1 // SPDX-License-Identifier: GPL-2.0+
3 * NVIDIA Tegra210 QSPI controller driver
5 * (C) Copyright 2015-2020 NVIDIA Corporation <www.nvidia.com>
13 #include <asm/global_data.h>
15 #include <asm/arch/clock.h>
16 #include <asm/arch-tegra/clk_rst.h>
19 #include <linux/bitops.h>
20 #include <linux/delay.h>
21 #include "tegra_spi.h"
23 DECLARE_GLOBAL_DATA_PTR;
26 #define QSPI_CMD1_GO BIT(31)
27 #define QSPI_CMD1_M_S BIT(30)
28 #define QSPI_CMD1_MODE_MASK GENMASK(1,0)
29 #define QSPI_CMD1_MODE_SHIFT 28
30 #define QSPI_CMD1_CS_SEL_MASK GENMASK(1,0)
31 #define QSPI_CMD1_CS_SEL_SHIFT 26
32 #define QSPI_CMD1_CS_POL_INACTIVE0 BIT(22)
33 #define QSPI_CMD1_CS_SW_HW BIT(21)
34 #define QSPI_CMD1_CS_SW_VAL BIT(20)
35 #define QSPI_CMD1_IDLE_SDA_MASK GENMASK(1,0)
36 #define QSPI_CMD1_IDLE_SDA_SHIFT 18
37 #define QSPI_CMD1_BIDIR BIT(17)
38 #define QSPI_CMD1_LSBI_FE BIT(16)
39 #define QSPI_CMD1_LSBY_FE BIT(15)
40 #define QSPI_CMD1_BOTH_EN_BIT BIT(14)
41 #define QSPI_CMD1_BOTH_EN_BYTE BIT(13)
42 #define QSPI_CMD1_RX_EN BIT(12)
43 #define QSPI_CMD1_TX_EN BIT(11)
44 #define QSPI_CMD1_PACKED BIT(5)
45 #define QSPI_CMD1_BITLEN_MASK GENMASK(4,0)
46 #define QSPI_CMD1_BITLEN_SHIFT 0
49 #define QSPI_CMD2_TX_CLK_TAP_DELAY_SHIFT 10
50 #define QSPI_CMD2_TX_CLK_TAP_DELAY_MASK GENMASK(14,10)
51 #define QSPI_CMD2_RX_CLK_TAP_DELAY_SHIFT 0
52 #define QSPI_CMD2_RX_CLK_TAP_DELAY_MASK GENMASK(7,0)
55 #define QSPI_XFER_STS_RDY BIT(30)
58 #define QSPI_FIFO_STS_CS_INACTIVE BIT(31)
59 #define QSPI_FIFO_STS_FRAME_END BIT(30)
60 #define QSPI_FIFO_STS_RX_FIFO_FLUSH BIT(15)
61 #define QSPI_FIFO_STS_TX_FIFO_FLUSH BIT(14)
62 #define QSPI_FIFO_STS_ERR BIT(8)
63 #define QSPI_FIFO_STS_TX_FIFO_OVF BIT(7)
64 #define QSPI_FIFO_STS_TX_FIFO_UNR BIT(6)
65 #define QSPI_FIFO_STS_RX_FIFO_OVF BIT(5)
66 #define QSPI_FIFO_STS_RX_FIFO_UNR BIT(4)
67 #define QSPI_FIFO_STS_TX_FIFO_FULL BIT(3)
68 #define QSPI_FIFO_STS_TX_FIFO_EMPTY BIT(2)
69 #define QSPI_FIFO_STS_RX_FIFO_FULL BIT(1)
70 #define QSPI_FIFO_STS_RX_FIFO_EMPTY BIT(0)
72 #define QSPI_TIMEOUT 1000
75 u32 command1; /* 000:QSPI_COMMAND1 register */
76 u32 command2; /* 004:QSPI_COMMAND2 register */
77 u32 timing1; /* 008:QSPI_CS_TIM1 register */
78 u32 timing2; /* 00c:QSPI_CS_TIM2 register */
79 u32 xfer_status;/* 010:QSPI_TRANS_STATUS register */
80 u32 fifo_status;/* 014:QSPI_FIFO_STATUS register */
81 u32 tx_data; /* 018:QSPI_TX_DATA register */
82 u32 rx_data; /* 01c:QSPI_RX_DATA register */
83 u32 dma_ctl; /* 020:QSPI_DMA_CTL register */
84 u32 dma_blk; /* 024:QSPI_DMA_BLK register */
85 u32 rsvd[56]; /* 028-107 reserved */
86 u32 tx_fifo; /* 108:QSPI_FIFO1 register */
87 u32 rsvd2[31]; /* 10c-187 reserved */
88 u32 rx_fifo; /* 188:QSPI_FIFO2 register */
89 u32 spare_ctl; /* 18c:QSPI_SPARE_CTRL register */
92 struct tegra210_qspi_priv {
93 struct qspi_regs *regs;
98 int last_transaction_us;
101 static int tegra210_qspi_of_to_plat(struct udevice *bus)
103 struct tegra_spi_plat *plat = dev_get_plat(bus);
105 plat->base = dev_read_addr(bus);
106 plat->periph_id = clock_decode_periph_id(bus);
108 if (plat->periph_id == PERIPH_ID_NONE) {
109 debug("%s: could not decode periph id %d\n", __func__,
111 return -FDT_ERR_NOTFOUND;
114 /* Use 500KHz as a suitable default */
115 plat->frequency = dev_read_u32_default(bus, "spi-max-frequency",
117 plat->deactivate_delay_us = dev_read_u32_default(bus,
118 "spi-deactivate-delay",
120 debug("%s: base=%#08lx, periph_id=%d, max-frequency=%d, deactivate_delay=%d\n",
121 __func__, plat->base, plat->periph_id, plat->frequency,
122 plat->deactivate_delay_us);
127 static int tegra210_qspi_probe(struct udevice *bus)
129 struct tegra_spi_plat *plat = dev_get_plat(bus);
130 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
132 priv->regs = (struct qspi_regs *)plat->base;
133 struct qspi_regs *regs = priv->regs;
135 priv->last_transaction_us = timer_get_us();
136 priv->freq = plat->frequency;
137 priv->periph_id = plat->periph_id;
139 debug("%s: Freq = %u, id = %d\n", __func__, priv->freq,
141 /* Change SPI clock to correct frequency, PLLP_OUT0 source */
142 clock_start_periph_pll(priv->periph_id, CLOCK_ID_PERIPH, priv->freq);
144 /* Set tap delays here, clock change above resets QSPI controller */
145 u32 reg = (0x09 << QSPI_CMD2_TX_CLK_TAP_DELAY_SHIFT) |
146 (0x0C << QSPI_CMD2_RX_CLK_TAP_DELAY_SHIFT);
147 writel(reg, ®s->command2);
148 debug("%s: COMMAND2 = %08x\n", __func__, readl(®s->command2));
153 static int tegra210_qspi_claim_bus(struct udevice *dev)
155 struct udevice *bus = dev->parent;
156 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
157 struct qspi_regs *regs = priv->regs;
159 debug("%s: FIFO STATUS = %08x\n", __func__, readl(®s->fifo_status));
161 /* Set master mode and sw controlled CS */
162 setbits_le32(®s->command1, QSPI_CMD1_M_S | QSPI_CMD1_CS_SW_HW |
163 (priv->mode << QSPI_CMD1_MODE_SHIFT));
164 debug("%s: COMMAND1 = %08x\n", __func__, readl(®s->command1));
170 * Activate the CS by driving it LOW
172 * @param slave Pointer to spi_slave to which controller has to
175 static void spi_cs_activate(struct udevice *dev)
177 struct udevice *bus = dev->parent;
178 struct tegra_spi_plat *pdata = dev_get_plat(bus);
179 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
181 /* If it's too soon to do another transaction, wait */
182 if (pdata->deactivate_delay_us &&
183 priv->last_transaction_us) {
184 ulong delay_us; /* The delay completed so far */
185 delay_us = timer_get_us() - priv->last_transaction_us;
186 if (delay_us < pdata->deactivate_delay_us)
187 udelay(pdata->deactivate_delay_us - delay_us);
190 clrbits_le32(&priv->regs->command1, QSPI_CMD1_CS_SW_VAL);
194 * Deactivate the CS by driving it HIGH
196 * @param slave Pointer to spi_slave to which controller has to
199 static void spi_cs_deactivate(struct udevice *dev)
201 struct udevice *bus = dev->parent;
202 struct tegra_spi_plat *pdata = dev_get_plat(bus);
203 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
205 setbits_le32(&priv->regs->command1, QSPI_CMD1_CS_SW_VAL);
207 /* Remember time of this transaction so we can honour the bus delay */
208 if (pdata->deactivate_delay_us)
209 priv->last_transaction_us = timer_get_us();
211 debug("Deactivate CS, bus '%s'\n", bus->name);
214 static int tegra210_qspi_xfer(struct udevice *dev, unsigned int bitlen,
215 const void *data_out, void *data_in,
218 struct udevice *bus = dev->parent;
219 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
220 struct qspi_regs *regs = priv->regs;
221 u32 reg, tmpdout, tmpdin = 0;
222 const u8 *dout = data_out;
224 int num_bytes, tm, ret;
226 debug("%s: slave %u:%u dout %p din %p bitlen %u\n",
227 __func__, dev_seq(bus), spi_chip_select(dev), dout, din, bitlen);
230 num_bytes = bitlen / 8;
234 /* clear all error status bits */
235 reg = readl(®s->fifo_status);
236 writel(reg, ®s->fifo_status);
238 /* flush RX/TX FIFOs */
239 setbits_le32(®s->fifo_status,
240 (QSPI_FIFO_STS_RX_FIFO_FLUSH |
241 QSPI_FIFO_STS_TX_FIFO_FLUSH));
244 while ((tm && readl(®s->fifo_status) &
245 (QSPI_FIFO_STS_RX_FIFO_FLUSH |
246 QSPI_FIFO_STS_TX_FIFO_FLUSH))) {
252 printf("%s: timeout during QSPI FIFO flush!\n",
259 * 1. don't set LSBY_FE, so no need to swap bytes from/to TX/RX FIFOs;
260 * 2. don't set RX_EN and TX_EN yet.
261 * (SW needs to make sure that while programming the blk_size,
262 * tx_en and rx_en bits must be zero)
263 * [TODO] I (Yen Lin) have problems when both RX/TX EN bits are set
264 * i.e., both dout and din are not NULL.
266 clrsetbits_le32(®s->command1,
267 (QSPI_CMD1_LSBI_FE | QSPI_CMD1_LSBY_FE |
268 QSPI_CMD1_RX_EN | QSPI_CMD1_TX_EN),
269 (spi_chip_select(dev) << QSPI_CMD1_CS_SEL_SHIFT));
271 /* set xfer size to 1 block (32 bits) */
272 writel(0, ®s->dma_blk);
274 if (flags & SPI_XFER_BEGIN)
275 spi_cs_activate(dev);
277 /* handle data in 32-bit chunks */
278 while (num_bytes > 0) {
282 bytes = (num_bytes > 4) ? 4 : num_bytes;
285 memcpy((void *)&tmpdout, (void *)dout, bytes);
288 writel(tmpdout, ®s->tx_fifo);
289 setbits_le32(®s->command1, QSPI_CMD1_TX_EN);
293 setbits_le32(®s->command1, QSPI_CMD1_RX_EN);
295 /* clear ready bit */
296 setbits_le32(®s->xfer_status, QSPI_XFER_STS_RDY);
298 clrsetbits_le32(®s->command1,
299 QSPI_CMD1_BITLEN_MASK << QSPI_CMD1_BITLEN_SHIFT,
300 (bytes * 8 - 1) << QSPI_CMD1_BITLEN_SHIFT);
302 /* Need to stabilize other reg bits before GO bit set.
304 * "For successful operation at various freq combinations,
305 * a minimum of 4-5 spi_clk cycle delay might be required
306 * before enabling the PIO or DMA bits. The worst case delay
307 * calculation can be done considering slowest qspi_clk as
308 * 1MHz. Based on that 1us delay should be enough before
309 * enabling PIO or DMA." Padded another 1us for safety.
312 setbits_le32(®s->command1, QSPI_CMD1_GO);
316 * Wait for SPI transmit FIFO to empty, or to time out.
317 * The RX FIFO status will be read and cleared last
319 for (tm = 0; tm < QSPI_TIMEOUT; ++tm) {
320 u32 fifo_status, xfer_status;
322 xfer_status = readl(®s->xfer_status);
323 if (!(xfer_status & QSPI_XFER_STS_RDY))
326 fifo_status = readl(®s->fifo_status);
327 if (fifo_status & QSPI_FIFO_STS_ERR) {
328 debug("%s: got a fifo error: ", __func__);
329 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_OVF)
330 debug("tx FIFO overflow ");
331 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_UNR)
332 debug("tx FIFO underrun ");
333 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_OVF)
334 debug("rx FIFO overflow ");
335 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_UNR)
336 debug("rx FIFO underrun ");
337 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_FULL)
338 debug("tx FIFO full ");
339 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_EMPTY)
340 debug("tx FIFO empty ");
341 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_FULL)
342 debug("rx FIFO full ");
343 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_EMPTY)
344 debug("rx FIFO empty ");
349 if (!(fifo_status & QSPI_FIFO_STS_RX_FIFO_EMPTY)) {
350 tmpdin = readl(®s->rx_fifo);
352 memcpy(din, &tmpdin, bytes);
360 if (tm >= QSPI_TIMEOUT)
363 /* clear ACK RDY, etc. bits */
364 writel(readl(®s->fifo_status), ®s->fifo_status);
367 if (flags & SPI_XFER_END)
368 spi_cs_deactivate(dev);
370 debug("%s: transfer ended. Value=%08x, fifo_status = %08x\n",
371 __func__, tmpdin, readl(®s->fifo_status));
374 printf("%s: timeout during SPI transfer, tm %d\n",
382 static int tegra210_qspi_set_speed(struct udevice *bus, uint speed)
384 struct tegra_spi_plat *plat = dev_get_plat(bus);
385 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
387 if (speed > plat->frequency)
388 speed = plat->frequency;
390 debug("%s: regs=%p, speed=%d\n", __func__, priv->regs, priv->freq);
395 static int tegra210_qspi_set_mode(struct udevice *bus, uint mode)
397 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
400 debug("%s: regs=%p, mode=%d\n", __func__, priv->regs, priv->mode);
405 static const struct dm_spi_ops tegra210_qspi_ops = {
406 .claim_bus = tegra210_qspi_claim_bus,
407 .xfer = tegra210_qspi_xfer,
408 .set_speed = tegra210_qspi_set_speed,
409 .set_mode = tegra210_qspi_set_mode,
411 * cs_info is not needed, since we require all chip selects to be
412 * in the device tree explicitly
416 static const struct udevice_id tegra210_qspi_ids[] = {
417 { .compatible = "nvidia,tegra210-qspi" },
421 U_BOOT_DRIVER(tegra210_qspi) = {
422 .name = "tegra210-qspi",
424 .of_match = tegra210_qspi_ids,
425 .ops = &tegra210_qspi_ops,
426 .of_to_plat = tegra210_qspi_of_to_plat,
427 .plat_auto = sizeof(struct tegra_spi_plat),
428 .priv_auto = sizeof(struct tegra210_qspi_priv),
429 .per_child_auto = sizeof(struct spi_slave),
430 .probe = tegra210_qspi_probe,