Re: [PATCH 1/3] net: fs_enet: merge NAPI RX and NAPI TX

2016-08-24 Thread Christophe Leroy
Le 24/08/2016 à 15:07, Eric Dumazet a écrit : On Wed, 2016-08-24 at 12:36 +0200, Christophe Leroy wrote: Initially, a NAPI TX routine has been implemented separately from NAPI RX, as done on the freescale/gianfar driver. By merging NAPI RX and NAPI TX, we reduce the amount of TX completion

Re: [PATCH 1/3] net: fs_enet: merge NAPI RX and NAPI TX

2016-08-24 Thread Christophe Leroy
Le 24/08/2016 à 15:07, Eric Dumazet a écrit : On Wed, 2016-08-24 at 12:36 +0200, Christophe Leroy wrote: Initially, a NAPI TX routine has been implemented separately from NAPI RX, as done on the freescale/gianfar driver. By merging NAPI RX and NAPI TX, we reduce the amount of TX completion

Re: [PATCH 1/3] net: fs_enet: merge NAPI RX and NAPI TX

2016-08-24 Thread Eric Dumazet
On Wed, 2016-08-24 at 06:07 -0700, Eric Dumazet wrote: > I am afraid you could live lock here on SMP. > > You should make sure you do not loop forever, not assuming cpu is faster > than NIC. You are protected by the tx_lock spinlock, but this is fragile as you could very well remove this

Re: [PATCH 1/3] net: fs_enet: merge NAPI RX and NAPI TX

2016-08-24 Thread Eric Dumazet
On Wed, 2016-08-24 at 06:07 -0700, Eric Dumazet wrote: > I am afraid you could live lock here on SMP. > > You should make sure you do not loop forever, not assuming cpu is faster > than NIC. You are protected by the tx_lock spinlock, but this is fragile as you could very well remove this

Re: [PATCH 1/3] net: fs_enet: merge NAPI RX and NAPI TX

2016-08-24 Thread Eric Dumazet
On Wed, 2016-08-24 at 12:36 +0200, Christophe Leroy wrote: > Initially, a NAPI TX routine has been implemented separately from > NAPI RX, as done on the freescale/gianfar driver. > > By merging NAPI RX and NAPI TX, we reduce the amount of TX completion > interrupts. > > Handling of the budget in

Re: [PATCH 1/3] net: fs_enet: merge NAPI RX and NAPI TX

2016-08-24 Thread Eric Dumazet
On Wed, 2016-08-24 at 12:36 +0200, Christophe Leroy wrote: > Initially, a NAPI TX routine has been implemented separately from > NAPI RX, as done on the freescale/gianfar driver. > > By merging NAPI RX and NAPI TX, we reduce the amount of TX completion > interrupts. > > Handling of the budget in

[PATCH 1/3] net: fs_enet: merge NAPI RX and NAPI TX

2016-08-24 Thread Christophe Leroy
Initially, a NAPI TX routine has been implemented separately from NAPI RX, as done on the freescale/gianfar driver. By merging NAPI RX and NAPI TX, we reduce the amount of TX completion interrupts. Handling of the budget in association with TX interrupts is based on indications provided at

[PATCH 1/3] net: fs_enet: merge NAPI RX and NAPI TX

2016-08-24 Thread Christophe Leroy
Initially, a NAPI TX routine has been implemented separately from NAPI RX, as done on the freescale/gianfar driver. By merging NAPI RX and NAPI TX, we reduce the amount of TX completion interrupts. Handling of the budget in association with TX interrupts is based on indications provided at