| Message ID | cover.1653000219.git.davthompson@nvidia.com |
|---|---|
| Headers | show |
| Series | mlxbf_gige: increase MDIO polling rate to 5us | expand |
This patch, though simple enough, has had no upstream review and should therefore be an 'UBUNTU: SAUCE:' patch until such time as it is included in net-next or linux-next. Furthermore, define a macro for the POLL time instead of duplicating magic numbers in the code. rtg On 5/19/22 16:49, David Thompson wrote: > BugLink: https://bugs.launchpad.net/bugs/1974246 > > SRU Justification: > > [Impact] > > There is MDIO logic within the mlxbf_gige driver that currently > polls for completed transactions every 100us. This has been > reviewed and tested in the lab and found to be inefficient. > > [Fix] > > The fix is to modify the driver logic to increase the MDIO > polling rate to 5us. With this change the amount of > time spent waiting for the MDIO BUSY signal to de-assert > drops from ~100us to ~27us for each operation. > > [Test Case] > > The driver should function as before, specifically: > * driver probes successfully > * oob_net0 link comes up, as seen with "ifconfig -a" > * reset and power cycle testing toggle PHY hardware, > and oob_net0 link still comes up > > [Regression Potential] > > This change affects all MDIO transactions to the board-level > PHY device, and as such, could cause a regression with PHY > communications. However, this change has been heavily tested. > > David Thompson (1): > mlxbf_gige: increase MDIO polling rate to 5us > > drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_mdio.c | 6 ++++-- > 1 file changed, 4 insertions(+), 2 deletions(-) >
Hello Tim, This patch was already applied to net-next (see attached email) - Dave > -----Original Message----- > From: Tim Gardner <tim.gardner@canonical.com> > Sent: Friday, May 20, 2022 8:35 AM > To: David Thompson <davthompson@nvidia.com>; kernel- > team@lists.ubuntu.com > Cc: Meriton Tuli <meriton@nvidia.com>; Khoa Vo <khoav@nvidia.com> > Subject: NAK: [SRU][F:linux-bluefield][PATCH v1 0/1] mlxbf_gige: increase MDIO > polling rate to 5us > > This patch, though simple enough, has had no upstream review and should > therefore be an 'UBUNTU: SAUCE:' patch until such time as it is included in net- > next or linux-next. Furthermore, define a macro for the POLL time instead of > duplicating magic numbers in the code. > > rtg > > On 5/19/22 16:49, David Thompson wrote: > > BugLink: https://bugs.launchpad.net/bugs/1974246 > > > > SRU Justification: > > > > [Impact] > > > > There is MDIO logic within the mlxbf_gige driver that currently polls > > for completed transactions every 100us. This has been reviewed and > > tested in the lab and found to be inefficient. > > > > [Fix] > > > > The fix is to modify the driver logic to increase the MDIO polling > > rate to 5us. With this change the amount of time spent waiting for the > > MDIO BUSY signal to de-assert drops from ~100us to ~27us for each > > operation. > > > > [Test Case] > > > > The driver should function as before, specifically: > > * driver probes successfully > > * oob_net0 link comes up, as seen with "ifconfig -a" > > * reset and power cycle testing toggle PHY hardware, > > and oob_net0 link still comes up > > > > [Regression Potential] > > > > This change affects all MDIO transactions to the board-level PHY > > device, and as such, could cause a regression with PHY communications. > > However, this change has been heavily tested. > > > > David Thompson (1): > > mlxbf_gige: increase MDIO polling rate to 5us > > > > drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_mdio.c | 6 ++++-- > > 1 file changed, 4 insertions(+), 2 deletions(-) > > > > -- > ----------- > Tim Gardner > Canonical, Inc Hello: This patch was applied to netdev/net-next.git (master) by Jakub Kicinski <kuba@kernel.org>: On Thu, 5 May 2022 12:23:09 -0400 you wrote: > This patch increases the polling rate used by the > mlxbf_gige driver on the MDIO bus. The previous > polling rate was every 100us, and the new rate is > every 5us. With this change the amount of time > spent waiting for the MDIO BUSY signal to de-assert > drops from ~100us to ~27us for each operation. > > [...] Here is the summary with links: - [net-next,v1] mlxbf_gige: increase MDIO polling rate to 5us https://git.kernel.org/netdev/net-next/c/0a02e282bad4 You are awesome, thank you! -- Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html
Acked-by: Tim Gardner <tim.gardner@canonical.com> The review should still have complained about the proliferation of magic numbers. Its one of my pet peeves. rtg On 5/20/22 08:46, David Thompson wrote: > Hello Tim, > > This patch was already applied to net-next (see attached email) > > - Dave > >> -----Original Message----- >> From: Tim Gardner <tim.gardner@canonical.com> >> Sent: Friday, May 20, 2022 8:35 AM >> To: David Thompson <davthompson@nvidia.com>; kernel- >> team@lists.ubuntu.com >> Cc: Meriton Tuli <meriton@nvidia.com>; Khoa Vo <khoav@nvidia.com> >> Subject: NAK: [SRU][F:linux-bluefield][PATCH v1 0/1] mlxbf_gige: increase MDIO >> polling rate to 5us >> >> This patch, though simple enough, has had no upstream review and should >> therefore be an 'UBUNTU: SAUCE:' patch until such time as it is included in net- >> next or linux-next. Furthermore, define a macro for the POLL time instead of >> duplicating magic numbers in the code. >> >> rtg >> >> On 5/19/22 16:49, David Thompson wrote: >>> BugLink: https://bugs.launchpad.net/bugs/1974246 >>> >>> SRU Justification: >>> >>> [Impact] >>> >>> There is MDIO logic within the mlxbf_gige driver that currently polls >>> for completed transactions every 100us. This has been reviewed and >>> tested in the lab and found to be inefficient. >>> >>> [Fix] >>> >>> The fix is to modify the driver logic to increase the MDIO polling >>> rate to 5us. With this change the amount of time spent waiting for the >>> MDIO BUSY signal to de-assert drops from ~100us to ~27us for each >>> operation. >>> >>> [Test Case] >>> >>> The driver should function as before, specifically: >>> * driver probes successfully >>> * oob_net0 link comes up, as seen with "ifconfig -a" >>> * reset and power cycle testing toggle PHY hardware, >>> and oob_net0 link still comes up >>> >>> [Regression Potential] >>> >>> This change affects all MDIO transactions to the board-level PHY >>> device, and as such, could cause a regression with PHY communications. >>> However, this change has been heavily tested. >>> >>> David Thompson (1): >>> mlxbf_gige: increase MDIO polling rate to 5us >>> >>> drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_mdio.c | 6 ++++-- >>> 1 file changed, 4 insertions(+), 2 deletions(-) >>> >> >> -- >> ----------- >> Tim Gardner >> Canonical, Inc
Applied to f:bluefield. Thanks, Zack On 5/19/22 6:49 PM, David Thompson wrote: > BugLink: https://bugs.launchpad.net/bugs/1974246 > > SRU Justification: > > [Impact] > > There is MDIO logic within the mlxbf_gige driver that currently > polls for completed transactions every 100us. This has been > reviewed and tested in the lab and found to be inefficient. > > [Fix] > > The fix is to modify the driver logic to increase the MDIO > polling rate to 5us. With this change the amount of > time spent waiting for the MDIO BUSY signal to de-assert > drops from ~100us to ~27us for each operation. > > [Test Case] > > The driver should function as before, specifically: > * driver probes successfully > * oob_net0 link comes up, as seen with "ifconfig -a" > * reset and power cycle testing toggle PHY hardware, > and oob_net0 link still comes up > > [Regression Potential] > > This change affects all MDIO transactions to the board-level > PHY device, and as such, could cause a regression with PHY > communications. However, this change has been heavily tested. > > David Thompson (1): > mlxbf_gige: increase MDIO polling rate to 5us > > drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_mdio.c | 6 ++++-- > 1 file changed, 4 insertions(+), 2 deletions(-) >