| Message ID | 20241205160602.3lIAsJRT@linutronix.de (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | preempt: Move PREEMPT_RT before PREEMPT in vermagic. | expand |
| Context | Check | Description |
|---|---|---|
| mcgrof/vmtest-modules-next-PR | fail | PR summary |
| mcgrof/vmtest-modules-next-VM_Test-0 | success | Logs for Run CI tests |
| mcgrof/vmtest-modules-next-VM_Test-2 | success | Logs for setup / Setup kdevops environment |
| mcgrof/vmtest-modules-next-VM_Test-1 | success | Logs for cleanup / Archive results and cleanup |
On 12/5/24 17:06, Sebastian Andrzej Siewior wrote: > Since the dynamic preemption has been enabled for PREEMPT_RT we have now > CONFIG_PREEMPT and CONFIG_PREEMPT_RT set simultaneously. This affects > the vermagic strings which comes now PREEMPT with PREEMPT_RT enabled. > > The PREEMPT_RT module usually can not be loaded on a PREEMPT kernel > because some symbols are missing. > However if the symbols are fine then it continues and it crashes later. > The problem is that the struct module has a different layout and the > num_exentries or init members are at a different position leading to a > crash later on. This is not necessary caught by the size check in > elf_validity_cache_index_mod() because the mem member has an alignment > requirement of __module_memory_align which is big enough keep the total > size unchanged. Therefore we should keep the string accurate instead of > removing it. > > Move the PREEMPT_RT check before the PREEMPT so that it takes precedence > if both symbols are enabled. > > Fixes: 35772d627b55c ("sched: Enable PREEMPT_DYNAMIC for PREEMPT_RT") > Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Reviewed-by: Petr Pavlu <petr.pavlu@suse.com> I'm going to wait for a few days if others want to comment and then plan to take this through the modules tree for 6.13-rcX fixes. [Luis asked me to look after the modules tree for a while, with the plan to rotate it.]
On 2024-12-10 14:52:23 [+0100], Petr Pavlu wrote: > > Reviewed-by: Petr Pavlu <petr.pavlu@suse.com> > > I'm going to wait for a few days if others want to comment and then plan > to take this through the modules tree for 6.13-rcX fixes. thank you. > [Luis asked me to look after the modules tree for a while, with the plan > to rotate it.] understood. Sebastian
diff --git a/include/linux/vermagic.h b/include/linux/vermagic.h index a54046bf37e55..939ceabcaf06f 100644 --- a/include/linux/vermagic.h +++ b/include/linux/vermagic.h @@ -15,10 +15,10 @@ #else #define MODULE_VERMAGIC_SMP "" #endif -#ifdef CONFIG_PREEMPT_BUILD -#define MODULE_VERMAGIC_PREEMPT "preempt " -#elif defined(CONFIG_PREEMPT_RT) +#ifdef CONFIG_PREEMPT_RT #define MODULE_VERMAGIC_PREEMPT "preempt_rt " +#elif defined(CONFIG_PREEMPT_BUILD) +#define MODULE_VERMAGIC_PREEMPT "preempt " #else #define MODULE_VERMAGIC_PREEMPT "" #endif
Since the dynamic preemption has been enabled for PREEMPT_RT we have now CONFIG_PREEMPT and CONFIG_PREEMPT_RT set simultaneously. This affects the vermagic strings which comes now PREEMPT with PREEMPT_RT enabled. The PREEMPT_RT module usually can not be loaded on a PREEMPT kernel because some symbols are missing. However if the symbols are fine then it continues and it crashes later. The problem is that the struct module has a different layout and the num_exentries or init members are at a different position leading to a crash later on. This is not necessary caught by the size check in elf_validity_cache_index_mod() because the mem member has an alignment requirement of __module_memory_align which is big enough keep the total size unchanged. Therefore we should keep the string accurate instead of removing it. Move the PREEMPT_RT check before the PREEMPT so that it takes precedence if both symbols are enabled. Fixes: 35772d627b55c ("sched: Enable PREEMPT_DYNAMIC for PREEMPT_RT") Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> --- include/linux/vermagic.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)