In more detail:
...
(gdb) PASS: gdb.arch/amd64-init-x87-values.exp:
check_setting_mxcsr_before_enable: step forward one instruction for mxcsr test
p/x $mxcsr^M
$1 = 0x1f80^M
(gdb) FAIL: gdb.arch/amd64-init-x87-values.exp:
check_setting_mxcsr_before_enable: check new value of MXCSR is still in place
...

This is with kernel-obs-build-5.9.11-1.1.

So the scenario is as follows:
- we set mxcsr to a non-default value
- then we stepi past a nop
- we check that mxcsr still has the same value
- it turns out, it doesn't: the register is back to default value

It sounds alot like the fix for
https://bugzilla.kernel.org/show_bug.cgi?id=207979 is active:
...
commit 7ad816762f9bf89e940e618ea40c43138b479e10
Author: Petteri Aimonen <jpa@git.mail.kapsi.fi>
Date:   Tue Jun 16 11:12:57 2020 +0200

    x86/fpu: Reset MXCSR to default in kernel_fpu_begin()
...

However, the commit has been in place since v5.8, and I don't see this problem
on a laptop with tumbleweed running kernel v5.9.10.