Looking for comments on Bottom-Half/Tasklet/SoftIRQ

Looking for comments on Bottom-Half/Tasklet/SoftIRQ

[chuckw]

Greetings,
	I was reading the unreliable guide to kernel hacking and was looking for
a little clarification on something.  2 Bottom halves cannot run at the same
time, why?  
	Also, could someone give me an example of a service which is a bottom half/
tasklet/SoftIRQ?

Thanks in advance,
Chuck

Re: Looking for comments on Bottom-Half/Tasklet/SoftIRQ

[Anders Peter Fugmann]

chuckw@ieee.org wrote:
> Greetings,
> 	I was reading the unreliable guide to kernel hacking and was looking for
> a little clarification on something.  2 Bottom halves cannot run at the same
> time, why? 

Per linux definition of bottom halves, there can only run one buttom 
half at one system wide. But dont use those - They are old and waists 
resources. Try tasklets instead. Multible tasklets can run in parrallel 
(but not the same tasklet)

> 	Also, could someone give me an example of a service which is a bottom half/
> tasklet/SoftIRQ?
Simple.

Imagine some hardware that generates interrupts.
Now we want to write a driver that keeps the hardware busy, so we 
implement a top half handler (IRQ-handler), and let it retrieve som data 
from the hardware. Instead of processing it right away, we shedule a 
tasklet to do that job. This way we can handle more interrupts/sec from 
the card, and the hardware is kept busy.


To summerize.
Buttom halves are the strictest (only one at a time.)
Takslets can run in parralel, but still no need to worry about reentrant 
code.
SoftIrq give no guarrentee at all, and should be used with great care
(code need to be reentrant).

Also try to readLinux device drivers by  A. Rubini:
http://www.xml.com/ldd/chapter/book/index.html

Hope it helps.
Anders Fugmann

> 
> Thanks in advance,
> Chuck
> -
> 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: Looking for comments on Bottom-Half/Tasklet/SoftIRQ

[chuckw]

Thanks.

	So, Bottom halves don't need to be re-entrant as do tasklets.  SoftIRQ's
need to be re-entrant.  The advantage of tasklets is that each tasklet can
be farmed out to different CPU's AND they don't need to be re-entrant 
because only one instance is allowed at a time.  I think I got it.

	Could you direct me to some code in the kernel which uses tasklets
so I can see the inner workings?

Thanks much,
Chuck

On Sun, Aug 19, 2001 at 06:59:22PM +0200, Anders Peter Fugmann wrote:
> 
> chuckw@ieee.org wrote:
> > Greetings,
> > 	I was reading the unreliable guide to kernel hacking and was looking for
> > a little clarification on something.  2 Bottom halves cannot run at the same
> > time, why? 
> 
> Per linux definition of bottom halves, there can only run one buttom 
> half at one system wide. But dont use those - They are old and waists 
> resources. Try tasklets instead. Multible tasklets can run in parrallel 
> (but not the same tasklet)
> 
> > 	Also, could someone give me an example of a service which is a bottom half/
> > tasklet/SoftIRQ?
> Simple.
> 
> Imagine some hardware that generates interrupts.
> Now we want to write a driver that keeps the hardware busy, so we 
> implement a top half handler (IRQ-handler), and let it retrieve som data 
> from the hardware. Instead of processing it right away, we shedule a 
> tasklet to do that job. This way we can handle more interrupts/sec from 
> the card, and the hardware is kept busy.
> 
> 
> To summerize.
> Buttom halves are the strictest (only one at a time.)
> Takslets can run in parralel, but still no need to worry about reentrant 
> code.
> SoftIrq give no guarrentee at all, and should be used with great care
> (code need to be reentrant).
> 
> Also try to readLinux device drivers by  A. Rubini:
> http://www.xml.com/ldd/chapter/book/index.html
> 
> Hope it helps.
> Anders Fugmann
> 
> > 
> > Thanks in advance,
> > Chuck
> > -
> > 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/
> > 
> > 
> 
> 
> -
> 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: Looking for comments on Bottom-Half/Tasklet/SoftIRQ

[Anders Peter Fugmann]

chuckw@ieee.org wrote:
> Thanks
> 
> 	So, Bottom halves don't need to be re-entrant as do tasklets.  SoftIRQ's
> need to be re-entrant.  The advantage of tasklets is that each tasklet can
> be farmed out to different CPU's AND they don't need to be re-entrant 
> because only one instance is allowed at a time.  I think I got it.

That is 100% correct.

> 
> 	Could you direct me to some code in the kernel which uses tasklets
> so I can see the inner workings?

Actually very few systems in the kernel has been rewritten to use 
tasklets instead og BH's.

But as they are very simillar to BH's, you should be able to use the 
same thinking, its just a new API.

Take a look at include/linux/interrupt.h
(or http://lxr.linux.no/source/include/linux/interrupt.h, an invaluable 
source when coding for linux).

Regards
Anders Fugmann

Re: Looking for comments on Bottom-Half/Tasklet/SoftIRQ

[chuckw]

Many thanks once again.

Chuck

On Sun, Aug 19, 2001 at 09:01:13PM +0200, Anders Peter Fugmann wrote:
> chuckw@ieee.org wrote:
> > Thanks
> > 
> > 	So, Bottom halves don't need to be re-entrant as do tasklets.  SoftIRQ's
> > need to be re-entrant.  The advantage of tasklets is that each tasklet can
> > be farmed out to different CPU's AND they don't need to be re-entrant 
> > because only one instance is allowed at a time.  I think I got it.
> 
> That is 100% correct.
> 
> > 
> > 	Could you direct me to some code in the kernel which uses tasklets
> > so I can see the inner workings?
> 
> Actually very few systems in the kernel has been rewritten to use 
> tasklets instead og BH's.
> 
> But as they are very simillar to BH's, you should be able to use the 
> same thinking, its just a new API.
> 
> Take a look at include/linux/interrupt.h
> (or http://lxr.linux.no/source/include/linux/interrupt.h, an invaluable 
> source when coding for linux).
> 
> Regards
> Anders Fugmann
> 

Re: Looking for comments on Bottom-Half/Tasklet/SoftIRQ

[george anzinger]

chuckw@ieee.org wrote:
> 
> Thanks.
> 
>         So, Bottom halves don't need to be re-entrant as do tasklets.  SoftIRQ's
> need to be re-entrant.  The advantage of tasklets is that each tasklet can
> be farmed out to different CPU's AND they don't need to be re-entrant
> because only one instance is allowed at a time.  I think I got it.
> 
>         Could you direct me to some code in the kernel which uses tasklets
> so I can see the inner workings?
> 
> Thanks much,
> Chuck
> 
> On Sun, Aug 19, 2001 at 06:59:22PM +0200, Anders Peter Fugmann wrote:
> >
> > chuckw@ieee.org wrote:
> > > Greetings,
> > >     I was reading the unreliable guide to kernel hacking and was looking for
> > > a little clarification on something.  2 Bottom halves cannot run at the same
> > > time, why?
> >
> > Per linux definition of bottom halves, there can only run one buttom
> > half at one system wide. But dont use those - They are old and waists
> > resources. Try tasklets instead. Multible tasklets can run in parrallel
> > (but not the same tasklet)
> >
> > >     Also, could someone give me an example of a service which is a bottom half/
> > > tasklet/SoftIRQ?
> > Simple.
> >
> > Imagine some hardware that generates interrupts.
> > Now we want to write a driver that keeps the hardware busy, so we
> > implement a top half handler (IRQ-handler), and let it retrieve som data
> > from the hardware. Instead of processing it right away, we shedule a
> > tasklet to do that job. This way we can handle more interrupts/sec from
> > the card, and the hardware is kept busy.
> >
> >
> > To summerize.
> > Buttom halves are the strictest (only one at a time.)
> > Takslets can run in parralel, but still no need to worry about reentrant
> > code.
> > SoftIrq give no guarrentee at all, and should be used with great care
> > (code need to be reentrant).
> >
> > Also try to readLinux device drivers by  A. Rubini:
> > http://www.xml.com/ldd/chapter/book/index.html
> >
> > Hope it helps.
> > Anders Fugmann
> >
> > >

A simple example is the ../kernel/timer.c code.  The "run_task_list()"
function is called from a tasklet.  "do_timer()" is called from
interrupt and "mark_bh(TIMER_BH)" puts the tasklet in the queue. 
"timer_bh()" (old names die hard) is the tasklet.

George

Re: Looking for comments on Bottom-Half/Tasklet/SoftIRQ

[chuckw]

Thanks all for the info.  You have been very helpful.

Many Thanks,
Chuck

On Mon, Aug 20, 2001 at 09:04:28AM -0700, george anzinger wrote:
> chuckw@ieee.org wrote:
> > 
> > Thanks.
> > 
> >         So, Bottom halves don't need to be re-entrant as do tasklets.  SoftIRQ's
> > need to be re-entrant.  The advantage of tasklets is that each tasklet can
> > be farmed out to different CPU's AND they don't need to be re-entrant
> > because only one instance is allowed at a time.  I think I got it.
> > 
> >         Could you direct me to some code in the kernel which uses tasklets
> > so I can see the inner workings?
> > 
> > Thanks much,
> > Chuck
> > 
> > On Sun, Aug 19, 2001 at 06:59:22PM +0200, Anders Peter Fugmann wrote:
> > >
> > > chuckw@ieee.org wrote:
> > > > Greetings,
> > > >     I was reading the unreliable guide to kernel hacking and was looking for
> > > > a little clarification on something.  2 Bottom halves cannot run at the same
> > > > time, why?
> > >
> > > Per linux definition of bottom halves, there can only run one buttom
> > > half at one system wide. But dont use those - They are old and waists
> > > resources. Try tasklets instead. Multible tasklets can run in parrallel
> > > (but not the same tasklet)
> > >
> > > >     Also, could someone give me an example of a service which is a bottom half/
> > > > tasklet/SoftIRQ?
> > > Simple.
> > >
> > > Imagine some hardware that generates interrupts.
> > > Now we want to write a driver that keeps the hardware busy, so we
> > > implement a top half handler (IRQ-handler), and let it retrieve som data
> > > from the hardware. Instead of processing it right away, we shedule a
> > > tasklet to do that job. This way we can handle more interrupts/sec from
> > > the card, and the hardware is kept busy.
> > >
> > >
> > > To summerize.
> > > Buttom halves are the strictest (only one at a time.)
> > > Takslets can run in parralel, but still no need to worry about reentrant
> > > code.
> > > SoftIrq give no guarrentee at all, and should be used with great care
> > > (code need to be reentrant).
> > >
> > > Also try to readLinux device drivers by  A. Rubini:
> > > http://www.xml.com/ldd/chapter/book/index.html
> > >
> > > Hope it helps.
> > > Anders Fugmann
> > >
> > > >
> 
> A simple example is the ../kernel/timer.c code.  The "run_task_list()"
> function is called from a tasklet.  "do_timer()" is called from
> interrupt and "mark_bh(TIMER_BH)" puts the tasklet in the queue. 
> "timer_bh()" (old names die hard) is the tasklet.
> 
> George

Re: Looking for comments on Bottom-Half/Tasklet/SoftIRQ

[Anders Peter Fugmann]

george anzinger wrote:
> 
> A simple example is the ../kernel/timer.c code.  The "run_task_list()"
> function is called from a tasklet.  "do_timer()" is called from
> interrupt and "mark_bh(TIMER_BH)" puts the tasklet in the queue. 
> "timer_bh()" (old names die hard) is the tasklet.
> 
mark_bh is the use of old BH's.
You want to use tasklet_schedule if a tasklet should be sheduled.
(_not_ tasklet_hi_schedule, those are the BH's)

Regards
Anders Fugmann

> George
> -
> 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/
> 
>