/dev/port BUG and possible workaround

Hello!

It seems that I've found out bug in subj
(sorry if this subj was already discussed here)

looking at sources of drivers/char/mem.c I found out:

static ssize_t read_port(struct file * file, char * buf,
size_t count, loff_t *ppos)

{
	unsigned long i = *ppos;
[snip]
	while (count-- > 0 && i < 65536) {
		if (__put_user(inb(i),tmp) < 0) 
			return -EFAULT;  

[snip]
}

static ssize_t write_port(struct file * file, const char * buf,
size_t count, loff_t *ppos)

{
	unsigned long i = *ppos;
[snip]
	while (count-- > 0 && i < 65536) {
		char c;
		if (__get_user(c, tmp)) 
			return -EFAULT; 
		outb(c,i);

[snip]
}

Well, when I'm trying to use this device in the way:
lseek(port_fd,port_idx,SEEK_SET); write(port_fd,&val32,4);
I get perfectly broken results due outl() != outb()*4 Please look at logs:
1. Correct log with using of inport/outport:
outb(CF8,0)
outb(CFA,0)
FF=inb(CF8)
FF=inb(CFA)
80FFFFFC=inl(CF8)
outl(CF8,80000000)
80000000=inl(CF8)
outl(CF8,80FFFFFC)
outl(CF8,80000000)
3051106=inl(CFC)
outl(CF8,80000004)
22100006=inl(CFC)
outl(CF8,80000008)
6000003=inl(CFC)
outl(CF8,8000000C)
800=inl(CFC)
2. Wrong log with using of /dev/port:
outb(CF8,0)
outb(CFA,0)
FF=inb(CF8)
FF=inb(CFA)
FFFFFFFF=inl(CF8)
outl(CF8,80000000)
FFFFFFFF=inl(CF8)
outl(CF8,FFFFFFFF)

As possible workaround it would be better to examine size of buffer and use corresponded insn within of these functions:
switch(size)
{
case 4: outl
case 2: outw
default: outb
}

But it seems that nobody uses this device. Then what is goal of implementing of this device?

Best regards! Nick

Uz.ytkownik Nick Kurshev napisa?:

...

...

Basically the goal is that contrary to some silly /proc stuff which is "en vouge" nowadays you have the ability to controll port access by using normal user permission control semantics of unix file access permissions, by giving /dev/port a proper group and so on. This is legacy crap of course, since the above goal can be reached by using a apache-suexec alike wrapper as well... even with more fine grained resolution of access controll.

Hello, Martin!

But this device is unworking in general. It simply unusable for programs which require 4-byte operations with ports. As I wrote: outl() != 4*outb() same as inl() != 4*inb()

Maybe after fixing these flaws it will be usable but not for now.

Best regards! Nick

It also allows you to write a watchdog driver in shell :-)

#!/bin/sh
# [email protected]
# This supports both the Ibase MB700 and Advantech PCM9576

usage() {
echo "Usage: basename $0 [pat] [playdead]"
exit 1
}

if [ $# != 1 ]; then usage; fi;
if [ $1 != "pat" ]; then

    if [ $1 != "playdead" ]; then
        usage
    fi

fi

HW_VERSION=cat /var/run/HW_VERSION

if [ "$HW_VERSION" == "IBASE MB700" ];
    ENABLE_PORT=`printf %d 0x443`
    DISABLE_PORT=`printf %d 0x441`

    if [ $1 == "pat" ]; then
        TIMEOUT='\x5' #20 seconds (0=30s, 1=28s, ..., F=0s)
        printf "$TIMEOUT" | dd bs=1 seek=$ENABLE_PORT of=/dev/port
    else
        #write any value to port to disable
        printf "\001" | dd bs=1 seek=$DISABLE_PORT of=/dev/port
    fi

elif [ "$HW_VERSION" == "Advantech PCM9576" ]; then
DISENABLE_PORT=printf %d 0x443

if [ $1 == "pat" ]; then
TIMEOUT='\x14' #20 seconds (1=1s, 2=2s, ..., 3E=62s) printf "$TIMEOUT" | dd bs=1 seek=$DISENABLE_PORT of=/dev/port
else
#read from port to disable
dd bs=1 count=1 skip=$DISENABLE_PORT if=/dev/port of=/dev/null
fi
fi