On Wed, Oct 18, 2023 at 12:38 PM Tummala, Sivaprasad
<sivaprasad.tumm...@amd.com> wrote:
>
> [AMD Official Use Only - General]
>
> Hi Jerin,
>
> > -----Original Message-----
> > From: Jerin Jacob <jerinjac...@gmail.com>
> > Sent: Tuesday, October 17, 2023 8:53 AM
> > To: Tummala, Sivaprasad <sivaprasad.tumm...@amd.com>
> > Cc: harry.van.haa...@intel.com; anatoly.bura...@intel.com; dev@dpdk.org;
> > Yigit,
> > Ferruh <ferruh.yi...@amd.com>; david.h...@intel.com
> > Subject: Re: [PATCH v1 5/6] power: add eventdev support for power management
> >
> > Caution: This message originated from an External Source. Use proper caution
> > when opening attachments, clicking links, or responding.
> >
> >
> > On Tue, Oct 17, 2023 at 2:27 AM Sivaprasad Tummala
> > <sivaprasad.tumm...@amd.com> wrote:
> > >
> > > Add eventdev support to enable power saving when no events are
> > > arriving. It is based on counting the number of empty polls and, when
> > > the number reaches a certain threshold, entering an
> > > architecture-defined optimized power state that will either wait until
> > > a TSC timestamp expires, or when events arrive.
> > >
> > > This API mandates a core-to-single-port mapping (i.e. one core polling
> > > multiple ports of event device is not supported). This should be ok as
> > > the general use case will have one CPU core using one port to
> > > enqueue/dequeue events from an eventdev.
> > >
> > > This design is using Eventdev PMD Dequeue callbacks.
> > >
> > > 1. MWAITX/MONITORX:
> > >
> > > When a certain threshold of empty polls is reached, the core will go
> > > into a power optimized sleep while waiting on an address of next RX
> > > descriptor to be written to.
> > >
> > > 2. Pause instruction
> > >
> > > This method uses the pause instruction to avoid busy polling.
> > >
> > > Signed-off-by: Sivaprasad Tummala <sivaprasad.tumm...@amd.com>
> >
> >
> > Hi Siva,
> >
> > It does not look it is aligned with previous discussion.
> >
> > I spend a couple of minutes to draft semantics. Please treat as reference.
> >
> > # IMO, only following public SLOW PATH eventdev API required.(Just share the
> > concept)
> >
> > enum rte_event_pmgmt_modes {
> > /** Default power management scheme */
> > RTE_EVENT_POWER_MGMT_TYPE_DEFAULT;
> > /** Use power-optimized monitoring to wait for incoming traffic */
> > RTE_EVENT_POWER_MGMT_TYPE_F_CPU_MONITOR = RTE_BIT(0),
> > /** Use power-optimized sleep to avoid busy polling */
> > RTE_EVENT_POWER_MGMT_TYPE_F_CPU_PAUSE = RTE_BIT(1),
> > /** HW based power management scheme found in ARM64 machines, where
> > core goes to sleep state till event available on dequeue */
> > RTE_EVENT_POWER_MGMT_TYPE_F_HW_WFE_ON_DEQUEUE = RTE_BIT(2),
> >
> > };
> >
> > int rte_event_port_pmgmt_type_supported_get(uint8_t dev_id, enum
> > rte_event_pmgmt_modes *mode_flags)
> > /** Device must be in stop state */
> > int rte_event_port_pmgmt_enable(uint8_t dev_id, uint8_t port_id, enum
> > rte_event_pmgmt_modes mode); int rte_event_port_pmgmt_disable(uint8_t
> > dev_id, uint8_t port_id);
> >
> > # It should be self-contained, No need to add to rte_power as it is CPU
> > only power
> > mgmt.(See RTE_EVENT_POWER_MGMT_TYPE_F_HW_WFE_ON_DEQUEUE
> > above)
> >
> > # Add: lib/eventdev/eventdev_pmd_pmgmt.c or so and have CPU based on power
> > management helper functions so that all SW PMD can be reused.
> > example:
> > eventdev_pmd_pmgmt_handle_monitor(uint8_t dev_id, uint8_t port_id, struct
> > rte_event ev[], uint16_t nb_events);
> > eventdev_pmd_pmgmt_handle_pause(uint8_t dev_id, uint8_t port_id, struct
> > rte_event ev[], uint16_t nb_events);
> >
> >
> > # In rte_event_dev_start(), Fixup dev->dequeue_burst if CPU based power
> > management is applicable,and it is selected.
> > ie. new dev->dequeue_burst is existing PMD's dev->dequeue_burst +
> > eventdev_pmd_pmgmt_handle_.. (based on power management mode selected)
>
> Thanks for the clarification. Will incorporate the changes in next version of
> the patch (to support power management on event port).
> With the time constraints, I will defer the power management support on event
> port to next release. However to avoid ABI breakage,
> I will split the Patchset and push the patches to support callbacks in this
> release, so we don't have to wait for next stable release to
> get these changes integrated.
if you follow this scheme, public callback API is not needed.
# In rte_event_dev_start(), Fixup dev->dequeue_burst if CPU based power
management is applicable,and it is selected.
ie. new dev->dequeue_burst is existing PMD's dev->dequeue_burst +
eventdev_pmd_pmgmt_handle_.. (based on power management mode selected)
>
> Please let me know your thoughts.
>
> Thanks & Regards,
> Sivaprasad
