|
| 1 | +.. SPDX-License-Identifier: GPL-2.0 |
| 2 | +.. include:: <isonum.txt> |
| 3 | + |
| 4 | +=========================================== |
| 5 | +Compute Express Link Subsystem Maturity Map |
| 6 | +=========================================== |
| 7 | + |
| 8 | +The Linux CXL subsystem tracks the dynamic `CXL specification |
| 9 | +<https://computeexpresslink.org/cxl-specification-landing-page>`_ that |
| 10 | +continues to respond to new use cases with new features, capability |
| 11 | +updates and fixes. At any given point some aspects of the subsystem are |
| 12 | +more mature than others. While the periodic pull requests summarize the |
| 13 | +`work being incorporated each merge window |
| 14 | +<https://lore.kernel.org/linux-cxl/?q=s%3APULL+s%3ACXL+tc%3Atorvalds+NOT+s%3ARe>`_, |
| 15 | +those do not always convey progress relative to a starting point and a |
| 16 | +future end goal. |
| 17 | + |
| 18 | +What follows is a coarse breakdown of the subsystem's major |
| 19 | +responsibilities along with a maturity score. The expectation is that |
| 20 | +the change-history of this document provides an overview summary of the |
| 21 | +subsystem maturation over time. |
| 22 | + |
| 23 | +The maturity scores are: |
| 24 | + |
| 25 | +- [3] Mature: Work in this area is complete and no changes on the horizon. |
| 26 | + Note that this score can regress from one kernel release to the next |
| 27 | + based on new test results or end user reports. |
| 28 | + |
| 29 | +- [2] Stabilizing: Major functionality operational, common cases are |
| 30 | + mature, but known corner cases are still a work in progress. |
| 31 | + |
| 32 | +- [1] Initial: Capability that has exited the Proof of Concept phase, but |
| 33 | + may still have significant gaps to close and fixes to apply as real |
| 34 | + world testing occurs. |
| 35 | + |
| 36 | +- [0] Known gap: Feature is on a medium to long term horizon to |
| 37 | + implement. If the specification has a feature that does not even have |
| 38 | + a '0' score in this document, there is a good chance that no one in |
| 39 | + the linux-cxl@vger.kernel.org community has started to look at it. |
| 40 | + |
| 41 | +- X: Out of scope for kernel enabling, or kernel enabling not required |
| 42 | + |
| 43 | +Feature and Capabilities |
| 44 | +======================== |
| 45 | + |
| 46 | +Enumeration / Provisioning |
| 47 | +-------------------------- |
| 48 | +All of the fundamental enumeration an object model of the subsystem is |
| 49 | +in place, but there are several corner cases that are pending closure. |
| 50 | + |
| 51 | + |
| 52 | +* [2] CXL Window Enumeration |
| 53 | + |
| 54 | + * [0] :ref:`Extended-linear memory-side cache <extended-linear>` |
| 55 | + * [0] Low Memory-hole |
| 56 | + * [0] Hetero-interleave |
| 57 | + |
| 58 | +* [2] Switch Enumeration |
| 59 | + |
| 60 | + * [0] CXL register enumeration link-up dependency |
| 61 | + |
| 62 | +* [2] HDM Decoder Configuration |
| 63 | + |
| 64 | + * [0] Decoder target and granularity constraints |
| 65 | + |
| 66 | +* [2] Performance enumeration |
| 67 | + |
| 68 | + * [3] Endpoint CDAT |
| 69 | + * [3] Switch CDAT |
| 70 | + * [1] CDAT to Core-mm integration |
| 71 | + |
| 72 | + * [1] x86 |
| 73 | + * [0] Arm64 |
| 74 | + * [0] All other arch. |
| 75 | + |
| 76 | + * [0] Shared link |
| 77 | + |
| 78 | +* [2] Hotplug |
| 79 | + (see CXL Window Enumeration) |
| 80 | + |
| 81 | + * [0] Handle Soft Reserved conflicts |
| 82 | + |
| 83 | +* [0] :ref:`RCH link status <rch-link-status>` |
| 84 | +* [0] Fabrics / G-FAM (chapter 7) |
| 85 | +* [0] Global Access Endpoint |
| 86 | + |
| 87 | + |
| 88 | +RAS |
| 89 | +--- |
| 90 | +In many ways CXL can be seen as a standardization of what would normally |
| 91 | +be handled by custom EDAC drivers. The open development here is |
| 92 | +mainly caused by the enumeration corner cases above. |
| 93 | + |
| 94 | +* [3] Component events (OS) |
| 95 | +* [2] Component events (FFM) |
| 96 | +* [1] Endpoint protocol errors (OS) |
| 97 | +* [1] Endpoint protocol errors (FFM) |
| 98 | +* [0] Switch protocol errors (OS) |
| 99 | +* [1] Switch protocol errors (FFM) |
| 100 | +* [2] DPA->HPA Address translation |
| 101 | + |
| 102 | + * [1] XOR Interleave translation |
| 103 | + (see CXL Window Enumeration) |
| 104 | + |
| 105 | +* [1] Memory Failure coordination |
| 106 | +* [0] Scrub control |
| 107 | +* [2] ACPI error injection EINJ |
| 108 | + |
| 109 | + * [0] EINJ v2 |
| 110 | + * [X] Compliance DOE |
| 111 | + |
| 112 | +* [2] Native error injection |
| 113 | +* [3] RCH error handling |
| 114 | +* [1] VH error handling |
| 115 | +* [0] PPR |
| 116 | +* [0] Sparing |
| 117 | +* [0] Device built in test |
| 118 | + |
| 119 | + |
| 120 | +Mailbox commands |
| 121 | +---------------- |
| 122 | + |
| 123 | +* [3] Firmware update |
| 124 | +* [3] Health / Alerts |
| 125 | +* [1] :ref:`Background commands <background-commands>` |
| 126 | +* [3] Sanitization |
| 127 | +* [3] Security commands |
| 128 | +* [3] RAW Command Debug Passthrough |
| 129 | +* [0] CEL-only-validation Passthrough |
| 130 | +* [0] Switch CCI |
| 131 | +* [3] Timestamp |
| 132 | +* [1] PMEM labels |
| 133 | +* [0] PMEM GPF / Dirty Shutdown |
| 134 | +* [0] Scan Media |
| 135 | + |
| 136 | +PMU |
| 137 | +--- |
| 138 | +* [1] Type 3 PMU |
| 139 | +* [0] Switch USP/ DSP, Root Port |
| 140 | + |
| 141 | +Security |
| 142 | +-------- |
| 143 | + |
| 144 | +* [X] CXL Trusted Execution Environment Security Protocol (TSP) |
| 145 | +* [X] CXL IDE (subsumed by TSP) |
| 146 | + |
| 147 | +Memory-pooling |
| 148 | +-------------- |
| 149 | + |
| 150 | +* [1] Hotplug of LDs (via PCI hotplug) |
| 151 | +* [0] Dynamic Capacity Device (DCD) Support |
| 152 | + |
| 153 | +Multi-host sharing |
| 154 | +------------------ |
| 155 | + |
| 156 | +* [0] Hardware coherent shared memory |
| 157 | +* [0] Software managed coherency shared memory |
| 158 | + |
| 159 | +Multi-host memory |
| 160 | +----------------- |
| 161 | + |
| 162 | +* [0] Dynamic Capacity Device Support |
| 163 | +* [0] Sharing |
| 164 | + |
| 165 | +Accelerator |
| 166 | +----------- |
| 167 | + |
| 168 | +* [0] Accelerator memory enumeration HDM-D (CXL 1.1/2.0 Type-2) |
| 169 | +* [0] Accelerator memory enumeration HDM-DB (CXL 3.0 Type-2) |
| 170 | +* [0] CXL.cache 68b (CXL 2.0) |
| 171 | +* [0] CXL.cache 256b Cache IDs (CXL 3.0) |
| 172 | + |
| 173 | +User Flow Support |
| 174 | +----------------- |
| 175 | + |
| 176 | +* [0] HPA->DPA Address translation (need xormaps export solution) |
| 177 | + |
| 178 | +Details |
| 179 | +======= |
| 180 | + |
| 181 | +.. _extended-linear: |
| 182 | + |
| 183 | +* **Extended-linear memory-side cache**: An HMAT proposal to enumerate the presence of a |
| 184 | + memory-side cache where the cache capacity extends the SRAT address |
| 185 | + range capacity. `See the ECN |
| 186 | + <https://lore.kernel.org/linux-cxl/6650e4f835a0e_195e294a8@dwillia2-mobl3.amr.corp.intel.com.notmuch/>`_ |
| 187 | + for more details: |
| 188 | + |
| 189 | +.. _rch-link-status: |
| 190 | + |
| 191 | +* **RCH Link Status**: RCH (Restricted CXL Host) topologies, end up |
| 192 | + hiding some standard registers like PCIe Link Status / Capabilities in |
| 193 | + the CXL RCRB (Root Complex Register Block). |
| 194 | + |
| 195 | +.. _background-commands: |
| 196 | + |
| 197 | +* **Background commands**: The CXL background command mechanism is |
| 198 | + awkward as the single slot is monopolized potentially indefinitely by |
| 199 | + various commands. A `cancel on conflict |
| 200 | + <http://lore.kernel.org/r/66035c2e8ba17_770232948b@dwillia2-xfh.jf.intel.com.notmuch>`_ |
| 201 | + facility is needed to make sure the kernel can ensure forward progress |
| 202 | + of priority commands. |
0 commit comments