Dead & Dying interfaces

Dead & Dying interfaces

[Matthew Wilcox]

We forgot to remove a lot of crap interfaces during 2.5 development.
Let's start a list now so we don't forget during 2.7.

This list is a combination of interfaces which have gone during 2.5 and
interfaces that should go during 2.7.  Think of it as a `updating your
driver/filesystem to sane code' guide.

sleep_on, sleep_on_timeout, interruptible_sleep_on,
	interruptible_sleep_on_timeout
 -> use wait_event interfaces

flush_page_to_ram
 -> Documentation/cachetlb.txt

cli, sti, save_flags
 -> Documentation/cli-sti-removal.txt

pcibios_*
 -> Documentation/pci.txt

virt_to_bus, bus_to_virt
 -> Documentation/DMA-mapping.txt

check_region
 -> Just use request_region and handle the failure

kmap
 -> Use kmap_atomic instead

task queues
 -> work queues

lock_kernel
 -> spinlocks

smp_processor_id
 -> get_cpu

-- 
Revolutions do not require corporate support.

Re: Dead & Dying interfaces

[William Lee Irwin III]

On Fri, Nov 15, 2002 at 06:47:25PM +0000, Matthew Wilcox wrote:
> We forgot to remove a lot of crap interfaces during 2.5 development.
> Let's start a list now so we don't forget during 2.7.
> This list is a combination of interfaces which have gone during 2.5 and
> interfaces that should go during 2.7.  Think of it as a `updating your
> driver/filesystem to sane code' guide.

It's very possible (and in fact bugfixing) to incrementally convert
callers of these interfaces, even during freezes or stable releases.
A bugfix is a bugfix. =)


Cheers,
Bill

Re: Dead & Dying interfaces

[Alan Cox]

On Fri, 2002-11-15 at 18:47, Matthew Wilcox wrote:
> 
> We forgot to remove a lot of crap interfaces during 2.5 development.
> Let's start a list now so we don't forget during 2.7.
> 
> This list is a combination of interfaces which have gone during 2.5 and
> interfaces that should go during 2.7.  Think of it as a `updating your
> driver/filesystem to sane code' guide.
> 
> sleep_on, sleep_on_timeout, interruptible_sleep_on,
> 	interruptible_sleep_on_timeout
>  -> use wait_event interfaces

We need to fix wait_event first, probably to do event variables of some
kind (wait_nonzero_interruptible() etc).

get_cpu is the wrong thing in several places too

Re: Dead & Dying interfaces

[Greg KH]

On Fri, Nov 15, 2002 at 06:47:25PM +0000, Matthew Wilcox wrote:
> 
> pcibios_*
>  -> Documentation/pci.txt

I still have patches that remove all of the instances of this.  I think
it's already in the -ac kernel.  I'll forward port it next week and send
it to Linus.

thanks,

greg k-h

Re: Dead & Dying interfaces

[Alan Cox]

On Sun, 2002-11-17 at 00:41, Greg KH wrote:
> On Fri, Nov 15, 2002 at 06:47:25PM +0000, Matthew Wilcox wrote:
> > 
> > pcibios_*
> >  -> Documentation/pci.txt
> 
> I still have patches that remove all of the instances of this.  I think
> it's already in the -ac kernel.  I'll forward port it next week and send
> it to Linus.

Its in the -ac tree. No problems reported except the pcmcia bug which
you sent a patch for

Re: Dead & Dying interfaces

[Dave Jones]

On Fri, Nov 15, 2002 at 06:47:25PM +0000, Matthew Wilcox wrote:

 > This list is a combination of interfaces which have gone during 2.5 and
 > interfaces that should go during 2.7.  Think of it as a `updating your
 > driver/filesystem to sane code' guide.

Adding printk (KERN_DEBUG "Usage of check_region() is deprecated");
to such interfaces may be an idea. For some of them, however it
is probably a bad idea if the logs get flooded with zillions of warnings
each boot.  Maybe just for the "We really should purge this crap next
time" functions ?

		Dave

-- 
| Dave Jones.        http://www.codemonkey.org.uk
| SuSE Labs

Re: Dead & Dying interfaces

[Adrian Bunk]

On Mon, Nov 18, 2002 at 05:55:35PM +0000, Dave Jones wrote:
> On Fri, Nov 15, 2002 at 06:47:25PM +0000, Matthew Wilcox wrote:
> 
>  > This list is a combination of interfaces which have gone during 2.5 and
>  > interfaces that should go during 2.7.  Think of it as a `updating your
>  > driver/filesystem to sane code' guide.
> 
> Adding printk (KERN_DEBUG "Usage of check_region() is deprecated");
> to such interfaces may be an idea. For some of them, however it
> is probably a bad idea if the logs get flooded with zillions of warnings
> each boot.  Maybe just for the "We really should purge this crap next
> time" functions ?

What about a #warning? With a #warning everyone compiling this code sees
that there's something that needs updating but it doesn't flood the logs
of users (#warning was already used for linux/malloc.h in 2.4).

> 		Dave

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed

Re: Dead & Dying interfaces

[William Lee Irwin III]

On Fri, Nov 15, 2002 at 06:47:25PM +0000, Matthew Wilcox wrote:
> We forgot to remove a lot of crap interfaces during 2.5 development.
> Let's start a list now so we don't forget during 2.7.
> This list is a combination of interfaces which have gone during 2.5 and
> interfaces that should go during 2.7.  Think of it as a `updating your
> driver/filesystem to sane code' guide.

do_each_thread()/for_each_process():
------------------------------------
(1) reserved for special "catastrophic" operations
	e.g. OOM, killall, cap_set_all(), etc.
(2) use for_each_task_pid() or maintain lists/tables of tasks etc. instead
(3) kernel API is missing some pieces to make it avoidable for all callers


Bill