Arch Linux User Repository

It seems to not compile at all using the latest GCC (9.1.0, just got out of testing today). I get way too many instances of -Werror=address-of-packed-member.

It's a known bug and afaik has been fixed in xen's git repo for a while now. Might be wise to incorporate the patch into the PKGBUILD until it gets merged into a proper release?

EDIT: I should add that this happens even when adding "--disable-werror" to the config params and "!debug" to the options à la dhlk's commment.

EDIT2: Even the latest, bleeding edge staging versions of xen and ipxe do not compile under GCC 9. I'm sure it'll be fixed in a few days or so, but the official Xen 4.13 release probably won't be coming for a while. If it ends up getting fixed and anyone is interested, I can create a xen-git package.

the http://google.com

For anyone else (and my future self) who forgot they enabled debug in makepkg.conf, debug CFLAGS break xen's builds.

Error looks something like this:

<command-line>: error: "__OBJECT_FILE__" redefined [-Werror]
<command-line>: note: this is the location of the previous definition
<command-line>: error: "__OBJECT_LABEL__" redefined [-Werror]
<command-line>: note: this is the location of the previous definition
cc1: all warnings being treated as errors
Makefile:69: recipe for target 'compat/callback.i' failed

Fix requires removing debug in the PKGBUILD:

-options=(!buildflags !strip)
+options=(!buildflags !strip !debug)

Works like a charm, Thank you.

@ArthurBorsboom: thank you! i'll have an AMD system to test it out on soon.

instead of flooding these comments with discussions, let's keep it in recent forum topic here:

https://bbs.archlinux.org/viewtopic.php?pid=1814467

In short, there's another way that we can ask the author of the intel-ucode package to resolve (posted in the link above).

I have found a solution in the past (and am still using) another way of loading the (AMD) microcode in Xen.

In essence:

• Create a symlink from /lib/firmware/amd-ucode/microcode_amd_fam15h.bin -> /boot/microcode_amd_fam15h.bin
• Edit the /etc/xen/grub.conf and add " ucode=-1"

• Manually update the file /etc/grub.d/09_xen by adding the following two lines (note that these lines differ a little bit from the forum post

  echo 'Loading AMD Microcode' module2 ${REAL_DIR}/microcode_amd_fam15h.bin

• rebuild grub conf file: grub-mkconfig -o /boot/grub/grub.cfg

• reboot

I have described it more accurately in this forum post, where you have to replace the word module (outdated) with module2.

https://bbs.archlinux.org/viewtopic.php?id=191828

So, no kernel panic while loading multiple ramdisks.

Does this help you to find a persistent solution?

EDIT: Created a bbs topic for discussion: https://bbs.archlinux.org/viewtopic.php?pid=1814467

Intel (and perhaps AMD, can't test that yet) microcode patching.

Problem:

xen.cfg for Xen EFI cannot use the intel-ucode.img microcode binary produced by the intel-ucode system package. Attempting to specify multiple ramdisk lines panics the kernel (Xen EFI only reads the first line); and, specifying multiple imgs on the single ramdisk line is invalid by Xen EFI and just exits the boot.

Resolution:

Create a "patched" initramfs-linux.img file:

# cat /boot/intel-ucode.img /boot/initramfs-linux.img > /boot/intel-ucode_initramfs-linux.img

And then change xen.cfg to use this new patched ramdisk.

You can verify the fix with:

$ sudo xl dmesg | grep microcode
(XEN) microcode: CPU0 updated from revision 0x28 to 0x32, date = 2018-05-11
(XEN) microcode: CPU2 updated from revision 0x28 to 0x32, date = 2018-05-11

The bbs discussion above is to figure out the best place to capture this information.

I confirm the error message reported by sniper7kills: "free(): invalid pointer".

To generate the error I went to tools/firmware/hvmloader and ran make while using the latest version of acpica

make -C ../../libacpi  ACPI_BUILD_DIR=/storage/xen/src/xen-4.11.0/tools/firmware/hvmloader DSDT_FILES="dsdt_anycpu.c dsdt_15cpu.c dsdt_anycpu_qemu_xen.c"
make[1]: Entering directory '/storage/xen/src/xen-4.11.0/tools/libacpi'
gcc -Wall -Werror -Wstrict-prototypes -O2 -fomit-frame-pointer -fno-strict-aliasing -Wdeclaration-after-statement -DCONFIG_X86 -I/storage/xen/src/xen-4.11.0/tools/libacpi/../../tools/include -D__XEN_TOOLS__ -o /storage/xen/src/xen-4.11.0/tools/firmware/hvmloader/mk_dsdt mk_dsdt.c
# Remove last bracket
awk 'NR > 1 {print s} {s=$0}' dsdt.asl > /storage/xen/src/xen-4.11.0/tools/firmware/hvmloader/dsdt_anycpu.asl.tmp
cat dsdt_acpi_info.asl >> /storage/xen/src/xen-4.11.0/tools/firmware/hvmloader/dsdt_anycpu.asl.tmp
/storage/xen/src/xen-4.11.0/tools/firmware/hvmloader/mk_dsdt --debug=n --maxcpu any  >> /storage/xen/src/xen-4.11.0/tools/firmware/hvmloader/dsdt_anycpu.asl.tmp
mv -f /storage/xen/src/xen-4.11.0/tools/firmware/hvmloader/dsdt_anycpu.asl.tmp /storage/xen/src/xen-4.11.0/tools/firmware/hvmloader/dsdt_anycpu.asl
iasl -vs -p /storage/xen/src/xen-4.11.0/tools/firmware/hvmloader/dsdt_anycpu.tmp -tc /storage/xen/src/xen-4.11.0/tools/firmware/hvmloader/dsdt_anycpu.asl
free(): invalid pointer
make[1]: *** [Makefile:78: /storage/xen/src/xen-4.11.0/tools/firmware/hvmloader/dsdt_anycpu.c] Aborted (core dumped)
make[1]: Leaving directory '/storage/xen/src/xen-4.11.0/tools/libacpi'
make: *** [Makefile:72: acpi] Error 2

Unlikely that this is a xen issue; but building fails when using the latest acpica package.

It was something to do with the hvmloader and unable to call free() due to an invalid pointer.

Downgrading acpica to version 20180629 allows the package to compile.

Once I recompile xen to fix my libiscsi issue, I'll upgrade acpica again to try and provide more information.