diff mbox series

[v2] x86/pci: Stop requiring ECAM to be declared in E820, ACPI or EFI

Message ID 20231215220343.22523-1-mario.limonciello@amd.com
State New
Headers show
Series [v2] x86/pci: Stop requiring ECAM to be declared in E820, ACPI or EFI | expand

Commit Message

Mario Limonciello Dec. 15, 2023, 10:03 p.m. UTC 
commit 7752d5cfe3d1 ("x86: validate against acpi motherboard resources")
introduced checks for ensuring that MCFG table also has memory region
reservations to ensure no conflicts were introduced from a buggy BIOS.

This has proceeded over time to add other types of reservation checks
for ACPI PNP resources and EFI MMIO memory type.  The PCI firmware spec
does say that these checks are only required when the operating system
doesn't comprehend the firmware region:

```
If the operating system does not natively comprehend reserving the MMCFG
region, the MMCFG region must be reserved by firmware. The address range
reported in the MCFG table or by _CBA method (see Section 4.1.3) must be
reserved by declaring a motherboard resource. For most systems, the
motherboard resource would appear at the root of the ACPI namespace
(under \_SB) in a node with a _HID of EISAID (PNP0C02), and the resources
in this case should not be claimed in the root PCI bus’s _CRS. The
resources can optionally be returned in Int15 E820h or EFIGetMemoryMap
as reserved memory but must always be reported through ACPI as a
motherboard resource.
```

Running this check causes problems with accessing extended PCI
configuration space on OEM laptops that don't specify the region in PNP
resources or in the EFI memory map. That later manifests as problems with
dGPU and accessing resizable BAR. Similar problems don't exist in Windows
11 with exact same laptop/firmware stack.

Due to the stability of the Windows ecosystem that x86 machines participate
it is unlikely that using the region specified in the MCFG table as
a reservation will cause a problem. The possible worst circumstance could
be that a buggy BIOS causes a larger hole in the memory map that is
unusable for devices than intended.

Change the default behavior to keep the region specified in MCFG even if
it's not specified in another source. This is expected to improve
machines that otherwise couldn't access PCI extended configuration space.

In case this change causes problems, add a kernel command line parameter
that can restore the previous behavior.

Link: https://members.pcisig.com/wg/PCI-SIG/document/15350
      PCI Firmware Specification 3.3
      Section 4.1.2 MCFG Table Description Note 2
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
---
v1->v2:
 * Rebase on pci/next
 * Add an escape hatch
 * Reword commit message
---
 .../admin-guide/kernel-parameters.txt         |  6 ++++++
 arch/x86/pci/mmconfig-shared.c                | 19 +++++++++++++++----
 2 files changed, 21 insertions(+), 4 deletions(-)


base-commit: 67e04d921cb6902e8c2abdbf748279d43f25213e

Comments

Mario Limonciello Jan. 17, 2024, 5:53 p.m. UTC | #1 
On 12/15/2023 16:03, Mario Limonciello wrote:
> commit 7752d5cfe3d1 ("x86: validate against acpi motherboard resources")
> introduced checks for ensuring that MCFG table also has memory region
> reservations to ensure no conflicts were introduced from a buggy BIOS.
> 
> This has proceeded over time to add other types of reservation checks
> for ACPI PNP resources and EFI MMIO memory type.  The PCI firmware spec
> does say that these checks are only required when the operating system
> doesn't comprehend the firmware region:
> 
> ```
> If the operating system does not natively comprehend reserving the MMCFG
> region, the MMCFG region must be reserved by firmware. The address range
> reported in the MCFG table or by _CBA method (see Section 4.1.3) must be
> reserved by declaring a motherboard resource. For most systems, the
> motherboard resource would appear at the root of the ACPI namespace
> (under \_SB) in a node with a _HID of EISAID (PNP0C02), and the resources
> in this case should not be claimed in the root PCI bus’s _CRS. The
> resources can optionally be returned in Int15 E820h or EFIGetMemoryMap
> as reserved memory but must always be reported through ACPI as a
> motherboard resource.
> ```
> 
> Running this check causes problems with accessing extended PCI
> configuration space on OEM laptops that don't specify the region in PNP
> resources or in the EFI memory map. That later manifests as problems with
> dGPU and accessing resizable BAR. Similar problems don't exist in Windows
> 11 with exact same laptop/firmware stack.
> 
> Due to the stability of the Windows ecosystem that x86 machines participate
> it is unlikely that using the region specified in the MCFG table as
> a reservation will cause a problem. The possible worst circumstance could
> be that a buggy BIOS causes a larger hole in the memory map that is
> unusable for devices than intended.
> 
> Change the default behavior to keep the region specified in MCFG even if
> it's not specified in another source. This is expected to improve
> machines that otherwise couldn't access PCI extended configuration space.
> 
> In case this change causes problems, add a kernel command line parameter
> that can restore the previous behavior.
> 
> Link: https://members.pcisig.com/wg/PCI-SIG/document/15350
>        PCI Firmware Specification 3.3
>        Section 4.1.2 MCFG Table Description Note 2
> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
> ---

Bjorn,

Any thoughts on this version since our last conversation on V1?

Thanks,

> v1->v2:
>   * Rebase on pci/next
>   * Add an escape hatch
>   * Reword commit message
> ---
>   .../admin-guide/kernel-parameters.txt         |  6 ++++++
>   arch/x86/pci/mmconfig-shared.c                | 19 +++++++++++++++----
>   2 files changed, 21 insertions(+), 4 deletions(-)
> 
> diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> index 65731b060e3f..eacd0c0521c2 100644
> --- a/Documentation/admin-guide/kernel-parameters.txt
> +++ b/Documentation/admin-guide/kernel-parameters.txt
> @@ -1473,6 +1473,12 @@
>   			(in particular on some ATI chipsets).
>   			The kernel tries to set a reasonable default.
>   
> +	enforce_ecam_resv [X86]
> +			Enforce requiring an ECAM reservation specified in
> +			BIOS for PCI devices.
> +			This parameter is only valid if CONFIG_PCI_MMCONFIG
> +			is enabled.
> +
>   	enforcing=	[SELINUX] Set initial enforcing status.
>   			Format: {"0" | "1"}
>   			See security/selinux/Kconfig help text.
> diff --git a/arch/x86/pci/mmconfig-shared.c b/arch/x86/pci/mmconfig-shared.c
> index 0cc9520666ef..aee117c6bbf9 100644
> --- a/arch/x86/pci/mmconfig-shared.c
> +++ b/arch/x86/pci/mmconfig-shared.c
> @@ -34,6 +34,15 @@ static DEFINE_MUTEX(pci_mmcfg_lock);
>   
>   LIST_HEAD(pci_mmcfg_list);
>   
> +static bool enforce_ecam_resv __read_mostly;
> +static int __init parse_ecam_options(char *str)
> +{
> +	enforce_ecam_resv = true;
> +
> +	return 1;
> +}
> +__setup("enforce_ecam_resv", parse_ecam_options);
> +
>   static void __init pci_mmconfig_remove(struct pci_mmcfg_region *cfg)
>   {
>   	if (cfg->res.parent)
> @@ -569,10 +578,12 @@ static void __init pci_mmcfg_reject_broken(int early)
>   
>   	list_for_each_entry(cfg, &pci_mmcfg_list, list) {
>   		if (!pci_mmcfg_reserved(NULL, cfg, early)) {
> -			pr_info("not using ECAM (%pR not reserved)\n",
> -				&cfg->res);
> -			free_all_mmcfg();
> -			return;
> +			pr_info("ECAM %pR not reserved, %s\n", &cfg->res,
> +				enforce_ecam_resv ? "ignoring" : "using anyway");
> +			if (enforce_ecam_resv) {
> +				free_all_mmcfg();
> +				return;
> +			}
>   		}
>   	}
>   }
> 
> base-commit: 67e04d921cb6902e8c2abdbf748279d43f25213e
Bjorn Helgaas Jan. 26, 2024, 12:35 a.m. UTC | #2 
On Wed, Jan 17, 2024 at 11:53:50AM -0600, Mario Limonciello wrote:
> On 12/15/2023 16:03, Mario Limonciello wrote:
> > commit 7752d5cfe3d1 ("x86: validate against acpi motherboard resources")
> > introduced checks for ensuring that MCFG table also has memory region
> > reservations to ensure no conflicts were introduced from a buggy BIOS.
> > 
> > This has proceeded over time to add other types of reservation checks
> > for ACPI PNP resources and EFI MMIO memory type.  The PCI firmware spec
> > does say that these checks are only required when the operating system
> > doesn't comprehend the firmware region:
> > 
> > ```
> > If the operating system does not natively comprehend reserving the MMCFG
> > region, the MMCFG region must be reserved by firmware. The address range
> > reported in the MCFG table or by _CBA method (see Section 4.1.3) must be
> > reserved by declaring a motherboard resource. For most systems, the
> > motherboard resource would appear at the root of the ACPI namespace
> > (under \_SB) in a node with a _HID of EISAID (PNP0C02), and the resources
> > in this case should not be claimed in the root PCI bus’s _CRS. The
> > resources can optionally be returned in Int15 E820h or EFIGetMemoryMap
> > as reserved memory but must always be reported through ACPI as a
> > motherboard resource.
> > ```
> > 
> > Running this check causes problems with accessing extended PCI
> > configuration space on OEM laptops that don't specify the region in PNP
> > resources or in the EFI memory map. That later manifests as problems with
> > dGPU and accessing resizable BAR. Similar problems don't exist in Windows
> > 11 with exact same laptop/firmware stack.
> > 
> > Due to the stability of the Windows ecosystem that x86 machines participate
> > it is unlikely that using the region specified in the MCFG table as
> > a reservation will cause a problem. The possible worst circumstance could
> > be that a buggy BIOS causes a larger hole in the memory map that is
> > unusable for devices than intended.
> > 
> > Change the default behavior to keep the region specified in MCFG even if
> > it's not specified in another source. This is expected to improve
> > machines that otherwise couldn't access PCI extended configuration space.
> > 
> > In case this change causes problems, add a kernel command line parameter
> > that can restore the previous behavior.
> > 
> > Link: https://members.pcisig.com/wg/PCI-SIG/document/15350
> >        PCI Firmware Specification 3.3
> >        Section 4.1.2 MCFG Table Description Note 2
> > Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
> > ---
> 
> Bjorn,
> 
> Any thoughts on this version since our last conversation on V1?

Just to let you know that I'm not ignoring this, and I'm trying to
formulate a useful response.  mmconfig-shared.c has grown into an
extremely complicated mess and is a continual source of problems, so
I'd really like to untangle it a tiny bit if we can.

One thing is that per spec, ACPI motherboard resources are the ONLY
way to reserve ECAM space.  I'd like to reclaim a little clarity about
that and separate out the E820 and EFI stuff as secondary hacks.  But
there's an insane amount of history that got us here.

The problem we have to avoid is assigning a BAR that overlaps ECAM.
We assign BARs for lots of reasons.  dGPU and resizable BARs are
examples but there are others, like hotplug and SR-IOV.  The fact that
Windows works is a red herring because it allocates BARs differently.

It's also little bit of a burr under my saddle to add things for a
problem on unspecified machines, where I don't even know whether the
machines are already in the field or the firmware could still be
fixed.

And of course, if there's any way to solve this safely without adding
yet another kernel parameter, that would be vastly preferable.

Sorry, nothing actionable here but wanted to let you know that it's
keeping me awake at night ;)

Bjorn

> > v1->v2:
> >   * Rebase on pci/next
> >   * Add an escape hatch
> >   * Reword commit message
> > ---
> >   .../admin-guide/kernel-parameters.txt         |  6 ++++++
> >   arch/x86/pci/mmconfig-shared.c                | 19 +++++++++++++++----
> >   2 files changed, 21 insertions(+), 4 deletions(-)
> > 
> > diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> > index 65731b060e3f..eacd0c0521c2 100644
> > --- a/Documentation/admin-guide/kernel-parameters.txt
> > +++ b/Documentation/admin-guide/kernel-parameters.txt
> > @@ -1473,6 +1473,12 @@
> >   			(in particular on some ATI chipsets).
> >   			The kernel tries to set a reasonable default.
> > +	enforce_ecam_resv [X86]
> > +			Enforce requiring an ECAM reservation specified in
> > +			BIOS for PCI devices.
> > +			This parameter is only valid if CONFIG_PCI_MMCONFIG
> > +			is enabled.
> > +
> >   	enforcing=	[SELINUX] Set initial enforcing status.
> >   			Format: {"0" | "1"}
> >   			See security/selinux/Kconfig help text.
> > diff --git a/arch/x86/pci/mmconfig-shared.c b/arch/x86/pci/mmconfig-shared.c
> > index 0cc9520666ef..aee117c6bbf9 100644
> > --- a/arch/x86/pci/mmconfig-shared.c
> > +++ b/arch/x86/pci/mmconfig-shared.c
> > @@ -34,6 +34,15 @@ static DEFINE_MUTEX(pci_mmcfg_lock);
> >   LIST_HEAD(pci_mmcfg_list);
> > +static bool enforce_ecam_resv __read_mostly;
> > +static int __init parse_ecam_options(char *str)
> > +{
> > +	enforce_ecam_resv = true;
> > +
> > +	return 1;
> > +}
> > +__setup("enforce_ecam_resv", parse_ecam_options);
> > +
> >   static void __init pci_mmconfig_remove(struct pci_mmcfg_region *cfg)
> >   {
> >   	if (cfg->res.parent)
> > @@ -569,10 +578,12 @@ static void __init pci_mmcfg_reject_broken(int early)
> >   	list_for_each_entry(cfg, &pci_mmcfg_list, list) {
> >   		if (!pci_mmcfg_reserved(NULL, cfg, early)) {
> > -			pr_info("not using ECAM (%pR not reserved)\n",
> > -				&cfg->res);
> > -			free_all_mmcfg();
> > -			return;
> > +			pr_info("ECAM %pR not reserved, %s\n", &cfg->res,
> > +				enforce_ecam_resv ? "ignoring" : "using anyway");
> > +			if (enforce_ecam_resv) {
> > +				free_all_mmcfg();
> > +				return;
> > +			}
> >   		}
> >   	}
> >   }
> > 
> > base-commit: 67e04d921cb6902e8c2abdbf748279d43f25213e
>
Bjorn Helgaas Feb. 12, 2024, 10:29 p.m. UTC | #3 
[+cc Ivan in case there's opportunity to improve FWTS]

On Wed, Jan 17, 2024 at 11:53:50AM -0600, Mario Limonciello wrote:
> On 12/15/2023 16:03, Mario Limonciello wrote:
> > commit 7752d5cfe3d1 ("x86: validate against acpi motherboard resources")
> > introduced checks for ensuring that MCFG table also has memory region
> > reservations to ensure no conflicts were introduced from a buggy BIOS.
> > 
> > This has proceeded over time to add other types of reservation checks
> > for ACPI PNP resources and EFI MMIO memory type.  The PCI firmware spec
> > does say that these checks are only required when the operating system
> > doesn't comprehend the firmware region:
> > 
> > ```
> > If the operating system does not natively comprehend reserving the MMCFG
> > region, the MMCFG region must be reserved by firmware. The address range
> > reported in the MCFG table or by _CBA method (see Section 4.1.3) must be
> > reserved by declaring a motherboard resource. For most systems, the
> > motherboard resource would appear at the root of the ACPI namespace
> > (under \_SB) in a node with a _HID of EISAID (PNP0C02), and the resources
> > in this case should not be claimed in the root PCI bus’s _CRS. The
> > resources can optionally be returned in Int15 E820h or EFIGetMemoryMap
> > as reserved memory but must always be reported through ACPI as a
> > motherboard resource.
> > ```
> > 
> > Running this check causes problems with accessing extended PCI
> > configuration space on OEM laptops that don't specify the region in PNP
> > resources or in the EFI memory map. That later manifests as problems with
> > dGPU and accessing resizable BAR. Similar problems don't exist in Windows
> > 11 with exact same laptop/firmware stack.
> > 
> > Due to the stability of the Windows ecosystem that x86 machines participate
> > it is unlikely that using the region specified in the MCFG table as
> > a reservation will cause a problem. The possible worst circumstance could
> > be that a buggy BIOS causes a larger hole in the memory map that is
> > unusable for devices than intended.
> > 
> > Change the default behavior to keep the region specified in MCFG even if
> > it's not specified in another source. This is expected to improve
> > machines that otherwise couldn't access PCI extended configuration space.
> > 
> > In case this change causes problems, add a kernel command line parameter
> > that can restore the previous behavior.
> > 
> > Link: https://members.pcisig.com/wg/PCI-SIG/document/15350
> >        PCI Firmware Specification 3.3
> >        Section 4.1.2 MCFG Table Description Note 2
> > Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
> > ---
> 
> Bjorn,
> 
> Any thoughts on this version since our last conversation on V1?

I really want to clarify the dmesg logging such that it's clear that
PNP0C02 reservation is the only valid way to reserve the space
described by MCFG.  Obviously we have to retain the fallbacks, but I
think there should be FW_BUG logging in that case.  We currently only
do FW_INFO for missing PNP0C02 reservations.

I think we should try to change FWTS so it validates MCFG addresses
against the PNP0C02 reservations required by spec, instead of
searching E820 for them.  The spec doesn't require MCFG regions to be
in E820, and I think searching there encourages the wrong behavior.
It probably also doesn't work at all on arm64, since it doesn't have
E820 at all.

The /sys/devices/pnp0/00:xx/resources files and "system 00:xx: [mem
...] has been reserved" lines in dmesg would be much better places to
check.

Bjorn
diff mbox series

Patch

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 65731b060e3f..eacd0c0521c2 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -1473,6 +1473,12 @@ 
 			(in particular on some ATI chipsets).
 			The kernel tries to set a reasonable default.
 
+	enforce_ecam_resv [X86]
+			Enforce requiring an ECAM reservation specified in
+			BIOS for PCI devices.
+			This parameter is only valid if CONFIG_PCI_MMCONFIG
+			is enabled.
+
 	enforcing=	[SELINUX] Set initial enforcing status.
 			Format: {"0" | "1"}
 			See security/selinux/Kconfig help text.
diff --git a/arch/x86/pci/mmconfig-shared.c b/arch/x86/pci/mmconfig-shared.c
index 0cc9520666ef..aee117c6bbf9 100644
--- a/arch/x86/pci/mmconfig-shared.c
+++ b/arch/x86/pci/mmconfig-shared.c
@@ -34,6 +34,15 @@  static DEFINE_MUTEX(pci_mmcfg_lock);
 
 LIST_HEAD(pci_mmcfg_list);
 
+static bool enforce_ecam_resv __read_mostly;
+static int __init parse_ecam_options(char *str)
+{
+	enforce_ecam_resv = true;
+
+	return 1;
+}
+__setup("enforce_ecam_resv", parse_ecam_options);
+
 static void __init pci_mmconfig_remove(struct pci_mmcfg_region *cfg)
 {
 	if (cfg->res.parent)
@@ -569,10 +578,12 @@  static void __init pci_mmcfg_reject_broken(int early)
 
 	list_for_each_entry(cfg, &pci_mmcfg_list, list) {
 		if (!pci_mmcfg_reserved(NULL, cfg, early)) {
-			pr_info("not using ECAM (%pR not reserved)\n",
-				&cfg->res);
-			free_all_mmcfg();
-			return;
+			pr_info("ECAM %pR not reserved, %s\n", &cfg->res,
+				enforce_ecam_resv ? "ignoring" : "using anyway");
+			if (enforce_ecam_resv) {
+				free_all_mmcfg();
+				return;
+			}
 		}
 	}
 }