| Message ID | 20241204115715.3148412-1-yoong.siang.song@intel.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [bpf-next,1/1] selftests/bpf: Enable Tx hwtstamp in xdp_hw_metadata | expand |
On 12/04, Song Yoong Siang wrote: > Set tx_type to HWTSTAMP_TX_ON to enable hardware timestamping for all > outgoing packets. > > Besides, set XDP_UMEM_TX_METADATA_LEN flag to reserve tx_metadata_len bytes > of per-chunk metadata. XDP_UMEM_TX_METADATA_LEN is missing after d5e726d9143c ("xsk: Require XDP_UMEM_TX_METADATA_LEN to actuate tx_metadata_len"), so that make sense. Maybe add a fixes tag? And I don't see mlx5 looking at HWTSTAMP_TX anywhere in the drivers, so I'm assuming that's why I didn't need HWTSTAMP_TX_ON during my tests.. Which device are you testing against? I do see some hwts_tx_en checks in the stfmmac at least... Can you add these details to the commit message and respin? With the above addressed: Acked-by: Stanislav Fomichev <sdf@fomichev.me>
On Wednesday, December 4, 2024 11:49 PM, Stanislav Fomichev <stfomichev@gmail.com> wrote: >On 12/04, Song Yoong Siang wrote: >> Set tx_type to HWTSTAMP_TX_ON to enable hardware timestamping for all >> outgoing packets. >> >> Besides, set XDP_UMEM_TX_METADATA_LEN flag to reserve tx_metadata_len bytes >> of per-chunk metadata. > >XDP_UMEM_TX_METADATA_LEN is missing after d5e726d9143c ("xsk: Require >XDP_UMEM_TX_METADATA_LEN to actuate tx_metadata_len"), so that make >sense. Maybe add a fixes tag? > Sure. I will add the fixes tag and submit with "PATCH bpf" prefix in next version. >And I don't see mlx5 looking at HWTSTAMP_TX anywhere in the drivers, >so I'm assuming that's why I didn't need HWTSTAMP_TX_ON during my tests.. >Which device are you testing against? I do see some hwts_tx_en >checks in the stfmmac at least... Can you add these details to the >commit message and respin? > I am testing on stmmac and igc drivers. You are right, stmmac needs it for hwts_tx_en check. Besides, igc needs it to set IGC_RING_FLAG_TX_HWTSTAMP flag. Without this patch, user will need to manually enable tx hwts using command: sudo hwstamp_ctl -i eth0 -t 1 -r 1 after start xdp_hw_metadata. Therefore, adding HWTSTAMP_TX_ON is not a bug fix solution. I will separate this as another new patch to "PATCH bpf-next" and provide detail in commit message. Btw, is mlx5 driver always enable Tx HWTS? >With the above addressed: >Acked-by: Stanislav Fomichev <sdf@fomichev.me> Thanks & Regards Siang
On 12/05, Song, Yoong Siang wrote: > On Wednesday, December 4, 2024 11:49 PM, Stanislav Fomichev <stfomichev@gmail.com> wrote: > >On 12/04, Song Yoong Siang wrote: > >> Set tx_type to HWTSTAMP_TX_ON to enable hardware timestamping for all > >> outgoing packets. > >> > >> Besides, set XDP_UMEM_TX_METADATA_LEN flag to reserve tx_metadata_len bytes > >> of per-chunk metadata. > > > >XDP_UMEM_TX_METADATA_LEN is missing after d5e726d9143c ("xsk: Require > >XDP_UMEM_TX_METADATA_LEN to actuate tx_metadata_len"), so that make > >sense. Maybe add a fixes tag? > > > > Sure. I will add the fixes tag and submit with "PATCH bpf" prefix > in next version. > > >And I don't see mlx5 looking at HWTSTAMP_TX anywhere in the drivers, > >so I'm assuming that's why I didn't need HWTSTAMP_TX_ON during my tests.. > >Which device are you testing against? I do see some hwts_tx_en > >checks in the stfmmac at least... Can you add these details to the > >commit message and respin? > > > > I am testing on stmmac and igc drivers. > You are right, stmmac needs it for hwts_tx_en check. > Besides, igc needs it to set IGC_RING_FLAG_TX_HWTSTAMP flag. > > Without this patch, user will need to manually enable tx hwts using > command: sudo hwstamp_ctl -i eth0 -t 1 -r 1 > after start xdp_hw_metadata. > > Therefore, adding HWTSTAMP_TX_ON is not a bug fix solution. > I will separate this as another new patch to "PATCH bpf-next" > and provide detail in commit message. > > Btw, is mlx5 driver always enable Tx HWTS? I don't remember doing anything special to enable it. And looking at the code I also don't see any conditionals on HWTSTAMP_TX_ON.
diff --git a/tools/testing/selftests/bpf/xdp_hw_metadata.c b/tools/testing/selftests/bpf/xdp_hw_metadata.c index 06266aad2f99..6f7b15d6c6ed 100644 --- a/tools/testing/selftests/bpf/xdp_hw_metadata.c +++ b/tools/testing/selftests/bpf/xdp_hw_metadata.c @@ -79,7 +79,7 @@ static int open_xsk(int ifindex, struct xsk *xsk, __u32 queue_id) .fill_size = XSK_RING_PROD__DEFAULT_NUM_DESCS, .comp_size = XSK_RING_CONS__DEFAULT_NUM_DESCS, .frame_size = XSK_UMEM__DEFAULT_FRAME_SIZE, - .flags = XSK_UMEM__DEFAULT_FLAGS, + .flags = XDP_UMEM_TX_METADATA_LEN, .tx_metadata_len = sizeof(struct xsk_tx_metadata), }; __u32 idx = 0; @@ -551,6 +551,7 @@ static void hwtstamp_enable(const char *ifname) { struct hwtstamp_config cfg = { .rx_filter = HWTSTAMP_FILTER_ALL, + .tx_type = HWTSTAMP_TX_ON, }; hwtstamp_ioctl(SIOCGHWTSTAMP, ifname, &saved_hwtstamp_cfg);
Set tx_type to HWTSTAMP_TX_ON to enable hardware timestamping for all outgoing packets. Besides, set XDP_UMEM_TX_METADATA_LEN flag to reserve tx_metadata_len bytes of per-chunk metadata. Signed-off-by: Song Yoong Siang <yoong.siang.song@intel.com> --- tools/testing/selftests/bpf/xdp_hw_metadata.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)