Re: PCI device driver question

Thanks for your help. I followed this advice:

Having said all that, if this is just a one-off or occasional
special-purpose administrative operation, you may not even need to
write a driver (on most platforms at least). You can use the lspci
program to find the physical address of the pci device's memory. Then
you open "/dev/mem", mmap the desired area and read / write data
directly to or from it in your user program.

Following GH's advice, I looked at lspci, and found that the device
Region 0 memory is mapped at a0000000 (32-bit, prefetch, size=512 M)

I wrote the following program:

#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/types.h>
#include <sys/mman.h>
#include <asm/io.h>

int main(){

int memory;
unsigned int data;
unsigned char *address;

memory = open("/dev/mem", O_RDWR);
address = mmap((void*)0, 0xafffffff, PORT_WRITE | PORT_READ, MAP_FIXED,
memory, 0xa0000000);

data = readl(address);
printf("Data: %x", data);
writel(0xFF, address);
data = readl(address);
printf("Data: %x", data);

munmap((void*)address, 0xafffffff);
close(memory);
return 0;

}

I just want to make sure that my syntax is correct. When I went to run
the makefile, however, it did not work. It was having problems
recognizing readl and writel even though both functions should be in
<asm/io.h>. My makefile is the following:
testIO: testIO.c
gcc -o $@ $<

I have the following errors/warnings:

In file included from /usr/include/asm/io.h:11 from testIO.c:6:
/usr/include/asm-i386/io.h:1:2: warning: #warning "You should include
<sys/io.h>. This time I will do it for you."

testIO.c: (.text+0x59): undefined reference to 'readl'
testIO.c: (.text+0x82): undefined reference to 'writel'
testIO.c: (.text+0x90): undefined reference to 'readl'

gil_hamilton@xxxxxxxxxxx wrote:
elliotng.ee@xxxxxxxxx wrote:
OK, after doing some research, it appears that I can use the char
device read and write functions in the pci device driver. To summarize
what I have, I have written a pci device driver. In userspace, I want
to be able to access the pci device driver, read from memory, and write
to memory. I have recently found out that I can try to read/write from
using file operations.

(1) You need to map the device memory into kernel virtual address space
with ioremap. Then, in the test_write function, you copy data from
user space into kernel space (possibly to a small temporary holding
area), then use writel to put it into the (mapped) device memory
(adding in the f_pos value).

test_read is simply the inverse, readl to read the device memory space,
then copy it to user space.

copy_to_user and copy_from_user can be used for data transfer between
user space and kernel space. Look for the book "Linux Device Drivers"
(3rd edition) -- which is available in pdf format for free online --
for much detailed info.

(2) You create a device node (e.g. "mknod /dev/mydevice c xxx 0" where
xxx is your device's major number from /proc/devices). Your program
then opens /dev/mydevice, lseek's to the proper offset and does read or
write calls.

Having said all that, if this is just a one-off or occasional
special-purpose administrative operation, you may not even need to
write a driver (on most platforms at least). You can use the lspci
program to find the physical address of the pci device's memory. Then
you open "/dev/mem", mmap the desired area and read / write data
directly to or from it in your user program.

GH

.

Relevant Pages

  • Re: sleeping and waiting and tasklets
    ... care must be used when accessing user space from kernel code. ... being addressed might not be currently present in memory, ... The net result for the driver writer is that any ... they also check whether the user space pointer is valid. ...
    (comp.os.linux.development.system)
  • Re: Why get_user_pages fails?
    ... memory above 512M to user space and got the user space virtual ... address for that memory. ... The get_user_pagesin my driver fails. ... You're bypassing the Linux ...
    (Linux-Kernel)
  • RE: Why get_user_pages fails?
    ... I totally agree with you that the best solution is to bypass the Linux VM. ... The reason to continue using get_user_pagesis that we can leverage the existing kernel driver - the SCST. ... Do I have a way to map a range of phy memory to user virtual address and have the page structures built for that memory region? ... memory above 512M to user space and got the user space virtual address ...
    (Linux-Kernel)
  • Re: user_to_phys() without mmap?
    ... |> |> ram, not some PCI or AGP video card), but mmap'ing from kernel space ... |> |> into user space is causing large latencies and unsightly artifacts. ... The /dev/mem driver ... |> | virtual memory locations from user space, ...
    (comp.os.linux.development.system)
  • Re: [PATCH] uio: User IRQ Mode
    ... In this mode the user space driver ... is responsible for acknowledging and re-enabling the interrupt. ... This can easily be done without your patch. ...
    (Linux-Kernel)