Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / mmc / host / mmci.c
diff --git a/kernel/drivers/mmc/host/mmci.c b/kernel/drivers/mmc/host/mmci.c
new file mode 100644 (file)
index 0000000..d42fc08
--- /dev/null
@@ -0,0 +1,1922 @@
+/*
+ *  linux/drivers/mmc/host/mmci.c - ARM PrimeCell MMCI PL180/1 driver
+ *
+ *  Copyright (C) 2003 Deep Blue Solutions, Ltd, All Rights Reserved.
+ *  Copyright (C) 2010 ST-Ericsson SA
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/init.h>
+#include <linux/ioport.h>
+#include <linux/device.h>
+#include <linux/io.h>
+#include <linux/interrupt.h>
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/delay.h>
+#include <linux/err.h>
+#include <linux/highmem.h>
+#include <linux/log2.h>
+#include <linux/mmc/pm.h>
+#include <linux/mmc/host.h>
+#include <linux/mmc/card.h>
+#include <linux/mmc/slot-gpio.h>
+#include <linux/amba/bus.h>
+#include <linux/clk.h>
+#include <linux/scatterlist.h>
+#include <linux/gpio.h>
+#include <linux/of_gpio.h>
+#include <linux/regulator/consumer.h>
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
+#include <linux/amba/mmci.h>
+#include <linux/pm_runtime.h>
+#include <linux/types.h>
+#include <linux/pinctrl/consumer.h>
+
+#include <asm/div64.h>
+#include <asm/io.h>
+#include <asm/sizes.h>
+
+#include "mmci.h"
+#include "mmci_qcom_dml.h"
+
+#define DRIVER_NAME "mmci-pl18x"
+
+static unsigned int fmax = 515633;
+
+/**
+ * struct variant_data - MMCI variant-specific quirks
+ * @clkreg: default value for MCICLOCK register
+ * @clkreg_enable: enable value for MMCICLOCK register
+ * @clkreg_8bit_bus_enable: enable value for 8 bit bus
+ * @clkreg_neg_edge_enable: enable value for inverted data/cmd output
+ * @datalength_bits: number of bits in the MMCIDATALENGTH register
+ * @fifosize: number of bytes that can be written when MMCI_TXFIFOEMPTY
+ *           is asserted (likewise for RX)
+ * @fifohalfsize: number of bytes that can be written when MCI_TXFIFOHALFEMPTY
+ *               is asserted (likewise for RX)
+ * @data_cmd_enable: enable value for data commands.
+ * @st_sdio: enable ST specific SDIO logic
+ * @st_clkdiv: true if using a ST-specific clock divider algorithm
+ * @datactrl_mask_ddrmode: ddr mode mask in datactrl register.
+ * @blksz_datactrl16: true if Block size is at b16..b30 position in datactrl register
+ * @blksz_datactrl4: true if Block size is at b4..b16 position in datactrl
+ *                  register
+ * @datactrl_mask_sdio: SDIO enable mask in datactrl register
+ * @pwrreg_powerup: power up value for MMCIPOWER register
+ * @f_max: maximum clk frequency supported by the controller.
+ * @signal_direction: input/out direction of bus signals can be indicated
+ * @pwrreg_clkgate: MMCIPOWER register must be used to gate the clock
+ * @busy_detect: true if busy detection on dat0 is supported
+ * @pwrreg_nopower: bits in MMCIPOWER don't controls ext. power supply
+ * @explicit_mclk_control: enable explicit mclk control in driver.
+ * @qcom_fifo: enables qcom specific fifo pio read logic.
+ * @qcom_dml: enables qcom specific dma glue for dma transfers.
+ * @reversed_irq_handling: handle data irq before cmd irq.
+ */
+struct variant_data {
+       unsigned int            clkreg;
+       unsigned int            clkreg_enable;
+       unsigned int            clkreg_8bit_bus_enable;
+       unsigned int            clkreg_neg_edge_enable;
+       unsigned int            datalength_bits;
+       unsigned int            fifosize;
+       unsigned int            fifohalfsize;
+       unsigned int            data_cmd_enable;
+       unsigned int            datactrl_mask_ddrmode;
+       unsigned int            datactrl_mask_sdio;
+       bool                    st_sdio;
+       bool                    st_clkdiv;
+       bool                    blksz_datactrl16;
+       bool                    blksz_datactrl4;
+       u32                     pwrreg_powerup;
+       u32                     f_max;
+       bool                    signal_direction;
+       bool                    pwrreg_clkgate;
+       bool                    busy_detect;
+       bool                    pwrreg_nopower;
+       bool                    explicit_mclk_control;
+       bool                    qcom_fifo;
+       bool                    qcom_dml;
+       bool                    reversed_irq_handling;
+};
+
+static struct variant_data variant_arm = {
+       .fifosize               = 16 * 4,
+       .fifohalfsize           = 8 * 4,
+       .datalength_bits        = 16,
+       .pwrreg_powerup         = MCI_PWR_UP,
+       .f_max                  = 100000000,
+       .reversed_irq_handling  = true,
+};
+
+static struct variant_data variant_arm_extended_fifo = {
+       .fifosize               = 128 * 4,
+       .fifohalfsize           = 64 * 4,
+       .datalength_bits        = 16,
+       .pwrreg_powerup         = MCI_PWR_UP,
+       .f_max                  = 100000000,
+};
+
+static struct variant_data variant_arm_extended_fifo_hwfc = {
+       .fifosize               = 128 * 4,
+       .fifohalfsize           = 64 * 4,
+       .clkreg_enable          = MCI_ARM_HWFCEN,
+       .datalength_bits        = 16,
+       .pwrreg_powerup         = MCI_PWR_UP,
+       .f_max                  = 100000000,
+};
+
+static struct variant_data variant_u300 = {
+       .fifosize               = 16 * 4,
+       .fifohalfsize           = 8 * 4,
+       .clkreg_enable          = MCI_ST_U300_HWFCEN,
+       .clkreg_8bit_bus_enable = MCI_ST_8BIT_BUS,
+       .datalength_bits        = 16,
+       .datactrl_mask_sdio     = MCI_ST_DPSM_SDIOEN,
+       .st_sdio                        = true,
+       .pwrreg_powerup         = MCI_PWR_ON,
+       .f_max                  = 100000000,
+       .signal_direction       = true,
+       .pwrreg_clkgate         = true,
+       .pwrreg_nopower         = true,
+};
+
+static struct variant_data variant_nomadik = {
+       .fifosize               = 16 * 4,
+       .fifohalfsize           = 8 * 4,
+       .clkreg                 = MCI_CLK_ENABLE,
+       .datalength_bits        = 24,
+       .datactrl_mask_sdio     = MCI_ST_DPSM_SDIOEN,
+       .st_sdio                = true,
+       .st_clkdiv              = true,
+       .pwrreg_powerup         = MCI_PWR_ON,
+       .f_max                  = 100000000,
+       .signal_direction       = true,
+       .pwrreg_clkgate         = true,
+       .pwrreg_nopower         = true,
+};
+
+static struct variant_data variant_ux500 = {
+       .fifosize               = 30 * 4,
+       .fifohalfsize           = 8 * 4,
+       .clkreg                 = MCI_CLK_ENABLE,
+       .clkreg_enable          = MCI_ST_UX500_HWFCEN,
+       .clkreg_8bit_bus_enable = MCI_ST_8BIT_BUS,
+       .clkreg_neg_edge_enable = MCI_ST_UX500_NEG_EDGE,
+       .datalength_bits        = 24,
+       .datactrl_mask_sdio     = MCI_ST_DPSM_SDIOEN,
+       .st_sdio                = true,
+       .st_clkdiv              = true,
+       .pwrreg_powerup         = MCI_PWR_ON,
+       .f_max                  = 100000000,
+       .signal_direction       = true,
+       .pwrreg_clkgate         = true,
+       .busy_detect            = true,
+       .pwrreg_nopower         = true,
+};
+
+static struct variant_data variant_ux500v2 = {
+       .fifosize               = 30 * 4,
+       .fifohalfsize           = 8 * 4,
+       .clkreg                 = MCI_CLK_ENABLE,
+       .clkreg_enable          = MCI_ST_UX500_HWFCEN,
+       .clkreg_8bit_bus_enable = MCI_ST_8BIT_BUS,
+       .clkreg_neg_edge_enable = MCI_ST_UX500_NEG_EDGE,
+       .datactrl_mask_ddrmode  = MCI_ST_DPSM_DDRMODE,
+       .datalength_bits        = 24,
+       .datactrl_mask_sdio     = MCI_ST_DPSM_SDIOEN,
+       .st_sdio                = true,
+       .st_clkdiv              = true,
+       .blksz_datactrl16       = true,
+       .pwrreg_powerup         = MCI_PWR_ON,
+       .f_max                  = 100000000,
+       .signal_direction       = true,
+       .pwrreg_clkgate         = true,
+       .busy_detect            = true,
+       .pwrreg_nopower         = true,
+};
+
+static struct variant_data variant_qcom = {
+       .fifosize               = 16 * 4,
+       .fifohalfsize           = 8 * 4,
+       .clkreg                 = MCI_CLK_ENABLE,
+       .clkreg_enable          = MCI_QCOM_CLK_FLOWENA |
+                                 MCI_QCOM_CLK_SELECT_IN_FBCLK,
+       .clkreg_8bit_bus_enable = MCI_QCOM_CLK_WIDEBUS_8,
+       .datactrl_mask_ddrmode  = MCI_QCOM_CLK_SELECT_IN_DDR_MODE,
+       .data_cmd_enable        = MCI_QCOM_CSPM_DATCMD,
+       .blksz_datactrl4        = true,
+       .datalength_bits        = 24,
+       .pwrreg_powerup         = MCI_PWR_UP,
+       .f_max                  = 208000000,
+       .explicit_mclk_control  = true,
+       .qcom_fifo              = true,
+       .qcom_dml               = true,
+};
+
+static int mmci_card_busy(struct mmc_host *mmc)
+{
+       struct mmci_host *host = mmc_priv(mmc);
+       unsigned long flags;
+       int busy = 0;
+
+       pm_runtime_get_sync(mmc_dev(mmc));
+
+       spin_lock_irqsave(&host->lock, flags);
+       if (readl(host->base + MMCISTATUS) & MCI_ST_CARDBUSY)
+               busy = 1;
+       spin_unlock_irqrestore(&host->lock, flags);
+
+       pm_runtime_mark_last_busy(mmc_dev(mmc));
+       pm_runtime_put_autosuspend(mmc_dev(mmc));
+
+       return busy;
+}
+
+/*
+ * Validate mmc prerequisites
+ */
+static int mmci_validate_data(struct mmci_host *host,
+                             struct mmc_data *data)
+{
+       if (!data)
+               return 0;
+
+       if (!is_power_of_2(data->blksz)) {
+               dev_err(mmc_dev(host->mmc),
+                       "unsupported block size (%d bytes)\n", data->blksz);
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+static void mmci_reg_delay(struct mmci_host *host)
+{
+       /*
+        * According to the spec, at least three feedback clock cycles
+        * of max 52 MHz must pass between two writes to the MMCICLOCK reg.
+        * Three MCLK clock cycles must pass between two MMCIPOWER reg writes.
+        * Worst delay time during card init is at 100 kHz => 30 us.
+        * Worst delay time when up and running is at 25 MHz => 120 ns.
+        */
+       if (host->cclk < 25000000)
+               udelay(30);
+       else
+               ndelay(120);
+}
+
+/*
+ * This must be called with host->lock held
+ */
+static void mmci_write_clkreg(struct mmci_host *host, u32 clk)
+{
+       if (host->clk_reg != clk) {
+               host->clk_reg = clk;
+               writel(clk, host->base + MMCICLOCK);
+       }
+}
+
+/*
+ * This must be called with host->lock held
+ */
+static void mmci_write_pwrreg(struct mmci_host *host, u32 pwr)
+{
+       if (host->pwr_reg != pwr) {
+               host->pwr_reg = pwr;
+               writel(pwr, host->base + MMCIPOWER);
+       }
+}
+
+/*
+ * This must be called with host->lock held
+ */
+static void mmci_write_datactrlreg(struct mmci_host *host, u32 datactrl)
+{
+       /* Keep ST Micro busy mode if enabled */
+       datactrl |= host->datactrl_reg & MCI_ST_DPSM_BUSYMODE;
+
+       if (host->datactrl_reg != datactrl) {
+               host->datactrl_reg = datactrl;
+               writel(datactrl, host->base + MMCIDATACTRL);
+       }
+}
+
+/*
+ * This must be called with host->lock held
+ */
+static void mmci_set_clkreg(struct mmci_host *host, unsigned int desired)
+{
+       struct variant_data *variant = host->variant;
+       u32 clk = variant->clkreg;
+
+       /* Make sure cclk reflects the current calculated clock */
+       host->cclk = 0;
+
+       if (desired) {
+               if (variant->explicit_mclk_control) {
+                       host->cclk = host->mclk;
+               } else if (desired >= host->mclk) {
+                       clk = MCI_CLK_BYPASS;
+                       if (variant->st_clkdiv)
+                               clk |= MCI_ST_UX500_NEG_EDGE;
+                       host->cclk = host->mclk;
+               } else if (variant->st_clkdiv) {
+                       /*
+                        * DB8500 TRM says f = mclk / (clkdiv + 2)
+                        * => clkdiv = (mclk / f) - 2
+                        * Round the divider up so we don't exceed the max
+                        * frequency
+                        */
+                       clk = DIV_ROUND_UP(host->mclk, desired) - 2;
+                       if (clk >= 256)
+                               clk = 255;
+                       host->cclk = host->mclk / (clk + 2);
+               } else {
+                       /*
+                        * PL180 TRM says f = mclk / (2 * (clkdiv + 1))
+                        * => clkdiv = mclk / (2 * f) - 1
+                        */
+                       clk = host->mclk / (2 * desired) - 1;
+                       if (clk >= 256)
+                               clk = 255;
+                       host->cclk = host->mclk / (2 * (clk + 1));
+               }
+
+               clk |= variant->clkreg_enable;
+               clk |= MCI_CLK_ENABLE;
+               /* This hasn't proven to be worthwhile */
+               /* clk |= MCI_CLK_PWRSAVE; */
+       }
+
+       /* Set actual clock for debug */
+       host->mmc->actual_clock = host->cclk;
+
+       if (host->mmc->ios.bus_width == MMC_BUS_WIDTH_4)
+               clk |= MCI_4BIT_BUS;
+       if (host->mmc->ios.bus_width == MMC_BUS_WIDTH_8)
+               clk |= variant->clkreg_8bit_bus_enable;
+
+       if (host->mmc->ios.timing == MMC_TIMING_UHS_DDR50 ||
+           host->mmc->ios.timing == MMC_TIMING_MMC_DDR52)
+               clk |= variant->clkreg_neg_edge_enable;
+
+       mmci_write_clkreg(host, clk);
+}
+
+static void
+mmci_request_end(struct mmci_host *host, struct mmc_request *mrq)
+{
+       writel(0, host->base + MMCICOMMAND);
+
+       BUG_ON(host->data);
+
+       host->mrq = NULL;
+       host->cmd = NULL;
+
+       mmc_request_done(host->mmc, mrq);
+
+       pm_runtime_mark_last_busy(mmc_dev(host->mmc));
+       pm_runtime_put_autosuspend(mmc_dev(host->mmc));
+}
+
+static void mmci_set_mask1(struct mmci_host *host, unsigned int mask)
+{
+       void __iomem *base = host->base;
+
+       if (host->singleirq) {
+               unsigned int mask0 = readl(base + MMCIMASK0);
+
+               mask0 &= ~MCI_IRQ1MASK;
+               mask0 |= mask;
+
+               writel(mask0, base + MMCIMASK0);
+       }
+
+       writel(mask, base + MMCIMASK1);
+}
+
+static void mmci_stop_data(struct mmci_host *host)
+{
+       mmci_write_datactrlreg(host, 0);
+       mmci_set_mask1(host, 0);
+       host->data = NULL;
+}
+
+static void mmci_init_sg(struct mmci_host *host, struct mmc_data *data)
+{
+       unsigned int flags = SG_MITER_ATOMIC;
+
+       if (data->flags & MMC_DATA_READ)
+               flags |= SG_MITER_TO_SG;
+       else
+               flags |= SG_MITER_FROM_SG;
+
+       sg_miter_start(&host->sg_miter, data->sg, data->sg_len, flags);
+}
+
+/*
+ * All the DMA operation mode stuff goes inside this ifdef.
+ * This assumes that you have a generic DMA device interface,
+ * no custom DMA interfaces are supported.
+ */
+#ifdef CONFIG_DMA_ENGINE
+static void mmci_dma_setup(struct mmci_host *host)
+{
+       const char *rxname, *txname;
+       struct variant_data *variant = host->variant;
+
+       host->dma_rx_channel = dma_request_slave_channel(mmc_dev(host->mmc), "rx");
+       host->dma_tx_channel = dma_request_slave_channel(mmc_dev(host->mmc), "tx");
+
+       /* initialize pre request cookie */
+       host->next_data.cookie = 1;
+
+       /*
+        * If only an RX channel is specified, the driver will
+        * attempt to use it bidirectionally, however if it is
+        * is specified but cannot be located, DMA will be disabled.
+        */
+       if (host->dma_rx_channel && !host->dma_tx_channel)
+               host->dma_tx_channel = host->dma_rx_channel;
+
+       if (host->dma_rx_channel)
+               rxname = dma_chan_name(host->dma_rx_channel);
+       else
+               rxname = "none";
+
+       if (host->dma_tx_channel)
+               txname = dma_chan_name(host->dma_tx_channel);
+       else
+               txname = "none";
+
+       dev_info(mmc_dev(host->mmc), "DMA channels RX %s, TX %s\n",
+                rxname, txname);
+
+       /*
+        * Limit the maximum segment size in any SG entry according to
+        * the parameters of the DMA engine device.
+        */
+       if (host->dma_tx_channel) {
+               struct device *dev = host->dma_tx_channel->device->dev;
+               unsigned int max_seg_size = dma_get_max_seg_size(dev);
+
+               if (max_seg_size < host->mmc->max_seg_size)
+                       host->mmc->max_seg_size = max_seg_size;
+       }
+       if (host->dma_rx_channel) {
+               struct device *dev = host->dma_rx_channel->device->dev;
+               unsigned int max_seg_size = dma_get_max_seg_size(dev);
+
+               if (max_seg_size < host->mmc->max_seg_size)
+                       host->mmc->max_seg_size = max_seg_size;
+       }
+
+       if (variant->qcom_dml && host->dma_rx_channel && host->dma_tx_channel)
+               if (dml_hw_init(host, host->mmc->parent->of_node))
+                       variant->qcom_dml = false;
+}
+
+/*
+ * This is used in or so inline it
+ * so it can be discarded.
+ */
+static inline void mmci_dma_release(struct mmci_host *host)
+{
+       if (host->dma_rx_channel)
+               dma_release_channel(host->dma_rx_channel);
+       if (host->dma_tx_channel)
+               dma_release_channel(host->dma_tx_channel);
+       host->dma_rx_channel = host->dma_tx_channel = NULL;
+}
+
+static void mmci_dma_data_error(struct mmci_host *host)
+{
+       dev_err(mmc_dev(host->mmc), "error during DMA transfer!\n");
+       dmaengine_terminate_all(host->dma_current);
+       host->dma_current = NULL;
+       host->dma_desc_current = NULL;
+       host->data->host_cookie = 0;
+}
+
+static void mmci_dma_unmap(struct mmci_host *host, struct mmc_data *data)
+{
+       struct dma_chan *chan;
+       enum dma_data_direction dir;
+
+       if (data->flags & MMC_DATA_READ) {
+               dir = DMA_FROM_DEVICE;
+               chan = host->dma_rx_channel;
+       } else {
+               dir = DMA_TO_DEVICE;
+               chan = host->dma_tx_channel;
+       }
+
+       dma_unmap_sg(chan->device->dev, data->sg, data->sg_len, dir);
+}
+
+static void mmci_dma_finalize(struct mmci_host *host, struct mmc_data *data)
+{
+       u32 status;
+       int i;
+
+       /* Wait up to 1ms for the DMA to complete */
+       for (i = 0; ; i++) {
+               status = readl(host->base + MMCISTATUS);
+               if (!(status & MCI_RXDATAAVLBLMASK) || i >= 100)
+                       break;
+               udelay(10);
+       }
+
+       /*
+        * Check to see whether we still have some data left in the FIFO -
+        * this catches DMA controllers which are unable to monitor the
+        * DMALBREQ and DMALSREQ signals while allowing us to DMA to non-
+        * contiguous buffers.  On TX, we'll get a FIFO underrun error.
+        */
+       if (status & MCI_RXDATAAVLBLMASK) {
+               mmci_dma_data_error(host);
+               if (!data->error)
+                       data->error = -EIO;
+       }
+
+       if (!data->host_cookie)
+               mmci_dma_unmap(host, data);
+
+       /*
+        * Use of DMA with scatter-gather is impossible.
+        * Give up with DMA and switch back to PIO mode.
+        */
+       if (status & MCI_RXDATAAVLBLMASK) {
+               dev_err(mmc_dev(host->mmc), "buggy DMA detected. Taking evasive action.\n");
+               mmci_dma_release(host);
+       }
+
+       host->dma_current = NULL;
+       host->dma_desc_current = NULL;
+}
+
+/* prepares DMA channel and DMA descriptor, returns non-zero on failure */
+static int __mmci_dma_prep_data(struct mmci_host *host, struct mmc_data *data,
+                               struct dma_chan **dma_chan,
+                               struct dma_async_tx_descriptor **dma_desc)
+{
+       struct variant_data *variant = host->variant;
+       struct dma_slave_config conf = {
+               .src_addr = host->phybase + MMCIFIFO,
+               .dst_addr = host->phybase + MMCIFIFO,
+               .src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES,
+               .dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES,
+               .src_maxburst = variant->fifohalfsize >> 2, /* # of words */
+               .dst_maxburst = variant->fifohalfsize >> 2, /* # of words */
+               .device_fc = false,
+       };
+       struct dma_chan *chan;
+       struct dma_device *device;
+       struct dma_async_tx_descriptor *desc;
+       enum dma_data_direction buffer_dirn;
+       int nr_sg;
+       unsigned long flags = DMA_CTRL_ACK;
+
+       if (data->flags & MMC_DATA_READ) {
+               conf.direction = DMA_DEV_TO_MEM;
+               buffer_dirn = DMA_FROM_DEVICE;
+               chan = host->dma_rx_channel;
+       } else {
+               conf.direction = DMA_MEM_TO_DEV;
+               buffer_dirn = DMA_TO_DEVICE;
+               chan = host->dma_tx_channel;
+       }
+
+       /* If there's no DMA channel, fall back to PIO */
+       if (!chan)
+               return -EINVAL;
+
+       /* If less than or equal to the fifo size, don't bother with DMA */
+       if (data->blksz * data->blocks <= variant->fifosize)
+               return -EINVAL;
+
+       device = chan->device;
+       nr_sg = dma_map_sg(device->dev, data->sg, data->sg_len, buffer_dirn);
+       if (nr_sg == 0)
+               return -EINVAL;
+
+       if (host->variant->qcom_dml)
+               flags |= DMA_PREP_INTERRUPT;
+
+       dmaengine_slave_config(chan, &conf);
+       desc = dmaengine_prep_slave_sg(chan, data->sg, nr_sg,
+                                           conf.direction, flags);
+       if (!desc)
+               goto unmap_exit;
+
+       *dma_chan = chan;
+       *dma_desc = desc;
+
+       return 0;
+
+ unmap_exit:
+       dma_unmap_sg(device->dev, data->sg, data->sg_len, buffer_dirn);
+       return -ENOMEM;
+}
+
+static inline int mmci_dma_prep_data(struct mmci_host *host,
+                                    struct mmc_data *data)
+{
+       /* Check if next job is already prepared. */
+       if (host->dma_current && host->dma_desc_current)
+               return 0;
+
+       /* No job were prepared thus do it now. */
+       return __mmci_dma_prep_data(host, data, &host->dma_current,
+                                   &host->dma_desc_current);
+}
+
+static inline int mmci_dma_prep_next(struct mmci_host *host,
+                                    struct mmc_data *data)
+{
+       struct mmci_host_next *nd = &host->next_data;
+       return __mmci_dma_prep_data(host, data, &nd->dma_chan, &nd->dma_desc);
+}
+
+static int mmci_dma_start_data(struct mmci_host *host, unsigned int datactrl)
+{
+       int ret;
+       struct mmc_data *data = host->data;
+
+       ret = mmci_dma_prep_data(host, host->data);
+       if (ret)
+               return ret;
+
+       /* Okay, go for it. */
+       dev_vdbg(mmc_dev(host->mmc),
+                "Submit MMCI DMA job, sglen %d blksz %04x blks %04x flags %08x\n",
+                data->sg_len, data->blksz, data->blocks, data->flags);
+       dmaengine_submit(host->dma_desc_current);
+       dma_async_issue_pending(host->dma_current);
+
+       if (host->variant->qcom_dml)
+               dml_start_xfer(host, data);
+
+       datactrl |= MCI_DPSM_DMAENABLE;
+
+       /* Trigger the DMA transfer */
+       mmci_write_datactrlreg(host, datactrl);
+
+       /*
+        * Let the MMCI say when the data is ended and it's time
+        * to fire next DMA request. When that happens, MMCI will
+        * call mmci_data_end()
+        */
+       writel(readl(host->base + MMCIMASK0) | MCI_DATAENDMASK,
+              host->base + MMCIMASK0);
+       return 0;
+}
+
+static void mmci_get_next_data(struct mmci_host *host, struct mmc_data *data)
+{
+       struct mmci_host_next *next = &host->next_data;
+
+       WARN_ON(data->host_cookie && data->host_cookie != next->cookie);
+       WARN_ON(!data->host_cookie && (next->dma_desc || next->dma_chan));
+
+       host->dma_desc_current = next->dma_desc;
+       host->dma_current = next->dma_chan;
+       next->dma_desc = NULL;
+       next->dma_chan = NULL;
+}
+
+static void mmci_pre_request(struct mmc_host *mmc, struct mmc_request *mrq,
+                            bool is_first_req)
+{
+       struct mmci_host *host = mmc_priv(mmc);
+       struct mmc_data *data = mrq->data;
+       struct mmci_host_next *nd = &host->next_data;
+
+       if (!data)
+               return;
+
+       BUG_ON(data->host_cookie);
+
+       if (mmci_validate_data(host, data))
+               return;
+
+       if (!mmci_dma_prep_next(host, data))
+               data->host_cookie = ++nd->cookie < 0 ? 1 : nd->cookie;
+}
+
+static void mmci_post_request(struct mmc_host *mmc, struct mmc_request *mrq,
+                             int err)
+{
+       struct mmci_host *host = mmc_priv(mmc);
+       struct mmc_data *data = mrq->data;
+
+       if (!data || !data->host_cookie)
+               return;
+
+       mmci_dma_unmap(host, data);
+
+       if (err) {
+               struct mmci_host_next *next = &host->next_data;
+               struct dma_chan *chan;
+               if (data->flags & MMC_DATA_READ)
+                       chan = host->dma_rx_channel;
+               else
+                       chan = host->dma_tx_channel;
+               dmaengine_terminate_all(chan);
+
+               if (host->dma_desc_current == next->dma_desc)
+                       host->dma_desc_current = NULL;
+
+               if (host->dma_current == next->dma_chan)
+                       host->dma_current = NULL;
+
+               next->dma_desc = NULL;
+               next->dma_chan = NULL;
+               data->host_cookie = 0;
+       }
+}
+
+#else
+/* Blank functions if the DMA engine is not available */
+static void mmci_get_next_data(struct mmci_host *host, struct mmc_data *data)
+{
+}
+static inline void mmci_dma_setup(struct mmci_host *host)
+{
+}
+
+static inline void mmci_dma_release(struct mmci_host *host)
+{
+}
+
+static inline void mmci_dma_unmap(struct mmci_host *host, struct mmc_data *data)
+{
+}
+
+static inline void mmci_dma_finalize(struct mmci_host *host,
+                                    struct mmc_data *data)
+{
+}
+
+static inline void mmci_dma_data_error(struct mmci_host *host)
+{
+}
+
+static inline int mmci_dma_start_data(struct mmci_host *host, unsigned int datactrl)
+{
+       return -ENOSYS;
+}
+
+#define mmci_pre_request NULL
+#define mmci_post_request NULL
+
+#endif
+
+static void mmci_start_data(struct mmci_host *host, struct mmc_data *data)
+{
+       struct variant_data *variant = host->variant;
+       unsigned int datactrl, timeout, irqmask;
+       unsigned long long clks;
+       void __iomem *base;
+       int blksz_bits;
+
+       dev_dbg(mmc_dev(host->mmc), "blksz %04x blks %04x flags %08x\n",
+               data->blksz, data->blocks, data->flags);
+
+       host->data = data;
+       host->size = data->blksz * data->blocks;
+       data->bytes_xfered = 0;
+
+       clks = (unsigned long long)data->timeout_ns * host->cclk;
+       do_div(clks, NSEC_PER_SEC);
+
+       timeout = data->timeout_clks + (unsigned int)clks;
+
+       base = host->base;
+       writel(timeout, base + MMCIDATATIMER);
+       writel(host->size, base + MMCIDATALENGTH);
+
+       blksz_bits = ffs(data->blksz) - 1;
+       BUG_ON(1 << blksz_bits != data->blksz);
+
+       if (variant->blksz_datactrl16)
+               datactrl = MCI_DPSM_ENABLE | (data->blksz << 16);
+       else if (variant->blksz_datactrl4)
+               datactrl = MCI_DPSM_ENABLE | (data->blksz << 4);
+       else
+               datactrl = MCI_DPSM_ENABLE | blksz_bits << 4;
+
+       if (data->flags & MMC_DATA_READ)
+               datactrl |= MCI_DPSM_DIRECTION;
+
+       if (host->mmc->card && mmc_card_sdio(host->mmc->card)) {
+               u32 clk;
+
+               datactrl |= variant->datactrl_mask_sdio;
+
+               /*
+                * The ST Micro variant for SDIO small write transfers
+                * needs to have clock H/W flow control disabled,
+                * otherwise the transfer will not start. The threshold
+                * depends on the rate of MCLK.
+                */
+               if (variant->st_sdio && data->flags & MMC_DATA_WRITE &&
+                   (host->size < 8 ||
+                    (host->size <= 8 && host->mclk > 50000000)))
+                       clk = host->clk_reg & ~variant->clkreg_enable;
+               else
+                       clk = host->clk_reg | variant->clkreg_enable;
+
+               mmci_write_clkreg(host, clk);
+       }
+
+       if (host->mmc->ios.timing == MMC_TIMING_UHS_DDR50 ||
+           host->mmc->ios.timing == MMC_TIMING_MMC_DDR52)
+               datactrl |= variant->datactrl_mask_ddrmode;
+
+       /*
+        * Attempt to use DMA operation mode, if this
+        * should fail, fall back to PIO mode
+        */
+       if (!mmci_dma_start_data(host, datactrl))
+               return;
+
+       /* IRQ mode, map the SG list for CPU reading/writing */
+       mmci_init_sg(host, data);
+
+       if (data->flags & MMC_DATA_READ) {
+               irqmask = MCI_RXFIFOHALFFULLMASK;
+
+               /*
+                * If we have less than the fifo 'half-full' threshold to
+                * transfer, trigger a PIO interrupt as soon as any data
+                * is available.
+                */
+               if (host->size < variant->fifohalfsize)
+                       irqmask |= MCI_RXDATAAVLBLMASK;
+       } else {
+               /*
+                * We don't actually need to include "FIFO empty" here
+                * since its implicit in "FIFO half empty".
+                */
+               irqmask = MCI_TXFIFOHALFEMPTYMASK;
+       }
+
+       mmci_write_datactrlreg(host, datactrl);
+       writel(readl(base + MMCIMASK0) & ~MCI_DATAENDMASK, base + MMCIMASK0);
+       mmci_set_mask1(host, irqmask);
+}
+
+static void
+mmci_start_command(struct mmci_host *host, struct mmc_command *cmd, u32 c)
+{
+       void __iomem *base = host->base;
+
+       dev_dbg(mmc_dev(host->mmc), "op %02x arg %08x flags %08x\n",
+           cmd->opcode, cmd->arg, cmd->flags);
+
+       if (readl(base + MMCICOMMAND) & MCI_CPSM_ENABLE) {
+               writel(0, base + MMCICOMMAND);
+               mmci_reg_delay(host);
+       }
+
+       c |= cmd->opcode | MCI_CPSM_ENABLE;
+       if (cmd->flags & MMC_RSP_PRESENT) {
+               if (cmd->flags & MMC_RSP_136)
+                       c |= MCI_CPSM_LONGRSP;
+               c |= MCI_CPSM_RESPONSE;
+       }
+       if (/*interrupt*/0)
+               c |= MCI_CPSM_INTERRUPT;
+
+       if (mmc_cmd_type(cmd) == MMC_CMD_ADTC)
+               c |= host->variant->data_cmd_enable;
+
+       host->cmd = cmd;
+
+       writel(cmd->arg, base + MMCIARGUMENT);
+       writel(c, base + MMCICOMMAND);
+}
+
+static void
+mmci_data_irq(struct mmci_host *host, struct mmc_data *data,
+             unsigned int status)
+{
+       /* Make sure we have data to handle */
+       if (!data)
+               return;
+
+       /* First check for errors */
+       if (status & (MCI_DATACRCFAIL|MCI_DATATIMEOUT|MCI_STARTBITERR|
+                     MCI_TXUNDERRUN|MCI_RXOVERRUN)) {
+               u32 remain, success;
+
+               /* Terminate the DMA transfer */
+               if (dma_inprogress(host)) {
+                       mmci_dma_data_error(host);
+                       mmci_dma_unmap(host, data);
+               }
+
+               /*
+                * Calculate how far we are into the transfer.  Note that
+                * the data counter gives the number of bytes transferred
+                * on the MMC bus, not on the host side.  On reads, this
+                * can be as much as a FIFO-worth of data ahead.  This
+                * matters for FIFO overruns only.
+                */
+               remain = readl(host->base + MMCIDATACNT);
+               success = data->blksz * data->blocks - remain;
+
+               dev_dbg(mmc_dev(host->mmc), "MCI ERROR IRQ, status 0x%08x at 0x%08x\n",
+                       status, success);
+               if (status & MCI_DATACRCFAIL) {
+                       /* Last block was not successful */
+                       success -= 1;
+                       data->error = -EILSEQ;
+               } else if (status & MCI_DATATIMEOUT) {
+                       data->error = -ETIMEDOUT;
+               } else if (status & MCI_STARTBITERR) {
+                       data->error = -ECOMM;
+               } else if (status & MCI_TXUNDERRUN) {
+                       data->error = -EIO;
+               } else if (status & MCI_RXOVERRUN) {
+                       if (success > host->variant->fifosize)
+                               success -= host->variant->fifosize;
+                       else
+                               success = 0;
+                       data->error = -EIO;
+               }
+               data->bytes_xfered = round_down(success, data->blksz);
+       }
+
+       if (status & MCI_DATABLOCKEND)
+               dev_err(mmc_dev(host->mmc), "stray MCI_DATABLOCKEND interrupt\n");
+
+       if (status & MCI_DATAEND || data->error) {
+               if (dma_inprogress(host))
+                       mmci_dma_finalize(host, data);
+               mmci_stop_data(host);
+
+               if (!data->error)
+                       /* The error clause is handled above, success! */
+                       data->bytes_xfered = data->blksz * data->blocks;
+
+               if (!data->stop || host->mrq->sbc) {
+                       mmci_request_end(host, data->mrq);
+               } else {
+                       mmci_start_command(host, data->stop, 0);
+               }
+       }
+}
+
+static void
+mmci_cmd_irq(struct mmci_host *host, struct mmc_command *cmd,
+            unsigned int status)
+{
+       void __iomem *base = host->base;
+       bool sbc, busy_resp;
+
+       if (!cmd)
+               return;
+
+       sbc = (cmd == host->mrq->sbc);
+       busy_resp = host->variant->busy_detect && (cmd->flags & MMC_RSP_BUSY);
+
+       if (!((status|host->busy_status) & (MCI_CMDCRCFAIL|MCI_CMDTIMEOUT|
+               MCI_CMDSENT|MCI_CMDRESPEND)))
+               return;
+
+       /* Check if we need to wait for busy completion. */
+       if (host->busy_status && (status & MCI_ST_CARDBUSY))
+               return;
+
+       /* Enable busy completion if needed and supported. */
+       if (!host->busy_status && busy_resp &&
+               !(status & (MCI_CMDCRCFAIL|MCI_CMDTIMEOUT)) &&
+               (readl(base + MMCISTATUS) & MCI_ST_CARDBUSY)) {
+               writel(readl(base + MMCIMASK0) | MCI_ST_BUSYEND,
+                       base + MMCIMASK0);
+               host->busy_status = status & (MCI_CMDSENT|MCI_CMDRESPEND);
+               return;
+       }
+
+       /* At busy completion, mask the IRQ and complete the request. */
+       if (host->busy_status) {
+               writel(readl(base + MMCIMASK0) & ~MCI_ST_BUSYEND,
+                       base + MMCIMASK0);
+               host->busy_status = 0;
+       }
+
+       host->cmd = NULL;
+
+       if (status & MCI_CMDTIMEOUT) {
+               cmd->error = -ETIMEDOUT;
+       } else if (status & MCI_CMDCRCFAIL && cmd->flags & MMC_RSP_CRC) {
+               cmd->error = -EILSEQ;
+       } else {
+               cmd->resp[0] = readl(base + MMCIRESPONSE0);
+               cmd->resp[1] = readl(base + MMCIRESPONSE1);
+               cmd->resp[2] = readl(base + MMCIRESPONSE2);
+               cmd->resp[3] = readl(base + MMCIRESPONSE3);
+       }
+
+       if ((!sbc && !cmd->data) || cmd->error) {
+               if (host->data) {
+                       /* Terminate the DMA transfer */
+                       if (dma_inprogress(host)) {
+                               mmci_dma_data_error(host);
+                               mmci_dma_unmap(host, host->data);
+                       }
+                       mmci_stop_data(host);
+               }
+               mmci_request_end(host, host->mrq);
+       } else if (sbc) {
+               mmci_start_command(host, host->mrq->cmd, 0);
+       } else if (!(cmd->data->flags & MMC_DATA_READ)) {
+               mmci_start_data(host, cmd->data);
+       }
+}
+
+static int mmci_get_rx_fifocnt(struct mmci_host *host, u32 status, int remain)
+{
+       return remain - (readl(host->base + MMCIFIFOCNT) << 2);
+}
+
+static int mmci_qcom_get_rx_fifocnt(struct mmci_host *host, u32 status, int r)
+{
+       /*
+        * on qcom SDCC4 only 8 words are used in each burst so only 8 addresses
+        * from the fifo range should be used
+        */
+       if (status & MCI_RXFIFOHALFFULL)
+               return host->variant->fifohalfsize;
+       else if (status & MCI_RXDATAAVLBL)
+               return 4;
+
+       return 0;
+}
+
+static int mmci_pio_read(struct mmci_host *host, char *buffer, unsigned int remain)
+{
+       void __iomem *base = host->base;
+       char *ptr = buffer;
+       u32 status = readl(host->base + MMCISTATUS);
+       int host_remain = host->size;
+
+       do {
+               int count = host->get_rx_fifocnt(host, status, host_remain);
+
+               if (count > remain)
+                       count = remain;
+
+               if (count <= 0)
+                       break;
+
+               /*
+                * SDIO especially may want to send something that is
+                * not divisible by 4 (as opposed to card sectors
+                * etc). Therefore make sure to always read the last bytes
+                * while only doing full 32-bit reads towards the FIFO.
+                */
+               if (unlikely(count & 0x3)) {
+                       if (count < 4) {
+                               unsigned char buf[4];
+                               ioread32_rep(base + MMCIFIFO, buf, 1);
+                               memcpy(ptr, buf, count);
+                       } else {
+                               ioread32_rep(base + MMCIFIFO, ptr, count >> 2);
+                               count &= ~0x3;
+                       }
+               } else {
+                       ioread32_rep(base + MMCIFIFO, ptr, count >> 2);
+               }
+
+               ptr += count;
+               remain -= count;
+               host_remain -= count;
+
+               if (remain == 0)
+                       break;
+
+               status = readl(base + MMCISTATUS);
+       } while (status & MCI_RXDATAAVLBL);
+
+       return ptr - buffer;
+}
+
+static int mmci_pio_write(struct mmci_host *host, char *buffer, unsigned int remain, u32 status)
+{
+       struct variant_data *variant = host->variant;
+       void __iomem *base = host->base;
+       char *ptr = buffer;
+
+       do {
+               unsigned int count, maxcnt;
+
+               maxcnt = status & MCI_TXFIFOEMPTY ?
+                        variant->fifosize : variant->fifohalfsize;
+               count = min(remain, maxcnt);
+
+               /*
+                * SDIO especially may want to send something that is
+                * not divisible by 4 (as opposed to card sectors
+                * etc), and the FIFO only accept full 32-bit writes.
+                * So compensate by adding +3 on the count, a single
+                * byte become a 32bit write, 7 bytes will be two
+                * 32bit writes etc.
+                */
+               iowrite32_rep(base + MMCIFIFO, ptr, (count + 3) >> 2);
+
+               ptr += count;
+               remain -= count;
+
+               if (remain == 0)
+                       break;
+
+               status = readl(base + MMCISTATUS);
+       } while (status & MCI_TXFIFOHALFEMPTY);
+
+       return ptr - buffer;
+}
+
+/*
+ * PIO data transfer IRQ handler.
+ */
+static irqreturn_t mmci_pio_irq(int irq, void *dev_id)
+{
+       struct mmci_host *host = dev_id;
+       struct sg_mapping_iter *sg_miter = &host->sg_miter;
+       struct variant_data *variant = host->variant;
+       void __iomem *base = host->base;
+       u32 status;
+
+       status = readl(base + MMCISTATUS);
+
+       dev_dbg(mmc_dev(host->mmc), "irq1 (pio) %08x\n", status);
+
+       do {
+               unsigned int remain, len;
+               char *buffer;
+
+               /*
+                * For write, we only need to test the half-empty flag
+                * here - if the FIFO is completely empty, then by
+                * definition it is more than half empty.
+                *
+                * For read, check for data available.
+                */
+               if (!(status & (MCI_TXFIFOHALFEMPTY|MCI_RXDATAAVLBL)))
+                       break;
+
+               if (!sg_miter_next(sg_miter))
+                       break;
+
+               buffer = sg_miter->addr;
+               remain = sg_miter->length;
+
+               len = 0;
+               if (status & MCI_RXACTIVE)
+                       len = mmci_pio_read(host, buffer, remain);
+               if (status & MCI_TXACTIVE)
+                       len = mmci_pio_write(host, buffer, remain, status);
+
+               sg_miter->consumed = len;
+
+               host->size -= len;
+               remain -= len;
+
+               if (remain)
+                       break;
+
+               status = readl(base + MMCISTATUS);
+       } while (1);
+
+       sg_miter_stop(sg_miter);
+
+       /*
+        * If we have less than the fifo 'half-full' threshold to transfer,
+        * trigger a PIO interrupt as soon as any data is available.
+        */
+       if (status & MCI_RXACTIVE && host->size < variant->fifohalfsize)
+               mmci_set_mask1(host, MCI_RXDATAAVLBLMASK);
+
+       /*
+        * If we run out of data, disable the data IRQs; this
+        * prevents a race where the FIFO becomes empty before
+        * the chip itself has disabled the data path, and
+        * stops us racing with our data end IRQ.
+        */
+       if (host->size == 0) {
+               mmci_set_mask1(host, 0);
+               writel(readl(base + MMCIMASK0) | MCI_DATAENDMASK, base + MMCIMASK0);
+       }
+
+       return IRQ_HANDLED;
+}
+
+/*
+ * Handle completion of command and data transfers.
+ */
+static irqreturn_t mmci_irq(int irq, void *dev_id)
+{
+       struct mmci_host *host = dev_id;
+       u32 status;
+       int ret = 0;
+
+       spin_lock(&host->lock);
+
+       do {
+               status = readl(host->base + MMCISTATUS);
+
+               if (host->singleirq) {
+                       if (status & readl(host->base + MMCIMASK1))
+                               mmci_pio_irq(irq, dev_id);
+
+                       status &= ~MCI_IRQ1MASK;
+               }
+
+               /*
+                * We intentionally clear the MCI_ST_CARDBUSY IRQ here (if it's
+                * enabled) since the HW seems to be triggering the IRQ on both
+                * edges while monitoring DAT0 for busy completion.
+                */
+               status &= readl(host->base + MMCIMASK0);
+               writel(status, host->base + MMCICLEAR);
+
+               dev_dbg(mmc_dev(host->mmc), "irq0 (data+cmd) %08x\n", status);
+
+               if (host->variant->reversed_irq_handling) {
+                       mmci_data_irq(host, host->data, status);
+                       mmci_cmd_irq(host, host->cmd, status);
+               } else {
+                       mmci_cmd_irq(host, host->cmd, status);
+                       mmci_data_irq(host, host->data, status);
+               }
+
+               /* Don't poll for busy completion in irq context. */
+               if (host->busy_status)
+                       status &= ~MCI_ST_CARDBUSY;
+
+               ret = 1;
+       } while (status);
+
+       spin_unlock(&host->lock);
+
+       return IRQ_RETVAL(ret);
+}
+
+static void mmci_request(struct mmc_host *mmc, struct mmc_request *mrq)
+{
+       struct mmci_host *host = mmc_priv(mmc);
+       unsigned long flags;
+
+       WARN_ON(host->mrq != NULL);
+
+       mrq->cmd->error = mmci_validate_data(host, mrq->data);
+       if (mrq->cmd->error) {
+               mmc_request_done(mmc, mrq);
+               return;
+       }
+
+       pm_runtime_get_sync(mmc_dev(mmc));
+
+       spin_lock_irqsave(&host->lock, flags);
+
+       host->mrq = mrq;
+
+       if (mrq->data)
+               mmci_get_next_data(host, mrq->data);
+
+       if (mrq->data && mrq->data->flags & MMC_DATA_READ)
+               mmci_start_data(host, mrq->data);
+
+       if (mrq->sbc)
+               mmci_start_command(host, mrq->sbc, 0);
+       else
+               mmci_start_command(host, mrq->cmd, 0);
+
+       spin_unlock_irqrestore(&host->lock, flags);
+}
+
+static void mmci_set_ios(struct mmc_host *mmc, struct mmc_ios *ios)
+{
+       struct mmci_host *host = mmc_priv(mmc);
+       struct variant_data *variant = host->variant;
+       u32 pwr = 0;
+       unsigned long flags;
+       int ret;
+
+       pm_runtime_get_sync(mmc_dev(mmc));
+
+       if (host->plat->ios_handler &&
+               host->plat->ios_handler(mmc_dev(mmc), ios))
+                       dev_err(mmc_dev(mmc), "platform ios_handler failed\n");
+
+       switch (ios->power_mode) {
+       case MMC_POWER_OFF:
+               if (!IS_ERR(mmc->supply.vmmc))
+                       mmc_regulator_set_ocr(mmc, mmc->supply.vmmc, 0);
+
+               if (!IS_ERR(mmc->supply.vqmmc) && host->vqmmc_enabled) {
+                       regulator_disable(mmc->supply.vqmmc);
+                       host->vqmmc_enabled = false;
+               }
+
+               break;
+       case MMC_POWER_UP:
+               if (!IS_ERR(mmc->supply.vmmc))
+                       mmc_regulator_set_ocr(mmc, mmc->supply.vmmc, ios->vdd);
+
+               /*
+                * The ST Micro variant doesn't have the PL180s MCI_PWR_UP
+                * and instead uses MCI_PWR_ON so apply whatever value is
+                * configured in the variant data.
+                */
+               pwr |= variant->pwrreg_powerup;
+
+               break;
+       case MMC_POWER_ON:
+               if (!IS_ERR(mmc->supply.vqmmc) && !host->vqmmc_enabled) {
+                       ret = regulator_enable(mmc->supply.vqmmc);
+                       if (ret < 0)
+                               dev_err(mmc_dev(mmc),
+                                       "failed to enable vqmmc regulator\n");
+                       else
+                               host->vqmmc_enabled = true;
+               }
+
+               pwr |= MCI_PWR_ON;
+               break;
+       }
+
+       if (variant->signal_direction && ios->power_mode != MMC_POWER_OFF) {
+               /*
+                * The ST Micro variant has some additional bits
+                * indicating signal direction for the signals in
+                * the SD/MMC bus and feedback-clock usage.
+                */
+               pwr |= host->pwr_reg_add;
+
+               if (ios->bus_width == MMC_BUS_WIDTH_4)
+                       pwr &= ~MCI_ST_DATA74DIREN;
+               else if (ios->bus_width == MMC_BUS_WIDTH_1)
+                       pwr &= (~MCI_ST_DATA74DIREN &
+                               ~MCI_ST_DATA31DIREN &
+                               ~MCI_ST_DATA2DIREN);
+       }
+
+       if (ios->bus_mode == MMC_BUSMODE_OPENDRAIN) {
+               if (host->hw_designer != AMBA_VENDOR_ST)
+                       pwr |= MCI_ROD;
+               else {
+                       /*
+                        * The ST Micro variant use the ROD bit for something
+                        * else and only has OD (Open Drain).
+                        */
+                       pwr |= MCI_OD;
+               }
+       }
+
+       /*
+        * If clock = 0 and the variant requires the MMCIPOWER to be used for
+        * gating the clock, the MCI_PWR_ON bit is cleared.
+        */
+       if (!ios->clock && variant->pwrreg_clkgate)
+               pwr &= ~MCI_PWR_ON;
+
+       if (host->variant->explicit_mclk_control &&
+           ios->clock != host->clock_cache) {
+               ret = clk_set_rate(host->clk, ios->clock);
+               if (ret < 0)
+                       dev_err(mmc_dev(host->mmc),
+                               "Error setting clock rate (%d)\n", ret);
+               else
+                       host->mclk = clk_get_rate(host->clk);
+       }
+       host->clock_cache = ios->clock;
+
+       spin_lock_irqsave(&host->lock, flags);
+
+       mmci_set_clkreg(host, ios->clock);
+       mmci_write_pwrreg(host, pwr);
+       mmci_reg_delay(host);
+
+       spin_unlock_irqrestore(&host->lock, flags);
+
+       pm_runtime_mark_last_busy(mmc_dev(mmc));
+       pm_runtime_put_autosuspend(mmc_dev(mmc));
+}
+
+static int mmci_get_cd(struct mmc_host *mmc)
+{
+       struct mmci_host *host = mmc_priv(mmc);
+       struct mmci_platform_data *plat = host->plat;
+       unsigned int status = mmc_gpio_get_cd(mmc);
+
+       if (status == -ENOSYS) {
+               if (!plat->status)
+                       return 1; /* Assume always present */
+
+               status = plat->status(mmc_dev(host->mmc));
+       }
+       return status;
+}
+
+static int mmci_sig_volt_switch(struct mmc_host *mmc, struct mmc_ios *ios)
+{
+       int ret = 0;
+
+       if (!IS_ERR(mmc->supply.vqmmc)) {
+
+               pm_runtime_get_sync(mmc_dev(mmc));
+
+               switch (ios->signal_voltage) {
+               case MMC_SIGNAL_VOLTAGE_330:
+                       ret = regulator_set_voltage(mmc->supply.vqmmc,
+                                               2700000, 3600000);
+                       break;
+               case MMC_SIGNAL_VOLTAGE_180:
+                       ret = regulator_set_voltage(mmc->supply.vqmmc,
+                                               1700000, 1950000);
+                       break;
+               case MMC_SIGNAL_VOLTAGE_120:
+                       ret = regulator_set_voltage(mmc->supply.vqmmc,
+                                               1100000, 1300000);
+                       break;
+               }
+
+               if (ret)
+                       dev_warn(mmc_dev(mmc), "Voltage switch failed\n");
+
+               pm_runtime_mark_last_busy(mmc_dev(mmc));
+               pm_runtime_put_autosuspend(mmc_dev(mmc));
+       }
+
+       return ret;
+}
+
+static struct mmc_host_ops mmci_ops = {
+       .request        = mmci_request,
+       .pre_req        = mmci_pre_request,
+       .post_req       = mmci_post_request,
+       .set_ios        = mmci_set_ios,
+       .get_ro         = mmc_gpio_get_ro,
+       .get_cd         = mmci_get_cd,
+       .start_signal_voltage_switch = mmci_sig_volt_switch,
+};
+
+static int mmci_of_parse(struct device_node *np, struct mmc_host *mmc)
+{
+       struct mmci_host *host = mmc_priv(mmc);
+       int ret = mmc_of_parse(mmc);
+
+       if (ret)
+               return ret;
+
+       if (of_get_property(np, "st,sig-dir-dat0", NULL))
+               host->pwr_reg_add |= MCI_ST_DATA0DIREN;
+       if (of_get_property(np, "st,sig-dir-dat2", NULL))
+               host->pwr_reg_add |= MCI_ST_DATA2DIREN;
+       if (of_get_property(np, "st,sig-dir-dat31", NULL))
+               host->pwr_reg_add |= MCI_ST_DATA31DIREN;
+       if (of_get_property(np, "st,sig-dir-dat74", NULL))
+               host->pwr_reg_add |= MCI_ST_DATA74DIREN;
+       if (of_get_property(np, "st,sig-dir-cmd", NULL))
+               host->pwr_reg_add |= MCI_ST_CMDDIREN;
+       if (of_get_property(np, "st,sig-pin-fbclk", NULL))
+               host->pwr_reg_add |= MCI_ST_FBCLKEN;
+
+       if (of_get_property(np, "mmc-cap-mmc-highspeed", NULL))
+               mmc->caps |= MMC_CAP_MMC_HIGHSPEED;
+       if (of_get_property(np, "mmc-cap-sd-highspeed", NULL))
+               mmc->caps |= MMC_CAP_SD_HIGHSPEED;
+
+       return 0;
+}
+
+static int mmci_probe(struct amba_device *dev,
+       const struct amba_id *id)
+{
+       struct mmci_platform_data *plat = dev->dev.platform_data;
+       struct device_node *np = dev->dev.of_node;
+       struct variant_data *variant = id->data;
+       struct mmci_host *host;
+       struct mmc_host *mmc;
+       int ret;
+
+       /* Must have platform data or Device Tree. */
+       if (!plat && !np) {
+               dev_err(&dev->dev, "No plat data or DT found\n");
+               return -EINVAL;
+       }
+
+       if (!plat) {
+               plat = devm_kzalloc(&dev->dev, sizeof(*plat), GFP_KERNEL);
+               if (!plat)
+                       return -ENOMEM;
+       }
+
+       mmc = mmc_alloc_host(sizeof(struct mmci_host), &dev->dev);
+       if (!mmc)
+               return -ENOMEM;
+
+       ret = mmci_of_parse(np, mmc);
+       if (ret)
+               goto host_free;
+
+       host = mmc_priv(mmc);
+       host->mmc = mmc;
+
+       host->hw_designer = amba_manf(dev);
+       host->hw_revision = amba_rev(dev);
+       dev_dbg(mmc_dev(mmc), "designer ID = 0x%02x\n", host->hw_designer);
+       dev_dbg(mmc_dev(mmc), "revision = 0x%01x\n", host->hw_revision);
+
+       host->clk = devm_clk_get(&dev->dev, NULL);
+       if (IS_ERR(host->clk)) {
+               ret = PTR_ERR(host->clk);
+               goto host_free;
+       }
+
+       ret = clk_prepare_enable(host->clk);
+       if (ret)
+               goto host_free;
+
+       if (variant->qcom_fifo)
+               host->get_rx_fifocnt = mmci_qcom_get_rx_fifocnt;
+       else
+               host->get_rx_fifocnt = mmci_get_rx_fifocnt;
+
+       host->plat = plat;
+       host->variant = variant;
+       host->mclk = clk_get_rate(host->clk);
+       /*
+        * According to the spec, mclk is max 100 MHz,
+        * so we try to adjust the clock down to this,
+        * (if possible).
+        */
+       if (host->mclk > variant->f_max) {
+               ret = clk_set_rate(host->clk, variant->f_max);
+               if (ret < 0)
+                       goto clk_disable;
+               host->mclk = clk_get_rate(host->clk);
+               dev_dbg(mmc_dev(mmc), "eventual mclk rate: %u Hz\n",
+                       host->mclk);
+       }
+
+       host->phybase = dev->res.start;
+       host->base = devm_ioremap_resource(&dev->dev, &dev->res);
+       if (IS_ERR(host->base)) {
+               ret = PTR_ERR(host->base);
+               goto clk_disable;
+       }
+
+       /*
+        * The ARM and ST versions of the block have slightly different
+        * clock divider equations which means that the minimum divider
+        * differs too.
+        * on Qualcomm like controllers get the nearest minimum clock to 100Khz
+        */
+       if (variant->st_clkdiv)
+               mmc->f_min = DIV_ROUND_UP(host->mclk, 257);
+       else if (variant->explicit_mclk_control)
+               mmc->f_min = clk_round_rate(host->clk, 100000);
+       else
+               mmc->f_min = DIV_ROUND_UP(host->mclk, 512);
+       /*
+        * If no maximum operating frequency is supplied, fall back to use
+        * the module parameter, which has a (low) default value in case it
+        * is not specified. Either value must not exceed the clock rate into
+        * the block, of course.
+        */
+       if (mmc->f_max)
+               mmc->f_max = variant->explicit_mclk_control ?
+                               min(variant->f_max, mmc->f_max) :
+                               min(host->mclk, mmc->f_max);
+       else
+               mmc->f_max = variant->explicit_mclk_control ?
+                               fmax : min(host->mclk, fmax);
+
+
+       dev_dbg(mmc_dev(mmc), "clocking block at %u Hz\n", mmc->f_max);
+
+       /* Get regulators and the supported OCR mask */
+       ret = mmc_regulator_get_supply(mmc);
+       if (ret == -EPROBE_DEFER)
+               goto clk_disable;
+
+       if (!mmc->ocr_avail)
+               mmc->ocr_avail = plat->ocr_mask;
+       else if (plat->ocr_mask)
+               dev_warn(mmc_dev(mmc), "Platform OCR mask is ignored\n");
+
+       /* DT takes precedence over platform data. */
+       if (!np) {
+               if (!plat->cd_invert)
+                       mmc->caps2 |= MMC_CAP2_CD_ACTIVE_HIGH;
+               mmc->caps2 |= MMC_CAP2_RO_ACTIVE_HIGH;
+       }
+
+       /* We support these capabilities. */
+       mmc->caps |= MMC_CAP_CMD23;
+
+       if (variant->busy_detect) {
+               mmci_ops.card_busy = mmci_card_busy;
+               mmci_write_datactrlreg(host, MCI_ST_DPSM_BUSYMODE);
+               mmc->caps |= MMC_CAP_WAIT_WHILE_BUSY;
+               mmc->max_busy_timeout = 0;
+       }
+
+       mmc->ops = &mmci_ops;
+
+       /* We support these PM capabilities. */
+       mmc->pm_caps |= MMC_PM_KEEP_POWER;
+
+       /*
+        * We can do SGIO
+        */
+       mmc->max_segs = NR_SG;
+
+       /*
+        * Since only a certain number of bits are valid in the data length
+        * register, we must ensure that we don't exceed 2^num-1 bytes in a
+        * single request.
+        */
+       mmc->max_req_size = (1 << variant->datalength_bits) - 1;
+
+       /*
+        * Set the maximum segment size.  Since we aren't doing DMA
+        * (yet) we are only limited by the data length register.
+        */
+       mmc->max_seg_size = mmc->max_req_size;
+
+       /*
+        * Block size can be up to 2048 bytes, but must be a power of two.
+        */
+       mmc->max_blk_size = 1 << 11;
+
+       /*
+        * Limit the number of blocks transferred so that we don't overflow
+        * the maximum request size.
+        */
+       mmc->max_blk_count = mmc->max_req_size >> 11;
+
+       spin_lock_init(&host->lock);
+
+       writel(0, host->base + MMCIMASK0);
+       writel(0, host->base + MMCIMASK1);
+       writel(0xfff, host->base + MMCICLEAR);
+
+       /*
+        * If:
+        * - not using DT but using a descriptor table, or
+        * - using a table of descriptors ALONGSIDE DT, or
+        * look up these descriptors named "cd" and "wp" right here, fail
+        * silently of these do not exist and proceed to try platform data
+        */
+       if (!np) {
+               ret = mmc_gpiod_request_cd(mmc, "cd", 0, false, 0, NULL);
+               if (ret < 0) {
+                       if (ret == -EPROBE_DEFER)
+                               goto clk_disable;
+                       else if (gpio_is_valid(plat->gpio_cd)) {
+                               ret = mmc_gpio_request_cd(mmc, plat->gpio_cd, 0);
+                               if (ret)
+                                       goto clk_disable;
+                       }
+               }
+
+               ret = mmc_gpiod_request_ro(mmc, "wp", 0, false, 0, NULL);
+               if (ret < 0) {
+                       if (ret == -EPROBE_DEFER)
+                               goto clk_disable;
+                       else if (gpio_is_valid(plat->gpio_wp)) {
+                               ret = mmc_gpio_request_ro(mmc, plat->gpio_wp);
+                               if (ret)
+                                       goto clk_disable;
+                       }
+               }
+       }
+
+       ret = devm_request_irq(&dev->dev, dev->irq[0], mmci_irq, IRQF_SHARED,
+                       DRIVER_NAME " (cmd)", host);
+       if (ret)
+               goto clk_disable;
+
+       if (!dev->irq[1])
+               host->singleirq = true;
+       else {
+               ret = devm_request_irq(&dev->dev, dev->irq[1], mmci_pio_irq,
+                               IRQF_SHARED, DRIVER_NAME " (pio)", host);
+               if (ret)
+                       goto clk_disable;
+       }
+
+       writel(MCI_IRQENABLE, host->base + MMCIMASK0);
+
+       amba_set_drvdata(dev, mmc);
+
+       dev_info(&dev->dev, "%s: PL%03x manf %x rev%u at 0x%08llx irq %d,%d (pio)\n",
+                mmc_hostname(mmc), amba_part(dev), amba_manf(dev),
+                amba_rev(dev), (unsigned long long)dev->res.start,
+                dev->irq[0], dev->irq[1]);
+
+       mmci_dma_setup(host);
+
+       pm_runtime_set_autosuspend_delay(&dev->dev, 50);
+       pm_runtime_use_autosuspend(&dev->dev);
+
+       mmc_add_host(mmc);
+
+       pm_runtime_put(&dev->dev);
+       return 0;
+
+ clk_disable:
+       clk_disable_unprepare(host->clk);
+ host_free:
+       mmc_free_host(mmc);
+       return ret;
+}
+
+static int mmci_remove(struct amba_device *dev)
+{
+       struct mmc_host *mmc = amba_get_drvdata(dev);
+
+       if (mmc) {
+               struct mmci_host *host = mmc_priv(mmc);
+
+               /*
+                * Undo pm_runtime_put() in probe.  We use the _sync
+                * version here so that we can access the primecell.
+                */
+               pm_runtime_get_sync(&dev->dev);
+
+               mmc_remove_host(mmc);
+
+               writel(0, host->base + MMCIMASK0);
+               writel(0, host->base + MMCIMASK1);
+
+               writel(0, host->base + MMCICOMMAND);
+               writel(0, host->base + MMCIDATACTRL);
+
+               mmci_dma_release(host);
+               clk_disable_unprepare(host->clk);
+               mmc_free_host(mmc);
+       }
+
+       return 0;
+}
+
+#ifdef CONFIG_PM
+static void mmci_save(struct mmci_host *host)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&host->lock, flags);
+
+       writel(0, host->base + MMCIMASK0);
+       if (host->variant->pwrreg_nopower) {
+               writel(0, host->base + MMCIDATACTRL);
+               writel(0, host->base + MMCIPOWER);
+               writel(0, host->base + MMCICLOCK);
+       }
+       mmci_reg_delay(host);
+
+       spin_unlock_irqrestore(&host->lock, flags);
+}
+
+static void mmci_restore(struct mmci_host *host)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&host->lock, flags);
+
+       if (host->variant->pwrreg_nopower) {
+               writel(host->clk_reg, host->base + MMCICLOCK);
+               writel(host->datactrl_reg, host->base + MMCIDATACTRL);
+               writel(host->pwr_reg, host->base + MMCIPOWER);
+       }
+       writel(MCI_IRQENABLE, host->base + MMCIMASK0);
+       mmci_reg_delay(host);
+
+       spin_unlock_irqrestore(&host->lock, flags);
+}
+
+static int mmci_runtime_suspend(struct device *dev)
+{
+       struct amba_device *adev = to_amba_device(dev);
+       struct mmc_host *mmc = amba_get_drvdata(adev);
+
+       if (mmc) {
+               struct mmci_host *host = mmc_priv(mmc);
+               pinctrl_pm_select_sleep_state(dev);
+               mmci_save(host);
+               clk_disable_unprepare(host->clk);
+       }
+
+       return 0;
+}
+
+static int mmci_runtime_resume(struct device *dev)
+{
+       struct amba_device *adev = to_amba_device(dev);
+       struct mmc_host *mmc = amba_get_drvdata(adev);
+
+       if (mmc) {
+               struct mmci_host *host = mmc_priv(mmc);
+               clk_prepare_enable(host->clk);
+               mmci_restore(host);
+               pinctrl_pm_select_default_state(dev);
+       }
+
+       return 0;
+}
+#endif
+
+static const struct dev_pm_ops mmci_dev_pm_ops = {
+       SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
+                               pm_runtime_force_resume)
+       SET_RUNTIME_PM_OPS(mmci_runtime_suspend, mmci_runtime_resume, NULL)
+};
+
+static struct amba_id mmci_ids[] = {
+       {
+               .id     = 0x00041180,
+               .mask   = 0xff0fffff,
+               .data   = &variant_arm,
+       },
+       {
+               .id     = 0x01041180,
+               .mask   = 0xff0fffff,
+               .data   = &variant_arm_extended_fifo,
+       },
+       {
+               .id     = 0x02041180,
+               .mask   = 0xff0fffff,
+               .data   = &variant_arm_extended_fifo_hwfc,
+       },
+       {
+               .id     = 0x00041181,
+               .mask   = 0x000fffff,
+               .data   = &variant_arm,
+       },
+       /* ST Micro variants */
+       {
+               .id     = 0x00180180,
+               .mask   = 0x00ffffff,
+               .data   = &variant_u300,
+       },
+       {
+               .id     = 0x10180180,
+               .mask   = 0xf0ffffff,
+               .data   = &variant_nomadik,
+       },
+       {
+               .id     = 0x00280180,
+               .mask   = 0x00ffffff,
+               .data   = &variant_u300,
+       },
+       {
+               .id     = 0x00480180,
+               .mask   = 0xf0ffffff,
+               .data   = &variant_ux500,
+       },
+       {
+               .id     = 0x10480180,
+               .mask   = 0xf0ffffff,
+               .data   = &variant_ux500v2,
+       },
+       /* Qualcomm variants */
+       {
+               .id     = 0x00051180,
+               .mask   = 0x000fffff,
+               .data   = &variant_qcom,
+       },
+       { 0, 0 },
+};
+
+MODULE_DEVICE_TABLE(amba, mmci_ids);
+
+static struct amba_driver mmci_driver = {
+       .drv            = {
+               .name   = DRIVER_NAME,
+               .pm     = &mmci_dev_pm_ops,
+       },
+       .probe          = mmci_probe,
+       .remove         = mmci_remove,
+       .id_table       = mmci_ids,
+};
+
+module_amba_driver(mmci_driver);
+
+module_param(fmax, uint, 0444);
+
+MODULE_DESCRIPTION("ARM PrimeCell PL180/181 Multimedia Card Interface driver");
+MODULE_LICENSE("GPL");