From 755fa6d02f8d0f80582722bc08db09a70519de49 Mon Sep 17 00:00:00 2001 From: Adam360x Date: Tue, 20 Oct 2026 20:19:47 -0500 Subject: [PATCH 1/3] docs(rdc): split the RDC 1.3.1 and 1.4.0 changelog sections RDC 1.4.0 ships in ROCm 10.2.0, not ROCm 10.1.0. ROCm 10.1.0 shipped RDC 1.3.1, which already contains the GetBlockNameStr() removal and the ECC totals fix (#10973). Move those two entries to a new "RDC 1.3.1 for ROCm 10.1.0" section and retitle the 1.4.0 section for ROCm 10.2.0. The RDC_FI_GPU_MEMORY_CUR_BANDWIDTH fix (#11634) stays under 1.4.0. It merged after the ROCm 10.1 branch point, so ROCm 10.1.0 does not contain it. Signed-off-by: Adam360x --- projects/rdc/CHANGELOG.md | 18 +++++++++++------- 1 file changed, 11 insertions(+), 7 deletions(-) diff --git a/projects/rdc/CHANGELOG.md b/projects/rdc/CHANGELOG.md index 204cb6c4def7..1bc4496abbf1 100644 --- a/projects/rdc/CHANGELOG.md +++ b/projects/rdc/CHANGELOG.md @@ -2,17 +2,13 @@ Full documentation for RDC is available at [ROCm DataCenter Tool User Guide](https://rocm.docs.amd.com/projects/rdc/en/latest/). -## RDC 1.4.0 for ROCm 10.1.0 +## RDC 1.4.0 for ROCm 10.2.0 ### Changed - **`rdc_health_get()` returns `RDC_ST_GROUP_NOT_FOUND` for a GPU group that does not exist**. - It returned `RDC_ST_OK` with no health watches, so `rdci health -f` showed every watch as Off and exited successfully, and `rdci health -c` failed asking for watches to be enabled. Both commands now fail with a message that the group does not exist. -### Removed - -- Removed the unused `GetBlockNameStr()` test helper and its GPU block name map. Nothing called it, and its `static_assert` on `AMDSMI_GPU_BLOCK_LAST` broke the RDC build whenever AMD SMI added an IP block (#10949). - ### Resolved Issues - `rdci health -s` on a GPU group that does not exist no longer leaves an unused field group in rdcd. Repeated failures filled the field group table, and commands that need a new field group, such as `rdci dmon`, then failed until rdcd was restarted. @@ -32,8 +28,6 @@ Full documentation for RDC is available at [ROCm DataCenter Tool User Guide](htt - `rdci topo` and `rdc_device_topology_get()` now report the links of the requested GPU. They reported links between other GPU pairs, which showed the wrong link types on systems where GPU pairs are not all linked the same way, such as XGMI hives joined by PCIe. - `rdci topo` now exits with an error code when the GPU index is out of range or the topology query fails, instead of exiting 0, and `rdc_device_topology_get()` reports link types RDC does not define as `RDC_IOLINK_TYPE_UNDEFINED`. -- `RDC_FI_ECC_CORRECT_TOTAL`, `RDC_FI_ECC_UNCORRECT_TOTAL`, and `RDC_FI_ECC_DEFERRED_TOTAL` no longer hang when AMD SMI defines GPU blocks above bit 31. The block iteration used a 32-bit counter that wrapped to zero instead of terminating. - - Opening or stopping an RDC handle no longer crashes the process when another handle is in use, and concurrent `rdc_connect()` calls no longer fail at random. - Certificates signed with the RDC certificate authority setup in `authentication/openssl.cnf` can no longer sign other certificates. A certificate request could ask to become a signing certificate, and its holder could then create client and server certificates that `rdcd` and `rdci` trust. @@ -48,6 +42,16 @@ Full documentation for RDC is available at [ROCm DataCenter Tool User Guide](htt - Stopping `rdcd`, or embedded RDC with `rdc_stop_embedded()`, no longer hangs. Rarely, the stop never returned and the process had to be killed. +## RDC 1.3.1 for ROCm 10.1.0 + +### Removed + +- Removed the unused `GetBlockNameStr()` test helper and its GPU block name map. Nothing called it, and its `static_assert` on `AMDSMI_GPU_BLOCK_LAST` broke the RDC build whenever AMD SMI added an IP block (#10949). + +### Resolved Issues + +- `RDC_FI_ECC_CORRECT_TOTAL`, `RDC_FI_ECC_UNCORRECT_TOTAL`, and `RDC_FI_ECC_DEFERRED_TOTAL` no longer hang when AMD SMI defines GPU blocks above bit 31. The block iteration used a 32-bit counter that wrapped to zero instead of terminating. + ## RDC 1.3.1 for ROCm 7.14.0 ### Added From f1f6dd17ed771c9fca847a89ee9fa52f36780b3a Mon Sep 17 00:00:00 2001 From: Adam360x Date: Thu, 22 Oct 2026 09:56:32 -0500 Subject: [PATCH 2/3] docs(rdc): reword the memory bandwidth entry as a 10.1.0 correction The ROCm 10.1.0 release notes already list the RDC_FI_GPU_MEMORY_CUR_BANDWIDTH fix, but ROCm 10.1.0 does not include it. Instead of repeating that entry in full, say that the fix is now included and that ROCm 10.1.0 does not include it. Signed-off-by: Adam360x --- projects/rdc/CHANGELOG.md | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/projects/rdc/CHANGELOG.md b/projects/rdc/CHANGELOG.md index 1bc4496abbf1..1cdba513c9ec 100644 --- a/projects/rdc/CHANGELOG.md +++ b/projects/rdc/CHANGELOG.md @@ -23,7 +23,8 @@ Full documentation for RDC is available at [ROCm DataCenter Tool User Guide](htt - RDC no longer reports another GPU's XGMI link states, or garbage values that could crash `rdcd`, for a GPU whose link status AMD SMI cannot read. That GPU now reports no links. - `rdci link` now prints an error and exits with a nonzero status when it cannot get the link status, instead of printing nothing and exiting with 0. -- `RDC_FI_GPU_MEMORY_CUR_BANDWIDTH` no longer reports zero under DMA or copy-only memory traffic on GPUs whose instantaneous UMC activity does not reflect DMA transfers. It now derives memory activity from the `mem_activity_acc` accumulator over firmware time, and falls back to the instantaneous `umc_activity` reading when the accumulator or firmware timestamp is unavailable. +- The `RDC_FI_GPU_MEMORY_CUR_BANDWIDTH` fix for zero readings under DMA or copy-only memory traffic is now included. + - The ROCm 10.1.0 release notes listed this fix, but ROCm 10.1.0 does not include it. - `rdci topo` and `rdc_device_topology_get()` now report the links of the requested GPU. They reported links between other GPU pairs, which showed the wrong link types on systems where GPU pairs are not all linked the same way, such as XGMI hives joined by PCIe. - `rdci topo` now exits with an error code when the GPU index is out of range or the topology query fails, instead of exiting 0, and `rdc_device_topology_get()` reports link types RDC does not define as `RDC_IOLINK_TYPE_UNDEFINED`. From 9bc5b923d617725eb8010f11e5529a051e01faa2 Mon Sep 17 00:00:00 2001 From: Adam360x Date: Thu, 22 Oct 2026 10:43:10 -0500 Subject: [PATCH 3/3] docs(rdc): drop the 10.1.0 note from the memory bandwidth entry Keep only the line saying that the RDC_FI_GPU_MEMORY_CUR_BANDWIDTH fix is now included, as the release notes review asked. Signed-off-by: Adam360x --- projects/rdc/CHANGELOG.md | 1 - 1 file changed, 1 deletion(-) diff --git a/projects/rdc/CHANGELOG.md b/projects/rdc/CHANGELOG.md index 1cdba513c9ec..4a7d96517ec2 100644 --- a/projects/rdc/CHANGELOG.md +++ b/projects/rdc/CHANGELOG.md @@ -24,7 +24,6 @@ Full documentation for RDC is available at [ROCm DataCenter Tool User Guide](htt - `rdci link` now prints an error and exits with a nonzero status when it cannot get the link status, instead of printing nothing and exiting with 0. - The `RDC_FI_GPU_MEMORY_CUR_BANDWIDTH` fix for zero readings under DMA or copy-only memory traffic is now included. - - The ROCm 10.1.0 release notes listed this fix, but ROCm 10.1.0 does not include it. - `rdci topo` and `rdc_device_topology_get()` now report the links of the requested GPU. They reported links between other GPU pairs, which showed the wrong link types on systems where GPU pairs are not all linked the same way, such as XGMI hives joined by PCIe. - `rdci topo` now exits with an error code when the GPU index is out of range or the topology query fails, instead of exiting 0, and `rdc_device_topology_get()` reports link types RDC does not define as `RDC_IOLINK_TYPE_UNDEFINED`.