Hi Greg --- Greg KH <greg@xxxxxxxxx> escreveu: > On Wed, Dec 06, 2006 at 11:14:50AM -0300, Mauricio Lin wrote: > > --- Greg KH <greg@xxxxxxxxx> escreveu: > > > On Tue, Dec 05, 2006 at 08:57:48PM +0000, Mauricio Lin wrote: > > > > Hi all, > > > > > > > > I implemented a module that sends signals to user space using > > > > kobject_uevent() when some conditions happen. That step is > working. > > > > > > > > When the signal is sent, a sysfs attribute is also changed > (switch > > > from > > > > 0 to 1 or vice-versa). This means that the corresponding file > under > > > > /sys/ has its contents modified as well (using > "_name"_show(struct > > > > subsystem *subsys, char *page) function) . > > > > > > > > The user space when receives the signal (sent by > kobject_uevent()) > > > it > > > > tries to read the file under /sys (that was changed to 1 when > the > > > > signal was sent), but the value is always 0. IMHO there is a > sync > > > > problem, because when user space gets the signal and it reads > the > > > file > > > > under /sys, the value set to 1 (notify moment) could be changed > > > back to > > > > 0 (read moment) later. > > > > > > Then userspace isn't getting there fast enough it seems :) > > > > Yes, that's right. > > Then your design is wrong, you need to rethink what you are wanting > to > achieve here. > > > > > So is there any way to send a signal and guarantee that the > > > attribute > > > > will be read by user space before being changed by subsequent > > > actions? > > > > Is sysfs_notify() a possible solution for that? > > > > > > How about just poll() on the sysfs file? That way your program > gets > > > woken up, and you read the proper data. > > > > Humm, is it efficient? It must be something that does not consume > much > > memory. > > Memory where? In the kernel or in userspace? Userspace. > > > > But what is the real thing you are trying to do here? It seems > you > > > are > > > trying to use sysfs in an odd way, perhaps you might want to do > > > something else? > > > > Humm, you gave me an idea. I have just to send the signal when the > > attribute is changed to 1 and not 0, so there is no need to include > an > > additional read from user space to check if the value is 1 again. > > > > Let me check the code again :) > > You didn't answer my question about what you are trying to do here... I want to avoid OOM Killer so when the memory RAM reaches a watermark (defined in advance) it notifies the user space applications to free resources. BR, Mauricio Lin. > > thanks, > > greg k-h > _______________________________________________________ Você quer respostas para suas perguntas? Ou você sabe muito e quer compartilhar seu conhecimento? Experimente o Yahoo! Respostas ! http://br.answers.yahoo.com/ -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
