355 lines
11 KiB
Diff
355 lines
11 KiB
Diff
diff -ruN a/drivers/spi/spi-geni-qcom.c b/drivers/spi/spi-geni-qcom.c
|
|
--- a/drivers/spi/spi-geni-qcom.c
|
|
+++ b/drivers/spi/spi-geni-qcom.c
|
|
@@ -2,6 +2,7 @@
|
|
// Copyright (c) 2017-2018, The Linux foundation. All rights reserved.
|
|
|
|
#include <linux/clk.h>
|
|
+#include <linux/delay.h>
|
|
#include <linux/dmaengine.h>
|
|
#include <linux/dma-mapping.h>
|
|
#include <linux/dma/qcom-gpi-dma.h>
|
|
@@ -75,9 +76,68 @@
|
|
#define GSI_CPHA BIT(4)
|
|
#define GSI_CPOL BIT(5)
|
|
|
|
+/* QSPI 1-4-4 support (added on top of the standard SPI controller). */
|
|
+#define QSPI_SE_PROTO 9
|
|
+
|
|
+#define GENI_IO_MUX_1_EN BIT(1)
|
|
+#define GENI_IO_MUX_2_EN BIT(2)
|
|
+#define GENI_IO_MUX_3_EN BIT(3)
|
|
+#define GENI_QSPI_IO_MUX_EN (GENI_IO_MUX_0_EN | GENI_IO_MUX_1_EN | \
|
|
+ GENI_IO_MUX_2_EN | GENI_IO_MUX_3_EN)
|
|
+
|
|
+#define SE_GSI_EVENT_EN 0xe18
|
|
+#define SE_IRQ_EN 0xe1c
|
|
+#define SE_DMA_TX_IRQ_CLR 0xc44
|
|
+#define SE_DMA_TX_IRQ_EN_SET 0xc4c
|
|
+#define SE_DMA_TX_IRQ_EN_CLR 0xc50
|
|
+#define SE_DMA_RX_IRQ_CLR 0xd44
|
|
+#define SE_DMA_RX_IRQ_EN_SET 0xd4c
|
|
+#define SE_DMA_RX_IRQ_EN_CLR 0xd50
|
|
+
|
|
+#define DMA_RX_EVENT_EN BIT(0)
|
|
+#define DMA_TX_EVENT_EN BIT(1)
|
|
+#define GENI_M_EVENT_EN BIT(2)
|
|
+#define GENI_S_EVENT_EN BIT(3)
|
|
+
|
|
+#define QSPI_M_IRQ_EN_GPI 0x33c00046
|
|
+#define QSPI_S_IRQ_EN_GPI 0x03001e06
|
|
+#define QSPI_DMA_TX_IRQ_EN 0x0d
|
|
+#define QSPI_DMA_RX_IRQ_EN 0x1d
|
|
+
|
|
+#define QSPI_SINGLE_SDR 0x000
|
|
+#define QSPI_QUAD_SDR BIT(9)
|
|
+
|
|
+/* Defaults for a quad read with 1-byte opcode + 3-byte address. */
|
|
+#define QSPI_DEFAULT_READ_OPCODE 0xEB
|
|
+#define QSPI_DEFAULT_DUMMY_CLK_CNT 8
|
|
+#define QSPI_DEFAULT_TX_CMD_LEN 4
|
|
+#define QSPI_DEFAULT_MAX_SPEED_HZ 20000000
|
|
+
|
|
+/**
|
|
+ * struct spi_geni_data - per-compatible behavioral flags
|
|
+ * @qspi_mode: controller runs in QSPI 1-4-4 mode with 4 data lanes
|
|
+ */
|
|
+struct spi_geni_data {
|
|
+ bool qspi_mode;
|
|
+};
|
|
+
|
|
+/**
|
|
+ * struct spi_geni_qspi_params - per-SE QSPI tunables read from DT
|
|
+ * @read_opcode: first TX byte that identifies a read transfer (e.g. 0xEB)
|
|
+ * @dummy_clk_cnt: dummy clocks inserted between address and read data
|
|
+ * @tx_cmd_len: number of TX bytes forming the read command (opcode+address)
|
|
+ */
|
|
+struct spi_geni_qspi_params {
|
|
+ u32 read_opcode;
|
|
+ u32 dummy_clk_cnt;
|
|
+ u32 tx_cmd_len;
|
|
+};
|
|
+
|
|
struct spi_geni_master {
|
|
struct geni_se se;
|
|
struct device *dev;
|
|
+ const struct spi_geni_data *data;
|
|
+ struct spi_geni_qspi_params qspi;
|
|
u32 tx_fifo_depth;
|
|
u32 fifo_width_bits;
|
|
u32 tx_wm;
|
|
@@ -104,6 +164,52 @@
|
|
int cur_xfer_mode;
|
|
};
|
|
|
|
+static inline bool spi_geni_is_qspi(const struct spi_geni_master *mas)
|
|
+{
|
|
+ return mas->data && mas->data->qspi_mode;
|
|
+}
|
|
+
|
|
+/* Enable all 4 data lanes on the GENI output mux for QSPI. */
|
|
+static void qspi_setup_io_mux(struct spi_geni_master *mas)
|
|
+{
|
|
+ struct geni_se *se = &mas->se;
|
|
+ u32 out;
|
|
+
|
|
+ out = readl(se->base + GENI_OUTPUT_CTRL);
|
|
+ out |= GENI_QSPI_IO_MUX_EN;
|
|
+ writel(out, se->base + GENI_OUTPUT_CTRL);
|
|
+}
|
|
+
|
|
+/*
|
|
+ * Reprogram the SE IRQ / DMA / event registers for GPI DMA.
|
|
+ *
|
|
+ * geni_se_resources_on() (called from runtime_resume) writes a fixed set of
|
|
+ * values into SE_IRQ_EN that are correct for FIFO/SE_DMA mode but clobber the
|
|
+ * GPI configuration. The QSPI SE_PROTO (9) also needs different masks than
|
|
+ * the defaults. Call this from prepare_message (post resume, pre transfer)
|
|
+ * to restore the GPI-friendly values.
|
|
+ */
|
|
+static void prep_se_for_gpi_dma(struct spi_geni_master *mas)
|
|
+{
|
|
+ void __iomem *base = mas->se.base;
|
|
+
|
|
+ writel(GENI_DMA_MODE_EN, base + SE_GENI_DMA_MODE_EN);
|
|
+ writel(0, base + SE_IRQ_EN);
|
|
+ writel(DMA_RX_EVENT_EN | DMA_TX_EVENT_EN |
|
|
+ GENI_M_EVENT_EN | GENI_S_EVENT_EN,
|
|
+ base + SE_GSI_EVENT_EN);
|
|
+ writel(QSPI_M_IRQ_EN_GPI, base + SE_GENI_M_IRQ_EN);
|
|
+ writel(QSPI_S_IRQ_EN_GPI, base + SE_GENI_S_IRQ_EN);
|
|
+ writel(0xf, base + SE_DMA_TX_IRQ_EN_CLR);
|
|
+ writel(QSPI_DMA_TX_IRQ_EN, base + SE_DMA_TX_IRQ_EN_SET);
|
|
+ writel(0xfff, base + SE_DMA_RX_IRQ_EN_CLR);
|
|
+ writel(QSPI_DMA_RX_IRQ_EN, base + SE_DMA_RX_IRQ_EN_SET);
|
|
+ writel(0xffc07fff, base + SE_GENI_M_IRQ_CLEAR);
|
|
+ writel(0x0fc07f3f, base + SE_GENI_S_IRQ_CLEAR);
|
|
+ writel(0xf, base + SE_DMA_TX_IRQ_CLR);
|
|
+ writel(0xfff, base + SE_DMA_RX_IRQ_CLR);
|
|
+}
|
|
+
|
|
static void spi_slv_setup(struct spi_geni_master *mas)
|
|
{
|
|
struct geni_se *se = &mas->se;
|
|
@@ -411,7 +517,20 @@
|
|
}
|
|
|
|
if (xfer->tx_buf && xfer->rx_buf) {
|
|
- peripheral.cmd = SPI_DUPLEX;
|
|
+ /*
|
|
+ * QSPI uses SPI_TX_RX (7) for TX-opcode-then-RX-data transfers;
|
|
+ * standard SPI uses SPI_DUPLEX (3) for true full-duplex.
|
|
+ *
|
|
+ * For QSPI_TX_RX the GSI firmware needs an explicit rx_len so
|
|
+ * it knows how many bytes to clock in after the TX command
|
|
+ * phase. SPI_DUPLEX uses xfer->len implicitly.
|
|
+ */
|
|
+ if (spi_geni_is_qspi(mas)) {
|
|
+ peripheral.cmd = SPI_TX_RX;
|
|
+ peripheral.rx_len = (xfer->len << 3) / xfer->bits_per_word;
|
|
+ } else {
|
|
+ peripheral.cmd = SPI_DUPLEX;
|
|
+ }
|
|
} else if (xfer->tx_buf) {
|
|
peripheral.cmd = SPI_TX;
|
|
peripheral.rx_len = 0;
|
|
@@ -445,6 +564,44 @@
|
|
peripheral.fragmentation = FRAGMENTATION;
|
|
}
|
|
|
|
+ if (spi_geni_is_qspi(mas)) {
|
|
+ bool multi = !list_is_singular(&spi->cur_msg->transfers);
|
|
+
|
|
+ peripheral.qspi_mode = true;
|
|
+
|
|
+ /*
|
|
+ * In a multi-transfer message the first write uses SINGLE_SDR
|
|
+ * (opcode+addr lane transition) while subsequent TX-only
|
|
+ * transfers stay in QUAD.
|
|
+ */
|
|
+ if (multi && xfer->tx_buf && !xfer->rx_buf &&
|
|
+ &xfer->transfer_list == spi->cur_msg->transfers.next)
|
|
+ peripheral.qspi_lane_flags = QSPI_SINGLE_SDR;
|
|
+ else
|
|
+ peripheral.qspi_lane_flags = QSPI_QUAD_SDR;
|
|
+
|
|
+ if (peripheral.cmd == SPI_TX_RX && xfer->tx_buf) {
|
|
+ const u8 *tx = xfer->tx_buf;
|
|
+
|
|
+ /*
|
|
+ * The configured read opcode is followed by an address
|
|
+ * then dummy clocks before the device drives the data
|
|
+ * lanes. Any other first byte means the host is issuing
|
|
+ * a write command, so downgrade to TX-only.
|
|
+ */
|
|
+ if (tx[0] == mas->qspi.read_opcode) {
|
|
+ peripheral.dummy_clk_cnt = mas->qspi.dummy_clk_cnt;
|
|
+ peripheral.tx_cmd_len = mas->qspi.tx_cmd_len;
|
|
+ } else {
|
|
+ peripheral.cmd = SPI_TX;
|
|
+ peripheral.dummy_clk_cnt = 0;
|
|
+ peripheral.rx_len = 0;
|
|
+ }
|
|
+ } else if (peripheral.cmd == SPI_RX) {
|
|
+ peripheral.dummy_clk_cnt = mas->qspi.dummy_clk_cnt;
|
|
+ }
|
|
+ }
|
|
+
|
|
if (peripheral.cmd & SPI_RX) {
|
|
dmaengine_slave_config(mas->rx, &config);
|
|
rx_desc = dmaengine_prep_slave_sg(mas->rx, xfer->rx_sg.sgl, xfer->rx_sg.nents,
|
|
@@ -467,8 +624,18 @@
|
|
return -EIO;
|
|
}
|
|
|
|
- tx_desc->callback_result = spi_gsi_callback_result;
|
|
- tx_desc->callback_param = spi;
|
|
+ /*
|
|
+ * In QSPI mode the Go TRE on the TX channel has no DMA TRE for SPI_RX
|
|
+ * (and for TX_RX completes TX-side early), so the real completion
|
|
+ * event is the IEOT on the RX channel. Attach the callback there.
|
|
+ */
|
|
+ if (spi_geni_is_qspi(mas) && (peripheral.cmd & SPI_RX)) {
|
|
+ rx_desc->callback_result = spi_gsi_callback_result;
|
|
+ rx_desc->callback_param = spi;
|
|
+ } else {
|
|
+ tx_desc->callback_result = spi_gsi_callback_result;
|
|
+ tx_desc->callback_param = spi;
|
|
+ }
|
|
|
|
if (peripheral.cmd & SPI_RX)
|
|
dmaengine_submit(rx_desc);
|
|
@@ -534,7 +701,13 @@
|
|
return ret;
|
|
|
|
case GENI_GPI_DMA:
|
|
- /* nothing to do for GPI DMA */
|
|
+ /*
|
|
+ * In QSPI mode, runtime_resume's call to geni_se_resources_on()
|
|
+ * clobbers the GPI-specific IRQ/DMA register layout. Restore
|
|
+ * it before every message.
|
|
+ */
|
|
+ if (spi_geni_is_qspi(mas))
|
|
+ prep_se_for_gpi_dma(mas);
|
|
return 0;
|
|
}
|
|
|
|
@@ -609,6 +782,18 @@
|
|
goto out_pm;
|
|
}
|
|
spi_slv_setup(mas);
|
|
+ } else if (spi_geni_is_qspi(mas)) {
|
|
+ /*
|
|
+ * QSPI SEs report protocol 9 in hardware. The GENI_SE_SPI
|
|
+ * firmware loader cannot be used here (no firmware for proto 9
|
|
+ * is shipped), so reject anything else.
|
|
+ */
|
|
+ if (proto != QSPI_SE_PROTO) {
|
|
+ dev_err(mas->dev, "Expected QSPI proto %d, got %d\n",
|
|
+ QSPI_SE_PROTO, proto);
|
|
+ goto out_pm;
|
|
+ }
|
|
+ qspi_setup_io_mux(mas);
|
|
} else if (proto == GENI_SE_INVALID_PROTO) {
|
|
ret = geni_load_se_firmware(se, GENI_SE_SPI);
|
|
if (ret) {
|
|
@@ -640,9 +825,28 @@
|
|
else
|
|
mas->oversampling = 1;
|
|
|
|
+ /*
|
|
+ * QSPI SEs cannot use FIFO mode (the FIFO path would try to drive
|
|
+ * a single-lane word format on a 4-lane QSPI bus), so force GPI DMA
|
|
+ * regardless of what GENI_IF_DISABLE_RO reports.
|
|
+ */
|
|
fifo_disable = readl(se->base + GENI_IF_DISABLE_RO) & FIFO_IF_DISABLE;
|
|
+ if (spi_geni_is_qspi(mas))
|
|
+ fifo_disable = 1;
|
|
+
|
|
switch (fifo_disable) {
|
|
case 1:
|
|
+ /*
|
|
+ * For QSPI, bring the SE up in SE_DMA first (which arms the
|
|
+ * DMA-related registers) and let it settle before switching
|
|
+ * to GPI_DMA and grabbing the GPII channels. Without this
|
|
+ * intermediate step the first GPI command after probe can
|
|
+ * hang on the CH STOP completion.
|
|
+ */
|
|
+ if (spi_geni_is_qspi(mas)) {
|
|
+ geni_se_select_mode(se, GENI_SE_DMA);
|
|
+ msleep(10);
|
|
+ }
|
|
ret = spi_geni_grab_gpi_chan(mas);
|
|
if (!ret) { /* success case */
|
|
mas->cur_xfer_mode = GENI_GPI_DMA;
|
|
@@ -653,6 +857,16 @@
|
|
goto out_pm;
|
|
}
|
|
/*
|
|
+ * For QSPI there is no usable FIFO fallback: FIFO mode cannot
|
|
+ * drive a 4-lane QSPI bus. Fail the probe instead of silently
|
|
+ * producing garbage.
|
|
+ */
|
|
+ if (spi_geni_is_qspi(mas)) {
|
|
+ dev_err(mas->dev, "Failed to grab GPI DMA channels for QSPI: %d\n",
|
|
+ ret);
|
|
+ goto out_pm;
|
|
+ }
|
|
+ /*
|
|
* in case of failure to get gpi dma channel, we can still do the
|
|
* FIFO mode, so fallthrough
|
|
*/
|
|
@@ -1052,11 +1266,24 @@
|
|
mas = spi_controller_get_devdata(spi);
|
|
mas->irq = irq;
|
|
mas->dev = dev;
|
|
+ mas->data = device_get_match_data(dev);
|
|
mas->se.dev = dev;
|
|
mas->se.wrapper = dev_get_drvdata(dev->parent);
|
|
mas->se.base = base;
|
|
mas->se.clk = clk;
|
|
|
|
+ if (spi_geni_is_qspi(mas)) {
|
|
+ mas->qspi.read_opcode = QSPI_DEFAULT_READ_OPCODE;
|
|
+ mas->qspi.dummy_clk_cnt = QSPI_DEFAULT_DUMMY_CLK_CNT;
|
|
+ mas->qspi.tx_cmd_len = QSPI_DEFAULT_TX_CMD_LEN;
|
|
+ device_property_read_u32(dev, "qcom,qspi-read-opcode",
|
|
+ &mas->qspi.read_opcode);
|
|
+ device_property_read_u32(dev, "qcom,qspi-read-dummy-clocks",
|
|
+ &mas->qspi.dummy_clk_cnt);
|
|
+ device_property_read_u32(dev, "qcom,qspi-read-cmd-bytes",
|
|
+ &mas->qspi.tx_cmd_len);
|
|
+ }
|
|
+
|
|
ret = devm_pm_opp_set_clkname(&pdev->dev, "se");
|
|
if (ret)
|
|
return ret;
|
|
@@ -1069,9 +1296,12 @@
|
|
|
|
spi->bus_num = -1;
|
|
spi->mode_bits = SPI_CPOL | SPI_CPHA | SPI_LOOP | SPI_CS_HIGH;
|
|
+ if (spi_geni_is_qspi(mas))
|
|
+ spi->mode_bits |= SPI_TX_QUAD | SPI_RX_QUAD;
|
|
spi->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 32);
|
|
spi->num_chipselect = 4;
|
|
- spi->max_speed_hz = 50000000;
|
|
+ spi->max_speed_hz = spi_geni_is_qspi(mas) ? QSPI_DEFAULT_MAX_SPEED_HZ
|
|
+ : 50000000;
|
|
spi->max_dma_len = 0xffff0; /* 24 bits for tx/rx dma length */
|
|
spi->prepare_message = spi_geni_prepare_message;
|
|
spi->transfer_one = spi_geni_transfer_one;
|
|
@@ -1197,8 +1427,13 @@
|
|
SET_SYSTEM_SLEEP_PM_OPS(spi_geni_suspend, spi_geni_resume)
|
|
};
|
|
|
|
+static const struct spi_geni_data spi_geni_qspi_data = {
|
|
+ .qspi_mode = true,
|
|
+};
|
|
+
|
|
static const struct of_device_id spi_geni_dt_match[] = {
|
|
{ .compatible = "qcom,geni-spi" },
|
|
+ { .compatible = "qcom,geni-spi-qspi", .data = &spi_geni_qspi_data },
|
|
{}
|
|
};
|
|
MODULE_DEVICE_TABLE(of, spi_geni_dt_match);
|