dect
/
linux-2.6
Archived
13
0
Fork 0
This repository has been archived on 2022-02-17. You can view files and clone it, but cannot push or open issues or pull requests.
linux-2.6/drivers/char/moxa.c

2587 lines
66 KiB
C
Raw Normal View History

/*****************************************************************************/
/*
* moxa.c -- MOXA Intellio family multiport serial driver.
*
* Copyright (C) 1999-2000 Moxa Technologies (support@moxa.com.tw).
*
* This code is loosely based on the Linux serial driver, written by
* Linus Torvalds, Theodore T'so and others.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*/
/*
* MOXA Intellio Series Driver
* for : LINUX
* date : 1999/1/7
* version : 5.1
*/
#include <linux/module.h>
#include <linux/types.h>
#include <linux/mm.h>
#include <linux/ioport.h>
#include <linux/errno.h>
#include <linux/firmware.h>
#include <linux/signal.h>
#include <linux/sched.h>
#include <linux/timer.h>
#include <linux/interrupt.h>
#include <linux/tty.h>
#include <linux/tty_flip.h>
#include <linux/major.h>
#include <linux/string.h>
#include <linux/fcntl.h>
#include <linux/ptrace.h>
#include <linux/serial.h>
#include <linux/tty_driver.h>
#include <linux/delay.h>
#include <linux/pci.h>
#include <linux/init.h>
#include <linux/bitops.h>
#include <linux/completion.h>
#include <asm/system.h>
#include <asm/io.h>
#include <asm/uaccess.h>
#include "moxa.h"
#define MOXA_VERSION "5.1k"
#define MOXA_FW_HDRLEN 32
#define MOXAMAJOR 172
#define MOXACUMAJOR 173
#define MAX_BOARDS 4 /* Don't change this value */
#define MAX_PORTS_PER_BOARD 32 /* Don't change this value */
#define MAX_PORTS (MAX_BOARDS * MAX_PORTS_PER_BOARD)
/*
* Define the Moxa PCI vendor and device IDs.
*/
#define MOXA_BUS_TYPE_ISA 0
#define MOXA_BUS_TYPE_PCI 1
enum {
MOXA_BOARD_C218_PCI = 1,
MOXA_BOARD_C218_ISA,
MOXA_BOARD_C320_PCI,
MOXA_BOARD_C320_ISA,
MOXA_BOARD_CP204J,
};
static char *moxa_brdname[] =
{
"C218 Turbo PCI series",
"C218 Turbo ISA series",
"C320 Turbo PCI series",
"C320 Turbo ISA series",
"CP-204J series",
};
#ifdef CONFIG_PCI
static struct pci_device_id moxa_pcibrds[] = {
{ PCI_DEVICE(PCI_VENDOR_ID_MOXA, PCI_DEVICE_ID_MOXA_C218),
.driver_data = MOXA_BOARD_C218_PCI },
{ PCI_DEVICE(PCI_VENDOR_ID_MOXA, PCI_DEVICE_ID_MOXA_C320),
.driver_data = MOXA_BOARD_C320_PCI },
{ PCI_DEVICE(PCI_VENDOR_ID_MOXA, PCI_DEVICE_ID_MOXA_CP204J),
.driver_data = MOXA_BOARD_CP204J },
{ 0 }
};
MODULE_DEVICE_TABLE(pci, moxa_pcibrds);
#endif /* CONFIG_PCI */
struct moxa_port;
static struct moxa_board_conf {
int boardType;
int numPorts;
int busType;
int loadstat;
struct moxa_port *ports;
void __iomem *basemem;
void __iomem *intNdx;
void __iomem *intPend;
void __iomem *intTable;
} moxa_boards[MAX_BOARDS];
struct mxser_mstatus {
tcflag_t cflag;
int cts;
int dsr;
int ri;
int dcd;
};
struct moxaq_str {
int inq;
int outq;
};
struct moxa_port {
int type;
int port;
int close_delay;
unsigned short closing_wait;
int count;
int blocked_open;
long event; /* long req'd for set_bit --RR */
int asyncflags;
unsigned long statusflags;
struct tty_struct *tty;
int cflag;
wait_queue_head_t open_wait;
struct completion close_wait;
struct timer_list emptyTimer;
char chkPort;
char lineCtrl;
void __iomem *tableAddr;
long curBaud;
char DCDState;
char lowChkFlag;
ushort breakCnt;
};
/* statusflags */
#define TXSTOPPED 0x1
#define LOWWAIT 0x2
#define EMPTYWAIT 0x4
#define THROTTLE 0x8
#define SERIAL_DO_RESTART
#define WAKEUP_CHARS 256
static int ttymajor = MOXAMAJOR;
static int moxaCard;
/* Variables for insmod */
#ifdef MODULE
static unsigned long baseaddr[MAX_BOARDS];
static unsigned int type[MAX_BOARDS];
static unsigned int numports[MAX_BOARDS];
#endif
MODULE_AUTHOR("William Chen");
MODULE_DESCRIPTION("MOXA Intellio Family Multiport Board Device Driver");
MODULE_LICENSE("GPL");
#ifdef MODULE
module_param_array(type, uint, NULL, 0);
MODULE_PARM_DESC(type, "card type: C218=2, C320=4");
module_param_array(baseaddr, ulong, NULL, 0);
MODULE_PARM_DESC(baseaddr, "base address");
module_param_array(numports, uint, NULL, 0);
MODULE_PARM_DESC(numports, "numports (ignored for C218)");
#endif
module_param(ttymajor, int, 0);
/*
* static functions:
*/
static int moxa_open(struct tty_struct *, struct file *);
static void moxa_close(struct tty_struct *, struct file *);
static int moxa_write(struct tty_struct *, const unsigned char *, int);
static int moxa_write_room(struct tty_struct *);
static void moxa_flush_buffer(struct tty_struct *);
static int moxa_chars_in_buffer(struct tty_struct *);
static void moxa_flush_chars(struct tty_struct *);
static void moxa_put_char(struct tty_struct *, unsigned char);
static int moxa_ioctl(struct tty_struct *, struct file *, unsigned int, unsigned long);
static void moxa_throttle(struct tty_struct *);
static void moxa_unthrottle(struct tty_struct *);
static void moxa_set_termios(struct tty_struct *, struct ktermios *);
static void moxa_stop(struct tty_struct *);
static void moxa_start(struct tty_struct *);
static void moxa_hangup(struct tty_struct *);
static int moxa_tiocmget(struct tty_struct *tty, struct file *file);
static int moxa_tiocmset(struct tty_struct *tty, struct file *file,
unsigned int set, unsigned int clear);
static void moxa_poll(unsigned long);
static void moxa_set_tty_param(struct tty_struct *, struct ktermios *);
static int moxa_block_till_ready(struct tty_struct *, struct file *,
struct moxa_port *);
static void moxa_setup_empty_event(struct tty_struct *);
static void moxa_check_xmit_empty(unsigned long);
static void moxa_shut_down(struct moxa_port *);
static void moxa_receive_data(struct moxa_port *);
/*
* moxa board interface functions:
*/
static int MoxaDriverIoctl(unsigned int, unsigned long, int);
static int MoxaDriverPoll(void);
static int MoxaPortsOfCard(int);
static int MoxaPortIsValid(int);
static void MoxaPortEnable(int);
static void MoxaPortDisable(int);
static long MoxaPortGetMaxBaud(int);
static long MoxaPortSetBaud(int, long);
static int MoxaPortSetTermio(int, struct ktermios *, speed_t);
static int MoxaPortGetLineOut(int, int *, int *);
static void MoxaPortLineCtrl(int, int, int);
static void MoxaPortFlowCtrl(int, int, int, int, int, int);
static int MoxaPortLineStatus(int);
static int MoxaPortDCDChange(int);
static int MoxaPortDCDON(int);
static void MoxaPortFlushData(int, int);
static int MoxaPortWriteData(int, unsigned char *, int);
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
static int MoxaPortReadData(int, struct tty_struct *tty);
static int MoxaPortTxQueue(int);
static int MoxaPortRxQueue(int);
static int MoxaPortTxFree(int);
static void MoxaPortTxDisable(int);
static void MoxaPortTxEnable(int);
static int MoxaPortResetBrkCnt(int);
static void MoxaPortSendBreak(int, int);
static int moxa_get_serial_info(struct moxa_port *, struct serial_struct __user *);
static int moxa_set_serial_info(struct moxa_port *, struct serial_struct __user *);
static void MoxaSetFifo(int port, int enable);
static const struct tty_operations moxa_ops = {
.open = moxa_open,
.close = moxa_close,
.write = moxa_write,
.write_room = moxa_write_room,
.flush_buffer = moxa_flush_buffer,
.chars_in_buffer = moxa_chars_in_buffer,
.flush_chars = moxa_flush_chars,
.put_char = moxa_put_char,
.ioctl = moxa_ioctl,
.throttle = moxa_throttle,
.unthrottle = moxa_unthrottle,
.set_termios = moxa_set_termios,
.stop = moxa_stop,
.start = moxa_start,
.hangup = moxa_hangup,
.tiocmget = moxa_tiocmget,
.tiocmset = moxa_tiocmset,
};
static struct tty_driver *moxaDriver;
static struct moxa_port moxa_ports[MAX_PORTS];
static DEFINE_TIMER(moxaTimer, moxa_poll, 0, 0);
static DEFINE_SPINLOCK(moxa_lock);
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
static int moxa_check_fw_model(struct moxa_board_conf *brd, u8 model)
{
switch (brd->boardType) {
case MOXA_BOARD_C218_ISA:
case MOXA_BOARD_C218_PCI:
if (model != 1)
goto err;
break;
case MOXA_BOARD_CP204J:
if (model != 3)
goto err;
break;
default:
if (model != 2)
goto err;
break;
}
return 0;
err:
return -EINVAL;
}
static int moxa_check_fw(const void *ptr)
{
const __le16 *lptr = ptr;
if (*lptr != cpu_to_le16(0x7980))
return -EINVAL;
return 0;
}
static int moxa_load_bios(struct moxa_board_conf *brd, const u8 *buf,
size_t len)
{
void __iomem *baseAddr = brd->basemem;
u16 tmp;
writeb(HW_reset, baseAddr + Control_reg); /* reset */
msleep(10);
memset_io(baseAddr, 0, 4096);
memcpy_toio(baseAddr, buf, len); /* download BIOS */
writeb(0, baseAddr + Control_reg); /* restart */
msleep(2000);
switch (brd->boardType) {
case MOXA_BOARD_C218_ISA:
case MOXA_BOARD_C218_PCI:
tmp = readw(baseAddr + C218_key);
if (tmp != C218_KeyCode)
goto err;
break;
case MOXA_BOARD_CP204J:
tmp = readw(baseAddr + C218_key);
if (tmp != CP204J_KeyCode)
goto err;
break;
default:
tmp = readw(baseAddr + C320_key);
if (tmp != C320_KeyCode)
goto err;
tmp = readw(baseAddr + C320_status);
if (tmp != STS_init) {
printk(KERN_ERR "moxa: bios upload failed -- CPU/Basic "
"module not found\n");
return -EIO;
}
break;
}
return 0;
err:
printk(KERN_ERR "moxa: bios upload failed -- board not found\n");
return -EIO;
}
static int moxa_load_320b(struct moxa_board_conf *brd, const u8 *ptr,
size_t len)
{
void __iomem *baseAddr = brd->basemem;
if (len < 7168) {
printk(KERN_ERR "moxa: invalid 320 bios -- too short\n");
return -EINVAL;
}
writew(len - 7168 - 2, baseAddr + C320bapi_len);
writeb(1, baseAddr + Control_reg); /* Select Page 1 */
memcpy_toio(baseAddr + DynPage_addr, ptr, 7168);
writeb(2, baseAddr + Control_reg); /* Select Page 2 */
memcpy_toio(baseAddr + DynPage_addr, ptr + 7168, len - 7168);
return 0;
}
static int moxa_real_load_code(struct moxa_board_conf *brd, const void *ptr,
size_t len)
{
void __iomem *baseAddr = brd->basemem;
const u16 *uptr = ptr;
size_t wlen, len2, j;
unsigned long key, loadbuf, loadlen, checksum, checksum_ok;
unsigned int i, retry, c320;
u16 usum, keycode;
c320 = brd->boardType == MOXA_BOARD_C320_PCI ||
brd->boardType == MOXA_BOARD_C320_ISA;
keycode = (brd->boardType == MOXA_BOARD_CP204J) ? CP204J_KeyCode :
C218_KeyCode;
switch (brd->boardType) {
case MOXA_BOARD_CP204J:
case MOXA_BOARD_C218_ISA:
case MOXA_BOARD_C218_PCI:
key = C218_key;
loadbuf = C218_LoadBuf;
loadlen = C218DLoad_len;
checksum = C218check_sum;
checksum_ok = C218chksum_ok;
break;
default:
key = C320_key;
keycode = C320_KeyCode;
loadbuf = C320_LoadBuf;
loadlen = C320DLoad_len;
checksum = C320check_sum;
checksum_ok = C320chksum_ok;
break;
}
usum = 0;
wlen = len >> 1;
for (i = 0; i < wlen; i++)
usum += le16_to_cpu(uptr[i]);
retry = 0;
do {
wlen = len >> 1;
j = 0;
while (wlen) {
len2 = (wlen > 2048) ? 2048 : wlen;
wlen -= len2;
memcpy_toio(baseAddr + loadbuf, ptr + j, len2 << 1);
j += len2 << 1;
writew(len2, baseAddr + loadlen);
writew(0, baseAddr + key);
for (i = 0; i < 100; i++) {
if (readw(baseAddr + key) == keycode)
break;
msleep(10);
}
if (readw(baseAddr + key) != keycode)
return -EIO;
}
writew(0, baseAddr + loadlen);
writew(usum, baseAddr + checksum);
writew(0, baseAddr + key);
for (i = 0; i < 100; i++) {
if (readw(baseAddr + key) == keycode)
break;
msleep(10);
}
retry++;
} while ((readb(baseAddr + checksum_ok) != 1) && (retry < 3));
if (readb(baseAddr + checksum_ok) != 1)
return -EIO;
writew(0, baseAddr + key);
for (i = 0; i < 600; i++) {
if (readw(baseAddr + Magic_no) == Magic_code)
break;
msleep(10);
}
if (readw(baseAddr + Magic_no) != Magic_code)
return -EIO;
if (c320) {
if (brd->busType == MOXA_BUS_TYPE_PCI) { /* ASIC board */
writew(0x3800, baseAddr + TMS320_PORT1);
writew(0x3900, baseAddr + TMS320_PORT2);
writew(28499, baseAddr + TMS320_CLOCK);
} else {
writew(0x3200, baseAddr + TMS320_PORT1);
writew(0x3400, baseAddr + TMS320_PORT2);
writew(19999, baseAddr + TMS320_CLOCK);
}
}
writew(1, baseAddr + Disable_IRQ);
writew(0, baseAddr + Magic_no);
for (i = 0; i < 500; i++) {
if (readw(baseAddr + Magic_no) == Magic_code)
break;
msleep(10);
}
if (readw(baseAddr + Magic_no) != Magic_code)
return -EIO;
if (c320) {
j = readw(baseAddr + Module_cnt);
if (j <= 0)
return -EIO;
brd->numPorts = j * 8;
writew(j, baseAddr + Module_no);
writew(0, baseAddr + Magic_no);
for (i = 0; i < 600; i++) {
if (readw(baseAddr + Magic_no) == Magic_code)
break;
msleep(10);
}
if (readw(baseAddr + Magic_no) != Magic_code)
return -EIO;
}
moxaCard = 1;
brd->intNdx = baseAddr + IRQindex;
brd->intPend = baseAddr + IRQpending;
brd->intTable = baseAddr + IRQtable;
return 0;
}
static int moxa_load_code(struct moxa_board_conf *brd, const void *ptr,
size_t len)
{
void __iomem *ofsAddr, *baseAddr = brd->basemem;
struct moxa_port *port;
int retval, i;
if (len % 2) {
printk(KERN_ERR "moxa: bios length is not even\n");
return -EINVAL;
}
retval = moxa_real_load_code(brd, ptr, len); /* may change numPorts */
if (retval)
return retval;
switch (brd->boardType) {
case MOXA_BOARD_C218_ISA:
case MOXA_BOARD_C218_PCI:
case MOXA_BOARD_CP204J:
port = brd->ports;
for (i = 0; i < brd->numPorts; i++, port++) {
port->chkPort = 1;
port->curBaud = 9600L;
port->DCDState = 0;
port->tableAddr = baseAddr + Extern_table +
Extern_size * i;
ofsAddr = port->tableAddr;
writew(C218rx_mask, ofsAddr + RX_mask);
writew(C218tx_mask, ofsAddr + TX_mask);
writew(C218rx_spage + i * C218buf_pageno, ofsAddr + Page_rxb);
writew(readw(ofsAddr + Page_rxb) + C218rx_pageno, ofsAddr + EndPage_rxb);
writew(C218tx_spage + i * C218buf_pageno, ofsAddr + Page_txb);
writew(readw(ofsAddr + Page_txb) + C218tx_pageno, ofsAddr + EndPage_txb);
}
break;
default:
port = brd->ports;
for (i = 0; i < brd->numPorts; i++, port++) {
port->chkPort = 1;
port->curBaud = 9600L;
port->DCDState = 0;
port->tableAddr = baseAddr + Extern_table +
Extern_size * i;
ofsAddr = port->tableAddr;
switch (brd->numPorts) {
case 8:
writew(C320p8rx_mask, ofsAddr + RX_mask);
writew(C320p8tx_mask, ofsAddr + TX_mask);
writew(C320p8rx_spage + i * C320p8buf_pgno, ofsAddr + Page_rxb);
writew(readw(ofsAddr + Page_rxb) + C320p8rx_pgno, ofsAddr + EndPage_rxb);
writew(C320p8tx_spage + i * C320p8buf_pgno, ofsAddr + Page_txb);
writew(readw(ofsAddr + Page_txb) + C320p8tx_pgno, ofsAddr + EndPage_txb);
break;
case 16:
writew(C320p16rx_mask, ofsAddr + RX_mask);
writew(C320p16tx_mask, ofsAddr + TX_mask);
writew(C320p16rx_spage + i * C320p16buf_pgno, ofsAddr + Page_rxb);
writew(readw(ofsAddr + Page_rxb) + C320p16rx_pgno, ofsAddr + EndPage_rxb);
writew(C320p16tx_spage + i * C320p16buf_pgno, ofsAddr + Page_txb);
writew(readw(ofsAddr + Page_txb) + C320p16tx_pgno, ofsAddr + EndPage_txb);
break;
case 24:
writew(C320p24rx_mask, ofsAddr + RX_mask);
writew(C320p24tx_mask, ofsAddr + TX_mask);
writew(C320p24rx_spage + i * C320p24buf_pgno, ofsAddr + Page_rxb);
writew(readw(ofsAddr + Page_rxb) + C320p24rx_pgno, ofsAddr + EndPage_rxb);
writew(C320p24tx_spage + i * C320p24buf_pgno, ofsAddr + Page_txb);
writew(readw(ofsAddr + Page_txb), ofsAddr + EndPage_txb);
break;
case 32:
writew(C320p32rx_mask, ofsAddr + RX_mask);
writew(C320p32tx_mask, ofsAddr + TX_mask);
writew(C320p32tx_ofs, ofsAddr + Ofs_txb);
writew(C320p32rx_spage + i * C320p32buf_pgno, ofsAddr + Page_rxb);
writew(readb(ofsAddr + Page_rxb), ofsAddr + EndPage_rxb);
writew(C320p32tx_spage + i * C320p32buf_pgno, ofsAddr + Page_txb);
writew(readw(ofsAddr + Page_txb), ofsAddr + EndPage_txb);
break;
}
}
break;
}
brd->loadstat = 1;
return 0;
}
static int moxa_load_fw(struct moxa_board_conf *brd, const struct firmware *fw)
{
void *ptr = fw->data;
char rsn[64];
u16 lens[5];
size_t len;
unsigned int a, lenp, lencnt;
int ret = -EINVAL;
struct {
__le32 magic; /* 0x34303430 */
u8 reserved1[2];
u8 type; /* UNIX = 3 */
u8 model; /* C218T=1, C320T=2, CP204=3 */
u8 reserved2[8];
__le16 len[5];
} *hdr = ptr;
BUILD_BUG_ON(ARRAY_SIZE(hdr->len) != ARRAY_SIZE(lens));
if (fw->size < MOXA_FW_HDRLEN) {
strcpy(rsn, "too short (even header won't fit)");
goto err;
}
if (hdr->magic != cpu_to_le32(0x30343034)) {
sprintf(rsn, "bad magic: %.8x", le32_to_cpu(hdr->magic));
goto err;
}
if (hdr->type != 3) {
sprintf(rsn, "not for linux, type is %u", hdr->type);
goto err;
}
if (moxa_check_fw_model(brd, hdr->model)) {
sprintf(rsn, "not for this card, model is %u", hdr->model);
goto err;
}
len = MOXA_FW_HDRLEN;
lencnt = hdr->model == 2 ? 5 : 3;
for (a = 0; a < ARRAY_SIZE(lens); a++) {
lens[a] = le16_to_cpu(hdr->len[a]);
if (lens[a] && len + lens[a] <= fw->size &&
moxa_check_fw(&fw->data[len]))
printk(KERN_WARNING "moxa firmware: unexpected input "
"at offset %u, but going on\n", (u32)len);
if (!lens[a] && a < lencnt) {
sprintf(rsn, "too few entries in fw file");
goto err;
}
len += lens[a];
}
if (len != fw->size) {
sprintf(rsn, "bad length: %u (should be %u)", (u32)fw->size,
(u32)len);
goto err;
}
ptr += MOXA_FW_HDRLEN;
lenp = 0; /* bios */
strcpy(rsn, "read above");
ret = moxa_load_bios(brd, ptr, lens[lenp]);
if (ret)
goto err;
/* we skip the tty section (lens[1]), since we don't need it */
ptr += lens[lenp] + lens[lenp + 1];
lenp += 2; /* comm */
if (hdr->model == 2) {
ret = moxa_load_320b(brd, ptr, lens[lenp]);
if (ret)
goto err;
/* skip another tty */
ptr += lens[lenp] + lens[lenp + 1];
lenp += 2;
}
ret = moxa_load_code(brd, ptr, lens[lenp]);
if (ret)
goto err;
return 0;
err:
printk(KERN_ERR "firmware failed to load, reason: %s\n", rsn);
return ret;
}
static int moxa_init_board(struct moxa_board_conf *brd, struct device *dev)
{
const struct firmware *fw;
const char *file;
int ret;
switch (brd->boardType) {
case MOXA_BOARD_C218_ISA:
case MOXA_BOARD_C218_PCI:
file = "c218tunx.cod";
break;
case MOXA_BOARD_CP204J:
file = "cp204unx.cod";
break;
default:
file = "c320tunx.cod";
break;
}
ret = request_firmware(&fw, file, dev);
if (ret) {
printk(KERN_ERR "request_firmware failed\n");
goto end;
}
ret = moxa_load_fw(brd, fw);
release_firmware(fw);
end:
return ret;
}
#ifdef CONFIG_PCI
static int __devinit moxa_pci_probe(struct pci_dev *pdev,
const struct pci_device_id *ent)
{
struct moxa_board_conf *board;
unsigned int i;
int board_type = ent->driver_data;
int retval;
retval = pci_enable_device(pdev);
if (retval) {
dev_err(&pdev->dev, "can't enable pci device\n");
goto err;
}
for (i = 0; i < MAX_BOARDS; i++)
if (moxa_boards[i].basemem == NULL)
break;
retval = -ENODEV;
if (i >= MAX_BOARDS) {
dev_warn(&pdev->dev, "more than %u MOXA Intellio family boards "
"found. Board is ignored.\n", MAX_BOARDS);
goto err;
}
board = &moxa_boards[i];
board->ports = &moxa_ports[i * MAX_PORTS_PER_BOARD];
retval = pci_request_region(pdev, 2, "moxa-base");
if (retval) {
dev_err(&pdev->dev, "can't request pci region 2\n");
goto err;
}
board->basemem = ioremap(pci_resource_start(pdev, 2), 0x4000);
if (board->basemem == NULL) {
dev_err(&pdev->dev, "can't remap io space 2\n");
goto err_reg;
}
board->boardType = board_type;
switch (board_type) {
case MOXA_BOARD_C218_ISA:
case MOXA_BOARD_C218_PCI:
board->numPorts = 8;
break;
case MOXA_BOARD_CP204J:
board->numPorts = 4;
break;
default:
board->numPorts = 0;
break;
}
board->busType = MOXA_BUS_TYPE_PCI;
retval = moxa_init_board(board, &pdev->dev);
if (retval)
goto err_base;
pci_set_drvdata(pdev, board);
return (0);
err_base:
iounmap(board->basemem);
board->basemem = NULL;
err_reg:
pci_release_region(pdev, 2);
err:
return retval;
}
static void __devexit moxa_pci_remove(struct pci_dev *pdev)
{
struct moxa_board_conf *brd = pci_get_drvdata(pdev);
iounmap(brd->basemem);
brd->basemem = NULL;
pci_release_region(pdev, 2);
}
static struct pci_driver moxa_pci_driver = {
.name = "moxa",
.id_table = moxa_pcibrds,
.probe = moxa_pci_probe,
.remove = __devexit_p(moxa_pci_remove)
};
#endif /* CONFIG_PCI */
static int __init moxa_init(void)
{
struct moxa_port *ch;
unsigned int i, isabrds = 0;
int retval = 0;
printk(KERN_INFO "MOXA Intellio family driver version %s\n",
MOXA_VERSION);
moxaDriver = alloc_tty_driver(MAX_PORTS + 1);
if (!moxaDriver)
return -ENOMEM;
moxaDriver->owner = THIS_MODULE;
moxaDriver->name = "ttyMX";
moxaDriver->major = ttymajor;
moxaDriver->minor_start = 0;
moxaDriver->type = TTY_DRIVER_TYPE_SERIAL;
moxaDriver->subtype = SERIAL_TYPE_NORMAL;
moxaDriver->init_termios = tty_std_termios;
moxaDriver->init_termios.c_cflag = B9600 | CS8 | CREAD | CLOCAL | HUPCL;
moxaDriver->init_termios.c_ispeed = 9600;
moxaDriver->init_termios.c_ospeed = 9600;
moxaDriver->flags = TTY_DRIVER_REAL_RAW;
tty_set_operations(moxaDriver, &moxa_ops);
for (i = 0, ch = moxa_ports; i < MAX_PORTS; i++, ch++) {
ch->type = PORT_16550A;
ch->port = i;
ch->close_delay = 5 * HZ / 10;
ch->closing_wait = 30 * HZ;
ch->cflag = B9600 | CS8 | CREAD | CLOCAL | HUPCL;
init_waitqueue_head(&ch->open_wait);
init_completion(&ch->close_wait);
setup_timer(&ch->emptyTimer, moxa_check_xmit_empty,
(unsigned long)ch);
}
pr_debug("Moxa tty devices major number = %d\n", ttymajor);
if (tty_register_driver(moxaDriver)) {
printk(KERN_ERR "Couldn't install MOXA Smartio family driver !\n");
put_tty_driver(moxaDriver);
return -1;
}
mod_timer(&moxaTimer, jiffies + HZ / 50);
/* Find the boards defined from module args. */
#ifdef MODULE
{
struct moxa_board_conf *brd = moxa_boards;
for (i = 0; i < MAX_BOARDS; i++) {
if (!baseaddr[i])
break;
if (type[i] == MOXA_BOARD_C218_ISA ||
type[i] == MOXA_BOARD_C320_ISA) {
pr_debug("Moxa board %2d: %s board(baseAddr=%lx)\n",
isabrds + 1, moxa_brdname[type[i] - 1],
baseaddr[i]);
brd->boardType = type[i];
brd->ports = &moxa_ports[isabrds * MAX_PORTS_PER_BOARD];
brd->numPorts = type[i] == MOXA_BOARD_C218_ISA ? 8 :
numports[i];
brd->busType = MOXA_BUS_TYPE_ISA;
brd->basemem = ioremap(baseaddr[i], 0x4000);
if (!brd->basemem) {
printk(KERN_ERR "moxa: can't remap %lx\n",
baseaddr[i]);
continue;
}
if (moxa_init_board(brd, NULL)) {
iounmap(brd->basemem);
brd->basemem = NULL;
continue;
}
brd++;
isabrds++;
}
}
}
#endif
#ifdef CONFIG_PCI
retval = pci_register_driver(&moxa_pci_driver);
if (retval) {
printk(KERN_ERR "Can't register moxa pci driver!\n");
if (isabrds)
retval = 0;
}
#endif
return retval;
}
static void __exit moxa_exit(void)
{
int i;
del_timer_sync(&moxaTimer);
for (i = 0; i < MAX_PORTS; i++)
del_timer_sync(&moxa_ports[i].emptyTimer);
if (tty_unregister_driver(moxaDriver))
printk(KERN_ERR "Couldn't unregister MOXA Intellio family "
"serial driver\n");
put_tty_driver(moxaDriver);
#ifdef CONFIG_PCI
pci_unregister_driver(&moxa_pci_driver);
#endif
for (i = 0; i < MAX_BOARDS; i++)
if (moxa_boards[i].basemem)
iounmap(moxa_boards[i].basemem);
}
module_init(moxa_init);
module_exit(moxa_exit);
static int moxa_open(struct tty_struct *tty, struct file *filp)
{
struct moxa_port *ch;
int port;
int retval;
port = tty->index;
if (port == MAX_PORTS) {
return (0);
}
if (!MoxaPortIsValid(port)) {
tty->driver_data = NULL;
return (-ENODEV);
}
ch = &moxa_ports[port];
ch->count++;
tty->driver_data = ch;
ch->tty = tty;
if (!(ch->asyncflags & ASYNC_INITIALIZED)) {
ch->statusflags = 0;
moxa_set_tty_param(tty, tty->termios);
MoxaPortLineCtrl(ch->port, 1, 1);
MoxaPortEnable(ch->port);
ch->asyncflags |= ASYNC_INITIALIZED;
}
retval = moxa_block_till_ready(tty, filp, ch);
moxa_unthrottle(tty);
if (ch->type == PORT_16550A) {
MoxaSetFifo(ch->port, 1);
} else {
MoxaSetFifo(ch->port, 0);
}
return (retval);
}
static void moxa_close(struct tty_struct *tty, struct file *filp)
{
struct moxa_port *ch;
int port;
port = tty->index;
if (port == MAX_PORTS) {
return;
}
if (!MoxaPortIsValid(port)) {
pr_debug("Invalid portno in moxa_close\n");
tty->driver_data = NULL;
return;
}
if (tty->driver_data == NULL) {
return;
}
if (tty_hung_up_p(filp)) {
return;
}
ch = (struct moxa_port *) tty->driver_data;
if ((tty->count == 1) && (ch->count != 1)) {
printk(KERN_WARNING "moxa_close: bad serial port count; "
"tty->count is 1, ch->count is %d\n", ch->count);
ch->count = 1;
}
if (--ch->count < 0) {
printk(KERN_WARNING "moxa_close: bad serial port count, "
"device=%s\n", tty->name);
ch->count = 0;
}
if (ch->count) {
return;
}
ch->asyncflags |= ASYNC_CLOSING;
ch->cflag = tty->termios->c_cflag;
if (ch->asyncflags & ASYNC_INITIALIZED) {
moxa_setup_empty_event(tty);
tty_wait_until_sent(tty, 30 * HZ); /* 30 seconds timeout */
del_timer_sync(&moxa_ports[ch->port].emptyTimer);
}
moxa_shut_down(ch);
MoxaPortFlushData(port, 2);
if (tty->driver->flush_buffer)
tty->driver->flush_buffer(tty);
tty_ldisc_flush(tty);
tty->closing = 0;
ch->event = 0;
ch->tty = NULL;
if (ch->blocked_open) {
if (ch->close_delay) {
msleep_interruptible(jiffies_to_msecs(ch->close_delay));
}
wake_up_interruptible(&ch->open_wait);
}
ch->asyncflags &= ~(ASYNC_NORMAL_ACTIVE | ASYNC_CLOSING);
complete_all(&ch->close_wait);
}
static int moxa_write(struct tty_struct *tty,
const unsigned char *buf, int count)
{
struct moxa_port *ch;
int len, port;
unsigned long flags;
ch = (struct moxa_port *) tty->driver_data;
if (ch == NULL)
return (0);
port = ch->port;
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
spin_lock_irqsave(&moxa_lock, flags);
len = MoxaPortWriteData(port, (unsigned char *) buf, count);
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
spin_unlock_irqrestore(&moxa_lock, flags);
/*********************************************
if ( !(ch->statusflags & LOWWAIT) &&
((len != count) || (MoxaPortTxFree(port) <= 100)) )
************************************************/
ch->statusflags |= LOWWAIT;
return (len);
}
static int moxa_write_room(struct tty_struct *tty)
{
struct moxa_port *ch;
if (tty->stopped)
return (0);
ch = (struct moxa_port *) tty->driver_data;
if (ch == NULL)
return (0);
return (MoxaPortTxFree(ch->port));
}
static void moxa_flush_buffer(struct tty_struct *tty)
{
struct moxa_port *ch = (struct moxa_port *) tty->driver_data;
if (ch == NULL)
return;
MoxaPortFlushData(ch->port, 1);
tty_wakeup(tty);
}
static int moxa_chars_in_buffer(struct tty_struct *tty)
{
int chars;
struct moxa_port *ch = (struct moxa_port *) tty->driver_data;
/*
* Sigh...I have to check if driver_data is NULL here, because
* if an open() fails, the TTY subsystem eventually calls
* tty_wait_until_sent(), which calls the driver's chars_in_buffer()
* routine. And since the open() failed, we return 0 here. TDJ
*/
if (ch == NULL)
return (0);
chars = MoxaPortTxQueue(ch->port);
if (chars) {
/*
* Make it possible to wakeup anything waiting for output
* in tty_ioctl.c, etc.
*/
if (!(ch->statusflags & EMPTYWAIT))
moxa_setup_empty_event(tty);
}
return (chars);
}
static void moxa_flush_chars(struct tty_struct *tty)
{
/*
* Don't think I need this, because this is called to empty the TX
* buffer for the 16450, 16550, etc.
*/
}
static void moxa_put_char(struct tty_struct *tty, unsigned char c)
{
struct moxa_port *ch;
int port;
unsigned long flags;
ch = (struct moxa_port *) tty->driver_data;
if (ch == NULL)
return;
port = ch->port;
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
spin_lock_irqsave(&moxa_lock, flags);
MoxaPortWriteData(port, &c, 1);
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
spin_unlock_irqrestore(&moxa_lock, flags);
/************************************************
if ( !(ch->statusflags & LOWWAIT) && (MoxaPortTxFree(port) <= 100) )
*************************************************/
ch->statusflags |= LOWWAIT;
}
static int moxa_tiocmget(struct tty_struct *tty, struct file *file)
{
struct moxa_port *ch = (struct moxa_port *) tty->driver_data;
int port;
int flag = 0, dtr, rts;
port = tty->index;
if ((port != MAX_PORTS) && (!ch))
return (-EINVAL);
MoxaPortGetLineOut(ch->port, &dtr, &rts);
if (dtr)
flag |= TIOCM_DTR;
if (rts)
flag |= TIOCM_RTS;
dtr = MoxaPortLineStatus(ch->port);
if (dtr & 1)
flag |= TIOCM_CTS;
if (dtr & 2)
flag |= TIOCM_DSR;
if (dtr & 4)
flag |= TIOCM_CD;
return flag;
}
static int moxa_tiocmset(struct tty_struct *tty, struct file *file,
unsigned int set, unsigned int clear)
{
struct moxa_port *ch = (struct moxa_port *) tty->driver_data;
int port;
int dtr, rts;
port = tty->index;
if ((port != MAX_PORTS) && (!ch))
return (-EINVAL);
MoxaPortGetLineOut(ch->port, &dtr, &rts);
if (set & TIOCM_RTS)
rts = 1;
if (set & TIOCM_DTR)
dtr = 1;
if (clear & TIOCM_RTS)
rts = 0;
if (clear & TIOCM_DTR)
dtr = 0;
MoxaPortLineCtrl(ch->port, dtr, rts);
return 0;
}
static int moxa_ioctl(struct tty_struct *tty, struct file *file,
unsigned int cmd, unsigned long arg)
{
struct moxa_port *ch = (struct moxa_port *) tty->driver_data;
register int port;
void __user *argp = (void __user *)arg;
int retval;
port = tty->index;
if ((port != MAX_PORTS) && (!ch))
return (-EINVAL);
switch (cmd) {
case TCSBRK: /* SVID version: non-zero arg --> no break */
retval = tty_check_change(tty);
if (retval)
return (retval);
moxa_setup_empty_event(tty);
tty_wait_until_sent(tty, 0);
if (!arg)
MoxaPortSendBreak(ch->port, 0);
return (0);
case TCSBRKP: /* support for POSIX tcsendbreak() */
retval = tty_check_change(tty);
if (retval)
return (retval);
moxa_setup_empty_event(tty);
tty_wait_until_sent(tty, 0);
MoxaPortSendBreak(ch->port, arg);
return (0);
case TIOCGSOFTCAR:
return put_user(C_CLOCAL(tty) ? 1 : 0, (int __user *)argp);
case TIOCSSOFTCAR:
if (get_user(retval, (int __user *)argp))
return -EFAULT;
arg = retval;
tty->termios->c_cflag = ((tty->termios->c_cflag & ~CLOCAL) |
(arg ? CLOCAL : 0));
if (C_CLOCAL(tty))
ch->asyncflags &= ~ASYNC_CHECK_CD;
else
ch->asyncflags |= ASYNC_CHECK_CD;
return (0);
case TIOCGSERIAL:
return moxa_get_serial_info(ch, argp);
case TIOCSSERIAL:
return moxa_set_serial_info(ch, argp);
default:
retval = MoxaDriverIoctl(cmd, arg, port);
}
return (retval);
}
static void moxa_throttle(struct tty_struct *tty)
{
struct moxa_port *ch = (struct moxa_port *) tty->driver_data;
ch->statusflags |= THROTTLE;
}
static void moxa_unthrottle(struct tty_struct *tty)
{
struct moxa_port *ch = (struct moxa_port *) tty->driver_data;
ch->statusflags &= ~THROTTLE;
}
static void moxa_set_termios(struct tty_struct *tty,
struct ktermios *old_termios)
{
struct moxa_port *ch = (struct moxa_port *) tty->driver_data;
if (ch == NULL)
return;
moxa_set_tty_param(tty, old_termios);
if (!(old_termios->c_cflag & CLOCAL) &&
(tty->termios->c_cflag & CLOCAL))
wake_up_interruptible(&ch->open_wait);
}
static void moxa_stop(struct tty_struct *tty)
{
struct moxa_port *ch = (struct moxa_port *) tty->driver_data;
if (ch == NULL)
return;
MoxaPortTxDisable(ch->port);
ch->statusflags |= TXSTOPPED;
}
static void moxa_start(struct tty_struct *tty)
{
struct moxa_port *ch = (struct moxa_port *) tty->driver_data;
if (ch == NULL)
return;
if (!(ch->statusflags & TXSTOPPED))
return;
MoxaPortTxEnable(ch->port);
ch->statusflags &= ~TXSTOPPED;
}
static void moxa_hangup(struct tty_struct *tty)
{
struct moxa_port *ch = (struct moxa_port *) tty->driver_data;
moxa_flush_buffer(tty);
moxa_shut_down(ch);
ch->event = 0;
ch->count = 0;
ch->asyncflags &= ~ASYNC_NORMAL_ACTIVE;
ch->tty = NULL;
wake_up_interruptible(&ch->open_wait);
}
static void moxa_poll(unsigned long ignored)
{
register int card;
struct moxa_port *ch;
struct tty_struct *tp;
int i, ports;
del_timer(&moxaTimer);
if (MoxaDriverPoll() < 0) {
mod_timer(&moxaTimer, jiffies + HZ / 50);
return;
}
for (card = 0; card < MAX_BOARDS; card++) {
if ((ports = MoxaPortsOfCard(card)) <= 0)
continue;
ch = &moxa_ports[card * MAX_PORTS_PER_BOARD];
for (i = 0; i < ports; i++, ch++) {
if ((ch->asyncflags & ASYNC_INITIALIZED) == 0)
continue;
if (!(ch->statusflags & THROTTLE) &&
(MoxaPortRxQueue(ch->port) > 0))
moxa_receive_data(ch);
if ((tp = ch->tty) == 0)
continue;
if (ch->statusflags & LOWWAIT) {
if (MoxaPortTxQueue(ch->port) <= WAKEUP_CHARS) {
if (!tp->stopped) {
ch->statusflags &= ~LOWWAIT;
tty_wakeup(tp);
}
}
}
if (!I_IGNBRK(tp) && (MoxaPortResetBrkCnt(ch->port) > 0)) {
tty_insert_flip_char(tp, 0, TTY_BREAK);
tty_schedule_flip(tp);
}
if (MoxaPortDCDChange(ch->port)) {
if (ch->asyncflags & ASYNC_CHECK_CD) {
if (MoxaPortDCDON(ch->port))
wake_up_interruptible(&ch->open_wait);
else {
tty_hangup(tp);
wake_up_interruptible(&ch->open_wait);
ch->asyncflags &= ~ASYNC_NORMAL_ACTIVE;
}
}
}
}
}
mod_timer(&moxaTimer, jiffies + HZ / 50);
}
/******************************************************************************/
static void moxa_set_tty_param(struct tty_struct *tty, struct ktermios *old_termios)
{
register struct ktermios *ts;
struct moxa_port *ch;
int rts, cts, txflow, rxflow, xany, baud;
ch = (struct moxa_port *) tty->driver_data;
ts = tty->termios;
if (ts->c_cflag & CLOCAL)
ch->asyncflags &= ~ASYNC_CHECK_CD;
else
ch->asyncflags |= ASYNC_CHECK_CD;
rts = cts = txflow = rxflow = xany = 0;
if (ts->c_cflag & CRTSCTS)
rts = cts = 1;
if (ts->c_iflag & IXON)
txflow = 1;
if (ts->c_iflag & IXOFF)
rxflow = 1;
if (ts->c_iflag & IXANY)
xany = 1;
/* Clear the features we don't support */
ts->c_cflag &= ~CMSPAR;
MoxaPortFlowCtrl(ch->port, rts, cts, txflow, rxflow, xany);
baud = MoxaPortSetTermio(ch->port, ts, tty_get_baud_rate(tty));
if (baud == -1)
baud = tty_termios_baud_rate(old_termios);
/* Not put the baud rate into the termios data */
tty_encode_baud_rate(tty, baud, baud);
}
static int moxa_block_till_ready(struct tty_struct *tty, struct file *filp,
struct moxa_port *ch)
{
DECLARE_WAITQUEUE(wait,current);
unsigned long flags;
int retval;
int do_clocal = C_CLOCAL(tty);
/*
* If the device is in the middle of being closed, then block
* until it's done, and then try again.
*/
if (tty_hung_up_p(filp) || (ch->asyncflags & ASYNC_CLOSING)) {
if (ch->asyncflags & ASYNC_CLOSING)
wait_for_completion_interruptible(&ch->close_wait);
#ifdef SERIAL_DO_RESTART
if (ch->asyncflags & ASYNC_HUP_NOTIFY)
return (-EAGAIN);
else
return (-ERESTARTSYS);
#else
return (-EAGAIN);
#endif
}
/*
* If non-blocking mode is set, then make the check up front
* and then exit.
*/
if (filp->f_flags & O_NONBLOCK) {
ch->asyncflags |= ASYNC_NORMAL_ACTIVE;
return (0);
}
/*
* Block waiting for the carrier detect and the line to become free
*/
retval = 0;
add_wait_queue(&ch->open_wait, &wait);
pr_debug("block_til_ready before block: ttys%d, count = %d\n",
ch->port, ch->count);
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
spin_lock_irqsave(&moxa_lock, flags);
if (!tty_hung_up_p(filp))
ch->count--;
ch->blocked_open++;
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
spin_unlock_irqrestore(&moxa_lock, flags);
while (1) {
set_current_state(TASK_INTERRUPTIBLE);
if (tty_hung_up_p(filp) ||
!(ch->asyncflags & ASYNC_INITIALIZED)) {
#ifdef SERIAL_DO_RESTART
if (ch->asyncflags & ASYNC_HUP_NOTIFY)
retval = -EAGAIN;
else
retval = -ERESTARTSYS;
#else
retval = -EAGAIN;
#endif
break;
}
if (!(ch->asyncflags & ASYNC_CLOSING) && (do_clocal ||
MoxaPortDCDON(ch->port)))
break;
if (signal_pending(current)) {
retval = -ERESTARTSYS;
break;
}
schedule();
}
set_current_state(TASK_RUNNING);
remove_wait_queue(&ch->open_wait, &wait);
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
spin_lock_irqsave(&moxa_lock, flags);
if (!tty_hung_up_p(filp))
ch->count++;
ch->blocked_open--;
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
spin_unlock_irqrestore(&moxa_lock, flags);
pr_debug("block_til_ready after blocking: ttys%d, count = %d\n",
ch->port, ch->count);
if (retval)
return (retval);
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
/* FIXME: review to see if we need to use set_bit on these */
ch->asyncflags |= ASYNC_NORMAL_ACTIVE;
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
return 0;
}
static void moxa_setup_empty_event(struct tty_struct *tty)
{
struct moxa_port *ch = tty->driver_data;
unsigned long flags;
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
spin_lock_irqsave(&moxa_lock, flags);
ch->statusflags |= EMPTYWAIT;
mod_timer(&moxa_ports[ch->port].emptyTimer, jiffies + HZ);
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
spin_unlock_irqrestore(&moxa_lock, flags);
}
static void moxa_check_xmit_empty(unsigned long data)
{
struct moxa_port *ch;
ch = (struct moxa_port *) data;
if (ch->tty && (ch->statusflags & EMPTYWAIT)) {
if (MoxaPortTxQueue(ch->port) == 0) {
ch->statusflags &= ~EMPTYWAIT;
tty_wakeup(ch->tty);
return;
}
mod_timer(&moxa_ports[ch->port].emptyTimer,
round_jiffies(jiffies + HZ));
} else
ch->statusflags &= ~EMPTYWAIT;
}
static void moxa_shut_down(struct moxa_port *ch)
{
struct tty_struct *tp;
if (!(ch->asyncflags & ASYNC_INITIALIZED))
return;
tp = ch->tty;
MoxaPortDisable(ch->port);
/*
* If we're a modem control device and HUPCL is on, drop RTS & DTR.
*/
if (tp->termios->c_cflag & HUPCL)
MoxaPortLineCtrl(ch->port, 0, 0);
ch->asyncflags &= ~ASYNC_INITIALIZED;
}
static void moxa_receive_data(struct moxa_port *ch)
{
struct tty_struct *tp;
struct ktermios *ts;
unsigned long flags;
ts = NULL;
tp = ch->tty;
if (tp)
ts = tp->termios;
/**************************************************
if ( !tp || !ts || !(ts->c_cflag & CREAD) ) {
*****************************************************/
if (!tp || !ts) {
MoxaPortFlushData(ch->port, 0);
return;
}
spin_lock_irqsave(&moxa_lock, flags);
MoxaPortReadData(ch->port, tp);
spin_unlock_irqrestore(&moxa_lock, flags);
tty_schedule_flip(tp);
}
/*
* Query
*/
struct mon_str {
int tick;
int rxcnt[MAX_PORTS];
int txcnt[MAX_PORTS];
};
#define DCD_changed 0x01
#define DCD_oldstate 0x80
static int moxaLowWaterChk;
static struct mon_str moxaLog;
static int moxaFuncTout = HZ / 2;
static void moxafunc(void __iomem *, int, ushort);
static void moxa_wait_finish(void __iomem *);
static void moxa_low_water_check(void __iomem *);
/*****************************************************************************
* Driver level functions: *
* 2. MoxaDriverIoctl(unsigned int cmd, unsigned long arg, int port); *
* 3. MoxaDriverPoll(void); *
*****************************************************************************/
#define MOXA 0x400
#define MOXA_GET_IQUEUE (MOXA + 1) /* get input buffered count */
#define MOXA_GET_OQUEUE (MOXA + 2) /* get output buffered count */
#define MOXA_GETDATACOUNT (MOXA + 23)
#define MOXA_GET_IOQUEUE (MOXA + 27)
#define MOXA_FLUSH_QUEUE (MOXA + 28)
#define MOXA_GET_CONF (MOXA + 35) /* configuration */
#define MOXA_GET_MAJOR (MOXA + 63)
#define MOXA_GET_CUMAJOR (MOXA + 64)
#define MOXA_GETMSTATUS (MOXA + 65)
void MoxaPortFlushData(int port, int mode)
{
void __iomem *ofsAddr;
if ((mode < 0) || (mode > 2))
return;
ofsAddr = moxa_ports[port].tableAddr;
moxafunc(ofsAddr, FC_FlushQueue, mode);
if (mode != 1) {
moxa_ports[port].lowChkFlag = 0;
moxa_low_water_check(ofsAddr);
}
}
int MoxaDriverIoctl(unsigned int cmd, unsigned long arg, int port)
{
int i;
int status;
int MoxaPortTxQueue(int), MoxaPortRxQueue(int);
void __user *argp = (void __user *)arg;
if (port == MAX_PORTS) {
if ((cmd != MOXA_GET_CONF) && (cmd != MOXA_GETDATACOUNT) &&
(cmd != MOXA_GET_IOQUEUE) && (cmd != MOXA_GET_MAJOR) &&
(cmd != MOXA_GET_CUMAJOR) && (cmd != MOXA_GETMSTATUS))
return (-EINVAL);
}
switch (cmd) {
case MOXA_GETDATACOUNT:
moxaLog.tick = jiffies;
if(copy_to_user(argp, &moxaLog, sizeof(struct mon_str)))
return -EFAULT;
return (0);
case MOXA_FLUSH_QUEUE:
MoxaPortFlushData(port, arg);
return (0);
case MOXA_GET_IOQUEUE: {
struct moxaq_str __user *argm = argp;
struct moxaq_str tmp;
for (i = 0; i < MAX_PORTS; i++, argm++) {
memset(&tmp, 0, sizeof(tmp));
if (moxa_ports[i].chkPort) {
tmp.inq = MoxaPortRxQueue(i);
tmp.outq = MoxaPortTxQueue(i);
}
if (copy_to_user(argm, &tmp, sizeof(tmp)))
return -EFAULT;
}
return (0);
} case MOXA_GET_OQUEUE:
i = MoxaPortTxQueue(port);
return put_user(i, (unsigned long __user *)argp);
case MOXA_GET_IQUEUE:
i = MoxaPortRxQueue(port);
return put_user(i, (unsigned long __user *)argp);
case MOXA_GET_MAJOR:
if(copy_to_user(argp, &ttymajor, sizeof(int)))
return -EFAULT;
return 0;
case MOXA_GET_CUMAJOR:
i = 0;
if(copy_to_user(argp, &i, sizeof(int)))
return -EFAULT;
return 0;
case MOXA_GETMSTATUS: {
struct mxser_mstatus __user *argm = argp;
struct mxser_mstatus tmp;
struct moxa_port *p;
for (i = 0; i < MAX_PORTS; i++, argm++) {
p = &moxa_ports[i];
memset(&tmp, 0, sizeof(tmp));
if (!p->chkPort) {
goto copy;
} else {
status = MoxaPortLineStatus(p->port);
if (status & 1)
tmp.cts = 1;
if (status & 2)
tmp.dsr = 1;
if (status & 4)
tmp.dcd = 1;
}
if (!p->tty || !p->tty->termios)
tmp.cflag = p->cflag;
else
tmp.cflag = p->tty->termios->c_cflag;
copy:
if (copy_to_user(argm, &tmp, sizeof(tmp)))
return -EFAULT;
}
return 0;
}
}
return -ENOIOCTLCMD;
}
int MoxaDriverPoll(void)
{
struct moxa_board_conf *brd;
register ushort temp;
register int card;
void __iomem *ofsAddr;
void __iomem *ip;
int port, p, ports;
if (moxaCard == 0)
return (-1);
for (card = 0; card < MAX_BOARDS; card++) {
brd = &moxa_boards[card];
if (brd->loadstat == 0)
continue;
if ((ports = brd->numPorts) == 0)
continue;
if (readb(brd->intPend) == 0xff) {
ip = brd->intTable + readb(brd->intNdx);
p = card * MAX_PORTS_PER_BOARD;
ports <<= 1;
for (port = 0; port < ports; port += 2, p++) {
if ((temp = readw(ip + port)) != 0) {
writew(0, ip + port);
ofsAddr = moxa_ports[p].tableAddr;
if (temp & IntrTx)
writew(readw(ofsAddr + HostStat) & ~WakeupTx, ofsAddr + HostStat);
if (temp & IntrBreak) {
moxa_ports[p].breakCnt++;
}
if (temp & IntrLine) {
if (readb(ofsAddr + FlagStat) & DCD_state) {
if ((moxa_ports[p].DCDState & DCD_oldstate) == 0)
moxa_ports[p].DCDState = (DCD_oldstate |
DCD_changed);
} else {
if (moxa_ports[p].DCDState & DCD_oldstate)
moxa_ports[p].DCDState = DCD_changed;
}
}
}
}
writeb(0, brd->intPend);
}
if (moxaLowWaterChk) {
p = card * MAX_PORTS_PER_BOARD;
for (port = 0; port < ports; port++, p++) {
if (moxa_ports[p].lowChkFlag) {
moxa_ports[p].lowChkFlag = 0;
ofsAddr = moxa_ports[p].tableAddr;
moxa_low_water_check(ofsAddr);
}
}
}
}
moxaLowWaterChk = 0;
return (0);
}
/*****************************************************************************
* Card level function: *
* 1. MoxaPortsOfCard(int cardno); *
*****************************************************************************/
int MoxaPortsOfCard(int cardno)
{
if (moxa_boards[cardno].boardType == 0)
return (0);
return (moxa_boards[cardno].numPorts);
}
/*****************************************************************************
* Port level functions: *
* 1. MoxaPortIsValid(int port); *
* 2. MoxaPortEnable(int port); *
* 3. MoxaPortDisable(int port); *
* 4. MoxaPortGetMaxBaud(int port); *
* 6. MoxaPortSetBaud(int port, long baud); *
* 8. MoxaPortSetTermio(int port, unsigned char *termio); *
* 9. MoxaPortGetLineOut(int port, int *dtrState, int *rtsState); *
* 10. MoxaPortLineCtrl(int port, int dtrState, int rtsState); *
* 11. MoxaPortFlowCtrl(int port, int rts, int cts, int rx, int tx,int xany); *
* 12. MoxaPortLineStatus(int port); *
* 13. MoxaPortDCDChange(int port); *
* 14. MoxaPortDCDON(int port); *
* 15. MoxaPortFlushData(int port, int mode); *
* 16. MoxaPortWriteData(int port, unsigned char * buffer, int length); *
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
* 17. MoxaPortReadData(int port, struct tty_struct *tty); *
* 20. MoxaPortTxQueue(int port); *
* 21. MoxaPortTxFree(int port); *
* 22. MoxaPortRxQueue(int port); *
* 24. MoxaPortTxDisable(int port); *
* 25. MoxaPortTxEnable(int port); *
* 27. MoxaPortResetBrkCnt(int port); *
* 30. MoxaPortSendBreak(int port, int ticks); *
*****************************************************************************/
/*
* Moxa Port Number Description:
*
* MOXA serial driver supports up to 4 MOXA-C218/C320 boards. And,
* the port number using in MOXA driver functions will be 0 to 31 for
* first MOXA board, 32 to 63 for second, 64 to 95 for third and 96
* to 127 for fourth. For example, if you setup three MOXA boards,
* first board is C218, second board is C320-16 and third board is
* C320-32. The port number of first board (C218 - 8 ports) is from
* 0 to 7. The port number of second board (C320 - 16 ports) is form
* 32 to 47. The port number of third board (C320 - 32 ports) is from
* 64 to 95. And those port numbers form 8 to 31, 48 to 63 and 96 to
* 127 will be invalid.
*
*
* Moxa Functions Description:
*
* Function 1: Driver initialization routine, this routine must be
* called when initialized driver.
* Syntax:
* void MoxaDriverInit();
*
*
* Function 2: Moxa driver private IOCTL command processing.
* Syntax:
* int MoxaDriverIoctl(unsigned int cmd, unsigned long arg, int port);
*
* unsigned int cmd : IOCTL command
* unsigned long arg : IOCTL argument
* int port : port number (0 - 127)
*
* return: 0 (OK)
* -EINVAL
* -ENOIOCTLCMD
*
*
* Function 3: Moxa driver polling process routine.
* Syntax:
* int MoxaDriverPoll(void);
*
* return: 0 ; polling O.K.
* -1 : no any Moxa card.
*
*
* Function 4: Get the ports of this card.
* Syntax:
* int MoxaPortsOfCard(int cardno);
*
* int cardno : card number (0 - 3)
*
* return: 0 : this card is invalid
* 8/16/24/32
*
*
* Function 5: Check this port is valid or invalid
* Syntax:
* int MoxaPortIsValid(int port);
* int port : port number (0 - 127, ref port description)
*
* return: 0 : this port is invalid
* 1 : this port is valid
*
*
* Function 6: Enable this port to start Tx/Rx data.
* Syntax:
* void MoxaPortEnable(int port);
* int port : port number (0 - 127)
*
*
* Function 7: Disable this port
* Syntax:
* void MoxaPortDisable(int port);
* int port : port number (0 - 127)
*
*
* Function 8: Get the maximun available baud rate of this port.
* Syntax:
* long MoxaPortGetMaxBaud(int port);
* int port : port number (0 - 127)
*
* return: 0 : this port is invalid
* 38400/57600/115200 bps
*
*
* Function 10: Setting baud rate of this port.
* Syntax:
* long MoxaPortSetBaud(int port, long baud);
* int port : port number (0 - 127)
* long baud : baud rate (50 - 115200)
*
* return: 0 : this port is invalid or baud < 50
* 50 - 115200 : the real baud rate set to the port, if
* the argument baud is large than maximun
* available baud rate, the real setting
* baud rate will be the maximun baud rate.
*
*
* Function 12: Configure the port.
* Syntax:
* int MoxaPortSetTermio(int port, struct ktermios *termio, speed_t baud);
* int port : port number (0 - 127)
* struct ktermios * termio : termio structure pointer
* speed_t baud : baud rate
*
* return: -1 : this port is invalid or termio == NULL
* 0 : setting O.K.
*
*
* Function 13: Get the DTR/RTS state of this port.
* Syntax:
* int MoxaPortGetLineOut(int port, int *dtrState, int *rtsState);
* int port : port number (0 - 127)
* int * dtrState : pointer to INT to receive the current DTR
* state. (if NULL, this function will not
* write to this address)
* int * rtsState : pointer to INT to receive the current RTS
* state. (if NULL, this function will not
* write to this address)
*
* return: -1 : this port is invalid
* 0 : O.K.
*
*
* Function 14: Setting the DTR/RTS output state of this port.
* Syntax:
* void MoxaPortLineCtrl(int port, int dtrState, int rtsState);
* int port : port number (0 - 127)
* int dtrState : DTR output state (0: off, 1: on)
* int rtsState : RTS output state (0: off, 1: on)
*
*
* Function 15: Setting the flow control of this port.
* Syntax:
* void MoxaPortFlowCtrl(int port, int rtsFlow, int ctsFlow, int rxFlow,
* int txFlow,int xany);
* int port : port number (0 - 127)
* int rtsFlow : H/W RTS flow control (0: no, 1: yes)
* int ctsFlow : H/W CTS flow control (0: no, 1: yes)
* int rxFlow : S/W Rx XON/XOFF flow control (0: no, 1: yes)
* int txFlow : S/W Tx XON/XOFF flow control (0: no, 1: yes)
* int xany : S/W XANY flow control (0: no, 1: yes)
*
*
* Function 16: Get ths line status of this port
* Syntax:
* int MoxaPortLineStatus(int port);
* int port : port number (0 - 127)
*
* return: Bit 0 - CTS state (0: off, 1: on)
* Bit 1 - DSR state (0: off, 1: on)
* Bit 2 - DCD state (0: off, 1: on)
*
*
* Function 17: Check the DCD state has changed since the last read
* of this function.
* Syntax:
* int MoxaPortDCDChange(int port);
* int port : port number (0 - 127)
*
* return: 0 : no changed
* 1 : DCD has changed
*
*
* Function 18: Check ths current DCD state is ON or not.
* Syntax:
* int MoxaPortDCDON(int port);
* int port : port number (0 - 127)
*
* return: 0 : DCD off
* 1 : DCD on
*
*
* Function 19: Flush the Rx/Tx buffer data of this port.
* Syntax:
* void MoxaPortFlushData(int port, int mode);
* int port : port number (0 - 127)
* int mode
* 0 : flush the Rx buffer
* 1 : flush the Tx buffer
* 2 : flush the Rx and Tx buffer
*
*
* Function 20: Write data.
* Syntax:
* int MoxaPortWriteData(int port, unsigned char * buffer, int length);
* int port : port number (0 - 127)
* unsigned char * buffer : pointer to write data buffer.
* int length : write data length
*
* return: 0 - length : real write data length
*
*
* Function 21: Read data.
* Syntax:
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
* int MoxaPortReadData(int port, struct tty_struct *tty);
* int port : port number (0 - 127)
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
* struct tty_struct *tty : tty for data
*
* return: 0 - length : real read data length
*
*
* Function 24: Get the Tx buffer current queued data bytes
* Syntax:
* int MoxaPortTxQueue(int port);
* int port : port number (0 - 127)
*
* return: .. : Tx buffer current queued data bytes
*
*
* Function 25: Get the Tx buffer current free space
* Syntax:
* int MoxaPortTxFree(int port);
* int port : port number (0 - 127)
*
* return: .. : Tx buffer current free space
*
*
* Function 26: Get the Rx buffer current queued data bytes
* Syntax:
* int MoxaPortRxQueue(int port);
* int port : port number (0 - 127)
*
* return: .. : Rx buffer current queued data bytes
*
*
* Function 28: Disable port data transmission.
* Syntax:
* void MoxaPortTxDisable(int port);
* int port : port number (0 - 127)
*
*
* Function 29: Enable port data transmission.
* Syntax:
* void MoxaPortTxEnable(int port);
* int port : port number (0 - 127)
*
*
* Function 31: Get the received BREAK signal count and reset it.
* Syntax:
* int MoxaPortResetBrkCnt(int port);
* int port : port number (0 - 127)
*
* return: 0 - .. : BREAK signal count
*
*
* Function 34: Send out a BREAK signal.
* Syntax:
* void MoxaPortSendBreak(int port, int ms100);
* int port : port number (0 - 127)
* int ms100 : break signal time interval.
* unit: 100 mini-second. if ms100 == 0, it will
* send out a about 250 ms BREAK signal.
*
*/
int MoxaPortIsValid(int port)
{
if (moxaCard == 0)
return (0);
if (moxa_ports[port].chkPort == 0)
return (0);
return (1);
}
void MoxaPortEnable(int port)
{
void __iomem *ofsAddr;
int MoxaPortLineStatus(int);
short lowwater = 512;
ofsAddr = moxa_ports[port].tableAddr;
writew(lowwater, ofsAddr + Low_water);
moxa_ports[port].breakCnt = 0;
if ((moxa_boards[port / MAX_PORTS_PER_BOARD].boardType == MOXA_BOARD_C320_ISA) ||
(moxa_boards[port / MAX_PORTS_PER_BOARD].boardType == MOXA_BOARD_C320_PCI)) {
moxafunc(ofsAddr, FC_SetBreakIrq, 0);
} else {
writew(readw(ofsAddr + HostStat) | WakeupBreak, ofsAddr + HostStat);
}
moxafunc(ofsAddr, FC_SetLineIrq, Magic_code);
moxafunc(ofsAddr, FC_FlushQueue, 2);
moxafunc(ofsAddr, FC_EnableCH, Magic_code);
MoxaPortLineStatus(port);
}
void MoxaPortDisable(int port)
{
void __iomem *ofsAddr = moxa_ports[port].tableAddr;
moxafunc(ofsAddr, FC_SetFlowCtl, 0); /* disable flow control */
moxafunc(ofsAddr, FC_ClrLineIrq, Magic_code);
writew(0, ofsAddr + HostStat);
moxafunc(ofsAddr, FC_DisableCH, Magic_code);
}
long MoxaPortGetMaxBaud(int port)
{
if ((moxa_boards[port / MAX_PORTS_PER_BOARD].boardType == MOXA_BOARD_C320_ISA) ||
(moxa_boards[port / MAX_PORTS_PER_BOARD].boardType == MOXA_BOARD_C320_PCI))
return (460800L);
else
return (921600L);
}
long MoxaPortSetBaud(int port, long baud)
{
void __iomem *ofsAddr;
long max, clock;
unsigned int val;
if ((baud < 50L) || ((max = MoxaPortGetMaxBaud(port)) == 0))
return (0);
ofsAddr = moxa_ports[port].tableAddr;
if (baud > max)
baud = max;
if (max == 38400L)
clock = 614400L; /* for 9.8304 Mhz : max. 38400 bps */
else if (max == 57600L)
clock = 691200L; /* for 11.0592 Mhz : max. 57600 bps */
else
clock = 921600L; /* for 14.7456 Mhz : max. 115200 bps */
val = clock / baud;
moxafunc(ofsAddr, FC_SetBaud, val);
baud = clock / val;
moxa_ports[port].curBaud = baud;
return (baud);
}
int MoxaPortSetTermio(int port, struct ktermios *termio, speed_t baud)
{
void __iomem *ofsAddr;
tcflag_t cflag;
tcflag_t mode = 0;
if (moxa_ports[port].chkPort == 0 || termio == 0)
return (-1);
ofsAddr = moxa_ports[port].tableAddr;
cflag = termio->c_cflag; /* termio->c_cflag */
mode = termio->c_cflag & CSIZE;
if (mode == CS5)
mode = MX_CS5;
else if (mode == CS6)
mode = MX_CS6;
else if (mode == CS7)
mode = MX_CS7;
else if (mode == CS8)
mode = MX_CS8;
if (termio->c_cflag & CSTOPB) {
if (mode == MX_CS5)
mode |= MX_STOP15;
else
mode |= MX_STOP2;
} else
mode |= MX_STOP1;
if (termio->c_cflag & PARENB) {
if (termio->c_cflag & PARODD)
mode |= MX_PARODD;
else
mode |= MX_PAREVEN;
} else
mode |= MX_PARNONE;
moxafunc(ofsAddr, FC_SetDataMode, (ushort) mode);
if ((moxa_boards[port / MAX_PORTS_PER_BOARD].boardType == MOXA_BOARD_C320_ISA) ||
(moxa_boards[port / MAX_PORTS_PER_BOARD].boardType == MOXA_BOARD_C320_PCI)) {
if (baud >= 921600L)
return (-1);
}
baud = MoxaPortSetBaud(port, baud);
if (termio->c_iflag & (IXON | IXOFF | IXANY)) {
writeb(termio->c_cc[VSTART], ofsAddr + FuncArg);
writeb(termio->c_cc[VSTOP], ofsAddr + FuncArg1);
writeb(FC_SetXonXoff, ofsAddr + FuncCode);
moxa_wait_finish(ofsAddr);
}
return (baud);
}
int MoxaPortGetLineOut(int port, int *dtrState, int *rtsState)
{
if (!MoxaPortIsValid(port))
return (-1);
if (dtrState) {
if (moxa_ports[port].lineCtrl & DTR_ON)
*dtrState = 1;
else
*dtrState = 0;
}
if (rtsState) {
if (moxa_ports[port].lineCtrl & RTS_ON)
*rtsState = 1;
else
*rtsState = 0;
}
return (0);
}
void MoxaPortLineCtrl(int port, int dtr, int rts)
{
void __iomem *ofsAddr;
int mode;
ofsAddr = moxa_ports[port].tableAddr;
mode = 0;
if (dtr)
mode |= DTR_ON;
if (rts)
mode |= RTS_ON;
moxa_ports[port].lineCtrl = mode;
moxafunc(ofsAddr, FC_LineControl, mode);
}
void MoxaPortFlowCtrl(int port, int rts, int cts, int txflow, int rxflow, int txany)
{
void __iomem *ofsAddr;
int mode;
ofsAddr = moxa_ports[port].tableAddr;
mode = 0;
if (rts)
mode |= RTS_FlowCtl;
if (cts)
mode |= CTS_FlowCtl;
if (txflow)
mode |= Tx_FlowCtl;
if (rxflow)
mode |= Rx_FlowCtl;
if (txany)
mode |= IXM_IXANY;
moxafunc(ofsAddr, FC_SetFlowCtl, mode);
}
int MoxaPortLineStatus(int port)
{
void __iomem *ofsAddr;
int val;
ofsAddr = moxa_ports[port].tableAddr;
if ((moxa_boards[port / MAX_PORTS_PER_BOARD].boardType == MOXA_BOARD_C320_ISA) ||
(moxa_boards[port / MAX_PORTS_PER_BOARD].boardType == MOXA_BOARD_C320_PCI)) {
moxafunc(ofsAddr, FC_LineStatus, 0);
val = readw(ofsAddr + FuncArg);
} else {
val = readw(ofsAddr + FlagStat) >> 4;
}
val &= 0x0B;
if (val & 8) {
val |= 4;
if ((moxa_ports[port].DCDState & DCD_oldstate) == 0)
moxa_ports[port].DCDState = (DCD_oldstate | DCD_changed);
} else {
if (moxa_ports[port].DCDState & DCD_oldstate)
moxa_ports[port].DCDState = DCD_changed;
}
val &= 7;
return (val);
}
int MoxaPortDCDChange(int port)
{
int n;
if (moxa_ports[port].chkPort == 0)
return (0);
n = moxa_ports[port].DCDState;
moxa_ports[port].DCDState &= ~DCD_changed;
n &= DCD_changed;
return (n);
}
int MoxaPortDCDON(int port)
{
int n;
if (moxa_ports[port].chkPort == 0)
return (0);
if (moxa_ports[port].DCDState & DCD_oldstate)
n = 1;
else
n = 0;
return (n);
}
int MoxaPortWriteData(int port, unsigned char * buffer, int len)
{
int c, total, i;
ushort tail;
int cnt;
ushort head, tx_mask, spage, epage;
ushort pageno, pageofs, bufhead;
void __iomem *baseAddr, *ofsAddr, *ofs;
ofsAddr = moxa_ports[port].tableAddr;
baseAddr = moxa_boards[port / MAX_PORTS_PER_BOARD].basemem;
tx_mask = readw(ofsAddr + TX_mask);
spage = readw(ofsAddr + Page_txb);
epage = readw(ofsAddr + EndPage_txb);
tail = readw(ofsAddr + TXwptr);
head = readw(ofsAddr + TXrptr);
c = (head > tail) ? (head - tail - 1)
: (head - tail + tx_mask);
if (c > len)
c = len;
moxaLog.txcnt[port] += c;
total = c;
if (spage == epage) {
bufhead = readw(ofsAddr + Ofs_txb);
writew(spage, baseAddr + Control_reg);
while (c > 0) {
if (head > tail)
len = head - tail - 1;
else
len = tx_mask + 1 - tail;
len = (c > len) ? len : c;
ofs = baseAddr + DynPage_addr + bufhead + tail;
for (i = 0; i < len; i++)
writeb(*buffer++, ofs + i);
tail = (tail + len) & tx_mask;
c -= len;
}
writew(tail, ofsAddr + TXwptr);
} else {
len = c;
pageno = spage + (tail >> 13);
pageofs = tail & Page_mask;
do {
cnt = Page_size - pageofs;
if (cnt > c)
cnt = c;
c -= cnt;
writeb(pageno, baseAddr + Control_reg);
ofs = baseAddr + DynPage_addr + pageofs;
for (i = 0; i < cnt; i++)
writeb(*buffer++, ofs + i);
if (c == 0) {
writew((tail + len) & tx_mask, ofsAddr + TXwptr);
break;
}
if (++pageno == epage)
pageno = spage;
pageofs = 0;
} while (1);
}
writeb(1, ofsAddr + CD180TXirq); /* start to send */
return (total);
}
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
int MoxaPortReadData(int port, struct tty_struct *tty)
{
register ushort head, pageofs;
int i, count, cnt, len, total, remain;
ushort tail, rx_mask, spage, epage;
ushort pageno, bufhead;
void __iomem *baseAddr, *ofsAddr, *ofs;
ofsAddr = moxa_ports[port].tableAddr;
baseAddr = moxa_boards[port / MAX_PORTS_PER_BOARD].basemem;
head = readw(ofsAddr + RXrptr);
tail = readw(ofsAddr + RXwptr);
rx_mask = readw(ofsAddr + RX_mask);
spage = readw(ofsAddr + Page_rxb);
epage = readw(ofsAddr + EndPage_rxb);
count = (tail >= head) ? (tail - head)
: (tail - head + rx_mask + 1);
if (count == 0)
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
return 0;
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
total = count;
remain = count - total;
moxaLog.rxcnt[port] += total;
count = total;
if (spage == epage) {
bufhead = readw(ofsAddr + Ofs_rxb);
writew(spage, baseAddr + Control_reg);
while (count > 0) {
if (tail >= head)
len = tail - head;
else
len = rx_mask + 1 - head;
len = (count > len) ? len : count;
ofs = baseAddr + DynPage_addr + bufhead + head;
for (i = 0; i < len; i++)
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
tty_insert_flip_char(tty, readb(ofs + i), TTY_NORMAL);
head = (head + len) & rx_mask;
count -= len;
}
writew(head, ofsAddr + RXrptr);
} else {
len = count;
pageno = spage + (head >> 13);
pageofs = head & Page_mask;
do {
cnt = Page_size - pageofs;
if (cnt > count)
cnt = count;
count -= cnt;
writew(pageno, baseAddr + Control_reg);
ofs = baseAddr + DynPage_addr + pageofs;
for (i = 0; i < cnt; i++)
[PATCH] TTY layer buffering revamp The API and code have been through various bits of initial review by serial driver people but they definitely need to live somewhere for a while so the unconverted drivers can get knocked into shape, existing drivers that have been updated can be better tuned and bugs whacked out. This replaces the tty flip buffers with kmalloc objects in rings. In the normal situation for an IRQ driven serial port at typical speeds the behaviour is pretty much the same, two buffers end up allocated and the kernel cycles between them as before. When there are delays or at high speed we now behave far better as the buffer pool can grow a bit rather than lose characters. This also means that we can operate at higher speeds reliably. For drivers that receive characters in blocks (DMA based, USB and especially virtualisation) the layer allows a lot of driver specific code that works around the tty layer with private secondary queues to be removed. The IBM folks need this sort of layer, the smart serial port people do, the virtualisers do (because a virtualised tty typically operates at infinite speed rather than emulating 9600 baud). Finally many drivers had invalid and unsafe attempts to avoid buffer overflows by directly invoking tty methods extracted out of the innards of work queue structs. These are no longer needed and all go away. That fixes various random hangs with serial ports on overflow. The other change in here is to optimise the receive_room path that is used by some callers. It turns out that only one ldisc uses receive room except asa constant and it updates it far far less than the value is read. We thus make it a variable not a function call. I expect the code to contain bugs due to the size alone but I'll be watching and squashing them and feeding out new patches as it goes. Because the buffers now dynamically expand you should only run out of buffering when the kernel runs out of memory for real. That means a lot of the horrible hacks high performance drivers used to do just aren't needed any more. Description: tty_insert_flip_char is an old API and continues to work as before, as does tty_flip_buffer_push() [this is why many drivers dont need modification]. It does now also return the number of chars inserted There are also tty_buffer_request_room(tty, len) which asks for a buffer block of the length requested and returns the space found. This improves efficiency with hardware that knows how much to transfer. and tty_insert_flip_string_flags(tty, str, flags, len) to insert a string of characters and flags For a smart interface the usual code is len = tty_request_buffer_room(tty, amount_hardware_says); tty_insert_flip_string(tty, buffer_from_card, len); More description! At the moment tty buffers are attached directly to the tty. This is causing a lot of the problems related to tty layer locking, also problems at high speed and also with bursty data (such as occurs in virtualised environments) I'm working on ripping out the flip buffers and replacing them with a pool of dynamically allocated buffers. This allows both for old style "byte I/O" devices and also helps virtualisation and smart devices where large blocks of data suddenely materialise and need storing. So far so good. Lots of drivers reference tty->flip.*. Several of them also call directly and unsafely into function pointers it provides. This will all break. Most drivers can use tty_insert_flip_char which can be kept as an API but others need more. At the moment I've added the following interfaces, if people think more will be needed now is a good time to say int tty_buffer_request_room(tty, size) Try and ensure at least size bytes are available, returns actual room (may be zero). At the moment it just uses the flipbuf space but that will change. Repeated calls without characters being added are not cumulative. (ie if you call it with 1, 1, 1, and then 4 you'll have four characters of space. The other functions will also try and grow buffers in future but this will be a more efficient way when you know block sizes. int tty_insert_flip_char(tty, ch, flag) As before insert a character if there is room. Now returns 1 for success, 0 for failure. int tty_insert_flip_string(tty, str, len) Insert a block of non error characters. Returns the number inserted. int tty_prepare_flip_string(tty, strptr, len) Adjust the buffer to allow len characters to be added. Returns a buffer pointer in strptr and the length available. This allows for hardware that needs to use functions like insl or mencpy_fromio. Signed-off-by: Alan Cox <alan@redhat.com> Cc: Paul Fulghum <paulkf@microgate.com> Signed-off-by: Hirokazu Takata <takata@linux-m32r.org> Signed-off-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Jeff Dike <jdike@addtoit.com> Signed-off-by: John Hawkes <hawkes@sgi.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-10 04:54:13 +00:00
tty_insert_flip_char(tty, readb(ofs + i), TTY_NORMAL);
if (count == 0) {
writew((head + len) & rx_mask, ofsAddr + RXrptr);
break;
}
if (++pageno == epage)
pageno = spage;
pageofs = 0;
} while (1);
}
if ((readb(ofsAddr + FlagStat) & Xoff_state) && (remain < LowWater)) {
moxaLowWaterChk = 1;
moxa_ports[port].lowChkFlag = 1;
}
return (total);
}
int MoxaPortTxQueue(int port)
{
void __iomem *ofsAddr;
ushort rptr, wptr, mask;
int len;
ofsAddr = moxa_ports[port].tableAddr;
rptr = readw(ofsAddr + TXrptr);
wptr = readw(ofsAddr + TXwptr);
mask = readw(ofsAddr + TX_mask);
len = (wptr - rptr) & mask;
return (len);
}
int MoxaPortTxFree(int port)
{
void __iomem *ofsAddr;
ushort rptr, wptr, mask;
int len;
ofsAddr = moxa_ports[port].tableAddr;
rptr = readw(ofsAddr + TXrptr);
wptr = readw(ofsAddr + TXwptr);
mask = readw(ofsAddr + TX_mask);
len = mask - ((wptr - rptr) & mask);
return (len);
}
int MoxaPortRxQueue(int port)
{
void __iomem *ofsAddr;
ushort rptr, wptr, mask;
int len;
ofsAddr = moxa_ports[port].tableAddr;
rptr = readw(ofsAddr + RXrptr);
wptr = readw(ofsAddr + RXwptr);
mask = readw(ofsAddr + RX_mask);
len = (wptr - rptr) & mask;
return (len);
}
void MoxaPortTxDisable(int port)
{
void __iomem *ofsAddr;
ofsAddr = moxa_ports[port].tableAddr;
moxafunc(ofsAddr, FC_SetXoffState, Magic_code);
}
void MoxaPortTxEnable(int port)
{
void __iomem *ofsAddr;
ofsAddr = moxa_ports[port].tableAddr;
moxafunc(ofsAddr, FC_SetXonState, Magic_code);
}
int MoxaPortResetBrkCnt(int port)
{
ushort cnt;
cnt = moxa_ports[port].breakCnt;
moxa_ports[port].breakCnt = 0;
return (cnt);
}
void MoxaPortSendBreak(int port, int ms100)
{
void __iomem *ofsAddr;
ofsAddr = moxa_ports[port].tableAddr;
if (ms100) {
moxafunc(ofsAddr, FC_SendBreak, Magic_code);
msleep(ms100 * 10);
} else {
moxafunc(ofsAddr, FC_SendBreak, Magic_code);
msleep(250);
}
moxafunc(ofsAddr, FC_StopBreak, Magic_code);
}
static int moxa_get_serial_info(struct moxa_port *info,
struct serial_struct __user *retinfo)
{
struct serial_struct tmp;
memset(&tmp, 0, sizeof(tmp));
tmp.type = info->type;
tmp.line = info->port;
tmp.port = 0;
tmp.irq = 0;
tmp.flags = info->asyncflags;
tmp.baud_base = 921600;
tmp.close_delay = info->close_delay;
tmp.closing_wait = info->closing_wait;
tmp.custom_divisor = 0;
tmp.hub6 = 0;
if(copy_to_user(retinfo, &tmp, sizeof(*retinfo)))
return -EFAULT;
return (0);
}
static int moxa_set_serial_info(struct moxa_port *info,
struct serial_struct __user *new_info)
{
struct serial_struct new_serial;
if(copy_from_user(&new_serial, new_info, sizeof(new_serial)))
return -EFAULT;
if ((new_serial.irq != 0) ||
(new_serial.port != 0) ||
// (new_serial.type != info->type) ||
(new_serial.custom_divisor != 0) ||
(new_serial.baud_base != 921600))
return (-EPERM);
if (!capable(CAP_SYS_ADMIN)) {
if (((new_serial.flags & ~ASYNC_USR_MASK) !=
(info->asyncflags & ~ASYNC_USR_MASK)))
return (-EPERM);
} else {
info->close_delay = new_serial.close_delay * HZ / 100;
info->closing_wait = new_serial.closing_wait * HZ / 100;
}
new_serial.flags = (new_serial.flags & ~ASYNC_FLAGS);
new_serial.flags |= (info->asyncflags & ASYNC_FLAGS);
if (new_serial.type == PORT_16550A) {
MoxaSetFifo(info->port, 1);
} else {
MoxaSetFifo(info->port, 0);
}
info->type = new_serial.type;
return (0);
}
/*****************************************************************************
* Static local functions: *
*****************************************************************************/
static void moxafunc(void __iomem *ofsAddr, int cmd, ushort arg)
{
writew(arg, ofsAddr + FuncArg);
writew(cmd, ofsAddr + FuncCode);
moxa_wait_finish(ofsAddr);
}
static void moxa_wait_finish(void __iomem *ofsAddr)
{
unsigned long i, j;
i = jiffies;
while (readw(ofsAddr + FuncCode) != 0) {
j = jiffies;
if ((j - i) > moxaFuncTout) {
return;
}
}
}
static void moxa_low_water_check(void __iomem *ofsAddr)
{
int len;
ushort rptr, wptr, mask;
if (readb(ofsAddr + FlagStat) & Xoff_state) {
rptr = readw(ofsAddr + RXrptr);
wptr = readw(ofsAddr + RXwptr);
mask = readw(ofsAddr + RX_mask);
len = (wptr - rptr) & mask;
if (len <= Low_water)
moxafunc(ofsAddr, FC_SendXon, 0);
}
}
static void MoxaSetFifo(int port, int enable)
{
void __iomem *ofsAddr = moxa_ports[port].tableAddr;
if (!enable) {
moxafunc(ofsAddr, FC_SetRxFIFOTrig, 0);
moxafunc(ofsAddr, FC_SetTxFIFOCnt, 1);
} else {
moxafunc(ofsAddr, FC_SetRxFIFOTrig, 3);
moxafunc(ofsAddr, FC_SetTxFIFOCnt, 16);
}
}