misc questions about kernel 2.4.x internals

misc questions about kernel 2.4.x internals

[Christian Widmer]

hi,

1) when using any functions that can block i need to do this in the context 
of a process. so a can't read, write to sockets in a bottom-half of a 
interrupt handler. thats why i need to use a kernel thread (i don't what to
use a user level process). my question now is - how long does it take until 
my kernel thread starts running? do i have a way to give it very high 
priority and force my thread to be scheduled so that i can be 'sure' to run 
just after softirq's, tasklets, ...?

2) for module writers there is documented and easy to use api how to use 
tasklets to schedule it's buttom-half for later (very soon) execution. 
are tasklets like tq_immedate in 2.2.x or tq_schedule? i mean is there a
current process or do they runn at interrupt time?
and am i right when i say: to add a new softirq i need to patch kernel 
sources?

3) i had a look at the ll_rw_block and realised that it can block when there 
are to many buffers locked. when i use generic_make_request can i be 
shure that i wont block so that i can call it in a tasklet and don't need to
switch to a kernel thread? i think that also needs that clustering function 
__make_request may not block. does it or does it not?

4) i was looking at the networking code in 2.4 because it is possible that
i need to write a new thin network protocoll which is optimised for disk-i/o.
i didn't find any documentation how to implement a new one in 2.4. does
anybody have some pointers to doc's or can give me some comments?

thanks for any help or pointers to further information
chris

Re: misc questions about kernel 2.4.x internals

[Daniel Phillips]

On August 20, 2001 04:52 pm, Christian Widmer wrote:
> 3) i had a look at the ll_rw_block and realised that it can block when there 
> are to many buffers locked. when i use generic_make_request can i be 
> shure that i wont block so that i can call it in a tasklet and don't need to
> switch to a kernel thread? i think that also needs that clustering function 
> __make_request may not block. does it or does it not?

It blocks on the availability of a struct request, look at __get_request_wait.
This is how we do IO throttling in 2.4.8/9.

--
Daniel

Re: misc questions about kernel 2.4.x internals

[george anzinger]

Christian Widmer wrote:
> 
> hi,
> 
> 1) when using any functions that can block i need to do this in the context
> of a process. so a can't read, write to sockets in a bottom-half of a
> interrupt handler. thats why i need to use a kernel thread (i don't what to
> use a user level process). my question now is - how long does it take until
> my kernel thread starts running? do i have a way to give it very high
> priority and force my thread to be scheduled so that i can be 'sure' to run
> just after softirq's, tasklets, ...?

You would/ could assign it a real time priority and make it SCHED_FIFO
or SCHED_RR.  If you do this, it is a good idea to make the priority
used available to be "tuned".  Not every one will agree that _your_
handler is as important as you think it is.
> 
> 2) for module writers there is documented and easy to use api how to use
> tasklets to schedule it's buttom-half for later (very soon) execution.
> are tasklets like tq_immedate in 2.2.x or tq_schedule? i mean is there a
> current process or do they runn at interrupt time?
> and am i right when i say: to add a new softirq i need to patch kernel
> sources?
> 
> 3) i had a look at the ll_rw_block and realised that it can block when there
> are to many buffers locked. when i use generic_make_request can i be
> shure that i wont block so that i can call it in a tasklet and don't need to
> switch to a kernel thread? i think that also needs that clustering function
> __make_request may not block. does it or does it not?
> 
> 4) i was looking at the networking code in 2.4 because it is possible that
> i need to write a new thin network protocoll which is optimised for disk-i/o.
> i didn't find any documentation how to implement a new one in 2.4. does
> anybody have some pointers to doc's or can give me some comments?
> 
> thanks for any help or pointers to further information
> chris
> -
> 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: misc questions about kernel 2.4.x internals

[Christian Widmer]

thanks 

> You would/ could assign it a real time priority and make it SCHED_FIFO
> or SCHED_RR.  

do you have references to an example how to to it?

> If you do this, it is a good idea to make the priority
> used available to be "tuned".  Not every one will agree that _your_
> handler is as important as you think it is.

ye i think so - in our cluster it's just fine at the moment :)

chris

Re: misc questions about kernel 2.4.x internals

[george anzinger]

Christian Widmer wrote:
> 
> thanks
> 
> > You would/ could assign it a real time priority and make it SCHED_FIFO
> > or SCHED_RR.
> 
> do you have references to an example how to to it?
> 
> > If you do this, it is a good idea to make the priority
> > used available to be "tuned".  Not every one will agree that _your_
> > handler is as important as you think it is.
> 
> ye i think so - in our cluster it's just fine at the moment :)
> 
> chris
Read the man page on sched_setscheduler(2).  If in the kernel, you would
need to make a system call.  I am afraid I don't know exactly how this
is done, but it is possible.  By the way, some scheduler patches allow
the real time priority to go well above 99, thus if you need this:

#ifndef MAX_PRI
#define MAX_PRI 99
#endif

if you want the maximum priority.

George

Re: misc questions about kernel 2.4.x internals

[Anders Peter Fugmann]

Christian Widmer wrote:
> hi,
> 
> 1) when using any functions that can block i need to do this in the context 
> of a process. so a can't read, write to sockets in a bottom-half of a 
> interrupt handler. thats why i need to use a kernel thread (i don't what to
> use a user level process). my question now is - how long does it take until 
> my kernel thread starts running? do i have a way to give it very high 
> priority and force my thread to be scheduled so that i can be 'sure' to run 
> just after softirq's, tasklets, ...?
> 
> 2) for module writers there is documented and easy to use api how to use 
> tasklets to schedule it's buttom-half for later (very soon) execution. 
> are tasklets like tq_immedate in 2.2.x or tq_schedule? i mean is there a
> current process or do they runn at interrupt time?
> and am i right when i say: to add a new softirq i need to patch kernel 
> source

Tasklets are startet by a softirq, so there is no current process (there 
is, but is random.) I'm not sure that softirq's alwarys run in 
interrupttime.
Do thins of tasklets like BH's that are much more fair to others (not 
blocking all other processes.)

To add a new sorftirq, yes you need to patch kernel source and therfore 
not reccomended (I believe that there are max. 16 softirq's, so its a 
sparse resource.)

> 3) i had a look at the ll_rw_block and realised that it can block when there 
> are to many buffers locked. when i use generic_make_request can i be 
> shure that i wont block so that i can call it in a tasklet and don't need to
> switch to a kernel thread? i think that also needs that clustering function 
> __make_request may not block. does it or does it not?
> 
> 4) i was looking at the networking code in 2.4 because it is possible that
> i need to write a new thin network protocoll which is optimised for disk-i/o.
> i didn't find any documentation how to implement a new one in 2.4. does
> anybody have some pointers to doc's or can give me some comments?
> 
> thanks for any help or pointers to further information
> chris
> -

Regards
Anders Fugmann