Re: [PATCH v2 2/5] x86/idle: Disable IBRS when cpu is offline

[Waiman Long <longman@redhat.com>]

On 6/21/23 03:23, Peter Zijlstra wrote:
> On Tue, Jun 20, 2023 at 10:06:22AM -0400, Waiman Long wrote:
>> Commit bf5835bcdb96 ("intel_idle: Disable IBRS during long idle")
>> disables IBRS when the CPU enters long idle. However, when a CPU becomes
>> offline, the IBRS bit is still set when X86_FEATURE_KERNEL_IBRS is
>> enabled. That will impact the performance of a sibling CPU. Mitigate
>> this performance impact by clearing all the mitigation bits in SPEC_CTRL
>> MSR when offline and restoring the value of the MSR when it becomes
>> online again.
>>
>> Signed-off-by: Waiman Long <longman@redhat.com>
>> ---
>>   arch/x86/kernel/smpboot.c | 13 +++++++++++++
>>   1 file changed, 13 insertions(+)
>>
>> diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
>> index 352f0ce1ece4..5ff82fef413c 100644
>> --- a/arch/x86/kernel/smpboot.c
>> +++ b/arch/x86/kernel/smpboot.c
>> @@ -84,6 +84,7 @@
>>   #include <asm/hw_irq.h>
>>   #include <asm/stackprotector.h>
>>   #include <asm/sev.h>
>> +#include <asm/nospec-branch.h>
>>   
>>   /* representing HT siblings of each logical CPU */
>>   DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_sibling_map);
>> @@ -1838,12 +1839,24 @@ void __noreturn hlt_play_dead(void)
>>   
>>   void native_play_dead(void)
>>   {
>> +	u64 spec_ctrl = spec_ctrl_current();
>> +
>> +	if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS)) {
>> +		this_cpu_write(x86_spec_ctrl_current, 0);
>> +		native_wrmsrl(MSR_IA32_SPEC_CTRL, 0);
>> +	}
>> +
>>   	play_dead_common();
>>   	tboot_shutdown(TB_SHUTDOWN_WFS);
>>   
>>   	mwait_play_dead();
>>   	if (cpuidle_play_dead())
>>   		hlt_play_dead();
>> +
>> +	if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS)) {
>> +		native_wrmsrl(MSR_IA32_SPEC_CTRL, spec_ctrl);
>> +		this_cpu_write(x86_spec_ctrl_current, spec_ctrl);
>> +	}
>>   }
> play_dead() is marked __noreturn

There are different versions of play_dead() in the kernel. Some of them 
are indeed marked __noreturn like the non-SMP one in 
arch/x86/kernel/process.c. The native_play_dead() that I am patching 
isn't one of those.

Cheers,
Longman