On Tue, Feb 25, 2025 at 06:25:52PM +0000, Gupta, Anshuman wrote: > > -----Original Message----- > > From: Bjorn Helgaas <helgaas@xxxxxxxxxx> > ... redundant headers snipped > > On Mon, Feb 24, 2025 at 10:18:44PM +0530, Anshuman Gupta wrote: > > > Implement _DSM method 10 and _DSM Method 11 as per PCI firmware > > specs > > > section 4.6.10 and 4.6.11. > > > > Please split into two patches, one for each _DSM. Include spec > > citations, e.g., PCI Firmware r3.3, sec 4.6.10. Section numbers > > are not guaranteed to stay consistent across spec revisions, so we > > need both the revision and section number. > > > > Include some descriptive words about the DSM in each subject line, > > e.g., "D3cold Aux Power Limit", "PERST# Assertion Delay". > > > > > Current assumption is only one PCIe Endpoint driver (XeKMD for > > > Battlemage GPU) will request for Aux Power Limit under a given Root > > > Port but theoretically it is possible that other Non-Intel GPU or > > > Non-GPU PCIe Endpoint driver can also request for Aux Power Limit and > > > request to block the core power removal under same Root Port. > > > That will disrupt the Battlemage GPU VRAM Self Refresh. > > > > I guess this is sort of an acknowledgement of the r3.3, sec 4.6.10 spec text > > about system software being responsible for tracking and aggregating > > requests when there are multiple functions below the Downstream Port? > AFAIU apart from multiple function below the Downstream Port (from > same PCIe Card), there can be possibility of another PCie card > connected via a switch to same root port like below topology. > > |-> PCIe PCIe Downstream Port -> End Point Device > Root Port -> PCIe Upstream Port |-> PCIe PCIe Downstream Port -> End Point Device > |-> PCIe PCIe Downstream Port -> PCIe Upstream Port -> PCIe Downstream Port -> *EndPoint Device > > *Endpoint Device from different PCIe card can also request to block the core power removal under same Root Port ? Of course. > How to document such limitation ? > > If so, remove the Battlemage-specific language and just say something about > > the fact that this implementation doesn't do any of that tracking and > > aggregation. ^^ Here's a hint about how to document this. My point is that this has nothing to do with Battlemage in particular, so the text about Battlemage-specific things is a distraction from the real point, which IIUC is this: Note that this implementation assumes only a single device below the Downstream Port because it does not track and aggregate requests from all child devices below the Downstream Port as required by sec 4.6.10. > > > One possible mitigation would be only allowing only first PCIe > > > Non-Bridge Endpoint Function 0 driver to call_DSM method 10.
