usage count in device driver and concurrency
Hi,
I am wondering how to handle device usage count in open and release call of device driver if hardware need to be initialized on first open and shutdown on last close. I have seen som code like
int open()
{
/*some code*/
device->usage++;
if(device->usage==1)
init_hardware();
/*rest of code*/
}
void release ()
{
/*some code*/
if(device->usage==1)
shutdown_hardware();
device->usage--;
/*rest of code*/
}However, it seems to me that this code can make problem.
If device->usage=0, and two process A,B execute line
device->usage++;
concurretly, device->usage will become 2 when they come to next line. This will result in hardware being used without initialization. Same things can happen in release call also, which will result in no shutdown of hardware.
I have seen this type of code in
/linux-2.6.8/drivers/ide/ide-disk.c and /linux-2.6.8/drivers/ide/ide-floppy.c
Please let me know if its bug or not before I
start working on patches.
Regards,
Tushar
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
Quoted 38 lines · Adeshara Tushar
Hi,
I am wondering how to handle device usage count in open and release call of device driver if hardware need to be initialized on first open and shutdown on last close. I have seen som code likeint
open(){ /*some code*/ device->usage++; if(device->usage==1) init_hardware(); /*rest of code*/ } void release () { /*some code*/ if(device->usage==1) shutdown_hardware(); device->usage--; /*rest of code*/ }However, it seems to me that this code can make problem.
If device->usage=0, and two process A,B execute linedevice->usage++;concurretly,
device->usagewill become 2 when they come to next line. This will result in hardware being used without initialization. Same things can happen in release call also, which will result in no shutdown of hardware.
I have seen this type of code in
/linux-2.6.8/drivers/ide/ide-disk.c and /linux-2.6.8/drivers/ide/ide-floppy.c
Quoted 2 lines
Please let me know if its bug or not before I
start working on patches.
Not a bug: ->open() and ->release() for block devices are never called concurrently (because of bdev->bd_sem).
Bartlomiej