| Message ID | 20200601171211.7272-1-simon@simonsouth.net |
|---|---|
| Headers | show |
On Mon, Jun 01, 2020 at 01:12:11PM -0400, Simon South wrote: > > Note I have found it unnecessary to use older versions of gcc and glibc when > building JamVM 1.5.1 on AArch64; the CPU's instruction cache is flushed > explicitly on this platform so it shouldn't be possible for the "invalid > instruction" situation mentioned in a comment to develop, and I haven't seen > it myself. > Can you test running bootstrap on jamvm-1-bootstrap (and jamvm@2) and see if that negates the need for an older version of glibc and gcc on other platforms? (basically just x86_64 for now, unless it now works on other architectures)
Efraim Flashner <efraim@flashner.co.il> writes: > Can you test running bootstrap on jamvm-1-bootstrap (and jamvm@2) and > see if that negates the need for an older version of glibc and gcc on > other platforms? I've tested this on x86_64 (building with a current gcc and glibc) and it doesn't work: I get the same "Illegal instruction" error during the configure phase of IcedTea. I'll do a bit of investigation and see if I can get it working normally on that platform, or come up with a different solution.
On Tue, Jun 02, 2020 at 12:04:15PM -0400, Simon South wrote: > Efraim Flashner <efraim@flashner.co.il> writes: > > Can you test running bootstrap on jamvm-1-bootstrap (and jamvm@2) and > > see if that negates the need for an older version of glibc and gcc on > > other platforms? > > I've tested this on x86_64 (building with a current gcc and glibc) and > it doesn't work: I get the same "Illegal instruction" error during the > configure phase of IcedTea. > > I'll do a bit of investigation and see if I can get it working normally > on that platform, or come up with a different solution. It can wait for whenever. I'm pretty sure autoconf and friends were already in the bootstrap chain before icedtea and it would be nice to get the older glibc and gcc out but I'd personally prioritize getting it working on more architectures first. Thanks for checking.
These patches add aarch64-linux support to the two versions of JamVM, a compact Java virtual machine, used during Guix's Java-bootstrap process (see the comment near the top of gnu/packages/java.scm). With them applied the process gets as far as starting to build the first IcedTea package (version 1.13.13), though this eventually fails with an error from gcc while compiling the Hotspot JVM (unrelated to my changes as far as I can tell). I'm still working on this issue. Note I have found it unnecessary to use older versions of gcc and glibc when building JamVM 1.5.1 on AArch64; the CPU's instruction cache is flushed explicitly on this platform so it shouldn't be possible for the "invalid instruction" situation mentioned in a comment to develop, and I haven't seen it myself. The "opcode guard" patch is necessary for floating-point operations to work correctly when JamVM is built with stack-caching enabled, as it is by default on AArch64 and elsewhere. Finally, I've renamed the existing "jamvm-arm.patch" file to "jamvm-1.5.1-arm.patch" for clarity as there are now separate patchsets for each JamVM release. -- Simon South simon@simonsouth.net Simon South (2): gnu: jamvm-1-bootstrap: Add aarch64-linux support. gnu: jamvm: Add aarch64-linux support. gnu/packages/java.scm | 41 +- .../patches/jamvm-1.5.1-aarch64-support.patch | 572 ++++++++++++++++ ...{jamvm-arm.patch => jamvm-1.5.1-arm.patch} | 0 .../patches/jamvm-2.0.0-aarch64-support.patch | 645 ++++++++++++++++++ .../patches/jamvm-2.0.0-opcode-guard.patch | 35 + 5 files changed, 1283 insertions(+), 10 deletions(-) create mode 100644 gnu/packages/patches/jamvm-1.5.1-aarch64-support.patch rename gnu/packages/patches/{jamvm-arm.patch => jamvm-1.5.1-arm.patch} (100%) create mode 100644 gnu/packages/patches/jamvm-2.0.0-aarch64-support.patch create mode 100644 gnu/packages/patches/jamvm-2.0.0-opcode-guard.patch