update qemu version to v8.0.4

[crazy-max]

changelog: https://wiki.qemu.org/ChangeLog/8.0

notable changes:

• QEMU 8.0 now supports Xen guests under KVM when running on the Linux 5.12 kernel or newer.
• System emulation on 32-bit x86 hosts has been deprecated. The 32-bit x86 host support for system emulation is deemed no longer effective.
• Due to Intel HAXM being retired, the "-accel hax" option has been deprecated.
• New ARM emulated CPU types of Cortex-A55 and Cortex-R52.
• QEMU 8.0 on RISC-V supports a number of new extensions like smstateen, Zawrs, the T-Head vendor extensions, Svadu, Zicond, and Zicbiom, ZCicboz, and Zicbop. QEMU 8.0 is also now capable of supporting ACPI on RISC-V.
• QEMU 8.0 adds a new CPU model for Intel Sapphire Rapids.
• A new "ide-cf" option for emulating a CompactFlash card block device.
• Support for the F13 and F14 keys have been added for input handling.
• Experimental VFIO migration support for the v2 VFIO migration protocol while removing v1 protocol support.

---

changelog: https://wiki.qemu.org/ChangeLog/7.2

notable changes:

• QEMU for x86/x86_64 now supports the CORE_THREAD_COUNT MSR.
• Support for passing a random seed to the Linux kernel when booted using the "-kernel" argument.
• The Tiny Code Generator (TCG) has performance improvements in full system emulation mode, fixes for its SSE implementation, and TCG now supports AVX / AVX2 / F16C / FMA3 / VAES instructions.
• Support for saving/restoring the XSAVE state in signal frames for the user-mode emulation code on x86_64.
• QEMU with 9pfs has a "massive general performance improvement" as much as 6~12x better performance. This massive speed-up in the QEMU 9pfs code comes from making use of a GHashTable for the FID table. With the change by Linus Heckemann, "This change uses a hashtable for this instead, significantly improving the performance of the 9p filesystem. The runtime of NixOS's simple installer test, which copies ~122k files totalling ~1.8GiB from 9p, decreased by a factor of about 10."
• QEMU's RISC-V code now supports Zihintpause, disas support for vector instructions, promoting zmmul out of being experimental, improvements to the debugger spec, and other RISC-V improvements.
• QEMU for s390x now supports SHA-512 via KIMD/KLMD instructions, fixed emulation of LZRF and VISTR instructions, and implementing other features.
• Stability improvements to the QEMU OpenRISC code as well as performance improvements.
• QEMU on Arm now supports emulating Enhanced Translation Synchronization, PMU Extensions v3.5, guest translation granule size, HAFDBS, and E0PD. The Arm Cortex-A35 CPU type can also now be emulated.
• LoongArch ISA support improvements as well as a number of machine additions.
• Big endian 32-bit MIPS hosts have been deprecated.

---

• cpu now sets to max by default on x86: qemu/qemu@d135f78
• linux-user: handle /proc/self/exe with execve() syscall: qemu/qemu@f07eb1c
• disable-blobs opts removed: qemu/qemu@10229ec
• linux-user/syscall: Extract do_execve() from do_syscall1(): qemu/qemu@156e1f6
• linux-user/syscall: Implement execveat(): qemu/qemu@55bbe4d

[crazy-max]

@tonistiigi About this change: https://github.com/tonistiigi/binfmt/actions/runs/5305194882/jobs/9602044053#step:4:385

#43 15.22 not ok 11 shebang-direct
#43 15.22 # (from function `assert_output' in file bats-assert/src/assert.bash, line 239,
#43 15.22 #  in test file test.bats, line 86)
#43 15.23 #   `assert_output "./printargs ./shebang.sh foo bar1"' failed
#43 15.23 #
#43 15.23 # -- output differs --
#43 15.23 # expected : ./printargs ./shebang.sh foo bar1
#43 15.23 # actual   : ./printargs /work/shebang.sh foo bar1
#43 15.23 # --
#43 15.23 #

linux-user now resolves the full path name of the started process at startup. See qemu/qemu@258bec3

[tonistiigi]

Isn't this a breaking change (at least for the direct mode without -P)? What if the program reads its args[0], it wouldn't expect that the value gets modified.

If this is how it behaves with QEMU_PRESERVE_ARGV0 in kernel then it just looks plain wrong. If it is just a problem with our direct exec patches then the patches may need updates.

[tonistiigi]

@hdeller are we missing some context in here?

[hdeller]

My change just makes the emulated binary work the same way as a native program would. This means /proc/self/exe will return the currently running executable. I don't think my patch modified args[] values.
I think if you would run your testsuite in the ./work directory you wouldn't see that changed.

[sando38]

Hello there,
Qemu is now available as version 8.1, which includes a bug fix, which solves an issue with our build pipelines.
Do you plan to release a new version of your Docker image soon, including version 8.1?
Thanks in advance!

[tonistiigi]

My suggestion is to go ahead with this and follow up with 8.1 BUT not update the latest tag before we have the args CI issue figured out. @crazy-max Do we need release pipeline changes for this?

[crazy-max]

Do we need release pipeline changes for this?

We can skip latest:

binfmt/.github/workflows/release.yml

Lines 19 to 22 in 82a943b

	latest:
	description: 'Create latest tag'
	required: true
	default: 'false'

[crazy-max]

Can be tested with tonistiigi/binfmt:qemu-v8.0.4