Message ID | 20161208001248.13098.93773.stgit@localhost.localdomain |
---|---|
State | Changes Requested |
Delegated to: | Jeff Kirsher |
Headers | show |
On 12/7/2016 4:14 PM, Alexander Duyck wrote: > From: Alexander Duyck <alexander.h.duyck@intel.com> > > Recently napi_complete_done was updated so that it will return zero if we > are using the queue vector for busy polling. When this is the case we > don't need to re-enable interrupts as the busy polling will reschedule the > NAPI instance when it is finally done polling. This patch is causing a regression with memcached when using epoll with busy polling. I think we need to do more investigation before pushing or re-working this patch. Thanks Sridhar > Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com> > --- > > v2: Updated code to place the IRQ enabling inside the if statment > This made it so that we are consistent with the recent return value changes > > v3: Added i40evf changes as well > > drivers/net/ethernet/intel/i40e/i40e_txrx.c | 32 +++++++++++++------------ > drivers/net/ethernet/intel/i40evf/i40e_txrx.c | 28 ++++++++++++---------- > 2 files changed, 32 insertions(+), 28 deletions(-) > > diff --git a/drivers/net/ethernet/intel/i40e/i40e_txrx.c b/drivers/net/ethernet/intel/i40e/i40e_txrx.c > index e88e335c34ef..f9ef8c2676e2 100644 > --- a/drivers/net/ethernet/intel/i40e/i40e_txrx.c > +++ b/drivers/net/ethernet/intel/i40e/i40e_txrx.c > @@ -2033,22 +2033,24 @@ int i40e_napi_poll(struct napi_struct *napi, int budget) > } > } > > - if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR) > - q_vector->arm_wb_state = false; > - > - /* Work is done so exit the polling mode and re-enable the interrupt */ > - napi_complete_done(napi, work_done); > - > - /* If we're prematurely stopping polling to fix the interrupt > - * affinity we want to make sure polling starts back up so we > - * issue a call to i40e_force_wb which triggers a SW interrupt. > + /* Work is done so exit the polling mode, > + * if not busy polling re-enable interrupts. > */ > - if (!clean_complete) > - i40e_force_wb(vsi, q_vector); > - else if (!(vsi->back->flags & I40E_FLAG_MSIX_ENABLED)) > - i40e_irq_dynamic_enable_icr0(vsi->back, false); > - else > - i40e_update_enable_itr(vsi, q_vector); > + if (napi_complete_done(napi, work_done)) { > + if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR) > + q_vector->arm_wb_state = false; > + > + /* If we're prematurely stopping polling to fix the interrupt > + * affinity we want to make sure polling starts back up so we > + * issue a call to i40e_force_wb which triggers a SW interrupt. > + */ > + if (!clean_complete) > + i40e_force_wb(vsi, q_vector); > + else if (!(vsi->back->flags & I40E_FLAG_MSIX_ENABLED)) > + i40e_irq_dynamic_enable_icr0(vsi->back, false); > + else > + i40e_update_enable_itr(vsi, q_vector); > + } > > return min(work_done, budget - 1); > } > diff --git a/drivers/net/ethernet/intel/i40evf/i40e_txrx.c b/drivers/net/ethernet/intel/i40evf/i40e_txrx.c > index 4870cb50dc33..3097a91bd4aa 100644 > --- a/drivers/net/ethernet/intel/i40evf/i40e_txrx.c > +++ b/drivers/net/ethernet/intel/i40evf/i40e_txrx.c > @@ -1494,20 +1494,22 @@ int i40evf_napi_poll(struct napi_struct *napi, int budget) > } > } > > - if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR) > - q_vector->arm_wb_state = false; > - > - /* Work is done so exit the polling mode and re-enable the interrupt */ > - napi_complete_done(napi, work_done); > - > - /* If we're prematurely stopping polling to fix the interrupt > - * affinity we want to make sure polling starts back up so we > - * issue a call to i40evf_force_wb which triggers a SW interrupt. > + /* Work is done so exit the polling mode, > + * if not busy polling re-enable interrupts. > */ > - if (!clean_complete) > - i40evf_force_wb(vsi, q_vector); > - else > - i40e_update_enable_itr(vsi, q_vector); > + if (napi_complete_done(napi, work_done)) { > + if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR) > + q_vector->arm_wb_state = false; > + > + /* If we're prematurely stopping polling to fix the interrupt > + * affinity we want to make sure polling starts back up so we > + * issue a call to i40evf_force_wb which triggers a SW interrupt. > + */ > + if (!clean_complete) > + i40evf_force_wb(vsi, q_vector); > + else > + i40e_update_enable_itr(vsi, q_vector); > + } > > return min(work_done, budget - 1); > } >
On Wed, Dec 7, 2016 at 4:14 PM, Alexander Duyck <alexander.duyck@gmail.com> wrote: > From: Alexander Duyck <alexander.h.duyck@intel.com> > > Recently napi_complete_done was updated so that it will return zero if we > are using the queue vector for busy polling. When this is the case we > don't need to re-enable interrupts as the busy polling will reschedule the > NAPI instance when it is finally done polling. > > Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com> > --- > > v2: Updated code to place the IRQ enabling inside the if statment > This made it so that we are consistent with the recent return value changes > > v3: Added i40evf changes as well > > drivers/net/ethernet/intel/i40e/i40e_txrx.c | 32 +++++++++++++------------ > drivers/net/ethernet/intel/i40evf/i40e_txrx.c | 28 ++++++++++++---------- > 2 files changed, 32 insertions(+), 28 deletions(-) > Jeff, This patch causes a performance regression due to the way it interacts with the i40e WB on ITR behavior. Can you please drop it from the dev-queue. Thanks. - Alex
On Thu, Jan 5, 2017 at 2:05 PM, Alexander Duyck <alexander.duyck@gmail.com> wrote: > On Wed, Dec 7, 2016 at 4:14 PM, Alexander Duyck > <alexander.duyck@gmail.com> wrote: >> From: Alexander Duyck <alexander.h.duyck@intel.com> >> >> Recently napi_complete_done was updated so that it will return zero if we >> are using the queue vector for busy polling. When this is the case we >> don't need to re-enable interrupts as the busy polling will reschedule the >> NAPI instance when it is finally done polling. >> >> Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com> >> --- >> >> v2: Updated code to place the IRQ enabling inside the if statment >> This made it so that we are consistent with the recent return value changes >> >> v3: Added i40evf changes as well >> >> drivers/net/ethernet/intel/i40e/i40e_txrx.c | 32 +++++++++++++------------ >> drivers/net/ethernet/intel/i40evf/i40e_txrx.c | 28 ++++++++++++---------- >> 2 files changed, 32 insertions(+), 28 deletions(-) >> > > Jeff, > > This patch causes a performance regression due to the way it interacts > with the i40e WB on ITR behavior. Can you please drop it from the > dev-queue. > > Thanks. > > - Alex Re-sending this time and making sure it is sent to Jeff. Thanks. - Alex
On Thu, 2017-01-05 at 14:05 -0800, Alexander Duyck wrote: > On Wed, Dec 7, 2016 at 4:14 PM, Alexander Duyck > <alexander.duyck@gmail.com> wrote: > > From: Alexander Duyck <alexander.h.duyck@intel.com> > > > > Recently napi_complete_done was updated so that it will return zero if > > we > > are using the queue vector for busy polling. When this is the case we > > don't need to re-enable interrupts as the busy polling will reschedule > > the > > NAPI instance when it is finally done polling. > > > > Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com> > > --- > > > > v2: Updated code to place the IRQ enabling inside the if statment > > This made it so that we are consistent with the recent return > > value changes > > > > v3: Added i40evf changes as well > > > > drivers/net/ethernet/intel/i40e/i40e_txrx.c | 32 +++++++++++++--- > > --------- > > drivers/net/ethernet/intel/i40evf/i40e_txrx.c | 28 ++++++++++++-- > > -------- > > 2 files changed, 32 insertions(+), 28 deletions(-) > > > > Jeff, > > This patch causes a performance regression due to the way it interacts > with the i40e WB on ITR behavior. Can you please drop it from the > dev-queue. I have dropped the patch from my queue/tree.
diff --git a/drivers/net/ethernet/intel/i40e/i40e_txrx.c b/drivers/net/ethernet/intel/i40e/i40e_txrx.c index e88e335c34ef..f9ef8c2676e2 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_txrx.c +++ b/drivers/net/ethernet/intel/i40e/i40e_txrx.c @@ -2033,22 +2033,24 @@ int i40e_napi_poll(struct napi_struct *napi, int budget) } } - if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR) - q_vector->arm_wb_state = false; - - /* Work is done so exit the polling mode and re-enable the interrupt */ - napi_complete_done(napi, work_done); - - /* If we're prematurely stopping polling to fix the interrupt - * affinity we want to make sure polling starts back up so we - * issue a call to i40e_force_wb which triggers a SW interrupt. + /* Work is done so exit the polling mode, + * if not busy polling re-enable interrupts. */ - if (!clean_complete) - i40e_force_wb(vsi, q_vector); - else if (!(vsi->back->flags & I40E_FLAG_MSIX_ENABLED)) - i40e_irq_dynamic_enable_icr0(vsi->back, false); - else - i40e_update_enable_itr(vsi, q_vector); + if (napi_complete_done(napi, work_done)) { + if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR) + q_vector->arm_wb_state = false; + + /* If we're prematurely stopping polling to fix the interrupt + * affinity we want to make sure polling starts back up so we + * issue a call to i40e_force_wb which triggers a SW interrupt. + */ + if (!clean_complete) + i40e_force_wb(vsi, q_vector); + else if (!(vsi->back->flags & I40E_FLAG_MSIX_ENABLED)) + i40e_irq_dynamic_enable_icr0(vsi->back, false); + else + i40e_update_enable_itr(vsi, q_vector); + } return min(work_done, budget - 1); } diff --git a/drivers/net/ethernet/intel/i40evf/i40e_txrx.c b/drivers/net/ethernet/intel/i40evf/i40e_txrx.c index 4870cb50dc33..3097a91bd4aa 100644 --- a/drivers/net/ethernet/intel/i40evf/i40e_txrx.c +++ b/drivers/net/ethernet/intel/i40evf/i40e_txrx.c @@ -1494,20 +1494,22 @@ int i40evf_napi_poll(struct napi_struct *napi, int budget) } } - if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR) - q_vector->arm_wb_state = false; - - /* Work is done so exit the polling mode and re-enable the interrupt */ - napi_complete_done(napi, work_done); - - /* If we're prematurely stopping polling to fix the interrupt - * affinity we want to make sure polling starts back up so we - * issue a call to i40evf_force_wb which triggers a SW interrupt. + /* Work is done so exit the polling mode, + * if not busy polling re-enable interrupts. */ - if (!clean_complete) - i40evf_force_wb(vsi, q_vector); - else - i40e_update_enable_itr(vsi, q_vector); + if (napi_complete_done(napi, work_done)) { + if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR) + q_vector->arm_wb_state = false; + + /* If we're prematurely stopping polling to fix the interrupt + * affinity we want to make sure polling starts back up so we + * issue a call to i40evf_force_wb which triggers a SW interrupt. + */ + if (!clean_complete) + i40evf_force_wb(vsi, q_vector); + else + i40e_update_enable_itr(vsi, q_vector); + } return min(work_done, budget - 1); }