Re: [PATCH v15 4/4] PCI: Centralize pci_aer_available() checking

[Kuppuswamy Sathyanarayanan]

Hi Bjorn,

On 10/2/2026 2:18 PM, Kuppuswamy Sathyanarayanan wrote:
> From: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>
> 
> "pci=noaer" tells us not to use AER.  pci_aer_available() reports that,
> and it also reports the other cases where the OS cannot use AER at all,
> namely CONFIG_PCIEAER=n and MSI being unavailable.
> 
> Set host_bridge->native_aer from pci_aer_available() when we initialize
> the host bridge, so callers only have to look at native_aer and we do not
> have to test pci_aer_available() separately in each of them.
> 
> Do this in pci_init_host_bridge() rather than in acpi_pci_root_create()
> so it also covers host bridges that are not described by ACPI and never
> reach acpi_pci_root_create().
> 
> This subsumes the CONFIG_PCIEPORTBUS check for native_aer, since
> pci_aer_available() is false when CONFIG_PCIEAER=n and PCIEAER depends on
> PCIEPORTBUS.
> 
> Signed-off-by: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>
> Co-developed-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx>
> Signed-off-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx>
> ---

Sashiko's comment [1] looks valid to me. quirk_disable_all_msi() calls
pci_no_msi() during pci_bus_add_devices(), after native_aer has been
set. That leaves native_aer stale. At boot, pcie_aer_init() checks
pci_aer_available() again, so the AER driver still won't register. But
other native_aer users see the stale value.

The patch below clears native_aer in pci_no_msi(). Would you like a v16
with it before this patch, or would you rather fold it in?

[1] https://lore.kernel.org/linux-pci/20261003013359.A38F61F00898@xxxxxxxxxxxxxxx/

Author: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx>
Date:   Mon Oct 5 10:29:19 2026 -0700

    PCI/AER: Clear native_aer when MSI is disabled after host bridge init

    pci_aer_available() reports that AER is unusable when MSI is disabled.
    A quirk such as quirk_disable_all_msi() can call pci_no_msi() during
    enumeration, after pci_init_host_bridge() and the _OSC negotiation have
    already set host_bridge->native_aer. In that case native_aer stays set
    even though the OS can no longer use AER.

    Clear native_aer on all registered host bridges when MSI is disabled.
    Host bridges added later start with native_aer cleared because
    pci_aer_available() is already false. _OSC negotiation and
    "pcie_ports=native" can only clear native_aer, so they cannot set it
    again.

    This keeps native_aer accurate so callers can rely on it instead of
    checking pci_aer_available() separately.

    Reported-by: sashiko-bot@xxxxxxxxxx
    Closes: https://lore.kernel.org/linux-pci/20261003013359.A38F61F00898@xxxxxxxxxxxxxxx/
    Signed-off-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx>

diff --git a/drivers/pci/msi/msi.c b/drivers/pci/msi/msi.c
index 80a9db417dc8..911e61fbda31 100644
--- a/drivers/pci/msi/msi.c
+++ b/drivers/pci/msi/msi.c
@@ -995,4 +995,5 @@ EXPORT_SYMBOL(msi_desc_to_pci_dev);
 void pci_no_msi(void)
 {
        pci_msi_enable = false;
+       pci_aer_no_msi();
 }
diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
index ba3c3fddddc2..11f47d5fda21 100644
--- a/drivers/pci/pci.h
+++ b/drivers/pci/pci.h
@@ -1331,6 +1331,7 @@ static inline void of_pci_remove_host_bridge_node(struct pci_host_bridge *bridge

 #ifdef CONFIG_PCIEAER
 void pci_no_aer(void);
+void pci_aer_no_msi(void);
 void pci_aer_init(struct pci_dev *dev);
 void pci_aer_exit(struct pci_dev *dev);
 extern const struct attribute_group aer_stats_attr_group;
@@ -1342,6 +1343,7 @@ void pci_save_aer_state(struct pci_dev *dev);
 void pci_restore_aer_state(struct pci_dev *dev);
 #else
 static inline void pci_no_aer(void) { }
+static inline void pci_aer_no_msi(void) { }
 static inline void pci_aer_init(struct pci_dev *d) { }
 static inline void pci_aer_exit(struct pci_dev *d) { }
 static inline void pci_aer_clear_fatal_status(struct pci_dev *dev) { }
diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
index e84dd686582a..494fdd20798d 100644
--- a/drivers/pci/pcie/aer.c
+++ b/drivers/pci/pcie/aer.c
@@ -154,6 +154,19 @@ bool pci_aer_available(void)
        return !pcie_aer_disable && pci_msi_enabled();
 }

+/*
+ * AER depends on MSI (see pci_aer_available()).  If MSI is disabled after
+ * host bridges have been initialized, e.g., by a quirk, the OS can no
+ * longer use AER on them.
+ */
+void pci_aer_no_msi(void)
+{
+       struct pci_bus *bus = NULL;
+
+       while ((bus = pci_find_next_bus(bus)))
+               pci_find_host_bridge(bus)->native_aer = 0;
+}
+
 #ifdef CONFIG_PCIE_ECRC

 #define ECRC_POLICY_DEFAULT 0          /* ECRC set by BIOS */


> Changes since v14
> 
>  * No change.
> 
> v14 posting
> https://lore.kernel.org/r/20260922204548.3884906-1-sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx
> 
> Changes since v13
> 
>  * No change.
> 
> v13 posting
> https://lore.kernel.org/r/20260919162655.3499010-1-sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx
> 
>  drivers/pci/pcie/portdrv.c | 3 +--
>  drivers/pci/probe.c        | 2 +-
>  2 files changed, 2 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/pci/pcie/portdrv.c b/drivers/pci/pcie/portdrv.c
> index 32fc623dd410..9f8c6dd434c5 100644
> --- a/drivers/pci/pcie/portdrv.c
> +++ b/drivers/pci/pcie/portdrv.c
> @@ -239,8 +239,7 @@ static int get_port_device_capability(struct pci_dev *dev)
>  #ifdef CONFIG_PCIEAER
>  	if ((pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT ||
>               pci_pcie_type(dev) == PCI_EXP_TYPE_RC_EC) &&
> -	    dev->aer_cap && pci_aer_available() &&
> -	    host->native_aer)
> +	    dev->aer_cap && host->native_aer)
>  		services |= PCIE_PORT_SERVICE_AER;
>  #endif
>  
> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> index 1e487a213eb0..e1ca8096bcd5 100644
> --- a/drivers/pci/probe.c
> +++ b/drivers/pci/probe.c
> @@ -670,7 +670,7 @@ static void pci_init_host_bridge(struct pci_host_bridge *bridge)
>  	 * may implement its own AER handling and use _OSC to prevent the
>  	 * OS from interfering.
>  	 */
> -	bridge->native_aer = port_services;
> +	bridge->native_aer = pci_aer_available();
>  	bridge->native_pcie_hotplug = port_services;
>  	bridge->native_shpc_hotplug = 1;
>  	bridge->native_pme = port_services;

-- 
Sathyanarayanan Kuppuswamy
Linux Kernel Developer