Re: [PATCH] tcm: Convert to cpu_relax() for atomic_read() in tightloops

[Nicholas A. Bellinger]

On Fri, 2010-09-10 at 08:32 +0200, Jiri Slaby wrote:
> On 09/09/2010 11:29 PM, Nicholas A. Bellinger wrote:
> > From: Nicholas Bellinger <nab@xxxxxxxxxxxxxxx>
> > 
> > Greetings all,
> > 
> > This patch converts the following functions from within shutdown and release
> > path code to use cpu_relax() instead of msleep() while doing atomic_read() in
> > a tight loop waiting for various PR and MIB reference count(s) to return to zero:
> > 
> > 	*) core_update_device_list_for_node() + struct se_dev_entry->pr_ref_count
> > 	*) core_release_port() + struct se_port->sep_tg_pt_ref_cnt
> > 	*) se_dev_stop() + struct se_hba->dev_mib_access_count
> > 	*) __core_scsi3_free_registration() + struct t10_pr_registration->pr_res_holders
> > 	*) core_tpg_wait_for_nacl_pr_ref() + struct se_node_acl->acl_pr_ref_count
> > 	*) core_tpg_wait_for_mib_ref() + struct se_node_acl->mib_ref_count
> > 	*) core_tpg_deregister() + struct se_portal_group->tpg_pr_ref_count
> > 	*) transport_deregister_session() + struct se_session->mib_ref_count
> > 
> > Thanks to Konrad Rzeszutek Wilk for mentioning cpu_relax() usage!
> > 
> > Signed-off-by: Nicholas A. Bellinger <nab@xxxxxxxxxxxxxxx>
> > ---
> >  drivers/target/target_core_device.c    |    6 +++---
> >  drivers/target/target_core_pr.c        |    2 +-
> >  drivers/target/target_core_tpg.c       |    6 +++---
> >  drivers/target/target_core_transport.c |    2 +-
> >  4 files changed, 8 insertions(+), 8 deletions(-)
> > 
> > diff --git a/drivers/target/target_core_device.c b/drivers/target/target_core_device.c
> > index 665b8e2..459f0f8 100644
> > --- a/drivers/target/target_core_device.c
> > +++ b/drivers/target/target_core_device.c
> > @@ -532,7 +532,7 @@ int core_update_device_list_for_node(
> >  	 */
> >  	spin_unlock_irq(&nacl->device_list_lock);
> >  	while (atomic_read(&deve->pr_ref_count) != 0)
> > -		msleep(100);
> > +		cpu_relax();
> >  	spin_lock_irq(&nacl->device_list_lock);
> >  	/*
> >  	 * Disable struct se_dev_entry LUN ACL mapping
> 
> Why would you want to convert those to busy-waiting? This should rather
> be converted to wait_event/wake_up, right?

Hi Jiri,

I agree that this is the one scenario within the msleep() -> cpu_relax()
conversion patch that involves something besides a simple shortly held
'non CDB I/O emulation path context ref/deref', and might probably be
better served by using wait_event/wake_up.

I will take a look at a conversion to use wait_queue_head_t in the near
future.

Thanks for your comments Jiri!

--nab


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/