Re: [Xen-devel] [PATCH v5 03/11] x86/mce: handle host LMCE

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 04.07.17 at 05:05, <haozhong.zhang@xxxxxxxxx> wrote:
> On 07/03/17 09:42 -0600, Jan Beulich wrote:
>> >>> On 03.07.17 at 05:46, <haozhong.zhang@xxxxxxxxx> wrote:
> [..]
>> >  void mctelem_process_deferred(unsigned int cpu,
>> > -                        int (*fn)(mctelem_cookie_t))
>> > +                        int (*fn)(mctelem_cookie_t),
>> > +                        bool lmce)
>> >  {
>> >    struct mctelem_ent *tep;
>> >    struct mctelem_ent *head, *prev;
>> > +  struct mc_telem_cpu_ctl *mctctl = &per_cpu(mctctl, cpu);
>> >    int ret;
>> >  
>> >    /*
>> >     * First, unhook the list of telemetry structures, and  
>> >     * hook it up to the processing list head for this CPU.
>> > +   *
>> > +   * If @lmce is true and a non-local MC# occurs before the
>> > +   * following atomic exchange, @lmce will not hold after
>> > +   * resumption, because all telemetries in @lmce_pending on
>> > +   * @cpu are moved to @pending on @cpu in mcheck_cmn_handler().
>> > +   * In such a case, no telemetries will be handled in this
>> > +   * function after resumption. Another round of MCE softirq,
>> > +   * which was raised by above mcheck_cmn_handler(), will handle
>> > +   * those moved telemetries in @pending on @cpu.
>> > +   *
>> > +   * If another MC# occurs after the following atomic exchange,
>> > +   * it will be handled by another round of MCE softirq.
>> 
>> This restates what the earlier paragraph of the comment already
>> says. With these taken care of
> 
> This paragraph is for the case that MC# occurs *after* the atomic
> exchange, while the previous paragraph is for the case that MC# occurs
> *before* the atomic exchange.

Oh, indeed. But isn't that obvious? In which case, re-phrasing it to
make clear this is nothing special may help. E.g. "Any MC# occurring
after the following atomic exchange will be handled by another round
of MCE softirq anyway"?

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel