[PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Sinan Kaya]

The change introduced in commit 103544d86976 ("ACPI,PCI,IRQ: reduce
resource requirements") removed PCI_USING penalty from
acpi_pci_link_allocate function as there is no longer a fixed size penalty
array for both PCI interrupts greater than 16.

The array size has been reduced to 16 and array name got prefixed as ISA
since it only is accountable for the ISA interrupts.

The original change in commit 103544d86976 ("ACPI,PCI,IRQ: reduce
resource requirements") removed penalty assignment in the code for PCI
thinking that we will add the penalty later in acpi_irq_pci_sharing_penalty
function.

However, this function only gets called if the IRQ number is greater than
16 and acpi_irq_get_penalty function gets called before ACPI start in
acpi_isa_irq_available and acpi_penalize_isa_irq functions. We can't rely
on iterating the link list.

We need to add the PCI_USING penalty for ISA interrupts too if the link is
in use and matches our ISA IRQ number.

Signed-off-by: Sinan Kaya <okaya@codeaurora.org>
---
 drivers/acpi/pci_link.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/acpi/pci_link.c b/drivers/acpi/pci_link.c
index c983bf7..a212709 100644
--- a/drivers/acpi/pci_link.c
+++ b/drivers/acpi/pci_link.c
@@ -619,6 +619,10 @@ static int acpi_pci_link_allocate(struct acpi_pci_link *link)
 			    acpi_device_bid(link->device));
 		return -ENODEV;
 	} else {
+		if (link->irq.active < ACPI_MAX_ISA_IRQS)
+			acpi_isa_irq_penalty[link->irq.active] +=
+				PIRQ_PENALTY_PCI_USING;
+
 		printk(KERN_WARNING PREFIX "%s [%s] enabled at IRQ %d\n",
 		       acpi_device_name(link->device),
 		       acpi_device_bid(link->device), link->irq.active);
-- 
1.9.1

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Rafael J. Wysocki]

On Sat, Oct 15, 2016 at 6:31 AM, Sinan Kaya <okaya@codeaurora.org> wrote:
> The change introduced in commit 103544d86976 ("ACPI,PCI,IRQ: reduce
> resource requirements") removed PCI_USING penalty from
> acpi_pci_link_allocate function as there is no longer a fixed size penalty
> array for both PCI interrupts greater than 16.
>
> The array size has been reduced to 16 and array name got prefixed as ISA
> since it only is accountable for the ISA interrupts.
>
> The original change in commit 103544d86976 ("ACPI,PCI,IRQ: reduce
> resource requirements") removed penalty assignment in the code for PCI
> thinking that we will add the penalty later in acpi_irq_pci_sharing_penalty
> function.

I'd write the above this way:

"Commit 103544d86976 (ACPI,PCI,IRQ: reduce resource requirements)
dropped the PCI_USING penalty from acpi_pci_link_allocate() with the
assumption that the penalty will be added later in
acpi_irq_pci_sharing_penalty()."

This conveys essentially the same information (up to some irrelevant
bits), but in a clearer way IMO.

>
> However, this function only gets called if the IRQ number is greater than
> 16 and acpi_irq_get_penalty function gets called before ACPI start in
> acpi_isa_irq_available and acpi_penalize_isa_irq functions. We can't rely
> on iterating the link list.

"However, acpi_irq_pci_sharing_penalty() is only called for IRQ
numbers above 15.  Moreover, acpi_irq_get_penalty() is invoked by
acpi_isa_irq_available() and acpi_penalize_isa_irq() before the ACPI
initialization and the PCI interrupt links list is not ready at that
point, so it cannot be relied on when computing the penalty."

>
> We need to add the PCI_USING penalty for ISA interrupts too if the link is
> in use and matches our ISA IRQ number.

"For this reason, the PCI_USING penalty has to be added in
acpi_pci_link_allocate() directly if the link has been enabled
successfully and the IRQ number is within the ISA range."

IIUC

>
> Signed-off-by: Sinan Kaya <okaya@codeaurora.org>
> ---
>  drivers/acpi/pci_link.c | 4 ++++
>  1 file changed, 4 insertions(+)
>
> diff --git a/drivers/acpi/pci_link.c b/drivers/acpi/pci_link.c
> index c983bf7..a212709 100644
> --- a/drivers/acpi/pci_link.c
> +++ b/drivers/acpi/pci_link.c
> @@ -619,6 +619,10 @@ static int acpi_pci_link_allocate(struct acpi_pci_link *link)
>                             acpi_device_bid(link->device));
>                 return -ENODEV;
>         } else {
> +               if (link->irq.active < ACPI_MAX_ISA_IRQS)
> +                       acpi_isa_irq_penalty[link->irq.active] +=
> +                               PIRQ_PENALTY_PCI_USING;
> +

There's no need to break the line here and I would put the above after
the printk().

Or even after the whole "else" branch (which is unnecessary, but let's
limit changes in this patch).

>                 printk(KERN_WARNING PREFIX "%s [%s] enabled at IRQ %d\n",
>                        acpi_device_name(link->device),
>                        acpi_device_bid(link->device), link->irq.active);
> --

Thanks,
Rafael

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Sinan Kaya]

Hi Rafael,

On 10/15/2016 8:39 AM, Rafael J. Wysocki wrote:
> On Sat, Oct 15, 2016 at 6:31 AM, Sinan Kaya <okaya@codeaurora.org> wrote:
>> The change introduced in commit 103544d86976 ("ACPI,PCI,IRQ: reduce
>> resource requirements") removed PCI_USING penalty from
>> acpi_pci_link_allocate function as there is no longer a fixed size penalty
>> array for both PCI interrupts greater than 16.
>>
>> The array size has been reduced to 16 and array name got prefixed as ISA
>> since it only is accountable for the ISA interrupts.
>>
>> The original change in commit 103544d86976 ("ACPI,PCI,IRQ: reduce
>> resource requirements") removed penalty assignment in the code for PCI
>> thinking that we will add the penalty later in acpi_irq_pci_sharing_penalty
>> function.
> 
> I'd write the above this way:
> 
> "Commit 103544d86976 (ACPI,PCI,IRQ: reduce resource requirements)
> dropped the PCI_USING penalty from acpi_pci_link_allocate() with the
> assumption that the penalty will be added later in
> acpi_irq_pci_sharing_penalty()."
> 
> This conveys essentially the same information (up to some irrelevant
> bits), but in a clearer way IMO.
> 
>>
>> However, this function only gets called if the IRQ number is greater than
>> 16 and acpi_irq_get_penalty function gets called before ACPI start in
>> acpi_isa_irq_available and acpi_penalize_isa_irq functions. We can't rely
>> on iterating the link list.
> 
> "However, acpi_irq_pci_sharing_penalty() is only called for IRQ
> numbers above 15.  Moreover, acpi_irq_get_penalty() is invoked by
> acpi_isa_irq_available() and acpi_penalize_isa_irq() before the ACPI
> initialization and the PCI interrupt links list is not ready at that
> point, so it cannot be relied on when computing the penalty."
> 
>>
>> We need to add the PCI_USING penalty for ISA interrupts too if the link is
>> in use and matches our ISA IRQ number.
> 
> "For this reason, the PCI_USING penalty has to be added in
> acpi_pci_link_allocate() directly if the link has been enabled
> successfully and the IRQ number is within the ISA range."
> 
> IIUC
> 
>>
>> Signed-off-by: Sinan Kaya <okaya@codeaurora.org>
>> ---
>>  drivers/acpi/pci_link.c | 4 ++++
>>  1 file changed, 4 insertions(+)
>>
>> diff --git a/drivers/acpi/pci_link.c b/drivers/acpi/pci_link.c
>> index c983bf7..a212709 100644
>> --- a/drivers/acpi/pci_link.c
>> +++ b/drivers/acpi/pci_link.c
>> @@ -619,6 +619,10 @@ static int acpi_pci_link_allocate(struct acpi_pci_link *link)
>>                             acpi_device_bid(link->device));
>>                 return -ENODEV;
>>         } else {
>> +               if (link->irq.active < ACPI_MAX_ISA_IRQS)
>> +                       acpi_isa_irq_penalty[link->irq.active] +=
>> +                               PIRQ_PENALTY_PCI_USING;
>> +
> 
> There's no need to break the line here and I would put the above after
> the printk().
> 
> Or even after the whole "else" branch (which is unnecessary, but let's
> limit changes in this patch).
> 
>>                 printk(KERN_WARNING PREFIX "%s [%s] enabled at IRQ %d\n",
>>                        acpi_device_name(link->device),
>>                        acpi_device_bid(link->device), link->irq.active);
>> --
> 

Thanks for the feedback. I can resubmit with the comments corrected. I'll wait
until I hear from Bjorn first.

> Thanks,
> Rafael
> 


-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Bjorn Helgaas]

On Sat, Oct 15, 2016 at 12:31:05AM -0400, Sinan Kaya wrote:
> The change introduced in commit 103544d86976 ("ACPI,PCI,IRQ: reduce
> resource requirements") removed PCI_USING penalty from
> acpi_pci_link_allocate function as there is no longer a fixed size penalty
> array for both PCI interrupts greater than 16.
> 
> The array size has been reduced to 16 and array name got prefixed as ISA
> since it only is accountable for the ISA interrupts.
> 
> The original change in commit 103544d86976 ("ACPI,PCI,IRQ: reduce
> resource requirements") removed penalty assignment in the code for PCI
> thinking that we will add the penalty later in acpi_irq_pci_sharing_penalty
> function.
> 
> However, this function only gets called if the IRQ number is greater than
> 16 and acpi_irq_get_penalty function gets called before ACPI start in
> acpi_isa_irq_available and acpi_penalize_isa_irq functions. We can't rely
> on iterating the link list.

It seems wrong to me that we call acpi_irq_get_penalty() from
acpi_irq_penalty_update() and acpi_penalize_isa_irq().  It seems like they
should just manipulate acpi_isa_irq_penalty[irq] directly.

acpi_irq_penalty_update() is for command-line parameters, so it certainly
doesn't need the acpi_irq_pci_sharing_penalty() information (the
acpi_link_list should be empty at the time we process the command-line
parameters).

acpi_penalize_isa_irq() is telling us that a PNP or ACPI device is using
the IRQ -- this should modify the IRQ's penalty, but it shouldn't depend on
the acpi_irq_pci_sharing_penalty() value at all.

> We need to add the PCI_USING penalty for ISA interrupts too if the link is
> in use and matches our ISA IRQ number.
> 
> Signed-off-by: Sinan Kaya <okaya@codeaurora.org>
> ---
>  drivers/acpi/pci_link.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/drivers/acpi/pci_link.c b/drivers/acpi/pci_link.c
> index c983bf7..a212709 100644
> --- a/drivers/acpi/pci_link.c
> +++ b/drivers/acpi/pci_link.c
> @@ -619,6 +619,10 @@ static int acpi_pci_link_allocate(struct acpi_pci_link *link)
>  			    acpi_device_bid(link->device));
>  		return -ENODEV;
>  	} else {
> +		if (link->irq.active < ACPI_MAX_ISA_IRQS)
> +			acpi_isa_irq_penalty[link->irq.active] +=
> +				PIRQ_PENALTY_PCI_USING;
> +
>  		printk(KERN_WARNING PREFIX "%s [%s] enabled at IRQ %d\n",
>  		       acpi_device_name(link->device),
>  		       acpi_device_bid(link->device), link->irq.active);
> -- 
> 1.9.1
> 
> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Sinan Kaya]

On 10/18/2016 6:59 AM, Bjorn Helgaas wrote:
>> However, this function only gets called if the IRQ number is greater than
>> > 16 and acpi_irq_get_penalty function gets called before ACPI start in
>> > acpi_isa_irq_available and acpi_penalize_isa_irq functions. We can't rely
>> > on iterating the link list.

Maybe, I am missing context here. I can add this paragraph to the commit. 

When we started cleaning the code we got rid of the acpi_irq_penalty_init function
in favor of acpi_irq_pci_sharing_penalty function as it does have some fair
amount of code duplication.

I tried putting back the acpi_irq_pci_sharing_penalty function into the ISA
IRQ path again during the debug and the machine died way too early. We couldn't
collect any debug message.

This is telling me that we can't even iterate the link list when these two API
is called. ISA IRQ need to be handled with special care due to calling order.


> It seems wrong to me that we call acpi_irq_get_penalty() from
> acpi_irq_penalty_update() and acpi_penalize_isa_irq().  It seems like they
> should just manipulate acpi_isa_irq_penalty[irq] directly.
> 
> acpi_irq_penalty_update() is for command-line parameters, so it certainly
> doesn't need the acpi_irq_pci_sharing_penalty() information (the
> acpi_link_list should be empty at the time we process the command-line
> parameters).
> 
> acpi_penalize_isa_irq() is telling us that a PNP or ACPI device is using
> the IRQ -- this should modify the IRQ's penalty, but it shouldn't depend on
> the acpi_irq_pci_sharing_penalty() value at all.
> 

acpi_irq_get_penalty function knows how to deal with ISA IRQ. So, it is harmless
to call it. Also, reading the acpi_isa_irq_penalty array directly isn't also right.
It doesn't contain the SCI penalty. So, it returns incorrect penalty value.

The rule of thumb is:
- all PCI/SCI penalty reads need to go through acpi_isa_irq_penalty function
- all ISA penalty writes need to go through acpi_isa_irq_penalty array directly.
- we do not support modifying the PCI IRQ penalties greater than the ISA IRQ numbers.
The original code supported this.


-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Sinan Kaya]

Sorry, I think I didn't have enough morning coffee.

Looking at these again and trying to be specific.

On 10/18/2016 8:20 AM, Sinan Kaya wrote:
> It seems wrong to me that we call acpi_irq_get_penalty() from
>> acpi_irq_penalty_update() and acpi_penalize_isa_irq().  It seems like they
>> should just manipulate acpi_isa_irq_penalty[irq] directly.
>> 
>> acpi_irq_penalty_update() is for command-line parameters, so it certainly
>> doesn't need the acpi_irq_pci_sharing_penalty() information (the
>> acpi_link_list should be empty at the time we process the command-line
>> parameters).

Calling acpi_irq_get_penalty for ISA IRQ is OK as long as it doesn't have
any dynamic IRQ calculation such that acpi_isa_irq_penalty[irq] = acpi_irq_get_penalty.

If this is broken, then we need special care so that we don't assign
dynamically calcualted sci_penalty back to acpi_isa_irq_penalty[irq]. This
results in returning incorrect penalty as

acpi_irq_get_penalty = acpi_isa_irq_original_penalty[irq] + 2 * sci_penalty.

Now that we added sci_penalty into the acpi_irq_get_penalty function,
calling acpi_irq_get_penalty is not correct anymore. This line here needs to
be replaced with acpi_isa_irq_penalty[irq] as you suggested.

                if (used)
                        new_penalty = acpi_irq_get_penalty(irq) +
                                        PIRQ_PENALTY_ISA_USED;
                else
                        new_penalty = 0;

                acpi_isa_irq_penalty[irq] = new_penalty;


>> 
>> acpi_penalize_isa_irq() is telling us that a PNP or ACPI device is using
>> the IRQ -- this should modify the IRQ's penalty, but it shouldn't depend on
>> the acpi_irq_pci_sharing_penalty() value at all.
>> 

Same problem here. This line will be broken after the sci_penalty change.

                acpi_isa_irq_penalty[irq] = acpi_irq_get_penalty(irq) +
                  (active ? PIRQ_PENALTY_ISA_USED : PIRQ_PENALTY_PCI_USING);




-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Bjorn Helgaas]

On Tue, Oct 18, 2016 at 08:32:44AM -0700, Sinan Kaya wrote:
> Sorry, I think I didn't have enough morning coffee.
> 
> Looking at these again and trying to be specific.
> 
> On 10/18/2016 8:20 AM, Sinan Kaya wrote:
> > It seems wrong to me that we call acpi_irq_get_penalty() from
> >> acpi_irq_penalty_update() and acpi_penalize_isa_irq().  It seems like they
> >> should just manipulate acpi_isa_irq_penalty[irq] directly.
> >> 
> >> acpi_irq_penalty_update() is for command-line parameters, so it certainly
> >> doesn't need the acpi_irq_pci_sharing_penalty() information (the
> >> acpi_link_list should be empty at the time we process the command-line
> >> parameters).
> 
> Calling acpi_irq_get_penalty for ISA IRQ is OK as long as it doesn't have
> any dynamic IRQ calculation such that acpi_isa_irq_penalty[irq] = acpi_irq_get_penalty.
> 
> If this is broken, then we need special care so that we don't assign
> dynamically calcualted sci_penalty back to acpi_isa_irq_penalty[irq]. This
> results in returning incorrect penalty as
> 
> acpi_irq_get_penalty = acpi_isa_irq_original_penalty[irq] + 2 * sci_penalty.
> 
> Now that we added sci_penalty into the acpi_irq_get_penalty function,
> calling acpi_irq_get_penalty is not correct anymore. This line here needs to
> be replaced with acpi_isa_irq_penalty[irq] as you suggested.
> 
>                 if (used)
>                         new_penalty = acpi_irq_get_penalty(irq) +
>                                         PIRQ_PENALTY_ISA_USED;
>                 else
>                         new_penalty = 0;
> 
>                 acpi_isa_irq_penalty[irq] = new_penalty;
> 
> 
> >> 
> >> acpi_penalize_isa_irq() is telling us that a PNP or ACPI device is using
> >> the IRQ -- this should modify the IRQ's penalty, but it shouldn't depend on
> >> the acpi_irq_pci_sharing_penalty() value at all.
> >> 
> 
> Same problem here. This line will be broken after the sci_penalty change.
> 
>                 acpi_isa_irq_penalty[irq] = acpi_irq_get_penalty(irq) +
>                   (active ? PIRQ_PENALTY_ISA_USED : PIRQ_PENALTY_PCI_USING);

I think the fragility of this code is an indication that we have a
design problem, so I want to step back from the nitty gritty details
for a bit and look at the overall design.

Let me restate the overall problem: We have a PCI device connected to
an interrupt link.  The interrupt link can be connected to one of
several IRQs, and we want to choose one of those IRQs to minimize IRQ
sharing.

That means we need information about which IRQs are used.
Historically we've started with a compiled-in table of common ISA IRQ
usage, and we also collect information about which IRQs are used and
which *might* be used.  So we have the following inputs:

  - Compiled-in ISA IRQ usage: the static acpi_isa_irq_penalty[]
    values.  ACPI is *supposed* to tell us about all these usages, so
    I don't know why we have the table.  But it's been there as long
    as I can remember.  The table is probably x86-specific, but we
    keep it in the supposedly generic pci_link.c.

  - The "acpi_irq_isa=" and "acpi_irq_pci=" command-line overrides via
    acpi_irq_pci().  I suppose these are for cases where we can't
    figure things out automatically.  I would resist adding parameters
    like this today (I would treat the need for them as a bug and look
    for a fix or a quirk), but we might be stuck with these.

  - SCI information from the ACPI FADT (acpi_penalize_sci_irq()).

  - PNPBIOS and PNPACPI device IRQ usage from _CRS and _PRS via
    acpi_penalize_isa_irq().  This is only for IRQs 0-15, and it does
    NOT include interrupt link (PNP0C0F) devices because we don't
    handle them as PNPACPI devices.  I think this is related to the
    fact that PNP0C0F doesn't appear in acpi_pnp_device_ids[].

  - For non-PNP0C0F, non-PNPACPI devices, we completely ignore IRQ
    information from _CRS and _PRS.  This seems sub-optimal and
    possibly buggy.

  - Interrupt link (PNP0C0F) IRQ usage from _CRS and _PRS via
    acpi_irq_penalty_init().  This is only for IRQs 0-15, and we only
    call this on x86.  If _PRS exists, we penalize each possible IRQ.
    If there's no _PRS but _CRS contains an active IRQ, we penalize
    it.

  - Interrupt link (PNP0C0F) IRQ usage from _CRS and _PRS when
    enabling a new link.  In acpi_irq_pci_sharing_penalty(), we
    penalize an IRQ if it appears in _CRS or _PRS of any link device
    in the system.
    
    For IRQs 0-15, this overlaps with the penalization done at
    boot-time by acpi_irq_penalty_init(): if a device has _PRS, we'll
    add the "possible" penalty twice (once in acpi_irq_penalty_init()
    and again in acpi_irq_pci_sharing_penalty()), and the "using"
    penalty once (in acpi_irq_pci_sharing_penalty()).

    If a device has no _PRS but has _CRS, the "using" penalty is
    applied twice (once in once in acpi_irq_penalty_init() and again
    in acpi_irq_pci_sharing_penalty())

I think this whole thing is baroque and grotesque.

Here's a strawman idea:

  - Maintain a mapping of (IRQ, penalty).  Initially all penalties are
    zero.  This is for *all* IRQs, not just ISA ones.  This could be a
    linked list, but the structure is not important as long as we can
    add things dynamically.

  - Add a "acpi_penalize_irq()" function similar to
    acpi_penalize_isa_irq(), but not restricted to ISA, so we can
    increase the penalty for any IRQ, and maybe we can specify how
    much penalty to add.

  - Make acpi_irq_get_penalty() a simple lookup in the mapping.  No
    iterating through all link devices.

  - If we think the compiled-in penalties are really necessary, move
    the table to x86 code and add a boot-time loop to use
    acpi_penalize_irq() to penalize these IRQs.  Same for the
    command-line options.

  - Change acpi_penalize_sci_irq() to use acpi_penalize_irq().
    Probably the mapping needs to pay attention to trigger/polarity
    somehow, too.

  - Figure out how to make the ACPI core use acpi_penalize_irq() to
    based on the _CRS and _PRS of every ACPI device, including
    PNPACPI, PNP0C0F, etc.  Then we can remove acpi_irq_penalty_init().

  - Change acpi_pci_link_set() to use acpi_penalize_irq() for the IRQ
    it is enabling.  Conceptually maybe this should be done in the
    acpi_set_current_resources() path so it happens whenever we use
    _CRS to enable an IRQ on *any* ACPI device.

I think the biggest issue is figuring out how to get the ACPI core to
look at the _CRS for *all* devices.  If we could do that, I think it
could substantially simplify this code.

Bjorn

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Rafael J. Wysocki]

On Thu, Oct 20, 2016 at 12:44 AM, Bjorn Helgaas <helgaas@kernel.org> wrote:
> On Tue, Oct 18, 2016 at 08:32:44AM -0700, Sinan Kaya wrote:
>> Sorry, I think I didn't have enough morning coffee.
>>
>> Looking at these again and trying to be specific.
>>
>> On 10/18/2016 8:20 AM, Sinan Kaya wrote:
>> > It seems wrong to me that we call acpi_irq_get_penalty() from
>> >> acpi_irq_penalty_update() and acpi_penalize_isa_irq().  It seems like they
>> >> should just manipulate acpi_isa_irq_penalty[irq] directly.
>> >>
>> >> acpi_irq_penalty_update() is for command-line parameters, so it certainly
>> >> doesn't need the acpi_irq_pci_sharing_penalty() information (the
>> >> acpi_link_list should be empty at the time we process the command-line
>> >> parameters).
>>
>> Calling acpi_irq_get_penalty for ISA IRQ is OK as long as it doesn't have
>> any dynamic IRQ calculation such that acpi_isa_irq_penalty[irq] = acpi_irq_get_penalty.
>>
>> If this is broken, then we need special care so that we don't assign
>> dynamically calcualted sci_penalty back to acpi_isa_irq_penalty[irq]. This
>> results in returning incorrect penalty as
>>
>> acpi_irq_get_penalty = acpi_isa_irq_original_penalty[irq] + 2 * sci_penalty.
>>
>> Now that we added sci_penalty into the acpi_irq_get_penalty function,
>> calling acpi_irq_get_penalty is not correct anymore. This line here needs to
>> be replaced with acpi_isa_irq_penalty[irq] as you suggested.
>>
>>                 if (used)
>>                         new_penalty = acpi_irq_get_penalty(irq) +
>>                                         PIRQ_PENALTY_ISA_USED;
>>                 else
>>                         new_penalty = 0;
>>
>>                 acpi_isa_irq_penalty[irq] = new_penalty;
>>
>>
>> >>
>> >> acpi_penalize_isa_irq() is telling us that a PNP or ACPI device is using
>> >> the IRQ -- this should modify the IRQ's penalty, but it shouldn't depend on
>> >> the acpi_irq_pci_sharing_penalty() value at all.
>> >>
>>
>> Same problem here. This line will be broken after the sci_penalty change.
>>
>>                 acpi_isa_irq_penalty[irq] = acpi_irq_get_penalty(irq) +
>>                   (active ? PIRQ_PENALTY_ISA_USED : PIRQ_PENALTY_PCI_USING);
>
> I think the fragility of this code is an indication that we have a
> design problem, so I want to step back from the nitty gritty details
> for a bit and look at the overall design.
>
> Let me restate the overall problem: We have a PCI device connected to
> an interrupt link.  The interrupt link can be connected to one of
> several IRQs, and we want to choose one of those IRQs to minimize IRQ
> sharing.
>
> That means we need information about which IRQs are used.
> Historically we've started with a compiled-in table of common ISA IRQ
> usage, and we also collect information about which IRQs are used and
> which *might* be used.  So we have the following inputs:
>
>   - Compiled-in ISA IRQ usage: the static acpi_isa_irq_penalty[]
>     values.  ACPI is *supposed* to tell us about all these usages, so
>     I don't know why we have the table.  But it's been there as long
>     as I can remember.  The table is probably x86-specific, but we
>     keep it in the supposedly generic pci_link.c.
>
>   - The "acpi_irq_isa=" and "acpi_irq_pci=" command-line overrides via
>     acpi_irq_pci().  I suppose these are for cases where we can't
>     figure things out automatically.  I would resist adding parameters
>     like this today (I would treat the need for them as a bug and look
>     for a fix or a quirk), but we might be stuck with these.
>
>   - SCI information from the ACPI FADT (acpi_penalize_sci_irq()).
>
>   - PNPBIOS and PNPACPI device IRQ usage from _CRS and _PRS via
>     acpi_penalize_isa_irq().  This is only for IRQs 0-15, and it does
>     NOT include interrupt link (PNP0C0F) devices because we don't
>     handle them as PNPACPI devices.  I think this is related to the
>     fact that PNP0C0F doesn't appear in acpi_pnp_device_ids[].
>
>   - For non-PNP0C0F, non-PNPACPI devices, we completely ignore IRQ
>     information from _CRS and _PRS.  This seems sub-optimal and
>     possibly buggy.
>
>   - Interrupt link (PNP0C0F) IRQ usage from _CRS and _PRS via
>     acpi_irq_penalty_init().  This is only for IRQs 0-15, and we only
>     call this on x86.  If _PRS exists, we penalize each possible IRQ.
>     If there's no _PRS but _CRS contains an active IRQ, we penalize
>     it.
>
>   - Interrupt link (PNP0C0F) IRQ usage from _CRS and _PRS when
>     enabling a new link.  In acpi_irq_pci_sharing_penalty(), we
>     penalize an IRQ if it appears in _CRS or _PRS of any link device
>     in the system.
>
>     For IRQs 0-15, this overlaps with the penalization done at
>     boot-time by acpi_irq_penalty_init(): if a device has _PRS, we'll
>     add the "possible" penalty twice (once in acpi_irq_penalty_init()
>     and again in acpi_irq_pci_sharing_penalty()), and the "using"
>     penalty once (in acpi_irq_pci_sharing_penalty()).
>
>     If a device has no _PRS but has _CRS, the "using" penalty is
>     applied twice (once in once in acpi_irq_penalty_init() and again
>     in acpi_irq_pci_sharing_penalty())
>
> I think this whole thing is baroque and grotesque.

While I agree, I also would like the regression introduced here to be
fixed ASAP.

So do you want me to revert all of the changes made here so far and start over?

Thanks,
Rafael

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Bjorn Helgaas]

On Thu, Oct 20, 2016 at 01:17:23AM +0200, Rafael J. Wysocki wrote:
> On Thu, Oct 20, 2016 at 12:44 AM, Bjorn Helgaas <helgaas@kernel.org> wrote:
> > On Tue, Oct 18, 2016 at 08:32:44AM -0700, Sinan Kaya wrote:
> >> Sorry, I think I didn't have enough morning coffee.
> >>
> >> Looking at these again and trying to be specific.
> >>
> >> On 10/18/2016 8:20 AM, Sinan Kaya wrote:
> >> > It seems wrong to me that we call acpi_irq_get_penalty() from
> >> >> acpi_irq_penalty_update() and acpi_penalize_isa_irq().  It seems like they
> >> >> should just manipulate acpi_isa_irq_penalty[irq] directly.
> >> >>
> >> >> acpi_irq_penalty_update() is for command-line parameters, so it certainly
> >> >> doesn't need the acpi_irq_pci_sharing_penalty() information (the
> >> >> acpi_link_list should be empty at the time we process the command-line
> >> >> parameters).
> >>
> >> Calling acpi_irq_get_penalty for ISA IRQ is OK as long as it doesn't have
> >> any dynamic IRQ calculation such that acpi_isa_irq_penalty[irq] = acpi_irq_get_penalty.
> >>
> >> If this is broken, then we need special care so that we don't assign
> >> dynamically calcualted sci_penalty back to acpi_isa_irq_penalty[irq]. This
> >> results in returning incorrect penalty as
> >>
> >> acpi_irq_get_penalty = acpi_isa_irq_original_penalty[irq] + 2 * sci_penalty.
> >>
> >> Now that we added sci_penalty into the acpi_irq_get_penalty function,
> >> calling acpi_irq_get_penalty is not correct anymore. This line here needs to
> >> be replaced with acpi_isa_irq_penalty[irq] as you suggested.
> >>
> >>                 if (used)
> >>                         new_penalty = acpi_irq_get_penalty(irq) +
> >>                                         PIRQ_PENALTY_ISA_USED;
> >>                 else
> >>                         new_penalty = 0;
> >>
> >>                 acpi_isa_irq_penalty[irq] = new_penalty;
> >>
> >>
> >> >>
> >> >> acpi_penalize_isa_irq() is telling us that a PNP or ACPI device is using
> >> >> the IRQ -- this should modify the IRQ's penalty, but it shouldn't depend on
> >> >> the acpi_irq_pci_sharing_penalty() value at all.
> >> >>
> >>
> >> Same problem here. This line will be broken after the sci_penalty change.
> >>
> >>                 acpi_isa_irq_penalty[irq] = acpi_irq_get_penalty(irq) +
> >>                   (active ? PIRQ_PENALTY_ISA_USED : PIRQ_PENALTY_PCI_USING);
> >
> > I think the fragility of this code is an indication that we have a
> > design problem, so I want to step back from the nitty gritty details
> > for a bit and look at the overall design.
> >
> > Let me restate the overall problem: We have a PCI device connected to
> > an interrupt link.  The interrupt link can be connected to one of
> > several IRQs, and we want to choose one of those IRQs to minimize IRQ
> > sharing.
> >
> > That means we need information about which IRQs are used.
> > Historically we've started with a compiled-in table of common ISA IRQ
> > usage, and we also collect information about which IRQs are used and
> > which *might* be used.  So we have the following inputs:
> >
> >   - Compiled-in ISA IRQ usage: the static acpi_isa_irq_penalty[]
> >     values.  ACPI is *supposed* to tell us about all these usages, so
> >     I don't know why we have the table.  But it's been there as long
> >     as I can remember.  The table is probably x86-specific, but we
> >     keep it in the supposedly generic pci_link.c.
> >
> >   - The "acpi_irq_isa=" and "acpi_irq_pci=" command-line overrides via
> >     acpi_irq_pci().  I suppose these are for cases where we can't
> >     figure things out automatically.  I would resist adding parameters
> >     like this today (I would treat the need for them as a bug and look
> >     for a fix or a quirk), but we might be stuck with these.
> >
> >   - SCI information from the ACPI FADT (acpi_penalize_sci_irq()).
> >
> >   - PNPBIOS and PNPACPI device IRQ usage from _CRS and _PRS via
> >     acpi_penalize_isa_irq().  This is only for IRQs 0-15, and it does
> >     NOT include interrupt link (PNP0C0F) devices because we don't
> >     handle them as PNPACPI devices.  I think this is related to the
> >     fact that PNP0C0F doesn't appear in acpi_pnp_device_ids[].
> >
> >   - For non-PNP0C0F, non-PNPACPI devices, we completely ignore IRQ
> >     information from _CRS and _PRS.  This seems sub-optimal and
> >     possibly buggy.
> >
> >   - Interrupt link (PNP0C0F) IRQ usage from _CRS and _PRS via
> >     acpi_irq_penalty_init().  This is only for IRQs 0-15, and we only
> >     call this on x86.  If _PRS exists, we penalize each possible IRQ.
> >     If there's no _PRS but _CRS contains an active IRQ, we penalize
> >     it.
> >
> >   - Interrupt link (PNP0C0F) IRQ usage from _CRS and _PRS when
> >     enabling a new link.  In acpi_irq_pci_sharing_penalty(), we
> >     penalize an IRQ if it appears in _CRS or _PRS of any link device
> >     in the system.
> >
> >     For IRQs 0-15, this overlaps with the penalization done at
> >     boot-time by acpi_irq_penalty_init(): if a device has _PRS, we'll
> >     add the "possible" penalty twice (once in acpi_irq_penalty_init()
> >     and again in acpi_irq_pci_sharing_penalty()), and the "using"
> >     penalty once (in acpi_irq_pci_sharing_penalty()).
> >
> >     If a device has no _PRS but has _CRS, the "using" penalty is
> >     applied twice (once in once in acpi_irq_penalty_init() and again
> >     in acpi_irq_pci_sharing_penalty())
> >
> > I think this whole thing is baroque and grotesque.
> 
> While I agree, I also would like the regression introduced here to be
> fixed ASAP.
> 
> So do you want me to revert all of the changes made here so far and start over?

You're right, of course.  We need to fix the regression first, then
worry about longer-term changes.  I don't think we necessarily need to
fix *all* the issues with the current scheme, because most of them
have been there forever and I don't think people are tripping over
them.

Bjorn

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Sinan Kaya]

On 10/19/2016 3:44 PM, Bjorn Helgaas wrote:
[cut]

>>
>> Same problem here. This line will be broken after the sci_penalty change.
>>
>>                 acpi_isa_irq_penalty[irq] = acpi_irq_get_penalty(irq) +
>>                   (active ? PIRQ_PENALTY_ISA_USED : PIRQ_PENALTY_PCI_USING);
> 
> I think the fragility of this code is an indication that we have a
> design problem, so I want to step back from the nitty gritty details
> for a bit and look at the overall design.

This is all because we started with a very simple design where we replaced
the array with a get_penalty function. Set penalty function would dynamically
reallocate the link list so that we have a design that works no matter
what the number of IRQs are. 

As we hit the first problem on existing platform, we realized that we can't
actually dynamically reallocate an interrupt because get_penalty function
gets called from early boot stages where heap is not initialized yet.

Then, we started restructuring the code so that we can discover the IRQs
by iterating the link list rather than using an array. While transitioning
into this new design, we forgot the fact that irq_get_penalty function
became a fixed array penalty + dynamically calculated penalty. 

This line was trying to increment the static array but ended up gathering
the dynamic pieces together.

This obviously is a bug and was forgotten in the code.

> 
> Let me restate the overall problem: We have a PCI device connected to
> an interrupt link.  The interrupt link can be connected to one of
> several IRQs, and we want to choose one of those IRQs to minimize IRQ
> sharing.
> 
> That means we need information about which IRQs are used.
> Historically we've started with a compiled-in table of common ISA IRQ
> usage, and we also collect information about which IRQs are used and
> which *might* be used.  So we have the following inputs:
> 
>   - Compiled-in ISA IRQ usage: the static acpi_isa_irq_penalty[]
>     values.  ACPI is *supposed* to tell us about all these usages, so
>     I don't know why we have the table.  But it's been there as long
>     as I can remember.  The table is probably x86-specific, but we
>     keep it in the supposedly generic pci_link.c.
> 

I think it is because the ISA IRQ functions get called in early boot stages
and there is no heap allocator at that point of execution.

>   - The "acpi_irq_isa=" and "acpi_irq_pci=" command-line overrides via
>     acpi_irq_pci().  I suppose these are for cases where we can't
>     figure things out automatically.  I would resist adding parameters
>     like this today (I would treat the need for them as a bug and look
>     for a fix or a quirk), but we might be stuck with these.
> 
>   - SCI information from the ACPI FADT (acpi_penalize_sci_irq()).
> 
>   - PNPBIOS and PNPACPI device IRQ usage from _CRS and _PRS via
>     acpi_penalize_isa_irq().  This is only for IRQs 0-15, and it does
>     NOT include interrupt link (PNP0C0F) devices because we don't
>     handle them as PNPACPI devices.  I think this is related to the
>     fact that PNP0C0F doesn't appear in acpi_pnp_device_ids[].

The original code supports the entire IRQ range (0..255). We limited it
to 16 during redesign.

> 
>   - For non-PNP0C0F, non-PNPACPI devices, we completely ignore IRQ
>     information from _CRS and _PRS.  This seems sub-optimal and
>     possibly buggy.
> 
>   - Interrupt link (PNP0C0F) IRQ usage from _CRS and _PRS via
>     acpi_irq_penalty_init().  This is only for IRQs 0-15, and we only
>     call this on x86.  If _PRS exists, we penalize each possible IRQ.
>     If there's no _PRS but _CRS contains an active IRQ, we penalize
>     it.

Correct.

> 
>   - Interrupt link (PNP0C0F) IRQ usage from _CRS and _PRS when
>     enabling a new link.  In acpi_irq_pci_sharing_penalty(), we
>     penalize an IRQ if it appears in _CRS or _PRS of any link device
>     in the system.
>     
>     For IRQs 0-15, this overlaps with the penalization done at
>     boot-time by acpi_irq_penalty_init(): if a device has _PRS, we'll
>     add the "possible" penalty twice (once in acpi_irq_penalty_init()
>     and again in acpi_irq_pci_sharing_penalty()), and the "using"
>     penalty once (in acpi_irq_pci_sharing_penalty()).
> 
>     If a device has no _PRS but has _CRS, the "using" penalty is
>     applied twice (once in once in acpi_irq_penalty_init() and again
>     in acpi_irq_pci_sharing_penalty())
> 
> I think this whole thing is baroque and grotesque.

Regardless of these, we hit four different bugs because we didn't understand
the usage model.

1. You can't use kmalloc in IRQ get penalty
2. SCI IRQ type cannot be gathered from IRQ API
3. acpi_irq_penalty cannot be called from early stages since it is iterating
the link list. If there is a penalty assignment requirement from early stages,
this needs to be done on a statically allocated array.
4. acpi_irq_penalty_init is redundant.

> 
> Here's a strawman idea:
> 
>   - Maintain a mapping of (IRQ, penalty).  Initially all penalties are
>     zero.  This is for *all* IRQs, not just ISA ones.  This could be a
>     linked list, but the structure is not important as long as we can
>     add things dynamically.

Dynamic allocation doesn't work due to early calls from x86 architecture.
This is the reason why we iterate the link objects.

> 
>   - Add a "acpi_penalize_irq()" function similar to
>     acpi_penalize_isa_irq(), but not restricted to ISA, so we can
>     increase the penalty for any IRQ, and maybe we can specify how
>     much penalty to add.

This was my first patch. It didn't work due to heap requirements.

> 
>   - Make acpi_irq_get_penalty() a simple lookup in the mapping.  No
>     iterating through all link devices.
> 
>   - If we think the compiled-in penalties are really necessary, move
>     the table to x86 code and add a boot-time loop to use
>     acpi_penalize_irq() to penalize these IRQs.  Same for the
>     command-line options.

I agree, we can move all of the ISA interrupt stuff out of this file
if we want to.

I really don't like the fact that we have a acpi_penalize_isa_irq function.
The penalty information should have been contained in this file. 

acpi_penalize_isa_irq gets called from arbitrary contexts.

I'm not sure adding a acpi_penalize_irq function is a good idea. All penalty
users should have been in this file. 

> 
>   - Change acpi_penalize_sci_irq() to use acpi_penalize_irq().
>     Probably the mapping needs to pay attention to trigger/polarity
>     somehow, too.
> 
>   - Figure out how to make the ACPI core use acpi_penalize_irq() to
>     based on the _CRS and _PRS of every ACPI device, including
>     PNPACPI, PNP0C0F, etc.  Then we can remove acpi_irq_penalty_init().
> 

I think this is a new feature.

>   - Change acpi_pci_link_set() to use acpi_penalize_irq() for the IRQ
>     it is enabling.  Conceptually maybe this should be done in the
>     acpi_set_current_resources() path so it happens whenever we use
>     _CRS to enable an IRQ on *any* ACPI device.
> 
> I think the biggest issue is figuring out how to get the ACPI core to
> look at the _CRS for *all* devices.  If we could do that, I think it
> could substantially simplify this code.

I think my V4 patch satisfies all of these requirements and also fixes
the existing regression. (I have test vectors captured from the machines
with issues.)

1. You can't use kmalloc in IRQ get penalty
2. SCI IRQ type cannot be gathered from IRQ API
3. acpi_irq_penalty cannot be called from early stages since it is iterating
the link list. If there is a penalty assignment requirement from early stages,
this needs to be done on a statically allocated array.
4. acpi_irq_penalty_init is redundant.

and your goal to have some common code that can be used with any IRQ we want.

If we want to move the ISA pieces out of this file, that can be done too. 
We can also add support for PNPACPI. I'm not a very big fan of scratch
everything and start from beginning approach. This refactoring effort already
failed 3 times. I'd like to close the issue and move on.

> 
> Bjorn
> --
> To unsubscribe from this list: send the line "unsubscribe linux-pci" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 


-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Rafael J. Wysocki]

On Thu, Oct 20, 2016 at 10:01 PM, Sinan Kaya <okaya@codeaurora.org> wrote:
> On 10/19/2016 3:44 PM, Bjorn Helgaas wrote:

[cut]

> If we want to move the ISA pieces out of this file, that can be done too.
> We can also add support for PNPACPI. I'm not a very big fan of scratch
> everything and start from beginning approach. This refactoring effort already
> failed 3 times. I'd like to close the issue and move on.

Understood, but we have broken this for too many times already.

Either we have a minimum fix that is known working or we are going back.

Thanks,
Rafael

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Sinan Kaya]

On 10/20/2016 2:08 PM, Rafael J. Wysocki wrote:
> On Thu, Oct 20, 2016 at 10:01 PM, Sinan Kaya <okaya@codeaurora.org> wrote:
>> On 10/19/2016 3:44 PM, Bjorn Helgaas wrote:
> 
> [cut]
> 
>> If we want to move the ISA pieces out of this file, that can be done too.
>> We can also add support for PNPACPI. I'm not a very big fan of scratch
>> everything and start from beginning approach. This refactoring effort already
>> failed 3 times. I'd like to close the issue and move on.
> 
> Understood, but we have broken this for too many times already.
> 
> Either we have a minimum fix that is known working or we are going back.

Agreed. I think my V4 patch satisfies our short term goals and fixes the issue
you are seeing. IMO, it is good enough for the moment.

> 
> Thanks,
> Rafael
> 


-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Bjorn Helgaas]

On Thu, Oct 20, 2016 at 01:01:04PM -0700, Sinan Kaya wrote:
> On 10/19/2016 3:44 PM, Bjorn Helgaas wrote:

> >   - Maintain a mapping of (IRQ, penalty).  Initially all penalties are
> >     zero.  This is for *all* IRQs, not just ISA ones.  This could be a
> >     linked list, but the structure is not important as long as we can
> >     add things dynamically.
> 
> Dynamic allocation doesn't work due to early calls from x86 architecture.
> This is the reason why we iterate the link objects.

Where exactly is this early penalization?  That seems to be the
biggest problem.  Well, maybe the question of ACPI core parsing of
_CRS/_PRS is a bigger structural problem, but the dynamic allocation
thing at least seems solvable.

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Sinan Kaya]

On 10/20/2016 7:41 PM, Bjorn Helgaas wrote:
> On Thu, Oct 20, 2016 at 01:01:04PM -0700, Sinan Kaya wrote:
>> On 10/19/2016 3:44 PM, Bjorn Helgaas wrote:
> 
>>>   - Maintain a mapping of (IRQ, penalty).  Initially all penalties are
>>>     zero.  This is for *all* IRQs, not just ISA ones.  This could be a
>>>     linked list, but the structure is not important as long as we can
>>>     add things dynamically.
>>
>> Dynamic allocation doesn't work due to early calls from x86 architecture.
>> This is the reason why we iterate the link objects.
> 
> Where exactly is this early penalization?  That seems to be the
> biggest problem.  Well, maybe the question of ACPI core parsing of
> _CRS/_PRS is a bigger structural problem, but the dynamic allocation
> thing at least seems solvable.
> 

http://marc.info/?l=linux-acpi&m=145580159209240&w=2


-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Sinan Kaya]

Bjorn,

On 10/18/2016 6:59 AM, Bjorn Helgaas wrote:
> It seems wrong to me that we call acpi_irq_get_penalty() from
> acpi_irq_penalty_update() and acpi_penalize_isa_irq().  It seems like they
> should just manipulate acpi_isa_irq_penalty[irq] directly.
> 
> acpi_irq_penalty_update() is for command-line parameters, so it certainly
> doesn't need the acpi_irq_pci_sharing_penalty() information (the
> acpi_link_list should be empty at the time we process the command-line
> parameters).
> 
> acpi_penalize_isa_irq() is telling us that a PNP or ACPI device is using
> the IRQ -- this should modify the IRQ's penalty, but it shouldn't depend on
> the acpi_irq_pci_sharing_penalty() value at all.

I posted v4 with this change and also went back to the original implementation
for sharing penalty calculation whether the IRQ is ISA or PCI.

Let us know what you think. 

I also realized that calculating sharing penalty while the link object is not 
initialized is not right.

Sinan

-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Bjorn Helgaas]

On Sat, Oct 15, 2016 at 12:31:05AM -0400, Sinan Kaya wrote:
> The change introduced in commit 103544d86976 ("ACPI,PCI,IRQ: reduce
> resource requirements") removed PCI_USING penalty from
> acpi_pci_link_allocate function as there is no longer a fixed size penalty
> array for both PCI interrupts greater than 16.
> 
> The array size has been reduced to 16 and array name got prefixed as ISA
> since it only is accountable for the ISA interrupts.
> 
> The original change in commit 103544d86976 ("ACPI,PCI,IRQ: reduce
> resource requirements") removed penalty assignment in the code for PCI
> thinking that we will add the penalty later in acpi_irq_pci_sharing_penalty
> function.
> 
> However, this function only gets called if the IRQ number is greater than
> 16 and acpi_irq_get_penalty function gets called before ACPI start in
> acpi_isa_irq_available and acpi_penalize_isa_irq functions. We can't rely
> on iterating the link list.
> 
> We need to add the PCI_USING penalty for ISA interrupts too if the link is
> in use and matches our ISA IRQ number.

I think the history about the array size is more than is necessary for this
changelog.  I think the useful part is something like this:

  ACPI: pci_link: Include PIRQ_PENALTY_PCI_USING for ISA IRQs

  103544d86976 ("ACPI,PCI,IRQ: reduce resource requirements") replaced
  the addition of PIRQ_PENALTY_PCI_USING in acpi_pci_link_allocate()
  with an addition in acpi_irq_pci_sharing_penalty(), but f7eca374f000
  ("ACPI,PCI,IRQ: separate ISA penalty calculation") removed the use
  of acpi_irq_pci_sharing_penalty() for ISA IRQs.

  Therefore, PIRQ_PENALTY_PCI_USING is missing from ISA IRQs used by
  interrupt links.  Include that penalty by adding it in the
  acpi_pci_link_allocate() path.

  Fixes: f7eca374f000 ("ACPI,PCI,IRQ: separate ISA penalty calculation")
  
> Signed-off-by: Sinan Kaya <okaya@codeaurora.org>

Acked-by: Bjorn Helgaas <bhelgaas@google.com>

> ---
>  drivers/acpi/pci_link.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/drivers/acpi/pci_link.c b/drivers/acpi/pci_link.c
> index c983bf7..a212709 100644
> --- a/drivers/acpi/pci_link.c
> +++ b/drivers/acpi/pci_link.c
> @@ -619,6 +619,10 @@ static int acpi_pci_link_allocate(struct acpi_pci_link *link)
>  			    acpi_device_bid(link->device));
>  		return -ENODEV;
>  	} else {
> +		if (link->irq.active < ACPI_MAX_ISA_IRQS)
> +			acpi_isa_irq_penalty[link->irq.active] +=
> +				PIRQ_PENALTY_PCI_USING;
> +
>  		printk(KERN_WARNING PREFIX "%s [%s] enabled at IRQ %d\n",
>  		       acpi_device_name(link->device),
>  		       acpi_device_bid(link->device), link->irq.active);
> -- 
> 1.9.1
> 
> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [PATCH V3 1/3] ACPI, PCI IRQ: add PCI_USING penalty for ISA interrupts

[Sinan Kaya]

On 10/22/2016 7:58 PM, Bjorn Helgaas wrote:
> On Sat, Oct 15, 2016 at 12:31:05AM -0400, Sinan Kaya wrote:
>> The change introduced in commit 103544d86976 ("ACPI,PCI,IRQ: reduce
>> resource requirements") removed PCI_USING penalty from
>> acpi_pci_link_allocate function as there is no longer a fixed size penalty
>> array for both PCI interrupts greater than 16.
>>
>> The array size has been reduced to 16 and array name got prefixed as ISA
>> since it only is accountable for the ISA interrupts.
>>
>> The original change in commit 103544d86976 ("ACPI,PCI,IRQ: reduce
>> resource requirements") removed penalty assignment in the code for PCI
>> thinking that we will add the penalty later in acpi_irq_pci_sharing_penalty
>> function.
>>
>> However, this function only gets called if the IRQ number is greater than
>> 16 and acpi_irq_get_penalty function gets called before ACPI start in
>> acpi_isa_irq_available and acpi_penalize_isa_irq functions. We can't rely
>> on iterating the link list.
>>
>> We need to add the PCI_USING penalty for ISA interrupts too if the link is
>> in use and matches our ISA IRQ number.
> 
> I think the history about the array size is more than is necessary for this
> changelog.  I think the useful part is something like this:
> 
>   ACPI: pci_link: Include PIRQ_PENALTY_PCI_USING for ISA IRQs
> 
>   103544d86976 ("ACPI,PCI,IRQ: reduce resource requirements") replaced
>   the addition of PIRQ_PENALTY_PCI_USING in acpi_pci_link_allocate()
>   with an addition in acpi_irq_pci_sharing_penalty(), but f7eca374f000
>   ("ACPI,PCI,IRQ: separate ISA penalty calculation") removed the use
>   of acpi_irq_pci_sharing_penalty() for ISA IRQs.
> 
>   Therefore, PIRQ_PENALTY_PCI_USING is missing from ISA IRQs used by
>   interrupt links.  Include that penalty by adding it in the
>   acpi_pci_link_allocate() path.
> 
>   Fixes: f7eca374f000 ("ACPI,PCI,IRQ: separate ISA penalty calculation")
>   
>> Signed-off-by: Sinan Kaya <okaya@codeaurora.org>
> 
> Acked-by: Bjorn Helgaas <bhelgaas@google.com>

OK. Updated as suggested.

> 
>> ---
>>  drivers/acpi/pci_link.c | 4 ++++
>>  1 file changed, 4 insertions(+)
>>
>> diff --git a/drivers/acpi/pci_link.c b/drivers/acpi/pci_link.c
>> index c983bf7..a212709 100644
>> --- a/drivers/acpi/pci_link.c
>> +++ b/drivers/acpi/pci_link.c
>> @@ -619,6 +619,10 @@ static int acpi_pci_link_allocate(struct acpi_pci_link *link)
>>  			    acpi_device_bid(link->device));
>>  		return -ENODEV;
>>  	} else {
>> +		if (link->irq.active < ACPI_MAX_ISA_IRQS)
>> +			acpi_isa_irq_penalty[link->irq.active] +=
>> +				PIRQ_PENALTY_PCI_USING;
>> +
>>  		printk(KERN_WARNING PREFIX "%s [%s] enabled at IRQ %d\n",
>>  		       acpi_device_name(link->device),
>>  		       acpi_device_bid(link->device), link->irq.active);
>> -- 
>> 1.9.1
>>
>>
>> _______________________________________________
>> linux-arm-kernel mailing list
>> linux-arm-kernel@lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
> 


-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

[PATCH V3 2/3] Revert "ACPI,PCI,IRQ: remove SCI penalize function"

[Sinan Kaya]

The SCI function was removed in two steps (first refactor and then remove).
This patch does the revert in one step.

The commit 103544d86976 ("ACPI,PCI,IRQ: reduce resource requirements")
refactored the original code so that SCI penalty is calculated dynamically
by the time get_penalty function is called. This patch does a partial
revert for the SCI functionality only.

The commit 9e5ed6d1fb87 ("ACPI,PCI,IRQ: remove SCI penalize function") is
for the removal of the function. SCI penalty API was replaced by the
runtime penalty calculation based on the value of
acpi_gbl_FADT.sci_interrupt.

The IRQ type does not get updated at the right time for some platforms and
results in incorrect penalty assignment for PCI IRQs as
irq_get_trigger_type returns the wrong type.

The register_gsi function delivers the IRQ found in the ACPI table to
the interrupt controller driver.  Penalties are calculated before a
link object is enabled to find out which interrupt has the least number
of users. By the time penalties are calculated, the IRQ is not registered
yet and the API returns the wrong type.

Tested-by: Jonathan Liu <net147@gmail.com>
Tested-by: Ondrej Zary <linux@rainbow-software.org>
Link: https://lkml.org/lkml/2016/10/4/283
Signed-off-by: Sinan Kaya <okaya@codeaurora.org>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
---
 arch/x86/kernel/acpi/boot.c |  1 +
 drivers/acpi/pci_link.c     | 32 +++++++++++++-------------------
 include/linux/acpi.h        |  1 +
 3 files changed, 15 insertions(+), 19 deletions(-)

diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
index 90d84c3..0ffd26e 100644
--- a/arch/x86/kernel/acpi/boot.c
+++ b/arch/x86/kernel/acpi/boot.c
@@ -453,6 +453,7 @@ static void __init acpi_sci_ioapic_setup(u8 bus_irq, u16 polarity, u16 trigger,
 		polarity = acpi_sci_flags & ACPI_MADT_POLARITY_MASK;
 
 	mp_override_legacy_irq(bus_irq, polarity, trigger, gsi);
+	acpi_penalize_sci_irq(bus_irq, trigger, polarity);
 
 	/*
 	 * stash over-ride to indicate we've been here
diff --git a/drivers/acpi/pci_link.c b/drivers/acpi/pci_link.c
index a212709..1934e2a 100644
--- a/drivers/acpi/pci_link.c
+++ b/drivers/acpi/pci_link.c
@@ -494,27 +494,10 @@ static int acpi_irq_pci_sharing_penalty(int irq)
 
 static int acpi_irq_get_penalty(int irq)
 {
-	int penalty = 0;
-
-	/*
-	* Penalize IRQ used by ACPI SCI. If ACPI SCI pin attributes conflict
-	* with PCI IRQ attributes, mark ACPI SCI as ISA_ALWAYS so it won't be
-	* use for PCI IRQs.
-	*/
-	if (irq == acpi_gbl_FADT.sci_interrupt) {
-		u32 type = irq_get_trigger_type(irq) & IRQ_TYPE_SENSE_MASK;
-
-		if (type != IRQ_TYPE_LEVEL_LOW)
-			penalty += PIRQ_PENALTY_ISA_ALWAYS;
-		else
-			penalty += PIRQ_PENALTY_PCI_USING;
-	}
-
 	if (irq < ACPI_MAX_ISA_IRQS)
-		return penalty + acpi_isa_irq_penalty[irq];
+		return acpi_isa_irq_penalty[irq];
 
-	penalty += acpi_irq_pci_sharing_penalty(irq);
-	return penalty;
+	return acpi_irq_pci_sharing_penalty(irq);
 }
 
 int __init acpi_irq_penalty_init(void)
@@ -885,6 +868,17 @@ bool acpi_isa_irq_available(int irq)
 		    acpi_irq_get_penalty(irq) < PIRQ_PENALTY_ISA_ALWAYS);
 }
 
+void acpi_penalize_sci_irq(int irq, int trigger, int polarity)
+{
+	if (irq >= 0 && irq < ARRAY_SIZE(acpi_isa_irq_penalty)) {
+		if (trigger != ACPI_MADT_TRIGGER_LEVEL ||
+		    polarity != ACPI_MADT_POLARITY_ACTIVE_LOW)
+			acpi_isa_irq_penalty[irq] += PIRQ_PENALTY_ISA_ALWAYS;
+		else
+			acpi_isa_irq_penalty[irq] += PIRQ_PENALTY_PCI_USING;
+	}
+}
+
 /*
  * Over-ride default table to reserve additional IRQs for use by ISA
  * e.g. acpi_irq_isa=5
diff --git a/include/linux/acpi.h b/include/linux/acpi.h
index c5eaf2f..67d1d3e 100644
--- a/include/linux/acpi.h
+++ b/include/linux/acpi.h
@@ -318,6 +318,7 @@ struct pci_dev;
 int acpi_pci_irq_enable (struct pci_dev *dev);
 void acpi_penalize_isa_irq(int irq, int active);
 bool acpi_isa_irq_available(int irq);
+void acpi_penalize_sci_irq(int irq, int trigger, int polarity);
 void acpi_pci_irq_disable (struct pci_dev *dev);
 
 extern int ec_read(u8 addr, u8 *val);
-- 
1.9.1

Re: [PATCH V3 2/3] Revert "ACPI,PCI,IRQ: remove SCI penalize function"

[Rafael J. Wysocki]

On Sat, Oct 15, 2016 at 6:31 AM, Sinan Kaya <okaya@codeaurora.org> wrote:
> The SCI function was removed in two steps (first refactor and then remove).
> This patch does the revert in one step.

"The SCI penalize function was removed in two steps (first refactor
and then remove) and these changes are reverted here in one go."

>
> The commit 103544d86976 ("ACPI,PCI,IRQ: reduce resource requirements")
> refactored the original code so that SCI penalty is calculated dynamically
> by the time get_penalty function is called. This patch does a partial
> revert for the SCI functionality only.

I'd write the last sentence this way:

"That change is partially reverted here, specifically for the SCI IRQ alone."

> The commit 9e5ed6d1fb87 ("ACPI,PCI,IRQ: remove SCI penalize function") is
> for the removal of the function. SCI penalty API was replaced by the
> runtime penalty calculation based on the value of
> acpi_gbl_FADT.sci_interrupt.

"The SCI penalize function was finally dropped by commit 9e5ed6d1fb87
(ACPI,PCI,IRQ: remove SCI penalize function) that replaced the old SCI
penalty API with penalty calculation carried out dynamically and based
on the acpi_gbl_FADT.sci_interrupt value."

> The IRQ type does not get updated at the right time for some platforms and
> results in incorrect penalty assignment for PCI IRQs as
> irq_get_trigger_type returns the wrong type.

"However, that new algorithm relies on the accurate setting of IRQ
types and that doesn't happen early enough on some platforms which
leads to incorrect penalty assignments for PCI IRQs.  In those cases,
irq_get_trigger_type() returns incorrect values for the IRQs in
question, because they have not been registered yet by the time the
penalties are calculated."

And then you don't need the paragraph below then.

> The register_gsi function delivers the IRQ found in the ACPI table to
> the interrupt controller driver.  Penalties are calculated before a
> link object is enabled to find out which interrupt has the least number
> of users. By the time penalties are calculated, the IRQ is not registered
> yet and the API returns the wrong type.

And I would add:

"For this reason, revert commit 9e5ed6d1fb87 (ACPI,PCI,IRQ: remove SCI
penalize function) completely to restore the correct behavior."

Also, please add Fixes: tags pointing to the commits you're fixing here.

> Tested-by: Jonathan Liu <net147@gmail.com>
> Tested-by: Ondrej Zary <linux@rainbow-software.org>
> Link: https://lkml.org/lkml/2016/10/4/283
> Signed-off-by: Sinan Kaya <okaya@codeaurora.org>
> Acked-by: Thomas Gleixner <tglx@linutronix.de>
> ---
>  arch/x86/kernel/acpi/boot.c |  1 +
>  drivers/acpi/pci_link.c     | 32 +++++++++++++-------------------
>  include/linux/acpi.h        |  1 +
>  3 files changed, 15 insertions(+), 19 deletions(-)
>
> diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
> index 90d84c3..0ffd26e 100644
> --- a/arch/x86/kernel/acpi/boot.c
> +++ b/arch/x86/kernel/acpi/boot.c
> @@ -453,6 +453,7 @@ static void __init acpi_sci_ioapic_setup(u8 bus_irq, u16 polarity, u16 trigger,
>                 polarity = acpi_sci_flags & ACPI_MADT_POLARITY_MASK;
>
>         mp_override_legacy_irq(bus_irq, polarity, trigger, gsi);
> +       acpi_penalize_sci_irq(bus_irq, trigger, polarity);
>
>         /*
>          * stash over-ride to indicate we've been here
> diff --git a/drivers/acpi/pci_link.c b/drivers/acpi/pci_link.c
> index a212709..1934e2a 100644
> --- a/drivers/acpi/pci_link.c
> +++ b/drivers/acpi/pci_link.c
> @@ -494,27 +494,10 @@ static int acpi_irq_pci_sharing_penalty(int irq)
>
>  static int acpi_irq_get_penalty(int irq)
>  {
> -       int penalty = 0;
> -
> -       /*
> -       * Penalize IRQ used by ACPI SCI. If ACPI SCI pin attributes conflict
> -       * with PCI IRQ attributes, mark ACPI SCI as ISA_ALWAYS so it won't be
> -       * use for PCI IRQs.
> -       */
> -       if (irq == acpi_gbl_FADT.sci_interrupt) {
> -               u32 type = irq_get_trigger_type(irq) & IRQ_TYPE_SENSE_MASK;
> -
> -               if (type != IRQ_TYPE_LEVEL_LOW)
> -                       penalty += PIRQ_PENALTY_ISA_ALWAYS;
> -               else
> -                       penalty += PIRQ_PENALTY_PCI_USING;
> -       }
> -
>         if (irq < ACPI_MAX_ISA_IRQS)
> -               return penalty + acpi_isa_irq_penalty[irq];
> +               return acpi_isa_irq_penalty[irq];
>
> -       penalty += acpi_irq_pci_sharing_penalty(irq);
> -       return penalty;
> +       return acpi_irq_pci_sharing_penalty(irq);
>  }
>
>  int __init acpi_irq_penalty_init(void)
> @@ -885,6 +868,17 @@ bool acpi_isa_irq_available(int irq)
>                     acpi_irq_get_penalty(irq) < PIRQ_PENALTY_ISA_ALWAYS);
>  }
>
> +void acpi_penalize_sci_irq(int irq, int trigger, int polarity)
> +{
> +       if (irq >= 0 && irq < ARRAY_SIZE(acpi_isa_irq_penalty)) {
> +               if (trigger != ACPI_MADT_TRIGGER_LEVEL ||
> +                   polarity != ACPI_MADT_POLARITY_ACTIVE_LOW)
> +                       acpi_isa_irq_penalty[irq] += PIRQ_PENALTY_ISA_ALWAYS;
> +               else
> +                       acpi_isa_irq_penalty[irq] += PIRQ_PENALTY_PCI_USING;
> +       }
> +}
> +
>  /*
>   * Over-ride default table to reserve additional IRQs for use by ISA
>   * e.g. acpi_irq_isa=5
> diff --git a/include/linux/acpi.h b/include/linux/acpi.h
> index c5eaf2f..67d1d3e 100644
> --- a/include/linux/acpi.h
> +++ b/include/linux/acpi.h
> @@ -318,6 +318,7 @@ struct pci_dev;
>  int acpi_pci_irq_enable (struct pci_dev *dev);
>  void acpi_penalize_isa_irq(int irq, int active);
>  bool acpi_isa_irq_available(int irq);
> +void acpi_penalize_sci_irq(int irq, int trigger, int polarity);
>  void acpi_pci_irq_disable (struct pci_dev *dev);
>
>  extern int ec_read(u8 addr, u8 *val);
> --

Re: [PATCH V3 2/3] Revert "ACPI,PCI,IRQ: remove SCI penalize function"

[Bjorn Helgaas]

On Sat, Oct 15, 2016 at 12:31:06AM -0400, Sinan Kaya wrote:
> The SCI function was removed in two steps (first refactor and then remove).
> This patch does the revert in one step.
> 
> The commit 103544d86976 ("ACPI,PCI,IRQ: reduce resource requirements")
> refactored the original code so that SCI penalty is calculated dynamically
> by the time get_penalty function is called. This patch does a partial
> revert for the SCI functionality only.
> 
> The commit 9e5ed6d1fb87 ("ACPI,PCI,IRQ: remove SCI penalize function") is
> for the removal of the function. SCI penalty API was replaced by the
> runtime penalty calculation based on the value of
> acpi_gbl_FADT.sci_interrupt.
> 
> The IRQ type does not get updated at the right time for some platforms and
> results in incorrect penalty assignment for PCI IRQs as
> irq_get_trigger_type returns the wrong type.
> 
> The register_gsi function delivers the IRQ found in the ACPI table to
> the interrupt controller driver.  Penalties are calculated before a
> link object is enabled to find out which interrupt has the least number
> of users. By the time penalties are calculated, the IRQ is not registered
> yet and the API returns the wrong type.

I think you should squash patches 2 & 3 and not bother with trying to
make this a revert followed by a fix.  Squashing them makes the git
history much easier to follow, plus you don't have to deal with the
headache of ensuring the intermediate state is correct and bisectable.

> Tested-by: Jonathan Liu <net147@gmail.com>
> Tested-by: Ondrej Zary <linux@rainbow-software.org>
> Link: https://lkml.org/lkml/2016/10/4/283
> Signed-off-by: Sinan Kaya <okaya@codeaurora.org>
> Acked-by: Thomas Gleixner <tglx@linutronix.de>
> ---
>  arch/x86/kernel/acpi/boot.c |  1 +
>  drivers/acpi/pci_link.c     | 32 +++++++++++++-------------------
>  include/linux/acpi.h        |  1 +
>  3 files changed, 15 insertions(+), 19 deletions(-)
> 
> diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
> index 90d84c3..0ffd26e 100644
> --- a/arch/x86/kernel/acpi/boot.c
> +++ b/arch/x86/kernel/acpi/boot.c
> @@ -453,6 +453,7 @@ static void __init acpi_sci_ioapic_setup(u8 bus_irq, u16 polarity, u16 trigger,
>  		polarity = acpi_sci_flags & ACPI_MADT_POLARITY_MASK;
>  
>  	mp_override_legacy_irq(bus_irq, polarity, trigger, gsi);
> +	acpi_penalize_sci_irq(bus_irq, trigger, polarity);
>  
>  	/*
>  	 * stash over-ride to indicate we've been here
> diff --git a/drivers/acpi/pci_link.c b/drivers/acpi/pci_link.c
> index a212709..1934e2a 100644
> --- a/drivers/acpi/pci_link.c
> +++ b/drivers/acpi/pci_link.c
> @@ -494,27 +494,10 @@ static int acpi_irq_pci_sharing_penalty(int irq)
>  
>  static int acpi_irq_get_penalty(int irq)
>  {
> -	int penalty = 0;
> -
> -	/*
> -	* Penalize IRQ used by ACPI SCI. If ACPI SCI pin attributes conflict
> -	* with PCI IRQ attributes, mark ACPI SCI as ISA_ALWAYS so it won't be
> -	* use for PCI IRQs.
> -	*/
> -	if (irq == acpi_gbl_FADT.sci_interrupt) {
> -		u32 type = irq_get_trigger_type(irq) & IRQ_TYPE_SENSE_MASK;
> -
> -		if (type != IRQ_TYPE_LEVEL_LOW)
> -			penalty += PIRQ_PENALTY_ISA_ALWAYS;
> -		else
> -			penalty += PIRQ_PENALTY_PCI_USING;
> -	}
> -
>  	if (irq < ACPI_MAX_ISA_IRQS)
> -		return penalty + acpi_isa_irq_penalty[irq];
> +		return acpi_isa_irq_penalty[irq];
>  
> -	penalty += acpi_irq_pci_sharing_penalty(irq);
> -	return penalty;
> +	return acpi_irq_pci_sharing_penalty(irq);
>  }
>  
>  int __init acpi_irq_penalty_init(void)
> @@ -885,6 +868,17 @@ bool acpi_isa_irq_available(int irq)
>  		    acpi_irq_get_penalty(irq) < PIRQ_PENALTY_ISA_ALWAYS);
>  }
>  
> +void acpi_penalize_sci_irq(int irq, int trigger, int polarity)
> +{
> +	if (irq >= 0 && irq < ARRAY_SIZE(acpi_isa_irq_penalty)) {
> +		if (trigger != ACPI_MADT_TRIGGER_LEVEL ||
> +		    polarity != ACPI_MADT_POLARITY_ACTIVE_LOW)
> +			acpi_isa_irq_penalty[irq] += PIRQ_PENALTY_ISA_ALWAYS;
> +		else
> +			acpi_isa_irq_penalty[irq] += PIRQ_PENALTY_PCI_USING;
> +	}
> +}
> +
>  /*
>   * Over-ride default table to reserve additional IRQs for use by ISA
>   * e.g. acpi_irq_isa=5
> diff --git a/include/linux/acpi.h b/include/linux/acpi.h
> index c5eaf2f..67d1d3e 100644
> --- a/include/linux/acpi.h
> +++ b/include/linux/acpi.h
> @@ -318,6 +318,7 @@ struct pci_dev;
>  int acpi_pci_irq_enable (struct pci_dev *dev);
>  void acpi_penalize_isa_irq(int irq, int active);
>  bool acpi_isa_irq_available(int irq);
> +void acpi_penalize_sci_irq(int irq, int trigger, int polarity);
>  void acpi_pci_irq_disable (struct pci_dev *dev);
>  
>  extern int ec_read(u8 addr, u8 *val);
> -- 
> 1.9.1
> 
> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [PATCH V3 2/3] Revert "ACPI,PCI,IRQ: remove SCI penalize function"

[Sinan Kaya]

On 10/18/2016 7:05 AM, Bjorn Helgaas wrote:
> I think you should squash patches 2 & 3 and not bother with trying to
> make this a revert followed by a fix.  Squashing them makes the git
> history much easier to follow, plus you don't have to deal with the
> headache of ensuring the intermediate state is correct and bisectable.

OK. I can do that.

-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

[PATCH V3 3/3] ACPI,PCI,IRQ: correct SCI penalty calculation

[Sinan Kaya]

It seems like the problem is that we removed acpi_penalize_sci_irq(),
which told us the polarity and trigger mode.  We tried to get that
information via irq_get_trigger_type(), but that didn't work in this
case because we use the acpi_irq_get_penalty() path before the SCI is
registered.

To fix this problem, we only need to fix the penalty for the SCI interrupt.
It seems better to add a single "sci_penalty" variable, set it to
PIRQ_PENALTY_PCI_USING if it's level/low or PIRQ_PENALTY_ISA_ALWAYS
otherwise, and add "sci_penalty" in when appropriate.  That should fix it
for *any* SCI IRQ, not just those less than 256, and we don't have to add
these extra penalty table entries that are all unused (except possibly for
one entry if we have an SCI in the 16-255 range).

Signed-off-by: Sinan Kaya <okaya@codeaurora.org>
---
 drivers/acpi/pci_link.c | 24 +++++++++++++++---------
 1 file changed, 15 insertions(+), 9 deletions(-)

diff --git a/drivers/acpi/pci_link.c b/drivers/acpi/pci_link.c
index 1934e2a..34bf527 100644
--- a/drivers/acpi/pci_link.c
+++ b/drivers/acpi/pci_link.c
@@ -87,6 +87,7 @@ struct acpi_pci_link {
 
 static LIST_HEAD(acpi_link_list);
 static DEFINE_MUTEX(acpi_link_lock);
+static int sci_irq = -1, sci_penalty;
 
 /* --------------------------------------------------------------------------
                             PCI Link Device Management
@@ -494,10 +495,15 @@ static int acpi_irq_pci_sharing_penalty(int irq)
 
 static int acpi_irq_get_penalty(int irq)
 {
+	int penalty = 0;
+
+	if (irq == sci_irq)
+		penalty += sci_penalty;
+
 	if (irq < ACPI_MAX_ISA_IRQS)
-		return acpi_isa_irq_penalty[irq];
+		return penalty + acpi_isa_irq_penalty[irq];
 
-	return acpi_irq_pci_sharing_penalty(irq);
+	return penalty + acpi_irq_pci_sharing_penalty(irq);
 }
 
 int __init acpi_irq_penalty_init(void)
@@ -870,13 +876,13 @@ bool acpi_isa_irq_available(int irq)
 
 void acpi_penalize_sci_irq(int irq, int trigger, int polarity)
 {
-	if (irq >= 0 && irq < ARRAY_SIZE(acpi_isa_irq_penalty)) {
-		if (trigger != ACPI_MADT_TRIGGER_LEVEL ||
-		    polarity != ACPI_MADT_POLARITY_ACTIVE_LOW)
-			acpi_isa_irq_penalty[irq] += PIRQ_PENALTY_ISA_ALWAYS;
-		else
-			acpi_isa_irq_penalty[irq] += PIRQ_PENALTY_PCI_USING;
-	}
+	sci_irq = irq;
+
+	if (trigger == ACPI_MADT_TRIGGER_LEVEL &&
+	    polarity == ACPI_MADT_POLARITY_ACTIVE_LOW)
+		sci_penalty = PIRQ_PENALTY_PCI_USING;
+	else
+		sci_penalty = PIRQ_PENALTY_ISA_ALWAYS;
 }
 
 /*
-- 
1.9.1