Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Bjorn Helgaas <helgaas@kernel.org>
To:     Mika Westerberg <mika.westerberg@linux.intel.com>
Cc:     Sathyanarayanan Kuppuswamy 
        <sathyanarayanan.kuppuswamy@linux.intel.com>,
        linux-pci@vger.kernel.org, linux-kernel@vger.kernel.org,
        Bjorn Helgaas <bhelgaas@google.com>
Subject: [PATCH v3] PCI/portdrv: Allow AER service only for Root Ports & RCECs
Date:   Fri,  9 Dec 2022 18:29:22 -0600
Message-Id: <20221210002922.1749403-1-helgaas@kernel.org>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

From: Bjorn Helgaas <bhelgaas@google.com>

Previously portdrv allowed the AER service for any device with an AER
capability (assuming Linux had control of AER) even though the AER service
driver only attaches to Root Port and RCECs.

Because get_port_device_capability() included AER for non-RP, non-RCEC
devices, we tried to initialize the AER IRQ even though these devices
don't generate AER interrupts.

Intel DG1 and DG2 discrete graphics cards contain a switch leading to a
GPU.  The switch supports AER but not MSI, so initializing an AER IRQ
failed, and portdrv failed to claim the switch port at all.  The GPU itself
could be suspended, but the switch could not be put in a low-power state
because it had no driver.

Don't allow the AER service on non-Root Port, non-Root Complex Event
Collector devices.  This means we won't enable Bus Mastering if the device
doesn't require MSI, the AER service will not appear in sysfs, and the AER
service driver will not bind to the device.

Link: https://lore.kernel.org/r/20221207084105.84947-1-mika.westerberg@linux.intel.com
Based-on-patch-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
---

This is a v3 based on Mika's patch at
https://lore.kernel.org/r/20221207084105.84947-1-mika.westerberg@linux.intel.com

I wouldn't normally kibbitz like this, but I'm hoping to squeeze this into
the v6.2 merge window.

Changes from v2:

  * Test the device type in get_port_device_capability() instead of
    pcie_init_service_irqs().  The benefits are to keep the device type
    checking together (this is similar to the PME test), avoid enabling Bus
    Mastering unnecessarily, avoid exposing the portdrv AER service in
    sysfs, and preventing the AER service driver from binding to devices it
    doesn't need to.

 drivers/pci/pcie/portdrv.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/pci/pcie/portdrv.c b/drivers/pci/pcie/portdrv.c
index a6c4225505d5..8b16e96ec15c 100644
--- a/drivers/pci/pcie/portdrv.c
+++ b/drivers/pci/pcie/portdrv.c
@@ -232,7 +232,9 @@ static int get_port_device_capability(struct pci_dev *dev)
 	}
 
 #ifdef CONFIG_PCIEAER
-	if (dev->aer_cap && pci_aer_available() &&
+	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() &&
 	    (pcie_ports_native || host->native_aer))
 		services |= PCIE_PORT_SERVICE_AER;
 #endif
-- 
2.25.1