Re: [PATCH 1/2] HSI: add ST-E HSI controller
From: Carlos Chinea
Date: Mon May 30 2011 - 10:41:26 EST
Hi,
My 2 cents for this patch bellow:
On Wed, 2011-05-25 at 18:27 +0200, ext Linus Walleij wrote:
> From: Pawel Szyszuk <pawel.szyszuk@xxxxxxxxxxxxxx>
>
> Signed-off-by: Pawel Szyszuk <pawel.szyszuk@xxxxxxxxxxxxxx>
> Signed-off-by: Linus Walleij <linus.walleij@xxxxxxxxxx>
> ---
> drivers/hsi/controllers/Kconfig | 10 +
> drivers/hsi/controllers/Makefile | 1 +
> drivers/hsi/controllers/ste_hsi.c | 1536 +++++++++++++++++++++++++++++++++++++
> 3 files changed, 1547 insertions(+), 0 deletions(-)
> create mode 100644 drivers/hsi/controllers/ste_hsi.c
>
> diff --git a/drivers/hsi/controllers/Kconfig b/drivers/hsi/controllers/Kconfig
> index 3efe0f0..76d339e 100644
> --- a/drivers/hsi/controllers/Kconfig
> +++ b/drivers/hsi/controllers/Kconfig
> @@ -3,6 +3,16 @@
> #
> comment "HSI controllers"
>
> +config STE_HSI
> + tristate "STE HSI controller driver"
> + depends on (ARCH_U8500 || ARCH_NOMADIK) && HSI
> + default n
> + help
> + ST-Ericsson HSI controller.
> + If you say Y here, you will enable the U8500 HSI hardware driver.
> +
> + If unsure, say N.
> +
> config OMAP_SSI
> tristate "OMAP SSI hardware driver"
> depends on ARCH_OMAP && HSI
> diff --git a/drivers/hsi/controllers/Makefile b/drivers/hsi/controllers/Makefile
> index c4ba2c2..475637a 100644
> --- a/drivers/hsi/controllers/Makefile
> +++ b/drivers/hsi/controllers/Makefile
> @@ -2,4 +2,5 @@
> # Makefile for HSI controllers drivers
> #
>
> +obj-$(CONFIG_STE_HSI) += ste_hsi.o
> obj-$(CONFIG_OMAP_SSI) += omap_ssi.o
> diff --git a/drivers/hsi/controllers/ste_hsi.c b/drivers/hsi/controllers/ste_hsi.c
> new file mode 100644
> index 0000000..1a16234
> --- /dev/null
> +++ b/drivers/hsi/controllers/ste_hsi.c
> @@ -0,0 +1,1536 @@
> +/*
> + * Copyright (C) ST-Ericsson SA 2010
> + *
> + * License Terms: GNU General Public License v2
> + * Author: Marcin Mielczarczyk <marcin.mielczarczyk@xxxxxxxxx> for ST-Ericsson
> + * Author: Lukasz Baj <lukasz.baj@xxxxxxxxx> for ST-Ericsson
> + */
> +#include <linux/kernel.h>
> +#include <linux/slab.h>
> +#include <linux/clk.h>
> +#include <linux/err.h>
> +#include <linux/interrupt.h>
> +#include <linux/io.h>
> +#include <linux/platform_device.h>
> +#include <linux/hsi/hsi.h>
> +
> +#ifdef CONFIG_STE_DMA40
> +#include <linux/dmaengine.h>
> +#include <linux/dma-mapping.h>
> +#endif
> +
> +#include <mach/hsi.h>
> +
> +/**
> + * struct ste_hsi_controller - Nomadik HSI controller data
> + * @dev: device associated to the controller (HSI controller)
> + * @rx_base: HSI receiver registers base address
> + * @tx_base: HSI transmitter registers base address
> + */
You may want to add few more fields ;)
> +struct ste_hsi_controller {
> + struct device *dev;
> + struct clk *tx_clk;
> + struct clk *rx_clk;
> + struct clk *ssitx_clk;
> + struct clk *ssirx_clk;
> + struct delayed_work clk_work;
> + unsigned char __iomem *rx_base;
> + unsigned char __iomem *tx_base;
Hmmm why the above two lines are not void __iomem * ?
That should also remove some typecasting later on.
> + int overrun_irq[STE_HSI_MAX_CHANNELS];
> + int ck_refcount;
> + spinlock_t ck_lock;
> + spinlock_t lock;
> + unsigned int use_dma:1;
> + unsigned int ck_on:1;
> + /* physical address of rx and tx controller */
> + dma_addr_t rx_dma_base;
> + dma_addr_t tx_dma_base;
> +};
> +
> +#ifdef CONFIG_STE_DMA40
> +struct ste_hsi_channel_dma {
> + struct dma_chan *dma_chan;
> + struct dma_async_tx_descriptor *desc;
> + dma_cookie_t cookie;
> +};
> +#endif
> +
> +struct ste_hsi_port {
> + struct device *dev;
> + struct list_head txqueue[STE_HSI_MAX_CHANNELS];
> + struct list_head rxqueue[STE_HSI_MAX_CHANNELS];
> + struct list_head brkqueue;
> + int tx_irq;
> + int rx_irq;
> + int excep_irq;
> + struct tasklet_struct rx_tasklet;
> + struct tasklet_struct tx_tasklet;
> + struct tasklet_struct exception_tasklet;
> + struct tasklet_struct overrun_tasklet;
> + unsigned char channels;
> +#ifdef CONFIG_STE_DMA40
> + struct ste_hsi_channel_dma tx_dma[STE_HSI_MAX_CHANNELS];
> + struct ste_hsi_channel_dma rx_dma[STE_HSI_MAX_CHANNELS];
> +#endif
> +};
> +
Adding some documentation for the above structs would be nice.
> +#define hsi_to_ste_port(port) (hsi_port_drvdata(port))
> +#define hsi_to_ste_controller(con) (hsi_controller_drvdata(con))
> +#define client_to_ste_port(cl) (hsi_port_drvdata(hsi_get_port(cl)))
> +#define client_to_hsi(cl) \
> + (to_hsi_controller(hsi_get_port(cl)->device.parent))
> +#define client_to_ste_controller(cl) \
> + (hsi_controller_drvdata(client_to_hsi(cl)))
> +#define ste_port_to_ste_controller(port) \
> + ((struct ste_hsi_controller *)hsi_controller_drvdata( \
> + to_hsi_controller(port->dev->parent)))
> +
> +static u32 ste_hsir_periphid[8] = { 0x2C, 0, 0x8, 0x18, 0xD, 0xF0, 0x5, 0xB1 };
> +static u32 ste_hsit_periphid[8] = { 0x2B, 0, 0x8, 0x18, 0xD, 0xF0, 0x5, 0xB1 };
Why do these two arrays need to be global ?
Moreover, I think they are potentially platform_data material.
> +
> +/*
> + * linux/amba/bus.h macros can not be used, because 8 bytes are validated:
> + * PERIPHID0..3 and PCELLID0..3 for HSIR and HSIT.
> + */
> +static inline int compare_periphid(u32 *id1, u32 *id2, int count)
> +{
> + while (count && *id1++ == *id2++)
> + count--;
> +
> + return count;
> +}
Seeing later what you are using this function for, I will recommend
something like:
static inline int compare_periphid(u32 *id1, void __iomem *id2, unsigned
int count)
{
while (count && *id1++ == readl(id2)) {
count--;
id2 += sizeof(*id1);
}
return count;
}
> +
> +static void ste_hsi_clk_free(struct clk **pclk)
> +{
> + if (IS_ERR(*pclk) && *pclk != NULL)
I guess this condition should be: (!IS_ERR(*pclk) && *pclk != NULL)
> + clk_put(*pclk);
> + *pclk = NULL;
> +}
> +
> +static void ste_hsi_clks_free(struct ste_hsi_controller *ste_hsi)
> +{
> + ste_hsi_clk_free(&ste_hsi->rx_clk);
> + ste_hsi_clk_free(&ste_hsi->tx_clk);
> + ste_hsi_clk_free(&ste_hsi->ssirx_clk);
> + ste_hsi_clk_free(&ste_hsi->ssitx_clk);
> +}
> +
> +static int ste_hsi_clock_enable(struct hsi_controller *hsi)
> +{
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + int err = 0;
> +
> + spin_lock_bh(&ste_hsi->ck_lock);
> + if (ste_hsi->ck_refcount++ || ste_hsi->ck_on)
If you have refcounting for the clocks, why do yo need the ck_on flag ?
> + goto out;
> +
> + err = clk_enable(ste_hsi->ssirx_clk);
> + if (unlikely(err))
> + goto out;
> +
> + err = clk_enable(ste_hsi->ssitx_clk);
> + if (unlikely(err))
> + clk_disable(ste_hsi->ssirx_clk);
> +
> + err = clk_enable(ste_hsi->rx_clk);
> + if (unlikely(err)) {
> + clk_disable(ste_hsi->ssitx_clk);
> + clk_disable(ste_hsi->ssirx_clk);
> + }
> +
> + err = clk_enable(ste_hsi->tx_clk);
> + if (unlikely(err)) {
> + clk_disable(ste_hsi->rx_clk);
> + clk_disable(ste_hsi->ssitx_clk);
> + clk_disable(ste_hsi->ssirx_clk);
> +
Maybe better to continue using goto statements here and in the above
conditions and have the whole cleanup at the end of the function.
> }
> +
> + ste_hsi->ck_on = 1;
> +out:
> + if (err)
> + ste_hsi->ck_refcount--;
> +
> + spin_unlock_bh(&ste_hsi->ck_lock);
> +
> + return err;
> +}
> +
> +static void ste_hsi_delayed_disable_clock(struct work_struct *work)
> +{
> + struct ste_hsi_controller *ste_hsi;
> + ste_hsi = container_of(work, struct ste_hsi_controller, clk_work.work);
> +
> + spin_lock_bh(&ste_hsi->ck_lock);
> +
> + /*
> + * If clock should not be off (enable clock called in meantime)
> + * or clock is already off nothing to do
> + */
> + if (ste_hsi->ck_refcount || !ste_hsi->ck_on)
Ohh I see why you want ck_on now, but yet again, why don't you increase
the refcounting when you delay the disabling of the clocks... ... then
this function will be pretty much a one liner calling
sti_hsi_clock_disable()
> + goto out;
> +
> + if (readl(ste_hsi->tx_base + STE_HSI_TX_STATE) != STE_HSI_STATE_IDLE ||
> + readl(ste_hsi->rx_base + STE_HSI_RX_STATE)
> + != STE_HSI_STATE_IDLE ||
> + readl(ste_hsi->rx_base + STE_HSI_RX_BUFSTATE) != 0) {
> + /* Try again later */
> + int err = schedule_delayed_work(&ste_hsi->clk_work, HZ);
> + if (err < 0)
> + dev_err(ste_hsi->dev, "Error scheduling work\n");
> + goto out;
> + }
> +
> + /* Actual clocks disable */
> + clk_disable(ste_hsi->tx_clk);
> + clk_disable(ste_hsi->rx_clk);
> + clk_disable(ste_hsi->ssitx_clk);
> + clk_disable(ste_hsi->ssirx_clk);
> + ste_hsi->ck_on = 0;
> +
> +out:
> + spin_unlock_bh(&ste_hsi->ck_lock);
> +}
> +
> +static void ste_hsi_clock_disable(struct hsi_controller *hsi)
> +{
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> +
> + spin_lock_bh(&ste_hsi->ck_lock);
> +
> + /* Sanity check */
> + if (ste_hsi->ck_refcount <= 0)
> + WARN_ON(ste_hsi->ck_refcount <= 0);
> +
The if statement is redundant. I will suggest also WARN_ON_ONCE().
Otherwise once this is trigger you can easily have an storm of them.
> + /* Need clock to be disable now? */
> + if (--ste_hsi->ck_refcount)
> + goto out;
> +
> + /*
> + * If receiver or transmitter is in the middle something delay clock off
> + */
> + if (readl(ste_hsi->tx_base + STE_HSI_TX_STATE) != STE_HSI_STATE_IDLE ||
> + readl(ste_hsi->rx_base + STE_HSI_RX_STATE)
> + != STE_HSI_STATE_IDLE ||
> + readl(ste_hsi->rx_base + STE_HSI_RX_BUFSTATE) != 0) {
> + int err = schedule_delayed_work(&ste_hsi->clk_work, HZ);
> + if (err < 0)
> + dev_err(&hsi->device, "Error scheduling work\n");
> +
> + goto out;
> + }
> +
> + /* Actual clocks disabled */
> + clk_disable(ste_hsi->tx_clk);
> + clk_disable(ste_hsi->rx_clk);
> + clk_disable(ste_hsi->ssitx_clk);
> + clk_disable(ste_hsi->ssirx_clk);
> + ste_hsi->ck_on = 0;
> +
> +out:
> + spin_unlock_bh(&ste_hsi->ck_lock);
> +}
> +
> +static int ste_hsi_start_irq(struct hsi_msg *msg)
> +{
> + struct hsi_port *port = hsi_get_port(msg->cl);
> + struct hsi_controller *hsi = to_hsi_controller(port->device.parent);
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + u32 val;
> + int err;
> +
> + err = ste_hsi_clock_enable(hsi);
> + if (unlikely(err))
> + return err;
> +
> + msg->actual_len = 0;
> + msg->status = HSI_STATUS_PROCEEDING;
> +
> + if (msg->ttype == HSI_MSG_WRITE) {
> + val = readl(ste_hsi->tx_base + STE_HSI_TX_WATERMARKIM) |
> + (1 << msg->channel);
> + writel(val, ste_hsi->tx_base + STE_HSI_TX_WATERMARKIM);
> + } else {
> + val = readl(ste_hsi->rx_base + STE_HSI_RX_WATERMARKIM) |
> + (1 << msg->channel);
> + writel(val, ste_hsi->rx_base + STE_HSI_RX_WATERMARKIM);
> +
> + val = readl(ste_hsi->rx_base + STE_HSI_RX_OVERRUNIM) |
> + (1 << msg->channel);
> + writel(val, ste_hsi->rx_base + STE_HSI_RX_OVERRUNIM);
> + }
> +
> + return 0;
> +}
> +
> +static int ste_hsi_start_transfer(struct ste_hsi_port *ste_port,
> + struct list_head *queue);
> +#ifdef CONFIG_STE_DMA40
> +static void ste_hsi_dma_callback(void *dma_async_param)
> +{
> + struct hsi_msg *msg = dma_async_param;
> + struct hsi_controller *hsi = client_to_hsi(msg->cl);
> + struct ste_hsi_port *ste_port = client_to_ste_port(msg->cl);
> + struct ste_hsi_controller *ste_hsi = client_to_ste_controller(msg->cl);
> + struct list_head *queue;
> + struct dma_chan *chan;
> + struct ste_hsi_channel_dma *hsi_dma_chan;
> + char *dma_enable_address;
> + enum dma_data_direction direction;
> + u32 dma_mask;
> +
> + /* Message finished, remove from list and notify client */
> + spin_lock_bh(&ste_hsi->lock);
> + list_del(&msg->link);
> +
> + if (msg->ttype == HSI_MSG_WRITE) {
> + queue = &ste_port->txqueue[msg->channel];
> + direction = DMA_TO_DEVICE;
> + dma_enable_address = ste_hsi->tx_base + STE_HSI_TX_DMAEN;
> + hsi_dma_chan = &ste_port->tx_dma[msg->channel];
> + } else {
> + queue = &ste_port->rxqueue[msg->channel];
> + direction = DMA_FROM_DEVICE;
> + dma_enable_address = ste_hsi->rx_base + STE_HSI_RX_DMAEN;
> + hsi_dma_chan = &ste_port->rx_dma[msg->channel];
> + }
> +
> + dma_sync_sg_for_cpu(&hsi->device, msg->sgt.sgl,
> + msg->sgt.nents, direction);
> + chan = hsi_dma_chan->dma_chan;
> +
> + /* disable DMA channel on HSI controller */
> + dma_mask = readl(dma_enable_address);
> + writel(dma_mask & ~(1 << msg->channel), dma_enable_address);
> +
> + hsi_dma_chan->desc = NULL;
> +
> + dma_unmap_sg(&hsi->device, msg->sgt.sgl, msg->sgt.nents, direction);
> +
> + msg->status = HSI_STATUS_COMPLETED;
> + msg->actual_len = sg_dma_len(msg->sgt.sgl);
> +
> + spin_unlock_bh(&ste_hsi->lock);
> +
> + msg->complete(msg);
> +
> + ste_hsi_clock_disable(hsi);
> +
> + spin_lock_bh(&ste_hsi->lock);
> + ste_hsi_start_transfer(ste_port, queue);
> + spin_unlock_bh(&ste_hsi->lock);
> +}
> +
> +static void dma_device_control(struct ste_hsi_channel_dma *chan,
> + enum dma_ctrl_cmd cmd, unsigned long arg)
> +{
> + chan->dma_chan->device->device_control(chan->dma_chan, cmd, arg);
> +}
> +
> +static void ste_hsi_terminate_dma_chan(struct ste_hsi_channel_dma *chan)
> +{
> + if (chan->desc) {
> + dma_device_control(chan, DMA_TERMINATE_ALL, 0);
> + chan->desc = NULL;
> + }
> + chan->cookie = 0;
> +}
> +
> +static void ste_hsi_terminate_dma(struct ste_hsi_port *ste_port)
> +{
> + int i;
> +
> + for (i = 0; i < ste_port->channels; ++i) {
> + ste_hsi_terminate_dma_chan(&ste_port->tx_dma[i]);
> + ste_hsi_terminate_dma_chan(&ste_port->rx_dma[i]);
> + }
> +}
> +
> +static int ste_hsi_start_dma(struct hsi_msg *msg)
> +{
> + struct hsi_controller *hsi = client_to_hsi(msg->cl);
> + struct ste_hsi_port *ste_port = client_to_ste_port(msg->cl);
> + struct ste_hsi_controller *ste_hsi = client_to_ste_controller(msg->cl);
> + struct dma_async_tx_descriptor *desc;
> + struct dma_chan *chan;
> + struct ste_hsi_channel_dma *hsi_dma_chan;
> + char *dma_enable_address;
> + enum dma_data_direction direction;
> + u32 dma_mask;
> + int err;
> +
> + err = ste_hsi_clock_enable(hsi);
> + if (unlikely(err))
> + return err;
> +
> + if (msg->ttype == HSI_MSG_WRITE) {
> + direction = DMA_TO_DEVICE;
> + dma_enable_address = ste_hsi->tx_base + STE_HSI_TX_DMAEN;
> + hsi_dma_chan = &ste_port->tx_dma[msg->channel];
> + } else {
> + u32 val;
> + direction = DMA_FROM_DEVICE;
> + dma_enable_address = ste_hsi->rx_base + STE_HSI_RX_DMAEN;
> + hsi_dma_chan = &ste_port->rx_dma[msg->channel];
> +
> + /* enable overrun for this channel */
> + val = readl(ste_hsi->rx_base + STE_HSI_RX_OVERRUNIM) |
> + (1 << msg->channel);
> + writel(val, ste_hsi->rx_base + STE_HSI_RX_OVERRUNIM);
> + }
> +
> + chan = hsi_dma_chan->dma_chan;
> +
> + if (0 == dma_map_sg(&hsi->device, msg->sgt.sgl, msg->sgt.nents,
> + direction)) {
> + dev_dbg(&hsi->device, "DMA map SG failed !\n");
> + err = -ENOMEM;
> + goto out;
> + }
> + /* Prepare the scatterlist */
> + desc = chan->device->device_prep_slave_sg(chan,
> + msg->sgt.sgl,
> + msg->sgt.nents,
> + direction,
> + DMA_PREP_INTERRUPT |
> + DMA_CTRL_ACK);
> +
> + if (!desc) {
> + dma_unmap_sg(&hsi->device, msg->sgt.sgl, msg->sgt.nents,
> + direction);
> + /* "Complete" DMA (errorpath) */
> + ste_hsi_terminate_dma_chan(hsi_dma_chan);
> + err = -EBUSY;
> + goto out;
> + }
> + desc->callback = ste_hsi_dma_callback;
> + desc->callback_param = msg;
> + hsi_dma_chan->cookie = desc->tx_submit(desc);
> + hsi_dma_chan->desc = desc;
> +
> + /* Fire the DMA transaction */
> + chan->device->device_issue_pending(chan);
> +
> + /* Enable DMA channel on HSI controller */
> + dma_mask = readl(dma_enable_address);
> + writel(dma_mask | 1 << msg->channel, dma_enable_address);
> +
> +out:
> + if (unlikely(err))
> + ste_hsi_clock_disable(hsi);
> +
> + return err;
> +}
> +
> +static void __init ste_hsi_init_dma(struct ste_hsi_platform_data *data,
> + struct hsi_controller *hsi)
> +{
> + struct hsi_port *port;
> + struct ste_hsi_port *ste_port;
> + struct ste_hsi_controller *ste_hsi = hsi_to_ste_controller(hsi);
> + dma_cap_mask_t mask;
> + int i, ch;
> +
> + ste_hsi->use_dma = 1;
> + /* Try to acquire a generic DMA engine slave channel */
> + dma_cap_zero(mask);
> + dma_cap_set(DMA_SLAVE, mask);
> +
> + for (i = 0; i < hsi->num_ports; ++i) {
> + port = &hsi->port[i];
> + ste_port = hsi_port_drvdata(port);
> +
> + for (ch = 0; ch < STE_HSI_MAX_CHANNELS; ++ch) {
> + ste_port->tx_dma[ch].dma_chan =
> + dma_request_channel(mask,
> + data->port_cfg[i].dma_filter,
> + &data->port_cfg[i].
> + dma_tx_cfg[ch]);
> +
> + ste_port->rx_dma[ch].dma_chan =
> + dma_request_channel(mask,
> + data->port_cfg[i].dma_filter,
> + &data->port_cfg[i].
> + dma_rx_cfg[ch]);
> + }
> + }
> +}
> +
> +static int ste_hsi_setup_dma(struct hsi_client *cl)
> +{
> + int i;
> + struct hsi_port *port = to_hsi_port(cl->device.parent);
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> + struct hsi_controller *hsi = to_hsi_controller(port->device.parent);
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + struct dma_slave_config rx_conf = {
> + .src_addr = 0, /* dynamic data */
> + .src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES,
> + .direction = DMA_FROM_DEVICE,
> + .src_maxburst = 1,
> + };
> + struct dma_slave_config tx_conf = {
> + .dst_addr = 0, /* dynamic data */
> + .dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES,
> + .direction = DMA_TO_DEVICE,
> + .dst_maxburst = 1,
> + };
> +
> + if (!ste_hsi->use_dma)
> + return 0;
> +
> + for (i = 0; i < ste_port->channels; ++i) {
> + struct dma_chan *chan;
> +
> + chan = ste_port->tx_dma[i].dma_chan;
> + tx_conf.dst_addr = (dma_addr_t) ste_hsi->tx_dma_base +
> + STE_HSI_TX_BUFFERX + 4 * i;
> + chan->device->device_control(chan,
> + DMA_SLAVE_CONFIG,
> + (unsigned long)&tx_conf);
> +
> + chan = ste_port->rx_dma[i].dma_chan;
> + rx_conf.src_addr = (dma_addr_t) ste_hsi->rx_dma_base +
> + STE_HSI_RX_BUFFERX + 4 * i;
> + chan->device->device_control(chan,
> + DMA_SLAVE_CONFIG,
> + (unsigned long)&rx_conf);
> + }
> +
> + return 0;
> +}
> +
> +#else
> +#define ste_hsi_init_dma(data, hsi) do { } while (0)
> +#define ste_hsi_start_dma ste_hsi_start_irq
> +#define ste_hsi_terminate_dma(ste_port) do { } while (0)
> +#define ste_hsi_setup_dma(cl) do { } while (0)
> +#endif
> +
> +static int ste_hsi_start_transfer(struct ste_hsi_port *ste_port,
> + struct list_head *queue)
> +{
> + struct hsi_msg *msg;
> + int err;
> +
> + if (list_empty(queue))
> + return 0;
> +
> + msg = list_first_entry(queue, struct hsi_msg, link);
> + if (msg->status != HSI_STATUS_QUEUED)
> + return 0;
> +
> + msg->actual_len = 0;
> + msg->status = HSI_STATUS_PROCEEDING;
> +
> + if (ste_port_to_ste_controller(ste_port)->use_dma)
> + err = ste_hsi_start_dma(msg);
> + els
> + err = ste_hsi_start_irq(msg);
> +
> + return err;
> +}
> +
> +static void ste_hsi_receive_data(struct hsi_port *port, unsigned int channel)
> +{
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> + struct hsi_controller *hsi = to_hsi_controller(port->device.parent);
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + struct list_head *queue = &ste_port->rxqueue[channel];
> + struct hsi_msg *msg;
> + char *bufferx;
> + u8 *buf;
> + int span;
> +
> + spin_lock_bh(&ste_hsi->lock);
spin_lock() is enough here, this is always called inside a tasklet,
right ?
> +
> + if (list_empty(queue))
> + goto out;
> +
> + msg = list_first_entry(queue, struct hsi_msg, link);
> + if ((!msg->sgt.nents) || (!msg->sgt.sgl->length)) {
> + msg->actual_len = 0;
> + msg->status = HSI_STATUS_PENDING;
> + }
> +
> + if (msg->status == HSI_STATUS_PROCEEDING && msg->ttype == HSI_MSG_READ) {
> + unsigned char len;
> + bufferx = ste_hsi->rx_base + STE_HSI_RX_BUFFERX + 4 * channel;
> +
> + len = readl(ste_hsi->rx_base + STE_HSI_RX_GAUGEX + 4 * channel);
> + buf = sg_virt(msg->sgt.sgl);
> + buf += msg->actual_len;
> + while (len--) {
> + *(u32 *) buf = readl(bufferx);
> + buf += 4;
> + msg->actual_len += 4;
> + if (msg->actual_len >= msg->sgt.sgl->length) {
> + msg->status = HSI_STATUS_COMPLETED;
> + break;
> + }
> + }
> + }
> +
> + /* re-enable interrupt by watermark manipulation */
> + span = readl(ste_hsi->rx_base + STE_HSI_RX_SPANX + 4 * channel);
> + writel(span, ste_hsi->rx_base + STE_HSI_RX_WATERMARKX + 4 * channel);
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_WATERMARKX + 4 * channel);
> +
> + /*
> + * If message was not transmitted completely enable interrupt for
> + * further work
> + */
> + if (msg->status == HSI_STATUS_PROCEEDING) {
> + u32 val;
> + val = readl(ste_hsi->rx_base + STE_HSI_RX_WATERMARKIM) |
> + (1 << channel);
> + writel(val, ste_hsi->rx_base + STE_HSI_RX_WATERMARKIM);
> + goto out;
> + }
> +
> + /* Message finished, remove from list and notify client */
> + list_del(&msg->link);
> + spin_unlock_bh(&ste_hsi->lock);
ditto
> + msg->complete(msg);
> +
> + ste_hsi_clock_disable(hsi);
> +
> + spin_lock_bh(&ste_hsi->lock);
> +
ditto
> + ste_hsi_start_transfer(ste_port, queue);
> +out:
> + spin_unlock_bh(&ste_hsi->lock);
> +}
> +
> +static void ste_hsi_transmit_data(struct hsi_port *port, unsigned int channel)
> +{
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> + struct hsi_controller *hsi = to_hsi_controller(port->device.parent);
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + struct list_head *queue = &ste_port->txqueue[channel];
> + struct hsi_msg *msg;
> + u8 *buf;
> + int span;
> +
> + if (list_empty(queue))
> + return;
> +
> + spin_lock_bh(&ste_hsi->lock);
ditto
> + msg = list_first_entry(queue, struct hsi_msg, link);
> + if ((!msg->sgt.nents) || (!msg->sgt.sgl->length)) {
> + msg->actual_len = 0;
> + msg->status = HSI_STATUS_PENDING;
> + }
> +
> + if (msg->status == HSI_STATUS_PROCEEDING &&
> + msg->ttype == HSI_MSG_WRITE) {
> + unsigned char free_space;
> +
> + free_space = readl(ste_hsi->tx_base +
> + STE_HSI_TX_GAUGEX + 4 * channel);
> + buf = sg_virt(msg->sgt.sgl);
> + buf += msg->actual_len;
> + while (free_space--) {
> + writel(*(u32 *) buf, ste_hsi->tx_base +
> + STE_HSI_TX_BUFFERX + 4 * channel);
> + buf += 4;
> + msg->actual_len += 4;
> + if (msg->actual_len >= msg->sgt.sgl->length) {
> + msg->status = HSI_STATUS_COMPLETED;
> + break;
> + }
> + }
> + }
> +
> + span = readl(ste_hsi->tx_base + STE_HSI_TX_SPANX + 4 * channel);
> + writel(span, ste_hsi->tx_base + STE_HSI_TX_WATERMARKX + 4 * channel);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_WATERMARKX + 4 * channel);
> +
> + if (msg->status == HSI_STATUS_PROCEEDING) {
> + u32 val;
> + val = readl(ste_hsi->tx_base + STE_HSI_TX_WATERMARKIM) |
> + (1 << channel);
> + writel(val, ste_hsi->tx_base + STE_HSI_TX_WATERMARKIM);
> + goto out;
> + }
> +
> + list_del(&msg->link);
> + spin_unlock_bh(&ste_hsi->lock);
> +
ditto
> msg->complete(msg);
> +
> + ste_hsi_clock_disable(hsi);
> +
> + spin_lock_bh(&ste_hsi->lock);
> +
ditto
> ste_hsi_start_transfer(ste_port, queue);
> +out:
> + spin_unlock_bh(&ste_hsi->lock);
ditto
> +}
> +
> +static void ste_hsi_rx_tasklet(unsigned long data)
> +{
> + struct hsi_port *port = (struct hsi_port *)data;
> + struct hsi_controller *hsi = to_hsi_controller(port->device.parent);
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> + u32 irq_status, irq_mask;
> + unsigned int i;
> +
> + irq_status = readl(ste_hsi->rx_base + STE_HSI_RX_WATERMARKIS);
> + if (!irq_status)
> + goto out;
> +
> + irq_mask = readl(ste_hsi->rx_base + STE_HSI_RX_WATERMARKIM);
> + writel(irq_mask & ~irq_status,
> + ste_hsi->rx_base + STE_HSI_RX_WATERMARKIM);
> + writel(irq_mask, ste_hsi->rx_base + STE_HSI_RX_WATERMARKIC);
> +
> + for (i = 0; i < STE_HSI_MAX_CHANNELS; ++i) {
> + if (irq_status & (1 << i))
> + ste_hsi_receive_data(port, i);
> + }
> +out:
> + enable_irq(ste_port->rx_irq);
> +}
> +
> +static irqreturn_t ste_hsi_rx_isr(int irq, void *data)
> +{
> + struct hsi_port *port = data;
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> +
> + disable_irq_nosync(irq);
> + tasklet_hi_schedule(&ste_port->rx_tasklet);
> +
> + return IRQ_HANDLED;
> +}
> +
> +static irqreturn_t ste_hsi_tx_isr(int irq, void *data)
> +{
> + struct hsi_port *port = data;
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> +
> + disable_irq_nosync(irq);
> + tasklet_hi_schedule(&ste_port->tx_tasklet);
> +
> + return IRQ_HANDLED;
> +}
> +
> +static void ste_hsi_tx_tasklet(unsigned long data)
> +{
> + struct hsi_port *port = (struct hsi_port *)data;
> + struct hsi_controller *hsi = to_hsi_controller(port->device.parent);
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> + u32 irq_status, irq_mask;
> + unsigned int i;
> +
> + irq_status = readl(ste_hsi->tx_base + STE_HSI_TX_WATERMARKIS);
> + if (!irq_status)
> + goto out;
> +
> + irq_mask = readl(ste_hsi->tx_base + STE_HSI_TX_WATERMARKIM);
> + writel(irq_mask & ~irq_status,
> + ste_hsi->tx_base + STE_HSI_TX_WATERMARKIM);
> + writel(irq_mask, ste_hsi->tx_base + STE_HSI_TX_WATERMARKIC);
> +
> + for (i = 0; i < STE_HSI_MAX_CHANNELS; ++i) {
> + if (irq_status & (1 << i))
> + ste_hsi_transmit_data(port, i);
> + }
> +out:
> + enable_irq(ste_port->tx_irq);
> +}
> +
> +static void ste_hsi_break_complete(struct hsi_port *port,
> + struct ste_hsi_controller *ste_hsi)
> +{
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> + struct hsi_msg *msg, *tmp;
> + u32 mask;
> +
> + dev_dbg(port->device.parent, "HWBREAK received\n");
> +
> + spin_lock_bh(&ste_hsi->lock);
ditto
> +
> + mask = readl(ste_hsi->rx_base + STE_HSI_RX_EXCEPIM);
> + writel(mask & ~STE_HSI_EXCEP_BREAK,
> + ste_hsi->rx_base + STE_HSI_RX_EXCEPIM);
> +
> + spin_unlock_bh(&ste_hsi->lock);
ditto
> +
> + list_for_each_entry_safe(msg, tmp, &ste_port->brkqueue, link) {
> + msg->status = HSI_STATUS_COMPLETED;
> + list_del(&msg->link);
> + msg->complete(msg);
> + }
This can create infinite loop if two clients rearm the HWBREAK in their
completion callback, plus list_del() should be protected with lock. The
same infinite loop issue is in the OMAP driver.
I will suggest something like:
LIST_HEAD(bq);
....
list_splice_init(&ste_port->brkqueue, bq);
spin_unlock(&ste_hsi->lock);
list_for_each_entry_safe(msg, tmp, &bq, link) {
...
> +}
> +
> +static void ste_hsi_error(struct hsi_port *port)
> +
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> + struct hsi_msg *msg;
> + unsigned int i;
> +
> + for (i = 0; i < ste_port->channels; i++) {
> + if (list_empty(&ste_port->rxqueue[i]))
> + continue;
> + msg = list_first_entry(&ste_port->rxqueue[i], struct hsi_msg,
> + link);
> + list_del(&msg->link);
> + msg->status = HSI_STATUS_ERROR;
> + msg->complete(msg);
> + /* Now restart queued reads if any */
> + ste_hsi_start_transfer(ste_port, &ste_port->rxqueue[i]);
> + }
> +}
> +
> +static void ste_hsi_exception_tasklet(unsigned long data)
> +{
> + struct hsi_port *port = (struct hsi_port *)data;
> + struct hsi_controller *hsi = to_hsi_controller(port->device.parent);
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> + u32 error_status;
> + u32 error_interrupts;
> +
> + error_status = readl(ste_hsi->rx_base + STE_HSI_RX_EXCEP);
> + /*
> + * sometimes interrupt that cause running this tasklet is already
> + * inactive so base handling of exception on masked interrupt status
> + * not on exception state register.
> + */
> + error_interrupts = readl(ste_hsi->rx_base + STE_HSI_RX_EXCEPMIS);
> +
> + if (error_interrupts & STE_HSI_EXCEP_BREAK)
> + ste_hsi_break_complete(port, ste_hsi);
> +
> + if (error_interrupts & STE_HSI_EXCEP_TIMEOUT)
> + dev_err(&hsi->device, "timeout exception occurred\n");
> + if (error_interrupts & STE_HSI_EXCEP_OVERRUN)
> + dev_err(&hsi->device, "overrun exception occurred\n");
> + if (error_interrupts & STE_HSI_EXCEP_PARITY)
> + dev_err(&hsi->device, "parity exception occurred\n");
> +
> + if (error_interrupts & ~STE_HSI_EXCEP_BREAK)
> + ste_hsi_error(port);
Shouldn't the spin_lock lock be claimed before calling ste_hsi_error()
and then released after it. Or then put the proper locking for accessing
the queues in ste_hsi_error() ?
> +
> + /* Acknowledge exception interrupts */
> + writel(error_status, ste_hsi->rx_base + STE_HSI_RX_ACK);
> +
> + enable_irq(ste_port->excep_irq);
> +}
> +
> +static irqreturn_t ste_hsi_exception_isr(int irq, void *data)
> +{
> + struct hsi_port *port = data;
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> +
> + disable_irq_nosync(irq);
> + tasklet_hi_schedule(&ste_port->exception_tasklet);
> +
> + return IRQ_HANDLED;
> +}
> +
> +static void ste_hsi_overrun_tasklet(unsigned long data)
> +{
> + struct hsi_controller *hsi = (struct hsi_controller *)data;
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + struct hsi_port *hsi_port = &hsi->port[0];
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(hsi_port);
> + struct hsi_msg *msg;
> +
> + unsigned int channel;
> + u8 rised_overrun;
> + u8 mask;
> + u8 blocked = 0;
> +
> + rised_overrun = (u8) readl(ste_hsi->rx_base + STE_HSI_RX_OVERRUNMIS);
I guess we could use readb() here and forget about the casting. Or there
is something arch dependent that prevents this ?
> + mask = rised_overrun;
> + for (channel = 0; mask; ++channel, mask >>= 1) {
> + if (!(mask & 1))
> + continue;
> +
> + do {
> + /*
> + * No more messages, block interrupt
> + */
> + if (list_empty(&ste_port->rxqueue[channel])) {
> + blocked |= 1 << channel;
> + break;
> + }
> + /*
> + * Complete with error
> + */
> + msg = list_first_entry(&ste_port->rxqueue[channel],
> + struct hsi_msg, link);
> + list_del(&msg->link);
Access to the lists should be protected with lock.
> + msg->status = HSI_STATUS_ERROR;
> + msg->complete(msg);
> +
> + /*
> + * Now restart queued reads if any
> + * If start_transfer fails, try with next message
> + */
> + if (ste_hsi_start_transfer(ste_port,
> + &ste_port->rxqueue[channel]))
> + continue;
> + } while (0);
do {} while(0) Why ?
> + }
> +
> + /* Overrun acknowledge */
> + writel(rised_overrun, ste_hsi->rx_base + STE_HSI_RX_OVERRUNACK);
> + writel(~blocked & readl(ste_hsi->rx_base + STE_HSI_RX_OVERRUNIM),
> + ste_hsi->rx_base + STE_HSI_RX_OVERRUNIM);
> +
> + /*
> + * Enable all that should not be blocked
> + */
> + mask = rised_overrun & ~blocked;
> + for (channel = 0; mask; ++channel, mask >>= 1)
> + enable_irq(ste_hsi->overrun_irq[channel]);
> +}
> +
> +static irqreturn_t ste_hsi_overrun_isr(int irq, void *data)
> +{
> + struct hsi_port *port = data;
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> +
> + disable_irq_nosync(irq);
> + tasklet_hi_schedule(&ste_port->overrun_tasklet);
> +
> + return IRQ_HANDLED;
> +}
> +
> +static void __init ste_hsi_queues_init(struct ste_hsi_port *ste_port)
> +{
> + unsigned int ch;
> +
> + for (ch = 0; ch < STE_HSI_MAX_CHANNELS; ch++) {
> + INIT_LIST_HEAD(&ste_port->txqueue[ch]);
> + INIT_LIST_HEAD(&ste_port->rxqueue[ch]);
> + }
> + INIT_LIST_HEAD(&ste_port->brkqueue);
> +}
> +
> +static int __init ste_hsi_get_iomem(struct platform_device *pdev,
> + const char *res_name,
> + unsigned char __iomem **base,
> + dma_addr_t *phy)
> +{
> + struct resource *mem;
> + struct resource *ioarea;
> +
> + mem = platform_get_resource_byname(pdev, IORESOURCE_MEM, res_name);
> + if (!mem) {
> + dev_err(&pdev->dev, "IO memory region missing!\n");
> + return -ENXIO;
> + }
> +
> + ioarea = devm_request_mem_region(&pdev->dev, mem->start,
> + resource_size(mem),
> + dev_name(&pdev->dev));
> + if (!ioarea) {
> + dev_err(&pdev->dev, "Can't request IO memory region!\n");
> + return -ENXIO;
> + }
> +
> + *base = devm_ioremap(&pdev->dev, mem->start, resource_size(mem));
> + if (!base) {
> + dev_err(&pdev->dev, "%s IO remap failed!\n", mem->name);
> + return -ENXIO;
> + }
> + if (phy)
> + *phy = (dma_addr_t) mem->start;
> +
> + return 0;
> +}
> +
> +static int __init ste_hsi_get_irq(struct platform_device *pdev,
> + const char *res_name,
> + irqreturn_t(*isr) (int, void *), void *data,
> + int *irq_number)
> +{
> + struct resource *irq;
> + int err;
> +
> + irq = platform_get_resource_byname(pdev, IORESOURCE_IRQ, res_name);
> + if (!irq) {
> + dev_err(&pdev->dev, "IO memory region missing!\n");
> + return -ENXIO;
> + }
> +
> + err = devm_request_irq(&pdev->dev, irq->start, isr,
> + IRQF_DISABLED, irq->name, data);
> + if (err)
> + dev_err(&pdev->dev, "%s IRQ request failed!\n", irq->name);
> +
> + if (irq_number)
> + *irq_number = irq->start;
> +
> + return err;
> +}
> +
> +static void ste_hsi_flush_queue(struct list_head *queue, struct hsi_client *cl)
> +{
> + struct list_head *node, *tmp;
> + struct hsi_msg *msg;
> +
> + list_for_each_safe(node, tmp, queue) {
> + msg = list_entry(node, struct hsi_msg, link);
> + if ((cl) && (cl != msg->cl))
> + continue;
> + list_del(node);
> +
> + if (msg->destructor)
> + msg->destructor(msg);
> + else
> + hsi_free_msg(msg);
> + }
> +}
> +
> +static int ste_hsi_async_break(struct hsi_msg *msg)
> +{
> + struct hsi_port *port = hsi_get_port(msg->cl);
> + struct ste_hsi_port *ste_port = hsi_to_ste_port(port);
> + struct hsi_controller *hsi = to_hsi_controller(port->device.parent);
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + int err;
> +
> + err = ste_hsi_clock_enable(hsi);
> + if (unlikely(err))
> + return err;
> +
> + if (msg->ttype == HSI_MSG_WRITE) {
> + if (port->tx_cfg.mode != HSI_MODE_FRAME) {
> + err = -EINVAL;
> + goto out;
> + }
> + writel(1, ste_hsi->tx_base + STE_HSI_TX_BREAK);
> + msg->status = HSI_STATUS_COMPLETED;
> + msg->complete(msg);
> + } else {
> + u32 mask;
> + if (port->rx_cfg.mode != HSI_MODE_FRAME) {
> + err = -EINVAL;
> + goto out;
> + }
> + spin_lock_bh(&ste_hsi->lock);
> + msg->status = HSI_STATUS_PROCEEDING;
> + mask = readl(ste_hsi->rx_base + STE_HSI_RX_EXCEPIM);
> + /* Enable break exception on controller */
> + if (!(mask & STE_HSI_EXCEP_BREAK))
> + writel(mask | STE_HSI_EXCEP_BREAK,
> + ste_hsi->rx_base + STE_HSI_RX_EXCEPIM);
> +
> + list_add_tail(&msg->link, &ste_port->brkqueue);
> + spin_unlock_bh(&ste_hsi->lock);
> + }
> +
> +out:
> + ste_hsi_clock_disable(hsi);
> + return err;
> +}
> +
> +static int ste_hsi_async(struct hsi_msg *msg)
> +{
> + struct ste_hsi_controller *ste_hsi;
> + struct ste_hsi_port *ste_port;
> + struct list_head *queue;
> + int err = 0;
> +
> + if (unlikely(!msg))
> + return -ENOSYS;
> +
> + if (msg->sgt.nents > 1)
> + return -ENOSYS;
> +
> + if (unlikely(msg->break_frame))
> + return ste_hsi_async_break(msg);
> +
> + ste_port = client_to_ste_port(msg->cl);
> + ste_hsi = client_to_ste_controller(msg->cl);
> +
> + if (msg->ttype == HSI_MSG_WRITE) {
> + /* TX transfer */
> + BUG_ON(msg->channel >= ste_port->channels);
> + queue = &ste_port->txqueue[msg->channel];
> + } else {
> + /* RX transfer */
> + queue = &ste_port->rxqueue[msg->channel];
> + }
> +
> + spin_lock_bh(&ste_hsi->lock);
> + list_add_tail(&msg->link, queue);
> + msg->status = HSI_STATUS_QUEUED;
> +
> + err = ste_hsi_start_transfer(ste_port, queue);
> + if (err)
> + list_del(&msg->link);
> +
> + spin_unlock_bh(&ste_hsi->lock);
> +
> + return err;
> +}
> +
> +static int ste_hsi_setup(struct hsi_client *cl)
> +{
> + struct hsi_port *port = to_hsi_port(cl->device.parent);
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> + struct hsi_controller *hsi = to_hsi_controller(port->device.parent);
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + int err, i, buffers;
> + u32 div = 0;
> +
> + err = ste_hsi_clock_enable(hsi);
> + if (unlikely(err))
> + return err;
> +
> + if (cl->tx_cfg.speed) {
> + div = clk_get_rate(ste_hsi->tx_clk) / 1000 / cl->tx_cfg.speed;
> + if (div)
> + --div;
> + }
> +
> + port->tx_cfg = cl->tx_cfg;
> + port->rx_cfg = cl->rx_cfg;
> + /* Configure TX */
> + writel(cl->tx_cfg.mode, ste_hsi->tx_base + STE_HSI_TX_MODE);
> + writel(div, ste_hsi->tx_base + STE_HSI_TX_DIVISOR);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_PARITY);
> + /* TODO: Wait for idle here */
> + writel(cl->tx_cfg.channels, ste_hsi->tx_base + STE_HSI_TX_CHANNELS);
> + /* Calculate buffers number per channel */
> + buffers = STE_HSI_MAX_BUFFERS / cl->tx_cfg.channels;
> + for (i = 0; i < cl->tx_cfg.channels; i++) {
> + /* Set 32 bit long frames */
> + writel(31, ste_hsi->tx_base + STE_HSI_TX_FRAMELENX + 4 * i);
> + writel(buffers * i,
> + ste_hsi->tx_base + STE_HSI_TX_BASEX + 4 * i);
> + writel(buffers - 1,
> + ste_hsi->tx_base + STE_HSI_TX_SPANX + 4 * i);
> + writel(buffers - 1,
> + ste_hsi->tx_base + STE_HSI_TX_WATERMARKX + 4 * i);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_WATERMARKX + 4 * i);
> + }
> +
> + /* Configure RX */
> + writel(cl->rx_cfg.mode, ste_hsi->rx_base + STE_HSI_RX_MODE);
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_PARITY);
> + writel(cl->rx_cfg.channels, ste_hsi->rx_base + STE_HSI_RX_CHANNELS);
> + /* Calculate buffers number per channel */
> + buffers = STE_HSI_MAX_BUFFERS / cl->rx_cfg.channels;
> + for (i = 0; i < cl->rx_cfg.channels; i++) {
> + /* Set 32 bit long frames */
> + writel(31, ste_hsi->rx_base + STE_HSI_RX_FRAMELENX + 4 * i);
> + writel(buffers * i,
> + ste_hsi->rx_base + STE_HSI_RX_BASEX + 4 * i);
> + writel(buffers - 1,
> + ste_hsi->rx_base + STE_HSI_RX_SPANX + 4 * i);
> + writel(buffers - 1,
> + ste_hsi->rx_base + STE_HSI_RX_WATERMARKX + 4 * i);
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_WATERMARKX + 4 * i);
> + }
> +
> + ste_port->channels = max(cl->tx_cfg.channels, cl->rx_cfg.channels);
> +
> + ste_hsi_setup_dma(cl);
> +
> + ste_hsi_clock_disable(hsi);
> + return err;
> +}
> +
> +static int ste_hsi_flush(struct hsi_client *cl)
> +{
> + struct hsi_port *port = to_hsi_port(cl->device.parent);
> + struct ste_hsi_port *ste_port = hsi_port_drvdata(port);
> + struct hsi_controller *hsi = to_hsi_controller(port->device.parent);
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + int i;
> +
> + ste_hsi_clock_enable(hsi);
> +
> + /* Enter sleep mode */
> + writel(STE_HSI_MODE_SLEEP, ste_hsi->rx_base + STE_HSI_RX_MODE);
> +
> + /* Disable DMA, and terminate all outstanding jobs */
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_DMAEN);
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_DMAEN);
> + ste_hsi_terminate_dma(ste_port);
> +
> + /* Flush all HSIR and HSIT buffers */
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_STATE);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_BUFSTATE);
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_STATE);
> + /*
> + * BUFSTATE is cleared twice on purpose:
> + * first time all fifos are cleared
> + * second time to clear data that was in pipline buffer
> + * and was transfered to fifos
> + */
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_BUFSTATE);
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_BUFSTATE);
> +
> + /* Flush all errors */
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_EXCEP);
> +
> + /* Clear interrupts */
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_WATERMARKIM);
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_WATERMARKIC);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_WATERMARKIM);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_WATERMARKIC);
> + writel(0xFF, ste_hsi->rx_base + STE_HSI_RX_OVERRUNACK);
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_OVERRUNIM);
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_EXCEPIM);
> + writel(0x0F, ste_hsi->rx_base + STE_HSI_RX_ACK);
> +
> + /* Dequeue all pending requests */
> + for (i = 0; i < ste_port->channels; i++) {
> + /* Release write clocks */
> + if (!list_empty(&ste_port->txqueue[i]))
> + ste_hsi_clock_disable(hsi);
> + if (!list_empty(&ste_port->rxqueue[i]))
> + ste_hsi_clock_disable(hsi);
> + ste_hsi_flush_queue(&ste_port->txqueue[i], NULL);
> + ste_hsi_flush_queue(&ste_port->rxqueue[i], NULL);
> + }
> + ste_hsi_flush_queue(&ste_port->brkqueue, NULL);
> +
> + ste_hsi_clock_disable(hsi);
> +
> + return 0;
> +}
> +
> +static int ste_hsi_start_tx(struct hsi_client *cl)
> +{
> + return 0;
> +}
> +
> +static int ste_hsi_stop_tx(struct hsi_client *cl)
> +{
> + return 0;
> +}
> +
The above two functions are not necessary. The framework already set
dummy callback functions for the ones that you don't need.
> +static int ste_hsi_release(struct hsi_client *cl)
> +{
> + int err;
> + struct ste_hsi_controller *ste_hsi = client_to_ste_controller(cl);
> +
> + err = ste_hsi_flush(cl);
> + cancel_delayed_work(&ste_hsi->clk_work);
Why do you need to cancel delayed work here ?
Isn't so that if you happen indeed to have clk_work scheduled, you will
break PM ? You will leave the HSI clocks enabled although no one is
using the hsi port anymore.
> +
> + return 0;
> +}
> +
> +static int ste_hsi_ports_init(struct hsi_controller *hsi,
> + struct platform_device *pdev)
> +{
> + struct hsi_port *port;
> + struct ste_hsi_port *ste_port;
> + unsigned int i;
> + char irq_name[20];
> + int err;
> +
> + for (i = 0; i < hsi->num_ports; i++) {
> + ste_port = devm_kzalloc(&pdev->dev, sizeof *ste_port,
> + GFP_KERNEL);
> + if (!ste_port)
> + return -ENOMEM;
> +
> + port = &hsi->port[i];
> + port->async = ste_hsi_async;
> + port->setup = ste_hsi_setup;
> + port->flush = ste_hsi_flush;
> + port->start_tx = ste_hsi_start_tx;
> + port->stop_tx = ste_hsi_stop_tx;
The above two lines are not needed. See my previous comment on the
matter.
> + port->release = ste_hsi_release;
> + hsi_port_set_drvdata(port, ste_port);
> + ste_port->dev = &port->device;
> +
> + sprintf(irq_name, "hsi_rx_irq%d", i);
> + err = ste_hsi_get_irq(pdev, irq_name, ste_hsi_rx_isr, port,
> + &ste_port->rx_irq);
> + if (err)
> + return err;
> +
> + sprintf(irq_name, "hsi_tx_irq%d", i);
> + err = ste_hsi_get_irq(pdev, irq_name, ste_hsi_tx_isr, port,
> + &ste_port->tx_irq);
> + if (err)
> + return err;
> +
> + tasklet_init(&ste_port->rx_tasklet, ste_hsi_rx_tasklet,
> + (unsigned long)port);
> +
> + tasklet_init(&ste_port->tx_tasklet, ste_hsi_tx_tasklet,
> + (unsigned long)port);
> +
> + tasklet_init(&ste_port->exception_tasklet,
> + ste_hsi_exception_tasklet, (unsigned long)port);
> +
> + tasklet_init(&ste_port->overrun_tasklet,
> + ste_hsi_overrun_tasklet, (unsigned long)port);
> +
> + sprintf(irq_name, "hsi_rx_excep%d", i);
> + err = ste_hsi_get_irq(pdev, irq_name, ste_hsi_exception_isr,
> + port, &ste_port->excep_irq);
> + if (err)
> + return err;
> +
> + ste_hsi_queues_init(ste_port);
> + }
> + return 0;
> +}
> +
> +static int __init ste_hsi_hw_init(struct hsi_controller *hsi)
> +{
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> + int err;
> +
> + err = ste_hsi_clock_enable(hsi);
> + if (unlikely(err))
> + return err;
> +
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_BUFSTATE);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_FLUSHBITS);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_PRIORITY);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_BURSTLEN);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_PREAMBLE);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_DATASWAP);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_DMAEN);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_WATERMARKID);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_WATERMARKIC);
> + writel(0, ste_hsi->tx_base + STE_HSI_TX_WATERMARKIM);
> +
> + /* 0x23 is reset value per DB8500 Design Spec */
> + writel(0x23, ste_hsi->rx_base + STE_HSI_RX_THRESHOLD);
> +
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_BUFSTATE);
> +
> + /* Bits 0,1,2 set to 1 to clear exception flags */
> + writel(0x07, ste_hsi->rx_base + STE_HSI_RX_ACK);
> +
> + /* Bits 0..7 set to 1 to clear OVERRUN IRQ */
> + writel(0xFF, ste_hsi->rx_base + STE_HSI_RX_OVERRUNACK);
> +
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_DMAEN);
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_WATERMARKIC);
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_WATERMARKIM);
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_OVERRUNIM);
> +
> + /* Flush all errors */
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_EXCEP);
> +
> + /* 2 is Flush state, no RX exception generated afterwards */
> + writel(2, ste_hsi->rx_base + STE_HSI_RX_STATE);
> +
> + writel(0, ste_hsi->rx_base + STE_HSI_RX_EXCEPIM);
> +
> + ste_hsi_clock_disable(hsi);
> +
> + return err;
> +}
> +
> +static int __init ste_hsi_add_controller(struct hsi_controller *hsi,
> + struct platform_device *pdev)
> +{
> + struct ste_hsi_controller *ste_hsi;
> + char overrun_name[] = "hsi_rx_overrun_chxxx";
> + unsigned char i;
> + int err;
> +
> + ste_hsi = kzalloc(sizeof(struct ste_hsi_controller), GFP_KERNEL);
it is ok, although I think "sizeof(*ste_hsi)" is better.
> + if (!ste_hsi) {
> + dev_err(&pdev->dev, "Not enough memory for ste_hsi!\n");
> + return -ENOMEM;
> + }
> +
> + spin_lock_init(&ste_hsi->lock);
> + spin_lock_init(&ste_hsi->ck_lock);
> + INIT_DELAYED_WORK(&ste_hsi->clk_work, ste_hsi_delayed_disable_clock);
> +
> + hsi->id = pdev->id;
> + hsi->device.parent = &pdev->dev;
> + dev_set_name(&hsi->device, "ste-hsi.%d", hsi->id);
> + ste_hsi->dev = &hsi->device;
> + hsi_controller_set_drvdata(hsi, ste_hsi);
> +
> + /* Get and reserve resources for receiver */
> + err = ste_hsi_get_iomem(pdev, "hsi_rx_base", &ste_hsi->rx_base,
> + &ste_hsi->rx_dma_base);
> + if (err)
> + goto err_free_mem;
> + dev_info(&pdev->dev, "hsi_rx_base = %p\n", ste_hsi->rx_base);
> +
> + /* Get and reserve resources for transmitter */
> + err = ste_hsi_get_iomem(pdev, "hsi_tx_base", &ste_hsi->tx_base,
> + &ste_hsi->tx_dma_base);
> + if (err)
> + goto err_free_mem;
> + dev_info(&pdev->dev, "hsi_tx_base = %p\n", ste_hsi->tx_base);
> +
> + /* Get HSIT HSITXCLK clock */
> + ste_hsi->tx_clk = clk_get(&pdev->dev, "hsit_hsitxclk");
> + if (IS_ERR(ste_hsi->tx_clk)) {
> + dev_err(&hsi->device, "Couldn't get HSIT HSITXCLK clock\n");
> + err = PTR_ERR(ste_hsi->tx_clk);
> + goto err_free_mem;
> + }
> +
> + /* Get HSIR HSIRXCLK clock */
> + ste_hsi->rx_clk = clk_get(&pdev->dev, "hsir_hsirxclk");
> + if (IS_ERR(ste_hsi->rx_clk)) {
> + dev_err(&hsi->device, "Couldn't get HSIR HSIRXCLK clock\n");
> + err = PTR_ERR(ste_hsi->rx_clk);
> + goto err_clk_free;
> + }
> +
> + /* Get HSIT HCLK clock */
> + ste_hsi->ssitx_clk = clk_get(&pdev->dev, "hsit_hclk");
> + if (IS_ERR(ste_hsi->ssitx_clk)) {
> + dev_err(&hsi->device, "Couldn't get HSIT HCLK clock\n");
> + err = PTR_ERR(ste_hsi->ssitx_clk);
> + goto err_clk_free;
> + }
> +
> + /* Get HSIR HCLK clock */
> + ste_hsi->ssirx_clk = clk_get(&pdev->dev, "hsir_hclk");
> + if (IS_ERR(ste_hsi->ssirx_clk)) {
> + dev_err(&hsi->device, "Couldn't get HSIR HCLK clock\n");
> + err = PTR_ERR(ste_hsi->ssirx_clk);
> + goto err_clk_free;
> + }
> +
> + err = ste_hsi_clock_enable(hsi);
> + if (unlikely(err))
> + goto err_clk_free;
> +
> + /* Check if controller is at specified address */
> + if (compare_periphid(ste_hsir_periphid,
> + (u32 *) (ste_hsi->rx_base + 0xFE0), 8)) {
> + dev_err(&pdev->dev, "No hsir controller at = %p\n",
> + ste_hsi->rx_base);
> + err = -ENXIO;
> + goto err_clk_free;
> + }
> +
> + /* Check if controller is at specified address */
> + if (compare_periphid(ste_hsit_periphid,
> + (u32 *) (ste_hsi->tx_base + 0xFE0), 8)) {
> + dev_err(&pdev->dev, "No hsit controller at = %p\n",
> + ste_hsi->tx_base);
> + err = -ENXIO;
> + goto err_clk_free;
> + }
> + ste_hsi_clock_disable(hsi);
> +
> + err = ste_hsi_hw_init(hsi);
> + if (err) {
> + dev_err(&pdev->dev, "Failed to init HSI controller!\n");
> + goto err_clk_free;
> + }
> +
> + for (i = 0; i < STE_HSI_MAX_CHANNELS; i++) {
> + sprintf(overrun_name, "hsi_rx_overrun_ch%d", i);
> + err = ste_hsi_get_irq(pdev, overrun_name, ste_hsi_overrun_isr,
> + hsi, &ste_hsi->overrun_irq[i]);
> + if (err)
> + goto err_clk_free;
> + }
> +
> + err = ste_hsi_ports_init(hsi, pdev);
> + if (err)
> + goto err_clk_free;
> +
> + err = hsi_register_controller(hsi);
> + if (err)
> + goto err_clk_free;
> +
> + return 0;
> +
> +err_clk_free:
> + ste_hsi_clks_free(ste_hsi);
> +err_free_mem:
> + kfree(ste_hsi);
> + return err;
> +}
> +
> +static int ste_hsi_remove_controller(struct hsi_controller *hsi,
> + struct platform_device *pdev)
> +{
> + struct ste_hsi_controller *ste_hsi = hsi_controller_drvdata(hsi);
> +
> + ste_hsi_clks_free(ste_hsi);
> + hsi_unregister_controller(hsi);
> +
> + return 0;
> +}
> +
> +static int __init ste_hsi_probe(struct platform_device *pdev)
> +{
> + struct hsi_controller *hsi;
> + struct ste_hsi_platform_data *pdata = pdev->dev.platform_data;
> + int err;
> +
> + if (!pdata) {
> + dev_err(&pdev->dev, "No HSI platform data!\n");
> + return -EINVAL;
> + }
> +
> + hsi = hsi_alloc_controller(pdata->num_ports, GFP_KERNEL);
> + if (!hsi) {
> + dev_err(&pdev->dev, "No memory to allocate HSI controller!\n");
> + return -ENOMEM;
> + }
> + platform_set_drvdata(pdev, hsi);
> +
> + err = ste_hsi_add_controller(hsi, pdev);
> + if (err < 0) {
> + dev_err(&pdev->dev, "Can't add HSI controller!\n");
> + goto err_free_controller;
> + }
> +
> + if (pdata->use_dma)
> + ste_hsi_init_dma(pdata, hsi);
> +
> + return 0;
> +
> +err_free_controller:
> + platform_set_drvdata(pdev, NULL);
> + hsi_free_controller(hsi);
> +
> + return err;
> +}
> +
> +static int __exit ste_hsi_remove(struct platform_device *pdev)
> +{
> + struct hsi_controller *hsi = platform_get_drvdata(pdev);
> +
> + ste_hsi_remove_controller(hsi, pdev);
> + hsi_free_controller(hsi);
> +
> + return 0;
> +}
> +
> +static struct platform_driver ste_hsi_driver __refdata = {
> + .driver = {
> + .name = "ste_hsi",
> + .owner = THIS_MODULE,
> + },
> + .remove = __exit_p(ste_hsi_remove),
> +};
> +
> +static int __init ste_hsi_init(void)
> +{
> + return platform_driver_probe(&ste_hsi_driver, ste_hsi_probe);
> +}
> +
> +module_init(ste_hsi_init)
> +
> +static void __exit ste_hsi_exit(void)
> +{
> + platform_driver_unregister(&ste_hsi_driver);
> +}
> +
> +module_exit(ste_hsi_exit)
> +
> + MODULE_AUTHOR("Lukasz Baj <lukasz.baj@xxxxxxxxx");
> + MODULE_DESCRIPTION("STE HSI driver.");
> + MODULE_LICENSE("GPL");
It is not clear to me: Is the license GPLv2 only ? If it is so then you
should say so also: MODULE_LICENSE("GPLv2") otherwise I will suggest to
update the header to GPLv2 or later.
Br,
--
Carlos Chinea <carlos.chinea@xxxxxxxxx>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/