https://bugzilla.suse.com/show_bug.cgi?id=1202727 https://bugzilla.suse.com/show_bug.cgi?id=1202727#c1 --- Comment #1 from Marcos de Souza --- I added the wrong error output. Now with the error: (none):/home/kgrtst/mpdesouza/qa_test_klp # ./run.sh == Patch caller of graph traced callee == [14:28:06] Test Case 10: Patch caller of graph traced callee [14:28:06] *** Compiling kernel live patch [14:28:07] make: Entering directory '/home/kgrtst/mpdesouza/linux' [14:28:26] CC [M] /tmp/live-patch/tc_10/klp_tc_10_livepatch.o [14:28:37] MODPOST /tmp/live-patch/tc_10/Module.symvers [14:28:38] CC [M] /tmp/live-patch/tc_10/klp_tc_10_livepatch.mod.o [14:28:42] LD [M] /tmp/live-patch/tc_10/klp_tc_10_livepatch.ko [14:28:42] make: Leaving directory '/home/kgrtst/mpdesouza/linux' [14:28:42] *** Compile test support module [14:28:42] make: Entering directory '/home/kgrtst/mpdesouza/linux' [14:29:03] CC [M] /tmp/live-patch/tc_10/klp_test_support_mod.o [14:29:14] MODPOST /tmp/live-patch/tc_10/Module.symvers [14:29:14] CC [M] /tmp/live-patch/tc_10/klp_test_support_mod.mod.o [14:29:18] LD [M] /tmp/live-patch/tc_10/klp_test_support_mod.ko [14:29:19] make: Leaving directory '/home/kgrtst/mpdesouza/linux' [14:29:19] *** Load test support module [ 246.523106] klp_test_support_mod: loading out-of-tree module taints kernel. [14:29:19] *** Enable graph tracing for orig_do_sleep [14:29:19] *** Starting uninterruptible sleeper [14:29:19] *** Inserting live patch [ 247.204554] klp_tc_10_livepatch: tainting kernel with TAINT_LIVEPATCH [ 247.205261] livepatch: enabling patch 'klp_tc_10_livepatch' [ 247.209206] livepatch: 'klp_tc_10_livepatch': starting patching transition [14:29:20] *** Check that live patch is blocked [ 248.318683] livepatch: 'klp_tc_10_livepatch': patching complete [14:29:21] TEST CASE ABORT patching finished prematurely [14:29:21] *** Removing patches [ 249.012780] livepatch: 'klp_tc_10_livepatch': starting unpatching transition [14:29:21] *** Disabling and removing module klp_tc_10_livepatch [ 250.402688] livepatch: 'klp_tc_10_livepatch': unpatching complete [14:29:24] *** Removing module klp_test_support_mod [14:29:24] rmmod: ERROR: Module klp_test_support_mod is in use -- You are receiving this mail because: You are the assignee for the bug.

https://bugzilla.suse.com/show_bug.cgi?id=1202727 https://bugzilla.suse.com/show_bug.cgi?id=1202727#c3 --- Comment #3 from Claudio Fontana --- Hi Marcos, does this problem affect bare metal also, or does this only show up with QEMU/kvm? -- You are receiving this mail because: You are the assignee for the bug.

https://bugzilla.suse.com/show_bug.cgi?id=1202727 https://bugzilla.suse.com/show_bug.cgi?id=1202727#c5 --- Comment #5 from Marcos de Souza --- To exemplify, I'm testing on a s390x machine. Using the current v6.0-rc4: commit 7e18e42e4b280c85b76967a9106a13ca61c16179 (HEAD, tag: v6.0-rc4, origin/master, origin/HEAD) Author: Linus Torvalds Date: Sun Sep 4 13:10:01 2022 -0700 Linux 6.0-rc4 And executing qemu-system-s390x, it does not show any output: qemu-system-s390x -kernel arch/s390/boot/vmlinux -nographic But, if the problematic commit (30de14b1884ba609fc1acfba5b40309e3a6ccefe) is reverted on top of 6.0-rc4, it panics as expected: commit 82364b4f5173954e5e6d324531a950e3b898db09 (HEAD) Author: Marcos Date: Mon Sep 5 17:45:27 2022 +0000 Revert "s390: current_stack_pointer shouldn't be a function" This reverts commit 30de14b1884ba609fc1acfba5b40309e3a6ccefe. qemu-system-s390x -kernel arch/s390/boot/vmlinux -nographic KASLR disabled: CPU has no PRNG [ 0.422349] Linux version 6.0.0-rc4-00001-g82364b4f5173 (kgrtst@s390zlpa) (gcc (SUSE Linux) 7.5.0, GNU ld (GNU Binutils; SUSE Linux Enterprise 15) 2.35.1.20201123-7.18) #36 SMP Mon Sep 5 17:51:20 UTC 2022 [ 0.423773] setup: Linux is running under KVM in 64-bit mode [ 0.462163] setup: The maximum memory size is 128MB [ 0.462679] setup: Relocating AMODE31 section of size 0x00003000 [ 0.464744] cpu: 1 configured CPUs, 0 standby CPUs ... I hope this helps to explain what is the problem and how to reproduce it. -- You are receiving this mail because: You are the assignee for the bug.

https://bugzilla.suse.com/show_bug.cgi?id=1202727 https://bugzilla.suse.com/show_bug.cgi?id=1202727#c7 --- Comment #7 from Marcos de Souza --- More info: The kernel is compiled with gcc 7.5.0, from SLE 15-SP1 -- You are receiving this mail because: You are the assignee for the bug.

https://bugzilla.suse.com/show_bug.cgi?id=1202727 https://bugzilla.suse.com/show_bug.cgi?id=1202727#c9 Marcos de Souza changed: What |Removed |Added ---------------------------------------------------------------------------- Flags| |needinfo?(ptesarik@suse.com | |) --- Comment #9 from Marcos de Souza --- (In reply to Petr Tesa����k from comment #8)

Let me think...

One difference between the inline function and a global register value is that the function contained an "asm volatile", so the compiler was not allowed to re-order execution. Is KLP using the value of "current_stack_pointer" in a context that might change %r15 by any chance?

If that's the case, you may have to save the stack pointer into another variable like this:

const unsigned long current_sp = current_stack_pointer

The "const" qualifier may be needed to tell the compiler that the value is initialized once and never changed.

But do you see a reason for that commit to make the boot process stuck? -- You are receiving this mail because: You are the assignee for the bug.