2017-11-06 18:11:51 +01:00
|
|
|
// SPDX-License-Identifier: GPL-2.0+
|
2011-11-17 16:17:04 +01:00
|
|
|
/*
|
|
|
|
* Driver for CSR SiRFprimaII onboard UARTs.
|
|
|
|
*
|
|
|
|
* Copyright (c) 2011 Cambridge Silicon Radio Limited, a CSR plc group company.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/ioport.h>
|
|
|
|
#include <linux/platform_device.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/sysrq.h>
|
|
|
|
#include <linux/console.h>
|
|
|
|
#include <linux/tty.h>
|
|
|
|
#include <linux/tty_flip.h>
|
|
|
|
#include <linux/serial_core.h>
|
|
|
|
#include <linux/serial.h>
|
|
|
|
#include <linux/clk.h>
|
|
|
|
#include <linux/of.h>
|
|
|
|
#include <linux/slab.h>
|
|
|
|
#include <linux/io.h>
|
2013-08-15 00:52:15 +02:00
|
|
|
#include <linux/of_gpio.h>
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
#include <linux/dmaengine.h>
|
|
|
|
#include <linux/dma-direction.h>
|
|
|
|
#include <linux/dma-mapping.h>
|
2011-11-17 16:17:04 +01:00
|
|
|
#include <asm/irq.h>
|
|
|
|
#include <asm/mach/irq.h>
|
|
|
|
|
|
|
|
#include "sirfsoc_uart.h"
|
|
|
|
|
|
|
|
static unsigned int
|
|
|
|
sirfsoc_uart_pio_tx_chars(struct sirfsoc_uart_port *sirfport, int count);
|
|
|
|
static unsigned int
|
|
|
|
sirfsoc_uart_pio_rx_chars(struct uart_port *port, unsigned int max_rx_count);
|
|
|
|
static struct uart_driver sirfsoc_uart_drv;
|
|
|
|
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
static void sirfsoc_uart_tx_dma_complete_callback(void *param);
|
2011-11-17 16:17:04 +01:00
|
|
|
static const struct sirfsoc_baudrate_to_regv baudrate_to_regv[] = {
|
|
|
|
{4000000, 2359296},
|
|
|
|
{3500000, 1310721},
|
|
|
|
{3000000, 1572865},
|
|
|
|
{2500000, 1245186},
|
|
|
|
{2000000, 1572866},
|
|
|
|
{1500000, 1245188},
|
|
|
|
{1152000, 1638404},
|
|
|
|
{1000000, 1572869},
|
|
|
|
{921600, 1114120},
|
|
|
|
{576000, 1245196},
|
|
|
|
{500000, 1245198},
|
|
|
|
{460800, 1572876},
|
|
|
|
{230400, 1310750},
|
|
|
|
{115200, 1310781},
|
|
|
|
{57600, 1310843},
|
|
|
|
{38400, 1114328},
|
|
|
|
{19200, 1114545},
|
|
|
|
{9600, 1114979},
|
|
|
|
};
|
|
|
|
|
2015-04-29 08:45:08 +02:00
|
|
|
static struct sirfsoc_uart_port *sirf_ports[SIRFSOC_UART_NR];
|
2011-11-17 16:17:04 +01:00
|
|
|
|
|
|
|
static inline struct sirfsoc_uart_port *to_sirfport(struct uart_port *port)
|
|
|
|
{
|
|
|
|
return container_of(port, struct sirfsoc_uart_port, port);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline unsigned int sirfsoc_uart_tx_empty(struct uart_port *port)
|
|
|
|
{
|
|
|
|
unsigned long reg;
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_fifo_status *ufifo_st = &sirfport->uart_reg->fifo_status;
|
|
|
|
reg = rd_regl(port, ureg->sirfsoc_tx_fifo_status);
|
2015-04-29 08:45:09 +02:00
|
|
|
return (reg & ufifo_st->ff_empty(port)) ? TIOCSER_TEMT : 0;
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static unsigned int sirfsoc_uart_get_mctrl(struct uart_port *port)
|
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
2013-08-15 00:52:15 +02:00
|
|
|
if (!sirfport->hw_flow_ctrl || !sirfport->ms_enabled)
|
2011-11-17 16:17:04 +01:00
|
|
|
goto cts_asserted;
|
2013-08-15 00:52:15 +02:00
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_REAL_UART) {
|
2013-08-12 12:15:35 +02:00
|
|
|
if (!(rd_regl(port, ureg->sirfsoc_afc_ctrl) &
|
|
|
|
SIRFUART_AFC_CTS_STATUS))
|
2011-11-17 16:17:04 +01:00
|
|
|
goto cts_asserted;
|
|
|
|
else
|
|
|
|
goto cts_deasserted;
|
2013-08-15 00:52:15 +02:00
|
|
|
} else {
|
|
|
|
if (!gpio_get_value(sirfport->cts_gpio))
|
|
|
|
goto cts_asserted;
|
|
|
|
else
|
|
|
|
goto cts_deasserted;
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
cts_deasserted:
|
|
|
|
return TIOCM_CAR | TIOCM_DSR;
|
|
|
|
cts_asserted:
|
|
|
|
return TIOCM_CAR | TIOCM_DSR | TIOCM_CTS;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_set_mctrl(struct uart_port *port, unsigned int mctrl)
|
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
2011-11-17 16:17:04 +01:00
|
|
|
unsigned int assert = mctrl & TIOCM_RTS;
|
|
|
|
unsigned int val = assert ? SIRFUART_AFC_CTRL_RX_THD : 0x0;
|
|
|
|
unsigned int current_val;
|
2013-08-15 00:52:15 +02:00
|
|
|
|
2015-05-14 08:45:25 +02:00
|
|
|
if (mctrl & TIOCM_LOOP) {
|
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_REAL_UART)
|
|
|
|
wr_regl(port, ureg->sirfsoc_line_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_line_ctrl) |
|
|
|
|
SIRFUART_LOOP_BACK);
|
|
|
|
else
|
|
|
|
wr_regl(port, ureg->sirfsoc_mode1,
|
|
|
|
rd_regl(port, ureg->sirfsoc_mode1) |
|
|
|
|
SIRFSOC_USP_LOOP_BACK_CTRL);
|
|
|
|
} else {
|
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_REAL_UART)
|
|
|
|
wr_regl(port, ureg->sirfsoc_line_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_line_ctrl) &
|
|
|
|
~SIRFUART_LOOP_BACK);
|
|
|
|
else
|
|
|
|
wr_regl(port, ureg->sirfsoc_mode1,
|
|
|
|
rd_regl(port, ureg->sirfsoc_mode1) &
|
|
|
|
~SIRFSOC_USP_LOOP_BACK_CTRL);
|
|
|
|
}
|
|
|
|
|
2013-08-15 00:52:15 +02:00
|
|
|
if (!sirfport->hw_flow_ctrl || !sirfport->ms_enabled)
|
|
|
|
return;
|
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_REAL_UART) {
|
2013-08-12 12:15:35 +02:00
|
|
|
current_val = rd_regl(port, ureg->sirfsoc_afc_ctrl) & ~0xFF;
|
2011-11-17 16:17:04 +01:00
|
|
|
val |= current_val;
|
2013-08-12 12:15:35 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_afc_ctrl, val);
|
2013-08-15 00:52:15 +02:00
|
|
|
} else {
|
|
|
|
if (!val)
|
|
|
|
gpio_set_value(sirfport->rts_gpio, 1);
|
|
|
|
else
|
|
|
|
gpio_set_value(sirfport->rts_gpio, 0);
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_stop_tx(struct uart_port *port)
|
|
|
|
{
|
2013-08-07 07:35:38 +02:00
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_int_en *uint_en = &sirfport->uart_reg->uart_int_en;
|
2013-08-07 07:35:38 +02:00
|
|
|
|
2014-01-30 06:57:29 +01:00
|
|
|
if (sirfport->tx_dma_chan) {
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
if (sirfport->tx_dma_state == TX_DMA_RUNNING) {
|
|
|
|
dmaengine_pause(sirfport->tx_dma_chan);
|
|
|
|
sirfport->tx_dma_state = TX_DMA_PAUSE;
|
|
|
|
} else {
|
2015-01-03 10:02:57 +01:00
|
|
|
if (!sirfport->is_atlas7)
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg) &
|
|
|
|
~uint_en->sirfsoc_txfifo_empty_en);
|
|
|
|
else
|
2015-05-14 08:45:21 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_clr_reg,
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
uint_en->sirfsoc_txfifo_empty_en);
|
|
|
|
}
|
|
|
|
} else {
|
2015-05-14 08:45:21 +02:00
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_USP_UART)
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_rx_en, rd_regl(port,
|
|
|
|
ureg->sirfsoc_tx_rx_en) & ~SIRFUART_TX_EN);
|
2015-01-03 10:02:57 +01:00
|
|
|
if (!sirfport->is_atlas7)
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg) &
|
|
|
|
~uint_en->sirfsoc_txfifo_empty_en);
|
|
|
|
else
|
2015-05-14 08:45:21 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_clr_reg,
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
uint_en->sirfsoc_txfifo_empty_en);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_tx_with_dma(struct sirfsoc_uart_port *sirfport)
|
|
|
|
{
|
|
|
|
struct uart_port *port = &sirfport->port;
|
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_int_en *uint_en = &sirfport->uart_reg->uart_int_en;
|
|
|
|
struct circ_buf *xmit = &port->state->xmit;
|
|
|
|
unsigned long tran_size;
|
|
|
|
unsigned long tran_start;
|
|
|
|
unsigned long pio_tx_size;
|
|
|
|
|
|
|
|
tran_size = CIRC_CNT_TO_END(xmit->head, xmit->tail, UART_XMIT_SIZE);
|
|
|
|
tran_start = (unsigned long)(xmit->buf + xmit->tail);
|
|
|
|
if (uart_circ_empty(xmit) || uart_tx_stopped(port) ||
|
|
|
|
!tran_size)
|
|
|
|
return;
|
|
|
|
if (sirfport->tx_dma_state == TX_DMA_PAUSE) {
|
|
|
|
dmaengine_resume(sirfport->tx_dma_chan);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
if (sirfport->tx_dma_state == TX_DMA_RUNNING)
|
|
|
|
return;
|
2015-01-03 10:02:57 +01:00
|
|
|
if (!sirfport->is_atlas7)
|
2013-08-12 12:15:35 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg)&
|
|
|
|
~(uint_en->sirfsoc_txfifo_empty_en));
|
|
|
|
else
|
2015-05-14 08:45:21 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_clr_reg,
|
2013-08-12 12:15:35 +02:00
|
|
|
uint_en->sirfsoc_txfifo_empty_en);
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
/*
|
|
|
|
* DMA requires buffer address and buffer length are both aligned with
|
|
|
|
* 4 bytes, so we use PIO for
|
|
|
|
* 1. if address is not aligned with 4bytes, use PIO for the first 1~3
|
|
|
|
* bytes, and move to DMA for the left part aligned with 4bytes
|
|
|
|
* 2. if buffer length is not aligned with 4bytes, use DMA for aligned
|
|
|
|
* part first, move to PIO for the left 1~3 bytes
|
|
|
|
*/
|
|
|
|
if (tran_size < 4 || BYTES_TO_ALIGN(tran_start)) {
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_op, SIRFUART_FIFO_STOP);
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_dma_io_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_tx_dma_io_ctrl)|
|
|
|
|
SIRFUART_IO_MODE);
|
|
|
|
if (BYTES_TO_ALIGN(tran_start)) {
|
|
|
|
pio_tx_size = sirfsoc_uart_pio_tx_chars(sirfport,
|
|
|
|
BYTES_TO_ALIGN(tran_start));
|
|
|
|
tran_size -= pio_tx_size;
|
|
|
|
}
|
|
|
|
if (tran_size < 4)
|
|
|
|
sirfsoc_uart_pio_tx_chars(sirfport, tran_size);
|
2015-01-03 10:02:57 +01:00
|
|
|
if (!sirfport->is_atlas7)
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg)|
|
|
|
|
uint_en->sirfsoc_txfifo_empty_en);
|
|
|
|
else
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
uint_en->sirfsoc_txfifo_empty_en);
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_op, SIRFUART_FIFO_START);
|
|
|
|
} else {
|
|
|
|
/* tx transfer mode switch into dma mode */
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_op, SIRFUART_FIFO_STOP);
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_dma_io_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_tx_dma_io_ctrl)&
|
|
|
|
~SIRFUART_IO_MODE);
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_op, SIRFUART_FIFO_START);
|
|
|
|
tran_size &= ~(0x3);
|
|
|
|
|
|
|
|
sirfport->tx_dma_addr = dma_map_single(port->dev,
|
|
|
|
xmit->buf + xmit->tail,
|
|
|
|
tran_size, DMA_TO_DEVICE);
|
|
|
|
sirfport->tx_dma_desc = dmaengine_prep_slave_single(
|
|
|
|
sirfport->tx_dma_chan, sirfport->tx_dma_addr,
|
|
|
|
tran_size, DMA_MEM_TO_DEV, DMA_PREP_INTERRUPT);
|
|
|
|
if (!sirfport->tx_dma_desc) {
|
|
|
|
dev_err(port->dev, "DMA prep slave single fail\n");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
sirfport->tx_dma_desc->callback =
|
|
|
|
sirfsoc_uart_tx_dma_complete_callback;
|
|
|
|
sirfport->tx_dma_desc->callback_param = (void *)sirfport;
|
|
|
|
sirfport->transfer_size = tran_size;
|
|
|
|
|
|
|
|
dmaengine_submit(sirfport->tx_dma_desc);
|
|
|
|
dma_async_issue_pending(sirfport->tx_dma_chan);
|
|
|
|
sirfport->tx_dma_state = TX_DMA_RUNNING;
|
|
|
|
}
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
2013-08-08 10:41:43 +02:00
|
|
|
static void sirfsoc_uart_start_tx(struct uart_port *port)
|
2011-11-17 16:17:04 +01:00
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_int_en *uint_en = &sirfport->uart_reg->uart_int_en;
|
2014-01-30 06:57:29 +01:00
|
|
|
if (sirfport->tx_dma_chan)
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
sirfsoc_uart_tx_with_dma(sirfport);
|
|
|
|
else {
|
2015-05-14 08:45:21 +02:00
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_USP_UART)
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_rx_en, rd_regl(port,
|
|
|
|
ureg->sirfsoc_tx_rx_en) | SIRFUART_TX_EN);
|
2015-05-26 11:35:58 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_op, SIRFUART_FIFO_STOP);
|
2015-04-29 08:45:09 +02:00
|
|
|
sirfsoc_uart_pio_tx_chars(sirfport, port->fifosize);
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_op, SIRFUART_FIFO_START);
|
2015-01-03 10:02:57 +01:00
|
|
|
if (!sirfport->is_atlas7)
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg)|
|
|
|
|
uint_en->sirfsoc_txfifo_empty_en);
|
|
|
|
else
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
uint_en->sirfsoc_txfifo_empty_en);
|
|
|
|
}
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_stop_rx(struct uart_port *port)
|
|
|
|
{
|
2013-08-07 07:35:38 +02:00
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_int_en *uint_en = &sirfport->uart_reg->uart_int_en;
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
|
2013-08-12 12:15:35 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_rx_fifo_op, 0);
|
2014-01-30 06:57:29 +01:00
|
|
|
if (sirfport->rx_dma_chan) {
|
2015-01-03 10:02:57 +01:00
|
|
|
if (!sirfport->is_atlas7)
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg) &
|
2015-05-14 08:45:21 +02:00
|
|
|
~(SIRFUART_RX_DMA_INT_EN(uint_en,
|
|
|
|
sirfport->uart_reg->uart_type) |
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
uint_en->sirfsoc_rx_done_en));
|
|
|
|
else
|
2015-05-14 08:45:21 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_clr_reg,
|
|
|
|
SIRFUART_RX_DMA_INT_EN(uint_en,
|
|
|
|
sirfport->uart_reg->uart_type)|
|
|
|
|
uint_en->sirfsoc_rx_done_en);
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
dmaengine_terminate_all(sirfport->rx_dma_chan);
|
|
|
|
} else {
|
2015-01-03 10:02:57 +01:00
|
|
|
if (!sirfport->is_atlas7)
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg)&
|
2015-05-14 08:45:21 +02:00
|
|
|
~(SIRFUART_RX_IO_INT_EN(uint_en,
|
|
|
|
sirfport->uart_reg->uart_type)));
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
else
|
2015-05-14 08:45:21 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_clr_reg,
|
|
|
|
SIRFUART_RX_IO_INT_EN(uint_en,
|
|
|
|
sirfport->uart_reg->uart_type));
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
}
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_disable_ms(struct uart_port *port)
|
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_int_en *uint_en = &sirfport->uart_reg->uart_int_en;
|
2013-08-07 07:35:38 +02:00
|
|
|
|
2011-11-17 16:17:04 +01:00
|
|
|
if (!sirfport->hw_flow_ctrl)
|
|
|
|
return;
|
2013-08-15 00:52:15 +02:00
|
|
|
sirfport->ms_enabled = false;
|
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_REAL_UART) {
|
|
|
|
wr_regl(port, ureg->sirfsoc_afc_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_afc_ctrl) & ~0x3FF);
|
2015-01-03 10:02:57 +01:00
|
|
|
if (!sirfport->is_atlas7)
|
2013-08-15 00:52:15 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg)&
|
|
|
|
~uint_en->sirfsoc_cts_en);
|
|
|
|
else
|
2015-05-14 08:45:21 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_clr_reg,
|
2013-08-15 00:52:15 +02:00
|
|
|
uint_en->sirfsoc_cts_en);
|
2013-08-12 12:15:35 +02:00
|
|
|
} else
|
2013-08-15 00:52:15 +02:00
|
|
|
disable_irq(gpio_to_irq(sirfport->cts_gpio));
|
|
|
|
}
|
|
|
|
|
|
|
|
static irqreturn_t sirfsoc_uart_usp_cts_handler(int irq, void *dev_id)
|
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport = (struct sirfsoc_uart_port *)dev_id;
|
|
|
|
struct uart_port *port = &sirfport->port;
|
2014-05-26 13:02:07 +02:00
|
|
|
spin_lock(&port->lock);
|
2013-08-15 00:52:15 +02:00
|
|
|
if (gpio_is_valid(sirfport->cts_gpio) && sirfport->ms_enabled)
|
|
|
|
uart_handle_cts_change(port,
|
|
|
|
!gpio_get_value(sirfport->cts_gpio));
|
2014-05-26 13:02:07 +02:00
|
|
|
spin_unlock(&port->lock);
|
2013-08-15 00:52:15 +02:00
|
|
|
return IRQ_HANDLED;
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_enable_ms(struct uart_port *port)
|
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_int_en *uint_en = &sirfport->uart_reg->uart_int_en;
|
2013-08-07 07:35:38 +02:00
|
|
|
|
2011-11-17 16:17:04 +01:00
|
|
|
if (!sirfport->hw_flow_ctrl)
|
|
|
|
return;
|
2013-08-15 00:52:15 +02:00
|
|
|
sirfport->ms_enabled = true;
|
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_REAL_UART) {
|
|
|
|
wr_regl(port, ureg->sirfsoc_afc_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_afc_ctrl) |
|
2015-05-14 08:45:22 +02:00
|
|
|
SIRFUART_AFC_TX_EN | SIRFUART_AFC_RX_EN |
|
|
|
|
SIRFUART_AFC_CTRL_RX_THD);
|
2015-01-03 10:02:57 +01:00
|
|
|
if (!sirfport->is_atlas7)
|
2013-08-15 00:52:15 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg)
|
|
|
|
| uint_en->sirfsoc_cts_en);
|
|
|
|
else
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
uint_en->sirfsoc_cts_en);
|
2013-08-12 12:15:35 +02:00
|
|
|
} else
|
2013-08-15 00:52:15 +02:00
|
|
|
enable_irq(gpio_to_irq(sirfport->cts_gpio));
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_break_ctl(struct uart_port *port, int break_state)
|
|
|
|
{
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_REAL_UART) {
|
|
|
|
unsigned long ulcon = rd_regl(port, ureg->sirfsoc_line_ctrl);
|
|
|
|
if (break_state)
|
|
|
|
ulcon |= SIRFUART_SET_BREAK;
|
|
|
|
else
|
|
|
|
ulcon &= ~SIRFUART_SET_BREAK;
|
|
|
|
wr_regl(port, ureg->sirfsoc_line_ctrl, ulcon);
|
|
|
|
}
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static unsigned int
|
|
|
|
sirfsoc_uart_pio_rx_chars(struct uart_port *port, unsigned int max_rx_count)
|
|
|
|
{
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_fifo_status *ufifo_st = &sirfport->uart_reg->fifo_status;
|
2011-11-17 16:17:04 +01:00
|
|
|
unsigned int ch, rx_count = 0;
|
2013-08-12 12:15:35 +02:00
|
|
|
struct tty_struct *tty;
|
|
|
|
tty = tty_port_tty_get(&port->state->port);
|
|
|
|
if (!tty)
|
|
|
|
return -ENODEV;
|
|
|
|
while (!(rd_regl(port, ureg->sirfsoc_rx_fifo_status) &
|
2015-04-29 08:45:09 +02:00
|
|
|
ufifo_st->ff_empty(port))) {
|
2013-08-12 12:15:35 +02:00
|
|
|
ch = rd_regl(port, ureg->sirfsoc_rx_fifo_data) |
|
|
|
|
SIRFUART_DUMMY_READ;
|
2011-11-17 16:17:04 +01:00
|
|
|
if (unlikely(uart_handle_sysrq_char(port, ch)))
|
|
|
|
continue;
|
|
|
|
uart_insert_char(port, 0, 0, ch, TTY_NORMAL);
|
|
|
|
rx_count++;
|
|
|
|
if (rx_count >= max_rx_count)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
port->icount.rx += rx_count;
|
2013-08-19 16:44:28 +02:00
|
|
|
|
2011-11-17 16:17:04 +01:00
|
|
|
return rx_count;
|
|
|
|
}
|
|
|
|
|
|
|
|
static unsigned int
|
|
|
|
sirfsoc_uart_pio_tx_chars(struct sirfsoc_uart_port *sirfport, int count)
|
|
|
|
{
|
|
|
|
struct uart_port *port = &sirfport->port;
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_fifo_status *ufifo_st = &sirfport->uart_reg->fifo_status;
|
2011-11-17 16:17:04 +01:00
|
|
|
struct circ_buf *xmit = &port->state->xmit;
|
|
|
|
unsigned int num_tx = 0;
|
|
|
|
while (!uart_circ_empty(xmit) &&
|
2013-08-12 12:15:35 +02:00
|
|
|
!(rd_regl(port, ureg->sirfsoc_tx_fifo_status) &
|
2015-04-29 08:45:09 +02:00
|
|
|
ufifo_st->ff_full(port)) &&
|
2011-11-17 16:17:04 +01:00
|
|
|
count--) {
|
2013-08-12 12:15:35 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_data,
|
|
|
|
xmit->buf[xmit->tail]);
|
2011-11-17 16:17:04 +01:00
|
|
|
xmit->tail = (xmit->tail + 1) & (UART_XMIT_SIZE - 1);
|
|
|
|
port->icount.tx++;
|
|
|
|
num_tx++;
|
|
|
|
}
|
|
|
|
if (uart_circ_chars_pending(xmit) < WAKEUP_CHARS)
|
|
|
|
uart_write_wakeup(port);
|
|
|
|
return num_tx;
|
|
|
|
}
|
|
|
|
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
static void sirfsoc_uart_tx_dma_complete_callback(void *param)
|
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport = (struct sirfsoc_uart_port *)param;
|
|
|
|
struct uart_port *port = &sirfport->port;
|
|
|
|
struct circ_buf *xmit = &port->state->xmit;
|
|
|
|
unsigned long flags;
|
|
|
|
|
2014-05-26 13:02:07 +02:00
|
|
|
spin_lock_irqsave(&port->lock, flags);
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
xmit->tail = (xmit->tail + sirfport->transfer_size) &
|
|
|
|
(UART_XMIT_SIZE - 1);
|
|
|
|
port->icount.tx += sirfport->transfer_size;
|
|
|
|
if (uart_circ_chars_pending(xmit) < WAKEUP_CHARS)
|
|
|
|
uart_write_wakeup(port);
|
|
|
|
if (sirfport->tx_dma_addr)
|
|
|
|
dma_unmap_single(port->dev, sirfport->tx_dma_addr,
|
|
|
|
sirfport->transfer_size, DMA_TO_DEVICE);
|
|
|
|
sirfport->tx_dma_state = TX_DMA_IDLE;
|
|
|
|
sirfsoc_uart_tx_with_dma(sirfport);
|
2014-05-26 13:02:07 +02:00
|
|
|
spin_unlock_irqrestore(&port->lock, flags);
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
}
|
|
|
|
|
2011-11-17 16:17:04 +01:00
|
|
|
static irqreturn_t sirfsoc_uart_isr(int irq, void *dev_id)
|
|
|
|
{
|
|
|
|
unsigned long intr_status;
|
|
|
|
unsigned long cts_status;
|
|
|
|
unsigned long flag = TTY_NORMAL;
|
|
|
|
struct sirfsoc_uart_port *sirfport = (struct sirfsoc_uart_port *)dev_id;
|
|
|
|
struct uart_port *port = &sirfport->port;
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_fifo_status *ufifo_st = &sirfport->uart_reg->fifo_status;
|
|
|
|
struct sirfsoc_int_status *uint_st = &sirfport->uart_reg->uart_int_st;
|
|
|
|
struct sirfsoc_int_en *uint_en = &sirfport->uart_reg->uart_int_en;
|
2011-11-17 16:17:04 +01:00
|
|
|
struct uart_state *state = port->state;
|
|
|
|
struct circ_buf *xmit = &port->state->xmit;
|
2012-12-25 10:32:04 +01:00
|
|
|
spin_lock(&port->lock);
|
2013-08-12 12:15:35 +02:00
|
|
|
intr_status = rd_regl(port, ureg->sirfsoc_int_st_reg);
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_st_reg, intr_status);
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
intr_status &= rd_regl(port, ureg->sirfsoc_int_en_reg);
|
2015-05-14 08:45:21 +02:00
|
|
|
if (unlikely(intr_status & (SIRFUART_ERR_INT_STAT(uint_st,
|
|
|
|
sirfport->uart_reg->uart_type)))) {
|
2013-08-12 12:15:35 +02:00
|
|
|
if (intr_status & uint_st->sirfsoc_rxd_brk) {
|
|
|
|
port->icount.brk++;
|
2011-11-17 16:17:04 +01:00
|
|
|
if (uart_handle_break(port))
|
|
|
|
goto recv_char;
|
|
|
|
}
|
2015-05-14 08:45:24 +02:00
|
|
|
if (intr_status & uint_st->sirfsoc_rx_oflow) {
|
2011-11-17 16:17:04 +01:00
|
|
|
port->icount.overrun++;
|
2015-05-14 08:45:24 +02:00
|
|
|
flag = TTY_OVERRUN;
|
|
|
|
}
|
2013-08-12 12:15:35 +02:00
|
|
|
if (intr_status & uint_st->sirfsoc_frm_err) {
|
2011-11-17 16:17:04 +01:00
|
|
|
port->icount.frame++;
|
|
|
|
flag = TTY_FRAME;
|
|
|
|
}
|
2015-05-14 08:45:24 +02:00
|
|
|
if (intr_status & uint_st->sirfsoc_parity_err) {
|
|
|
|
port->icount.parity++;
|
2011-11-17 16:17:04 +01:00
|
|
|
flag = TTY_PARITY;
|
2015-05-14 08:45:24 +02:00
|
|
|
}
|
2013-08-12 12:15:35 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_rx_fifo_op, SIRFUART_FIFO_RESET);
|
|
|
|
wr_regl(port, ureg->sirfsoc_rx_fifo_op, 0);
|
|
|
|
wr_regl(port, ureg->sirfsoc_rx_fifo_op, SIRFUART_FIFO_START);
|
2011-11-17 16:17:04 +01:00
|
|
|
intr_status &= port->read_status_mask;
|
|
|
|
uart_insert_char(port, intr_status,
|
2013-08-12 12:15:35 +02:00
|
|
|
uint_en->sirfsoc_rx_oflow_en, 0, flag);
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
recv_char:
|
2013-08-12 12:15:35 +02:00
|
|
|
if ((sirfport->uart_reg->uart_type == SIRF_REAL_UART) &&
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
(intr_status & SIRFUART_CTS_INT_ST(uint_st)) &&
|
|
|
|
!sirfport->tx_dma_state) {
|
2013-08-12 12:15:35 +02:00
|
|
|
cts_status = rd_regl(port, ureg->sirfsoc_afc_ctrl) &
|
|
|
|
SIRFUART_AFC_CTS_STATUS;
|
|
|
|
if (cts_status != 0)
|
|
|
|
cts_status = 0;
|
|
|
|
else
|
|
|
|
cts_status = 1;
|
|
|
|
uart_handle_cts_change(port, cts_status);
|
|
|
|
wake_up_interruptible(&state->port.delta_msr_wait);
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
if (!sirfport->rx_dma_chan &&
|
|
|
|
(intr_status & SIRFUART_RX_IO_INT_ST(uint_st))) {
|
2015-05-14 08:45:21 +02:00
|
|
|
/*
|
|
|
|
* chip will trigger continuous RX_TIMEOUT interrupt
|
|
|
|
* in RXFIFO empty and not trigger if RXFIFO recevice
|
|
|
|
* data in limit time, original method use RX_TIMEOUT
|
|
|
|
* will trigger lots of useless interrupt in RXFIFO
|
|
|
|
* empty.RXFIFO received one byte will trigger RX_DONE
|
|
|
|
* interrupt.use RX_DONE to wait for data received
|
|
|
|
* into RXFIFO, use RX_THD/RX_FULL for lots data receive
|
|
|
|
* and use RX_TIMEOUT for the last left data.
|
|
|
|
*/
|
|
|
|
if (intr_status & uint_st->sirfsoc_rx_done) {
|
|
|
|
if (!sirfport->is_atlas7) {
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg)
|
|
|
|
& ~(uint_en->sirfsoc_rx_done_en));
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg)
|
|
|
|
| (uint_en->sirfsoc_rx_timeout_en));
|
|
|
|
} else {
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_clr_reg,
|
|
|
|
uint_en->sirfsoc_rx_done_en);
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
uint_en->sirfsoc_rx_timeout_en);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (intr_status & uint_st->sirfsoc_rx_timeout) {
|
|
|
|
if (!sirfport->is_atlas7) {
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg)
|
|
|
|
& ~(uint_en->sirfsoc_rx_timeout_en));
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg)
|
|
|
|
| (uint_en->sirfsoc_rx_done_en));
|
|
|
|
} else {
|
|
|
|
wr_regl(port,
|
|
|
|
ureg->sirfsoc_int_en_clr_reg,
|
|
|
|
uint_en->sirfsoc_rx_timeout_en);
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
uint_en->sirfsoc_rx_done_en);
|
|
|
|
}
|
|
|
|
}
|
2015-04-29 08:45:09 +02:00
|
|
|
sirfsoc_uart_pio_rx_chars(port, port->fifosize);
|
2015-05-14 08:45:21 +02:00
|
|
|
}
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
}
|
2014-05-26 13:02:07 +02:00
|
|
|
spin_unlock(&port->lock);
|
|
|
|
tty_flip_buffer_push(&state->port);
|
|
|
|
spin_lock(&port->lock);
|
2013-08-12 12:15:35 +02:00
|
|
|
if (intr_status & uint_st->sirfsoc_txfifo_empty) {
|
2014-01-30 06:57:29 +01:00
|
|
|
if (sirfport->tx_dma_chan)
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
sirfsoc_uart_tx_with_dma(sirfport);
|
|
|
|
else {
|
|
|
|
if (uart_circ_empty(xmit) || uart_tx_stopped(port)) {
|
|
|
|
spin_unlock(&port->lock);
|
|
|
|
return IRQ_HANDLED;
|
|
|
|
} else {
|
|
|
|
sirfsoc_uart_pio_tx_chars(sirfport,
|
2015-04-29 08:45:09 +02:00
|
|
|
port->fifosize);
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
if ((uart_circ_empty(xmit)) &&
|
2013-08-12 12:15:35 +02:00
|
|
|
(rd_regl(port, ureg->sirfsoc_tx_fifo_status) &
|
2015-04-29 08:45:09 +02:00
|
|
|
ufifo_st->ff_empty(port)))
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
sirfsoc_uart_stop_tx(port);
|
|
|
|
}
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
}
|
2012-12-25 10:32:04 +01:00
|
|
|
spin_unlock(&port->lock);
|
2014-05-26 13:02:07 +02:00
|
|
|
|
2011-11-17 16:17:04 +01:00
|
|
|
return IRQ_HANDLED;
|
|
|
|
}
|
|
|
|
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
static void sirfsoc_uart_rx_dma_complete_callback(void *param)
|
|
|
|
{
|
|
|
|
}
|
|
|
|
|
|
|
|
/* submit rx dma task into dmaengine */
|
|
|
|
static void sirfsoc_uart_start_next_rx_dma(struct uart_port *port)
|
2011-11-17 16:17:04 +01:00
|
|
|
{
|
2013-08-07 07:35:38 +02:00
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_int_en *uint_en = &sirfport->uart_reg->uart_int_en;
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_rx_dma_io_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_rx_dma_io_ctrl) &
|
|
|
|
~SIRFUART_IO_MODE);
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
sirfport->rx_dma_items.xmit.tail =
|
|
|
|
sirfport->rx_dma_items.xmit.head = 0;
|
|
|
|
sirfport->rx_dma_items.desc =
|
|
|
|
dmaengine_prep_dma_cyclic(sirfport->rx_dma_chan,
|
|
|
|
sirfport->rx_dma_items.dma_addr, SIRFSOC_RX_DMA_BUF_SIZE,
|
|
|
|
SIRFSOC_RX_DMA_BUF_SIZE / 2,
|
|
|
|
DMA_DEV_TO_MEM, DMA_PREP_INTERRUPT);
|
|
|
|
if (IS_ERR_OR_NULL(sirfport->rx_dma_items.desc)) {
|
|
|
|
dev_err(port->dev, "DMA slave single fail\n");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
sirfport->rx_dma_items.desc->callback =
|
|
|
|
sirfsoc_uart_rx_dma_complete_callback;
|
|
|
|
sirfport->rx_dma_items.desc->callback_param = sirfport;
|
|
|
|
sirfport->rx_dma_items.cookie =
|
|
|
|
dmaengine_submit(sirfport->rx_dma_items.desc);
|
|
|
|
dma_async_issue_pending(sirfport->rx_dma_chan);
|
2015-01-03 10:02:57 +01:00
|
|
|
if (!sirfport->is_atlas7)
|
2013-08-12 12:15:35 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg) |
|
2015-05-14 08:45:21 +02:00
|
|
|
SIRFUART_RX_DMA_INT_EN(uint_en,
|
|
|
|
sirfport->uart_reg->uart_type));
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
else
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
2015-05-14 08:45:21 +02:00
|
|
|
SIRFUART_RX_DMA_INT_EN(uint_en,
|
|
|
|
sirfport->uart_reg->uart_type));
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
}
|
|
|
|
|
2013-08-12 12:15:35 +02:00
|
|
|
static unsigned int
|
|
|
|
sirfsoc_usp_calc_sample_div(unsigned long set_rate,
|
|
|
|
unsigned long ioclk_rate, unsigned long *sample_reg)
|
|
|
|
{
|
|
|
|
unsigned long min_delta = ~0UL;
|
|
|
|
unsigned short sample_div;
|
|
|
|
unsigned long ioclk_div = 0;
|
|
|
|
unsigned long temp_delta;
|
|
|
|
|
2015-04-29 08:45:09 +02:00
|
|
|
for (sample_div = SIRF_USP_MIN_SAMPLE_DIV;
|
2013-08-12 12:15:35 +02:00
|
|
|
sample_div <= SIRF_MAX_SAMPLE_DIV; sample_div++) {
|
|
|
|
temp_delta = ioclk_rate -
|
|
|
|
(ioclk_rate + (set_rate * sample_div) / 2)
|
|
|
|
/ (set_rate * sample_div) * set_rate * sample_div;
|
2013-08-07 07:35:38 +02:00
|
|
|
|
2013-08-12 12:15:35 +02:00
|
|
|
temp_delta = (temp_delta > 0) ? temp_delta : -temp_delta;
|
|
|
|
if (temp_delta < min_delta) {
|
|
|
|
ioclk_div = (2 * ioclk_rate /
|
|
|
|
(set_rate * sample_div) + 1) / 2 - 1;
|
|
|
|
if (ioclk_div > SIRF_IOCLK_DIV_MAX)
|
|
|
|
continue;
|
|
|
|
min_delta = temp_delta;
|
|
|
|
*sample_reg = sample_div;
|
|
|
|
if (!temp_delta)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return ioclk_div;
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static unsigned int
|
2013-08-12 12:15:35 +02:00
|
|
|
sirfsoc_uart_calc_sample_div(unsigned long baud_rate,
|
|
|
|
unsigned long ioclk_rate, unsigned long *set_baud)
|
2011-11-17 16:17:04 +01:00
|
|
|
{
|
|
|
|
unsigned long min_delta = ~0UL;
|
|
|
|
unsigned short sample_div;
|
|
|
|
unsigned int regv = 0;
|
|
|
|
unsigned long ioclk_div;
|
|
|
|
unsigned long baud_tmp;
|
|
|
|
int temp_delta;
|
|
|
|
|
|
|
|
for (sample_div = SIRF_MIN_SAMPLE_DIV;
|
|
|
|
sample_div <= SIRF_MAX_SAMPLE_DIV; sample_div++) {
|
|
|
|
ioclk_div = (ioclk_rate / (baud_rate * (sample_div + 1))) - 1;
|
|
|
|
if (ioclk_div > SIRF_IOCLK_DIV_MAX)
|
|
|
|
continue;
|
|
|
|
baud_tmp = ioclk_rate / ((ioclk_div + 1) * (sample_div + 1));
|
|
|
|
temp_delta = baud_tmp - baud_rate;
|
|
|
|
temp_delta = (temp_delta > 0) ? temp_delta : -temp_delta;
|
|
|
|
if (temp_delta < min_delta) {
|
|
|
|
regv = regv & (~SIRF_IOCLK_DIV_MASK);
|
|
|
|
regv = regv | ioclk_div;
|
|
|
|
regv = regv & (~SIRF_SAMPLE_DIV_MASK);
|
|
|
|
regv = regv | (sample_div << SIRF_SAMPLE_DIV_SHIFT);
|
|
|
|
min_delta = temp_delta;
|
2013-08-12 12:15:35 +02:00
|
|
|
*set_baud = baud_tmp;
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
return regv;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_set_termios(struct uart_port *port,
|
|
|
|
struct ktermios *termios,
|
|
|
|
struct ktermios *old)
|
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_int_en *uint_en = &sirfport->uart_reg->uart_int_en;
|
2011-11-17 16:17:04 +01:00
|
|
|
unsigned long config_reg = 0;
|
|
|
|
unsigned long baud_rate;
|
2013-08-12 12:15:35 +02:00
|
|
|
unsigned long set_baud;
|
2011-11-17 16:17:04 +01:00
|
|
|
unsigned long flags;
|
|
|
|
unsigned long ic;
|
|
|
|
unsigned int clk_div_reg = 0;
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
unsigned long txfifo_op_reg, ioclk_rate;
|
2011-11-17 16:17:04 +01:00
|
|
|
unsigned long rx_time_out;
|
|
|
|
int threshold_div;
|
2013-08-12 12:15:35 +02:00
|
|
|
u32 data_bit_len, stop_bit_len, len_val;
|
|
|
|
unsigned long sample_div_reg = 0xf;
|
|
|
|
ioclk_rate = port->uartclk;
|
2011-11-17 16:17:04 +01:00
|
|
|
|
|
|
|
switch (termios->c_cflag & CSIZE) {
|
|
|
|
default:
|
|
|
|
case CS8:
|
2013-08-12 12:15:35 +02:00
|
|
|
data_bit_len = 8;
|
2011-11-17 16:17:04 +01:00
|
|
|
config_reg |= SIRFUART_DATA_BIT_LEN_8;
|
|
|
|
break;
|
|
|
|
case CS7:
|
2013-08-12 12:15:35 +02:00
|
|
|
data_bit_len = 7;
|
2011-11-17 16:17:04 +01:00
|
|
|
config_reg |= SIRFUART_DATA_BIT_LEN_7;
|
|
|
|
break;
|
|
|
|
case CS6:
|
2013-08-12 12:15:35 +02:00
|
|
|
data_bit_len = 6;
|
2011-11-17 16:17:04 +01:00
|
|
|
config_reg |= SIRFUART_DATA_BIT_LEN_6;
|
|
|
|
break;
|
|
|
|
case CS5:
|
2013-08-12 12:15:35 +02:00
|
|
|
data_bit_len = 5;
|
2011-11-17 16:17:04 +01:00
|
|
|
config_reg |= SIRFUART_DATA_BIT_LEN_5;
|
|
|
|
break;
|
|
|
|
}
|
2013-08-12 12:15:35 +02:00
|
|
|
if (termios->c_cflag & CSTOPB) {
|
2011-11-17 16:17:04 +01:00
|
|
|
config_reg |= SIRFUART_STOP_BIT_LEN_2;
|
2013-08-12 12:15:35 +02:00
|
|
|
stop_bit_len = 2;
|
|
|
|
} else
|
|
|
|
stop_bit_len = 1;
|
|
|
|
|
2011-11-17 16:17:04 +01:00
|
|
|
spin_lock_irqsave(&port->lock, flags);
|
2013-08-12 12:15:35 +02:00
|
|
|
port->read_status_mask = uint_en->sirfsoc_rx_oflow_en;
|
2011-11-17 16:17:04 +01:00
|
|
|
port->ignore_status_mask = 0;
|
2013-08-12 12:15:35 +02:00
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_REAL_UART) {
|
|
|
|
if (termios->c_iflag & INPCK)
|
|
|
|
port->read_status_mask |= uint_en->sirfsoc_frm_err_en |
|
|
|
|
uint_en->sirfsoc_parity_err_en;
|
2013-08-15 00:52:15 +02:00
|
|
|
} else {
|
2013-08-12 12:15:35 +02:00
|
|
|
if (termios->c_iflag & INPCK)
|
|
|
|
port->read_status_mask |= uint_en->sirfsoc_frm_err_en;
|
|
|
|
}
|
2014-06-16 14:10:41 +02:00
|
|
|
if (termios->c_iflag & (IGNBRK | BRKINT | PARMRK))
|
2013-08-12 12:15:35 +02:00
|
|
|
port->read_status_mask |= uint_en->sirfsoc_rxd_brk_en;
|
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_REAL_UART) {
|
|
|
|
if (termios->c_iflag & IGNPAR)
|
|
|
|
port->ignore_status_mask |=
|
|
|
|
uint_en->sirfsoc_frm_err_en |
|
|
|
|
uint_en->sirfsoc_parity_err_en;
|
|
|
|
if (termios->c_cflag & PARENB) {
|
|
|
|
if (termios->c_cflag & CMSPAR) {
|
|
|
|
if (termios->c_cflag & PARODD)
|
|
|
|
config_reg |= SIRFUART_STICK_BIT_MARK;
|
|
|
|
else
|
|
|
|
config_reg |= SIRFUART_STICK_BIT_SPACE;
|
|
|
|
} else {
|
2015-05-14 08:45:24 +02:00
|
|
|
if (termios->c_cflag & PARODD)
|
|
|
|
config_reg |= SIRFUART_STICK_BIT_ODD;
|
|
|
|
else
|
|
|
|
config_reg |= SIRFUART_STICK_BIT_EVEN;
|
2013-08-12 12:15:35 +02:00
|
|
|
}
|
|
|
|
}
|
2013-08-15 00:52:15 +02:00
|
|
|
} else {
|
2013-08-12 12:15:35 +02:00
|
|
|
if (termios->c_iflag & IGNPAR)
|
|
|
|
port->ignore_status_mask |=
|
|
|
|
uint_en->sirfsoc_frm_err_en;
|
|
|
|
if (termios->c_cflag & PARENB)
|
|
|
|
dev_warn(port->dev,
|
|
|
|
"USP-UART not support parity err\n");
|
|
|
|
}
|
|
|
|
if (termios->c_iflag & IGNBRK) {
|
2011-11-17 16:17:04 +01:00
|
|
|
port->ignore_status_mask |=
|
2013-08-12 12:15:35 +02:00
|
|
|
uint_en->sirfsoc_rxd_brk_en;
|
|
|
|
if (termios->c_iflag & IGNPAR)
|
|
|
|
port->ignore_status_mask |=
|
|
|
|
uint_en->sirfsoc_rx_oflow_en;
|
|
|
|
}
|
2011-11-17 16:17:04 +01:00
|
|
|
if ((termios->c_cflag & CREAD) == 0)
|
|
|
|
port->ignore_status_mask |= SIRFUART_DUMMY_READ;
|
|
|
|
/* Hardware Flow Control Settings */
|
|
|
|
if (UART_ENABLE_MS(port, termios->c_cflag)) {
|
|
|
|
if (!sirfport->ms_enabled)
|
|
|
|
sirfsoc_uart_enable_ms(port);
|
|
|
|
} else {
|
|
|
|
if (sirfport->ms_enabled)
|
|
|
|
sirfsoc_uart_disable_ms(port);
|
|
|
|
}
|
2013-08-12 12:15:35 +02:00
|
|
|
baud_rate = uart_get_baud_rate(port, termios, old, 0, 4000000);
|
|
|
|
if (ioclk_rate == 150000000) {
|
2013-01-16 07:49:27 +01:00
|
|
|
for (ic = 0; ic < SIRF_BAUD_RATE_SUPPORT_NR; ic++)
|
|
|
|
if (baud_rate == baudrate_to_regv[ic].baud_rate)
|
|
|
|
clk_div_reg = baudrate_to_regv[ic].reg_val;
|
|
|
|
}
|
2013-08-12 12:15:35 +02:00
|
|
|
set_baud = baud_rate;
|
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_REAL_UART) {
|
|
|
|
if (unlikely(clk_div_reg == 0))
|
|
|
|
clk_div_reg = sirfsoc_uart_calc_sample_div(baud_rate,
|
|
|
|
ioclk_rate, &set_baud);
|
|
|
|
wr_regl(port, ureg->sirfsoc_divisor, clk_div_reg);
|
2013-08-15 00:52:15 +02:00
|
|
|
} else {
|
2013-08-12 12:15:35 +02:00
|
|
|
clk_div_reg = sirfsoc_usp_calc_sample_div(baud_rate,
|
|
|
|
ioclk_rate, &sample_div_reg);
|
|
|
|
sample_div_reg--;
|
|
|
|
set_baud = ((ioclk_rate / (clk_div_reg+1) - 1) /
|
|
|
|
(sample_div_reg + 1));
|
|
|
|
/* setting usp mode 2 */
|
2013-08-25 14:18:40 +02:00
|
|
|
len_val = ((1 << SIRFSOC_USP_MODE2_RXD_DELAY_OFFSET) |
|
|
|
|
(1 << SIRFSOC_USP_MODE2_TXD_DELAY_OFFSET));
|
|
|
|
len_val |= ((clk_div_reg & SIRFSOC_USP_MODE2_CLK_DIVISOR_MASK)
|
|
|
|
<< SIRFSOC_USP_MODE2_CLK_DIVISOR_OFFSET);
|
|
|
|
wr_regl(port, ureg->sirfsoc_mode2, len_val);
|
2013-08-12 12:15:35 +02:00
|
|
|
}
|
2011-11-17 16:17:04 +01:00
|
|
|
if (tty_termios_baud_rate(termios))
|
2013-08-12 12:15:35 +02:00
|
|
|
tty_termios_encode_baud_rate(termios, set_baud, set_baud);
|
|
|
|
/* set receive timeout && data bits len */
|
|
|
|
rx_time_out = SIRFSOC_UART_RX_TIMEOUT(set_baud, 20000);
|
|
|
|
rx_time_out = SIRFUART_RECV_TIMEOUT_VALUE(rx_time_out);
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
txfifo_op_reg = rd_regl(port, ureg->sirfsoc_tx_fifo_op);
|
2013-08-12 12:15:35 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_op,
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
(txfifo_op_reg & ~SIRFUART_FIFO_START));
|
2013-08-12 12:15:35 +02:00
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_REAL_UART) {
|
2015-05-14 08:45:21 +02:00
|
|
|
config_reg |= SIRFUART_UART_RECV_TIMEOUT(rx_time_out);
|
2013-08-12 12:15:35 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_line_ctrl, config_reg);
|
2013-08-15 00:52:15 +02:00
|
|
|
} else {
|
2013-08-12 12:15:35 +02:00
|
|
|
/*tx frame ctrl*/
|
2013-08-25 14:18:40 +02:00
|
|
|
len_val = (data_bit_len - 1) << SIRFSOC_USP_TX_DATA_LEN_OFFSET;
|
|
|
|
len_val |= (data_bit_len + 1 + stop_bit_len - 1) <<
|
|
|
|
SIRFSOC_USP_TX_FRAME_LEN_OFFSET;
|
|
|
|
len_val |= ((data_bit_len - 1) <<
|
|
|
|
SIRFSOC_USP_TX_SHIFTER_LEN_OFFSET);
|
|
|
|
len_val |= (((clk_div_reg & 0xc00) >> 10) <<
|
|
|
|
SIRFSOC_USP_TX_CLK_DIVISOR_OFFSET);
|
2013-08-12 12:15:35 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_tx_frame_ctrl, len_val);
|
|
|
|
/*rx frame ctrl*/
|
2013-08-25 14:18:40 +02:00
|
|
|
len_val = (data_bit_len - 1) << SIRFSOC_USP_RX_DATA_LEN_OFFSET;
|
|
|
|
len_val |= (data_bit_len + 1 + stop_bit_len - 1) <<
|
|
|
|
SIRFSOC_USP_RX_FRAME_LEN_OFFSET;
|
|
|
|
len_val |= (data_bit_len - 1) <<
|
|
|
|
SIRFSOC_USP_RX_SHIFTER_LEN_OFFSET;
|
|
|
|
len_val |= (((clk_div_reg & 0xf000) >> 12) <<
|
|
|
|
SIRFSOC_USP_RX_CLK_DIVISOR_OFFSET);
|
2013-08-12 12:15:35 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_rx_frame_ctrl, len_val);
|
|
|
|
/*async param*/
|
|
|
|
wr_regl(port, ureg->sirfsoc_async_param_reg,
|
2015-05-14 08:45:21 +02:00
|
|
|
(SIRFUART_USP_RECV_TIMEOUT(rx_time_out)) |
|
2013-08-25 14:18:40 +02:00
|
|
|
(sample_div_reg & SIRFSOC_USP_ASYNC_DIV2_MASK) <<
|
|
|
|
SIRFSOC_USP_ASYNC_DIV2_OFFSET);
|
2013-08-12 12:15:35 +02:00
|
|
|
}
|
2014-01-30 06:57:29 +01:00
|
|
|
if (sirfport->tx_dma_chan)
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_tx_dma_io_ctrl, SIRFUART_DMA_MODE);
|
|
|
|
else
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_dma_io_ctrl, SIRFUART_IO_MODE);
|
2014-01-30 06:57:29 +01:00
|
|
|
if (sirfport->rx_dma_chan)
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_rx_dma_io_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_rx_dma_io_ctrl) &
|
|
|
|
~SIRFUART_IO_MODE);
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
else
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_rx_dma_io_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_rx_dma_io_ctrl) |
|
|
|
|
SIRFUART_IO_MODE);
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
sirfport->rx_period_time = 20000000;
|
2011-11-17 16:17:04 +01:00
|
|
|
/* Reset Rx/Tx FIFO Threshold level for proper baudrate */
|
2013-08-12 12:15:35 +02:00
|
|
|
if (set_baud < 1000000)
|
2011-11-17 16:17:04 +01:00
|
|
|
threshold_div = 1;
|
|
|
|
else
|
|
|
|
threshold_div = 2;
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_ctrl,
|
|
|
|
SIRFUART_FIFO_THD(port) / threshold_div);
|
|
|
|
wr_regl(port, ureg->sirfsoc_rx_fifo_ctrl,
|
|
|
|
SIRFUART_FIFO_THD(port) / threshold_div);
|
|
|
|
txfifo_op_reg |= SIRFUART_FIFO_START;
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_op, txfifo_op_reg);
|
2013-08-12 12:15:35 +02:00
|
|
|
uart_update_timeout(port, termios->c_cflag, set_baud);
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_rx_en, SIRFUART_TX_EN | SIRFUART_RX_EN);
|
2011-11-17 16:17:04 +01:00
|
|
|
spin_unlock_irqrestore(&port->lock, flags);
|
|
|
|
}
|
|
|
|
|
2014-01-03 08:44:07 +01:00
|
|
|
static void sirfsoc_uart_pm(struct uart_port *port, unsigned int state,
|
|
|
|
unsigned int oldstate)
|
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
2015-04-20 10:10:22 +02:00
|
|
|
if (!state)
|
2014-01-03 08:44:07 +01:00
|
|
|
clk_prepare_enable(sirfport->clk);
|
2015-04-20 10:10:22 +02:00
|
|
|
else
|
2014-01-03 08:44:07 +01:00
|
|
|
clk_disable_unprepare(sirfport->clk);
|
|
|
|
}
|
|
|
|
|
2011-11-17 16:17:04 +01:00
|
|
|
static int sirfsoc_uart_startup(struct uart_port *port)
|
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
2013-08-19 05:47:52 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
2015-07-14 02:52:23 +02:00
|
|
|
struct sirfsoc_int_en *uint_en = &sirfport->uart_reg->uart_int_en;
|
2011-11-17 16:17:04 +01:00
|
|
|
unsigned int index = port->line;
|
|
|
|
int ret;
|
2015-06-09 20:26:39 +02:00
|
|
|
irq_modify_status(port->irq, IRQ_NOREQUEST, IRQ_NOAUTOEN);
|
2011-11-17 16:17:04 +01:00
|
|
|
ret = request_irq(port->irq,
|
|
|
|
sirfsoc_uart_isr,
|
|
|
|
0,
|
|
|
|
SIRFUART_PORT_NAME,
|
|
|
|
sirfport);
|
|
|
|
if (ret != 0) {
|
|
|
|
dev_err(port->dev, "UART%d request IRQ line (%d) failed.\n",
|
|
|
|
index, port->irq);
|
|
|
|
goto irq_err;
|
|
|
|
}
|
2013-08-19 05:47:52 +02:00
|
|
|
/* initial hardware settings */
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_dma_io_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_tx_dma_io_ctrl) |
|
|
|
|
SIRFUART_IO_MODE);
|
|
|
|
wr_regl(port, ureg->sirfsoc_rx_dma_io_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_rx_dma_io_ctrl) |
|
|
|
|
SIRFUART_IO_MODE);
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_rx_dma_io_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_rx_dma_io_ctrl) &
|
|
|
|
~SIRFUART_RX_DMA_FLUSH);
|
2013-08-19 05:47:52 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_tx_dma_io_len, 0);
|
|
|
|
wr_regl(port, ureg->sirfsoc_rx_dma_io_len, 0);
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_rx_en, SIRFUART_RX_EN | SIRFUART_TX_EN);
|
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_USP_UART)
|
|
|
|
wr_regl(port, ureg->sirfsoc_mode1,
|
|
|
|
SIRFSOC_USP_ENDIAN_CTRL_LSBF |
|
|
|
|
SIRFSOC_USP_EN);
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_op, SIRFUART_FIFO_RESET);
|
|
|
|
wr_regl(port, ureg->sirfsoc_rx_fifo_op, SIRFUART_FIFO_RESET);
|
|
|
|
wr_regl(port, ureg->sirfsoc_rx_fifo_op, 0);
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_ctrl, SIRFUART_FIFO_THD(port));
|
|
|
|
wr_regl(port, ureg->sirfsoc_rx_fifo_ctrl, SIRFUART_FIFO_THD(port));
|
2014-01-30 06:57:29 +01:00
|
|
|
if (sirfport->rx_dma_chan)
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_rx_fifo_level_chk,
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
SIRFUART_RX_FIFO_CHK_SC(port->line, 0x1) |
|
|
|
|
SIRFUART_RX_FIFO_CHK_LC(port->line, 0x2) |
|
|
|
|
SIRFUART_RX_FIFO_CHK_HC(port->line, 0x4));
|
2014-01-30 06:57:29 +01:00
|
|
|
if (sirfport->tx_dma_chan) {
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
sirfport->tx_dma_state = TX_DMA_IDLE;
|
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_level_chk,
|
|
|
|
SIRFUART_TX_FIFO_CHK_SC(port->line, 0x1b) |
|
|
|
|
SIRFUART_TX_FIFO_CHK_LC(port->line, 0xe) |
|
|
|
|
SIRFUART_TX_FIFO_CHK_HC(port->line, 0x4));
|
|
|
|
}
|
2013-08-15 00:52:15 +02:00
|
|
|
sirfport->ms_enabled = false;
|
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_USP_UART &&
|
|
|
|
sirfport->hw_flow_ctrl) {
|
2015-06-09 20:26:39 +02:00
|
|
|
irq_modify_status(gpio_to_irq(sirfport->cts_gpio),
|
|
|
|
IRQ_NOREQUEST, IRQ_NOAUTOEN);
|
2013-08-15 00:52:15 +02:00
|
|
|
ret = request_irq(gpio_to_irq(sirfport->cts_gpio),
|
|
|
|
sirfsoc_uart_usp_cts_handler, IRQF_TRIGGER_FALLING |
|
|
|
|
IRQF_TRIGGER_RISING, "usp_cts_irq", sirfport);
|
|
|
|
if (ret != 0) {
|
|
|
|
dev_err(port->dev, "UART-USP:request gpio irq fail\n");
|
|
|
|
goto init_rx_err;
|
|
|
|
}
|
|
|
|
}
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_REAL_UART &&
|
|
|
|
sirfport->rx_dma_chan)
|
|
|
|
wr_regl(port, ureg->sirfsoc_swh_dma_io,
|
|
|
|
SIRFUART_CLEAR_RX_ADDR_EN);
|
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_USP_UART &&
|
|
|
|
sirfport->rx_dma_chan)
|
|
|
|
wr_regl(port, ureg->sirfsoc_rx_dma_io_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_rx_dma_io_ctrl) |
|
|
|
|
SIRFSOC_USP_FRADDR_CLR_EN);
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
if (sirfport->rx_dma_chan && !sirfport->is_hrt_enabled) {
|
|
|
|
sirfport->is_hrt_enabled = true;
|
|
|
|
sirfport->rx_period_time = 20000000;
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
sirfport->rx_last_pos = -1;
|
|
|
|
sirfport->pio_fetch_cnt = 0;
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
sirfport->rx_dma_items.xmit.tail =
|
|
|
|
sirfport->rx_dma_items.xmit.head = 0;
|
|
|
|
hrtimer_start(&sirfport->hrt,
|
|
|
|
ns_to_ktime(sirfport->rx_period_time),
|
|
|
|
HRTIMER_MODE_REL);
|
|
|
|
}
|
2015-07-14 02:52:23 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_rx_fifo_op, SIRFUART_FIFO_START);
|
|
|
|
if (sirfport->rx_dma_chan)
|
|
|
|
sirfsoc_uart_start_next_rx_dma(port);
|
|
|
|
else {
|
|
|
|
if (!sirfport->is_atlas7)
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
rd_regl(port, ureg->sirfsoc_int_en_reg) |
|
|
|
|
SIRFUART_RX_IO_INT_EN(uint_en,
|
|
|
|
sirfport->uart_reg->uart_type));
|
|
|
|
else
|
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg,
|
|
|
|
SIRFUART_RX_IO_INT_EN(uint_en,
|
|
|
|
sirfport->uart_reg->uart_type));
|
|
|
|
}
|
|
|
|
enable_irq(port->irq);
|
2013-08-15 00:52:15 +02:00
|
|
|
|
2013-08-19 05:47:52 +02:00
|
|
|
return 0;
|
2013-08-15 00:52:15 +02:00
|
|
|
init_rx_err:
|
|
|
|
free_irq(port->irq, sirfport);
|
2011-11-17 16:17:04 +01:00
|
|
|
irq_err:
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_shutdown(struct uart_port *port)
|
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
struct circ_buf *xmit;
|
|
|
|
|
|
|
|
xmit = &sirfport->rx_dma_items.xmit;
|
2015-01-03 10:02:57 +01:00
|
|
|
if (!sirfport->is_atlas7)
|
2013-08-12 12:15:35 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_reg, 0);
|
2013-08-07 07:35:38 +02:00
|
|
|
else
|
2015-05-14 08:45:21 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_int_en_clr_reg, ~0UL);
|
2013-08-07 07:35:38 +02:00
|
|
|
|
2011-11-17 16:17:04 +01:00
|
|
|
free_irq(port->irq, sirfport);
|
2013-08-15 00:52:15 +02:00
|
|
|
if (sirfport->ms_enabled)
|
2011-11-17 16:17:04 +01:00
|
|
|
sirfsoc_uart_disable_ms(port);
|
2013-08-15 00:52:15 +02:00
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_USP_UART &&
|
|
|
|
sirfport->hw_flow_ctrl) {
|
|
|
|
gpio_set_value(sirfport->rts_gpio, 1);
|
|
|
|
free_irq(gpio_to_irq(sirfport->cts_gpio), sirfport);
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
2014-01-30 06:57:29 +01:00
|
|
|
if (sirfport->tx_dma_chan)
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
sirfport->tx_dma_state = TX_DMA_IDLE;
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
if (sirfport->rx_dma_chan && sirfport->is_hrt_enabled) {
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
while (((rd_regl(port, ureg->sirfsoc_rx_fifo_status) &
|
|
|
|
SIRFUART_RX_FIFO_MASK) > sirfport->pio_fetch_cnt) &&
|
|
|
|
!CIRC_CNT(xmit->head, xmit->tail,
|
|
|
|
SIRFSOC_RX_DMA_BUF_SIZE))
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
;
|
|
|
|
sirfport->is_hrt_enabled = false;
|
|
|
|
hrtimer_cancel(&sirfport->hrt);
|
|
|
|
}
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static const char *sirfsoc_uart_type(struct uart_port *port)
|
|
|
|
{
|
|
|
|
return port->type == SIRFSOC_PORT_TYPE ? SIRFUART_PORT_NAME : NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int sirfsoc_uart_request_port(struct uart_port *port)
|
|
|
|
{
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
|
|
|
struct sirfsoc_uart_param *uart_param = &sirfport->uart_reg->uart_param;
|
2011-11-17 16:17:04 +01:00
|
|
|
void *ret;
|
|
|
|
ret = request_mem_region(port->mapbase,
|
2013-08-12 12:15:35 +02:00
|
|
|
SIRFUART_MAP_SIZE, uart_param->port_name);
|
2011-11-17 16:17:04 +01:00
|
|
|
return ret ? 0 : -EBUSY;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_release_port(struct uart_port *port)
|
|
|
|
{
|
|
|
|
release_mem_region(port->mapbase, SIRFUART_MAP_SIZE);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_config_port(struct uart_port *port, int flags)
|
|
|
|
{
|
|
|
|
if (flags & UART_CONFIG_TYPE) {
|
|
|
|
port->type = SIRFSOC_PORT_TYPE;
|
|
|
|
sirfsoc_uart_request_port(port);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-01-25 18:48:52 +01:00
|
|
|
static const struct uart_ops sirfsoc_uart_ops = {
|
2011-11-17 16:17:04 +01:00
|
|
|
.tx_empty = sirfsoc_uart_tx_empty,
|
|
|
|
.get_mctrl = sirfsoc_uart_get_mctrl,
|
|
|
|
.set_mctrl = sirfsoc_uart_set_mctrl,
|
|
|
|
.stop_tx = sirfsoc_uart_stop_tx,
|
|
|
|
.start_tx = sirfsoc_uart_start_tx,
|
|
|
|
.stop_rx = sirfsoc_uart_stop_rx,
|
|
|
|
.enable_ms = sirfsoc_uart_enable_ms,
|
|
|
|
.break_ctl = sirfsoc_uart_break_ctl,
|
|
|
|
.startup = sirfsoc_uart_startup,
|
|
|
|
.shutdown = sirfsoc_uart_shutdown,
|
|
|
|
.set_termios = sirfsoc_uart_set_termios,
|
2014-01-03 08:44:07 +01:00
|
|
|
.pm = sirfsoc_uart_pm,
|
2011-11-17 16:17:04 +01:00
|
|
|
.type = sirfsoc_uart_type,
|
|
|
|
.release_port = sirfsoc_uart_release_port,
|
|
|
|
.request_port = sirfsoc_uart_request_port,
|
|
|
|
.config_port = sirfsoc_uart_config_port,
|
|
|
|
};
|
|
|
|
|
|
|
|
#ifdef CONFIG_SERIAL_SIRFSOC_CONSOLE
|
2013-08-12 12:15:35 +02:00
|
|
|
static int __init
|
|
|
|
sirfsoc_uart_console_setup(struct console *co, char *options)
|
2011-11-17 16:17:04 +01:00
|
|
|
{
|
|
|
|
unsigned int baud = 115200;
|
|
|
|
unsigned int bits = 8;
|
|
|
|
unsigned int parity = 'n';
|
|
|
|
unsigned int flow = 'n';
|
2015-04-29 08:45:08 +02:00
|
|
|
struct sirfsoc_uart_port *sirfport;
|
|
|
|
struct sirfsoc_register *ureg;
|
2011-11-17 16:17:04 +01:00
|
|
|
if (co->index < 0 || co->index >= SIRFSOC_UART_NR)
|
2015-05-14 08:45:26 +02:00
|
|
|
co->index = 1;
|
2015-04-29 08:45:08 +02:00
|
|
|
sirfport = sirf_ports[co->index];
|
|
|
|
if (!sirfport)
|
|
|
|
return -ENODEV;
|
|
|
|
ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
if (!sirfport->port.mapbase)
|
2011-11-17 16:17:04 +01:00
|
|
|
return -ENODEV;
|
|
|
|
|
2013-08-12 12:15:35 +02:00
|
|
|
/* enable usp in mode1 register */
|
|
|
|
if (sirfport->uart_reg->uart_type == SIRF_USP_UART)
|
2015-04-29 08:45:08 +02:00
|
|
|
wr_regl(&sirfport->port, ureg->sirfsoc_mode1, SIRFSOC_USP_EN |
|
2013-08-12 12:15:35 +02:00
|
|
|
SIRFSOC_USP_ENDIAN_CTRL_LSBF);
|
2011-11-17 16:17:04 +01:00
|
|
|
if (options)
|
|
|
|
uart_parse_options(options, &baud, &parity, &bits, &flow);
|
2015-04-29 08:45:08 +02:00
|
|
|
sirfport->port.cons = co;
|
2013-08-12 12:15:35 +02:00
|
|
|
|
serial: sirf: add DMA support using dmaengine APIs
if we get the valid dma channels from dts, move to use dmaengine to do
rx/tx. because the dma hardware requires dma address and length to be
4bytes aligned, in this driver, we will still use PIO for non-aligned
bytes, and use dma for aligned bytes.
for rx, to keep the dmaengine always active, we use double-buffer, so
we issue two dma_desc at first, and maintain the status of both
1. dma transfer done: update in rx dma finish callback
2. dma buffer is inserted into tty: update in rx dma finish tasklet and
rx timeout tasklet
so we re-issue the dma_desc only if both 1&2 are finished.
for tx, as we know the actual length for every transfer, we don't need
the above double buffering.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2013-08-19 05:47:53 +02:00
|
|
|
/* default console tx/rx transfer using io mode */
|
2014-01-30 06:57:29 +01:00
|
|
|
sirfport->rx_dma_chan = NULL;
|
|
|
|
sirfport->tx_dma_chan = NULL;
|
2015-04-29 08:45:08 +02:00
|
|
|
return uart_set_options(&sirfport->port, co, baud, parity, bits, flow);
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_console_putchar(struct uart_port *port, int ch)
|
|
|
|
{
|
2013-08-12 12:15:35 +02:00
|
|
|
struct sirfsoc_uart_port *sirfport = to_sirfport(port);
|
|
|
|
struct sirfsoc_register *ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
struct sirfsoc_fifo_status *ufifo_st = &sirfport->uart_reg->fifo_status;
|
2015-04-29 08:45:09 +02:00
|
|
|
while (rd_regl(port, ureg->sirfsoc_tx_fifo_status) &
|
|
|
|
ufifo_st->ff_full(port))
|
2011-11-17 16:17:04 +01:00
|
|
|
cpu_relax();
|
2014-05-05 02:05:51 +02:00
|
|
|
wr_regl(port, ureg->sirfsoc_tx_fifo_data, ch);
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static void sirfsoc_uart_console_write(struct console *co, const char *s,
|
|
|
|
unsigned int count)
|
|
|
|
{
|
2015-04-29 08:45:08 +02:00
|
|
|
struct sirfsoc_uart_port *sirfport = sirf_ports[co->index];
|
|
|
|
|
|
|
|
uart_console_write(&sirfport->port, s, count,
|
|
|
|
sirfsoc_uart_console_putchar);
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct console sirfsoc_uart_console = {
|
|
|
|
.name = SIRFSOC_UART_NAME,
|
|
|
|
.device = uart_console_device,
|
|
|
|
.flags = CON_PRINTBUFFER,
|
|
|
|
.index = -1,
|
|
|
|
.write = sirfsoc_uart_console_write,
|
|
|
|
.setup = sirfsoc_uart_console_setup,
|
|
|
|
.data = &sirfsoc_uart_drv,
|
|
|
|
};
|
|
|
|
|
|
|
|
static int __init sirfsoc_uart_console_init(void)
|
|
|
|
{
|
|
|
|
register_console(&sirfsoc_uart_console);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
console_initcall(sirfsoc_uart_console_init);
|
|
|
|
#endif
|
|
|
|
|
|
|
|
static struct uart_driver sirfsoc_uart_drv = {
|
|
|
|
.owner = THIS_MODULE,
|
|
|
|
.driver_name = SIRFUART_PORT_NAME,
|
|
|
|
.nr = SIRFSOC_UART_NR,
|
|
|
|
.dev_name = SIRFSOC_UART_NAME,
|
|
|
|
.major = SIRFSOC_UART_MAJOR,
|
|
|
|
.minor = SIRFSOC_UART_MINOR,
|
|
|
|
#ifdef CONFIG_SERIAL_SIRFSOC_CONSOLE
|
|
|
|
.cons = &sirfsoc_uart_console,
|
|
|
|
#else
|
|
|
|
.cons = NULL,
|
|
|
|
#endif
|
|
|
|
};
|
|
|
|
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
static enum hrtimer_restart
|
|
|
|
sirfsoc_uart_rx_dma_hrtimer_callback(struct hrtimer *hrt)
|
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport;
|
|
|
|
struct uart_port *port;
|
|
|
|
int count, inserted;
|
|
|
|
struct dma_tx_state tx_state;
|
|
|
|
struct tty_struct *tty;
|
|
|
|
struct sirfsoc_register *ureg;
|
|
|
|
struct circ_buf *xmit;
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
struct sirfsoc_fifo_status *ufifo_st;
|
|
|
|
int max_pio_cnt;
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
|
|
|
|
sirfport = container_of(hrt, struct sirfsoc_uart_port, hrt);
|
|
|
|
port = &sirfport->port;
|
|
|
|
inserted = 0;
|
|
|
|
tty = port->state->port.tty;
|
|
|
|
ureg = &sirfport->uart_reg->uart_reg;
|
|
|
|
xmit = &sirfport->rx_dma_items.xmit;
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
ufifo_st = &sirfport->uart_reg->fifo_status;
|
|
|
|
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
dmaengine_tx_status(sirfport->rx_dma_chan,
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
sirfport->rx_dma_items.cookie, &tx_state);
|
|
|
|
if (SIRFSOC_RX_DMA_BUF_SIZE - tx_state.residue !=
|
|
|
|
sirfport->rx_last_pos) {
|
|
|
|
xmit->head = SIRFSOC_RX_DMA_BUF_SIZE - tx_state.residue;
|
|
|
|
sirfport->rx_last_pos = xmit->head;
|
|
|
|
sirfport->pio_fetch_cnt = 0;
|
|
|
|
}
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
count = CIRC_CNT_TO_END(xmit->head, xmit->tail,
|
|
|
|
SIRFSOC_RX_DMA_BUF_SIZE);
|
|
|
|
while (count > 0) {
|
|
|
|
inserted = tty_insert_flip_string(tty->port,
|
|
|
|
(const unsigned char *)&xmit->buf[xmit->tail], count);
|
|
|
|
if (!inserted)
|
|
|
|
goto next_hrt;
|
|
|
|
port->icount.rx += inserted;
|
|
|
|
xmit->tail = (xmit->tail + inserted) &
|
|
|
|
(SIRFSOC_RX_DMA_BUF_SIZE - 1);
|
|
|
|
count = CIRC_CNT_TO_END(xmit->head, xmit->tail,
|
|
|
|
SIRFSOC_RX_DMA_BUF_SIZE);
|
|
|
|
tty_flip_buffer_push(tty->port);
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* if RX DMA buffer data have all push into tty buffer, and there is
|
|
|
|
* only little data(less than a dma transfer unit) left in rxfifo,
|
|
|
|
* fetch it out in pio mode and switch back to dma immediately
|
|
|
|
*/
|
|
|
|
if (!inserted && !count &&
|
|
|
|
((rd_regl(port, ureg->sirfsoc_rx_fifo_status) &
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
SIRFUART_RX_FIFO_MASK) > sirfport->pio_fetch_cnt)) {
|
|
|
|
dmaengine_pause(sirfport->rx_dma_chan);
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
/* switch to pio mode */
|
|
|
|
wr_regl(port, ureg->sirfsoc_rx_dma_io_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_rx_dma_io_ctrl) |
|
|
|
|
SIRFUART_IO_MODE);
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
/*
|
|
|
|
* UART controller SWH_DMA_IO register have CLEAR_RX_ADDR_EN
|
|
|
|
* When found changing I/O to DMA mode, it clears
|
|
|
|
* two low bits of read point;
|
|
|
|
* USP have similar FRADDR_CLR_EN bit in USP_RX_DMA_IO_CTRL.
|
|
|
|
* Fetch data out from rxfifo into DMA buffer in PIO mode,
|
|
|
|
* while switch back to DMA mode, the data fetched will override
|
|
|
|
* by DMA, as hardware have a strange behaviour:
|
|
|
|
* after switch back to DMA mode, check rxfifo status it will
|
|
|
|
* be the number PIO fetched, so record the fetched data count
|
|
|
|
* to avoid the repeated fetch
|
|
|
|
*/
|
|
|
|
max_pio_cnt = 3;
|
|
|
|
while (!(rd_regl(port, ureg->sirfsoc_rx_fifo_status) &
|
|
|
|
ufifo_st->ff_empty(port)) && max_pio_cnt--) {
|
|
|
|
xmit->buf[xmit->head] =
|
|
|
|
rd_regl(port, ureg->sirfsoc_rx_fifo_data);
|
|
|
|
xmit->head = (xmit->head + 1) &
|
|
|
|
(SIRFSOC_RX_DMA_BUF_SIZE - 1);
|
|
|
|
sirfport->pio_fetch_cnt++;
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
}
|
|
|
|
/* switch back to dma mode */
|
|
|
|
wr_regl(port, ureg->sirfsoc_rx_dma_io_ctrl,
|
|
|
|
rd_regl(port, ureg->sirfsoc_rx_dma_io_ctrl) &
|
|
|
|
~SIRFUART_IO_MODE);
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
dmaengine_resume(sirfport->rx_dma_chan);
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
}
|
|
|
|
next_hrt:
|
|
|
|
hrtimer_forward_now(hrt, ns_to_ktime(sirfport->rx_period_time));
|
|
|
|
return HRTIMER_RESTART;
|
|
|
|
}
|
|
|
|
|
2017-06-15 13:38:40 +02:00
|
|
|
static const struct of_device_id sirfsoc_uart_ids[] = {
|
2013-08-12 12:15:35 +02:00
|
|
|
{ .compatible = "sirf,prima2-uart", .data = &sirfsoc_uart,},
|
2015-01-03 10:02:57 +01:00
|
|
|
{ .compatible = "sirf,atlas7-uart", .data = &sirfsoc_uart},
|
2013-08-12 12:15:35 +02:00
|
|
|
{ .compatible = "sirf,prima2-usp-uart", .data = &sirfsoc_usp},
|
2015-05-14 08:45:21 +02:00
|
|
|
{ .compatible = "sirf,atlas7-usp-uart", .data = &sirfsoc_usp},
|
2013-08-12 12:15:35 +02:00
|
|
|
{}
|
|
|
|
};
|
|
|
|
MODULE_DEVICE_TABLE(of, sirfsoc_uart_ids);
|
|
|
|
|
2013-08-08 10:41:43 +02:00
|
|
|
static int sirfsoc_uart_probe(struct platform_device *pdev)
|
2011-11-17 16:17:04 +01:00
|
|
|
{
|
2016-04-22 17:22:23 +02:00
|
|
|
struct device_node *np = pdev->dev.of_node;
|
2011-11-17 16:17:04 +01:00
|
|
|
struct sirfsoc_uart_port *sirfport;
|
|
|
|
struct uart_port *port;
|
|
|
|
struct resource *res;
|
|
|
|
int ret;
|
2014-01-30 06:57:29 +01:00
|
|
|
struct dma_slave_config slv_cfg = {
|
serial: sirf: workaround rx process to avoid possible data loss
when UART works in DMA mode and left bytes in rx fifo less than
a dma transfer unit, DMA engine can't transfer the bytes out
to rx DMA buffer. so it need a way to fetch them out and
flush them into tty buffer in time.
in the above case, we want UART switch from DMA mode to PIO mode and
fetch && flush bytes into tty layer buffer until rxfifo become empty,
after that done let UART switch from PIO mode back to DMA mode.
(record as method1)
method1 result in the next receive result wrong. for example in PIO part
of method1, we fetched && pushed X1...X3 bytes, when UART rxfifo newly
received Y1...Y4 bytes, UART trigger a DMA unit transfer, the DMA unit's
content is X1...X3Y1 and rxfifo fifo status is empty, so X1X2X3 pushed
twice by PIO way and DMA way also the bytes Y2Y3Y4 missed. add rxfifo
reset operation before UART switch back to DMA mode would resolve the
issue. ([method1 + do fifo reset] record as method2)
before the commit, UART driver use method2. but methd2 have a risk of
data loss, as if UART's shift register receive a complete byte and
transfer it into rxfifo before rxfifo reset operation the byte will
loss.
UART and USP have the similar bits CLEAR_RX_ADDR_EN(uart)/FRADDR_CLR_EN(usp),
When found UART controller changing I/O to DMA mode, UART controller
clears the two low bits of read point (rx_fifo_addr[1:0]).
when enable the bit + method1(record as method3), in above example
the DMA unit's content is X1...X3Y1 and there are Y2Y3Y4 in rxfifo by
experiment, we just push bytes in rx DMA buffer.
BTW, the workaround works only for UART receive DMA channel use SINGLE
DMA mode.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-07-14 02:52:22 +02:00
|
|
|
.src_maxburst = 1,
|
2014-01-30 06:57:29 +01:00
|
|
|
};
|
|
|
|
struct dma_slave_config tx_slv_cfg = {
|
|
|
|
.dst_maxburst = 2,
|
|
|
|
};
|
2013-08-12 12:15:35 +02:00
|
|
|
const struct of_device_id *match;
|
2011-11-17 16:17:04 +01:00
|
|
|
|
2016-04-22 17:22:23 +02:00
|
|
|
match = of_match_node(sirfsoc_uart_ids, np);
|
2015-04-29 08:45:08 +02:00
|
|
|
sirfport = devm_kzalloc(&pdev->dev, sizeof(*sirfport), GFP_KERNEL);
|
|
|
|
if (!sirfport) {
|
|
|
|
ret = -ENOMEM;
|
2011-11-17 16:17:04 +01:00
|
|
|
goto err;
|
|
|
|
}
|
2016-04-22 17:22:23 +02:00
|
|
|
sirfport->port.line = of_alias_get_id(np, "serial");
|
2018-02-23 14:38:36 +01:00
|
|
|
if (sirfport->port.line >= ARRAY_SIZE(sirf_ports)) {
|
|
|
|
dev_err(&pdev->dev, "serial%d out of range\n",
|
|
|
|
sirfport->port.line);
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
2015-04-29 08:45:08 +02:00
|
|
|
sirf_ports[sirfport->port.line] = sirfport;
|
|
|
|
sirfport->port.iotype = UPIO_MEM;
|
|
|
|
sirfport->port.flags = UPF_BOOT_AUTOCONF;
|
2011-11-17 16:17:04 +01:00
|
|
|
port = &sirfport->port;
|
|
|
|
port->dev = &pdev->dev;
|
|
|
|
port->private_data = sirfport;
|
2013-08-12 12:15:35 +02:00
|
|
|
sirfport->uart_reg = (struct sirfsoc_uart_register *)match->data;
|
2011-11-17 16:17:04 +01:00
|
|
|
|
2016-04-22 17:22:23 +02:00
|
|
|
sirfport->hw_flow_ctrl =
|
2016-04-22 17:22:24 +02:00
|
|
|
of_property_read_bool(np, "uart-has-rtscts") ||
|
|
|
|
of_property_read_bool(np, "sirf,uart-has-rtscts") /* deprecated */;
|
2016-04-22 17:22:23 +02:00
|
|
|
if (of_device_is_compatible(np, "sirf,prima2-uart") ||
|
|
|
|
of_device_is_compatible(np, "sirf,atlas7-uart"))
|
2013-08-12 12:15:35 +02:00
|
|
|
sirfport->uart_reg->uart_type = SIRF_REAL_UART;
|
2016-04-22 17:22:23 +02:00
|
|
|
if (of_device_is_compatible(np, "sirf,prima2-usp-uart") ||
|
|
|
|
of_device_is_compatible(np, "sirf,atlas7-usp-uart")) {
|
2013-08-12 12:15:35 +02:00
|
|
|
sirfport->uart_reg->uart_type = SIRF_USP_UART;
|
2013-08-15 00:52:15 +02:00
|
|
|
if (!sirfport->hw_flow_ctrl)
|
|
|
|
goto usp_no_flow_control;
|
2016-04-22 17:22:23 +02:00
|
|
|
if (of_find_property(np, "cts-gpios", NULL))
|
|
|
|
sirfport->cts_gpio =
|
|
|
|
of_get_named_gpio(np, "cts-gpios", 0);
|
2013-08-15 00:52:15 +02:00
|
|
|
else
|
|
|
|
sirfport->cts_gpio = -1;
|
2016-04-22 17:22:23 +02:00
|
|
|
if (of_find_property(np, "rts-gpios", NULL))
|
|
|
|
sirfport->rts_gpio =
|
|
|
|
of_get_named_gpio(np, "rts-gpios", 0);
|
2013-08-15 00:52:15 +02:00
|
|
|
else
|
|
|
|
sirfport->rts_gpio = -1;
|
|
|
|
|
|
|
|
if ((!gpio_is_valid(sirfport->cts_gpio) ||
|
|
|
|
!gpio_is_valid(sirfport->rts_gpio))) {
|
|
|
|
ret = -EINVAL;
|
|
|
|
dev_err(&pdev->dev,
|
2013-08-19 05:47:51 +02:00
|
|
|
"Usp flow control must have cts and rts gpio");
|
2013-08-15 00:52:15 +02:00
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
ret = devm_gpio_request(&pdev->dev, sirfport->cts_gpio,
|
2013-08-19 05:47:51 +02:00
|
|
|
"usp-cts-gpio");
|
2013-08-15 00:52:15 +02:00
|
|
|
if (ret) {
|
2013-08-19 05:47:51 +02:00
|
|
|
dev_err(&pdev->dev, "Unable request cts gpio");
|
2013-08-15 00:52:15 +02:00
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
gpio_direction_input(sirfport->cts_gpio);
|
|
|
|
ret = devm_gpio_request(&pdev->dev, sirfport->rts_gpio,
|
2013-08-19 05:47:51 +02:00
|
|
|
"usp-rts-gpio");
|
2013-08-15 00:52:15 +02:00
|
|
|
if (ret) {
|
2013-08-19 05:47:51 +02:00
|
|
|
dev_err(&pdev->dev, "Unable request rts gpio");
|
2013-08-15 00:52:15 +02:00
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
gpio_direction_output(sirfport->rts_gpio, 1);
|
|
|
|
}
|
|
|
|
usp_no_flow_control:
|
2016-04-22 17:22:23 +02:00
|
|
|
if (of_device_is_compatible(np, "sirf,atlas7-uart") ||
|
|
|
|
of_device_is_compatible(np, "sirf,atlas7-usp-uart"))
|
2015-01-03 10:02:57 +01:00
|
|
|
sirfport->is_atlas7 = true;
|
2013-08-07 07:35:38 +02:00
|
|
|
|
2016-04-22 17:22:23 +02:00
|
|
|
if (of_property_read_u32(np, "fifosize", &port->fifosize)) {
|
2011-11-17 16:17:04 +01:00
|
|
|
dev_err(&pdev->dev,
|
|
|
|
"Unable to find fifosize in uart node.\n");
|
|
|
|
ret = -EFAULT;
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
|
|
|
|
if (res == NULL) {
|
|
|
|
dev_err(&pdev->dev, "Insufficient resources.\n");
|
|
|
|
ret = -EFAULT;
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
port->mapbase = res->start;
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
port->membase = devm_ioremap(&pdev->dev,
|
|
|
|
res->start, resource_size(res));
|
2011-11-17 16:17:04 +01:00
|
|
|
if (!port->membase) {
|
|
|
|
dev_err(&pdev->dev, "Cannot remap resource.\n");
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
|
|
|
|
if (res == NULL) {
|
|
|
|
dev_err(&pdev->dev, "Insufficient resources.\n");
|
|
|
|
ret = -EFAULT;
|
2012-09-01 18:33:09 +02:00
|
|
|
goto err;
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
port->irq = res->start;
|
|
|
|
|
2015-04-20 10:10:23 +02:00
|
|
|
sirfport->clk = devm_clk_get(&pdev->dev, NULL);
|
2013-01-16 07:49:27 +01:00
|
|
|
if (IS_ERR(sirfport->clk)) {
|
|
|
|
ret = PTR_ERR(sirfport->clk);
|
2013-08-15 00:52:14 +02:00
|
|
|
goto err;
|
2013-01-16 07:49:27 +01:00
|
|
|
}
|
|
|
|
port->uartclk = clk_get_rate(sirfport->clk);
|
|
|
|
|
2011-11-17 16:17:04 +01:00
|
|
|
port->ops = &sirfsoc_uart_ops;
|
|
|
|
spin_lock_init(&port->lock);
|
|
|
|
|
|
|
|
platform_set_drvdata(pdev, sirfport);
|
|
|
|
ret = uart_add_one_port(&sirfsoc_uart_drv, port);
|
|
|
|
if (ret != 0) {
|
|
|
|
dev_err(&pdev->dev, "Cannot add UART port(%d).\n", pdev->id);
|
2015-04-20 10:10:23 +02:00
|
|
|
goto err;
|
2011-11-17 16:17:04 +01:00
|
|
|
}
|
|
|
|
|
2014-01-30 06:57:29 +01:00
|
|
|
sirfport->rx_dma_chan = dma_request_slave_channel(port->dev, "rx");
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
sirfport->rx_dma_items.xmit.buf =
|
|
|
|
dma_alloc_coherent(port->dev, SIRFSOC_RX_DMA_BUF_SIZE,
|
|
|
|
&sirfport->rx_dma_items.dma_addr, GFP_KERNEL);
|
|
|
|
if (!sirfport->rx_dma_items.xmit.buf) {
|
|
|
|
dev_err(port->dev, "Uart alloc bufa failed\n");
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto alloc_coherent_err;
|
2014-01-30 06:57:29 +01:00
|
|
|
}
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
sirfport->rx_dma_items.xmit.head =
|
|
|
|
sirfport->rx_dma_items.xmit.tail = 0;
|
2014-01-30 06:57:29 +01:00
|
|
|
if (sirfport->rx_dma_chan)
|
|
|
|
dmaengine_slave_config(sirfport->rx_dma_chan, &slv_cfg);
|
|
|
|
sirfport->tx_dma_chan = dma_request_slave_channel(port->dev, "tx");
|
|
|
|
if (sirfport->tx_dma_chan)
|
|
|
|
dmaengine_slave_config(sirfport->tx_dma_chan, &tx_slv_cfg);
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
if (sirfport->rx_dma_chan) {
|
|
|
|
hrtimer_init(&sirfport->hrt, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
|
|
|
|
sirfport->hrt.function = sirfsoc_uart_rx_dma_hrtimer_callback;
|
|
|
|
sirfport->is_hrt_enabled = false;
|
|
|
|
}
|
2011-11-17 16:17:04 +01:00
|
|
|
|
2014-01-30 06:57:29 +01:00
|
|
|
return 0;
|
|
|
|
alloc_coherent_err:
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
dma_free_coherent(port->dev, SIRFSOC_RX_DMA_BUF_SIZE,
|
|
|
|
sirfport->rx_dma_items.xmit.buf,
|
|
|
|
sirfport->rx_dma_items.dma_addr);
|
2014-01-30 06:57:29 +01:00
|
|
|
dma_release_channel(sirfport->rx_dma_chan);
|
2011-11-17 16:17:04 +01:00
|
|
|
err:
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int sirfsoc_uart_remove(struct platform_device *pdev)
|
|
|
|
{
|
|
|
|
struct sirfsoc_uart_port *sirfport = platform_get_drvdata(pdev);
|
|
|
|
struct uart_port *port = &sirfport->port;
|
|
|
|
uart_remove_one_port(&sirfsoc_uart_drv, port);
|
2014-01-30 06:57:29 +01:00
|
|
|
if (sirfport->rx_dma_chan) {
|
|
|
|
dmaengine_terminate_all(sirfport->rx_dma_chan);
|
|
|
|
dma_release_channel(sirfport->rx_dma_chan);
|
serial: sirf: use hrtimer for data rx
when the serial works as a bluetooth sink, due to audio realtime
requirement, the driver should have something similar with ALSA:
1. one big DMA buffer to easy the schedule jitter
2. split this big DMA buffer to multiple small periods, for each
period, we get a DMA interrupt, then push the data to userspace.
the small periods will easy the audio latency.
so ALSA generally uses a cyclic chained DMA.
but for sirfsoc, the dma hardware has the limitation: we have
only two loops in the cyclic mode, so we can only support two
small periods to switch. if we make the DMA buffer too big, we
get long latency, if we make the DMA buffer too little, we get
miss in scheduling for audio realtime.
so this patch moves to use a hrtimer to simulate the cyclic
DMA, then we can have a big buffer, and also have a timely
data push to users as the hrtimer can generate in small period
then actual HW interrupts.
with this patch, we also delete a lot of complex codes to handle
loop buffers, and RX timeout interrupt since the RX work can be
completely handled from hrtimer interrupt.
tests show using this way will make our bad audio streaming be-
come smooth.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-05-26 11:36:00 +02:00
|
|
|
dma_free_coherent(port->dev, SIRFSOC_RX_DMA_BUF_SIZE,
|
|
|
|
sirfport->rx_dma_items.xmit.buf,
|
|
|
|
sirfport->rx_dma_items.dma_addr);
|
2014-01-30 06:57:29 +01:00
|
|
|
}
|
|
|
|
if (sirfport->tx_dma_chan) {
|
|
|
|
dmaengine_terminate_all(sirfport->tx_dma_chan);
|
|
|
|
dma_release_channel(sirfport->tx_dma_chan);
|
|
|
|
}
|
2011-11-17 16:17:04 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2014-01-03 08:44:06 +01:00
|
|
|
#ifdef CONFIG_PM_SLEEP
|
2011-11-17 16:17:04 +01:00
|
|
|
static int
|
2014-01-03 08:44:06 +01:00
|
|
|
sirfsoc_uart_suspend(struct device *pdev)
|
2011-11-17 16:17:04 +01:00
|
|
|
{
|
2014-01-03 08:44:06 +01:00
|
|
|
struct sirfsoc_uart_port *sirfport = dev_get_drvdata(pdev);
|
2011-11-17 16:17:04 +01:00
|
|
|
struct uart_port *port = &sirfport->port;
|
|
|
|
uart_suspend_port(&sirfsoc_uart_drv, port);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2014-01-03 08:44:06 +01:00
|
|
|
static int sirfsoc_uart_resume(struct device *pdev)
|
2011-11-17 16:17:04 +01:00
|
|
|
{
|
2014-01-03 08:44:06 +01:00
|
|
|
struct sirfsoc_uart_port *sirfport = dev_get_drvdata(pdev);
|
2011-11-17 16:17:04 +01:00
|
|
|
struct uart_port *port = &sirfport->port;
|
|
|
|
uart_resume_port(&sirfsoc_uart_drv, port);
|
|
|
|
return 0;
|
|
|
|
}
|
2014-01-03 08:44:06 +01:00
|
|
|
#endif
|
|
|
|
|
|
|
|
static const struct dev_pm_ops sirfsoc_uart_pm_ops = {
|
|
|
|
SET_SYSTEM_SLEEP_PM_OPS(sirfsoc_uart_suspend, sirfsoc_uart_resume)
|
|
|
|
};
|
2011-11-17 16:17:04 +01:00
|
|
|
|
|
|
|
static struct platform_driver sirfsoc_uart_driver = {
|
|
|
|
.probe = sirfsoc_uart_probe,
|
2012-11-19 19:21:34 +01:00
|
|
|
.remove = sirfsoc_uart_remove,
|
2011-11-17 16:17:04 +01:00
|
|
|
.driver = {
|
|
|
|
.name = SIRFUART_PORT_NAME,
|
|
|
|
.of_match_table = sirfsoc_uart_ids,
|
2014-01-03 08:44:06 +01:00
|
|
|
.pm = &sirfsoc_uart_pm_ops,
|
2011-11-17 16:17:04 +01:00
|
|
|
},
|
|
|
|
};
|
|
|
|
|
|
|
|
static int __init sirfsoc_uart_init(void)
|
|
|
|
{
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
ret = uart_register_driver(&sirfsoc_uart_drv);
|
|
|
|
if (ret)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
ret = platform_driver_register(&sirfsoc_uart_driver);
|
|
|
|
if (ret)
|
|
|
|
uart_unregister_driver(&sirfsoc_uart_drv);
|
|
|
|
out:
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
module_init(sirfsoc_uart_init);
|
|
|
|
|
|
|
|
static void __exit sirfsoc_uart_exit(void)
|
|
|
|
{
|
|
|
|
platform_driver_unregister(&sirfsoc_uart_driver);
|
|
|
|
uart_unregister_driver(&sirfsoc_uart_drv);
|
|
|
|
}
|
|
|
|
module_exit(sirfsoc_uart_exit);
|
|
|
|
|
|
|
|
MODULE_LICENSE("GPL v2");
|
|
|
|
MODULE_AUTHOR("Bin Shi <Bin.Shi@csr.com>, Rong Wang<Rong.Wang@csr.com>");
|
|
|
|
MODULE_DESCRIPTION("CSR SiRFprimaII Uart Driver");
|