Re: [resend PATCH 2/2] dim: pass dim_sample to net_dim() by reference
From: Caleb Sander <hidden>
Date: 2024-10-31 17:20:07
Also in:
intel-wired-lan, linux-doc, linux-mediatek, linux-rdma, lkml, netdev, virtualization
On Thu, Oct 31, 2024 at 5:49 AM Louis Peens [off-list ref] wrote:
quoted hunk ↗ jump to hunk
On Wed, Oct 30, 2024 at 06:23:26PM -0600, Caleb Sander Mateos wrote:quoted
net_dim() is currently passed a struct dim_sample argument by value. struct dim_sample is 24 bytes. Since this is greater 16 bytes, x86-64 passes it on the stack. All callers have already initialized dim_sample on the stack, so passing it by value requires pushing a duplicated copy to the stack. Either witing to the stack and immediately reading it, or perhaps dereferencing addresses relative to the stack pointer in a chain of push instructions, seems to perform quite poorly. In a heavy TCP workload, mlx5e_handle_rx_dim() consumes 3% of CPU time, 94% of which is attributed to the first push instruction to copy dim_sample on the stack for the call to net_dim(): // Call ktime_get() 0.26 |4ead2: call 4ead7 <mlx5e_handle_rx_dim+0x47> // Pass the address of struct dim in %rdi |4ead7: lea 0x3d0(%rbx),%rdi // Set dim_sample.pkt_ctr |4eade: mov %r13d,0x8(%rsp) // Set dim_sample.byte_ctr |4eae3: mov %r12d,0xc(%rsp) // Set dim_sample.event_ctr 0.15 |4eae8: mov %bp,0x10(%rsp) // Duplicate dim_sample on the stack 94.16 |4eaed: push 0x10(%rsp) 2.79 |4eaf1: push 0x10(%rsp) 0.07 |4eaf5: push %rax // Call net_dim() 0.21 |4eaf6: call 4eafb <mlx5e_handle_rx_dim+0x6b> To allow the caller to reuse the struct dim_sample already on the stack, pass the struct dim_sample by reference to net_dim(). Signed-off-by: Caleb Sander Mateos <redacted> --- Documentation/networking/net_dim.rst | 2 +- drivers/net/ethernet/amazon/ena/ena_netdev.c | 2 +- drivers/net/ethernet/broadcom/bcmsysport.c | 2 +- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 4 ++-- drivers/net/ethernet/broadcom/genet/bcmgenet.c | 2 +- drivers/net/ethernet/freescale/enetc/enetc.c | 2 +- drivers/net/ethernet/hisilicon/hns3/hns3_enet.c | 4 ++-- drivers/net/ethernet/intel/ice/ice_txrx.c | 4 ++-- drivers/net/ethernet/intel/idpf/idpf_txrx.c | 4 ++-- drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.c | 2 +- drivers/net/ethernet/mediatek/mtk_eth_soc.c | 4 ++-- drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c | 4 ++-- drivers/net/ethernet/netronome/nfp/nfd3/dp.c | 4 ++-- drivers/net/ethernet/netronome/nfp/nfdk/dp.c | 4 ++-- drivers/net/ethernet/pensando/ionic/ionic_txrx.c | 2 +- drivers/net/virtio_net.c | 2 +- drivers/soc/fsl/dpio/dpio-service.c | 2 +- include/linux/dim.h | 2 +- lib/dim/net_dim.c | 10 +++++----- 19 files changed, 31 insertions(+), 31 deletions(-)--- snip --quoted
diff --git a/drivers/net/ethernet/netronome/nfp/nfd3/dp.c b/drivers/net/ethernet/netronome/nfp/nfd3/dp.c index d215efc6cad0..f1c6c47564b1 100644 --- a/drivers/net/ethernet/netronome/nfp/nfd3/dp.c +++ b/drivers/net/ethernet/netronome/nfp/nfd3/dp.c@@ -1177,11 +1177,11 @@ int nfp_nfd3_poll(struct napi_struct *napi, int budget) pkts = r_vec->rx_pkts; bytes = r_vec->rx_bytes; } while (u64_stats_fetch_retry(&r_vec->rx_sync, start)); dim_update_sample(r_vec->event_ctr, pkts, bytes, &dim_sample); - net_dim(&r_vec->rx_dim, dim_sample); + net_dim(&r_vec->rx_dim, &dim_sample); } if (r_vec->nfp_net->tx_coalesce_adapt_on && r_vec->tx_ring) { struct dim_sample dim_sample = {}; unsigned int start;@@ -1192,11 +1192,11 @@ int nfp_nfd3_poll(struct napi_struct *napi, int budget) pkts = r_vec->tx_pkts; bytes = r_vec->tx_bytes; } while (u64_stats_fetch_retry(&r_vec->tx_sync, start)); dim_update_sample(r_vec->event_ctr, pkts, bytes, &dim_sample); - net_dim(&r_vec->tx_dim, dim_sample); + net_dim(&r_vec->tx_dim, &dim_sample); } return pkts_polled; }diff --git a/drivers/net/ethernet/netronome/nfp/nfdk/dp.c b/drivers/net/ethernet/netronome/nfp/nfdk/dp.c index dae5af7d1845..ebeb6ab4465c 100644 --- a/drivers/net/ethernet/netronome/nfp/nfdk/dp.c +++ b/drivers/net/ethernet/netronome/nfp/nfdk/dp.c@@ -1287,11 +1287,11 @@ int nfp_nfdk_poll(struct napi_struct *napi, int budget) pkts = r_vec->rx_pkts; bytes = r_vec->rx_bytes; } while (u64_stats_fetch_retry(&r_vec->rx_sync, start)); dim_update_sample(r_vec->event_ctr, pkts, bytes, &dim_sample); - net_dim(&r_vec->rx_dim, dim_sample); + net_dim(&r_vec->rx_dim, &dim_sample); } if (r_vec->nfp_net->tx_coalesce_adapt_on && r_vec->tx_ring) { struct dim_sample dim_sample = {}; unsigned int start;@@ -1302,11 +1302,11 @@ int nfp_nfdk_poll(struct napi_struct *napi, int budget) pkts = r_vec->tx_pkts; bytes = r_vec->tx_bytes; } while (u64_stats_fetch_retry(&r_vec->tx_sync, start)); dim_update_sample(r_vec->event_ctr, pkts, bytes, &dim_sample); - net_dim(&r_vec->tx_dim, dim_sample); + net_dim(&r_vec->tx_dim, &dim_sample); } return pkts_polled; }--- snip ---Hi Caleb. Looks like a fair enough update to me in general, but I am not an expert on 'dim'. For the corresponding nfp driver changes feel free to add: Signed-off-by: Louis Peens <redacted>
Hi Louis, Thanks for the review. Did you mean "Reviewed-by"? If there was a change you were suggesting, I missed it. Best, Caleb