Remove usages of Ubuntu 16.04 from pipelines

[am11]

Ubuntu Xenial 16.04 has reached the end of its five year LTS window.

Summary:

• Update to Ubuntu 18.04 VM or 22.04 host for alpine cross images. The main difference in cross images is the newer host compiler, otherwise rootfs has remained the same.
• Simplify pipeline scripts; RuntimeOS is inferred except for linux-bionic.
• Delete old jenkins scripts from src/tests/Common/scripts, which don't work. The blob URLs in those scripts return 404 and all deleted scripts have replacements.

[ghost]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @dotnet/runtime-infrastructure
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Ubuntu Xenial 16.04 has reached the end of its five year LTS window.

Summary:

• Update to Ubuntu 18.04 VM or 22.04 host for alpine cross images. The main difference in cross images is the newer host compiler, otherwise rootfs has remains the same.
• Simplify pipeline scripts; RuntimeOS is inferred except for linux-bionic.
• Delete old jenkins scripts from src/tests/Common/scripts, which don't work. The blob URLs in those scripts return 404 and all deleted scripts have replacements.

Author:	am11
Assignees:	-
Labels:	area-Infrastructure, community-contribution
Milestone:	-

[am11]

llvmaot failure is intermittent and unrelated (e.g. it failed on main branch in 2bc4f61 as well). I think the problem is it picks the incompatible helix machine for hwintrinsic tests. cc @vargaz

[am11]

Handled /p:CrossBuild=true this way to support multiple property cases like ./build.sh /p:'RuntimeFlavor=coreclr;CrossBuild=true;Configuration=release'. Ideally, all cross-building CI legs should explicitly pass -cross (which appends /p:CrossBuild=true), but an explicit /p:CrossBuild=true should have the same effect.

[jkotas]

LGTM. Thanks!

[am11]

The latest binutils 2.39 linker turned out to be stricter than its predecessors:

  Generating native code

  /usr/bin/aarch64-alpine-linux-musl-ld(0,0): error : (NETCORE_ENGINEERING_TELEMETRY=Build) 
  /__w/1/s/artifacts/bin/coreclr/Linux.arm64.Debug/crossgen2/linux-musl-arm64/native/crossgen2
  has a LOAD segment with RWX permissions

Refs:

• https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=ba951afb99912da01a6e8434126b8fac7aa75107
• https://www.redhat.com/en/blog/linkers-warnings-about-executable-stacks-and-segments

We have couple of options:

• try to figure out which segment it is complaining about and fix it
• add -Wl,--no-warn-rwx-segments suppression and filter the unrecognized warning on stderr for older linkers (Remove usages of Ubuntu 16.04 from pipelines #75513 (comment))
• change Add step for binutils build in cross-arm{64}-alpine dotnet-buildtools-prereqs-docker#695 to use 2.38 (instead of the latest 2.39)
• switch to llvm's ld again

@jkotas, which option do you suggest? (I think we must fix it or suppress it for consumers, regardless of the option selected for our CI leg, as consumer can have latest binutils in their PATH)

[jkotas]

It would be useful to figure out which segment it is complaining about. It should not be hard to fix.

[am11]

With retain,used lld is happy but ld is not. Since ld is default on linux distros, I have reverted it for now.

Ideally we should find a way to align them with retain,used, but I am not sure how that can be done with annotations; it might be that we need to resort to inline assembly for that.

[jkotas]

More details about this preference are in #52244.

We do the page protection flipping around GS cookie in nativeaot too, so it sounds like we will have a problem with Linux arm64 with 64kB pages. The page protection flipping around GS cookie seems very fragile. I am wondering whether we should stop doing that, in nativeaot at least.

[am11]

One problem is that we have hardcoded assumptions about page sizes:

runtime/src/coreclr/nativeaot/Runtime/CommonMacros.h

Lines 154 to 161 in cb6fb60

	#elif defined(HOST_ARM64)
	#define DATA_ALIGNMENT 8
	#ifndef OS_PAGE_SIZE
	#ifdef HOST_OSX
	#define OS_PAGE_SIZE 0x4000
	#else
	#define OS_PAGE_SIZE 0x1000

whereas we should instead be querying the page size from OS at run-time, as we do in GC (int getpagesize(void) from unistd.h). As it stands, 64K paged linux-arm64 kernels are currently not supported in NativeAOT regardless of the linker flavor.

System.Speech failures on Windows are unrelated (#75732).

[jkotas]

@dotnet/runtime-infrastructure Could you please review as well?

[am11]

CI failure is #75767. It's ready to merge.