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