Arend van Spriel | 5b435de | 2011-10-05 13:19:03 +0200 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2010 Broadcom Corporation |
| 3 | * |
| 4 | * Permission to use, copy, modify, and/or distribute this software for any |
| 5 | * purpose with or without fee is hereby granted, provided that the above |
| 6 | * copyright notice and this permission notice appear in all copies. |
| 7 | * |
| 8 | * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES |
| 9 | * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF |
| 10 | * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY |
| 11 | * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES |
| 12 | * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION |
| 13 | * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN |
| 14 | * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. |
| 15 | */ |
| 16 | #include <linux/slab.h> |
| 17 | #include <linux/skbuff.h> |
| 18 | #include <linux/delay.h> |
| 19 | #include <linux/pci.h> |
| 20 | |
| 21 | #include <brcmu_utils.h> |
| 22 | #include <aiutils.h> |
| 23 | #include "types.h" |
| 24 | #include "dma.h" |
| 25 | |
| 26 | /* |
| 27 | * DMA hardware requires each descriptor ring to be 8kB aligned, and fit within |
| 28 | * a contiguous 8kB physical address. |
| 29 | */ |
| 30 | #define D64RINGALIGN_BITS 13 |
| 31 | #define D64MAXRINGSZ (1 << D64RINGALIGN_BITS) |
| 32 | #define D64RINGALIGN (1 << D64RINGALIGN_BITS) |
| 33 | |
| 34 | #define D64MAXDD (D64MAXRINGSZ / sizeof(struct dma64desc)) |
| 35 | |
| 36 | /* transmit channel control */ |
| 37 | #define D64_XC_XE 0x00000001 /* transmit enable */ |
| 38 | #define D64_XC_SE 0x00000002 /* transmit suspend request */ |
| 39 | #define D64_XC_LE 0x00000004 /* loopback enable */ |
| 40 | #define D64_XC_FL 0x00000010 /* flush request */ |
| 41 | #define D64_XC_PD 0x00000800 /* parity check disable */ |
| 42 | #define D64_XC_AE 0x00030000 /* address extension bits */ |
| 43 | #define D64_XC_AE_SHIFT 16 |
| 44 | |
| 45 | /* transmit descriptor table pointer */ |
| 46 | #define D64_XP_LD_MASK 0x00000fff /* last valid descriptor */ |
| 47 | |
| 48 | /* transmit channel status */ |
| 49 | #define D64_XS0_CD_MASK 0x00001fff /* current descriptor pointer */ |
| 50 | #define D64_XS0_XS_MASK 0xf0000000 /* transmit state */ |
| 51 | #define D64_XS0_XS_SHIFT 28 |
| 52 | #define D64_XS0_XS_DISABLED 0x00000000 /* disabled */ |
| 53 | #define D64_XS0_XS_ACTIVE 0x10000000 /* active */ |
| 54 | #define D64_XS0_XS_IDLE 0x20000000 /* idle wait */ |
| 55 | #define D64_XS0_XS_STOPPED 0x30000000 /* stopped */ |
| 56 | #define D64_XS0_XS_SUSP 0x40000000 /* suspend pending */ |
| 57 | |
| 58 | #define D64_XS1_AD_MASK 0x00001fff /* active descriptor */ |
| 59 | #define D64_XS1_XE_MASK 0xf0000000 /* transmit errors */ |
| 60 | #define D64_XS1_XE_SHIFT 28 |
| 61 | #define D64_XS1_XE_NOERR 0x00000000 /* no error */ |
| 62 | #define D64_XS1_XE_DPE 0x10000000 /* descriptor protocol error */ |
| 63 | #define D64_XS1_XE_DFU 0x20000000 /* data fifo underrun */ |
| 64 | #define D64_XS1_XE_DTE 0x30000000 /* data transfer error */ |
| 65 | #define D64_XS1_XE_DESRE 0x40000000 /* descriptor read error */ |
| 66 | #define D64_XS1_XE_COREE 0x50000000 /* core error */ |
| 67 | |
| 68 | /* receive channel control */ |
| 69 | /* receive enable */ |
| 70 | #define D64_RC_RE 0x00000001 |
| 71 | /* receive frame offset */ |
| 72 | #define D64_RC_RO_MASK 0x000000fe |
| 73 | #define D64_RC_RO_SHIFT 1 |
| 74 | /* direct fifo receive (pio) mode */ |
| 75 | #define D64_RC_FM 0x00000100 |
| 76 | /* separate rx header descriptor enable */ |
| 77 | #define D64_RC_SH 0x00000200 |
| 78 | /* overflow continue */ |
| 79 | #define D64_RC_OC 0x00000400 |
| 80 | /* parity check disable */ |
| 81 | #define D64_RC_PD 0x00000800 |
| 82 | /* address extension bits */ |
| 83 | #define D64_RC_AE 0x00030000 |
| 84 | #define D64_RC_AE_SHIFT 16 |
| 85 | |
| 86 | /* flags for dma controller */ |
| 87 | /* partity enable */ |
| 88 | #define DMA_CTRL_PEN (1 << 0) |
| 89 | /* rx overflow continue */ |
| 90 | #define DMA_CTRL_ROC (1 << 1) |
| 91 | /* allow rx scatter to multiple descriptors */ |
| 92 | #define DMA_CTRL_RXMULTI (1 << 2) |
| 93 | /* Unframed Rx/Tx data */ |
| 94 | #define DMA_CTRL_UNFRAMED (1 << 3) |
| 95 | |
| 96 | /* receive descriptor table pointer */ |
| 97 | #define D64_RP_LD_MASK 0x00000fff /* last valid descriptor */ |
| 98 | |
| 99 | /* receive channel status */ |
| 100 | #define D64_RS0_CD_MASK 0x00001fff /* current descriptor pointer */ |
| 101 | #define D64_RS0_RS_MASK 0xf0000000 /* receive state */ |
| 102 | #define D64_RS0_RS_SHIFT 28 |
| 103 | #define D64_RS0_RS_DISABLED 0x00000000 /* disabled */ |
| 104 | #define D64_RS0_RS_ACTIVE 0x10000000 /* active */ |
| 105 | #define D64_RS0_RS_IDLE 0x20000000 /* idle wait */ |
| 106 | #define D64_RS0_RS_STOPPED 0x30000000 /* stopped */ |
| 107 | #define D64_RS0_RS_SUSP 0x40000000 /* suspend pending */ |
| 108 | |
| 109 | #define D64_RS1_AD_MASK 0x0001ffff /* active descriptor */ |
| 110 | #define D64_RS1_RE_MASK 0xf0000000 /* receive errors */ |
| 111 | #define D64_RS1_RE_SHIFT 28 |
| 112 | #define D64_RS1_RE_NOERR 0x00000000 /* no error */ |
| 113 | #define D64_RS1_RE_DPO 0x10000000 /* descriptor protocol error */ |
| 114 | #define D64_RS1_RE_DFU 0x20000000 /* data fifo overflow */ |
| 115 | #define D64_RS1_RE_DTE 0x30000000 /* data transfer error */ |
| 116 | #define D64_RS1_RE_DESRE 0x40000000 /* descriptor read error */ |
| 117 | #define D64_RS1_RE_COREE 0x50000000 /* core error */ |
| 118 | |
| 119 | /* fifoaddr */ |
| 120 | #define D64_FA_OFF_MASK 0xffff /* offset */ |
| 121 | #define D64_FA_SEL_MASK 0xf0000 /* select */ |
| 122 | #define D64_FA_SEL_SHIFT 16 |
| 123 | #define D64_FA_SEL_XDD 0x00000 /* transmit dma data */ |
| 124 | #define D64_FA_SEL_XDP 0x10000 /* transmit dma pointers */ |
| 125 | #define D64_FA_SEL_RDD 0x40000 /* receive dma data */ |
| 126 | #define D64_FA_SEL_RDP 0x50000 /* receive dma pointers */ |
| 127 | #define D64_FA_SEL_XFD 0x80000 /* transmit fifo data */ |
| 128 | #define D64_FA_SEL_XFP 0x90000 /* transmit fifo pointers */ |
| 129 | #define D64_FA_SEL_RFD 0xc0000 /* receive fifo data */ |
| 130 | #define D64_FA_SEL_RFP 0xd0000 /* receive fifo pointers */ |
| 131 | #define D64_FA_SEL_RSD 0xe0000 /* receive frame status data */ |
| 132 | #define D64_FA_SEL_RSP 0xf0000 /* receive frame status pointers */ |
| 133 | |
| 134 | /* descriptor control flags 1 */ |
| 135 | #define D64_CTRL_COREFLAGS 0x0ff00000 /* core specific flags */ |
| 136 | #define D64_CTRL1_EOT ((u32)1 << 28) /* end of descriptor table */ |
| 137 | #define D64_CTRL1_IOC ((u32)1 << 29) /* interrupt on completion */ |
| 138 | #define D64_CTRL1_EOF ((u32)1 << 30) /* end of frame */ |
| 139 | #define D64_CTRL1_SOF ((u32)1 << 31) /* start of frame */ |
| 140 | |
| 141 | /* descriptor control flags 2 */ |
| 142 | /* buffer byte count. real data len must <= 16KB */ |
| 143 | #define D64_CTRL2_BC_MASK 0x00007fff |
| 144 | /* address extension bits */ |
| 145 | #define D64_CTRL2_AE 0x00030000 |
| 146 | #define D64_CTRL2_AE_SHIFT 16 |
| 147 | /* parity bit */ |
| 148 | #define D64_CTRL2_PARITY 0x00040000 |
| 149 | |
| 150 | /* control flags in the range [27:20] are core-specific and not defined here */ |
| 151 | #define D64_CTRL_CORE_MASK 0x0ff00000 |
| 152 | |
| 153 | #define D64_RX_FRM_STS_LEN 0x0000ffff /* frame length mask */ |
| 154 | #define D64_RX_FRM_STS_OVFL 0x00800000 /* RxOverFlow */ |
| 155 | #define D64_RX_FRM_STS_DSCRCNT 0x0f000000 /* no. of descriptors used - 1 */ |
| 156 | #define D64_RX_FRM_STS_DATATYPE 0xf0000000 /* core-dependent data type */ |
| 157 | |
| 158 | /* |
| 159 | * packet headroom necessary to accommodate the largest header |
| 160 | * in the system, (i.e TXOFF). By doing, we avoid the need to |
| 161 | * allocate an extra buffer for the header when bridging to WL. |
| 162 | * There is a compile time check in wlc.c which ensure that this |
| 163 | * value is at least as big as TXOFF. This value is used in |
| 164 | * dma_rxfill(). |
| 165 | */ |
| 166 | |
| 167 | #define BCMEXTRAHDROOM 172 |
| 168 | |
| 169 | /* debug/trace */ |
| 170 | #ifdef BCMDBG |
| 171 | #define DMA_ERROR(args) \ |
| 172 | do { \ |
| 173 | if (!(*di->msg_level & 1)) \ |
| 174 | ; \ |
| 175 | else \ |
| 176 | printk args; \ |
| 177 | } while (0) |
| 178 | #define DMA_TRACE(args) \ |
| 179 | do { \ |
| 180 | if (!(*di->msg_level & 2)) \ |
| 181 | ; \ |
| 182 | else \ |
| 183 | printk args; \ |
| 184 | } while (0) |
| 185 | #else |
| 186 | #define DMA_ERROR(args) |
| 187 | #define DMA_TRACE(args) |
| 188 | #endif /* BCMDBG */ |
| 189 | |
| 190 | #define DMA_NONE(args) |
| 191 | |
| 192 | #define MAXNAMEL 8 /* 8 char names */ |
| 193 | |
| 194 | /* macros to convert between byte offsets and indexes */ |
| 195 | #define B2I(bytes, type) ((bytes) / sizeof(type)) |
| 196 | #define I2B(index, type) ((index) * sizeof(type)) |
| 197 | |
| 198 | #define PCI32ADDR_HIGH 0xc0000000 /* address[31:30] */ |
| 199 | #define PCI32ADDR_HIGH_SHIFT 30 /* address[31:30] */ |
| 200 | |
| 201 | #define PCI64ADDR_HIGH 0x80000000 /* address[63] */ |
| 202 | #define PCI64ADDR_HIGH_SHIFT 31 /* address[63] */ |
| 203 | |
| 204 | /* |
| 205 | * DMA Descriptor |
| 206 | * Descriptors are only read by the hardware, never written back. |
| 207 | */ |
| 208 | struct dma64desc { |
| 209 | __le32 ctrl1; /* misc control bits & bufcount */ |
| 210 | __le32 ctrl2; /* buffer count and address extension */ |
| 211 | __le32 addrlow; /* memory address of the date buffer, bits 31:0 */ |
| 212 | __le32 addrhigh; /* memory address of the date buffer, bits 63:32 */ |
| 213 | }; |
| 214 | |
| 215 | /* dma engine software state */ |
| 216 | struct dma_info { |
| 217 | struct dma_pub dma; /* exported structure */ |
| 218 | uint *msg_level; /* message level pointer */ |
| 219 | char name[MAXNAMEL]; /* callers name for diag msgs */ |
| 220 | |
| 221 | struct pci_dev *pbus; /* bus handle */ |
| 222 | |
| 223 | bool dma64; /* this dma engine is operating in 64-bit mode */ |
| 224 | bool addrext; /* this dma engine supports DmaExtendedAddrChanges */ |
| 225 | |
| 226 | /* 64-bit dma tx engine registers */ |
| 227 | struct dma64regs __iomem *d64txregs; |
| 228 | /* 64-bit dma rx engine registers */ |
| 229 | struct dma64regs __iomem *d64rxregs; |
| 230 | /* pointer to dma64 tx descriptor ring */ |
| 231 | struct dma64desc *txd64; |
| 232 | /* pointer to dma64 rx descriptor ring */ |
| 233 | struct dma64desc *rxd64; |
| 234 | |
| 235 | u16 dmadesc_align; /* alignment requirement for dma descriptors */ |
| 236 | |
| 237 | u16 ntxd; /* # tx descriptors tunable */ |
| 238 | u16 txin; /* index of next descriptor to reclaim */ |
| 239 | u16 txout; /* index of next descriptor to post */ |
| 240 | /* pointer to parallel array of pointers to packets */ |
| 241 | struct sk_buff **txp; |
| 242 | /* Aligned physical address of descriptor ring */ |
| 243 | dma_addr_t txdpa; |
| 244 | /* Original physical address of descriptor ring */ |
| 245 | dma_addr_t txdpaorig; |
| 246 | u16 txdalign; /* #bytes added to alloc'd mem to align txd */ |
| 247 | u32 txdalloc; /* #bytes allocated for the ring */ |
| 248 | u32 xmtptrbase; /* When using unaligned descriptors, the ptr register |
| 249 | * is not just an index, it needs all 13 bits to be |
| 250 | * an offset from the addr register. |
| 251 | */ |
| 252 | |
| 253 | u16 nrxd; /* # rx descriptors tunable */ |
| 254 | u16 rxin; /* index of next descriptor to reclaim */ |
| 255 | u16 rxout; /* index of next descriptor to post */ |
| 256 | /* pointer to parallel array of pointers to packets */ |
| 257 | struct sk_buff **rxp; |
| 258 | /* Aligned physical address of descriptor ring */ |
| 259 | dma_addr_t rxdpa; |
| 260 | /* Original physical address of descriptor ring */ |
| 261 | dma_addr_t rxdpaorig; |
| 262 | u16 rxdalign; /* #bytes added to alloc'd mem to align rxd */ |
| 263 | u32 rxdalloc; /* #bytes allocated for the ring */ |
| 264 | u32 rcvptrbase; /* Base for ptr reg when using unaligned descriptors */ |
| 265 | |
| 266 | /* tunables */ |
| 267 | unsigned int rxbufsize; /* rx buffer size in bytes, not including |
| 268 | * the extra headroom |
| 269 | */ |
| 270 | uint rxextrahdrroom; /* extra rx headroom, reverseved to assist upper |
| 271 | * stack, e.g. some rx pkt buffers will be |
| 272 | * bridged to tx side without byte copying. |
| 273 | * The extra headroom needs to be large enough |
| 274 | * to fit txheader needs. Some dongle driver may |
| 275 | * not need it. |
| 276 | */ |
| 277 | uint nrxpost; /* # rx buffers to keep posted */ |
| 278 | unsigned int rxoffset; /* rxcontrol offset */ |
| 279 | /* add to get dma address of descriptor ring, low 32 bits */ |
| 280 | uint ddoffsetlow; |
| 281 | /* high 32 bits */ |
| 282 | uint ddoffsethigh; |
| 283 | /* add to get dma address of data buffer, low 32 bits */ |
| 284 | uint dataoffsetlow; |
| 285 | /* high 32 bits */ |
| 286 | uint dataoffsethigh; |
| 287 | /* descriptor base need to be aligned or not */ |
| 288 | bool aligndesc_4k; |
| 289 | }; |
| 290 | |
| 291 | /* |
| 292 | * default dma message level (if input msg_level |
| 293 | * pointer is null in dma_attach()) |
| 294 | */ |
| 295 | static uint dma_msg_level; |
| 296 | |
| 297 | /* Check for odd number of 1's */ |
| 298 | static u32 parity32(__le32 data) |
| 299 | { |
| 300 | /* no swap needed for counting 1's */ |
| 301 | u32 par_data = *(u32 *)&data; |
| 302 | |
| 303 | par_data ^= par_data >> 16; |
| 304 | par_data ^= par_data >> 8; |
| 305 | par_data ^= par_data >> 4; |
| 306 | par_data ^= par_data >> 2; |
| 307 | par_data ^= par_data >> 1; |
| 308 | |
| 309 | return par_data & 1; |
| 310 | } |
| 311 | |
| 312 | static bool dma64_dd_parity(struct dma64desc *dd) |
| 313 | { |
| 314 | return parity32(dd->addrlow ^ dd->addrhigh ^ dd->ctrl1 ^ dd->ctrl2); |
| 315 | } |
| 316 | |
| 317 | /* descriptor bumping functions */ |
| 318 | |
| 319 | static uint xxd(uint x, uint n) |
| 320 | { |
| 321 | return x & (n - 1); /* faster than %, but n must be power of 2 */ |
| 322 | } |
| 323 | |
| 324 | static uint txd(struct dma_info *di, uint x) |
| 325 | { |
| 326 | return xxd(x, di->ntxd); |
| 327 | } |
| 328 | |
| 329 | static uint rxd(struct dma_info *di, uint x) |
| 330 | { |
| 331 | return xxd(x, di->nrxd); |
| 332 | } |
| 333 | |
| 334 | static uint nexttxd(struct dma_info *di, uint i) |
| 335 | { |
| 336 | return txd(di, i + 1); |
| 337 | } |
| 338 | |
| 339 | static uint prevtxd(struct dma_info *di, uint i) |
| 340 | { |
| 341 | return txd(di, i - 1); |
| 342 | } |
| 343 | |
| 344 | static uint nextrxd(struct dma_info *di, uint i) |
| 345 | { |
| 346 | return txd(di, i + 1); |
| 347 | } |
| 348 | |
| 349 | static uint ntxdactive(struct dma_info *di, uint h, uint t) |
| 350 | { |
| 351 | return txd(di, t-h); |
| 352 | } |
| 353 | |
| 354 | static uint nrxdactive(struct dma_info *di, uint h, uint t) |
| 355 | { |
| 356 | return rxd(di, t-h); |
| 357 | } |
| 358 | |
| 359 | static uint _dma_ctrlflags(struct dma_info *di, uint mask, uint flags) |
| 360 | { |
Arend van Spriel | ae8e467 | 2011-10-29 11:30:15 +0200 | [diff] [blame] | 361 | uint dmactrlflags; |
Arend van Spriel | 5b435de | 2011-10-05 13:19:03 +0200 | [diff] [blame] | 362 | |
| 363 | if (di == NULL) { |
Arend van Spriel | ae8e467 | 2011-10-29 11:30:15 +0200 | [diff] [blame] | 364 | DMA_ERROR(("_dma_ctrlflags: NULL dma handle\n")); |
Arend van Spriel | 5b435de | 2011-10-05 13:19:03 +0200 | [diff] [blame] | 365 | return 0; |
| 366 | } |
| 367 | |
Arend van Spriel | ae8e467 | 2011-10-29 11:30:15 +0200 | [diff] [blame] | 368 | dmactrlflags = di->dma.dmactrlflags; |
Arend van Spriel | 5b435de | 2011-10-05 13:19:03 +0200 | [diff] [blame] | 369 | dmactrlflags &= ~mask; |
| 370 | dmactrlflags |= flags; |
| 371 | |
| 372 | /* If trying to enable parity, check if parity is actually supported */ |
| 373 | if (dmactrlflags & DMA_CTRL_PEN) { |
| 374 | u32 control; |
| 375 | |
| 376 | control = R_REG(&di->d64txregs->control); |
| 377 | W_REG(&di->d64txregs->control, |
| 378 | control | D64_XC_PD); |
| 379 | if (R_REG(&di->d64txregs->control) & D64_XC_PD) |
| 380 | /* We *can* disable it so it is supported, |
| 381 | * restore control register |
| 382 | */ |
| 383 | W_REG(&di->d64txregs->control, |
| 384 | control); |
| 385 | else |
| 386 | /* Not supported, don't allow it to be enabled */ |
| 387 | dmactrlflags &= ~DMA_CTRL_PEN; |
| 388 | } |
| 389 | |
| 390 | di->dma.dmactrlflags = dmactrlflags; |
| 391 | |
| 392 | return dmactrlflags; |
| 393 | } |
| 394 | |
| 395 | static bool _dma64_addrext(struct dma64regs __iomem *dma64regs) |
| 396 | { |
| 397 | u32 w; |
| 398 | OR_REG(&dma64regs->control, D64_XC_AE); |
| 399 | w = R_REG(&dma64regs->control); |
| 400 | AND_REG(&dma64regs->control, ~D64_XC_AE); |
| 401 | return (w & D64_XC_AE) == D64_XC_AE; |
| 402 | } |
| 403 | |
| 404 | /* |
| 405 | * return true if this dma engine supports DmaExtendedAddrChanges, |
| 406 | * otherwise false |
| 407 | */ |
| 408 | static bool _dma_isaddrext(struct dma_info *di) |
| 409 | { |
| 410 | /* DMA64 supports full 32- or 64-bit operation. AE is always valid */ |
| 411 | |
| 412 | /* not all tx or rx channel are available */ |
| 413 | if (di->d64txregs != NULL) { |
| 414 | if (!_dma64_addrext(di->d64txregs)) |
| 415 | DMA_ERROR(("%s: _dma_isaddrext: DMA64 tx doesn't have " |
| 416 | "AE set\n", di->name)); |
| 417 | return true; |
| 418 | } else if (di->d64rxregs != NULL) { |
| 419 | if (!_dma64_addrext(di->d64rxregs)) |
| 420 | DMA_ERROR(("%s: _dma_isaddrext: DMA64 rx doesn't have " |
| 421 | "AE set\n", di->name)); |
| 422 | return true; |
| 423 | } |
| 424 | |
| 425 | return false; |
| 426 | } |
| 427 | |
| 428 | static bool _dma_descriptor_align(struct dma_info *di) |
| 429 | { |
| 430 | u32 addrl; |
| 431 | |
| 432 | /* Check to see if the descriptors need to be aligned on 4K/8K or not */ |
| 433 | if (di->d64txregs != NULL) { |
| 434 | W_REG(&di->d64txregs->addrlow, 0xff0); |
| 435 | addrl = R_REG(&di->d64txregs->addrlow); |
| 436 | if (addrl != 0) |
| 437 | return false; |
| 438 | } else if (di->d64rxregs != NULL) { |
| 439 | W_REG(&di->d64rxregs->addrlow, 0xff0); |
| 440 | addrl = R_REG(&di->d64rxregs->addrlow); |
| 441 | if (addrl != 0) |
| 442 | return false; |
| 443 | } |
| 444 | return true; |
| 445 | } |
| 446 | |
| 447 | /* |
| 448 | * Descriptor table must start at the DMA hardware dictated alignment, so |
| 449 | * allocated memory must be large enough to support this requirement. |
| 450 | */ |
| 451 | static void *dma_alloc_consistent(struct pci_dev *pdev, uint size, |
| 452 | u16 align_bits, uint *alloced, |
| 453 | dma_addr_t *pap) |
| 454 | { |
| 455 | if (align_bits) { |
| 456 | u16 align = (1 << align_bits); |
| 457 | if (!IS_ALIGNED(PAGE_SIZE, align)) |
| 458 | size += align; |
| 459 | *alloced = size; |
| 460 | } |
| 461 | return pci_alloc_consistent(pdev, size, pap); |
| 462 | } |
| 463 | |
| 464 | static |
| 465 | u8 dma_align_sizetobits(uint size) |
| 466 | { |
| 467 | u8 bitpos = 0; |
| 468 | while (size >>= 1) |
| 469 | bitpos++; |
| 470 | return bitpos; |
| 471 | } |
| 472 | |
| 473 | /* This function ensures that the DMA descriptor ring will not get allocated |
| 474 | * across Page boundary. If the allocation is done across the page boundary |
| 475 | * at the first time, then it is freed and the allocation is done at |
| 476 | * descriptor ring size aligned location. This will ensure that the ring will |
| 477 | * not cross page boundary |
| 478 | */ |
| 479 | static void *dma_ringalloc(struct dma_info *di, u32 boundary, uint size, |
| 480 | u16 *alignbits, uint *alloced, |
| 481 | dma_addr_t *descpa) |
| 482 | { |
| 483 | void *va; |
| 484 | u32 desc_strtaddr; |
| 485 | u32 alignbytes = 1 << *alignbits; |
| 486 | |
| 487 | va = dma_alloc_consistent(di->pbus, size, *alignbits, alloced, descpa); |
| 488 | |
| 489 | if (NULL == va) |
| 490 | return NULL; |
| 491 | |
| 492 | desc_strtaddr = (u32) roundup((unsigned long)va, alignbytes); |
| 493 | if (((desc_strtaddr + size - 1) & boundary) != (desc_strtaddr |
| 494 | & boundary)) { |
| 495 | *alignbits = dma_align_sizetobits(size); |
| 496 | pci_free_consistent(di->pbus, size, va, *descpa); |
| 497 | va = dma_alloc_consistent(di->pbus, size, *alignbits, |
| 498 | alloced, descpa); |
| 499 | } |
| 500 | return va; |
| 501 | } |
| 502 | |
| 503 | static bool dma64_alloc(struct dma_info *di, uint direction) |
| 504 | { |
| 505 | u16 size; |
| 506 | uint ddlen; |
| 507 | void *va; |
| 508 | uint alloced = 0; |
| 509 | u16 align; |
| 510 | u16 align_bits; |
| 511 | |
| 512 | ddlen = sizeof(struct dma64desc); |
| 513 | |
| 514 | size = (direction == DMA_TX) ? (di->ntxd * ddlen) : (di->nrxd * ddlen); |
| 515 | align_bits = di->dmadesc_align; |
| 516 | align = (1 << align_bits); |
| 517 | |
| 518 | if (direction == DMA_TX) { |
| 519 | va = dma_ringalloc(di, D64RINGALIGN, size, &align_bits, |
| 520 | &alloced, &di->txdpaorig); |
| 521 | if (va == NULL) { |
| 522 | DMA_ERROR(("%s: dma64_alloc: DMA_ALLOC_CONSISTENT(ntxd)" |
| 523 | " failed\n", di->name)); |
| 524 | return false; |
| 525 | } |
| 526 | align = (1 << align_bits); |
| 527 | di->txd64 = (struct dma64desc *) |
| 528 | roundup((unsigned long)va, align); |
| 529 | di->txdalign = (uint) ((s8 *)di->txd64 - (s8 *) va); |
| 530 | di->txdpa = di->txdpaorig + di->txdalign; |
| 531 | di->txdalloc = alloced; |
| 532 | } else { |
| 533 | va = dma_ringalloc(di, D64RINGALIGN, size, &align_bits, |
| 534 | &alloced, &di->rxdpaorig); |
| 535 | if (va == NULL) { |
| 536 | DMA_ERROR(("%s: dma64_alloc: DMA_ALLOC_CONSISTENT(nrxd)" |
| 537 | " failed\n", di->name)); |
| 538 | return false; |
| 539 | } |
| 540 | align = (1 << align_bits); |
| 541 | di->rxd64 = (struct dma64desc *) |
| 542 | roundup((unsigned long)va, align); |
| 543 | di->rxdalign = (uint) ((s8 *)di->rxd64 - (s8 *) va); |
| 544 | di->rxdpa = di->rxdpaorig + di->rxdalign; |
| 545 | di->rxdalloc = alloced; |
| 546 | } |
| 547 | |
| 548 | return true; |
| 549 | } |
| 550 | |
| 551 | static bool _dma_alloc(struct dma_info *di, uint direction) |
| 552 | { |
| 553 | return dma64_alloc(di, direction); |
| 554 | } |
| 555 | |
| 556 | struct dma_pub *dma_attach(char *name, struct si_pub *sih, |
| 557 | void __iomem *dmaregstx, void __iomem *dmaregsrx, |
| 558 | uint ntxd, uint nrxd, |
| 559 | uint rxbufsize, int rxextheadroom, |
| 560 | uint nrxpost, uint rxoffset, uint *msg_level) |
| 561 | { |
| 562 | struct dma_info *di; |
| 563 | uint size; |
| 564 | |
| 565 | /* allocate private info structure */ |
| 566 | di = kzalloc(sizeof(struct dma_info), GFP_ATOMIC); |
| 567 | if (di == NULL) |
| 568 | return NULL; |
| 569 | |
| 570 | di->msg_level = msg_level ? msg_level : &dma_msg_level; |
| 571 | |
| 572 | |
| 573 | di->dma64 = ((ai_core_sflags(sih, 0, 0) & SISF_DMA64) == SISF_DMA64); |
| 574 | |
| 575 | /* init dma reg pointer */ |
| 576 | di->d64txregs = (struct dma64regs __iomem *) dmaregstx; |
| 577 | di->d64rxregs = (struct dma64regs __iomem *) dmaregsrx; |
| 578 | |
| 579 | /* |
| 580 | * Default flags (which can be changed by the driver calling |
| 581 | * dma_ctrlflags before enable): For backwards compatibility |
| 582 | * both Rx Overflow Continue and Parity are DISABLED. |
| 583 | */ |
| 584 | _dma_ctrlflags(di, DMA_CTRL_ROC | DMA_CTRL_PEN, 0); |
| 585 | |
| 586 | DMA_TRACE(("%s: dma_attach: %s flags 0x%x ntxd %d nrxd %d " |
| 587 | "rxbufsize %d rxextheadroom %d nrxpost %d rxoffset %d " |
| 588 | "dmaregstx %p dmaregsrx %p\n", name, "DMA64", |
| 589 | di->dma.dmactrlflags, ntxd, nrxd, rxbufsize, |
| 590 | rxextheadroom, nrxpost, rxoffset, dmaregstx, dmaregsrx)); |
| 591 | |
| 592 | /* make a private copy of our callers name */ |
| 593 | strncpy(di->name, name, MAXNAMEL); |
| 594 | di->name[MAXNAMEL - 1] = '\0'; |
| 595 | |
| 596 | di->pbus = ((struct si_info *)sih)->pbus; |
| 597 | |
| 598 | /* save tunables */ |
| 599 | di->ntxd = (u16) ntxd; |
| 600 | di->nrxd = (u16) nrxd; |
| 601 | |
| 602 | /* the actual dma size doesn't include the extra headroom */ |
| 603 | di->rxextrahdrroom = |
| 604 | (rxextheadroom == -1) ? BCMEXTRAHDROOM : rxextheadroom; |
| 605 | if (rxbufsize > BCMEXTRAHDROOM) |
| 606 | di->rxbufsize = (u16) (rxbufsize - di->rxextrahdrroom); |
| 607 | else |
| 608 | di->rxbufsize = (u16) rxbufsize; |
| 609 | |
| 610 | di->nrxpost = (u16) nrxpost; |
| 611 | di->rxoffset = (u8) rxoffset; |
| 612 | |
| 613 | /* |
| 614 | * figure out the DMA physical address offset for dd and data |
| 615 | * PCI/PCIE: they map silicon backplace address to zero |
| 616 | * based memory, need offset |
| 617 | * Other bus: use zero SI_BUS BIGENDIAN kludge: use sdram |
| 618 | * swapped region for data buffer, not descriptor |
| 619 | */ |
| 620 | di->ddoffsetlow = 0; |
| 621 | di->dataoffsetlow = 0; |
| 622 | /* add offset for pcie with DMA64 bus */ |
| 623 | di->ddoffsetlow = 0; |
| 624 | di->ddoffsethigh = SI_PCIE_DMA_H32; |
| 625 | di->dataoffsetlow = di->ddoffsetlow; |
| 626 | di->dataoffsethigh = di->ddoffsethigh; |
| 627 | /* WAR64450 : DMACtl.Addr ext fields are not supported in SDIOD core. */ |
| 628 | if ((ai_coreid(sih) == SDIOD_CORE_ID) |
| 629 | && ((ai_corerev(sih) > 0) && (ai_corerev(sih) <= 2))) |
| 630 | di->addrext = 0; |
| 631 | else if ((ai_coreid(sih) == I2S_CORE_ID) && |
| 632 | ((ai_corerev(sih) == 0) || (ai_corerev(sih) == 1))) |
| 633 | di->addrext = 0; |
| 634 | else |
| 635 | di->addrext = _dma_isaddrext(di); |
| 636 | |
| 637 | /* does the descriptor need to be aligned and if yes, on 4K/8K or not */ |
| 638 | di->aligndesc_4k = _dma_descriptor_align(di); |
| 639 | if (di->aligndesc_4k) { |
| 640 | di->dmadesc_align = D64RINGALIGN_BITS; |
| 641 | if ((ntxd < D64MAXDD / 2) && (nrxd < D64MAXDD / 2)) |
| 642 | /* for smaller dd table, HW relax alignment reqmnt */ |
| 643 | di->dmadesc_align = D64RINGALIGN_BITS - 1; |
| 644 | } else { |
| 645 | di->dmadesc_align = 4; /* 16 byte alignment */ |
| 646 | } |
| 647 | |
| 648 | DMA_NONE(("DMA descriptor align_needed %d, align %d\n", |
| 649 | di->aligndesc_4k, di->dmadesc_align)); |
| 650 | |
| 651 | /* allocate tx packet pointer vector */ |
| 652 | if (ntxd) { |
| 653 | size = ntxd * sizeof(void *); |
| 654 | di->txp = kzalloc(size, GFP_ATOMIC); |
| 655 | if (di->txp == NULL) |
| 656 | goto fail; |
| 657 | } |
| 658 | |
| 659 | /* allocate rx packet pointer vector */ |
| 660 | if (nrxd) { |
| 661 | size = nrxd * sizeof(void *); |
| 662 | di->rxp = kzalloc(size, GFP_ATOMIC); |
| 663 | if (di->rxp == NULL) |
| 664 | goto fail; |
| 665 | } |
| 666 | |
| 667 | /* |
| 668 | * allocate transmit descriptor ring, only need ntxd descriptors |
| 669 | * but it must be aligned |
| 670 | */ |
| 671 | if (ntxd) { |
| 672 | if (!_dma_alloc(di, DMA_TX)) |
| 673 | goto fail; |
| 674 | } |
| 675 | |
| 676 | /* |
| 677 | * allocate receive descriptor ring, only need nrxd descriptors |
| 678 | * but it must be aligned |
| 679 | */ |
| 680 | if (nrxd) { |
| 681 | if (!_dma_alloc(di, DMA_RX)) |
| 682 | goto fail; |
| 683 | } |
| 684 | |
| 685 | if ((di->ddoffsetlow != 0) && !di->addrext) { |
| 686 | if (di->txdpa > SI_PCI_DMA_SZ) { |
| 687 | DMA_ERROR(("%s: dma_attach: txdpa 0x%x: addrext not " |
| 688 | "supported\n", di->name, (u32)di->txdpa)); |
| 689 | goto fail; |
| 690 | } |
| 691 | if (di->rxdpa > SI_PCI_DMA_SZ) { |
| 692 | DMA_ERROR(("%s: dma_attach: rxdpa 0x%x: addrext not " |
| 693 | "supported\n", di->name, (u32)di->rxdpa)); |
| 694 | goto fail; |
| 695 | } |
| 696 | } |
| 697 | |
| 698 | DMA_TRACE(("ddoffsetlow 0x%x ddoffsethigh 0x%x dataoffsetlow 0x%x " |
| 699 | "dataoffsethigh " "0x%x addrext %d\n", di->ddoffsetlow, |
| 700 | di->ddoffsethigh, di->dataoffsetlow, di->dataoffsethigh, |
| 701 | di->addrext)); |
| 702 | |
| 703 | return (struct dma_pub *) di; |
| 704 | |
| 705 | fail: |
| 706 | dma_detach((struct dma_pub *)di); |
| 707 | return NULL; |
| 708 | } |
| 709 | |
| 710 | static inline void |
| 711 | dma64_dd_upd(struct dma_info *di, struct dma64desc *ddring, |
| 712 | dma_addr_t pa, uint outidx, u32 *flags, u32 bufcount) |
| 713 | { |
| 714 | u32 ctrl2 = bufcount & D64_CTRL2_BC_MASK; |
| 715 | |
| 716 | /* PCI bus with big(>1G) physical address, use address extension */ |
| 717 | if ((di->dataoffsetlow == 0) || !(pa & PCI32ADDR_HIGH)) { |
| 718 | ddring[outidx].addrlow = cpu_to_le32(pa + di->dataoffsetlow); |
| 719 | ddring[outidx].addrhigh = cpu_to_le32(di->dataoffsethigh); |
| 720 | ddring[outidx].ctrl1 = cpu_to_le32(*flags); |
| 721 | ddring[outidx].ctrl2 = cpu_to_le32(ctrl2); |
| 722 | } else { |
| 723 | /* address extension for 32-bit PCI */ |
| 724 | u32 ae; |
| 725 | |
| 726 | ae = (pa & PCI32ADDR_HIGH) >> PCI32ADDR_HIGH_SHIFT; |
| 727 | pa &= ~PCI32ADDR_HIGH; |
| 728 | |
| 729 | ctrl2 |= (ae << D64_CTRL2_AE_SHIFT) & D64_CTRL2_AE; |
| 730 | ddring[outidx].addrlow = cpu_to_le32(pa + di->dataoffsetlow); |
| 731 | ddring[outidx].addrhigh = cpu_to_le32(di->dataoffsethigh); |
| 732 | ddring[outidx].ctrl1 = cpu_to_le32(*flags); |
| 733 | ddring[outidx].ctrl2 = cpu_to_le32(ctrl2); |
| 734 | } |
| 735 | if (di->dma.dmactrlflags & DMA_CTRL_PEN) { |
| 736 | if (dma64_dd_parity(&ddring[outidx])) |
| 737 | ddring[outidx].ctrl2 = |
| 738 | cpu_to_le32(ctrl2 | D64_CTRL2_PARITY); |
| 739 | } |
| 740 | } |
| 741 | |
| 742 | /* !! may be called with core in reset */ |
| 743 | void dma_detach(struct dma_pub *pub) |
| 744 | { |
| 745 | struct dma_info *di = (struct dma_info *)pub; |
| 746 | |
| 747 | DMA_TRACE(("%s: dma_detach\n", di->name)); |
| 748 | |
| 749 | /* free dma descriptor rings */ |
| 750 | if (di->txd64) |
| 751 | pci_free_consistent(di->pbus, di->txdalloc, |
| 752 | ((s8 *)di->txd64 - di->txdalign), |
| 753 | (di->txdpaorig)); |
| 754 | if (di->rxd64) |
| 755 | pci_free_consistent(di->pbus, di->rxdalloc, |
| 756 | ((s8 *)di->rxd64 - di->rxdalign), |
| 757 | (di->rxdpaorig)); |
| 758 | |
| 759 | /* free packet pointer vectors */ |
| 760 | kfree(di->txp); |
| 761 | kfree(di->rxp); |
| 762 | |
| 763 | /* free our private info structure */ |
| 764 | kfree(di); |
| 765 | |
| 766 | } |
| 767 | |
| 768 | /* initialize descriptor table base address */ |
| 769 | static void |
| 770 | _dma_ddtable_init(struct dma_info *di, uint direction, dma_addr_t pa) |
| 771 | { |
| 772 | if (!di->aligndesc_4k) { |
| 773 | if (direction == DMA_TX) |
| 774 | di->xmtptrbase = pa; |
| 775 | else |
| 776 | di->rcvptrbase = pa; |
| 777 | } |
| 778 | |
| 779 | if ((di->ddoffsetlow == 0) |
| 780 | || !(pa & PCI32ADDR_HIGH)) { |
| 781 | if (direction == DMA_TX) { |
| 782 | W_REG(&di->d64txregs->addrlow, pa + di->ddoffsetlow); |
| 783 | W_REG(&di->d64txregs->addrhigh, di->ddoffsethigh); |
| 784 | } else { |
| 785 | W_REG(&di->d64rxregs->addrlow, pa + di->ddoffsetlow); |
| 786 | W_REG(&di->d64rxregs->addrhigh, di->ddoffsethigh); |
| 787 | } |
| 788 | } else { |
| 789 | /* DMA64 32bits address extension */ |
| 790 | u32 ae; |
| 791 | |
| 792 | /* shift the high bit(s) from pa to ae */ |
| 793 | ae = (pa & PCI32ADDR_HIGH) >> PCI32ADDR_HIGH_SHIFT; |
| 794 | pa &= ~PCI32ADDR_HIGH; |
| 795 | |
| 796 | if (direction == DMA_TX) { |
| 797 | W_REG(&di->d64txregs->addrlow, pa + di->ddoffsetlow); |
| 798 | W_REG(&di->d64txregs->addrhigh, di->ddoffsethigh); |
| 799 | SET_REG(&di->d64txregs->control, |
| 800 | D64_XC_AE, (ae << D64_XC_AE_SHIFT)); |
| 801 | } else { |
| 802 | W_REG(&di->d64rxregs->addrlow, pa + di->ddoffsetlow); |
| 803 | W_REG(&di->d64rxregs->addrhigh, di->ddoffsethigh); |
| 804 | SET_REG(&di->d64rxregs->control, |
| 805 | D64_RC_AE, (ae << D64_RC_AE_SHIFT)); |
| 806 | } |
| 807 | } |
| 808 | } |
| 809 | |
| 810 | static void _dma_rxenable(struct dma_info *di) |
| 811 | { |
| 812 | uint dmactrlflags = di->dma.dmactrlflags; |
| 813 | u32 control; |
| 814 | |
| 815 | DMA_TRACE(("%s: dma_rxenable\n", di->name)); |
| 816 | |
| 817 | control = |
| 818 | (R_REG(&di->d64rxregs->control) & D64_RC_AE) | |
| 819 | D64_RC_RE; |
| 820 | |
| 821 | if ((dmactrlflags & DMA_CTRL_PEN) == 0) |
| 822 | control |= D64_RC_PD; |
| 823 | |
| 824 | if (dmactrlflags & DMA_CTRL_ROC) |
| 825 | control |= D64_RC_OC; |
| 826 | |
| 827 | W_REG(&di->d64rxregs->control, |
| 828 | ((di->rxoffset << D64_RC_RO_SHIFT) | control)); |
| 829 | } |
| 830 | |
| 831 | void dma_rxinit(struct dma_pub *pub) |
| 832 | { |
| 833 | struct dma_info *di = (struct dma_info *)pub; |
| 834 | |
| 835 | DMA_TRACE(("%s: dma_rxinit\n", di->name)); |
| 836 | |
| 837 | if (di->nrxd == 0) |
| 838 | return; |
| 839 | |
| 840 | di->rxin = di->rxout = 0; |
| 841 | |
| 842 | /* clear rx descriptor ring */ |
| 843 | memset(di->rxd64, '\0', di->nrxd * sizeof(struct dma64desc)); |
| 844 | |
| 845 | /* DMA engine with out alignment requirement requires table to be inited |
| 846 | * before enabling the engine |
| 847 | */ |
| 848 | if (!di->aligndesc_4k) |
| 849 | _dma_ddtable_init(di, DMA_RX, di->rxdpa); |
| 850 | |
| 851 | _dma_rxenable(di); |
| 852 | |
| 853 | if (di->aligndesc_4k) |
| 854 | _dma_ddtable_init(di, DMA_RX, di->rxdpa); |
| 855 | } |
| 856 | |
| 857 | static struct sk_buff *dma64_getnextrxp(struct dma_info *di, bool forceall) |
| 858 | { |
| 859 | uint i, curr; |
| 860 | struct sk_buff *rxp; |
| 861 | dma_addr_t pa; |
| 862 | |
| 863 | i = di->rxin; |
| 864 | |
| 865 | /* return if no packets posted */ |
| 866 | if (i == di->rxout) |
| 867 | return NULL; |
| 868 | |
| 869 | curr = |
| 870 | B2I(((R_REG(&di->d64rxregs->status0) & D64_RS0_CD_MASK) - |
| 871 | di->rcvptrbase) & D64_RS0_CD_MASK, struct dma64desc); |
| 872 | |
| 873 | /* ignore curr if forceall */ |
| 874 | if (!forceall && (i == curr)) |
| 875 | return NULL; |
| 876 | |
| 877 | /* get the packet pointer that corresponds to the rx descriptor */ |
| 878 | rxp = di->rxp[i]; |
| 879 | di->rxp[i] = NULL; |
| 880 | |
| 881 | pa = le32_to_cpu(di->rxd64[i].addrlow) - di->dataoffsetlow; |
| 882 | |
| 883 | /* clear this packet from the descriptor ring */ |
| 884 | pci_unmap_single(di->pbus, pa, di->rxbufsize, PCI_DMA_FROMDEVICE); |
| 885 | |
| 886 | di->rxd64[i].addrlow = cpu_to_le32(0xdeadbeef); |
| 887 | di->rxd64[i].addrhigh = cpu_to_le32(0xdeadbeef); |
| 888 | |
| 889 | di->rxin = nextrxd(di, i); |
| 890 | |
| 891 | return rxp; |
| 892 | } |
| 893 | |
| 894 | static struct sk_buff *_dma_getnextrxp(struct dma_info *di, bool forceall) |
| 895 | { |
| 896 | if (di->nrxd == 0) |
| 897 | return NULL; |
| 898 | |
| 899 | return dma64_getnextrxp(di, forceall); |
| 900 | } |
| 901 | |
| 902 | /* |
| 903 | * !! rx entry routine |
| 904 | * returns a pointer to the next frame received, or NULL if there are no more |
| 905 | * if DMA_CTRL_RXMULTI is defined, DMA scattering(multiple buffers) is |
| 906 | * supported with pkts chain |
| 907 | * otherwise, it's treated as giant pkt and will be tossed. |
| 908 | * The DMA scattering starts with normal DMA header, followed by first |
| 909 | * buffer data. After it reaches the max size of buffer, the data continues |
| 910 | * in next DMA descriptor buffer WITHOUT DMA header |
| 911 | */ |
| 912 | struct sk_buff *dma_rx(struct dma_pub *pub) |
| 913 | { |
| 914 | struct dma_info *di = (struct dma_info *)pub; |
| 915 | struct sk_buff *p, *head, *tail; |
| 916 | uint len; |
| 917 | uint pkt_len; |
| 918 | int resid = 0; |
| 919 | |
| 920 | next_frame: |
| 921 | head = _dma_getnextrxp(di, false); |
| 922 | if (head == NULL) |
| 923 | return NULL; |
| 924 | |
| 925 | len = le16_to_cpu(*(__le16 *) (head->data)); |
| 926 | DMA_TRACE(("%s: dma_rx len %d\n", di->name, len)); |
| 927 | dma_spin_for_len(len, head); |
| 928 | |
| 929 | /* set actual length */ |
| 930 | pkt_len = min((di->rxoffset + len), di->rxbufsize); |
| 931 | __skb_trim(head, pkt_len); |
| 932 | resid = len - (di->rxbufsize - di->rxoffset); |
| 933 | |
| 934 | /* check for single or multi-buffer rx */ |
| 935 | if (resid > 0) { |
| 936 | tail = head; |
| 937 | while ((resid > 0) && (p = _dma_getnextrxp(di, false))) { |
| 938 | tail->next = p; |
| 939 | pkt_len = min_t(uint, resid, di->rxbufsize); |
| 940 | __skb_trim(p, pkt_len); |
| 941 | |
| 942 | tail = p; |
| 943 | resid -= di->rxbufsize; |
| 944 | } |
| 945 | |
| 946 | #ifdef BCMDBG |
| 947 | if (resid > 0) { |
| 948 | uint cur; |
| 949 | cur = |
| 950 | B2I(((R_REG(&di->d64rxregs->status0) & |
| 951 | D64_RS0_CD_MASK) - |
| 952 | di->rcvptrbase) & D64_RS0_CD_MASK, |
| 953 | struct dma64desc); |
| 954 | DMA_ERROR(("dma_rx, rxin %d rxout %d, hw_curr %d\n", |
| 955 | di->rxin, di->rxout, cur)); |
| 956 | } |
| 957 | #endif /* BCMDBG */ |
| 958 | |
| 959 | if ((di->dma.dmactrlflags & DMA_CTRL_RXMULTI) == 0) { |
| 960 | DMA_ERROR(("%s: dma_rx: bad frame length (%d)\n", |
| 961 | di->name, len)); |
| 962 | brcmu_pkt_buf_free_skb(head); |
| 963 | di->dma.rxgiants++; |
| 964 | goto next_frame; |
| 965 | } |
| 966 | } |
| 967 | |
| 968 | return head; |
| 969 | } |
| 970 | |
| 971 | static bool dma64_rxidle(struct dma_info *di) |
| 972 | { |
| 973 | DMA_TRACE(("%s: dma_rxidle\n", di->name)); |
| 974 | |
| 975 | if (di->nrxd == 0) |
| 976 | return true; |
| 977 | |
| 978 | return ((R_REG(&di->d64rxregs->status0) & D64_RS0_CD_MASK) == |
| 979 | (R_REG(&di->d64rxregs->ptr) & D64_RS0_CD_MASK)); |
| 980 | } |
| 981 | |
| 982 | /* |
| 983 | * post receive buffers |
| 984 | * return false is refill failed completely and ring is empty this will stall |
| 985 | * the rx dma and user might want to call rxfill again asap. This unlikely |
| 986 | * happens on memory-rich NIC, but often on memory-constrained dongle |
| 987 | */ |
| 988 | bool dma_rxfill(struct dma_pub *pub) |
| 989 | { |
| 990 | struct dma_info *di = (struct dma_info *)pub; |
| 991 | struct sk_buff *p; |
| 992 | u16 rxin, rxout; |
| 993 | u32 flags = 0; |
| 994 | uint n; |
| 995 | uint i; |
| 996 | dma_addr_t pa; |
| 997 | uint extra_offset = 0; |
| 998 | bool ring_empty; |
| 999 | |
| 1000 | ring_empty = false; |
| 1001 | |
| 1002 | /* |
| 1003 | * Determine how many receive buffers we're lacking |
| 1004 | * from the full complement, allocate, initialize, |
| 1005 | * and post them, then update the chip rx lastdscr. |
| 1006 | */ |
| 1007 | |
| 1008 | rxin = di->rxin; |
| 1009 | rxout = di->rxout; |
| 1010 | |
| 1011 | n = di->nrxpost - nrxdactive(di, rxin, rxout); |
| 1012 | |
| 1013 | DMA_TRACE(("%s: dma_rxfill: post %d\n", di->name, n)); |
| 1014 | |
| 1015 | if (di->rxbufsize > BCMEXTRAHDROOM) |
| 1016 | extra_offset = di->rxextrahdrroom; |
| 1017 | |
| 1018 | for (i = 0; i < n; i++) { |
| 1019 | /* |
| 1020 | * the di->rxbufsize doesn't include the extra headroom, |
| 1021 | * we need to add it to the size to be allocated |
| 1022 | */ |
| 1023 | p = brcmu_pkt_buf_get_skb(di->rxbufsize + extra_offset); |
| 1024 | |
| 1025 | if (p == NULL) { |
| 1026 | DMA_ERROR(("%s: dma_rxfill: out of rxbufs\n", |
| 1027 | di->name)); |
| 1028 | if (i == 0 && dma64_rxidle(di)) { |
| 1029 | DMA_ERROR(("%s: rxfill64: ring is empty !\n", |
| 1030 | di->name)); |
| 1031 | ring_empty = true; |
| 1032 | } |
| 1033 | di->dma.rxnobuf++; |
| 1034 | break; |
| 1035 | } |
| 1036 | /* reserve an extra headroom, if applicable */ |
| 1037 | if (extra_offset) |
| 1038 | skb_pull(p, extra_offset); |
| 1039 | |
| 1040 | /* Do a cached write instead of uncached write since DMA_MAP |
| 1041 | * will flush the cache. |
| 1042 | */ |
| 1043 | *(u32 *) (p->data) = 0; |
| 1044 | |
| 1045 | pa = pci_map_single(di->pbus, p->data, |
| 1046 | di->rxbufsize, PCI_DMA_FROMDEVICE); |
| 1047 | |
| 1048 | /* save the free packet pointer */ |
| 1049 | di->rxp[rxout] = p; |
| 1050 | |
| 1051 | /* reset flags for each descriptor */ |
| 1052 | flags = 0; |
| 1053 | if (rxout == (di->nrxd - 1)) |
| 1054 | flags = D64_CTRL1_EOT; |
| 1055 | |
| 1056 | dma64_dd_upd(di, di->rxd64, pa, rxout, &flags, |
| 1057 | di->rxbufsize); |
| 1058 | rxout = nextrxd(di, rxout); |
| 1059 | } |
| 1060 | |
| 1061 | di->rxout = rxout; |
| 1062 | |
| 1063 | /* update the chip lastdscr pointer */ |
| 1064 | W_REG(&di->d64rxregs->ptr, |
| 1065 | di->rcvptrbase + I2B(rxout, struct dma64desc)); |
| 1066 | |
| 1067 | return ring_empty; |
| 1068 | } |
| 1069 | |
| 1070 | void dma_rxreclaim(struct dma_pub *pub) |
| 1071 | { |
| 1072 | struct dma_info *di = (struct dma_info *)pub; |
| 1073 | struct sk_buff *p; |
| 1074 | |
| 1075 | DMA_TRACE(("%s: dma_rxreclaim\n", di->name)); |
| 1076 | |
| 1077 | while ((p = _dma_getnextrxp(di, true))) |
| 1078 | brcmu_pkt_buf_free_skb(p); |
| 1079 | } |
| 1080 | |
| 1081 | void dma_counterreset(struct dma_pub *pub) |
| 1082 | { |
| 1083 | /* reset all software counters */ |
| 1084 | pub->rxgiants = 0; |
| 1085 | pub->rxnobuf = 0; |
| 1086 | pub->txnobuf = 0; |
| 1087 | } |
| 1088 | |
| 1089 | /* get the address of the var in order to change later */ |
| 1090 | unsigned long dma_getvar(struct dma_pub *pub, const char *name) |
| 1091 | { |
| 1092 | struct dma_info *di = (struct dma_info *)pub; |
| 1093 | |
| 1094 | if (!strcmp(name, "&txavail")) |
| 1095 | return (unsigned long)&(di->dma.txavail); |
| 1096 | return 0; |
| 1097 | } |
| 1098 | |
| 1099 | /* 64-bit DMA functions */ |
| 1100 | |
| 1101 | void dma_txinit(struct dma_pub *pub) |
| 1102 | { |
| 1103 | struct dma_info *di = (struct dma_info *)pub; |
| 1104 | u32 control = D64_XC_XE; |
| 1105 | |
| 1106 | DMA_TRACE(("%s: dma_txinit\n", di->name)); |
| 1107 | |
| 1108 | if (di->ntxd == 0) |
| 1109 | return; |
| 1110 | |
| 1111 | di->txin = di->txout = 0; |
| 1112 | di->dma.txavail = di->ntxd - 1; |
| 1113 | |
| 1114 | /* clear tx descriptor ring */ |
| 1115 | memset(di->txd64, '\0', (di->ntxd * sizeof(struct dma64desc))); |
| 1116 | |
| 1117 | /* DMA engine with out alignment requirement requires table to be inited |
| 1118 | * before enabling the engine |
| 1119 | */ |
| 1120 | if (!di->aligndesc_4k) |
| 1121 | _dma_ddtable_init(di, DMA_TX, di->txdpa); |
| 1122 | |
| 1123 | if ((di->dma.dmactrlflags & DMA_CTRL_PEN) == 0) |
| 1124 | control |= D64_XC_PD; |
| 1125 | OR_REG(&di->d64txregs->control, control); |
| 1126 | |
| 1127 | /* DMA engine with alignment requirement requires table to be inited |
| 1128 | * before enabling the engine |
| 1129 | */ |
| 1130 | if (di->aligndesc_4k) |
| 1131 | _dma_ddtable_init(di, DMA_TX, di->txdpa); |
| 1132 | } |
| 1133 | |
| 1134 | void dma_txsuspend(struct dma_pub *pub) |
| 1135 | { |
| 1136 | struct dma_info *di = (struct dma_info *)pub; |
| 1137 | |
| 1138 | DMA_TRACE(("%s: dma_txsuspend\n", di->name)); |
| 1139 | |
| 1140 | if (di->ntxd == 0) |
| 1141 | return; |
| 1142 | |
| 1143 | OR_REG(&di->d64txregs->control, D64_XC_SE); |
| 1144 | } |
| 1145 | |
| 1146 | void dma_txresume(struct dma_pub *pub) |
| 1147 | { |
| 1148 | struct dma_info *di = (struct dma_info *)pub; |
| 1149 | |
| 1150 | DMA_TRACE(("%s: dma_txresume\n", di->name)); |
| 1151 | |
| 1152 | if (di->ntxd == 0) |
| 1153 | return; |
| 1154 | |
| 1155 | AND_REG(&di->d64txregs->control, ~D64_XC_SE); |
| 1156 | } |
| 1157 | |
| 1158 | bool dma_txsuspended(struct dma_pub *pub) |
| 1159 | { |
| 1160 | struct dma_info *di = (struct dma_info *)pub; |
| 1161 | |
| 1162 | return (di->ntxd == 0) || |
| 1163 | ((R_REG(&di->d64txregs->control) & D64_XC_SE) == |
| 1164 | D64_XC_SE); |
| 1165 | } |
| 1166 | |
| 1167 | void dma_txreclaim(struct dma_pub *pub, enum txd_range range) |
| 1168 | { |
| 1169 | struct dma_info *di = (struct dma_info *)pub; |
| 1170 | struct sk_buff *p; |
| 1171 | |
| 1172 | DMA_TRACE(("%s: dma_txreclaim %s\n", di->name, |
| 1173 | (range == DMA_RANGE_ALL) ? "all" : |
| 1174 | ((range == |
| 1175 | DMA_RANGE_TRANSMITTED) ? "transmitted" : |
| 1176 | "transferred"))); |
| 1177 | |
| 1178 | if (di->txin == di->txout) |
| 1179 | return; |
| 1180 | |
| 1181 | while ((p = dma_getnexttxp(pub, range))) { |
| 1182 | /* For unframed data, we don't have any packets to free */ |
| 1183 | if (!(di->dma.dmactrlflags & DMA_CTRL_UNFRAMED)) |
| 1184 | brcmu_pkt_buf_free_skb(p); |
| 1185 | } |
| 1186 | } |
| 1187 | |
| 1188 | bool dma_txreset(struct dma_pub *pub) |
| 1189 | { |
| 1190 | struct dma_info *di = (struct dma_info *)pub; |
| 1191 | u32 status; |
| 1192 | |
| 1193 | if (di->ntxd == 0) |
| 1194 | return true; |
| 1195 | |
| 1196 | /* suspend tx DMA first */ |
| 1197 | W_REG(&di->d64txregs->control, D64_XC_SE); |
| 1198 | SPINWAIT(((status = |
| 1199 | (R_REG(&di->d64txregs->status0) & D64_XS0_XS_MASK)) |
| 1200 | != D64_XS0_XS_DISABLED) && (status != D64_XS0_XS_IDLE) |
| 1201 | && (status != D64_XS0_XS_STOPPED), 10000); |
| 1202 | |
| 1203 | W_REG(&di->d64txregs->control, 0); |
| 1204 | SPINWAIT(((status = |
| 1205 | (R_REG(&di->d64txregs->status0) & D64_XS0_XS_MASK)) |
| 1206 | != D64_XS0_XS_DISABLED), 10000); |
| 1207 | |
| 1208 | /* wait for the last transaction to complete */ |
| 1209 | udelay(300); |
| 1210 | |
| 1211 | return status == D64_XS0_XS_DISABLED; |
| 1212 | } |
| 1213 | |
| 1214 | bool dma_rxreset(struct dma_pub *pub) |
| 1215 | { |
| 1216 | struct dma_info *di = (struct dma_info *)pub; |
| 1217 | u32 status; |
| 1218 | |
| 1219 | if (di->nrxd == 0) |
| 1220 | return true; |
| 1221 | |
| 1222 | W_REG(&di->d64rxregs->control, 0); |
| 1223 | SPINWAIT(((status = |
| 1224 | (R_REG(&di->d64rxregs->status0) & D64_RS0_RS_MASK)) |
| 1225 | != D64_RS0_RS_DISABLED), 10000); |
| 1226 | |
| 1227 | return status == D64_RS0_RS_DISABLED; |
| 1228 | } |
| 1229 | |
| 1230 | /* |
| 1231 | * !! tx entry routine |
| 1232 | * WARNING: call must check the return value for error. |
| 1233 | * the error(toss frames) could be fatal and cause many subsequent hard |
| 1234 | * to debug problems |
| 1235 | */ |
| 1236 | int dma_txfast(struct dma_pub *pub, struct sk_buff *p0, bool commit) |
| 1237 | { |
| 1238 | struct dma_info *di = (struct dma_info *)pub; |
| 1239 | struct sk_buff *p, *next; |
| 1240 | unsigned char *data; |
| 1241 | uint len; |
| 1242 | u16 txout; |
| 1243 | u32 flags = 0; |
| 1244 | dma_addr_t pa; |
| 1245 | |
| 1246 | DMA_TRACE(("%s: dma_txfast\n", di->name)); |
| 1247 | |
| 1248 | txout = di->txout; |
| 1249 | |
| 1250 | /* |
| 1251 | * Walk the chain of packet buffers |
| 1252 | * allocating and initializing transmit descriptor entries. |
| 1253 | */ |
| 1254 | for (p = p0; p; p = next) { |
| 1255 | data = p->data; |
| 1256 | len = p->len; |
| 1257 | next = p->next; |
| 1258 | |
| 1259 | /* return nonzero if out of tx descriptors */ |
| 1260 | if (nexttxd(di, txout) == di->txin) |
| 1261 | goto outoftxd; |
| 1262 | |
| 1263 | if (len == 0) |
| 1264 | continue; |
| 1265 | |
| 1266 | /* get physical address of buffer start */ |
| 1267 | pa = pci_map_single(di->pbus, data, len, PCI_DMA_TODEVICE); |
| 1268 | |
| 1269 | flags = 0; |
| 1270 | if (p == p0) |
| 1271 | flags |= D64_CTRL1_SOF; |
| 1272 | |
| 1273 | /* With a DMA segment list, Descriptor table is filled |
| 1274 | * using the segment list instead of looping over |
| 1275 | * buffers in multi-chain DMA. Therefore, EOF for SGLIST |
| 1276 | * is when end of segment list is reached. |
| 1277 | */ |
| 1278 | if (next == NULL) |
| 1279 | flags |= (D64_CTRL1_IOC | D64_CTRL1_EOF); |
| 1280 | if (txout == (di->ntxd - 1)) |
| 1281 | flags |= D64_CTRL1_EOT; |
| 1282 | |
| 1283 | dma64_dd_upd(di, di->txd64, pa, txout, &flags, len); |
| 1284 | |
| 1285 | txout = nexttxd(di, txout); |
| 1286 | } |
| 1287 | |
| 1288 | /* if last txd eof not set, fix it */ |
| 1289 | if (!(flags & D64_CTRL1_EOF)) |
| 1290 | di->txd64[prevtxd(di, txout)].ctrl1 = |
| 1291 | cpu_to_le32(flags | D64_CTRL1_IOC | D64_CTRL1_EOF); |
| 1292 | |
| 1293 | /* save the packet */ |
| 1294 | di->txp[prevtxd(di, txout)] = p0; |
| 1295 | |
| 1296 | /* bump the tx descriptor index */ |
| 1297 | di->txout = txout; |
| 1298 | |
| 1299 | /* kick the chip */ |
| 1300 | if (commit) |
| 1301 | W_REG(&di->d64txregs->ptr, |
| 1302 | di->xmtptrbase + I2B(txout, struct dma64desc)); |
| 1303 | |
| 1304 | /* tx flow control */ |
| 1305 | di->dma.txavail = di->ntxd - ntxdactive(di, di->txin, di->txout) - 1; |
| 1306 | |
| 1307 | return 0; |
| 1308 | |
| 1309 | outoftxd: |
| 1310 | DMA_ERROR(("%s: dma_txfast: out of txds !!!\n", di->name)); |
| 1311 | brcmu_pkt_buf_free_skb(p0); |
| 1312 | di->dma.txavail = 0; |
| 1313 | di->dma.txnobuf++; |
| 1314 | return -1; |
| 1315 | } |
| 1316 | |
| 1317 | /* |
| 1318 | * Reclaim next completed txd (txds if using chained buffers) in the range |
| 1319 | * specified and return associated packet. |
| 1320 | * If range is DMA_RANGE_TRANSMITTED, reclaim descriptors that have be |
| 1321 | * transmitted as noted by the hardware "CurrDescr" pointer. |
| 1322 | * If range is DMA_RANGE_TRANSFERED, reclaim descriptors that have be |
| 1323 | * transferred by the DMA as noted by the hardware "ActiveDescr" pointer. |
| 1324 | * If range is DMA_RANGE_ALL, reclaim all txd(s) posted to the ring and |
| 1325 | * return associated packet regardless of the value of hardware pointers. |
| 1326 | */ |
| 1327 | struct sk_buff *dma_getnexttxp(struct dma_pub *pub, enum txd_range range) |
| 1328 | { |
| 1329 | struct dma_info *di = (struct dma_info *)pub; |
| 1330 | u16 start, end, i; |
| 1331 | u16 active_desc; |
| 1332 | struct sk_buff *txp; |
| 1333 | |
| 1334 | DMA_TRACE(("%s: dma_getnexttxp %s\n", di->name, |
| 1335 | (range == DMA_RANGE_ALL) ? "all" : |
| 1336 | ((range == |
| 1337 | DMA_RANGE_TRANSMITTED) ? "transmitted" : |
| 1338 | "transferred"))); |
| 1339 | |
| 1340 | if (di->ntxd == 0) |
| 1341 | return NULL; |
| 1342 | |
| 1343 | txp = NULL; |
| 1344 | |
| 1345 | start = di->txin; |
| 1346 | if (range == DMA_RANGE_ALL) |
| 1347 | end = di->txout; |
| 1348 | else { |
| 1349 | struct dma64regs __iomem *dregs = di->d64txregs; |
| 1350 | |
| 1351 | end = (u16) (B2I(((R_REG(&dregs->status0) & |
| 1352 | D64_XS0_CD_MASK) - |
| 1353 | di->xmtptrbase) & D64_XS0_CD_MASK, |
| 1354 | struct dma64desc)); |
| 1355 | |
| 1356 | if (range == DMA_RANGE_TRANSFERED) { |
| 1357 | active_desc = |
| 1358 | (u16) (R_REG(&dregs->status1) & |
| 1359 | D64_XS1_AD_MASK); |
| 1360 | active_desc = |
| 1361 | (active_desc - di->xmtptrbase) & D64_XS0_CD_MASK; |
| 1362 | active_desc = B2I(active_desc, struct dma64desc); |
| 1363 | if (end != active_desc) |
| 1364 | end = prevtxd(di, active_desc); |
| 1365 | } |
| 1366 | } |
| 1367 | |
| 1368 | if ((start == 0) && (end > di->txout)) |
| 1369 | goto bogus; |
| 1370 | |
| 1371 | for (i = start; i != end && !txp; i = nexttxd(di, i)) { |
| 1372 | dma_addr_t pa; |
| 1373 | uint size; |
| 1374 | |
| 1375 | pa = le32_to_cpu(di->txd64[i].addrlow) - di->dataoffsetlow; |
| 1376 | |
| 1377 | size = |
| 1378 | (le32_to_cpu(di->txd64[i].ctrl2) & |
| 1379 | D64_CTRL2_BC_MASK); |
| 1380 | |
| 1381 | di->txd64[i].addrlow = cpu_to_le32(0xdeadbeef); |
| 1382 | di->txd64[i].addrhigh = cpu_to_le32(0xdeadbeef); |
| 1383 | |
| 1384 | txp = di->txp[i]; |
| 1385 | di->txp[i] = NULL; |
| 1386 | |
| 1387 | pci_unmap_single(di->pbus, pa, size, PCI_DMA_TODEVICE); |
| 1388 | } |
| 1389 | |
| 1390 | di->txin = i; |
| 1391 | |
| 1392 | /* tx flow control */ |
| 1393 | di->dma.txavail = di->ntxd - ntxdactive(di, di->txin, di->txout) - 1; |
| 1394 | |
| 1395 | return txp; |
| 1396 | |
| 1397 | bogus: |
| 1398 | DMA_NONE(("dma_getnexttxp: bogus curr: start %d end %d txout %d " |
| 1399 | "force %d\n", start, end, di->txout, forceall)); |
| 1400 | return NULL; |
| 1401 | } |
| 1402 | |
| 1403 | /* |
| 1404 | * Mac80211 initiated actions sometimes require packets in the DMA queue to be |
| 1405 | * modified. The modified portion of the packet is not under control of the DMA |
| 1406 | * engine. This function calls a caller-supplied function for each packet in |
| 1407 | * the caller specified dma chain. |
| 1408 | */ |
| 1409 | void dma_walk_packets(struct dma_pub *dmah, void (*callback_fnc) |
| 1410 | (void *pkt, void *arg_a), void *arg_a) |
| 1411 | { |
| 1412 | struct dma_info *di = (struct dma_info *) dmah; |
| 1413 | uint i = di->txin; |
| 1414 | uint end = di->txout; |
| 1415 | struct sk_buff *skb; |
| 1416 | struct ieee80211_tx_info *tx_info; |
| 1417 | |
| 1418 | while (i != end) { |
| 1419 | skb = (struct sk_buff *)di->txp[i]; |
| 1420 | if (skb != NULL) { |
| 1421 | tx_info = (struct ieee80211_tx_info *)skb->cb; |
| 1422 | (callback_fnc)(tx_info, arg_a); |
| 1423 | } |
| 1424 | i = nexttxd(di, i); |
| 1425 | } |
| 1426 | } |