Re: PLIP is still broken :(
< said: > This is also something that should be changed since we are to modify > things anyway, it introduces un-needed delays (and jitter) in the > interrupt delivery. Plus, the resource manager has built-in hooks to manage resources which are time-shared by multiple devices. (The thing it's lacking, but ought to have, is a ``gang'' notion to claim several resources at once.) -GAWollman -- Garrett A. Wollman | O Siem / We are all family / O Siem / We're all the same [EMAIL PROTECTED] | O Siem / The fires of freedom Opinions not those of| Dance in the burning flame MIT, LCS, CRS, or NSA| - Susan Aglukark and Chad Irschick To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
In message <[EMAIL PROTECTED]>, Nicolas Souchu writes: >Moreover, in the ppbus model, the ppc_intr() function is _always_ >the registered interrupt handler and it dispatches the interrupt >depending on the device driver which currently owns the bus when >the interrupt occurs. This is also something that should be changed since we are to modify things anyway, it introduces un-needed delays (and jitter) in the interrupt delivery. -- Poul-Henning Kamp FreeBSD coreteam member [EMAIL PROTECTED] "Real hackers run -current on their laptop." FreeBSD -- It will take a long time before progress goes too far! To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
On Mon, Jul 26, 1999 at 05:04:42AM +1000, Bruce Evans wrote:
>
>>>Otherwise,
>>>the generic code is missing mainly update of the interrupt masks when
>>>an interrupt is unregistered.
>>
>>For the low level side, we could consider something like the following code.
>>But this shall be called by the nexus layer and then needs generic newbus
>>support (as you said above, didn't you?).
>
>>/*
>> * Switch an irq from a maskptr to another without unregistering the irq
>> * handler.
>> * This function is supposed to work with only one handler per irq.
>> */
>>void
>>switch_masks(intrmask_t *oldmaskptr, intrmask_t *newmaskptr, int irq)
>
>
>I don't like most of this. Driver level code won't even know the
>correct maskptrs. (irq, maskptr) pairs depend on i386 implementation
>details for uniqueness. Use { s = splhigh(); BUS_TEARDOWN_INTR(...);
>BUS_SETUP_INTR(...); splx(s); } until/unless the newbus level provides
This is what I meant when I said the nexus layer shall do the switch_masks()
call. The current implementation of SETUP_INTR/TEARDOWN involves the overhead
of unregistering and registering the interrupt. This is why I propose
switch_masks() which doesn't.
Moreover, in the ppbus model, the ppc_intr() function is _always_ the registered
interrupt handler and it dispatches the interrupt depending on the device driver
which currently owns the bus when the interrupt occurs.
>a better interface. The problem with the masks not being updated when
>interrupts are unregistered should be fixed in update*_masks().
I agree if unregistering/registering is declared better than switching masks
for the current registered handler.
>
>Bruce
>
--
[EMAIL PROTECTED] / [EMAIL PROTECTED]
FreeBSD - Turning PCs into workstations - http://www.FreeBSD.org
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
>>Otherwise,
>>the generic code is missing mainly update of the interrupt masks when
>>an interrupt is unregistered.
>
>For the low level side, we could consider something like the following code.
>But this shall be called by the nexus layer and then needs generic newbus
>support (as you said above, didn't you?).
>/*
> * Switch an irq from a maskptr to another without unregistering the irq
> * handler.
> * This function is supposed to work with only one handler per irq.
> */
>void
>switch_masks(intrmask_t *oldmaskptr, intrmask_t *newmaskptr, int irq)
...
I don't like most of this. Driver level code won't even know the
correct maskptrs. (irq, maskptr) pairs depend on i386 implementation
details for uniqueness. Use { s = splhigh(); BUS_TEARDOWN_INTR(...);
BUS_SETUP_INTR(...); splx(s); } until/unless the newbus level provides
a better interface. The problem with the masks not being updated when
interrupts are unregistered should be fixed in update*_masks().
Bruce
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
On Sun, Jul 25, 1999 at 09:35:36AM +1000, Bruce Evans wrote:
>
>>>Possible quick fix (hack): change all the spltty()'s in lpt.c to
>>>splnet()'s. lpt isn't a tty driver; it just abuses spltty(). Abusing
>>>splnet() instead should work OK for lpt and fix if_plip.
>>
>>This seems good until the intr stuff handle dynamic update of a interrupt spl.
>>Is there some work in progress on that?
>
>Not much. ppc needs to do most of the work by registering its interrupt
>with the correct interrupt maskptr for the currently attached device.
>This may involve unregistering the interrupt when the device changes.
>The generic code could help here by supporting atomic changing of
>interrupt maskptrs without unregistering the interrupt. Otherwise,
>the generic code is missing mainly update of the interrupt masks when
>an interrupt is unregistered.
For the low level side, we could consider something like the following code.
But this shall be called by the nexus layer and then needs generic newbus
support (as you said above, didn't you?).
/*
* Switch an irq from a maskptr to another without unregistering the irq
* handler.
* This function is supposed to work with only one handler per irq.
*/
void
switch_masks(intrmask_t *oldmaskptr, intrmask_t *newmaskptr, int irq)
{
int s;
intrec *idesc;
intrmask_t mask = 1 << irq;
if ((oldmaskptr == NULL) || (newmaskptr == NULL))
return;
if (((idesc = find_idesc(oldmaskptr, irq)) == NULL) ||
(find_idesc(newmaskptr, irq) != NULL))
return;
/* block all interrupts */
s = splhigh();
/* update the irq mask ptr */
idesc->maskptr = newmaskptr;
/* remove the irq from the old mask and add it to the new one */
INTRUNMASK(*oldmaskptr, mask);
INTRMASK(*newmaskptr, mask);
/* we need to update all values in the intr_mask[irq] array */
update_intr_masks();
/* update mask in chains of the interrupt multiplex handler as well */
update_mux_masks();
/* restore interrupts */
splx(s);
return;
}
Your opinion?
Nicholas.
--
[EMAIL PROTECTED] / [EMAIL PROTECTED]
FreeBSD - Turning PCs into workstations - http://www.FreeBSD.org
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
>>Possible quick fix (hack): change all the spltty()'s in lpt.c to >>splnet()'s. lpt isn't a tty driver; it just abuses spltty(). Abusing >>splnet() instead should work OK for lpt and fix if_plip. > >This seems good until the intr stuff handle dynamic update of a interrupt spl. >Is there some work in progress on that? Not much. ppc needs to do most of the work by registering its interrupt with the correct interrupt maskptr for the currently attached device. This may involve unregistering the interrupt when the device changes. The generic code could help here by supporting atomic changing of interrupt maskptrs without unregistering the interrupt. Otherwise, the generic code is missing mainly update of the interrupt masks when an interrupt is unregistered. Bruce To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
On Mon, Jul 19, 1999 at 07:37:02PM +1000, Bruce Evans wrote: > >>You misunderstood what Bruce wrote. PLIP has always been broken. It >>used to be possible to hack around the brokenness by setting the >>interrupt mask to net instead of tty. With newbus, this hack is no >>longer possible (it was never correct anyway; it broke printing). Or we shall consider changing isa_compat.c if we choose splnet for lpt. > >Or by statically configuring SLIP (which forced tty = net), or maybe >by dynamically configuring PPP. The tty = net hack went away with >old-bus, so SLIP is broken in much the same way as PLIP. > >>The problem with PLIP is that it tries to do splnet stuff in at >>spltty. If you force the parallell port driver to run at splnet, PLIP >>works but you get panics when you print because it tries to do spltty >>stuff at splnet. > >Possible quick fix (hack): change all the spltty()'s in lpt.c to >splnet()'s. lpt isn't a tty driver; it just abuses spltty(). Abusing >splnet() instead should work OK for lpt and fix if_plip. This seems good until the intr stuff handle dynamic update of a interrupt spl. Is there some work in progress on that? > >Bruce > > >To Unsubscribe: send mail to [EMAIL PROTECTED] >with "unsubscribe freebsd-current" in the body of the message > -- [EMAIL PROTECTED] / [EMAIL PROTECTED] FreeBSD - Turning PCs into workstations - http://www.FreeBSD.org To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
On Mon, Jul 19, 1999 at 11:15:24AM +0200, Poul-Henning Kamp wrote: > > >This is actually a deficiency in the ppbus stuff, there is no >telling what SPL level the subdriver wants to use, so the interrupt This is changing. I'm currently working on porting ppbus to newbus. >should actually be released back to the system when no subdrivers What do you exactly call back to the system? Not served? If so, this is the current behaviour of the generic ppbus interrupt dispatcher (eg ppb_intr()) >are open and be grabbed the way the subdriver wants it once it >aquires the bus. But the newbus which relies on the old machdep intr stuff doesn't seem to offer such a service. Doesn't it? Why was the SLIP hack removed then? And I noticed the PPP workaround is still alive in net/ppp_tty.c... Nicholas. -- [EMAIL PROTECTED] / [EMAIL PROTECTED] FreeBSD - Turning PCs into workstations - http://www.FreeBSD.org To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
>> Possible quick fix (hack): change all the spltty()'s in lpt.c to >> splnet()'s. lpt isn't a tty driver; it just abuses spltty(). Abusing >> splnet() instead should work OK for lpt and fix if_plip. > >What about vpo? vpo uses only polled mode. I think its "interrupt" handler is attached to a cam software interrupt handler of timeout handler. In any case, it is not attached to ppcintr(), so ppc's interrupt class doesn't affect it. Bruce To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
Dag-Erling Smorgrav wrote: > Maxim Sobolev <[EMAIL PROTECTED]> writes: > > Bruce Evans wrote: > > > Possible quick fix (hack): change all the spltty()'s in lpt.c to > > > splnet()'s. lpt isn't a tty driver; it just abuses spltty(). Abusing > > > splnet() instead should work OK for lpt and fix if_plip. > > It doesn't help much because I'm not using lpt device in my kernel so > > lpt.c never get compiled in it ;). > > ppc.c But ppc.c doesn't contain spltty()'s in it... -Max To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
Maxim Sobolev <[EMAIL PROTECTED]> writes: > Bruce Evans wrote: > > Possible quick fix (hack): change all the spltty()'s in lpt.c to > > splnet()'s. lpt isn't a tty driver; it just abuses spltty(). Abusing > > splnet() instead should work OK for lpt and fix if_plip. > It doesn't help much because I'm not using lpt device in my kernel so > lpt.c never get compiled in it ;). ppc.c DES -- Dag-Erling Smorgrav - [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
Bruce Evans wrote: > Possible quick fix (hack): change all the spltty()'s in lpt.c to > splnet()'s. lpt isn't a tty driver; it just abuses spltty(). Abusing > splnet() instead should work OK for lpt and fix if_plip. It doesn't help much because I'm not using lpt device in my kernel so lpt.c never get compiled in it ;). Maxim To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
> >The problem with PLIP is that it tries to do splnet stuff in at > >spltty. If you force the parallell port driver to run at splnet, PLIP > >works but you get panics when you print because it tries to do spltty > >stuff at splnet. > > Possible quick fix (hack): change all the spltty()'s in lpt.c to > splnet()'s. lpt isn't a tty driver; it just abuses spltty(). Abusing > splnet() instead should work OK for lpt and fix if_plip. What about vpo? -- \\ The mind's the standard \\ Mike Smith \\ of the man. \\ [EMAIL PROTECTED] \\-- Joseph Merrick \\ [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
>You misunderstood what Bruce wrote. PLIP has always been broken. It >used to be possible to hack around the brokenness by setting the >interrupt mask to net instead of tty. With newbus, this hack is no >longer possible (it was never correct anyway; it broke printing). Or by statically configuring SLIP (which forced tty = net), or maybe by dynamically configuring PPP. The tty = net hack went away with old-bus, so SLIP is broken in much the same way as PLIP. >The problem with PLIP is that it tries to do splnet stuff in at >spltty. If you force the parallell port driver to run at splnet, PLIP >works but you get panics when you print because it tries to do spltty >stuff at splnet. Possible quick fix (hack): change all the spltty()'s in lpt.c to splnet()'s. lpt isn't a tty driver; it just abuses spltty(). Abusing splnet() instead should work OK for lpt and fix if_plip. Bruce To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
This is actually a deficiency in the ppbus stuff, there is no telling what SPL level the subdriver wants to use, so the interrupt should actually be released back to the system when no subdrivers are open and be grabbed the way the subdriver wants it once it aquires the bus. The ZIP driver would probably want splcam/splbio, the pps wants splabsolutelynobody() (and FAST_IRQ) and plip wants splnet(). In message <[EMAIL PROTECTED]>, Dag-Erling Smorgrav writes: >Maxim Sobolev <[EMAIL PROTECTED]> writes: >> Some time ago I >> had rised this question and Bruce Evans told that this problem arise >> because of switch to newbus. > >You misunderstood what Bruce wrote. PLIP has always been broken. It >used to be possible to hack around the brokenness by setting the >interrupt mask to net instead of tty. With newbus, this hack is no >longer possible (it was never correct anyway; it broke printing). > >The problem with PLIP is that it tries to do splnet stuff in at >spltty. If you force the parallell port driver to run at splnet, PLIP >works but you get panics when you print because it tries to do spltty >stuff at splnet. > >SLIP and PPPD do black magic with interrupt masks so spltty and splnet >become essentially equivalen (or so I understand). They do this >because they have the exact same problems as PLIP - they need to do >splnet stuff at spltty. > >DES >-- >Dag-Erling Smorgrav - [EMAIL PROTECTED] > > >To Unsubscribe: send mail to [EMAIL PROTECTED] >with "unsubscribe freebsd-current" in the body of the message > -- Poul-Henning Kamp FreeBSD coreteam member [EMAIL PROTECTED] "Real hackers run -current on their laptop." FreeBSD -- It will take a long time before progress goes too far! To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
Maxim Sobolev <[EMAIL PROTECTED]> writes: > Some time ago I > had rised this question and Bruce Evans told that this problem arise > because of switch to newbus. You misunderstood what Bruce wrote. PLIP has always been broken. It used to be possible to hack around the brokenness by setting the interrupt mask to net instead of tty. With newbus, this hack is no longer possible (it was never correct anyway; it broke printing). The problem with PLIP is that it tries to do splnet stuff in at spltty. If you force the parallell port driver to run at splnet, PLIP works but you get panics when you print because it tries to do spltty stuff at splnet. SLIP and PPPD do black magic with interrupt masks so spltty and splnet become essentially equivalen (or so I understand). They do this because they have the exact same problems as PLIP - they need to do splnet stuff at spltty. DES -- Dag-Erling Smorgrav - [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
> Does anybody have a plans to fix plip code which is broken a quite > awhile (several months or so)? Since I used it just last week on two -CURRENT boxes, I'd say there is some other problem you are experecing. -- -- David([EMAIL PROTECTED] -or- [EMAIL PROTECTED]) To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: PLIP is still broken :(
David O'Brien wrote: > > Does anybody have a plans to fix plip code which is broken a quite > > awhile (several months or so)? > > Since I used it just last week on two -CURRENT boxes, I'd say there is > some other problem you are experecing. I'm also using two -current boxes (P133 and K6-2/300) and when trying to do more or less massive file transfer one of the boxes dying with panic (usually one with a slower processor, but not alvays). Changing ppc flags also doesn't resolve a problem. If you can please try to test it by doing "ping -f -s 8000 other_side" - this test actually bring one side to its knees it several seconds (at lease for me). Some time ago I had rised this question and Bruce Evans told that this problem arise because of switch to newbus. Maybe you can look at his original meassage at: http://www.freebsd.org/cgi/getmsg.cgi?fetch=1692956+1694453+/usr/local/www/db/text/1999/freebsd-current/19990607.freebsd-current -Maxim To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
