[9fans] stupid kernel question

[9fans] stupid kernel question

[Steve Simon]

how does kernel code call plan9 system calls, does it perform a
trap from kernel back into kernel space and make the request just
like a user process or is there a more direct, subtle route?

thanks,

-Steve

Re: [9fans] stupid kernel question

[erik quanstrom]

On Fri Oct  1 15:25:16 EDT 2010, steve@quintile.net wrote:
> how does kernel code call plan9 system calls, does it perform a
> trap from kernel back into kernel space and make the request just
> like a user process or is there a more direct, subtle route?

the short answer is that it doesn't.  but read/write/open/close/stat
are done via channels, and often with Blocks.  for example
(hopefully without typos)

	c = namec("/sys/src/9/pc/mkfile", Aopen, OREAD, 0);		/* can call error */
	b = devtab[c->type]->bread(c, 64*1024, 0);
	/* b->wp points to data if successful BLEN(b) is the length */

- erik

Re: [9fans] stupid kernel question

[Francisco J Ballesteros]

I'm kind of sleepy. But if I managed to read your mail correctly...

when the kernel wants to do something that can be done with a system
call (mostly read, write, open, ...), it uses the device tab directly, like in

devtab[c->type].write(chan, ....),

using Chan instead of fds in this case.

I mean, I think that almost always you use the internal interface wrapped
by the system call code.

On Fri, Oct 1, 2010 at 9:17 PM, Steve Simon <steve@quintile.net> wrote:
> how does kernel code call plan9 system calls, does it perform a
> trap from kernel back into kernel space and make the request just
> like a user process or is there a more direct, subtle route?
>
> thanks,
>
> -Steve
>
>

Re: [9fans] stupid kernel question

[Steve Simon]

Thanks to all who replied on-list and off it.

using namec and calling through the device tab looks like what I need,
calling sysopen() with an array of integers seemed just too crufty, I
knew there must be a better way ☺

-Steve