1 // SPDX-License-Identifier: GPL-2.0+
3 * NVIDIA Tegra210 QSPI controller driver
5 * (C) Copyright 2015-2019 NVIDIA Corporation <www.nvidia.com>
13 #include <asm/arch/clock.h>
14 #include <asm/arch-tegra/clk_rst.h>
17 #include "tegra_spi.h"
19 DECLARE_GLOBAL_DATA_PTR;
22 #define QSPI_CMD1_GO BIT(31)
23 #define QSPI_CMD1_M_S BIT(30)
24 #define QSPI_CMD1_MODE_MASK GENMASK(1,0)
25 #define QSPI_CMD1_MODE_SHIFT 28
26 #define QSPI_CMD1_CS_SEL_MASK GENMASK(1,0)
27 #define QSPI_CMD1_CS_SEL_SHIFT 26
28 #define QSPI_CMD1_CS_POL_INACTIVE0 BIT(22)
29 #define QSPI_CMD1_CS_SW_HW BIT(21)
30 #define QSPI_CMD1_CS_SW_VAL BIT(20)
31 #define QSPI_CMD1_IDLE_SDA_MASK GENMASK(1,0)
32 #define QSPI_CMD1_IDLE_SDA_SHIFT 18
33 #define QSPI_CMD1_BIDIR BIT(17)
34 #define QSPI_CMD1_LSBI_FE BIT(16)
35 #define QSPI_CMD1_LSBY_FE BIT(15)
36 #define QSPI_CMD1_BOTH_EN_BIT BIT(14)
37 #define QSPI_CMD1_BOTH_EN_BYTE BIT(13)
38 #define QSPI_CMD1_RX_EN BIT(12)
39 #define QSPI_CMD1_TX_EN BIT(11)
40 #define QSPI_CMD1_PACKED BIT(5)
41 #define QSPI_CMD1_BITLEN_MASK GENMASK(4,0)
42 #define QSPI_CMD1_BITLEN_SHIFT 0
45 #define QSPI_CMD2_TX_CLK_TAP_DELAY BIT(6)
46 #define QSPI_CMD2_TX_CLK_TAP_DELAY_MASK GENMASK(11,6)
47 #define QSPI_CMD2_RX_CLK_TAP_DELAY BIT(0)
48 #define QSPI_CMD2_RX_CLK_TAP_DELAY_MASK GENMASK(5,0)
51 #define QSPI_XFER_STS_RDY BIT(30)
54 #define QSPI_FIFO_STS_CS_INACTIVE BIT(31)
55 #define QSPI_FIFO_STS_FRAME_END BIT(30)
56 #define QSPI_FIFO_STS_RX_FIFO_FLUSH BIT(15)
57 #define QSPI_FIFO_STS_TX_FIFO_FLUSH BIT(14)
58 #define QSPI_FIFO_STS_ERR BIT(8)
59 #define QSPI_FIFO_STS_TX_FIFO_OVF BIT(7)
60 #define QSPI_FIFO_STS_TX_FIFO_UNR BIT(6)
61 #define QSPI_FIFO_STS_RX_FIFO_OVF BIT(5)
62 #define QSPI_FIFO_STS_RX_FIFO_UNR BIT(4)
63 #define QSPI_FIFO_STS_TX_FIFO_FULL BIT(3)
64 #define QSPI_FIFO_STS_TX_FIFO_EMPTY BIT(2)
65 #define QSPI_FIFO_STS_RX_FIFO_FULL BIT(1)
66 #define QSPI_FIFO_STS_RX_FIFO_EMPTY BIT(0)
68 #define QSPI_TIMEOUT 1000
71 u32 command1; /* 000:QSPI_COMMAND1 register */
72 u32 command2; /* 004:QSPI_COMMAND2 register */
73 u32 timing1; /* 008:QSPI_CS_TIM1 register */
74 u32 timing2; /* 00c:QSPI_CS_TIM2 register */
75 u32 xfer_status;/* 010:QSPI_TRANS_STATUS register */
76 u32 fifo_status;/* 014:QSPI_FIFO_STATUS register */
77 u32 tx_data; /* 018:QSPI_TX_DATA register */
78 u32 rx_data; /* 01c:QSPI_RX_DATA register */
79 u32 dma_ctl; /* 020:QSPI_DMA_CTL register */
80 u32 dma_blk; /* 024:QSPI_DMA_BLK register */
81 u32 rsvd[56]; /* 028-107 reserved */
82 u32 tx_fifo; /* 108:QSPI_FIFO1 register */
83 u32 rsvd2[31]; /* 10c-187 reserved */
84 u32 rx_fifo; /* 188:QSPI_FIFO2 register */
85 u32 spare_ctl; /* 18c:QSPI_SPARE_CTRL register */
88 struct tegra210_qspi_priv {
89 struct qspi_regs *regs;
94 int last_transaction_us;
97 static int tegra210_qspi_ofdata_to_platdata(struct udevice *bus)
99 struct tegra_spi_platdata *plat = bus->platdata;
100 const void *blob = gd->fdt_blob;
101 int node = dev_of_offset(bus);
103 plat->base = devfdt_get_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 = fdtdec_get_int(blob, node, "spi-max-frequency",
115 plat->deactivate_delay_us = fdtdec_get_int(blob, node,
116 "spi-deactivate-delay", 0);
117 debug("%s: base=%#08lx, periph_id=%d, max-frequency=%d, deactivate_delay=%d\n",
118 __func__, plat->base, plat->periph_id, plat->frequency,
119 plat->deactivate_delay_us);
124 static int tegra210_qspi_probe(struct udevice *bus)
126 struct tegra_spi_platdata *plat = dev_get_platdata(bus);
127 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
129 priv->regs = (struct qspi_regs *)plat->base;
131 priv->last_transaction_us = timer_get_us();
132 priv->freq = plat->frequency;
133 priv->periph_id = plat->periph_id;
135 /* Change SPI clock to correct frequency, PLLP_OUT0 source */
136 clock_start_periph_pll(priv->periph_id, CLOCK_ID_PERIPH, priv->freq);
141 static int tegra210_qspi_claim_bus(struct udevice *dev)
143 struct udevice *bus = dev->parent;
144 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
145 struct qspi_regs *regs = priv->regs;
147 /* Change SPI clock to correct frequency, PLLP_OUT0 source */
148 clock_start_periph_pll(priv->periph_id, CLOCK_ID_PERIPH, priv->freq);
150 debug("%s: FIFO STATUS = %08x\n", __func__, readl(®s->fifo_status));
152 /* Set master mode and sw controlled CS */
153 setbits_le32(®s->command1, QSPI_CMD1_M_S | QSPI_CMD1_CS_SW_HW |
154 (priv->mode << QSPI_CMD1_MODE_SHIFT));
155 debug("%s: COMMAND1 = %08x\n", __func__, readl(®s->command1));
161 * Activate the CS by driving it LOW
163 * @param slave Pointer to spi_slave to which controller has to
166 static void spi_cs_activate(struct udevice *dev)
168 struct udevice *bus = dev->parent;
169 struct tegra_spi_platdata *pdata = dev_get_platdata(bus);
170 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
172 /* If it's too soon to do another transaction, wait */
173 if (pdata->deactivate_delay_us &&
174 priv->last_transaction_us) {
175 ulong delay_us; /* The delay completed so far */
176 delay_us = timer_get_us() - priv->last_transaction_us;
177 if (delay_us < pdata->deactivate_delay_us)
178 udelay(pdata->deactivate_delay_us - delay_us);
181 clrbits_le32(&priv->regs->command1, QSPI_CMD1_CS_SW_VAL);
185 * Deactivate the CS by driving it HIGH
187 * @param slave Pointer to spi_slave to which controller has to
190 static void spi_cs_deactivate(struct udevice *dev)
192 struct udevice *bus = dev->parent;
193 struct tegra_spi_platdata *pdata = dev_get_platdata(bus);
194 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
196 setbits_le32(&priv->regs->command1, QSPI_CMD1_CS_SW_VAL);
198 /* Remember time of this transaction so we can honour the bus delay */
199 if (pdata->deactivate_delay_us)
200 priv->last_transaction_us = timer_get_us();
202 debug("Deactivate CS, bus '%s'\n", bus->name);
205 static int tegra210_qspi_xfer(struct udevice *dev, unsigned int bitlen,
206 const void *data_out, void *data_in,
209 struct udevice *bus = dev->parent;
210 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
211 struct qspi_regs *regs = priv->regs;
212 u32 reg, tmpdout, tmpdin = 0;
213 const u8 *dout = data_out;
215 int num_bytes, tm, ret;
217 debug("%s: slave %u:%u dout %p din %p bitlen %u\n",
218 __func__, bus->seq, spi_chip_select(dev), dout, din, bitlen);
221 num_bytes = bitlen / 8;
225 /* clear all error status bits */
226 reg = readl(®s->fifo_status);
227 writel(reg, ®s->fifo_status);
229 /* flush RX/TX FIFOs */
230 setbits_le32(®s->fifo_status,
231 (QSPI_FIFO_STS_RX_FIFO_FLUSH |
232 QSPI_FIFO_STS_TX_FIFO_FLUSH));
235 while ((tm && readl(®s->fifo_status) &
236 (QSPI_FIFO_STS_RX_FIFO_FLUSH |
237 QSPI_FIFO_STS_TX_FIFO_FLUSH))) {
243 printf("%s: timeout during QSPI FIFO flush!\n",
250 * 1. don't set LSBY_FE, so no need to swap bytes from/to TX/RX FIFOs;
251 * 2. don't set RX_EN and TX_EN yet.
252 * (SW needs to make sure that while programming the blk_size,
253 * tx_en and rx_en bits must be zero)
254 * [TODO] I (Yen Lin) have problems when both RX/TX EN bits are set
255 * i.e., both dout and din are not NULL.
257 clrsetbits_le32(®s->command1,
258 (QSPI_CMD1_LSBI_FE | QSPI_CMD1_LSBY_FE |
259 QSPI_CMD1_RX_EN | QSPI_CMD1_TX_EN),
260 (spi_chip_select(dev) << QSPI_CMD1_CS_SEL_SHIFT));
262 /* set xfer size to 1 block (32 bits) */
263 writel(0, ®s->dma_blk);
265 if (flags & SPI_XFER_BEGIN)
266 spi_cs_activate(dev);
268 /* handle data in 32-bit chunks */
269 while (num_bytes > 0) {
273 bytes = (num_bytes > 4) ? 4 : num_bytes;
276 memcpy((void *)&tmpdout, (void *)dout, bytes);
279 writel(tmpdout, ®s->tx_fifo);
280 setbits_le32(®s->command1, QSPI_CMD1_TX_EN);
284 setbits_le32(®s->command1, QSPI_CMD1_RX_EN);
286 /* clear ready bit */
287 setbits_le32(®s->xfer_status, QSPI_XFER_STS_RDY);
289 clrsetbits_le32(®s->command1,
290 QSPI_CMD1_BITLEN_MASK << QSPI_CMD1_BITLEN_SHIFT,
291 (bytes * 8 - 1) << QSPI_CMD1_BITLEN_SHIFT);
293 /* Need to stabilize other reg bits before GO bit set.
295 * "For successful operation at various freq combinations,
296 * a minimum of 4-5 spi_clk cycle delay might be required
297 * before enabling the PIO or DMA bits. The worst case delay
298 * calculation can be done considering slowest qspi_clk as
299 * 1MHz. Based on that 1us delay should be enough before
300 * enabling PIO or DMA." Padded another 1us for safety.
303 setbits_le32(®s->command1, QSPI_CMD1_GO);
307 * Wait for SPI transmit FIFO to empty, or to time out.
308 * The RX FIFO status will be read and cleared last
310 for (tm = 0; tm < QSPI_TIMEOUT; ++tm) {
311 u32 fifo_status, xfer_status;
313 xfer_status = readl(®s->xfer_status);
314 if (!(xfer_status & QSPI_XFER_STS_RDY))
317 fifo_status = readl(®s->fifo_status);
318 if (fifo_status & QSPI_FIFO_STS_ERR) {
319 debug("%s: got a fifo error: ", __func__);
320 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_OVF)
321 debug("tx FIFO overflow ");
322 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_UNR)
323 debug("tx FIFO underrun ");
324 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_OVF)
325 debug("rx FIFO overflow ");
326 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_UNR)
327 debug("rx FIFO underrun ");
328 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_FULL)
329 debug("tx FIFO full ");
330 if (fifo_status & QSPI_FIFO_STS_TX_FIFO_EMPTY)
331 debug("tx FIFO empty ");
332 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_FULL)
333 debug("rx FIFO full ");
334 if (fifo_status & QSPI_FIFO_STS_RX_FIFO_EMPTY)
335 debug("rx FIFO empty ");
340 if (!(fifo_status & QSPI_FIFO_STS_RX_FIFO_EMPTY)) {
341 tmpdin = readl(®s->rx_fifo);
343 memcpy(din, &tmpdin, bytes);
351 if (tm >= QSPI_TIMEOUT)
354 /* clear ACK RDY, etc. bits */
355 writel(readl(®s->fifo_status), ®s->fifo_status);
358 if (flags & SPI_XFER_END)
359 spi_cs_deactivate(dev);
361 debug("%s: transfer ended. Value=%08x, fifo_status = %08x\n",
362 __func__, tmpdin, readl(®s->fifo_status));
365 printf("%s: timeout during SPI transfer, tm %d\n",
373 static int tegra210_qspi_set_speed(struct udevice *bus, uint speed)
375 struct tegra_spi_platdata *plat = bus->platdata;
376 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
378 if (speed > plat->frequency)
379 speed = plat->frequency;
381 debug("%s: regs=%p, speed=%d\n", __func__, priv->regs, priv->freq);
386 static int tegra210_qspi_set_mode(struct udevice *bus, uint mode)
388 struct tegra210_qspi_priv *priv = dev_get_priv(bus);
391 debug("%s: regs=%p, mode=%d\n", __func__, priv->regs, priv->mode);
396 static const struct dm_spi_ops tegra210_qspi_ops = {
397 .claim_bus = tegra210_qspi_claim_bus,
398 .xfer = tegra210_qspi_xfer,
399 .set_speed = tegra210_qspi_set_speed,
400 .set_mode = tegra210_qspi_set_mode,
402 * cs_info is not needed, since we require all chip selects to be
403 * in the device tree explicitly
407 static const struct udevice_id tegra210_qspi_ids[] = {
408 { .compatible = "nvidia,tegra210-qspi" },
412 U_BOOT_DRIVER(tegra210_qspi) = {
413 .name = "tegra210-qspi",
415 .of_match = tegra210_qspi_ids,
416 .ops = &tegra210_qspi_ops,
417 .ofdata_to_platdata = tegra210_qspi_ofdata_to_platdata,
418 .platdata_auto_alloc_size = sizeof(struct tegra_spi_platdata),
419 .priv_auto_alloc_size = sizeof(struct tegra210_qspi_priv),
420 .per_child_auto_alloc_size = sizeof(struct spi_slave),
421 .probe = tegra210_qspi_probe,