Re: [PATCH -next for tip:x86/pti] x86/tlb: drop unneeded local vars in enable_l1d_flush_for_task()

[Thomas Gleixner]

On Thu, Oct 01 2020 at 10:48, Balbir Singh wrote:
> On 1/10/20 9:49 am, Singh, Balbir wrote:
>>  
>> +static void l1d_flush_kill(struct callback_head *ch)
>> +{
>> +    clear_ti_thread_flag(&current->thread_info, TIF_SPEC_L1D_FLUSH);
>> +    force_signal(SIGBUS);
>> +}
>> +
>>  void switch_mm(struct mm_struct *prev, struct mm_struct *next,
>>             struct task_struct *tsk)
>>  {
>> @@ -443,12 +438,14 @@ static void cond_mitigation(struct task_struct *next)
>>      }
>>  
>>      /*
>> -     * Flush only if SMT is disabled as per the contract, which is checked
>> -     * when the feature is enabled.
>> +     * Flush only if SMT is disabled, if flushing is enabled
>> +     * and we are on an SMT enabled core, kill the task
>>       */
>> -    if (sched_smt_active() && !this_cpu_read(cpu_info.smt_active) &&
>> -            (prev_mm & LAST_USER_MM_L1D_FLUSH))
>> -            l1d_flush_hw();
>> +    if (unlikely(prev_mm & LAST_USER_MM_L1D_FLUSH)) {
>> +            if (!this_cpu_read(cpu_info.smt_active))
>> +                    l1d_flush_hw();
>> +            else
>> +                    task_work_add(prev, l1d_flush_kill, true);
>
> We have no access the to the previous task and mm->owner depends on MEMCG :)
> We can do the magic in mm_mangle_tif_spec_bits(), I suppose

No, because we don't have access to prev task there either. Interesting
problem to solve.

Thanks,

        tglx