1 // SPDX-License-Identifier: GPL-2.0+
3 * NVIDIA Tegra210 QSPI controller driver
5 * (C) Copyright 2015-2020 NVIDIA Corporation <www.nvidia.com>
14 #include <asm/arch/clock.h>
15 #include <asm/arch-tegra/clk_rst.h>
18 #include <linux/delay.h>
19 #include "tegra_spi.h"
21 DECLARE_GLOBAL_DATA_PTR;
24 #define QSPI_CMD1_GO BIT(31)
25 #define QSPI_CMD1_M_S BIT(30)
26 #define QSPI_CMD1_MODE_MASK GENMASK(1,0)
27 #define QSPI_CMD1_MODE_SHIFT 28
28 #define QSPI_CMD1_CS_SEL_MASK GENMASK(1,0)
29 #define QSPI_CMD1_CS_SEL_SHIFT 26
30 #define QSPI_CMD1_CS_POL_INACTIVE0 BIT(22)
31 #define QSPI_CMD1_CS_SW_HW BIT(21)
32 #define QSPI_CMD1_CS_SW_VAL BIT(20)
33 #define QSPI_CMD1_IDLE_SDA_MASK GENMASK(1,0)
34 #define QSPI_CMD1_IDLE_SDA_SHIFT 18
35 #define QSPI_CMD1_BIDIR BIT(17)
36 #define QSPI_CMD1_LSBI_FE BIT(16)
37 #define QSPI_CMD1_LSBY_FE BIT(15)
38 #define QSPI_CMD1_BOTH_EN_BIT BIT(14)
39 #define QSPI_CMD1_BOTH_EN_BYTE BIT(13)
40 #define QSPI_CMD1_RX_EN BIT(12)
41 #define QSPI_CMD1_TX_EN BIT(11)
42 #define QSPI_CMD1_PACKED BIT(5)
43 #define QSPI_CMD1_BITLEN_MASK GENMASK(4,0)
44 #define QSPI_CMD1_BITLEN_SHIFT 0
47 #define QSPI_CMD2_TX_CLK_TAP_DELAY_SHIFT 10
48 #define QSPI_CMD2_TX_CLK_TAP_DELAY_MASK GENMASK(14,10)
49 #define QSPI_CMD2_RX_CLK_TAP_DELAY_SHIFT 0
50 #define QSPI_CMD2_RX_CLK_TAP_DELAY_MASK GENMASK(7,0)
53 #define QSPI_XFER_STS_RDY BIT(30)
56 #define QSPI_FIFO_STS_CS_INACTIVE BIT(31)
57 #define QSPI_FIFO_STS_FRAME_END BIT(30)
58 #define QSPI_FIFO_STS_RX_FIFO_FLUSH BIT(15)
59 #define QSPI_FIFO_STS_TX_FIFO_FLUSH BIT(14)
60 #define QSPI_FIFO_STS_ERR BIT(8)
61 #define QSPI_FIFO_STS_TX_FIFO_OVF BIT(7)
62 #define QSPI_FIFO_STS_TX_FIFO_UNR BIT(6)
63 #define QSPI_FIFO_STS_RX_FIFO_OVF BIT(5)
64 #define QSPI_FIFO_STS_RX_FIFO_UNR BIT(4)
65 #define QSPI_FIFO_STS_TX_FIFO_FULL BIT(3)
66 #define QSPI_FIFO_STS_TX_FIFO_EMPTY BIT(2)
67 #define QSPI_FIFO_STS_RX_FIFO_FULL BIT(1)
68 #define QSPI_FIFO_STS_RX_FIFO_EMPTY BIT(0)
70 #define QSPI_TIMEOUT 1000
73 u32 command1; /* 000:QSPI_COMMAND1 register */
74 u32 command2; /* 004:QSPI_COMMAND2 register */
75 u32 timing1; /* 008:QSPI_CS_TIM1 register */
76 u32 timing2; /* 00c:QSPI_CS_TIM2 register */
77 u32 xfer_status;/* 010:QSPI_TRANS_STATUS register */
78 u32 fifo_status;/* 014:QSPI_FIFO_STATUS register */
79 u32 tx_data; /* 018:QSPI_TX_DATA register */
80 u32 rx_data; /* 01c:QSPI_RX_DATA register */
81 u32 dma_ctl; /* 020:QSPI_DMA_CTL register */
82 u32 dma_blk; /* 024:QSPI_DMA_BLK register */
83 u32 rsvd[56]; /* 028-107 reserved */
84 u32 tx_fifo; /* 108:QSPI_FIFO1 register */
85 u32 rsvd2[31]; /* 10c-187 reserved */
86 u32 rx_fifo; /* 188:QSPI_FIFO2 register */
87 u32 spare_ctl; /* 18c:QSPI_SPARE_CTRL register */
90 struct tegra210_qspi_priv {
91 struct qspi_regs *regs;
96 int last_transaction_us;
99 static int tegra210_qspi_ofdata_to_platdata(struct udevice *bus)
101 struct tegra_spi_platdata *plat = bus->platdata;
103 plat->base = dev_read_addr(bus);
104 plat->periph_id = clock_decode_periph_id(bus);
106 if (plat->periph_id == PERIPH_ID_NONE) {
107 debug("%s: could not decode periph id %d\n", __func__,
109 return -FDT_ERR_NOTFOUND;
112 /* Use 500KHz as a suitable default */
113 plat->frequency = dev_read_u32_default(bus, "spi-max-frequency",
115 plat->deactivate_delay_us = dev_read_u32_default(bus,
116 "spi-deactivate-delay",
118 debug("%s: base=%#08lx, periph_id=%d, max-frequency=%d, deactivate_delay=%d\n",
119 __func__, plat->base, plat->periph_id, plat->frequency,
120 plat->deactivate_delay_us);
125 static int tegra210_qspi_probe(struct udevice *bus)
127 struct tegra_spi_platdata *plat = dev_get_platdata(bus);
128 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
130 priv->regs = (struct qspi_regs *)plat->base;
131 struct qspi_regs *regs = priv->regs;
133 priv->last_transaction_us = timer_get_us();
134 priv->freq = plat->frequency;
135 priv->periph_id = plat->periph_id;
137 debug("%s: Freq = %u, id = %d\n", __func__, priv->freq,
139 /* Change SPI clock to correct frequency, PLLP_OUT0 source */
140 clock_start_periph_pll(priv->periph_id, CLOCK_ID_PERIPH, priv->freq);
142 /* Set tap delays here, clock change above resets QSPI controller */
143 u32 reg = (0x09 << QSPI_CMD2_TX_CLK_TAP_DELAY_SHIFT) |
144 (0x0C << QSPI_CMD2_RX_CLK_TAP_DELAY_SHIFT);
145 writel(reg, ®s->command2);
146 debug("%s: COMMAND2 = %08x\n", __func__, readl(®s->command2));
151 static int tegra210_qspi_claim_bus(struct udevice *dev)
153 struct udevice *bus = dev->parent;
154 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
155 struct qspi_regs *regs = priv->regs;
157 debug("%s: FIFO STATUS = %08x\n", __func__, readl(®s->fifo_status));
159 /* Set master mode and sw controlled CS */
160 setbits_le32(®s->command1, QSPI_CMD1_M_S | QSPI_CMD1_CS_SW_HW |
161 (priv->mode << QSPI_CMD1_MODE_SHIFT));
162 debug("%s: COMMAND1 = %08x\n", __func__, readl(®s->command1));
168 * Activate the CS by driving it LOW
170 * @param slave Pointer to spi_slave to which controller has to
173 static void spi_cs_activate(struct udevice *dev)
175 struct udevice *bus = dev->parent;
176 struct tegra_spi_platdata *pdata = dev_get_platdata(bus);
177 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
179 /* If it's too soon to do another transaction, wait */
180 if (pdata->deactivate_delay_us &&
181 priv->last_transaction_us) {
182 ulong delay_us; /* The delay completed so far */
183 delay_us = timer_get_us() - priv->last_transaction_us;
184 if (delay_us < pdata->deactivate_delay_us)
185 udelay(pdata->deactivate_delay_us - delay_us);
188 clrbits_le32(&priv->regs->command1, QSPI_CMD1_CS_SW_VAL);
192 * Deactivate the CS by driving it HIGH
194 * @param slave Pointer to spi_slave to which controller has to
197 static void spi_cs_deactivate(struct udevice *dev)
199 struct udevice *bus = dev->parent;
200 struct tegra_spi_platdata *pdata = dev_get_platdata(bus);
201 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
203 setbits_le32(&priv->regs->command1, QSPI_CMD1_CS_SW_VAL);
205 /* Remember time of this transaction so we can honour the bus delay */
206 if (pdata->deactivate_delay_us)
207 priv->last_transaction_us = timer_get_us();
209 debug("Deactivate CS, bus '%s'\n", bus->name);
212 static int tegra210_qspi_xfer(struct udevice *dev, unsigned int bitlen,
213 const void *data_out, void *data_in,
216 struct udevice *bus = dev->parent;
217 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
218 struct qspi_regs *regs = priv->regs;
219 u32 reg, tmpdout, tmpdin = 0;
220 const u8 *dout = data_out;
222 int num_bytes, tm, ret;
224 debug("%s: slave %u:%u dout %p din %p bitlen %u\n",
225 __func__, bus->seq, spi_chip_select(dev), dout, din, bitlen);
228 num_bytes = bitlen / 8;
232 /* clear all error status bits */
233 reg = readl(®s->fifo_status);
234 writel(reg, ®s->fifo_status);
236 /* flush RX/TX FIFOs */
237 setbits_le32(®s->fifo_status,
238 (QSPI_FIFO_STS_RX_FIFO_FLUSH |
239 QSPI_FIFO_STS_TX_FIFO_FLUSH));
242 while ((tm && readl(®s->fifo_status) &
243 (QSPI_FIFO_STS_RX_FIFO_FLUSH |
244 QSPI_FIFO_STS_TX_FIFO_FLUSH))) {
250 printf("%s: timeout during QSPI FIFO flush!\n",
257 * 1. don't set LSBY_FE, so no need to swap bytes from/to TX/RX FIFOs;
258 * 2. don't set RX_EN and TX_EN yet.
259 * (SW needs to make sure that while programming the blk_size,
260 * tx_en and rx_en bits must be zero)
261 * [TODO] I (Yen Lin) have problems when both RX/TX EN bits are set
262 * i.e., both dout and din are not NULL.
264 clrsetbits_le32(®s->command1,
265 (QSPI_CMD1_LSBI_FE | QSPI_CMD1_LSBY_FE |
266 QSPI_CMD1_RX_EN | QSPI_CMD1_TX_EN),
267 (spi_chip_select(dev) << QSPI_CMD1_CS_SEL_SHIFT));
269 /* set xfer size to 1 block (32 bits) */
270 writel(0, ®s->dma_blk);
272 if (flags & SPI_XFER_BEGIN)
273 spi_cs_activate(dev);
275 /* handle data in 32-bit chunks */
276 while (num_bytes > 0) {
280 bytes = (num_bytes > 4) ? 4 : num_bytes;
283 memcpy((void *)&tmpdout, (void *)dout, bytes);
286 writel(tmpdout, ®s->tx_fifo);
287 setbits_le32(®s->command1, QSPI_CMD1_TX_EN);
291 setbits_le32(®s->command1, QSPI_CMD1_RX_EN);
293 /* clear ready bit */
294 setbits_le32(®s->xfer_status, QSPI_XFER_STS_RDY);
296 clrsetbits_le32(®s->command1,
297 QSPI_CMD1_BITLEN_MASK << QSPI_CMD1_BITLEN_SHIFT,
298 (bytes * 8 - 1) << QSPI_CMD1_BITLEN_SHIFT);
300 /* Need to stabilize other reg bits before GO bit set.
302 * "For successful operation at various freq combinations,
303 * a minimum of 4-5 spi_clk cycle delay might be required
304 * before enabling the PIO or DMA bits. The worst case delay
305 * calculation can be done considering slowest qspi_clk as
306 * 1MHz. Based on that 1us delay should be enough before
307 * enabling PIO or DMA." Padded another 1us for safety.
310 setbits_le32(®s->command1, QSPI_CMD1_GO);
314 * Wait for SPI transmit FIFO to empty, or to time out.
315 * The RX FIFO status will be read and cleared last
317 for (tm = 0; tm < QSPI_TIMEOUT; ++tm) {
318 u32 fifo_status, xfer_status;
320 xfer_status = readl(®s->xfer_status);
321 if (!(xfer_status & QSPI_XFER_STS_RDY))
324 fifo_status = readl(®s->fifo_status);
325 if (fifo_status & QSPI_FIFO_STS_ERR) {
326 debug("%s: got a fifo error: ", __func__);
327 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_OVF)
328 debug("tx FIFO overflow ");
329 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_UNR)
330 debug("tx FIFO underrun ");
331 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_OVF)
332 debug("rx FIFO overflow ");
333 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_UNR)
334 debug("rx FIFO underrun ");
335 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_FULL)
336 debug("tx FIFO full ");
337 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_EMPTY)
338 debug("tx FIFO empty ");
339 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_FULL)
340 debug("rx FIFO full ");
341 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_EMPTY)
342 debug("rx FIFO empty ");
347 if (!(fifo_status & QSPI_FIFO_STS_RX_FIFO_EMPTY)) {
348 tmpdin = readl(®s->rx_fifo);
350 memcpy(din, &tmpdin, bytes);
358 if (tm >= QSPI_TIMEOUT)
361 /* clear ACK RDY, etc. bits */
362 writel(readl(®s->fifo_status), ®s->fifo_status);
365 if (flags & SPI_XFER_END)
366 spi_cs_deactivate(dev);
368 debug("%s: transfer ended. Value=%08x, fifo_status = %08x\n",
369 __func__, tmpdin, readl(®s->fifo_status));
372 printf("%s: timeout during SPI transfer, tm %d\n",
380 static int tegra210_qspi_set_speed(struct udevice *bus, uint speed)
382 struct tegra_spi_platdata *plat = bus->platdata;
383 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
385 if (speed > plat->frequency)
386 speed = plat->frequency;
388 debug("%s: regs=%p, speed=%d\n", __func__, priv->regs, priv->freq);
393 static int tegra210_qspi_set_mode(struct udevice *bus, uint mode)
395 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
398 debug("%s: regs=%p, mode=%d\n", __func__, priv->regs, priv->mode);
403 static const struct dm_spi_ops tegra210_qspi_ops = {
404 .claim_bus = tegra210_qspi_claim_bus,
405 .xfer = tegra210_qspi_xfer,
406 .set_speed = tegra210_qspi_set_speed,
407 .set_mode = tegra210_qspi_set_mode,
409 * cs_info is not needed, since we require all chip selects to be
410 * in the device tree explicitly
414 static const struct udevice_id tegra210_qspi_ids[] = {
415 { .compatible = "nvidia,tegra210-qspi" },
419 U_BOOT_DRIVER(tegra210_qspi) = {
420 .name = "tegra210-qspi",
422 .of_match = tegra210_qspi_ids,
423 .ops = &tegra210_qspi_ops,
424 .ofdata_to_platdata = tegra210_qspi_ofdata_to_platdata,
425 .platdata_auto_alloc_size = sizeof(struct tegra_spi_platdata),
426 .priv_auto_alloc_size = sizeof(struct tegra210_qspi_priv),
427 .per_child_auto_alloc_size = sizeof(struct spi_slave),
428 .probe = tegra210_qspi_probe,