1 /* 2 * NVIDIA Tegra210 QSPI controller driver 3 * 4 * (C) Copyright 2015 NVIDIA Corporation <www.nvidia.com> 5 * 6 * SPDX-License-Identifier: GPL-2.0+ 7 */ 8 9 #include <common.h> 10 #include <dm.h> 11 #include <asm/io.h> 12 #include <asm/arch/clock.h> 13 #include <asm/arch-tegra/clk_rst.h> 14 #include <spi.h> 15 #include <fdtdec.h> 16 #include "tegra_spi.h" 17 18 DECLARE_GLOBAL_DATA_PTR; 19 20 /* COMMAND1 */ 21 #define QSPI_CMD1_GO BIT(31) 22 #define QSPI_CMD1_M_S BIT(30) 23 #define QSPI_CMD1_MODE_MASK GENMASK(1,0) 24 #define QSPI_CMD1_MODE_SHIFT 28 25 #define QSPI_CMD1_CS_SEL_MASK GENMASK(1,0) 26 #define QSPI_CMD1_CS_SEL_SHIFT 26 27 #define QSPI_CMD1_CS_POL_INACTIVE0 BIT(22) 28 #define QSPI_CMD1_CS_SW_HW BIT(21) 29 #define QSPI_CMD1_CS_SW_VAL BIT(20) 30 #define QSPI_CMD1_IDLE_SDA_MASK GENMASK(1,0) 31 #define QSPI_CMD1_IDLE_SDA_SHIFT 18 32 #define QSPI_CMD1_BIDIR BIT(17) 33 #define QSPI_CMD1_LSBI_FE BIT(16) 34 #define QSPI_CMD1_LSBY_FE BIT(15) 35 #define QSPI_CMD1_BOTH_EN_BIT BIT(14) 36 #define QSPI_CMD1_BOTH_EN_BYTE BIT(13) 37 #define QSPI_CMD1_RX_EN BIT(12) 38 #define QSPI_CMD1_TX_EN BIT(11) 39 #define QSPI_CMD1_PACKED BIT(5) 40 #define QSPI_CMD1_BITLEN_MASK GENMASK(4,0) 41 #define QSPI_CMD1_BITLEN_SHIFT 0 42 43 /* COMMAND2 */ 44 #define QSPI_CMD2_TX_CLK_TAP_DELAY BIT(6) 45 #define QSPI_CMD2_TX_CLK_TAP_DELAY_MASK GENMASK(11,6) 46 #define QSPI_CMD2_RX_CLK_TAP_DELAY BIT(0) 47 #define QSPI_CMD2_RX_CLK_TAP_DELAY_MASK GENMASK(5,0) 48 49 /* TRANSFER STATUS */ 50 #define QSPI_XFER_STS_RDY BIT(30) 51 52 /* FIFO STATUS */ 53 #define QSPI_FIFO_STS_CS_INACTIVE BIT(31) 54 #define QSPI_FIFO_STS_FRAME_END BIT(30) 55 #define QSPI_FIFO_STS_RX_FIFO_FLUSH BIT(15) 56 #define QSPI_FIFO_STS_TX_FIFO_FLUSH BIT(14) 57 #define QSPI_FIFO_STS_ERR BIT(8) 58 #define QSPI_FIFO_STS_TX_FIFO_OVF BIT(7) 59 #define QSPI_FIFO_STS_TX_FIFO_UNR BIT(6) 60 #define QSPI_FIFO_STS_RX_FIFO_OVF BIT(5) 61 #define QSPI_FIFO_STS_RX_FIFO_UNR BIT(4) 62 #define QSPI_FIFO_STS_TX_FIFO_FULL BIT(3) 63 #define QSPI_FIFO_STS_TX_FIFO_EMPTY BIT(2) 64 #define QSPI_FIFO_STS_RX_FIFO_FULL BIT(1) 65 #define QSPI_FIFO_STS_RX_FIFO_EMPTY BIT(0) 66 67 #define QSPI_TIMEOUT 1000 68 69 struct qspi_regs { 70 u32 command1; /* 000:QSPI_COMMAND1 register */ 71 u32 command2; /* 004:QSPI_COMMAND2 register */ 72 u32 timing1; /* 008:QSPI_CS_TIM1 register */ 73 u32 timing2; /* 00c:QSPI_CS_TIM2 register */ 74 u32 xfer_status;/* 010:QSPI_TRANS_STATUS register */ 75 u32 fifo_status;/* 014:QSPI_FIFO_STATUS register */ 76 u32 tx_data; /* 018:QSPI_TX_DATA register */ 77 u32 rx_data; /* 01c:QSPI_RX_DATA register */ 78 u32 dma_ctl; /* 020:QSPI_DMA_CTL register */ 79 u32 dma_blk; /* 024:QSPI_DMA_BLK register */ 80 u32 rsvd[56]; /* 028-107 reserved */ 81 u32 tx_fifo; /* 108:QSPI_FIFO1 register */ 82 u32 rsvd2[31]; /* 10c-187 reserved */ 83 u32 rx_fifo; /* 188:QSPI_FIFO2 register */ 84 u32 spare_ctl; /* 18c:QSPI_SPARE_CTRL register */ 85 }; 86 87 struct tegra210_qspi_priv { 88 struct qspi_regs *regs; 89 unsigned int freq; 90 unsigned int mode; 91 int periph_id; 92 int valid; 93 int last_transaction_us; 94 }; 95 96 static int tegra210_qspi_ofdata_to_platdata(struct udevice *bus) 97 { 98 struct tegra_spi_platdata *plat = bus->platdata; 99 const void *blob = gd->fdt_blob; 100 int node = dev_of_offset(bus); 101 102 plat->base = devfdt_get_addr(bus); 103 plat->periph_id = clock_decode_periph_id(bus); 104 105 if (plat->periph_id == PERIPH_ID_NONE) { 106 debug("%s: could not decode periph id %d\n", __func__, 107 plat->periph_id); 108 return -FDT_ERR_NOTFOUND; 109 } 110 111 /* Use 500KHz as a suitable default */ 112 plat->frequency = fdtdec_get_int(blob, node, "spi-max-frequency", 113 500000); 114 plat->deactivate_delay_us = fdtdec_get_int(blob, node, 115 "spi-deactivate-delay", 0); 116 debug("%s: base=%#08lx, periph_id=%d, max-frequency=%d, deactivate_delay=%d\n", 117 __func__, plat->base, plat->periph_id, plat->frequency, 118 plat->deactivate_delay_us); 119 120 return 0; 121 } 122 123 static int tegra210_qspi_probe(struct udevice *bus) 124 { 125 struct tegra_spi_platdata *plat = dev_get_platdata(bus); 126 struct tegra210_qspi_priv *priv = dev_get_priv(bus); 127 128 priv->regs = (struct qspi_regs *)plat->base; 129 130 priv->last_transaction_us = timer_get_us(); 131 priv->freq = plat->frequency; 132 priv->periph_id = plat->periph_id; 133 134 /* Change SPI clock to correct frequency, PLLP_OUT0 source */ 135 clock_start_periph_pll(priv->periph_id, CLOCK_ID_PERIPH, priv->freq); 136 137 return 0; 138 } 139 140 static int tegra210_qspi_claim_bus(struct udevice *bus) 141 { 142 struct tegra210_qspi_priv *priv = dev_get_priv(bus); 143 struct qspi_regs *regs = priv->regs; 144 145 /* Change SPI clock to correct frequency, PLLP_OUT0 source */ 146 clock_start_periph_pll(priv->periph_id, CLOCK_ID_PERIPH, priv->freq); 147 148 debug("%s: FIFO STATUS = %08x\n", __func__, readl(®s->fifo_status)); 149 150 /* Set master mode and sw controlled CS */ 151 setbits_le32(®s->command1, QSPI_CMD1_M_S | QSPI_CMD1_CS_SW_HW | 152 (priv->mode << QSPI_CMD1_MODE_SHIFT)); 153 debug("%s: COMMAND1 = %08x\n", __func__, readl(®s->command1)); 154 155 return 0; 156 } 157 158 /** 159 * Activate the CS by driving it LOW 160 * 161 * @param slave Pointer to spi_slave to which controller has to 162 * communicate with 163 */ 164 static void spi_cs_activate(struct udevice *dev) 165 { 166 struct udevice *bus = dev->parent; 167 struct tegra_spi_platdata *pdata = dev_get_platdata(bus); 168 struct tegra210_qspi_priv *priv = dev_get_priv(bus); 169 170 /* If it's too soon to do another transaction, wait */ 171 if (pdata->deactivate_delay_us && 172 priv->last_transaction_us) { 173 ulong delay_us; /* The delay completed so far */ 174 delay_us = timer_get_us() - priv->last_transaction_us; 175 if (delay_us < pdata->deactivate_delay_us) 176 udelay(pdata->deactivate_delay_us - delay_us); 177 } 178 179 clrbits_le32(&priv->regs->command1, QSPI_CMD1_CS_SW_VAL); 180 } 181 182 /** 183 * Deactivate the CS by driving it HIGH 184 * 185 * @param slave Pointer to spi_slave to which controller has to 186 * communicate with 187 */ 188 static void spi_cs_deactivate(struct udevice *dev) 189 { 190 struct udevice *bus = dev->parent; 191 struct tegra_spi_platdata *pdata = dev_get_platdata(bus); 192 struct tegra210_qspi_priv *priv = dev_get_priv(bus); 193 194 setbits_le32(&priv->regs->command1, QSPI_CMD1_CS_SW_VAL); 195 196 /* Remember time of this transaction so we can honour the bus delay */ 197 if (pdata->deactivate_delay_us) 198 priv->last_transaction_us = timer_get_us(); 199 200 debug("Deactivate CS, bus '%s'\n", bus->name); 201 } 202 203 static int tegra210_qspi_xfer(struct udevice *dev, unsigned int bitlen, 204 const void *data_out, void *data_in, 205 unsigned long flags) 206 { 207 struct udevice *bus = dev->parent; 208 struct tegra210_qspi_priv *priv = dev_get_priv(bus); 209 struct qspi_regs *regs = priv->regs; 210 u32 reg, tmpdout, tmpdin = 0; 211 const u8 *dout = data_out; 212 u8 *din = data_in; 213 int num_bytes, tm, ret; 214 215 debug("%s: slave %u:%u dout %p din %p bitlen %u\n", 216 __func__, bus->seq, spi_chip_select(dev), dout, din, bitlen); 217 if (bitlen % 8) 218 return -1; 219 num_bytes = bitlen / 8; 220 221 ret = 0; 222 223 /* clear all error status bits */ 224 reg = readl(®s->fifo_status); 225 writel(reg, ®s->fifo_status); 226 227 /* flush RX/TX FIFOs */ 228 setbits_le32(®s->fifo_status, 229 (QSPI_FIFO_STS_RX_FIFO_FLUSH | 230 QSPI_FIFO_STS_TX_FIFO_FLUSH)); 231 232 tm = QSPI_TIMEOUT; 233 while ((tm && readl(®s->fifo_status) & 234 (QSPI_FIFO_STS_RX_FIFO_FLUSH | 235 QSPI_FIFO_STS_TX_FIFO_FLUSH))) { 236 tm--; 237 udelay(1); 238 } 239 240 if (!tm) { 241 printf("%s: timeout during QSPI FIFO flush!\n", 242 __func__); 243 return -1; 244 } 245 246 /* 247 * Notes: 248 * 1. don't set LSBY_FE, so no need to swap bytes from/to TX/RX FIFOs; 249 * 2. don't set RX_EN and TX_EN yet. 250 * (SW needs to make sure that while programming the blk_size, 251 * tx_en and rx_en bits must be zero) 252 * [TODO] I (Yen Lin) have problems when both RX/TX EN bits are set 253 * i.e., both dout and din are not NULL. 254 */ 255 clrsetbits_le32(®s->command1, 256 (QSPI_CMD1_LSBI_FE | QSPI_CMD1_LSBY_FE | 257 QSPI_CMD1_RX_EN | QSPI_CMD1_TX_EN), 258 (spi_chip_select(dev) << QSPI_CMD1_CS_SEL_SHIFT)); 259 260 /* set xfer size to 1 block (32 bits) */ 261 writel(0, ®s->dma_blk); 262 263 if (flags & SPI_XFER_BEGIN) 264 spi_cs_activate(dev); 265 266 /* handle data in 32-bit chunks */ 267 while (num_bytes > 0) { 268 int bytes; 269 270 tmpdout = 0; 271 bytes = (num_bytes > 4) ? 4 : num_bytes; 272 273 if (dout != NULL) { 274 memcpy((void *)&tmpdout, (void *)dout, bytes); 275 dout += bytes; 276 num_bytes -= bytes; 277 writel(tmpdout, ®s->tx_fifo); 278 setbits_le32(®s->command1, QSPI_CMD1_TX_EN); 279 } 280 281 if (din != NULL) 282 setbits_le32(®s->command1, QSPI_CMD1_RX_EN); 283 284 /* clear ready bit */ 285 setbits_le32(®s->xfer_status, QSPI_XFER_STS_RDY); 286 287 clrsetbits_le32(®s->command1, 288 QSPI_CMD1_BITLEN_MASK << QSPI_CMD1_BITLEN_SHIFT, 289 (bytes * 8 - 1) << QSPI_CMD1_BITLEN_SHIFT); 290 291 /* Need to stabilize other reg bits before GO bit set. 292 * As per the TRM: 293 * "For successful operation at various freq combinations, 294 * a minimum of 4-5 spi_clk cycle delay might be required 295 * before enabling the PIO or DMA bits. The worst case delay 296 * calculation can be done considering slowest qspi_clk as 297 * 1MHz. Based on that 1us delay should be enough before 298 * enabling PIO or DMA." Padded another 1us for safety. 299 */ 300 udelay(2); 301 setbits_le32(®s->command1, QSPI_CMD1_GO); 302 udelay(1); 303 304 /* 305 * Wait for SPI transmit FIFO to empty, or to time out. 306 * The RX FIFO status will be read and cleared last 307 */ 308 for (tm = 0; tm < QSPI_TIMEOUT; ++tm) { 309 u32 fifo_status, xfer_status; 310 311 xfer_status = readl(®s->xfer_status); 312 if (!(xfer_status & QSPI_XFER_STS_RDY)) 313 continue; 314 315 fifo_status = readl(®s->fifo_status); 316 if (fifo_status & QSPI_FIFO_STS_ERR) { 317 debug("%s: got a fifo error: ", __func__); 318 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_OVF) 319 debug("tx FIFO overflow "); 320 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_UNR) 321 debug("tx FIFO underrun "); 322 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_OVF) 323 debug("rx FIFO overflow "); 324 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_UNR) 325 debug("rx FIFO underrun "); 326 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_FULL) 327 debug("tx FIFO full "); 328 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_EMPTY) 329 debug("tx FIFO empty "); 330 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_FULL) 331 debug("rx FIFO full "); 332 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_EMPTY) 333 debug("rx FIFO empty "); 334 debug("\n"); 335 break; 336 } 337 338 if (!(fifo_status & QSPI_FIFO_STS_RX_FIFO_EMPTY)) { 339 tmpdin = readl(®s->rx_fifo); 340 if (din != NULL) { 341 memcpy(din, &tmpdin, bytes); 342 din += bytes; 343 num_bytes -= bytes; 344 } 345 } 346 break; 347 } 348 349 if (tm >= QSPI_TIMEOUT) 350 ret = tm; 351 352 /* clear ACK RDY, etc. bits */ 353 writel(readl(®s->fifo_status), ®s->fifo_status); 354 } 355 356 if (flags & SPI_XFER_END) 357 spi_cs_deactivate(dev); 358 359 debug("%s: transfer ended. Value=%08x, fifo_status = %08x\n", 360 __func__, tmpdin, readl(®s->fifo_status)); 361 362 if (ret) { 363 printf("%s: timeout during SPI transfer, tm %d\n", 364 __func__, ret); 365 return -1; 366 } 367 368 return ret; 369 } 370 371 static int tegra210_qspi_set_speed(struct udevice *bus, uint speed) 372 { 373 struct tegra_spi_platdata *plat = bus->platdata; 374 struct tegra210_qspi_priv *priv = dev_get_priv(bus); 375 376 if (speed > plat->frequency) 377 speed = plat->frequency; 378 priv->freq = speed; 379 debug("%s: regs=%p, speed=%d\n", __func__, priv->regs, priv->freq); 380 381 return 0; 382 } 383 384 static int tegra210_qspi_set_mode(struct udevice *bus, uint mode) 385 { 386 struct tegra210_qspi_priv *priv = dev_get_priv(bus); 387 388 priv->mode = mode; 389 debug("%s: regs=%p, mode=%d\n", __func__, priv->regs, priv->mode); 390 391 return 0; 392 } 393 394 static const struct dm_spi_ops tegra210_qspi_ops = { 395 .claim_bus = tegra210_qspi_claim_bus, 396 .xfer = tegra210_qspi_xfer, 397 .set_speed = tegra210_qspi_set_speed, 398 .set_mode = tegra210_qspi_set_mode, 399 /* 400 * cs_info is not needed, since we require all chip selects to be 401 * in the device tree explicitly 402 */ 403 }; 404 405 static const struct udevice_id tegra210_qspi_ids[] = { 406 { .compatible = "nvidia,tegra210-qspi" }, 407 { } 408 }; 409 410 U_BOOT_DRIVER(tegra210_qspi) = { 411 .name = "tegra210-qspi", 412 .id = UCLASS_SPI, 413 .of_match = tegra210_qspi_ids, 414 .ops = &tegra210_qspi_ops, 415 .ofdata_to_platdata = tegra210_qspi_ofdata_to_platdata, 416 .platdata_auto_alloc_size = sizeof(struct tegra_spi_platdata), 417 .priv_auto_alloc_size = sizeof(struct tegra210_qspi_priv), 418 .per_child_auto_alloc_size = sizeof(struct spi_slave), 419 .probe = tegra210_qspi_probe, 420 }; 421