[v2] PCI: Expose hidden NVIDIA HDA controllers

Submitted by Daniel Drake on July 8, 2019, 5:17 a.m.


Message ID 20190708051744.24039-1-drake@endlessm.com
State New
Headers show
Series "PCI: Expose hidden NVIDIA HDA controllers" ( rev: 2 ) in Nouveau

Not browsing as part of any series.

Commit Message

Daniel Drake July 8, 2019, 5:17 a.m.
From: Lukas Wunner <lukas@wunner.de>

The integrated HDA controller on Nvidia GPUs can be hidden with a bit in
the GPU's config space. Information about this scheme was provided by
NVIDIA on their forums.

Many laptops now ship with this device hidden, meaning that Linux users
of affected platforms (where the HDMI connector comes off the NVIDIA GPU)
cannot use HDMI audio functionality.

Some platforms have ACPI DSDT code that will make the device visible if
the HDMI cable was connected at boot time, but this does not handle the
hotplug case, and this limitation has also been confirmed under Windows.

Avoid this issue by exposing the HDMI audio device on device enumeration
and resume.

The GPU and HDA controller are two functions of the same PCI device
(VGA class device on function 0 and audio device on function 1).
The multifunction flag in the GPU's Header Type register is cleared when
the HDA controller is hidden and set if it's exposed, so reread the flag
after exposing the HDA.

According to Ilia Mirkin, the HDA controller is only present on GPUs with
PCI ID values from MCP89's onwards, so do not touch config space on older

This quirk is limited to NVIDIA PCI devices with the VGA Controller
device class. This is expected to correspond to product configurations
where the NVIDIA GPU has connectors attached. Other products where the
device class is 3D Controller are expected to correspond to configurations
where the NVIDIA GPU is dedicated (dGPU) and has no connectors.

It's sensible to avoid exposing the HDA controller on dGPU setups,
especially because we've seen cases where the PCI BARs are not set
up correctly by the platform in this case, causing Linux to log
errors if the device is visible. This assumption of device class
accurately corresponding to product configuration is true for 6 of 6
laptops recently checked at the Endless lab, and there are also signs of
agreement checking the data from 74 previously tested products, however
Ilia Mirkin comments that he's seen cases where it is not true. Anyway, it
looks like this quirk should fix audio support for the majority of
affected users.

This commit takes inspiration from an earlier patch by Daniel Drake.

Link: https://devtalk.nvidia.com/default/topic/1024022
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=75985
Cc: Aaron Plattner <aplattner@nvidia.com>
Cc: Peter Wu <peter@lekensteyn.nl>
Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: Karol Herbst <kherbst@redhat.com>
Cc: Maik Freudenberg <hhfeuer@gmx.de>
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Daniel Drake <drake@endlessm.com>

     - Mention in commit message that the ACPI code that controls this bit
       is insufficient (also confirmed on Windows on the buglink)
     - Tweak commit message to clarify the MCP89 comparison, thanks to Ilia

 drivers/pci/quirks.c    | 28 ++++++++++++++++++++++++++++
 include/linux/pci_ids.h |  1 +
 2 files changed, 29 insertions(+)

Patch hide | download patch | download mbox

diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
index 0f16acc323c6..52046b517e2e 100644
--- a/drivers/pci/quirks.c
+++ b/drivers/pci/quirks.c
 			      PCI_CLASS_MULTIMEDIA_HD_AUDIO, 8, quirk_gpu_hda);
+ * Many laptop BIOSes hide the integrated HDA controller on NVIDIA GPUs
+ * via a special bit. This prevents Linux from seeing and using it.
+ * Unhide it here.
+ * https://devtalk.nvidia.com/default/topic/1024022
+ */
+static void quirk_nvidia_hda(struct pci_dev *gpu)
+	u8 hdr_type;
+	u32 val;
+	/* there was no integrated HDA controller before MCP89 */
+	if (gpu->device < PCI_DEVICE_ID_NVIDIA_GEFORCE_320M)
+		return;
+	/* bit 25 at offset 0x488 hides or exposes the HDA controller */
+	pci_read_config_dword(gpu, 0x488, &val);
+	pci_write_config_dword(gpu, 0x488, val | BIT(25));
+	/* the GPU becomes a multifunction device when the HDA is exposed */
+	pci_read_config_byte(gpu, PCI_HEADER_TYPE, &hdr_type);
+	gpu->multifunction = !!(hdr_type & BIT(7));
+			       PCI_BASE_CLASS_DISPLAY, 16, quirk_nvidia_hda);
+			       PCI_BASE_CLASS_DISPLAY, 16, quirk_nvidia_hda);
  * Some IDT switches incorrectly flag an ACS Source Validation error on
  * completions for config read requests even though PCIe r4.0, sec
diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h
index 70e86148cb1e..66898463b81f 100644
--- a/include/linux/pci_ids.h
+++ b/include/linux/pci_ids.h
@@ -1336,6 +1336,7 @@ 
 #define PCI_DEVICE_ID_NVIDIA_NFORCE_MCP77_IDE       0x0759
+#define PCI_DEVICE_ID_NVIDIA_GEFORCE_320M           0x08A0