diff mbox

[15/18] staging: fsl-dpaa2/eth: Reorder priv fields

Message ID 20170606150041.10955-16-ruxandra.radulescu@nxp.com (mailing list archive)
State New, archived
Headers show

Commit Message

Ioana Ciocoi Radulescu June 6, 2017, 3 p.m. UTC
Reorder fields in the device's private structure such that
the ones most used on the hotpath are at the beginning of
the structure.

This brings a small performance improvement due to better
cache usage.

Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
---
 drivers/staging/fsl-dpaa2/ethernet/dpaa2-eth.h | 26 +++++++++++---------------
 1 file changed, 11 insertions(+), 15 deletions(-)

Comments

Greg Kroah-Hartman June 13, 2017, 10:03 a.m. UTC | #1
On Tue, Jun 06, 2017 at 10:00:38AM -0500, Ioana Radulescu wrote:
> Reorder fields in the device's private structure such that
> the ones most used on the hotpath are at the beginning of
> the structure.
> 
> This brings a small performance improvement due to better
> cache usage.

Really?  What kind of performance improvement?  Always be specific.

Please fix up and resend.

thanks,

greg k-h
Ioana Ciocoi Radulescu June 14, 2017, 1:09 p.m. UTC | #2
> -----Original Message-----
> From: Greg KH [mailto:gregkh@linuxfoundation.org]
> Sent: Tuesday, June 13, 2017 1:04 PM
> To: Ruxandra Ioana Radulescu <ruxandra.radulescu@nxp.com>
> Cc: devel@driverdev.osuosl.org; stuyoder@gmail.com; arnd@arndb.de;
> linux-kernel@vger.kernel.org; agraf@suse.de; Bharat Bhushan
> <bharat.bhushan@nxp.com>; Bogdan Purcareata
> <bogdan.purcareata@nxp.com>; linux-arm-kernel@lists.infradead.org;
> Laurentiu Tudor <laurentiu.tudor@nxp.com>
> Subject: Re: [PATCH 15/18] staging: fsl-dpaa2/eth: Reorder priv fields
> 
> On Tue, Jun 06, 2017 at 10:00:38AM -0500, Ioana Radulescu wrote:
> > Reorder fields in the device's private structure such that
> > the ones most used on the hotpath are at the beginning of
> > the structure.
> >
> > This brings a small performance improvement due to better
> > cache usage.
> 
> Really?  What kind of performance improvement?  Always be specific.

Fair point, thanks.

The test scenario was ipfwd with small frames and the improvement ~1%,
but I took this patch from an internal repo, so I need to redo the
measurements on the current tree before resending.

> 
> Please fix up and resend.

We're preparing a larger patchset with several optimizations, I'll
probably include the v2 there.

Thanks,
Ioana
diff mbox

Patch

diff --git a/drivers/staging/fsl-dpaa2/ethernet/dpaa2-eth.h b/drivers/staging/fsl-dpaa2/ethernet/dpaa2-eth.h
index 886a0681fee1..c760e9b77e22 100644
--- a/drivers/staging/fsl-dpaa2/ethernet/dpaa2-eth.h
+++ b/drivers/staging/fsl-dpaa2/ethernet/dpaa2-eth.h
@@ -290,34 +290,30 @@  struct dpaa2_eth_hash_fields {
 /* Driver private data */
 struct dpaa2_eth_priv {
 	struct net_device *net_dev;
+	/* Standard statistics */
+	struct rtnl_link_stats64 __percpu *percpu_stats;
+	/* Extra stats, in addition to the ones known by the kernel */
+	struct dpaa2_eth_drv_stats __percpu *percpu_extras;
+	struct iommu_domain *iommu_domain;
+	u16 tx_data_offset;
+	u16 bpid;
+	u16 tx_qdid;
 
 	u8 num_fqs;
-	struct dpaa2_eth_fq fq[DPAA2_ETH_MAX_QUEUES];
-
 	u8 num_channels;
+	struct dpaa2_eth_fq fq[DPAA2_ETH_MAX_QUEUES];
 	struct dpaa2_eth_channel *channel[DPAA2_ETH_MAX_DPCONS];
 
+	struct fsl_mc_io *mc_io;
+	u16 mc_token;
 	struct dpni_attr dpni_attrs;
-	u16 tx_data_offset;
-
 	struct fsl_mc_device *dpbp_dev;
-	u16 bpid;
-	struct iommu_domain *iommu_domain;
 
-	u16 tx_qdid;
-	struct fsl_mc_io *mc_io;
 	/* Cores which have an affine DPIO/DPCON.
 	 * This is the cpu set on which Rx and Tx conf frames are processed
 	 */
 	struct cpumask dpio_cpumask;
 
-	/* Standard statistics */
-	struct rtnl_link_stats64 __percpu *percpu_stats;
-	/* Extra stats, in addition to the ones known by the kernel */
-	struct dpaa2_eth_drv_stats __percpu *percpu_extras;
-
-	u16 mc_token;
-
 	struct dpni_link_state link_state;
 	bool do_link_poll;
 	struct task_struct *poll_thread;