[net-next] octeontx2-af: cn10k: Increase outstanding LMTST transactions
Commit Message
From: Pavan Nikhilesh <pbhagavatula@marvell.com>
Currently the number of outstanding store transactions issued by AP as
a part of LMTST operation is set to 1 i.e default value.
This patch set to max supported value to increase the performance.
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Signed-off-by: Geetha sowjanya <gakula@marvell.com>
---
drivers/net/ethernet/marvell/octeontx2/af/rvu.c | 3 +++
drivers/net/ethernet/marvell/octeontx2/af/rvu.h | 1 +
drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c | 9 +++++++++
3 files changed, 13 insertions(+)
Comments
On Wed, Nov 29, 2023 at 04:51:55PM +0530, Geetha sowjanya wrote:
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Currently the number of outstanding store transactions issued by AP as
> a part of LMTST operation is set to 1 i.e default value.
> This patch set to max supported value to increase the performance.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Signed-off-by: Geetha sowjanya <gakula@marvell.com>
...
> diff --git a/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c b/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
> index 0e74c5a2231e..93fedabfe31e 100644
> --- a/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
> +++ b/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
> @@ -559,3 +559,12 @@ void rvu_nix_block_cn10k_init(struct rvu *rvu, struct nix_hw *nix_hw)
> cfg |= BIT_ULL(1) | BIT_ULL(2);
> rvu_write64(rvu, blkaddr, NIX_AF_CFG, cfg);
> }
> +
> +void rvu_apr_block_cn10k_init(struct rvu *rvu)
> +{
> + u64 reg;
> +
> + reg = rvu_read64(rvu, BLKADDR_APR, APR_AF_LMT_CFG);
> + reg |= 0xFULL << 35;
Hi Pavan and Geetha,
I think it would be best to avoid the magic value 35 here.
Best would probably be to use GENMASK_ULL and FIELD_PREP.
Else defining something similar to APR_LMT_MAP_ENT_SCH_ENA_SHIFT.
It might also be nice to avoid the magic value 0xFULL using a #define.
> + rvu_write64(rvu, BLKADDR_APR, APR_AF_LMT_CFG, reg);
> +}
> --
> 2.25.1
>
> -----Original Message-----
> From: Simon Horman <horms@kernel.org>
> Sent: Friday, December 1, 2023 5:05 PM
> To: Geethasowjanya Akula <gakula@marvell.com>
> Cc: netdev@vger.kernel.org; linux-kernel@vger.kernel.org; kuba@kernel.org;
> davem@davemloft.net; pabeni@redhat.com; edumazet@google.com; Sunil
> Kovvuri Goutham <sgoutham@marvell.com>; Linu Cherian
> <lcherian@marvell.com>; Jerin Jacob Kollanukkaran <jerinj@marvell.com>;
> Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>; Subbaraya
> Sundeep Bhatta <sbhatta@marvell.com>; Hariprasad Kelam
> <hkelam@marvell.com>
> Subject: [EXT] Re: [net-next PATCH] octeontx2-af: cn10k: Increase
> outstanding LMTST transactions
>
> External Email
>
> ----------------------------------------------------------------------
> On Wed, Nov 29, 2023 at 04:51:55PM +0530, Geetha sowjanya wrote:
> > From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >
> > Currently the number of outstanding store transactions issued by AP as
> > a part of LMTST operation is set to 1 i.e default value.
> > This patch set to max supported value to increase the performance.
> >
> > Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> > Signed-off-by: Geetha sowjanya <gakula@marvell.com>
>
> ...
>
> > diff --git a/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
> > b/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
> > index 0e74c5a2231e..93fedabfe31e 100644
> > --- a/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
> > +++ b/drivers/net/ethernet/marvell/octeontx2/af/rvu_cn10k.c
> > @@ -559,3 +559,12 @@ void rvu_nix_block_cn10k_init(struct rvu *rvu,
> struct nix_hw *nix_hw)
> > cfg |= BIT_ULL(1) | BIT_ULL(2);
> > rvu_write64(rvu, blkaddr, NIX_AF_CFG, cfg); }
> > +
> > +void rvu_apr_block_cn10k_init(struct rvu *rvu) {
> > + u64 reg;
> > +
> > + reg = rvu_read64(rvu, BLKADDR_APR, APR_AF_LMT_CFG);
> > + reg |= 0xFULL << 35;
>
> Hi Pavan and Geetha,
>
> I think it would be best to avoid the magic value 35 here.
>
> Best would probably be to use GENMASK_ULL and FIELD_PREP.
> Else defining something similar to APR_LMT_MAP_ENT_SCH_ENA_SHIFT.
>
> It might also be nice to avoid the magic value 0xFULL using a #define.
Ack. Will submit next version.
>
> > + rvu_write64(rvu, BLKADDR_APR, APR_AF_LMT_CFG, reg); }
> > --
> > 2.25.1
> >
@@ -935,6 +935,9 @@ static int rvu_setup_hw_resources(struct rvu *rvu)
hw->total_vfs = (cfg >> 20) & 0xFFF;
hw->max_vfs_per_pf = (cfg >> 40) & 0xFF;
+ if (!is_rvu_otx2(rvu))
+ rvu_apr_block_cn10k_init(rvu);
+
/* Init NPA LF's bitmap */
block = &hw->block[BLKADDR_NPA];
if (!block->implemented)
@@ -940,6 +940,7 @@ void rvu_nix_block_cn10k_init(struct rvu *rvu, struct nix_hw *nix_hw);
/* CN10K RVU - LMT*/
void rvu_reset_lmt_map_tbl(struct rvu *rvu, u16 pcifunc);
+void rvu_apr_block_cn10k_init(struct rvu *rvu);
#ifdef CONFIG_DEBUG_FS
void rvu_dbg_init(struct rvu *rvu);
@@ -559,3 +559,12 @@ void rvu_nix_block_cn10k_init(struct rvu *rvu, struct nix_hw *nix_hw)
cfg |= BIT_ULL(1) | BIT_ULL(2);
rvu_write64(rvu, blkaddr, NIX_AF_CFG, cfg);
}
+
+void rvu_apr_block_cn10k_init(struct rvu *rvu)
+{
+ u64 reg;
+
+ reg = rvu_read64(rvu, BLKADDR_APR, APR_AF_LMT_CFG);
+ reg |= 0xFULL << 35;
+ rvu_write64(rvu, BLKADDR_APR, APR_AF_LMT_CFG, reg);
+}