Re: Issues with "PCI/LINK: Report degraded links via link bandwidth notification"

From: Alex Deucher
Date: Tue Feb 04 2020 - 09:48:05 EST


On Mon, Feb 3, 2020 at 11:38 PM Lukas Wunner <lukas@xxxxxxxxx> wrote:
>
> On Mon, Feb 03, 2020 at 04:16:36PM -0500, Alex Deucher wrote:
> > AMD has had a micro-controller on the GPU handling pcie link speeds
> > and widths dynamically (in addition to GPU clocks and voltages) for
> > about 12 years or so at this point to save power when the GPU is idle
> > and improve performance when it's required. The micro-controller
> > changes the link parameters dynamically based on load independent of
> > the driver. The driver can tweak the heuristics, or even disable the
> > dynamic changes, but by default it's enabled when the driver loads.
> > The ucode for this micro-controller is loaded by the driver so you'll
> > see fixed clocks and widths prior to the driver loading. We'd need
> > some sort of opt out I suppose for periods when the driver has enabled
> > dynamic pcie power management in the micro-controller.
>
> Note that there are *two* bits in the Link Status Register:
>
> * Link Autonomous Bandwidth Status
> "This bit is Set by hardware to indicate that hardware has
> autonomously changed Link speed or width, without the Port
> transitioning through DL_Down status, for reasons other than to
> attempt to correct unreliable Link operation. This bit must be set if
> the Physical Layer reports a speed or width change was initiated by
> the Downstream component that was indicated as an autonomous change."
>
> * Link Bandwidth Management Status
> "This bit is Set by hardware to indicate that either of the
> following has occurred without the Port transitioning through
> DL_Down status. [...] Hardware has changed Link speed or width to
> attempt to correct unreliable Link operation, either through an
> LTSSM timeout or a higher level process."
>
> See PCIe Base Spec 4.0 sec 7.8.8, 7.8.7, 4.2.6.3.3.1.
>
> The two bits generate *separate* interrupts. We only enable the
> interrupt for the latter.
>
> If AMD GPUs generate a Link Bandwidth Management Interrupt upon
> autonomously changing bandwidth for power management reasons
> (instead of to correct unreliability issues), that would be a
> spec violation.
>
> So the question is, do your GPUs violate the spec in this regard?

I don't know off hand. I can ask the firmware team. That said, I
haven't seen any reports of problems with our GPUs with this code
enabled.

Alex