg.address_hi = (0 + CVMX_SLI_PCIE_MSI_RCV) >> 32; |
| 173 |
break; |
| 174 |
default: |
| 175 |
panic("arch_setup_msi_irq: Invalid octeon_dma_bar_type"); |
| 176 |
} |
| 177 |
msg.data = irq - OCTEON_IRQ_MSI_BIT0; |
| 178 |
|
| 179 |
/* Update the number of IRQs the device has available to it */ |
| 180 |
control &= ~PCI_MSI_FLAGS_QSIZE; |
| 181 |
control |= request_private_bits << 4; |
| 182 |
pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control); |
| 183 |
|
| 184 |
irq_set_msi_desc(irq, desc); |
| 185 |
pci_write_msi_msg(irq, &msg); |
| 186 |
return 0; |
| 187 |
} |
| 188 |
|
| 189 |
/** |
| 190 |
* arch_teardown_msi_irq() - release MSI IRQs for a device |
| 191 |
* @irq: The devices first irq number. There may be multiple in sequence. |
| 192 |
* |
| 193 |
* Called when a device no longer needs its MSI interrupts. All |
| 194 |
* MSI interrupts for the device are freed. |
| 195 |
*/ |
| 196 |
void arch_teardown_msi_irq(unsigned int irq) |
| 197 |
{ |
| 198 |
int number_irqs; |
| 199 |
u64 bitmask; |
| 200 |
int index = 0; |
| 201 |
int irq0; |
| 202 |
|
| 203 |
if ((irq < OCTEON_IRQ_MSI_BIT0) |
| 204 |
|| (irq > msi_irq_size + OCTEON_IRQ_MSI_BIT0)) |
| 205 |
panic("arch_teardown_msi_irq: Attempted to teardown illegal " |
| 206 |
"MSI interrupt (%d)", irq); |
| 207 |
|
| 208 |
irq -= OCTEON_IRQ_MSI_BIT0; |
| 209 |
index = irq / 64; |
| 210 |
irq0 = irq % 64; |
| 211 |
|
| 212 |
/* |
| 213 |
* Count the number of IRQs we need to free by looking at the |
| 214 |
* msi_multiple_irq_bitmask. Each bit set means that the next |
| 215 |
* IRQ is also owned by this device. |
| 216 |
*/ |
| 217 |
number_irqs = 0; |
| 218 |
while ((irq0 + number_irqs < 64) && |
| 219 |
(msi_multiple_irq_bitmask[index] |
| 220 |
& (1ull << (irq0 + number_irqs)))) |
| 221 |
number_irqs++; |
| 222 |
number_irqs++; |
| 223 |
/* Mask with one bit for each IRQ */ |
| 224 |
bitmask = (1 << number_irqs) - 1; |
| 225 |
/* Shift the mask to the correct bit location */ |
| 226 |
bitmask <<= irq0; |
| 227 |
if ((msi_free_irq_bitmask[index] & bitmask) != bitmask) |
| 228 |
panic("arch_teardown_msi_irq: Attempted to teardown MSI " |
| 229 |
"interrupt (%d) not in use", irq); |
| 230 |
|
| 231 |
/* Checks are done, update the in use bitmask */ |
| 232 |
spin_lock(&msi_free_irq_bitmask_lock); |
| 233 |
msi_free_irq_bitmask[index] &= ~bitmask; |
| 234 |
msi_multiple_irq_bitmask[index] &= ~bitmask; |
| 235 |
spin_unlock(&msi_free_irq_bitmask_lock); |
| 236 |
} |
| 237 |
|
| 238 |
static DEFINE_RAW_SPINLOCK(octeon_irq_msi_lock); |
| 239 |
|
| 240 |
static u64 msi_rcv_reg[4]; |
| 241 |
static u64 mis_ena_reg[4]; |
| 242 |
|
| 243 |
static void octeon_irq_msi_enable_pcie(struct irq_data *data) |
| 244 |
{ |
| 245 |
u64 en; |
| 246 |
unsigned long flags; |
| 247 |
int msi_number = data->irq - OCTEON_IRQ_MSI_BIT0; |
| 248 |
int irq_index = msi_number >> 6; |
| 249 |
int irq_bit = msi_number & 0x3f; |
| 250 |
|
| 251 |
raw_spin_lock_irqsave(&octeon_irq_msi_lock, flags); |
| 252 |
en = cvmx_read_csr(mis_ena_reg[irq_index]); |
| 253 |
en |= 1ull << irq_bit; |
| 254 |
cvmx_write_csr(mis_ena_reg[irq_index], en); |
| 255 |
cvmx_read_csr(mis_ena_reg[irq_index]); |
| 256 |
raw_spin_unlock_irqrestore(&octeon_irq_msi_lock, flags); |
| 257 |
} |
| 258 |
|
| 259 |
static void octeon_irq_msi_disable_pcie(struct irq_data *data) |
| 260 |
{ |
| 261 |
u64 en; |
| 262 |
unsigned long flags; |
| 263 |
int msi_number = data->irq - OCTEON_IRQ_MSI_BIT0; |
| 264 |
int irq_index = msi_number >> 6; |
| 265 |
int irq_bit = msi_number & 0x3f; |
| 266 |
|
| 267 |
raw_spin_lock_irqsave(&octeon_irq_msi_lock, flags); |
| 268 |
en = cvmx_read_csr(mis_ena_reg[irq_index]); |
| 269 |
en &= ~(1ull << irq_bit); |
| 270 |
cvmx_write_csr(mis_ena_reg[irq_index], en); |
| 271 |
cvmx_read_csr(mis_ena_reg[irq_index]); |
| 272 |
raw_spin_unlock_irqrestore(&octeon_irq_msi_lock, flags); |
| 273 |
} |
| 274 |
|
| 275 |
static struct irq_chip octeon_irq_chip_msi_pcie = { |
| 276 |
.name = "MSI", |
| 277 |
.irq_enable = octeon_irq_msi_enable_pcie, |
| 278 |
.irq_disable = octeon_irq_msi_disable_pcie, |
| 279 |
}; |
| 280 |
|
| 281 |
static void octeon_irq_msi_enable_pci(struct irq_data *data) |
| 282 |
{ |
| 283 |
/* |
| 284 |
* Octeon PCI doesn't have the ability to mask/unmask MSI |
| 285 |
* interrupts individually. Instead of masking/unmasking them |
| 286 |
* in groups of 16, we simple assume MSI devices are well |
| 287 |
* behaved. MSI interrupts are always enable and the ACK is |
| 288 |
* assumed to be enough |
| 289 |
*/ |
| 290 |
} |
| 291 |
|
| 292 |
static void octeon_irq_msi_disable_pci(struct irq_data *data) |
| 293 |
{ |
| 294 |
/* See comment in enable */ |
| 295 |
} |
| 296 |
|
| 297 |
static struct irq_chip octeon_irq_chip_msi_pci = { |
| 298 |
.name = "MSI", |
| 299 |
.irq_enable = octeon_irq_msi_enable_pci, |
| 300 |
.irq_disable = octeon_irq_msi_disable_pci, |
| 301 |
}; |
| 302 |
|
| 303 |
/* |
| 304 |
* Called by the interrupt handling code when an MSI interrupt |
| 305 |
* occurs. |
| 306 |
*/ |
| 307 |
static irqreturn_t __octeon_msi_do_interrupt(int index, u64 msi_bits) |
| 308 |
{ |
| 309 |
int irq; |
| 310 |
int bit; |
| 311 |
|
| 312 |
bit = fls64(msi_bits); |
| 313 |
if (bit) { |
| 314 |
bit--; |
| 315 |
/* Acknowledge it first. */ |
| 316 |
cvmx_write_csr(msi_rcv_reg[index], 1ull << bit); |
| 317 |
|
| 318 |
irq = bit + OCTEON_IRQ_MSI_BIT0 + 64 * index; |
| 319 |
do_IRQ(irq); |
| 320 |
return IRQ_HANDLED; |
| 321 |
} |
| 322 |
return IRQ_NONE; |
| 323 |
} |
| 324 |
|
| 325 |
#define OCTEON_MSI_INT_HANDLER_X(x) \ |
| 326 |
static irqreturn_t octeon_msi_interrupt##x(int cpl, void *dev_id) \ |
| 327 |
{ \ |
| 328 |
u64 msi_bits = cvmx_read_csr(msi_rcv_reg[(x)]); \ |
| 329 |
return __octeon_msi_do_interrupt((x), msi_bits); \ |
| 330 |
} |
| 331 |
|
| 332 |
/* |
| 333 |
* Create octeon_msi_interrupt{0-3} function body |
| 334 |
*/ |
| 335 |
OCTEON_MSI_INT_HANDLER_X(0); |
| 336 |
OCTEON_MSI_INT_HANDLER_X(1); |
| 337 |
OCTEON_MSI_INT_HANDLER_X(2); |
| 338 |
OCTEON_MSI_INT_HANDLER_X(3); |
| 339 |
|
| 340 |
/* |
| 341 |
* Initializes the MSI interrupt handling code |
| 342 |
*/ |
| 343 |
int __init octeon_msi_initialize(void) |
| 344 |
{ |
| 345 |
int irq; |
| 346 |
struct irq_chip *msi; |
| 347 |
|
| 348 |
if (octeon_dma_bar_type == OCTEON_DMA_BAR_TYPE_INVALID) { |
| 349 |
return 0; |
| 350 |
} else if (octeon_dma_bar_type == OCTEON_DMA_BAR_TYPE_PCIE) { |
| 351 |
msi_rcv_reg[0] = CVMX_PEXP_NPEI_MSI_RCV0; |
| 352 |
msi_rcv_reg[1] = CVMX_PEXP_NPEI_MSI_RCV1; |
| 353 |
msi_rcv_reg[2] = CVMX_PEXP_NPEI_MSI_RCV2; |
| 354 |
msi_rcv_reg[3] = CVMX_PEXP_NPEI_MSI_RCV3; |
| 355 |
mis_ena_reg[0] = CVMX_PEXP_NPEI_MSI_ENB0; |
| 356 |
mis_ena_reg[1] = CVMX_PEXP_NPEI_MSI_ENB1; |
| 357 |
mis_ena_reg[2] = CVMX_PEXP_NPEI_MSI_ENB2; |
| 358 |
mis_ena_reg[3] = CVMX_PEXP_NPEI_MSI_ENB3; |
| 359 |
msi = &octeon_irq_chip_msi_pcie; |
| 360 |
} else { |
| 361 |
msi_rcv_reg[0] = CVMX_NPI_NPI_MSI_RCV; |
| 362 |
#define INVALID_GENERATE_ADE 0x8700000000000000ULL; |
| 363 |
msi_rcv_reg[1] = INVALID_GENERATE_ADE; |
| 364 |
msi_rcv_reg[2] = INVALID_GENERATE_ADE; |
| 365 |
msi_rcv_reg[3] = INVALID_GENERATE_ADE; |
| 366 |
mis_ena_reg[0] = INVALID_GENERATE_ADE; |
| 367 |
mis_ena_reg[1] = INVALID_GENERATE_ADE; |
| 368 |
mis_ena_reg[2] = INVALID_GENERATE_ADE; |
| 369 |
mis_ena_reg[3] = INVALID_GENERATE_ADE; |
| 370 |
msi = &octeon_irq_chip_msi_pci; |
| 371 |
} |
| 372 |
|
| 373 |
for (irq = OCTEON_IRQ_MSI_BIT0; irq <= OCTEON_IRQ_MSI_LAST; irq++) |
| 374 |
irq_set_chip_and_handler(irq, msi, handle_simple_irq); |
| 375 |
|
| 376 |
if (octeon_has_feature(OCTEON_FEATURE_PCIE)) { |
| 377 |
if (request_irq(OCTEON_IRQ_PCI_MSI0, octeon_msi_interrupt0, |
| 378 |
0, "MSI[0:63]", octeon_msi_interrupt0)) |
| 379 |
panic("request_irq(OCTEON_IRQ_PCI_MSI0) failed"); |
| 380 |
|
| 381 |
if (request_irq(OCTEON_IRQ_PCI_MSI1, octeon_msi_interrupt1, |
| 382 |
0, "MSI[64:127]", octeon_msi_interrupt1)) |
| 383 |
panic("request_irq(OCTEON_IRQ_PCI_MSI1) failed"); |
| 384 |
|
| 385 |
if (request_irq(OCTEON_IRQ_PCI_MSI2, octeon_msi_interrupt2, |
| 386 |
0, "MSI[127:191]", octeon_msi_interrupt2)) |
| 387 |
panic("request_irq(OCTEON_IRQ_PCI_MSI2) failed"); |
| 388 |
|
| 389 |
if (request_irq(OCTEON_IRQ_PCI_MSI3, octeon_msi_interrupt3, |
| 390 |
0, "MSI[192:255]", octeon_msi_interrupt3)) |
| 391 |
panic("request_irq(OCTEON_IRQ_PCI_MSI3) failed"); |
| 392 |
|
| 393 |
msi_irq_size = 256; |
| 394 |
} else if (octeon_is_pci_host()) { |
| 395 |
if (request_irq(OCTEON_IRQ_PCI_MSI0, octeon_msi_interrupt0, |
| 396 |
0, "MSI[0:15]", octeon_msi_interrupt0)) |
| 397 |
panic("request_irq(OCTEON_IRQ_PCI_MSI0) failed"); |
| 398 |
|
| 399 |
if (request_irq(OCTEON_IRQ_PCI_MSI1, octeon_msi_interrupt0, |
| 400 |
0, "MSI[16:31]", octeon_msi_interrupt0)) |
| 401 |
panic("request_irq(OCTEON_IRQ_PCI_MSI1) failed"); |
| 402 |
|
| 403 |
if (request_irq(OCTEON_IRQ_PCI_MSI2, octeon_msi_interrupt0, |
| 404 |
0, "MSI[32:47]", octeon_msi_interrupt0)) |
| 405 |
panic("request_irq(OCTEON_IRQ_PCI_MSI2) failed"); |
| 406 |
|
| 407 |
if (request_irq(OCTEON_IRQ_PCI_MSI3, octeon_msi_interrupt0, |
| 408 |
0, "MSI[48:63]", octeon_msi_interrupt0)) |
| 409 |
panic("request_irq(OCTEON_IRQ_PCI_MSI3) failed"); |
| 410 |
msi_irq_size = 64; |
| 411 |
} |
| 412 |
return 0; |
| 413 |
} |
| 414 |
subsys_initcall(octeon_msi_initialize); |