Future of pci-mvebu

Future of pci-mvebu

[Pali Rohár]

Hello, I have just one question. What do you want to do with pci-mvebu
driver? It is already marked as broken for 3 kernel releases and I do
not see any progress from anybody (and you rejected my fixes). How long
do you want it to have marked as broken?

Re: Future of pci-mvebu

[Serge Semin]

To += Mani

* He is now supposed to watch for the DW PCIe drivers.

On Tue, Jul 18, 2023 at 12:03:17AM +0200, Pali Rohár wrote:
> Hello, I have just one question. What do you want to do with pci-mvebu
> driver? It is already marked as broken for 3 kernel releases and I do
> not see any progress from anybody (and you rejected my fixes). How long
> do you want it to have marked as broken?

Re: Future of pci-mvebu

[Bjorn Helgaas]

[+cc Radu]

On Tue, Jul 18, 2023 at 12:03:17AM +0200, Pali Rohár wrote:
> Hello, I have just one question. What do you want to do with pci-mvebu
> driver? It is already marked as broken for 3 kernel releases and I do
> not see any progress from anybody (and you rejected my fixes). How long
> do you want it to have marked as broken?

I don't think "depends on BROKEN" necessarily means that we plan to
remove the driver.  I think it just means that it's currently broken,
but we hope to fix it eventually.

I think the problem here is the regular vs chained interrupt handlers,
right?  Radu has been looking at that recently, too, so maybe we can
have another go at it.

Bjorn

Re: Future of pci-mvebu

[Manivannan Sadhasivam]

On Tue, Jul 18, 2023 at 06:19:52AM -0500, Bjorn Helgaas wrote:
> [+cc Radu]
> 
> On Tue, Jul 18, 2023 at 12:03:17AM +0200, Pali Rohár wrote:
> > Hello, I have just one question. What do you want to do with pci-mvebu
> > driver? It is already marked as broken for 3 kernel releases and I do
> > not see any progress from anybody (and you rejected my fixes). How long
> > do you want it to have marked as broken?
> 
> I don't think "depends on BROKEN" necessarily means that we plan to
> remove the driver.  I think it just means that it's currently broken,
> but we hope to fix it eventually.
> 
> I think the problem here is the regular vs chained interrupt handlers,
> right?  Radu has been looking at that recently, too, so maybe we can
> have another go at it.
> 

We (Linaro and Redhat) had a discussion a while ago on this topic and I just got
pointed to Radu's series. So yes, we should instead work on that direction
instead of driver hacks which brings maintainers disagreement.

- Mani

> Bjorn

-- 
மணிவண்ணன் சதாசிவம்

Re: Future of pci-mvebu

[Manivannan Sadhasivam]

On Tue, Jul 18, 2023 at 05:21:01PM +0530, Manivannan Sadhasivam wrote:
> On Tue, Jul 18, 2023 at 06:19:52AM -0500, Bjorn Helgaas wrote:
> > [+cc Radu]
> > 
> > On Tue, Jul 18, 2023 at 12:03:17AM +0200, Pali Rohár wrote:
> > > Hello, I have just one question. What do you want to do with pci-mvebu
> > > driver? It is already marked as broken for 3 kernel releases and I do
> > > not see any progress from anybody (and you rejected my fixes). How long
> > > do you want it to have marked as broken?
> > 
> > I don't think "depends on BROKEN" necessarily means that we plan to
> > remove the driver.  I think it just means that it's currently broken,
> > but we hope to fix it eventually.
> > 
> > I think the problem here is the regular vs chained interrupt handlers,
> > right?  Radu has been looking at that recently, too, so maybe we can
> > have another go at it.
> > 
> 
> We (Linaro and Redhat) had a discussion a while ago on this topic and I just got
> pointed to Radu's series. So yes, we should instead work on that direction
> instead of driver hacks which brings maintainers disagreement.
> 

FYI: These are the two series/patches that I got pointed to:

https://lore.kernel.org/all/20230530214550.864894-1-rrendec@redhat.com/
https://lore.kernel.org/lkml/20230629183019.1992819-1-rrendec@redhat.com/

- Mani

> - Mani
> 
> > Bjorn
> 
> -- 
> மணிவண்ணன் சதாசிவம்

-- 
மணிவண்ணன் சதாசிவம்

Re: Future of pci-mvebu

[Pali Rohár]

On Tuesday 18 July 2023 06:19:52 Bjorn Helgaas wrote:
> [+cc Radu]
> 
> On Tue, Jul 18, 2023 at 12:03:17AM +0200, Pali Rohár wrote:
> > Hello, I have just one question. What do you want to do with pci-mvebu
> > driver? It is already marked as broken for 3 kernel releases and I do
> > not see any progress from anybody (and you rejected my fixes). How long
> > do you want it to have marked as broken?
> 
> I don't think "depends on BROKEN" necessarily means that we plan to
> remove the driver.  I think it just means that it's currently broken,
> but we hope to fix it eventually.
> 
> I think the problem here is the regular vs chained interrupt handlers,
> right?  Radu has been looking at that recently, too, so maybe we can
> have another go at it.
> 
> Bjorn

I guess that this is the main issue as all other fixes and improvements
are stalled. Just to remind that we have there shared interrupt source
(which can be requested by more HW drivers) and consumer is PCIe INTX
which is de-facto chained handler. And I have not seen anything for
shared interrupt source except request_irq(IRQF_SHARED) which you do not
want to use for shared interrupts anymore. Also setting of PCIe INTX
affinity is broken which worked fine with the previous kernel versions.
And you also rejected fixes for this regression.

Re: Future of pci-mvebu

[Bjorn Helgaas]

On Tue, Jul 18, 2023 at 06:57:03PM +0200, Pali Rohár wrote:
> On Tuesday 18 July 2023 06:19:52 Bjorn Helgaas wrote:
> > On Tue, Jul 18, 2023 at 12:03:17AM +0200, Pali Rohár wrote:
> > > Hello, I have just one question. What do you want to do with pci-mvebu
> > > driver? It is already marked as broken for 3 kernel releases and I do
> > > not see any progress from anybody (and you rejected my fixes). How long
> > > do you want it to have marked as broken?
> > 
> > I don't think "depends on BROKEN" necessarily means that we plan to
> > remove the driver.  I think it just means that it's currently broken,
> > but we hope to fix it eventually.
> > 
> > I think the problem here is the regular vs chained interrupt handlers,
> > right?  Radu has been looking at that recently, too, so maybe we can
> > have another go at it.
> 
> I guess that this is the main issue as all other fixes and improvements
> are stalled. 

If these other things don't depend on this IRQ issue, maybe we could
still make progress on them?  Maybe post them again (rebasing to
v6.5-rc1 if necessary)?  Often I forget or miss things, so it doesn't
hurt to try again if you don't hear anything.

> Just to remind that we have there shared interrupt source
> (which can be requested by more HW drivers) and consumer is PCIe INTX
> which is de-facto chained handler. And I have not seen anything for
> shared interrupt source except request_irq(IRQF_SHARED) which you do not
> want to use for shared interrupts anymore. Also setting of PCIe INTX
> affinity is broken which worked fine with the previous kernel versions.
> And you also rejected fixes for this regression.

This sounds like the issue where we haven't figured out how to make
both Marc and Thomas happy at the same time.  I don't know much about
IRQs, but I'm still optimistic that it may be possible because I don't
think there's anything really unique about mvebu here.

Bjorn

Re: Future of pci-mvebu

[Pali Rohár]

On Tuesday 18 July 2023 14:02:52 Bjorn Helgaas wrote:
> On Tue, Jul 18, 2023 at 06:57:03PM +0200, Pali Rohár wrote:
> > On Tuesday 18 July 2023 06:19:52 Bjorn Helgaas wrote:
> > > On Tue, Jul 18, 2023 at 12:03:17AM +0200, Pali Rohár wrote:
> > > > Hello, I have just one question. What do you want to do with pci-mvebu
> > > > driver? It is already marked as broken for 3 kernel releases and I do
> > > > not see any progress from anybody (and you rejected my fixes). How long
> > > > do you want it to have marked as broken?
> > > 
> > > I don't think "depends on BROKEN" necessarily means that we plan to
> > > remove the driver.  I think it just means that it's currently broken,
> > > but we hope to fix it eventually.
> > > 
> > > I think the problem here is the regular vs chained interrupt handlers,
> > > right?  Radu has been looking at that recently, too, so maybe we can
> > > have another go at it.
> > 
> > I guess that this is the main issue as all other fixes and improvements
> > are stalled. 
> 
> If these other things don't depend on this IRQ issue, maybe we could
> still make progress on them?  Maybe post them again (rebasing to
> v6.5-rc1 if necessary)?  Often I forget or miss things, so it doesn't
> hurt to try again if you don't hear anything.

In past I (re)sent changes at least 3 times. And would not do it again.

> > Just to remind that we have there shared interrupt source
> > (which can be requested by more HW drivers) and consumer is PCIe INTX
> > which is de-facto chained handler. And I have not seen anything for
> > shared interrupt source except request_irq(IRQF_SHARED) which you do not
> > want to use for shared interrupts anymore. Also setting of PCIe INTX
> > affinity is broken which worked fine with the previous kernel versions.
> > And you also rejected fixes for this regression.
> 
> This sounds like the issue where we haven't figured out how to make
> both Marc and Thomas happy at the same time.  I don't know much about
> IRQs, but I'm still optimistic that it may be possible because I don't
> think there's anything really unique about mvebu here.
> 
> Bjorn

At I pointed out, nothing happened for 3 kernel releases and also
nothing happened since my last attempt there. I'm not going to waste
more time on it as I already saw results and I'm not going to repeat
them. Look at the current result - it does not work.

Re: Future of pci-mvebu

[Bjorn Helgaas]

On Tue, Jul 18, 2023 at 09:21:52PM +0200, Pali Rohár wrote:
> On Tuesday 18 July 2023 14:02:52 Bjorn Helgaas wrote:
> > On Tue, Jul 18, 2023 at 06:57:03PM +0200, Pali Rohár wrote:
> > > On Tuesday 18 July 2023 06:19:52 Bjorn Helgaas wrote:
> > > > On Tue, Jul 18, 2023 at 12:03:17AM +0200, Pali Rohár wrote:
> > > > > Hello, I have just one question. What do you want to do with pci-mvebu
> > > > > driver? It is already marked as broken for 3 kernel releases and I do
> > > > > not see any progress from anybody (and you rejected my fixes). How long
> > > > > do you want it to have marked as broken?
> > > > 
> > > > I don't think "depends on BROKEN" necessarily means that we plan to
> > > > remove the driver.  I think it just means that it's currently broken,
> > > > but we hope to fix it eventually.
> > > > 
> > > > I think the problem here is the regular vs chained interrupt handlers,
> > > > right?  Radu has been looking at that recently, too, so maybe we can
> > > > have another go at it.
> > > 
> > > I guess that this is the main issue as all other fixes and improvements
> > > are stalled. 
> > 
> > If these other things don't depend on this IRQ issue, maybe we could
> > still make progress on them?  Maybe post them again (rebasing to
> > v6.5-rc1 if necessary)?  Often I forget or miss things, so it doesn't
> > hurt to try again if you don't hear anything.
> 
> In past I (re)sent changes at least 3 times. And would not do it again.

Ouch, sorry.  How about links to them?

I browsed through the archives, and I did see a couple things that
apparently depended on the IRQ issue, but it wasn't obvious to me what
the others were.

Bjorn

Re: Future of pci-mvebu

[Pali Rohár]

On Tuesday 18 July 2023 14:28:19 Bjorn Helgaas wrote:
> On Tue, Jul 18, 2023 at 09:21:52PM +0200, Pali Rohár wrote:
> > On Tuesday 18 July 2023 14:02:52 Bjorn Helgaas wrote:
> > > On Tue, Jul 18, 2023 at 06:57:03PM +0200, Pali Rohár wrote:
> > > > On Tuesday 18 July 2023 06:19:52 Bjorn Helgaas wrote:
> > > > > On Tue, Jul 18, 2023 at 12:03:17AM +0200, Pali Rohár wrote:
> > > > > > Hello, I have just one question. What do you want to do with pci-mvebu
> > > > > > driver? It is already marked as broken for 3 kernel releases and I do
> > > > > > not see any progress from anybody (and you rejected my fixes). How long
> > > > > > do you want it to have marked as broken?
> > > > > 
> > > > > I don't think "depends on BROKEN" necessarily means that we plan to
> > > > > remove the driver.  I think it just means that it's currently broken,
> > > > > but we hope to fix it eventually.
> > > > > 
> > > > > I think the problem here is the regular vs chained interrupt handlers,
> > > > > right?  Radu has been looking at that recently, too, so maybe we can
> > > > > have another go at it.
> > > > 
> > > > I guess that this is the main issue as all other fixes and improvements
> > > > are stalled. 
> > > 
> > > If these other things don't depend on this IRQ issue, maybe we could
> > > still make progress on them?  Maybe post them again (rebasing to
> > > v6.5-rc1 if necessary)?  Often I forget or miss things, so it doesn't
> > > hurt to try again if you don't hear anything.
> > 
> > In past I (re)sent changes at least 3 times. And would not do it again.
> 
> Ouch, sorry.  How about links to them?

Some of them are in my git tree:
https://git.kernel.org/pub/scm/linux/kernel/git/pali/linux.git/log/?h=pci-mvebu-pending

> I browsed through the archives, and I did see a couple things that
> apparently depended on the IRQ issue, but it wasn't obvious to me what
> the others were.
> 
> Bjorn