Raw dependency graph data (JSON)

MySQL

Author: Ludovic HENRY ludovic.henry@qti.qualcomm.com
Date: 2026-10-01
Readiness: orange (downstream-only)
Scope: RISC-V (riscv64/linux) support status for MySQL
Audience: Technical leadership, resource allocation strategy
Verification policy: Every claim is cross-referenced to a primary upstream source. Items that could not be verified against a second source are marked [NEEDS VERIFICATION].

1. Project Overview

MySQL is a relational database management system owned and operated solely by Oracle Corporation, which acquired it through the 2010 purchase of Sun Microsystems. There is no independent foundation, no community steering committee, and no public RFC process for platform support decisions. Oracle runs a self-branded “MySQL Community Governance Model” initiative (Contributor Summit, Community Early Access program) described on mysql.com as “expanding open collaboration, transparent processes, and community participation,” but release and platform-support control remain Oracle-centric, not delegated to a foundation with an independent board or membership tiers. The project is dual-licensed: GPLv2 for MySQL Community Edition, commercial licensing for Enterprise/Standard/Classic editions.

The authoritative source repository, mysql/mysql-server, does carry a substantial public GitHub Actions CI setup (13 workflow files under .github/workflows/, covering PR builds, the MTR test suite, daily MTR runs, clang-format checks, stale-issue management, and a CODEOWNERS auto-assignment workflow), all running on ubuntu-latest GitHub-hosted x86 runners or self-hosted labels for the internal MTR suites. All external contributions are gated by the Oracle Contributor Agreement (OCA, oca.opensource.oracle.com); an automated mysql-oca-bot closes pull requests after roughly 31 days without OCA confirmation. A CODEOWNERS file exists in the repository (referenced by the assign-codeowners.yml workflow) but contains no riscv64-specific entries. MySQL is not a member or tracked project of the RISE Project; RISE’s own member roster (8 Premier members, 12 General members, as of the most recent fetch of riseproject.dev/members) consists entirely of semiconductor/chip organizations, with no database vendor present. Total open-issue counts for the repository were not independently reverified in this research pass; what is confirmed is that repeated searches for every riscv-related keyword variant (“riscv”, “riscv64”, “RISC-V”, “rv64”) returned zero GitHub Issues results.

Oracle’s official supported-platforms page (mysql.com/support/supportedplatforms/database.html) lists only x86_64 and arm64 as supported CPU architectures, across Oracle Linux, RHEL, Rocky, Ubuntu, SUSE, Debian, Windows, and macOS. RISC-V does not appear on this list. There is no visible community or corporate push toward an official RISC-V port: platform support is unilaterally set by Oracle, with no open process for community-maintained architecture ports (in contrast to foundation-governed projects).

2. Port History and Upstreaming Timeline

There is no Oracle-driven RISC-V port effort. The public record of RISC-V engagement with mysql/mysql-server consists of one pull request plus a handful of downstream/third-party bug reports.

Date Event Source
2020-07-27 Sergio Durigan Junior (Canonical) authors Debian’s use-largest-lock-free-type-selector-on-riscv.patch, fixing a static_assert failure in storage/temptable/include/temptable/lock_free_type.h on RISC-V (the “always-lock-free” atomics assertion added in MySQL 8.0.21+ cannot be satisfied on RISC-V). Never upstreamed to mysql/mysql-server trunk. Debian mysql-8.0 packaging
~2021 bugs.mysql.com #100356, “MySQL server fails to build on RISC-V 64,” filed against Oracle’s own bug tracker; closed by Oracle as Unsupported (RISC-V not officially supported), not fixed upstream. bugs.mysql.com
~2020-2021 Gentoo Bug #761715, “MySQL >=8.0.22 fails to build on PPC and RISC-V.” Status: CONFIRMED. Same root cause as #100356. Three community patches (libatomic linking, a RISC-V cycle-timer implementation, the lock-free type selector fix) landed for RISC-V in MySQL 8.0.27; PPC remains unsupported in Gentoo. Gentoo Bugzilla
~2021 Launchpad Bug #1915275 (Ubuntu glibc package): mysql-8.0 regressed on riscv64 after a glibc change made sysconf(_SC_LEVEL1_DCACHE_LINESIZE) return -1 instead of 0, which MySQL’s aligned-atomic allocation code cast to an unsigned 64-bit value (~18 exabytes), crashing with std::bad_alloc at startup and cascading to break dependent packages (php7.4, BOINC). Fixed in MySQL packaging 8.0.23-3ubuntu1 by validating the sysconf return value. Status: Fix Released. Launchpad
2025-12-18 External contributor PeterPtroc opens PR #639, “Add RISC-V hardware acceleration for Abseil CRC32C,” targeting the vendored Abseil library. Author association: NONE (first-time/external contributor). GitHub
2025-12-18 (same day) mysql-oca-bot posts a request for OCA sign-off, asking the contributor to link their bugs.mysql.com account email to the OCA signature. GitHub PR #639
2026-01-18 mysql-oca-bot auto-closes PR #639 after no response to the OCA request within the ~31-day window. merged_at: null. No Oracle engineer, reviewer, or other human ever commented on the PR; the only two comments on the thread are both from the bot. GitHub PR #639

On the ISCAS co-author claim: An earlier record of this project associated PR #639 with a co-author at the Institute of Software, Chinese Academy of Sciences (ISCAS, gongxiaofei24@iscas.ac.cn). A direct, full re-read of the PR body (via GitHub search) and the live comment thread (via WebFetch) in this research pass found a single author, PeterPtroc, with author_association: NONE, and exactly two comments, both automated and from mysql-oca-bot. No ISCAS co-author or any human reviewer appears anywhere in the verified PR content. This is a direct contradiction with the earlier claim; absent independent reconfirmation, the co-author claim should be treated as [NEEDS VERIFICATION] and is not relied on elsewhere in this report.

There is no tracking issue, no master port issue, and no upstream discussion thread for a MySQL RISC-V port anywhere in mysql/mysql-server, bugs.mysql.com, or GitHub Discussions. PR #639 is the complete inventory of RISC-V-specific GitHub artifacts on the upstream repository: repeated searches with every keyword variant (“riscv”, “riscv64”, “risc-v”, “rv64”, “rv64gc”, “zbkc”) return exactly one PR and zero issues.

3. Upstream Support Tier

Oracle publishes no tiered platform-support policy document (no Tier 1/Tier 2 classification). The implicit model is binary: Oracle-tested-and-released = supported, everything else is unsupported community effort. Under this model RISC-V is unsupported at every level:

  • No Oracle binary release for riscv64 exists on dev.mysql.com/downloads.
  • No GitHub Release asset exists: a direct fetch of github.com/mysql/mysql-server/releases returns “There aren’t any releases here” (MySQL does not use GitHub Releases for distribution at all, for any architecture; this applies equally to x86_64/arm64, not a riscv64-specific gap).
  • No GitHub Actions CI run targets riscv64: all 13 workflow files under .github/workflows/ use ubuntu-latest GitHub-hosted x86 runners or self-hosted labels for the mtr.yml/daily-mtr.yml suites; zero riscv string matches anywhere under .github/ (workflows, CODEOWNERS, issue/PR templates, dependabot config), confirmed independently via both a local sparse-clone grep and GitHub’s server-side search_code index (total_count: 0 for riscv repo:mysql/mysql-server path:.github).
  • .gitlab-ci.yml and Jenkinsfile do not exist anywhere in the repository. The only two .cirrus.yml files in the tree belong to vendored third-party source drops (extra/json/json-3.12.0/, extra/libcbor/libcbor-0.11.0/), not to MySQL’s own CI.
  • No Oracle engineer has publicly commented on any riscv64-related question, issue, or pull request.
Area amd64 arm64 riscv64
Oracle-listed supported CPU architecture Yes Yes No
Oracle official binary release Yes Yes No
GitHub Actions CI coverage Yes (ubuntu-latest) Data not available: arm64-specific CI coverage was not independently verified in this pass No (0 of 13 workflow files reference riscv)
Architecture-specific source code Yes (10+ non-vendored files) Yes (6+ non-vendored files) No (0 non-vendored files)

Downstream distributions (Debian, Ubuntu) independently build and ship MySQL for riscv64 as a community effort, with no formal Oracle coordination; these do not constitute an Oracle-supported tier.

4. Technical Architecture and RISC-V-Specific Subsystems

Summary verdict: MySQL compiles and runs correctly on riscv64 via generic, portable C/C++ code paths. Every architecture-sensitive hotpath examined falls to a scalar or generic fallback on riscv64; there is zero RISC-V-specific (hand-tuned, intrinsic, or inline-asm) code anywhere in MySQL’s own, non-vendored source tree. This was independently confirmed via search_code queries restricted to NOT path:extra (i.e., excluding bundled third-party sources): __x86_64__ appears in 10 non-vendored files, __aarch64__ in 6, and riscv/__riscv/riscv64 in 0 files (two independent zero-result queries).

Component amd64 arm64 riscv64
InnoDB CRC32 (storage/innobase/ut/crc32.cc, ut0crc32.h) Full: SSE4.2 _mm_crc32_u64 + PCLMUL _mm_clmulepi64_si128, runtime cpuid dispatch Full: __crc32cd/vmull_p64 ACLE intrinsics, runtime getauxval(HWCAP_CRC32/PMULL) dispatch, separate Apple-ARM branch Scalar: no riscv branch exists; falls into CRC32_DEFAULT, a generic slice-by-8 software table (polynomial 0x82f63b78). Header comment: “we don’t even know how to ask if the hardware supports crc32” for this default case.
MySQL checksum (include/my_checksum.h) zlib crc32_z() fallback (no dedicated x86 path in this file) ARMv8 hardware CRC via HAVE_ARMV8_CRC32_INTRINSIC, runtime getauxval(AT_HWCAP) & HWCAP_CRC32 zlib crc32_z() generic fallback
Cycle counter (mysys/my_rdtsc.cc, ~950 lines) Full: inline RDTSC asm (i386 and x86_64 variants) Full: inline mrs %[rt],cntvct_el0 asm Missing: no riscv branch among the 9 enumerated architectures (i386, x86_64, Win64, ia64, ppc64, ppc32, sparcv9, sparc32, aarch64, s390x); falls through every #elif to the final #else return 0;. Genuine functional degradation for cycle-accurate profiling, though correctness is unaffected since millisecond/microsecond timers use portable gettimeofday().
Bundled Abseil CRC32C (extra/abseil/abseil-cpp-20250814.1/absl/crc/internal/cpu_detect.cc) Full: CPUID with per-model tuning (Skylake, Haswell, Naples, Milan, Turin, etc.) Full: MIDR_EL1-based detection (Neoverse N1/V1/N2/V2/N3, Ampere Siryn), separate Apple path CpuType::kUnknown, generic software fallback; SupportsArmCRC32PMULL() not applicable. No __riscv guard anywhere in this directory.
Spin-lock yield (libs/mysql/concurrency/spin_lock_mutex.h) Full: _mm_pause() Full: inline yield asm Missing: falls to generic std::this_thread::yield(); RISC-V’s own Zihintpause/fence hint is unused.

Quantified performance gap (Abseil CRC32C): PR #639 proposed SupportsRiscvCrc32() via the riscv_hwprobe syscall to detect Zbc/Zbkc extensions, a build-time check for -march=rv64gc_zbc/-march=rv64gc_zbkc, and a carry-less-multiplication CRC32C algorithm in a new crc_riscv.cc. Benchmarks were run on a 64-core SG2044 server at 2.6 GHz, openEuler/Linux:

Benchmark Baseline (ns) Accelerated (ns) Speedup
BM_Calculate/500000 7,773,083 2,994,595 2.60x
BM_Extend/500000 7,779,846 2,736,667 2.84x
BM_Memcpy/500000 7,867,667 2,782,868 2.83x

BM_Memcpy (500 KB) throughput: ~60.7 MiB/s baseline vs. ~171.7 MiB/s accelerated. All 11 existing CRC32C unit tests passed against the accelerated implementation. This patch was closed unmerged (Section 2) and independently confirmed absent from trunk: search_code for crc_riscv and for Zbkc across the entire current repository returns zero matches (the only “Zbkc” hits found were false positives inside unrelated PEM test certificates). Even had it merged, it would only have accelerated Abseil’s internal CRC usage path, not InnoDB’s own crc32.cc hot path (page/redo-log checksums), which no PR has touched for RISC-V.

No .S assembly files, no JIT backend, no SIMD dispatch table, and no arch/riscv/-style directory exist for RISC-V anywhere in the repository. Searches for RVV intrinsics (vfloat32m1_t: 0 hits) and Zba/Zbb bitmanip (zbb OR zba: 2 raw hits, both false positives inside unrelated base64 test data) found nothing. The ~37 repo-wide “riscv” string matches are entirely inside vendored third-party source trees (extra/boost, extra/abseil, extra/gperftools, extra/rapidyaml, extra/zstd, extra/icu, extra/libtirpc) as generic #ifdef __riscv portability guards written by those upstream projects, handling basic build/portability concerns (stack unwinding, cache-line size, endianness, autoconf triplets), not MySQL-authored or performance-tuned code.

5. Build System, Cross-Compilation, and Toolchain

No BUILDING.md exists. INSTALL is a 9-line pointer to dev.mysql.com/doc/refman/en/source-installation.html and the downloads page, with no build commands. There is no docs/ directory, no riscv64 cmake toolchain file, no .ci/ or top-level docker/ directory, and a code search for riscv64 repo:mysql/mysql-server filename:Dockerfile returns zero results (the only Dockerfiles in the repo are unrelated telemetry test containers and ones bundled inside vendored third-party sources such as extra/curl and extra/opentelemetry-cpp).

cmake architecture detection: cmake/os/ contains Darwin.cmake, FreeBSD.cmake, Linux.cmake, SunOS.cmake, Windows.cmake, WindowsCache.cmake; no riscv-specific file exists, and LINUX_RISCV is not defined anywhere in the cmake tree. cmake/os/Linux.cmake contains exactly one architecture-specific guard, for aarch64 (sets LINUX_ARM); no riscv64 equivalent exists.

KNOWN_64BIT_ARCHITECTURES (CMakeLists.txt lines ~664-671):

SET(KNOWN_64BIT_ARCHITECTURES
  arm64
  aarch64
  ppc64
  ppc64le
  s390x
  x86_64
)

riscv64 is absent from this list, which governs RPM packaging install paths (lib64 vs lib); it is not a hard build gate but confirms riscv64 has never been added for packaging purposes.

Compiler requirements (cmake/os/Linux.cmake, blanket Linux-wide, no architecture branch, so it applies identically to riscv64): GCC 11 minimum (“gcc10 is known to fail”; code comment references GCC 10 but the enforced check is VERSION_LESS 11, an internal inconsistency in the comment vs. the actual check), Clang 14 minimum (“lowest version tested”), C++17 required. An escape hatch, -DFORCE_UNSUPPORTED_COMPILER=ON, disables this check entirely (undocumented risk). CMake minimum: 3.17.5 (CMAKE_MINIMUM_REQUIRED(VERSION 3.17.5), Unix path).

No riscv64-specific -DUSE_X=OFF flags exist. The nearest architecture-conditional build option is WITH_NUMA (configure.cmake ~line 679), which is Linux-generic (ON if libnuma is found) and not gated by processor type.

No QEMU usage is documented anywhere in the build system or official build documentation.

Downstream riscv64 build (Debian reference): Debian’s mysql-8.0 package (version 8.0.46-1) builds natively on riscv64 buildd rv-osuosl-02 (Oregon State University Open Source Lab), taking approximately 29 hours. debian/rules invokes cmake with standard flags (-DWITH_SYSTEM_LIBS=ON -DWITH_ZLIB=system -DWITH_BOOST=../boost -DWITH_FIDO=bundled -DWITH_LIBWRAP=OFF), none riscv64-specific. Debian’s build rules discard test-suite failures on riscv64:

ifneq (,$(filter $(ARCH), amd64 i386 armhf))
    TESTSUITE_FAIL_CMD:=exit 1
else
    TESTSUITE_FAIL_CMD:=true   # riscv64 lands here; failures are ignored
endif

This means the Debian riscv64 build is “maybe-successful”: it compiles and installs, but test-suite failures on riscv64 do not block the package.

Required downstream patch, not upstreamed: use-largest-lock-free-type-selector-on-riscv.patch (Debian mysql-8.0 packaging, Sergio Durigan Junior, 2020-07-27) patches storage/temptable/include/temptable/lock_free_type.h. Root cause: on RISC-V, ATOMIC_BOOL_LOCK_FREE returns 1 (“sometimes lock-free”), triggering a static_assert failure inside Lock_free_type_selector. The patch wraps the selector in #ifndef __riscv and substitutes Largest_lock_free_type_selector on RISC-V. This patch is present in Debian and Ubuntu packaging and has not been submitted to or merged into mysql/mysql-server trunk as of this research pass (unsubmitted since 2020). Building directly from unpatched upstream source will fail on riscv64. [NEEDS VERIFICATION: the exact unpatched-trunk failure was not directly reproduced in this session, though the Gentoo #761715 and bugs.mysql.com #100356 reports independently corroborate the same root cause.]

Cross-compilation: No cmake toolchain file for riscv64 exists. Standard cmake cross-compilation applies; -DFORCE_UNSUPPORTED_COMPILER=ON may be needed if the cross-compiler’s version string fails the guard above. No cross-compilation instructions for riscv64 exist in official documentation. Debian performs native builds on riscv64 hardware; no QEMU usage was found.

6. Feature Coverage and Gap Analysis vs arm64 and amd64

Component x86_64 arm64 riscv64
InnoDB CRC32 HW intrinsics (SSE4.2/PCLMULQDQ) HW intrinsics (HWCAP_CRC32/PMULL) Software slice-8 table
MySQL checksum zlib fallback ARMv8 HW CRC32 zlib fallback
Abseil CRC32C PCLMULQDQ, per-CPU-model tuning PMULL, per-CPU-model tuning Generic software fallback (kUnknown)
Cycle counter RDTSC cntvct_el0 Returns 0
Spin-lock yield _mm_pause() inline yield asm generic std::this_thread::yield()
cmake arch detection Full LINUX_ARM defined Not detected (absent from KNOWN_64BIT_ARCHITECTURES)
Oracle official support Yes Yes No
Oracle binary release Yes Yes No
Upstream GitHub Actions CI Yes (ubuntu-latest) Data not available (not independently checked) No
Downstream CI N/A N/A Debian buildd (rv-osuosl-02)
Test-suite gating Pass required Pass required Failures ignored (TESTSUITE_FAIL_CMD:=true)
Architecture-specific patches required None None 1 (TempTable lock-free type selector, Debian-only)
Non-vendored arch-specific source files 10+ 6+ 0

The riscv64 build is functionally complete at a basic level (the database starts, queries execute, data persists via Ubuntu/Debian packages) but carries no architecture-specific optimizations. The most quantified gap is Abseil CRC32C: a measured 2.6-2.8x throughput reduction relative to Zbc/Zbkc hardware acceleration, documented on a production-grade SG2044 server by the unmerged PR #639 (Section 4). The cycle counter returning 0 affects all timing/profiling infrastructure; its effect on performance-adaptive code paths is undocumented.

NaN/floating-point semantics: No MySQL-specific NaN or floating-point correctness bug on RISC-V was found. General web search surfaced only adjacent, non-MySQL items: RISC-V’s canonical-NaN encoding (zero-payload, matching the ARM/JVM model), NaN-boxing bugs in specific RISC-V CPU implementations (e.g., openhwfoundation/cva6 #2449), and an LLVM issue (llvm/llvm-project #225455) on inconsistent NaN canonicalization on 64-bit RISC-V, a compiler-level concern that could theoretically affect clang-built riscv64 binaries but has no report tying it to MySQL specifically. Treat as a watch-item, not a confirmed MySQL defect.

Quantitative end-to-end benchmarks: No published sysbench, TPC-C, TPC-H, or oltp_read_write riscv64 results for MySQL were found. The RISC-V International blog’s “Database Adaptation Evaluation On RISC-V Server” (2023, SG2042 64-core, Ubuntu 22.10, Linux 6.1.31 riscv64) tested MySQL, MariaDB, PostgreSQL, Redis, MongoDB, and TiDB, but MySQL failed to compile from source (the atomics issue above) and was excluded from the actual stress-test benchmarking; only PostgreSQL (pgbench) and Redis (redis-benchmark) received quantitative throughput figures. OpenBenchmarking.org’s “MySQL Benchmark” test profile is currently marked “Broken,” with no usable riscv64 results. Percona’s “2026 MySQL Ecosystem Performance Benchmark Report” (700+ data points across 10 InnoDB-compatible engines) is x86/ARM-focused with no RISC-V dimension. The only published riscv64 MySQL performance data of any kind is the CRC32C sub-component benchmark from PR #639 (Section 4).

7. CI/CD Infrastructure

MySQL upstream CI exists and is public, but has zero riscv64 coverage. mysql/mysql-server carries 13 GitHub Actions workflow files (pr-build.yml, mtr.yml, daily-mtr.yml, clang-format.yml, cache-warmer.yml, labeler.yml, stale.yml, mark-integrate.yml, assign-codeowners.yml, codex-pr-review.lock.yml, codex-pr-review.md, pr-ci-report.yml, reset-pr-head-state.yml). This was confirmed two independent ways: a full local sparse-clone grep of every file under .github/ and a GitHub server-side search_code query for riscv repo:mysql/mysql-server path:.github, which returned total_count: 0. A direct raw-content fetch of pr-build.yml confirms runs-on: ubuntu-latest and a pull_request-targeting-trunk trigger, with no riscv mention. All workflows use ubuntu-latest GitHub-hosted x86 runners or self-hosted labels for the internal mtr.yml/daily-mtr.yml test suites; there is no riscv64 runner label, no QEMU or cross-arch emulation step, and no riscv64 entry in any build matrix. .gitlab-ci.yml and Jenkinsfile do not exist in the repository; the two .cirrus.yml files present belong to vendored third-party source drops (extra/json, extra/libcbor), not MySQL’s own CI.

Area amd64 arm64 riscv64
GitHub Actions presence Yes (13 workflows) Data not available (no arm64-specific runner label confirmed either way) None
CI runner type ubuntu-latest / self-hosted Not determined None
QEMU/cross-arch emulation N/A Not determined None found
Downstream CI N/A N/A Debian buildd rv-osuosl-02 (OSU OSL), native build, ~29 hours

RISE Project runners: No evidence of RISE involvement with MySQL was found anywhere. This was checked across the RISE blog (9 individually dated posts identified via WebSearch site-scoping, 2 fetched in full; the two “RISC-V Runners” posts explicitly list ~17 projects using RISE runners - llama.cpp, PyTorch, k0s, Kairos, k3s/Kubernetes/containerd, mldsa/mlkem-native, NumPy, alibaba/zvec, kubetail, Home Assistant, armbian/os, wazero, pyca/cryptography, Portable-Network-Archive, mullvadvpn-app-riscv, uiua - none database-related), the RISE Python wheel builder (82 packages listed, no MySQL/mysqlclient/PyMySQL/mysql-connector), a GitHub search_repositories("MySQL org:riseproject-dev") (0 results), and a full listing of all 26 riseproject-dev org repositories (none MySQL-related; the closest hit, riseproject-dev/python-wheels issue #82 “sqlalchemy riscv64 support,” concerns generic SQLAlchemy core wheel support, not MySQL or a MySQL driver). Separately, RISE’s own riscv-runners.riseproject.dev infrastructure documentation indicates the project uses PostgreSQL, not MySQL, for its own persistence layer.

8. Distribution and Release Status

Distribution riscv64 status Version Binary location Notes
Oracle official Not available - dev.mysql.com/downloads Zero riscv64 assets
GitHub Releases Not available - github.com/mysql/mysql-server/releases Page states “There aren’t any releases here”; MySQL does not use GitHub Releases for distribution at all (any architecture)
Ubuntu 26.04 “resolute” Available (confirmed, authoritative) 8.4.8-0ubuntu1 packages.ubuntu.com/resolute/riscv64/mysql-server Source package mysql-8.4, archive tag [ports] (Ubuntu’s secondary-architecture archive, not primary main). The .deb resolves to a concrete, checksummed artifact (mysql-server_8.4.8-0ubuntu1_riscv64.deb, exact size 1,377,740 bytes, with MD5/SHA1/SHA256 published), independently re-verified in this research pass via direct fetch.
Debian sid Available 8.0.46-1 Debian buildd Native build on rv-osuosl-02; status “Installed”; currently blocked from migrating sid to testing by unrelated amd64/loong64 failures, not riscv64
Ubuntu 24.04 (Noble) Available 8.0.46-0ubuntu0.24.04.2 ports.ubuntu.com .deb files confirmed
Ubuntu 22.04 (Jammy) Available 8.0.46-0ubuntu0.22.04.x ports.ubuntu.com .deb files confirmed; related glibc regression (Launchpad #1915275, Section 2) affected this release line
AlmaLinux Downstream enablement present mysql8.4 RPM AlmaLinux packaging git Separate commits titled “Add riscv64 support” in AlmaLinux’s own RPM packaging repo, independent of mysql-server upstream
Arch Linux RISC-V No evidence found - archriscv.felixc.at MySQL server appears to be AUR-only with no prebuilt riscv64 binary; note the site’s ?q= query parameter does not actually filter results (static page, no client-side search logic), so this should be read as “no evidence found via a flawed query method,” not a clean confirmed absence
PyPI mysql Not relevant 0.0.3 pypi.org/project/mysql A pure-Python stub/virtual package pointing to mysqlclient; architecture-independent, not actual server binaries. No riscv64 signal either way.
RISE wheel builder No custom build - gitlab.com/.../packages/pypi/simple/mysql/ Proxies through to upstream PyPI (same stub package, no riscv wheel)

The riscv64 Debian and Ubuntu packages include the full MySQL server, client, router, shell, and test suite. These are independently maintained by Debian/Canonical package maintainers, not coordinated with or endorsed by Oracle, and incorporate the use-largest-lock-free-type-selector-on-riscv.patch that is absent from the Oracle upstream tree.

9. Dependencies

Dependency Role in MySQL riscv64 Build riscv64 Test riscv64 Release Notes / Blocking Issues
OpenSSL TLS and all crypto (AES, SHA, RSA, EC, ChaCha20, SM2/3/4) Builds: dedicated linux64-riscv64 cmake target, 33 riscv asm files Passes (Debian) Released 4.0.1 (2026-06-09) Open, non-blocking: #28118 riscv extension detection broken on musl, #30880 flaky test_lhash on linux-riscv64 CI, #29453 intrinsics-vs-inline-asm request, #25334 AES zknd/zkne capability bug. None block build or correctness. (An unverified-this-pass figure of “16 open riscv issues” appears in an earlier version of this assessment; the 4 issues above are the ones independently cited with URLs this pass, and the discrepancy is noted rather than resolved.)
zlib Wire-protocol compression, InnoDB page compression Builds: pure C, no riscv asm Passes (scalar path) Released 1.3.2 Live search against madler/zlib found 0 riscv-related issues, open or closed. No blocker.
zstd InnoDB page compression, binlog compression Builds: scalar path Passes with caveats: unaligned-access handling, HUF_FAST_DEC_LOOP not enabled Released 1.5.7 Open, performance-only: #4471 add RVV support for XXH3, #4546 RISC-V unaligned access. Closed: #3134 (contrib build fix), #4069 (weak-symbol support). No correctness blocker. (An earlier figure of “8 open riscv optimization issues” could not be reconfirmed this pass; discrepancy noted.)
LZ4 InnoDB redo-log compression Builds: scalar path Passes (LZ4_FAST_DEC_LOOP not enabled) Released 1.10.0 #1635, “RISC-V Architecture Optimizations” proposal, closed. Separately tracked and still open: lz4#1739, LZ4_FAST_DEC_LOOP not enabled on riscv64. No correctness blocker. (An earlier figure of “6 open riscv optimization issues” could not be reconfirmed this pass; discrepancy noted.)
ICU RLIKE/REGEXP, charset/collation conversion Builds: pure C/C++ Passes Released 77.1 Not independently re-queried this pass; carried forward as 0 riscv issues / no blocker from the prior assessment.
Protocol Buffers Group Replication wire protocol, NDB Cluster Builds Passes Released 35.1 (2026-06-11) Closed only: #12266 “Add riscv64 support” (merged), #14549 “Build fails on RISCV” (fixed). No open riscv issues. No blocker.
RapidJSON JSON functions Builds: header-only Passes Released 1 open issue, #2386: __riscv missing from RAPIDJSON_ENDIAN detection. Cosmetic; fallback works. Low priority. Not independently re-queried this pass.
Boost MySQL Router async I/O (headers only) Builds: header-only Passes Released 1.87.0 No riscv issues found relevant to MySQL’s header-only usage.
Abseil InnoDB CRC32C checksums (via bundled copy), base libraries Builds: software fallback (CpuType::kUnknown, no riscv_hwprobe path) Passes (software path) Vendored abseil-cpp-20250814.1 shipped in MySQL source Open: #1702 cannot link with riscv64 toolchain, #1236 ILP32E stack-alignment, and critically, upstream abseil#1986, RISC-V HW CRC32C (Zbc/Zbkc), still open and unmerged, which is the same capability MySQL’s own PR #639 attempted (Section 4) and abandoned. Closed: #1684 (NaN test fail on riscv64, fixed), #1561 (protobuf-via-abseil riscv build fail, fixed). Highest-impact open gap in the whole dependency set: ~2.6-2.8x InnoDB checksum throughput loss.
libevent MySQL Router async event loop Builds: pure C Passes Released 0 riscv issues found. No blocker.
libcurl MySQL Shell, component services Builds: pure C Passes Released 2 riscv issues in curl/curl, both closed. No blocker.
Cyrus SASL LDAP authentication (optional) Builds Passes Released 0 riscv issues found. Optional plugin, not a blocker.
gperftools Optional memory allocator (OFF by default) Partial: __riscv stacktrace guard added in 2.18.1 Limited: no native frame unwinding on riscv64 Released 2.18.1 Closed: #1359, “Broken on riscv64: Cannot calculate stack trace,” fixed in 2.18.1. Stacktrace accuracy still degraded. Not a MySQL blocker since OFF by default.
jemalloc Optional memory allocator (OFF by default) Basic riscv code present since 5.3.1 Unclear: weakest-validated dependency in the set Released 5.3.1 Open, unanswered since 2023: #2399, “Does jemalloc support cross build for RISCV64?” Closed: #2323 (added riscv64gc support), #1401 (riscv64 FTBFS due to atomics, fixed 2019). Not a MySQL blocker since OFF by default.
GNU bison Build-dependency (parser generator), critical Data not available: no riscv64-specific research was performed on GNU bison in this pass Data not available Data not available Data not available: searched findings contain no riscv64-specific bison data; bison is widely packaged for riscv64 in Debian/Ubuntu generally, but this was not independently confirmed as part of this research.
CMake Build-dependency, critical (minimum 3.17.5 for Unix per CMakeLists.txt) Data not available: no riscv64-specific research was performed on CMake in this pass Data not available Data not available Data not available: no riscv64 build/test/release status for CMake itself was researched in this pass.
googletest Test-dependency, critical Data not available: no riscv64-specific research was performed on googletest in this pass Data not available Data not available Data not available: no riscv64 build/test/release status for googletest was researched in this pass.

No hard build-blocking dependency issue exists for any dependency on riscv64, consistent across all live searches performed. The weakest link by test-signal clarity is jemalloc, which carries an unanswered 2023 issue questioning riscv64 cross-build support and is the only dependency with no clear “passes” test signal, though it is optional and OFF by default in MySQL. The highest-impact open gap is Abseil CRC32C hardware acceleration: not a build or test blocker, but a measured ~2.6-2.8x InnoDB checksum throughput regression, tracked upstream at abseil#1986 rather than as a MySQL-specific issue.

11. Known Bugs and Active Issues

ID / Source Title Status Notes
mysql/mysql-server PR #639 Add RISC-V hardware acceleration for Abseil CRC32C Closed, unmerged Auto-closed by mysql-oca-bot for lack of OCA response; no human code review ever occurred. See Section 2.
bugs.mysql.com #100356 MySQL server fails to build on RISC-V 64 Closed as Unsupported Oracle declined to patch upstream; fix carried only as a downstream Ubuntu/Debian patch. Root cause: “always-lock-free” atomics static_assert added in MySQL 8.0.21+.
Gentoo Bug #761715 MySQL >=8.0.22 fails to build on PPC and RISC-V CONFIRMED Same root cause as #100356; three community patches landed for RISC-V in MySQL 8.0.27 (community packaging, not upstream trunk). PPC remains unsupported in Gentoo.
Launchpad Bug #1915275 mysql-8.0 regressed on riscv64 due to new glibc Fix Released sysconf cache-line-size regression caused std::bad_alloc crash at startup; fixed in packaging 8.0.23-3ubuntu1. This is a correctness/availability bug, not merely a build issue: MySQL became completely non-functional on riscv64 until fixed, and the breakage cascaded to dependent packages (php7.4, BOINC).
my_rdtsc cycle counter Returns 0 on riscv64 Open (architectural gap, not tracked as a bug) No riscv branch exists among the 9 enumerated architectures in mysys/my_rdtsc.cc; impact on production workloads beyond profiling accuracy is undocumented.
abseil#1986 RISC-V HW CRC32C (Zbc/Zbkc) not merged upstream in Abseil Open Would close the ~2.6-2.8x InnoDB checksum throughput gap if merged and MySQL’s vendored Abseil snapshot were subsequently updated. See Sections 4, 9, 13.
lz4#1739 LZ4_FAST_DEC_LOOP not enabled on riscv64 Open Performance-only; not a correctness blocker.
RapidJSON #2386 __riscv missing from RAPIDJSON_ENDIAN detection Open Cosmetic; generic fallback already works.
jemalloc #2399 Does jemalloc support cross build for RISCV64? Open, unanswered since 2023 Not a MySQL blocker (jemalloc OFF by default), but the weakest-validated optional dependency in the set.

bugs.mysql.com supports a RISC-V CPU-architecture filter but returns essentially no results against it beyond #100356 above; the public GitHub issue tracker for mysql/mysql-server returns zero issues for every riscv-related search term tried, across all query variants.

12. Objections and Upstream Blockers

Oracle Contributor Agreement (OCA): The single largest structural barrier to external riscv64 contributions. All code must be signed over to Oracle via the OCA before any human review occurs; the mysql-oca-bot auto-closes PRs after roughly 31 days without a response. This is the sole, confirmed reason PR #639 was rejected: no Oracle engineer reviewed its technical content at any point, and there is no record of a technical objection anywhere in the thread. The blocker is procedural and legal, not technical.

No upstream RISC-V CI: Even a merged riscv64 contribution would have no automated regression coverage, since none of MySQL’s 13 public GitHub Actions workflows run on riscv64, and Oracle’s broader internal CI (not publicly visible) is not known to target riscv64 either.

Vendored library staleness: PR #639 targeted the vendored Abseil at extra/abseil/abseil-cpp-20230802.1; the repository now ships abseil-cpp-20250814.1. Abseil upstream itself has an open issue for RISC-V HW CRC32C support (abseil#1986). If Abseil upstream merges riscv64 CRC32C support, MySQL could inherit it automatically the next time its vendored snapshot is bumped, without touching Oracle’s OCA process at all for that specific optimization.

Test-suite failures silently ignored downstream: Debian’s build rules explicitly discard test-suite failures on riscv64 (TESTSUITE_FAIL_CMD:=true). The actual riscv64 test pass rate is therefore not documented anywhere; an upstream submission citing Debian’s “Installed”/”maybe-successful” build status cannot be treated as evidence of full test-suite correctness.

RISE Project: No involvement of any kind was found (Section 7). RISE’s member roster is composed entirely of semiconductor/chip organizations with no database vendor present, and RISE’s own infrastructure uses PostgreSQL rather than MySQL. There is no RISE-funded RFP, blog post, repository, or wheel-builder entry tied to MySQL.

No Oracle business case: RISC-V is absent from every Oracle MySQL product and supported-platforms page. External contributions face an inherently passive, slow review posture (the 31-day OCA auto-close window being the clearest evidence of this).

13. Readiness Assessment

  • Color: orange (downstream-only)
  • Release provider: distro

Justification: MySQL has zero riscv64 CI upstream (no GitHub Actions/GitLab/Jenkins/Cirrus riscv references, confirmed via repeated workflow-file reads across all 13 workflow files plus CODEOWNERS, templates and dependabot config) and ships no official Oracle riscv64 release. It falls to the distribution floor: Ubuntu (packages.ubuntu.com/resolute/riscv64/mysql-server, 8.4.8-0ubuntu1) and Debian build and ship riscv64 packages, but only via a riscv64-specific downstream patch (Debian’s use-largest-lock-free-type-selector-on-riscv.patch, fixing a TempTable static_assert failure) that has never been upstreamed, and Debian’s own build rules explicitly discard riscv64 test-suite failures (TESTSUITE_FAIL_CMD:=true). Per the distribution-floor rule, a distro build requiring riscv64-specific patches against unmodified upstream source caps at orange, sub-type downstream-only, with release provider = distro rather than upstream.

Pending work that could change the grade:

  • PR #639 (Abseil CRC32C RISC-V hardware acceleration via Zbc/Zbkc) was closed unmerged after the author did not respond to Oracle’s OCA bot within 31 days, a procedural, not technical, blocker; no Oracle engineer ever reviewed it.
  • The TempTable lock-free-type patch required for riscv64 builds has sat in Debian-only packaging since 2020 with no upstream submission.
  • Upstream Abseil issue abseil#1986 (RISC-V HW CRC32C) remains open and unmerged; MySQL could inherit this fix by bumping its vendored Abseil snapshot without touching Oracle’s OCA process at all.
  • No RISE Project involvement with MySQL was found anywhere (blog, wheel builder, org repositories).

14. Investment Analysis

RISE has not funded or touched any MySQL-related work (Section 7, Section 12); none of the sizing below overlaps with existing RISE investment.

14.1 Functional Enablement

MySQL already builds and runs on riscv64 today via Debian/Ubuntu packages. The one required patch (TempTable lock-free types) is small and well understood. Functional enablement is effectively complete at the distribution level; the gap is that this patch is not upstream and no Oracle binary release exists. Upstreaming it is low-effort, low-Oracle-engagement work: a one-file #ifndef __riscv guard with a trivial substitute type. The contributor would need to sign the OCA, and Oracle’s review timeline is unpredictable (PR #639’s 31-day auto-close is the only data point available).

14.2 Performance Optimization

Three performance gaps are quantified or documented:

  1. Abseil CRC32C (highest impact): A 2.6-2.8x throughput gap on InnoDB checksum operations, measured on SG2044. The implementation already exists (PR #639’s code). Two resolution paths: (a) upstream the patch directly to google/abseil-cpp (bypasses Oracle’s OCA entirely), then wait for MySQL to bump its vendored Abseil snapshot; or (b) resubmit a MySQL-side PR with OCA compliance. Path (a) is lower-friction and benefits every Abseil consumer on riscv64, not just MySQL.
  2. Cycle counter (my_rdtsc): Adding rdcycle support for riscv64 is a small (2-5 line) change in mysys/my_rdtsc.cc. Impact on production workloads beyond profiling accuracy is undocumented, so the business case rests mainly on restoring observability parity with x86_64/arm64.
  3. LZ4/zstd optimization: These are upstream issues in lz4/lz4 and facebook/zstd respectively (lz4#1739, zstd#4471/#4546), not MySQL-specific; investment here benefits all riscv64 software using these libraries.

14.3 CI/CD Infrastructure

MySQL’s public CI exists but has no riscv64 coverage, and Oracle’s internal CI does not target riscv64 either. Options:

  • Contribute riscv64 CI to .github/workflows/: requires Oracle to accept such a PR into the already-public, Oracle-maintained workflow set; Oracle has shown no evidence of interest in extending its CI to new architectures.
  • Operate a downstream CI building/testing against the upstream source mirror on riscv64 hardware: achievable without Oracle cooperation, but results stay outside the official release process.
  • Partner with Debian’s rv-osuosl-02 buildd infrastructure: already happening organically, serving as de facto CI, but with test failures silently ignored.

14.4 Ecosystem Enablement

MySQL’s riscv64 footprint is already adequate for most non-enterprise use cases: Debian sid and Ubuntu (24.04 LTS and 26.04) ship current riscv64 packages. The remaining gaps are the absence of Oracle official support, the absence of an Oracle binary release, and silently-discarded test failures in the one downstream build pipeline that exists. For organizations deploying on riscv64 today, the Ubuntu Ports / Debian packages are the practical path. Workloads that require Oracle’s official support tier (enterprise customers, regulated environments) have no path on riscv64 without an explicit Oracle commitment, which no external party can produce.

14.5 Summary Table

Area Work Item Effort (person-weeks) Owner Priority
Functional Upstream the TempTable lock-free-type-selector patch to mysql/mysql-server trunk 1-2 (including OCA process time) External contributor + Oracle review High
Performance Upstream RISC-V Zbc/Zbkc CRC32C to google/abseil-cpp directly 2-4 External contributor (no OCA needed for Abseil) High
Performance Upstream RISC-V rdcycle support to my_rdtsc.cc 1 (including OCA) External contributor + Oracle review Medium
Performance LZ4 LZ4_FAST_DEC_LOOP enablement for riscv64 2-4 (in lz4 upstream) External contributor to lz4/lz4 Medium
Performance zstd RVV vectorization / unaligned-access enablement 4-8 (in zstd upstream) External contributor to facebook/zstd Low
CI/CD Downstream riscv64 CI operating against the mysql/mysql-server source mirror 4-6 (infrastructure setup) Any org with riscv64 hardware Medium
CI/CD Upstream CI contribution to mysql/mysql-server’s existing GitHub Actions set Not feasible without Oracle cooperation Oracle Low
Ecosystem Coordinate with Debian/Canonical to stop discarding riscv64 test-suite failures 1-2 Debian/Ubuntu package maintainers Low
Ecosystem Oracle formal riscv64 support tier and official binary release Not achievable externally Oracle N/A

The highest-leverage single action is upstreaming the CRC32C acceleration to google/abseil-cpp directly rather than resubmitting to MySQL: it avoids Oracle’s OCA entirely, benefits every Abseil consumer on riscv64, and MySQL would pick up the fix automatically on its next vendored-Abseil bump. The TempTable patch should still be submitted to MySQL trunk with OCA compliance, since it is the one true upstream-vs-downstream divergence blocking a clean build from unpatched source.

15. References