New: _PRS/_SRS methods should be optional
A message that is not in the archive (sent off-list or not yet received): [email protected]/
Hi Pierre,
Thanks a lot for the report!
Quoted 18 lines · On Wed, Feb 02, 2022 at 10:20:44AM +0000, [email protected] wrote:
https://bugzilla.kernel.org/show_bug.cgi?id=215560
Bug ID: 215560 Summary: _PRS/_SRS methods should be optional Product: Drivers Version: 2.5 Kernel Version: v5.17-rc2 Hardware: All OS: Linux Tree: Mainline Status: NEW Severity: enhancement Priority: P1 Component: PCI Assignee: [email protected] Reporter: [email protected] Regression: No
Quoted 3 lines
The PCI legacy interrupts can be described with link devices, cf ACPI 6.4, s6.2.13 "
_PRT(PCI Routing Table)".
Link devices can have optional _SRS/_PRS methods to set the interrupt.
Is this a direct quote? I don't see text similar to this in ACPI v6.4.
I do see this in sec 6.2.13:
There are two ways that _PRT can be used. Typically, the interrupt input that a given PCI interrupt is on is configurable. For example, a given PCI interrupt might be configured for either IRQ 10 or 11 on an 8259 interrupt controller. In this model, each interrupt is represented in the ACPI namespace as a PCI Interrupt Link Device.
These objects have _PRS, _CRS, _SRS, and _DIS control methods to allocate the interrupt. Then, OSPM handles the interrupts not as interrupt inputs on the interrupt controller, but as PCI interrupt pins. The driver looks up the device’s pins in the _PRT to determine which device objects allocate the interrupts. To move the PCI interrupt to a different interrupt input on the interrupt controller, OSPM uses _PRS, _CRS, _SRS, and _DIS control methods for the PCI Interrupt Link Device.
In the second model, the PCI interrupts are hardwired to specific interrupt inputs on the interrupt controller and are not configurable. In this case, the Source field in _PRT does not reference a device, but instead contains the value zero, and the Source Index field contains the global system interrupt to which the PCI interrupt is hardwired.
For the first model (configurable inputs), it says "These objects have _PRS, _CRS, _SRS, and _DIS," which could be read as requiring those objects.
For the second model (hardwired inputs), the interrupts are not configurable, and I don't think there would be any reason to have an interrupt link device at all.
Quoted 8 lines
In PCI Firmware Specification Revision 3.3, s4.3.2.1. "Resource Setting":
"""
A non-configurable device only specifies_CRS. However, if they are configurable, devices include_PRSto indicate the possible resource setting and_SRSto allow OSPM to specify a new resource
allocation for the device.
"""
My copy of the PCI Firmware spec r3.3 (dated Jan 20, 2021), sec 4.3.2.1 says:
Host bridges resources programming is communicated to the operating system using ACPI methods _CRS, _SRS, and _PRS. _CRS indicates the current resource setting for the host bridge. This includes I/O space, memory space, and bus range assigned to the bridge by platform firmware.
A non-configurable device only specifies _CRS. However, if they are configurable, devices include _PRS to indicate the possible resource setting and _SRS to allow OSPM to specify a new resource allocation for the device.
So this is specifically talking about methods of a PCI host bridge (PNP0A03 or PNP0A08), not about methods of an interrupt link device (PNP0C0F).
Quoted 5 lines
However, _PRS/_SRS methods are checked in
drivers/acpi/pci_link.c, and the driver aborts if they are absent.
E.g.: When_PRSis missing:
ACPI: \_SB_.PCI0.LNKA:_CRS36 not found in_PRS
ACPI: \_SB_.PCI0.LNKA: No IRQ available. Try pci=noacpi or acpi=off
I assume this bug report is because something isn't working. Can you update the bugzilla with a note about what specifically isn't working and also attach a complete dmesg log and acpidump?
Bjorn
Hi Bjorn
Quoted 23 lines · Bjorn Helgaas
Hi Pierre,
Thanks a lot for the report!
Quoted 18 lines · On Wed, Feb 02, 2022 at 10:20:44AM +0000, [email protected] wrote:
https://bugzilla.kernel.org/show_bug.cgi?id=215560
Bug ID: 215560 Summary: _PRS/_SRS methods should be optional Product: Drivers Version: 2.5 Kernel Version: v5.17-rc2 Hardware: All OS: Linux Tree: Mainline Status: NEW Severity: enhancement Priority: P1 Component: PCI Assignee: [email protected] Reporter: [email protected] Regression: No
Quoted 7 lines
Quoted 3 lines
The PCI legacy interrupts can be described with link devices, cf ACPI 6.4, s6.2.13 "
_PRT(PCI Routing Table)".
Link devices can have optional _SRS/_PRS methods to set the interrupt.Is this a direct quote? I don't see text similar to this in ACPI v6.4.
No it was not a direct quote. However from ACPI 6.4, s6.2.12 "_PRS (Possible Resource Settings)":
'''This optional object evaluates [...]'''
and similarly at s6.2.16 "_SRS (Set Resource Settings)" '''This optional control method [...]'''
Quoted 50 lines
I do see this in sec 6.2.13:
There are two ways that
_PRTcan be used. Typically, the interrupt input that a given PCI interrupt is on is configurable. For example, a given PCI interrupt might be configured for either IRQ 10 or 11 on an 8259 interrupt controller. In this model, each interrupt is represented in the ACPI namespace as a PCI Interrupt Link Device.These objects have
_PRS,_CRS,_SRS, and_DIScontrol methods to allocate the interrupt. Then, OSPM handles the interrupts not as interrupt inputs on the interrupt controller, but as PCI interrupt pins. The driver looks up the device’s pins in the_PRTto determine which device objects allocate the interrupts. To move the PCI interrupt to a different interrupt input on the interrupt controller, OSPM uses_PRS,_CRS,_SRS, and_DIScontrol methods for the PCI Interrupt Link Device.In the second model, the PCI interrupts are hardwired to specific interrupt inputs on the interrupt controller and are not configurable. In this case, the Source field in
_PRTdoes not reference a device, but instead contains the value zero, and the Source Index field contains the global system interrupt to which the PCI interrupt is hardwired.For the first model (configurable inputs), it says "These objects have
_PRS,_CRS,_SRS, and_DIS," which could be read as requiring those objects.For the second model (hardwired inputs), the interrupts are not configurable, and I don't think there would be any reason to have an interrupt link device at all.
Quoted 8 lines
In PCI Firmware Specification Revision 3.3, s4.3.2.1. "Resource Setting":
"""
A non-configurable device only specifies_CRS. However, if they are configurable, devices include_PRSto indicate the possible resource setting and_SRSto allow OSPM to specify a new resource
allocation for the device.
"""My copy of the PCI Firmware spec r3.3 (dated Jan 20, 2021), sec 4.3.2.1 says:
Host bridges resources programming is communicated to the operating system using ACPI methods
_CRS,_SRS, and_PRS._CRSindicates the current resource setting for the host bridge. This includes I/O space, memory space, and bus range assigned to the bridge by platform firmware.
Quoted 9 lines
A non-configurable device only specifies
_CRS. However, if they are configurable, devices include_PRSto indicate the possible resource setting and_SRSto allow OSPM to specify a new resource allocation for the device.So this is specifically talking about methods of a PCI host bridge (PNP0A03 or PNP0A08), not about methods of an interrupt link device (PNP0C0F).
Yes indeed, this quote was not relevant.
Quoted 10 lines
Quoted 5 lines
However, _PRS/_SRS methods are checked in
drivers/acpi/pci_link.c, and the driver aborts if they are absent.
E.g.: When_PRSis missing:
ACPI: \_SB_.PCI0.LNKA:_CRS36 not found in_PRS
ACPI: \_SB_.PCI0.LNKA: No IRQ available. Try pci=noacpi or acpi=offI assume this bug report is because something isn't working. Can you update the bugzilla with a note about what specifically isn't working and also attach a complete dmesg log and acpidump?
The question arose while writing link devices code, so there is no platform with missing _PRS/_SRS methods that I know.
The question was more about spec compliance and the necessity to have these methods when legacy interrupts are not configurable.
The message above (_CRS XXX not found in _PRS) can be generated for a Juno for instance, and the ACPI tables are at:
https://github.com/tianocore/edk2-platforms/blob/master/Platform/ARM/JunoPkg/AcpiTables/AcpiSsdtRootPci.asl
The ACPI table need to be modified (remove _PRS and set _CRS correctly).
If the conclusion is that _PRS/_SRS are mandatory, even for hard-wired interrupts, then the bugzilla can be closed.
Thanks for the quick answer,
Pierre
Quoted 2 lines
Bjorn
[+cc Rafael, Alex, Marc]
Quoted 30 lines · Pierre Gondois
Quoted 14 lines · Bjorn Helgaas
Quoted 7 lines · On Wed, Feb 02, 2022 at 10:20:44AM +0000, [email protected] wrote:
https://bugzilla.kernel.org/show_bug.cgi?id=215560
The PCI legacy interrupts can be described with link devices, cf ACPI 6.4, s6.2.13 "
_PRT(PCI Routing Table)".
Link devices can have optional _SRS/_PRS methods to set the interrupt.
...Quoted 5 lines
However, _PRS/_SRS methods are checked in
drivers/acpi/pci_link.c, and the driver aborts if they are absent.
E.g.: When_PRSis missing:
ACPI: \_SB_.PCI0.LNKA:_CRS36 not found in_PRS
ACPI: \_SB_.PCI0.LNKA: No IRQ available. Try pci=noacpi or acpi=offQuoted 3 lines
I assume this bug report is because something isn't working. Can you update the bugzilla with a note about what specifically isn't working and also attach a complete dmesg log and acpidump?
The question arose while writing link devices code, so there is no platform with missing _PRS/_SRS methods that I know.
The question was more about spec compliance and the necessity to have these methods when legacy interrupts are not configurable. The message above (
_CRSXXX not found in_PRS) can be generated for a Juno for instance, and the ACPI tables are at:
https://github.com/tianocore/edk2-platforms/blob/master/Platform/ARM/JunoPkg/AcpiTables/AcpiSsdtRootPci.asl
The ACPI table need to be modified (remove_PRSand set_CRScorrectly).
Quoted 2 lines
If the conclusion is that _PRS/_SRS are mandatory, even for hard-wired interrupts, then the bugzilla can be closed.
OK, so if I understand correctly, you're using Interrupt Link devices not because it's possible to connect a PCI interrupt to one of several inputs on the interrupt controller, but because the PCI default of "level triggered, active low" is not compatible with GICv2.
The Interrupt Link device gives you a chance to specify "level triggered, active *high*". If you used a Source of 0 (where there is no Interrupt Link), there would be no way to specify that.
Since this use of Interrupt Links only conveys triggering information and nothing is configurable, I think the OS should get that info from _CRS, and _PRS and _SRS should not be required.
Alex made a change [1] along that line a while ago, but maybe there's more we should do.
Bjorn
Quoted 49 lines · Bjorn Helgaas
[+cc Rafael, Alex, Marc]
Quoted 30 lines · Pierre Gondois
Quoted 14 lines · Bjorn Helgaas
Quoted 7 lines · On Wed, Feb 02, 2022 at 10:20:44AM +0000, [email protected] wrote:
https://bugzilla.kernel.org/show_bug.cgi?id=215560
The PCI legacy interrupts can be described with link devices, cf ACPI 6.4, s6.2.13 "
_PRT(PCI Routing Table)".
Link devices can have optional _SRS/_PRS methods to set the interrupt.
...Quoted 5 lines
However, _PRS/_SRS methods are checked in
drivers/acpi/pci_link.c, and the driver aborts if they are absent.
E.g.: When_PRSis missing:
ACPI: \_SB_.PCI0.LNKA:_CRS36 not found in_PRS
ACPI: \_SB_.PCI0.LNKA: No IRQ available. Try pci=noacpi or acpi=offQuoted 3 lines
I assume this bug report is because something isn't working. Can you update the bugzilla with a note about what specifically isn't working and also attach a complete dmesg log and acpidump?
The question arose while writing link devices code, so there is no platform with missing _PRS/_SRS methods that I know.
The question was more about spec compliance and the necessity to have these methods when legacy interrupts are not configurable. The message above (
_CRSXXX not found in_PRS) can be generated for a Juno for instance, and the ACPI tables are at:
https://github.com/tianocore/edk2-platforms/blob/master/Platform/ARM/JunoPkg/AcpiTables/AcpiSsdtRootPci.asl
The ACPI table need to be modified (remove_PRSand set_CRScorrectly).Quoted 2 lines
If the conclusion is that _PRS/_SRS are mandatory, even for hard-wired interrupts, then the bugzilla can be closed.
OK, so if I understand correctly, you're using Interrupt Link devices not because it's possible to connect a PCI interrupt to one of several inputs on the interrupt controller, but because the PCI default of "level triggered, active low" is not compatible with GICv2.
The Interrupt Link device gives you a chance to specify "level triggered, active *high*". If you used a Source of 0 (where there is no Interrupt Link), there would be no way to specify that.
Since this use of Interrupt Links only conveys triggering information and nothing is configurable, I think the OS should get that info from
_CRS, and_PRSand_SRSshould not be required.
Quoted 7 lines
Alex made a change [1] along that line a while ago, but maybe there's more we should do.
Bjorn
Yes, this is exactly the situation.
The interrupt advertised in _CRS is checked to be in _PRS at:
https://github.com/torvalds/linux/blob/26291c54e111ff6ba87a164d85d4a4e134b7315c/drivers/acpi/pci_link.c#L549
and the _SRS method is also evaluated.
I can submit a patch if necessary,
Pierre
Quoted 55 lines · Pierre Gondois
Quoted 47 lines · Bjorn Helgaas
Quoted 30 lines · Pierre Gondois
Quoted 14 lines · Bjorn Helgaas
Quoted 7 lines · On Wed, Feb 02, 2022 at 10:20:44AM +0000, [email protected] wrote:
https://bugzilla.kernel.org/show_bug.cgi?id=215560
The PCI legacy interrupts can be described with link devices, cf ACPI 6.4, s6.2.13 "
_PRT(PCI Routing Table)".
Link devices can have optional _SRS/_PRS methods to set the interrupt.
...Quoted 5 lines
However, _PRS/_SRS methods are checked in
drivers/acpi/pci_link.c, and the driver aborts if they are absent.
E.g.: When_PRSis missing:
ACPI: \_SB_.PCI0.LNKA:_CRS36 not found in_PRS
ACPI: \_SB_.PCI0.LNKA: No IRQ available. Try pci=noacpi or acpi=offQuoted 3 lines
I assume this bug report is because something isn't working. Can you update the bugzilla with a note about what specifically isn't working and also attach a complete dmesg log and acpidump?
The question arose while writing link devices code, so there is no platform with missing _PRS/_SRS methods that I know.
The question was more about spec compliance and the necessity to have these methods when legacy interrupts are not configurable. The message above (
_CRSXXX not found in_PRS) can be generated for a Juno for instance, and the ACPI tables are at:
https://github.com/tianocore/edk2-platforms/blob/master/Platform/ARM/JunoPkg/AcpiTables/AcpiSsdtRootPci.asl
The ACPI table need to be modified (remove_PRSand set_CRScorrectly).Quoted 2 lines
If the conclusion is that _PRS/_SRS are mandatory, even for hard-wired interrupts, then the bugzilla can be closed.
OK, so if I understand correctly, you're using Interrupt Link devices not because it's possible to connect a PCI interrupt to one of several inputs on the interrupt controller, but because the PCI default of "level triggered, active low" is not compatible with GICv2.
The Interrupt Link device gives you a chance to specify "level triggered, active *high*". If you used a Source of 0 (where there is no Interrupt Link), there would be no way to specify that.
Since this use of Interrupt Links only conveys triggering information and nothing is configurable, I think the OS should get that info from
_CRS, and_PRSand_SRSshould not be required.Quoted 6 lines
Alex made a change [1] along that line a while ago, but maybe there's more we should do.
Bjorn
Quoted 7 lines
Yes, this is exactly the situation.
The interrupt advertised in
_CRSis checked to be in_PRSat:
https://github.com/torvalds/linux/blob/26291c54e111ff6ba87a164d85d4a4e134b7315c/drivers/acpi/pci_link.c#L549
and the_SRSmethod is also evaluated.I can submit a patch if necessary,
That would be awesome. Thanks for pushing on this!
Bjorn