What if there is a CXL switch with a Generic Initiator and CXL type 3 memory which could be hotplugged in. I forget if the GI to type 3 memory path is ONLY through the host bridge today in 2.0 or also allowed through the switch. In future we would want it allowed through the switch for sure, just like PCIe p2p. So how would the switch route latency/BW be discovered? In the current CDAT spec, a coherent switch is modelled as having two or more bidirectional ports that carry traffic between host CPUs, coherent accelerators, and coherent memory devices. Switch CDAT returns latency and BW between various port pairs and can describe what you are looking for. Thank you, Mahesh Natu Datacenter Platform Architect Intel Corporation -----Original Message----- From: Vikram Sethi <vsethi@xxxxxxxxxx> Sent: Wednesday, February 10, 2021 9:02 AM To: Williams, Dan J <dan.j.williams@xxxxxxxxx>; linux-cxl@xxxxxxxxxxxxxxx; Linux ACPI <linux-acpi@xxxxxxxxxxxxxxx> Cc: Natu, Mahesh <mahesh.natu@xxxxxxxxx>; Douglas, Chet R <chet.r.douglas@xxxxxxxxx>; Widawsky, Ben <ben.widawsky@xxxxxxxxx>; Verma, Vishal L <vishal.l.verma@xxxxxxxxx> Subject: RE: [RFC] ACPI Code First ECR: Generic Target Hi Dan, > -----Original Message----- > From: Dan Williams <dan.j.williams@xxxxxxxxx> > Sent: Tuesday, February 9, 2021 9:55 PM > To: linux-cxl@xxxxxxxxxxxxxxx; Linux ACPI <linux-acpi@xxxxxxxxxxxxxxx> > Cc: Natu, Mahesh <mahesh.natu@xxxxxxxxx>; Chet R Douglas > <chet.r.douglas@xxxxxxxxx>; Ben Widawsky <ben.widawsky@xxxxxxxxx>; > Vishal L Verma <vishal.l.verma@xxxxxxxxx> > Subject: [RFC] ACPI Code First ECR: Generic Target > > External email: Use caution opening links or attachments > > > While the platform BIOS is able to describe the performance > characteristics of CXL memory that is present at boot, it is unable to > statically enumerate the performance of CXL memory hot inserted > post-boot. The OS can enumerate most of the characteristics from link > registers and CDAT, but the performance from the CPU to the host > bridge, for example, is not enumerated by PCIE or CXL. Introduce an > ACPI mechanism for this purpose. Critically this is achieved with a > small tweak to how the existing Generic Initiator proximity domain is > utilized in the HMAT. > > --- > > # Title: Introduce a Generic Target for CXL > > # Status: Draft > > # Document: ACPI Specification 6.4 > > # License > SPDX-License Identifier: CC-BY-4.0 > > # Submitter: > * Sponsor: Dan Williams, Intel > * Creators/Contributors: > * Mahesh Natu, Intel > * Chet Douglas, Intel > * Deepak Shivakumar, Intel > > # Summary of the Change > Introduce a "Generic Target" concept to the SRAT to describe the root > performance parameters in the path to dynamically discovered (outside > of ACPI enumeration) CXL memory target endpoints. > > # Benefits of the Change > Consider the case of a system with a set of CXL host bridges > (ACPI0016), and no devices attached at initial system power-on. In > this scenario platform firmware is unable to perform the end-to-end > enumeration necessary to populate SRAT and HMAT for the endpoints that > may be hot-inserted behind those bridges post power-on. The > address-range is unknown so SRAT can not be pre-populated, the > performance is unknown (no CDAT nor interleave configuration) so HMAT can not be pre-populated. > > However, what is known to platform firmware that generates the SRAT > and HMAT is the performance characteristics of the path between CPU > and Generic Initiators to the CXL host bridge target. With either > CPU-to-Generic-Target, or Generic-Initiator-to-Generic-Target entries > in the HMAT the OS CXL subsystem can enumerate the remaining details > (PCIE link status, device CDAT, interleave configuration) to calculate > the bandwidth and latency of a dynamically discovered CXL memory target. > What if there is a CXL switch with a Generic Initiator and CXL type 3 memory which could be hotplugged in. I forget if the GI to type 3 memory path is ONLY through the host bridge today in 2.0 or also allowed through the switch. In future we would want it allowed through the switch for sure, just like PCIe p2p. So how would the switch route latency/BW be discovered? Also, an example with numbers of what would be in HMAT may help understand the case where everything is only connected via host bridge also. > # Impact of the Change > The existing Generic Initiator Affinity Structure (ACPI 6.4 Section > 5.2.16.6) already contains all the fields necessary to enumerate a > generic target proximity domain. All that is missing is the > interpretation of that proximity domain optionally as a target > identifier in the HMAT. > > Given that the OS still needs to dynamically enumerate and instantiate > the memory ranges behind the host bridge. The assumption is that > operating systems that do not support native CXL enumeration will > ignore this data in the HMAT, while CXL native enumeration aware > environments will use this fragment of the performance path to > calculate the performance characteristics. > > # References > * Compute Express Link Specification v2.0, > <https://www.computeexpresslink.org/> > > # Detailed Description of the Change > > * Replace "Generic Initiator" with "Generic Initiator / Target" in all > locations except where an "initiator" or "target" is implied. > Specifically 5.2.27.3 "Memory Proximity Domain Attributes Structure" > need not replace occurrences of "generic initiator" in field: > "Proximity Domain for Attached Initiator". Additionally field: > "Proximity Domain for the Memory" must be renamed to "Proximity Domain > for the Memory / Generic Target" with a new description "Integer that > represents the memory / generic target proximity domain to which this memory belongs." > > * Revise "5.2.16.6 Generic Initiator Affinity Structure" to make it > consistent with being referenced as either a target or initiator. > > * Description: (replace all text) > > > The Generic Initiator / Target Affinity Structure provides the > > association between a Generic Initiator and a Memory Proximity > > Domain, or another Generic Target Proximity Domain. The > > distinction as to whether this structure represents an > > Initiator, a Target, or both depends on how it is referenced > > in the HMAT. See Section 5.2.27.3 for details. > > > Support of Generic Initiator / Target Affinity Structures by > > OSPM is optional, and the platform may query whether the OS > > supports it via the _OSC method. See Section 6.2.11.2. > > * Architectural transactions: (append after current text) > > > If this proximity domain is referenced as a target then it > > supports all the transaction types inferred above. > > * Other updates are simple Initiator => Initiator / Target > replacements.