Re: [PATCH] r8169: enable ASPM on Dell platforms

[+cc linux-pci]

On Mon, Sep 15, 2025 at 09:25:39PM +0200, Heiner Kallweit wrote:

On 9/15/2025 3:37 AM, Chia-Lin Kao (AceLan) wrote:

quoted

On Fri, Sep 12, 2025 at 05:30:52PM +0200, Heiner Kallweit wrote:

quoted

On 9/12/2025 9:29 AM, Chia-Lin Kao (AceLan) wrote:

quoted

Enable PCIe ASPM for RTL8169 NICs on Dell platforms that have been
verified to work reliably with this power management feature. The
r8169 driver traditionally disables ASPM to prevent random link
failures and system hangs on problematic hardware.

Dell has validated these product families to work correctly with
RTL NIC ASPM and commits to addressing any ASPM-related issues
with RTL hardware in collaboration with Realtek.

This change enables ASPM for the following Dell product families:
- Alienware
- Dell Laptops/Pro Laptops/Pro Max Laptops
- Dell Desktops/Pro Desktops/Pro Max Desktops
- Dell Pro Rugged Laptops

I'd like to avoid DMI-based whitelists in kernel code. If more system
vendors do it the same way, then this becomes hard to maintain.

I totally understand your point; I also don’t like constantly adding DMI
info to the list. But this list isn’t for a single product name, it’s a
product family that covers a series of products, and it probably won’t
change anytime soon.

quoted

There is already a mechanism for vendors to flag that they successfully
tested ASPM. See c217ab7a3961 ("r8169: enable ASPM L1.2 if system vendor
flags it as safe").

Right, but writing the flag is not applicable for Dell manufacturing
processes.

Can you elaborate on why this doesn't work for Dell?

quoted

quoted

Last but not least ASPM can be (re-)enabled from userspace, using sysfs.

That doesn't sound like a good solution to push the list to userspace.

Dell has already been working with Canonical for more than a decade to
ship their products with r8169 ASPM enabled. Dell has also had lengthy
discussions with Realtek to have this feature enabled by default in the
r8169 driver. I think this is a good opportunity for Dell to work with
Realtek to spot bugs and refine the r8169 driver.

One more option to avoid having to change kernel code with each new
and successfully ASPM-tested system family from any system vendor:

We could define a device property which states that ASPM has been
successfully tested on a system. This device property could be set
via device tree or via ACPI. Then a simple device_property_present()
in the driver would be sufficient.
A device property would also have the advantage that vendors can
control behavior per device, not only per device family.
An ACPI device property could be rolled out via normal BIOS update
for existing systems.

See also:
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/Documentation/firmware-guide/acpi/DSD-properties-rules.rst?h=v6.16.7

Related conversation:
https://lore.kernel.org/r/5b00870c-be1a-42d6-8857-48b89716d5e2@gmail.com (local)