mmap

mmap

[mdaljeet]

I am doing the following:

   malloc some memory is user space
   pass its pointer to some kernel module
   in the kernel module...do a pci_alloc_consistent so that i get a memory
   region for PCI DMA operations

now the problem is that i want to remap the address range pointed by the
user space pointer to the memory region allocated by the
'pci_alloc_consistent' inside the module. I think this is possible..need
some hints....

thanks,
Daljeet Maini
IBM Global Services Ltd. - Bangalore
Ph. No. - 5267117 Extn 2954

Re: mmap

[Gerd Knorr]

mdaljeet@in.ibm.com wrote:
>  I am doing the following:
>  
>     malloc some memory is user space
>     pass its pointer to some kernel module
>     in the kernel module...do a pci_alloc_consistent so that i get a memory
>     region for PCI DMA operations

Wrong approach, you can use kiobufs if you want DMA to the malloc()ed
userspace memory:

 * lock down the user memory using map_user_kiobuf() + lock_kiovec()
   (see linux/iobuf.h).
 * translate the iobuf->maplist into a scatterlist [1]
 * feed pci_map_sg() with the scatterlist to get DMA addresses.
   you can pass to the hardware.

And the reverse to free everything when you are done of course.

  Gerd

[1] IMHO it would be more useful if iobufs would use a scatterlist
    instead of an struct page* array.


-- 
Gerd Knorr <kraxel@bytesex.org>  --  SuSE Labs, Außenstelle Berlin

Re: mmap

[Alan Cox]

> now the problem is that i want to remap the address range pointed by the
> user space pointer to the memory region allocated by the
> 'pci_alloc_consistent' inside the module. I think this is possible..need
> some hints....

Wrong way around. Ask the device to create its mapping and reply with the size
mmap the object

Re: mmap

[mdaljeet]

[-- Attachment #1: Type: text/plain, Size: 1077 bytes --]

When I malloc the memory in user space, the memory may be discontinuous for
large chunks of memory say 16k or 32k. Does the 'kiobuf' interface take
care of this or it assumes it to be continuous?

regards,
Daljeet Maini
IBM Global Services Ltd. - Bangalore
Ph. No. - 5267117 Extn 2954


|--------+----------------------->
|        |          Gerd Knorr   |
|        |          <kraxel@bytes|
|        |          ex.org>      |
|        |                       |
|        |          05/15/01     |
|        |          01:03 PM     |
|        |          Please       |
|        |          respond to   |
|        |          Gerd Knorr   |
|        |                       |
|--------+----------------------->
  >--------------------------------------------------------|
  |                                                        |
  |       To:     linux-kernel@vger.kernel.org             |
  |       cc:     (bcc: Daljeet Maini/India/IBM)           |
  |       Subject:     Re: mmap                            |
  >--------------------------------------------------------|





[-- Attachment #2: Type: text/plain, Size: 1125 bytes --]



mdaljeet@in.ibm.com wrote:
>  I am doing the following:
>
>     malloc some memory is user space
>     pass its pointer to some kernel module
>     in the kernel module...do a pci_alloc_consistent so that i get a
memory
>     region for PCI DMA operations

Wrong approach, you can use kiobufs if you want DMA to the malloc()ed
userspace memory:

 * lock down the user memory using map_user_kiobuf() + lock_kiovec()
   (see linux/iobuf.h).
 * translate the iobuf->maplist into a scatterlist [1]
 * feed pci_map_sg() with the scatterlist to get DMA addresses.
   you can pass to the hardware.

And the reverse to free everything when you are done of course.

  Gerd

[1] IMHO it would be more useful if iobufs would use a scatterlist
    instead of an struct page* array.


--
Gerd Knorr <kraxel@bytesex.org>  --  SuSE Labs, Außenstelle Berlin
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Re: mmap

[mdaljeet]

[-- Attachment #1: Type: text/plain, Size: 1313 bytes --]

I use the 'map_user_kiobuf' and 'lock_kiovec' kernel routines in a module
for 'user space memory'. After that if I pass the
'(iobuf->maplist[0])-mem_map) <<  PAGE_SHIFT)' to the hardware for DMA
operations and it works fine for Intel platforms. Now how can I use the
'iobuf' struct obtained after lock_kiovec operation to get a PCI bus
address that I can pass to hardware for DMA operations on my Apple
machine.?

thanks,
Daljeet.


|--------+----------------------->
|        |          Gerd Knorr   |
|        |          <kraxel@bytes|
|        |          ex.org>      |
|        |                       |
|        |          05/15/01     |
|        |          01:03 PM     |
|        |          Please       |
|        |          respond to   |
|        |          Gerd Knorr   |
|        |                       |
|--------+----------------------->
  >-----------------------------------------------------------------------|
  |                                                                       |
  |       To:     linux-kernel@vger.kernel.org                            |
  |       cc:     (bcc: Daljeet Maini/India/IBM)                          |
  |       Subject:     Re: mmap                                           |
  >-----------------------------------------------------------------------|





[-- Attachment #2: Type: text/plain, Size: 1125 bytes --]



mdaljeet@in.ibm.com wrote:
>  I am doing the following:
>
>     malloc some memory is user space
>     pass its pointer to some kernel module
>     in the kernel module...do a pci_alloc_consistent so that i get a
memory
>     region for PCI DMA operations

Wrong approach, you can use kiobufs if you want DMA to the malloc()ed
userspace memory:

 * lock down the user memory using map_user_kiobuf() + lock_kiovec()
   (see linux/iobuf.h).
 * translate the iobuf->maplist into a scatterlist [1]
 * feed pci_map_sg() with the scatterlist to get DMA addresses.
   you can pass to the hardware.

And the reverse to free everything when you are done of course.

  Gerd

[1] IMHO it would be more useful if iobufs would use a scatterlist
    instead of an struct page* array.


--
Gerd Knorr <kraxel@bytesex.org>  --  SuSE Labs, Außenstelle Berlin
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Re: mmap

[Jens Axboe]

On Mon, Jul 02 2001, mdaljeet@in.ibm.com wrote:
> [1] IMHO it would be more useful if iobufs would use a scatterlist
>     instead of an struct page* array.

No that would be horrible, at least with the current scatterlist. A page
based scatterlist would be alright though -- but this boils down to the
per-page offset debate again.

-- 
Jens Axboe

mmap

[Sébastien Paumier]

[-- Attachment #1: Type: text/plain, Size: 699 bytes --]

Hi,
I have a question about mmap's behavior when one tries to map a file asking for 
a length greater than the actual file size. When I run the attached code on a 
100 bytes file, I have the following output:

(... file content followed by zeros...)
n=4096
write: Bad address

So, it seems that the actual memory area provided by mmap is one page large and 
not the requested length of filesize+10000. I guess that 'write' writes less 
than requested because it was interrupted by the SIGBUS signal. And my question is:

shouldn't mmap either complain about the requested length or provide an 
accessible area of the requested length, instead of silently failing ?

Best regards,
Sébastien Paumier

[-- Attachment #2: mmap.c --]
[-- Type: text/x-csrc, Size: 884 bytes --]

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <string.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/mman.h>
#include <sys/wait.h>
#include <signal.h>
#include <time.h>
#include <errno.h>

int main(int argc, char* argv[]){
if(argc != 2) {
	fprintf(stderr,"Usage: %s <file>\n",argv[0]);
	exit(1);
}
int fd = open(argv[argc-1],O_RDONLY);
if (fd==-1) {
	perror("open");
	return 1;
}
struct stat status;
if(-1 == fstat(fd,&status)){
	perror("fstat");
	return 1;
}
char* mem=(char*) mmap(NULL, status.st_size+10000, PROT_READ, MAP_PRIVATE,fd,0);
if(mem == MAP_FAILED){
	perror("mmap");
	return 1;
}
int n;
if ( -1 == (n=write(1,mem, status.st_size+10000))) {
	perror("write");
	return 1;
}
printf("\nn=%d\n",n);
if( -1 == (n=write(1,mem+n, status.st_size+10000-n))) {
	perror("write");
	return 1;
}
printf("\nn=%d\n",n);
return 0;
}

Re: mmap

[Eric Dumazet]

Le vendredi 02 décembre 2011 à 13:20 +0100, Sébastien Paumier a écrit :
> Hi,
> I have a question about mmap's behavior when one tries to map a file asking for 
> a length greater than the actual file size. When I run the attached code on a 
> 100 bytes file, I have the following output:
> 
> (... file content followed by zeros...)
> n=4096
> write: Bad address
> 
> So, it seems that the actual memory area provided by mmap is one page large and 
> not the requested length of filesize+10000. I guess that 'write' writes less 
> than requested because it was interrupted by the SIGBUS signal. And my question is:
> 
> shouldn't mmap either complain about the requested length or provide an 
> accessible area of the requested length, instead of silently failing ?


Accessing non existing memory leads to SIGBUS signal, not a silent
failure. Its documented behavior.

man mmap

       SIGBUS Attempted access to a portion of the buffer that does not corre‐
              spond to the file (for example, beyond  the  end  of  the  file,
              including  the  case  where  another  process  has truncated the
              file).