Re: [PATCH 04/15] block: prevent elevator switch during updating nr_hw_queues

[Christoph Hellwig <hch@lst.de>]

On Mon, Apr 14, 2025 at 08:54:36AM +0800, Ming Lei wrote:
> > >  	elv_iosched_load_module(name);
> > >  
> > > +	idx = srcu_read_lock(&set->update_nr_hwq_srcu);
> > > +
> > > +	if (set->flags & BLK_MQ_F_UPDATE_HW_QUEUES) {
> > 
> > What provides atomicity for field modifications vs reading of set->flags?
> > i.e. does this need to switch using test/set_bit?
> 
> WRITE is serialized via tag set lock with synchronize_srcu().
> 
> READ is covered by srcu read lock.
> 
> It is typical RCU usage, one writer vs. multiple writer.

No, (S)RCU does not help you with atomicy of bitfields.

> > Also mixing internal state with driver provided flags is always
> > a bad idea.  So this should probably be a new state field in the
> > tag_set and not reuse flags.
>  
> That is fine, but BLK_MQ_F_TAG_QUEUE_SHARED is used in this way too.

Yes, that should also move over to the state field.  Amd rnbd needs
to be fixed to not set it diretly which is a good example for why
they should not be mixed..