1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright (c) 2017-2018, The Linux foundation. All rights reserved.
5 #include <linux/dmaengine.h>
6 #include <linux/dma-mapping.h>
7 #include <linux/dma/qcom-gpi-dma.h>
8 #include <linux/interrupt.h>
10 #include <linux/log2.h>
11 #include <linux/module.h>
12 #include <linux/platform_device.h>
13 #include <linux/pm_opp.h>
14 #include <linux/pm_runtime.h>
15 #include <linux/soc/qcom/geni-se.h>
16 #include <linux/spi/spi.h>
17 #include <linux/spinlock.h>
19 /* SPI SE specific registers and respective register fields */
20 #define SE_SPI_CPHA 0x224
23 #define SE_SPI_LOOPBACK 0x22c
24 #define LOOPBACK_ENABLE 0x1
25 #define NORMAL_MODE 0x0
26 #define LOOPBACK_MSK GENMASK(1, 0)
28 #define SE_SPI_CPOL 0x230
31 #define SE_SPI_DEMUX_OUTPUT_INV 0x24c
32 #define CS_DEMUX_OUTPUT_INV_MSK GENMASK(3, 0)
34 #define SE_SPI_DEMUX_SEL 0x250
35 #define CS_DEMUX_OUTPUT_SEL GENMASK(3, 0)
37 #define SE_SPI_TRANS_CFG 0x25c
38 #define CS_TOGGLE BIT(0)
40 #define SE_SPI_WORD_LEN 0x268
41 #define WORD_LEN_MSK GENMASK(9, 0)
42 #define MIN_WORD_LEN 4
44 #define SE_SPI_TX_TRANS_LEN 0x26c
45 #define SE_SPI_RX_TRANS_LEN 0x270
46 #define TRANS_LEN_MSK GENMASK(23, 0)
48 #define SE_SPI_PRE_POST_CMD_DLY 0x274
50 #define SE_SPI_DELAY_COUNTERS 0x278
51 #define SPI_INTER_WORDS_DELAY_MSK GENMASK(9, 0)
52 #define SPI_CS_CLK_DELAY_MSK GENMASK(19, 10)
53 #define SPI_CS_CLK_DELAY_SHFT 10
55 /* M_CMD OP codes for SPI */
59 #define SPI_CS_ASSERT 8
60 #define SPI_CS_DEASSERT 9
61 #define SPI_SCK_ONLY 10
62 /* M_CMD params for SPI */
63 #define SPI_PRE_CMD_DELAY BIT(0)
64 #define TIMESTAMP_BEFORE BIT(1)
65 #define FRAGMENTATION BIT(2)
66 #define TIMESTAMP_AFTER BIT(3)
67 #define POST_CMD_DELAY BIT(4)
69 #define GSI_LOOPBACK_EN BIT(0)
70 #define GSI_CS_TOGGLE BIT(3)
71 #define GSI_CPHA BIT(4)
72 #define GSI_CPOL BIT(5)
74 struct spi_geni_master {
81 unsigned long cur_speed_hz;
82 unsigned long cur_sclk_hz;
83 unsigned int cur_bits_per_word;
84 unsigned int tx_rem_bytes;
85 unsigned int rx_rem_bytes;
86 const struct spi_transfer *cur_xfer;
87 struct completion cs_done;
88 struct completion cancel_done;
89 struct completion abort_done;
90 struct completion tx_reset_done;
91 struct completion rx_reset_done;
92 unsigned int oversampling;
100 dma_addr_t tx_se_dma;
101 dma_addr_t rx_se_dma;
104 static int get_spi_clk_cfg(unsigned int speed_hz,
105 struct spi_geni_master *mas,
106 unsigned int *clk_idx,
107 unsigned int *clk_div)
109 unsigned long sclk_freq;
110 unsigned int actual_hz;
113 ret = geni_se_clk_freq_match(&mas->se,
114 speed_hz * mas->oversampling,
115 clk_idx, &sclk_freq, false);
117 dev_err(mas->dev, "Failed(%d) to find src clk for %dHz\n",
122 *clk_div = DIV_ROUND_UP(sclk_freq, mas->oversampling * speed_hz);
123 actual_hz = sclk_freq / (mas->oversampling * *clk_div);
125 dev_dbg(mas->dev, "req %u=>%u sclk %lu, idx %d, div %d\n", speed_hz,
126 actual_hz, sclk_freq, *clk_idx, *clk_div);
127 ret = dev_pm_opp_set_rate(mas->dev, sclk_freq);
129 dev_err(mas->dev, "dev_pm_opp_set_rate failed %d\n", ret);
131 mas->cur_sclk_hz = sclk_freq;
136 static void handle_se_timeout(struct spi_master *spi,
137 struct spi_message *msg)
139 struct spi_geni_master *mas = spi_master_get_devdata(spi);
140 unsigned long time_left;
141 struct geni_se *se = &mas->se;
142 const struct spi_transfer *xfer;
144 spin_lock_irq(&mas->lock);
145 reinit_completion(&mas->cancel_done);
146 if (mas->cur_xfer_mode == GENI_SE_FIFO)
147 writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
149 xfer = mas->cur_xfer;
150 mas->cur_xfer = NULL;
151 geni_se_cancel_m_cmd(se);
152 spin_unlock_irq(&mas->lock);
154 time_left = wait_for_completion_timeout(&mas->cancel_done, HZ);
158 spin_lock_irq(&mas->lock);
159 reinit_completion(&mas->abort_done);
160 geni_se_abort_m_cmd(se);
161 spin_unlock_irq(&mas->lock);
163 time_left = wait_for_completion_timeout(&mas->abort_done, HZ);
165 dev_err(mas->dev, "Failed to cancel/abort m_cmd\n");
168 * No need for a lock since SPI core has a lock and we never
169 * access this from an interrupt.
171 mas->abort_failed = true;
175 if (mas->cur_xfer_mode == GENI_SE_DMA) {
177 if (xfer->tx_buf && mas->tx_se_dma) {
178 spin_lock_irq(&mas->lock);
179 reinit_completion(&mas->tx_reset_done);
180 writel(1, se->base + SE_DMA_TX_FSM_RST);
181 spin_unlock_irq(&mas->lock);
182 time_left = wait_for_completion_timeout(&mas->tx_reset_done, HZ);
184 dev_err(mas->dev, "DMA TX RESET failed\n");
185 geni_se_tx_dma_unprep(se, mas->tx_se_dma, xfer->len);
187 if (xfer->rx_buf && mas->rx_se_dma) {
188 spin_lock_irq(&mas->lock);
189 reinit_completion(&mas->rx_reset_done);
190 writel(1, se->base + SE_DMA_RX_FSM_RST);
191 spin_unlock_irq(&mas->lock);
192 time_left = wait_for_completion_timeout(&mas->rx_reset_done, HZ);
194 dev_err(mas->dev, "DMA RX RESET failed\n");
195 geni_se_rx_dma_unprep(se, mas->rx_se_dma, xfer->len);
199 * This can happen if a timeout happened and we had to wait
200 * for lock in this function because isr was holding the lock
201 * and handling transfer completion at that time.
203 dev_warn(mas->dev, "Cancel/Abort on completed SPI transfer\n");
208 static void handle_gpi_timeout(struct spi_master *spi, struct spi_message *msg)
210 struct spi_geni_master *mas = spi_master_get_devdata(spi);
212 dmaengine_terminate_sync(mas->tx);
213 dmaengine_terminate_sync(mas->rx);
216 static void spi_geni_handle_err(struct spi_master *spi, struct spi_message *msg)
218 struct spi_geni_master *mas = spi_master_get_devdata(spi);
220 switch (mas->cur_xfer_mode) {
223 handle_se_timeout(spi, msg);
226 handle_gpi_timeout(spi, msg);
229 dev_err(mas->dev, "Abort on Mode:%d not supported", mas->cur_xfer_mode);
233 static bool spi_geni_is_abort_still_pending(struct spi_geni_master *mas)
235 struct geni_se *se = &mas->se;
238 if (!mas->abort_failed)
242 * The only known case where a transfer times out and then a cancel
243 * times out then an abort times out is if something is blocking our
244 * interrupt handler from running. Avoid starting any new transfers
245 * until that sorts itself out.
247 spin_lock_irq(&mas->lock);
248 m_irq = readl(se->base + SE_GENI_M_IRQ_STATUS);
249 m_irq_en = readl(se->base + SE_GENI_M_IRQ_EN);
250 spin_unlock_irq(&mas->lock);
252 if (m_irq & m_irq_en) {
253 dev_err(mas->dev, "Interrupts pending after abort: %#010x\n",
259 * If we're here the problem resolved itself so no need to check more
260 * on future transfers.
262 mas->abort_failed = false;
267 static void spi_geni_set_cs(struct spi_device *slv, bool set_flag)
269 struct spi_geni_master *mas = spi_master_get_devdata(slv->master);
270 struct spi_master *spi = dev_get_drvdata(mas->dev);
271 struct geni_se *se = &mas->se;
272 unsigned long time_left;
274 if (!(slv->mode & SPI_CS_HIGH))
275 set_flag = !set_flag;
277 if (set_flag == mas->cs_flag)
280 pm_runtime_get_sync(mas->dev);
282 if (spi_geni_is_abort_still_pending(mas)) {
283 dev_err(mas->dev, "Can't set chip select\n");
287 spin_lock_irq(&mas->lock);
289 dev_err(mas->dev, "Can't set CS when prev xfer running\n");
290 spin_unlock_irq(&mas->lock);
294 mas->cs_flag = set_flag;
295 /* set xfer_mode to FIFO to complete cs_done in isr */
296 mas->cur_xfer_mode = GENI_SE_FIFO;
297 reinit_completion(&mas->cs_done);
299 geni_se_setup_m_cmd(se, SPI_CS_ASSERT, 0);
301 geni_se_setup_m_cmd(se, SPI_CS_DEASSERT, 0);
302 spin_unlock_irq(&mas->lock);
304 time_left = wait_for_completion_timeout(&mas->cs_done, HZ);
306 dev_warn(mas->dev, "Timeout setting chip select\n");
307 handle_se_timeout(spi, NULL);
311 pm_runtime_put(mas->dev);
314 static void spi_setup_word_len(struct spi_geni_master *mas, u16 mode,
315 unsigned int bits_per_word)
317 unsigned int pack_words;
318 bool msb_first = (mode & SPI_LSB_FIRST) ? false : true;
319 struct geni_se *se = &mas->se;
323 * If bits_per_word isn't a byte aligned value, set the packing to be
324 * 1 SPI word per FIFO word.
326 if (!(mas->fifo_width_bits % bits_per_word))
327 pack_words = mas->fifo_width_bits / bits_per_word;
330 geni_se_config_packing(&mas->se, bits_per_word, pack_words, msb_first,
332 word_len = (bits_per_word - MIN_WORD_LEN) & WORD_LEN_MSK;
333 writel(word_len, se->base + SE_SPI_WORD_LEN);
336 static int geni_spi_set_clock_and_bw(struct spi_geni_master *mas,
337 unsigned long clk_hz)
339 u32 clk_sel, m_clk_cfg, idx, div;
340 struct geni_se *se = &mas->se;
343 if (clk_hz == mas->cur_speed_hz)
346 ret = get_spi_clk_cfg(clk_hz, mas, &idx, &div);
348 dev_err(mas->dev, "Err setting clk to %lu: %d\n", clk_hz, ret);
353 * SPI core clock gets configured with the requested frequency
354 * or the frequency closer to the requested frequency.
355 * For that reason requested frequency is stored in the
356 * cur_speed_hz and referred in the consecutive transfer instead
357 * of calling clk_get_rate() API.
359 mas->cur_speed_hz = clk_hz;
361 clk_sel = idx & CLK_SEL_MSK;
362 m_clk_cfg = (div << CLK_DIV_SHFT) | SER_CLK_EN;
363 writel(clk_sel, se->base + SE_GENI_CLK_SEL);
364 writel(m_clk_cfg, se->base + GENI_SER_M_CLK_CFG);
366 /* Set BW quota for CPU as driver supports FIFO mode only. */
367 se->icc_paths[CPU_TO_GENI].avg_bw = Bps_to_icc(mas->cur_speed_hz);
368 ret = geni_icc_set_bw(se);
375 static int setup_fifo_params(struct spi_device *spi_slv,
376 struct spi_master *spi)
378 struct spi_geni_master *mas = spi_master_get_devdata(spi);
379 struct geni_se *se = &mas->se;
380 u32 loopback_cfg = 0, cpol = 0, cpha = 0, demux_output_inv = 0;
383 if (mas->last_mode != spi_slv->mode) {
384 if (spi_slv->mode & SPI_LOOP)
385 loopback_cfg = LOOPBACK_ENABLE;
387 if (spi_slv->mode & SPI_CPOL)
390 if (spi_slv->mode & SPI_CPHA)
393 if (spi_slv->mode & SPI_CS_HIGH)
394 demux_output_inv = BIT(spi_get_chipselect(spi_slv, 0));
396 demux_sel = spi_get_chipselect(spi_slv, 0);
397 mas->cur_bits_per_word = spi_slv->bits_per_word;
399 spi_setup_word_len(mas, spi_slv->mode, spi_slv->bits_per_word);
400 writel(loopback_cfg, se->base + SE_SPI_LOOPBACK);
401 writel(demux_sel, se->base + SE_SPI_DEMUX_SEL);
402 writel(cpha, se->base + SE_SPI_CPHA);
403 writel(cpol, se->base + SE_SPI_CPOL);
404 writel(demux_output_inv, se->base + SE_SPI_DEMUX_OUTPUT_INV);
406 mas->last_mode = spi_slv->mode;
409 return geni_spi_set_clock_and_bw(mas, spi_slv->max_speed_hz);
413 spi_gsi_callback_result(void *cb, const struct dmaengine_result *result)
415 struct spi_master *spi = cb;
417 spi->cur_msg->status = -EIO;
418 if (result->result != DMA_TRANS_NOERROR) {
419 dev_err(&spi->dev, "DMA txn failed: %d\n", result->result);
420 spi_finalize_current_transfer(spi);
424 if (!result->residue) {
425 spi->cur_msg->status = 0;
426 dev_dbg(&spi->dev, "DMA txn completed\n");
428 dev_err(&spi->dev, "DMA xfer has pending: %d\n", result->residue);
431 spi_finalize_current_transfer(spi);
434 static int setup_gsi_xfer(struct spi_transfer *xfer, struct spi_geni_master *mas,
435 struct spi_device *spi_slv, struct spi_master *spi)
437 unsigned long flags = DMA_PREP_INTERRUPT | DMA_CTRL_ACK;
438 struct dma_slave_config config = {};
439 struct gpi_spi_config peripheral = {};
440 struct dma_async_tx_descriptor *tx_desc, *rx_desc;
443 config.peripheral_config = &peripheral;
444 config.peripheral_size = sizeof(peripheral);
445 peripheral.set_config = true;
447 if (xfer->bits_per_word != mas->cur_bits_per_word ||
448 xfer->speed_hz != mas->cur_speed_hz) {
449 mas->cur_bits_per_word = xfer->bits_per_word;
450 mas->cur_speed_hz = xfer->speed_hz;
453 if (xfer->tx_buf && xfer->rx_buf) {
454 peripheral.cmd = SPI_DUPLEX;
455 } else if (xfer->tx_buf) {
456 peripheral.cmd = SPI_TX;
457 peripheral.rx_len = 0;
458 } else if (xfer->rx_buf) {
459 peripheral.cmd = SPI_RX;
460 if (!(mas->cur_bits_per_word % MIN_WORD_LEN)) {
461 peripheral.rx_len = ((xfer->len << 3) / mas->cur_bits_per_word);
463 int bytes_per_word = (mas->cur_bits_per_word / BITS_PER_BYTE) + 1;
465 peripheral.rx_len = (xfer->len / bytes_per_word);
469 peripheral.loopback_en = !!(spi_slv->mode & SPI_LOOP);
470 peripheral.clock_pol_high = !!(spi_slv->mode & SPI_CPOL);
471 peripheral.data_pol_high = !!(spi_slv->mode & SPI_CPHA);
472 peripheral.cs = spi_get_chipselect(spi_slv, 0);
473 peripheral.pack_en = true;
474 peripheral.word_len = xfer->bits_per_word - MIN_WORD_LEN;
476 ret = get_spi_clk_cfg(mas->cur_speed_hz, mas,
477 &peripheral.clk_src, &peripheral.clk_div);
479 dev_err(mas->dev, "Err in get_spi_clk_cfg() :%d\n", ret);
483 if (!xfer->cs_change) {
484 if (!list_is_last(&xfer->transfer_list, &spi->cur_msg->transfers))
485 peripheral.fragmentation = FRAGMENTATION;
488 if (peripheral.cmd & SPI_RX) {
489 dmaengine_slave_config(mas->rx, &config);
490 rx_desc = dmaengine_prep_slave_sg(mas->rx, xfer->rx_sg.sgl, xfer->rx_sg.nents,
491 DMA_DEV_TO_MEM, flags);
493 dev_err(mas->dev, "Err setting up rx desc\n");
499 * Prepare the TX always, even for RX or tx_buf being null, we would
500 * need TX to be prepared per GSI spec
502 dmaengine_slave_config(mas->tx, &config);
503 tx_desc = dmaengine_prep_slave_sg(mas->tx, xfer->tx_sg.sgl, xfer->tx_sg.nents,
504 DMA_MEM_TO_DEV, flags);
506 dev_err(mas->dev, "Err setting up tx desc\n");
510 tx_desc->callback_result = spi_gsi_callback_result;
511 tx_desc->callback_param = spi;
513 if (peripheral.cmd & SPI_RX)
514 dmaengine_submit(rx_desc);
515 dmaengine_submit(tx_desc);
517 if (peripheral.cmd & SPI_RX)
518 dma_async_issue_pending(mas->rx);
520 dma_async_issue_pending(mas->tx);
524 static bool geni_can_dma(struct spi_controller *ctlr,
525 struct spi_device *slv, struct spi_transfer *xfer)
527 struct spi_geni_master *mas = spi_master_get_devdata(slv->master);
530 * Return true if transfer needs to be mapped prior to
531 * calling transfer_one which is the case only for GPI_DMA.
532 * For SE_DMA mode, map/unmap is done in geni_se_*x_dma_prep.
534 return mas->cur_xfer_mode == GENI_GPI_DMA;
537 static int spi_geni_prepare_message(struct spi_master *spi,
538 struct spi_message *spi_msg)
540 struct spi_geni_master *mas = spi_master_get_devdata(spi);
543 switch (mas->cur_xfer_mode) {
546 if (spi_geni_is_abort_still_pending(mas))
548 ret = setup_fifo_params(spi_msg->spi, spi);
550 dev_err(mas->dev, "Couldn't select mode %d\n", ret);
554 /* nothing to do for GPI DMA */
558 dev_err(mas->dev, "Mode not supported %d", mas->cur_xfer_mode);
562 static int spi_geni_grab_gpi_chan(struct spi_geni_master *mas)
566 mas->tx = dma_request_chan(mas->dev, "tx");
567 if (IS_ERR(mas->tx)) {
568 ret = dev_err_probe(mas->dev, PTR_ERR(mas->tx),
569 "Failed to get tx DMA ch\n");
573 mas->rx = dma_request_chan(mas->dev, "rx");
574 if (IS_ERR(mas->rx)) {
575 ret = dev_err_probe(mas->dev, PTR_ERR(mas->rx),
576 "Failed to get rx DMA ch\n");
584 dma_release_channel(mas->tx);
590 static void spi_geni_release_dma_chan(struct spi_geni_master *mas)
593 dma_release_channel(mas->rx);
598 dma_release_channel(mas->tx);
603 static int spi_geni_init(struct spi_geni_master *mas)
605 struct geni_se *se = &mas->se;
606 unsigned int proto, major, minor, ver;
607 u32 spi_tx_cfg, fifo_disable;
610 pm_runtime_get_sync(mas->dev);
612 proto = geni_se_read_proto(se);
613 if (proto != GENI_SE_SPI) {
614 dev_err(mas->dev, "Invalid proto %d\n", proto);
617 mas->tx_fifo_depth = geni_se_get_tx_fifo_depth(se);
619 /* Width of Tx and Rx FIFO is same */
620 mas->fifo_width_bits = geni_se_get_tx_fifo_width(se);
623 * Hardware programming guide suggests to configure
624 * RX FIFO RFR level to fifo_depth-2.
626 geni_se_init(se, mas->tx_fifo_depth - 3, mas->tx_fifo_depth - 2);
627 /* Transmit an entire FIFO worth of data per IRQ */
629 ver = geni_se_get_qup_hw_version(se);
630 major = GENI_SE_VERSION_MAJOR(ver);
631 minor = GENI_SE_VERSION_MINOR(ver);
633 if (major == 1 && minor == 0)
634 mas->oversampling = 2;
636 mas->oversampling = 1;
638 fifo_disable = readl(se->base + GENI_IF_DISABLE_RO) & FIFO_IF_DISABLE;
639 switch (fifo_disable) {
641 ret = spi_geni_grab_gpi_chan(mas);
642 if (!ret) { /* success case */
643 mas->cur_xfer_mode = GENI_GPI_DMA;
644 geni_se_select_mode(se, GENI_GPI_DMA);
645 dev_dbg(mas->dev, "Using GPI DMA mode for SPI\n");
649 * in case of failure to get gpi dma channel, we can still do the
650 * FIFO mode, so fallthrough
652 dev_warn(mas->dev, "FIFO mode disabled, but couldn't get DMA, fall back to FIFO mode\n");
656 mas->cur_xfer_mode = GENI_SE_FIFO;
657 geni_se_select_mode(se, GENI_SE_FIFO);
662 /* We always control CS manually */
663 spi_tx_cfg = readl(se->base + SE_SPI_TRANS_CFG);
664 spi_tx_cfg &= ~CS_TOGGLE;
665 writel(spi_tx_cfg, se->base + SE_SPI_TRANS_CFG);
668 pm_runtime_put(mas->dev);
672 static unsigned int geni_byte_per_fifo_word(struct spi_geni_master *mas)
675 * Calculate how many bytes we'll put in each FIFO word. If the
676 * transfer words don't pack cleanly into a FIFO word we'll just put
677 * one transfer word in each FIFO word. If they do pack we'll pack 'em.
679 if (mas->fifo_width_bits % mas->cur_bits_per_word)
680 return roundup_pow_of_two(DIV_ROUND_UP(mas->cur_bits_per_word,
683 return mas->fifo_width_bits / BITS_PER_BYTE;
686 static bool geni_spi_handle_tx(struct spi_geni_master *mas)
688 struct geni_se *se = &mas->se;
689 unsigned int max_bytes;
691 unsigned int bytes_per_fifo_word = geni_byte_per_fifo_word(mas);
694 /* Stop the watermark IRQ if nothing to send */
695 if (!mas->cur_xfer) {
696 writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
700 max_bytes = (mas->tx_fifo_depth - mas->tx_wm) * bytes_per_fifo_word;
701 if (mas->tx_rem_bytes < max_bytes)
702 max_bytes = mas->tx_rem_bytes;
704 tx_buf = mas->cur_xfer->tx_buf + mas->cur_xfer->len - mas->tx_rem_bytes;
705 while (i < max_bytes) {
707 unsigned int bytes_to_write;
709 u8 *fifo_byte = (u8 *)&fifo_word;
711 bytes_to_write = min(bytes_per_fifo_word, max_bytes - i);
712 for (j = 0; j < bytes_to_write; j++)
713 fifo_byte[j] = tx_buf[i++];
714 iowrite32_rep(se->base + SE_GENI_TX_FIFOn, &fifo_word, 1);
716 mas->tx_rem_bytes -= max_bytes;
717 if (!mas->tx_rem_bytes) {
718 writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
724 static void geni_spi_handle_rx(struct spi_geni_master *mas)
726 struct geni_se *se = &mas->se;
728 unsigned int rx_bytes;
729 unsigned int rx_last_byte_valid;
731 unsigned int bytes_per_fifo_word = geni_byte_per_fifo_word(mas);
734 rx_fifo_status = readl(se->base + SE_GENI_RX_FIFO_STATUS);
735 rx_bytes = (rx_fifo_status & RX_FIFO_WC_MSK) * bytes_per_fifo_word;
736 if (rx_fifo_status & RX_LAST) {
737 rx_last_byte_valid = rx_fifo_status & RX_LAST_BYTE_VALID_MSK;
738 rx_last_byte_valid >>= RX_LAST_BYTE_VALID_SHFT;
739 if (rx_last_byte_valid && rx_last_byte_valid < 4)
740 rx_bytes -= bytes_per_fifo_word - rx_last_byte_valid;
743 /* Clear out the FIFO and bail if nowhere to put it */
744 if (!mas->cur_xfer) {
745 for (i = 0; i < DIV_ROUND_UP(rx_bytes, bytes_per_fifo_word); i++)
746 readl(se->base + SE_GENI_RX_FIFOn);
750 if (mas->rx_rem_bytes < rx_bytes)
751 rx_bytes = mas->rx_rem_bytes;
753 rx_buf = mas->cur_xfer->rx_buf + mas->cur_xfer->len - mas->rx_rem_bytes;
754 while (i < rx_bytes) {
756 u8 *fifo_byte = (u8 *)&fifo_word;
757 unsigned int bytes_to_read;
760 bytes_to_read = min(bytes_per_fifo_word, rx_bytes - i);
761 ioread32_rep(se->base + SE_GENI_RX_FIFOn, &fifo_word, 1);
762 for (j = 0; j < bytes_to_read; j++)
763 rx_buf[i++] = fifo_byte[j];
765 mas->rx_rem_bytes -= rx_bytes;
768 static int setup_se_xfer(struct spi_transfer *xfer,
769 struct spi_geni_master *mas,
770 u16 mode, struct spi_master *spi)
774 struct geni_se *se = &mas->se;
778 * Ensure that our interrupt handler isn't still running from some
779 * prior command before we start messing with the hardware behind
780 * its back. We don't need to _keep_ the lock here since we're only
781 * worried about racing with out interrupt handler. The SPI core
782 * already handles making sure that we're not trying to do two
783 * transfers at once or setting a chip select and doing a transfer
786 * NOTE: we actually _can't_ hold the lock here because possibly we
787 * might call clk_set_rate() which needs to be able to sleep.
789 spin_lock_irq(&mas->lock);
790 spin_unlock_irq(&mas->lock);
792 if (xfer->bits_per_word != mas->cur_bits_per_word) {
793 spi_setup_word_len(mas, mode, xfer->bits_per_word);
794 mas->cur_bits_per_word = xfer->bits_per_word;
797 /* Speed and bits per word can be overridden per transfer */
798 ret = geni_spi_set_clock_and_bw(mas, xfer->speed_hz);
802 mas->tx_rem_bytes = 0;
803 mas->rx_rem_bytes = 0;
805 if (!(mas->cur_bits_per_word % MIN_WORD_LEN))
806 len = xfer->len * BITS_PER_BYTE / mas->cur_bits_per_word;
808 len = xfer->len / (mas->cur_bits_per_word / BITS_PER_BYTE + 1);
809 len &= TRANS_LEN_MSK;
811 mas->cur_xfer = xfer;
813 m_cmd |= SPI_TX_ONLY;
814 mas->tx_rem_bytes = xfer->len;
815 writel(len, se->base + SE_SPI_TX_TRANS_LEN);
819 m_cmd |= SPI_RX_ONLY;
820 writel(len, se->base + SE_SPI_RX_TRANS_LEN);
821 mas->rx_rem_bytes = xfer->len;
824 /* Select transfer mode based on transfer length */
825 fifo_size = mas->tx_fifo_depth * mas->fifo_width_bits / mas->cur_bits_per_word;
826 mas->cur_xfer_mode = (len <= fifo_size) ? GENI_SE_FIFO : GENI_SE_DMA;
827 geni_se_select_mode(se, mas->cur_xfer_mode);
830 * Lock around right before we start the transfer since our
831 * interrupt could come in at any time now.
833 spin_lock_irq(&mas->lock);
834 geni_se_setup_m_cmd(se, m_cmd, FRAGMENTATION);
836 if (mas->cur_xfer_mode == GENI_SE_DMA) {
837 if (m_cmd & SPI_RX_ONLY) {
838 ret = geni_se_rx_dma_prep(se, xfer->rx_buf,
839 xfer->len, &mas->rx_se_dma);
841 dev_err(mas->dev, "Failed to setup Rx dma %d\n", ret);
843 goto unlock_and_return;
846 if (m_cmd & SPI_TX_ONLY) {
847 ret = geni_se_tx_dma_prep(se, (void *)xfer->tx_buf,
848 xfer->len, &mas->tx_se_dma);
850 dev_err(mas->dev, "Failed to setup Tx dma %d\n", ret);
852 if (m_cmd & SPI_RX_ONLY) {
853 /* Unmap rx buffer if duplex transfer */
854 geni_se_rx_dma_unprep(se, mas->rx_se_dma, xfer->len);
857 goto unlock_and_return;
860 } else if (m_cmd & SPI_TX_ONLY) {
861 if (geni_spi_handle_tx(mas))
862 writel(mas->tx_wm, se->base + SE_GENI_TX_WATERMARK_REG);
866 spin_unlock_irq(&mas->lock);
870 static int spi_geni_transfer_one(struct spi_master *spi,
871 struct spi_device *slv,
872 struct spi_transfer *xfer)
874 struct spi_geni_master *mas = spi_master_get_devdata(spi);
877 if (spi_geni_is_abort_still_pending(mas))
880 /* Terminate and return success for 0 byte length transfer */
884 if (mas->cur_xfer_mode == GENI_SE_FIFO || mas->cur_xfer_mode == GENI_SE_DMA) {
885 ret = setup_se_xfer(xfer, mas, slv->mode, spi);
886 /* SPI framework expects +ve ret code to wait for transfer complete */
891 return setup_gsi_xfer(xfer, mas, slv, spi);
894 static irqreturn_t geni_spi_isr(int irq, void *data)
896 struct spi_master *spi = data;
897 struct spi_geni_master *mas = spi_master_get_devdata(spi);
898 struct geni_se *se = &mas->se;
901 m_irq = readl(se->base + SE_GENI_M_IRQ_STATUS);
905 if (m_irq & (M_CMD_OVERRUN_EN | M_ILLEGAL_CMD_EN | M_CMD_FAILURE_EN |
906 M_RX_FIFO_RD_ERR_EN | M_RX_FIFO_WR_ERR_EN |
907 M_TX_FIFO_RD_ERR_EN | M_TX_FIFO_WR_ERR_EN))
908 dev_warn(mas->dev, "Unexpected IRQ err status %#010x\n", m_irq);
910 spin_lock(&mas->lock);
912 if (mas->cur_xfer_mode == GENI_SE_FIFO) {
913 if ((m_irq & M_RX_FIFO_WATERMARK_EN) || (m_irq & M_RX_FIFO_LAST_EN))
914 geni_spi_handle_rx(mas);
916 if (m_irq & M_TX_FIFO_WATERMARK_EN)
917 geni_spi_handle_tx(mas);
919 if (m_irq & M_CMD_DONE_EN) {
921 spi_finalize_current_transfer(spi);
922 mas->cur_xfer = NULL;
924 * If this happens, then a CMD_DONE came before all the
925 * Tx buffer bytes were sent out. This is unusual, log
926 * this condition and disable the WM interrupt to
927 * prevent the system from stalling due an interrupt
930 * If this happens when all Rx bytes haven't been
931 * received, log the condition. The only known time
932 * this can happen is if bits_per_word != 8 and some
933 * registers that expect xfer lengths in num spi_words
934 * weren't written correctly.
936 if (mas->tx_rem_bytes) {
937 writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
938 dev_err(mas->dev, "Premature done. tx_rem = %d bpw%d\n",
939 mas->tx_rem_bytes, mas->cur_bits_per_word);
941 if (mas->rx_rem_bytes)
942 dev_err(mas->dev, "Premature done. rx_rem = %d bpw%d\n",
943 mas->rx_rem_bytes, mas->cur_bits_per_word);
945 complete(&mas->cs_done);
948 } else if (mas->cur_xfer_mode == GENI_SE_DMA) {
949 const struct spi_transfer *xfer = mas->cur_xfer;
950 u32 dma_tx_status = readl_relaxed(se->base + SE_DMA_TX_IRQ_STAT);
951 u32 dma_rx_status = readl_relaxed(se->base + SE_DMA_RX_IRQ_STAT);
954 writel(dma_tx_status, se->base + SE_DMA_TX_IRQ_CLR);
956 writel(dma_rx_status, se->base + SE_DMA_RX_IRQ_CLR);
957 if (dma_tx_status & TX_DMA_DONE)
958 mas->tx_rem_bytes = 0;
959 if (dma_rx_status & RX_DMA_DONE)
960 mas->rx_rem_bytes = 0;
961 if (dma_tx_status & TX_RESET_DONE)
962 complete(&mas->tx_reset_done);
963 if (dma_rx_status & RX_RESET_DONE)
964 complete(&mas->rx_reset_done);
965 if (!mas->tx_rem_bytes && !mas->rx_rem_bytes && xfer) {
966 if (xfer->tx_buf && mas->tx_se_dma) {
967 geni_se_tx_dma_unprep(se, mas->tx_se_dma, xfer->len);
970 if (xfer->rx_buf && mas->rx_se_dma) {
971 geni_se_rx_dma_unprep(se, mas->rx_se_dma, xfer->len);
974 spi_finalize_current_transfer(spi);
975 mas->cur_xfer = NULL;
979 if (m_irq & M_CMD_CANCEL_EN)
980 complete(&mas->cancel_done);
981 if (m_irq & M_CMD_ABORT_EN)
982 complete(&mas->abort_done);
985 * It's safe or a good idea to Ack all of our interrupts at the end
986 * of the function. Specifically:
987 * - M_CMD_DONE_EN / M_RX_FIFO_LAST_EN: Edge triggered interrupts and
988 * clearing Acks. Clearing at the end relies on nobody else having
989 * started a new transfer yet or else we could be clearing _their_
990 * done bit, but everyone grabs the spinlock before starting a new
992 * - M_RX_FIFO_WATERMARK_EN / M_TX_FIFO_WATERMARK_EN: These appear
993 * to be "latched level" interrupts so it's important to clear them
994 * _after_ you've handled the condition and always safe to do so
995 * since they'll re-assert if they're still happening.
997 writel(m_irq, se->base + SE_GENI_M_IRQ_CLEAR);
999 spin_unlock(&mas->lock);
1004 static int spi_geni_probe(struct platform_device *pdev)
1007 struct spi_master *spi;
1008 struct spi_geni_master *mas;
1011 struct device *dev = &pdev->dev;
1013 irq = platform_get_irq(pdev, 0);
1017 ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64));
1019 return dev_err_probe(dev, ret, "could not set DMA mask\n");
1021 base = devm_platform_ioremap_resource(pdev, 0);
1023 return PTR_ERR(base);
1025 clk = devm_clk_get(dev, "se");
1027 return PTR_ERR(clk);
1029 spi = devm_spi_alloc_master(dev, sizeof(*mas));
1033 platform_set_drvdata(pdev, spi);
1034 mas = spi_master_get_devdata(spi);
1038 mas->se.wrapper = dev_get_drvdata(dev->parent);
1039 mas->se.base = base;
1042 ret = devm_pm_opp_set_clkname(&pdev->dev, "se");
1045 /* OPP table is optional */
1046 ret = devm_pm_opp_of_add_table(&pdev->dev);
1047 if (ret && ret != -ENODEV) {
1048 dev_err(&pdev->dev, "invalid OPP table in device tree\n");
1053 spi->dev.of_node = dev->of_node;
1054 spi->mode_bits = SPI_CPOL | SPI_CPHA | SPI_LOOP | SPI_CS_HIGH;
1055 spi->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 32);
1056 spi->num_chipselect = 4;
1057 spi->max_speed_hz = 50000000;
1058 spi->prepare_message = spi_geni_prepare_message;
1059 spi->transfer_one = spi_geni_transfer_one;
1060 spi->can_dma = geni_can_dma;
1061 spi->dma_map_dev = dev->parent;
1062 spi->auto_runtime_pm = true;
1063 spi->handle_err = spi_geni_handle_err;
1064 spi->use_gpio_descriptors = true;
1066 init_completion(&mas->cs_done);
1067 init_completion(&mas->cancel_done);
1068 init_completion(&mas->abort_done);
1069 init_completion(&mas->tx_reset_done);
1070 init_completion(&mas->rx_reset_done);
1071 spin_lock_init(&mas->lock);
1072 pm_runtime_use_autosuspend(&pdev->dev);
1073 pm_runtime_set_autosuspend_delay(&pdev->dev, 250);
1074 pm_runtime_enable(dev);
1076 ret = geni_icc_get(&mas->se, NULL);
1078 goto spi_geni_probe_runtime_disable;
1079 /* Set the bus quota to a reasonable value for register access */
1080 mas->se.icc_paths[GENI_TO_CORE].avg_bw = Bps_to_icc(CORE_2X_50_MHZ);
1081 mas->se.icc_paths[CPU_TO_GENI].avg_bw = GENI_DEFAULT_BW;
1083 ret = geni_icc_set_bw(&mas->se);
1085 goto spi_geni_probe_runtime_disable;
1087 ret = spi_geni_init(mas);
1089 goto spi_geni_probe_runtime_disable;
1092 * check the mode supported and set_cs for fifo mode only
1093 * for dma (gsi) mode, the gsi will set cs based on params passed in
1096 if (mas->cur_xfer_mode == GENI_SE_FIFO)
1097 spi->set_cs = spi_geni_set_cs;
1099 ret = request_irq(mas->irq, geni_spi_isr, 0, dev_name(dev), spi);
1101 goto spi_geni_release_dma;
1103 ret = spi_register_master(spi);
1105 goto spi_geni_probe_free_irq;
1108 spi_geni_probe_free_irq:
1109 free_irq(mas->irq, spi);
1110 spi_geni_release_dma:
1111 spi_geni_release_dma_chan(mas);
1112 spi_geni_probe_runtime_disable:
1113 pm_runtime_disable(dev);
1117 static void spi_geni_remove(struct platform_device *pdev)
1119 struct spi_master *spi = platform_get_drvdata(pdev);
1120 struct spi_geni_master *mas = spi_master_get_devdata(spi);
1122 /* Unregister _before_ disabling pm_runtime() so we stop transfers */
1123 spi_unregister_master(spi);
1125 spi_geni_release_dma_chan(mas);
1127 free_irq(mas->irq, spi);
1128 pm_runtime_disable(&pdev->dev);
1131 static int __maybe_unused spi_geni_runtime_suspend(struct device *dev)
1133 struct spi_master *spi = dev_get_drvdata(dev);
1134 struct spi_geni_master *mas = spi_master_get_devdata(spi);
1137 /* Drop the performance state vote */
1138 dev_pm_opp_set_rate(dev, 0);
1140 ret = geni_se_resources_off(&mas->se);
1144 return geni_icc_disable(&mas->se);
1147 static int __maybe_unused spi_geni_runtime_resume(struct device *dev)
1149 struct spi_master *spi = dev_get_drvdata(dev);
1150 struct spi_geni_master *mas = spi_master_get_devdata(spi);
1153 ret = geni_icc_enable(&mas->se);
1157 ret = geni_se_resources_on(&mas->se);
1161 return dev_pm_opp_set_rate(mas->dev, mas->cur_sclk_hz);
1164 static int __maybe_unused spi_geni_suspend(struct device *dev)
1166 struct spi_master *spi = dev_get_drvdata(dev);
1169 ret = spi_master_suspend(spi);
1173 ret = pm_runtime_force_suspend(dev);
1175 spi_master_resume(spi);
1180 static int __maybe_unused spi_geni_resume(struct device *dev)
1182 struct spi_master *spi = dev_get_drvdata(dev);
1185 ret = pm_runtime_force_resume(dev);
1189 ret = spi_master_resume(spi);
1191 pm_runtime_force_suspend(dev);
1196 static const struct dev_pm_ops spi_geni_pm_ops = {
1197 SET_RUNTIME_PM_OPS(spi_geni_runtime_suspend,
1198 spi_geni_runtime_resume, NULL)
1199 SET_SYSTEM_SLEEP_PM_OPS(spi_geni_suspend, spi_geni_resume)
1202 static const struct of_device_id spi_geni_dt_match[] = {
1203 { .compatible = "qcom,geni-spi" },
1206 MODULE_DEVICE_TABLE(of, spi_geni_dt_match);
1208 static struct platform_driver spi_geni_driver = {
1209 .probe = spi_geni_probe,
1210 .remove_new = spi_geni_remove,
1213 .pm = &spi_geni_pm_ops,
1214 .of_match_table = spi_geni_dt_match,
1217 module_platform_driver(spi_geni_driver);
1219 MODULE_DESCRIPTION("SPI driver for GENI based QUP cores");
1220 MODULE_LICENSE("GPL v2");