| Message ID | 20220909062813.1511575-1-bjorn@mork.no |
|---|---|
| State | Accepted |
| Delegated to: | Sander Vanheule |
| Headers | show |
| Series | realtek: fix RTL839x receive tag decoding | expand |
On Fri, 2022-09-09 at 08:28 +0200, Bjørn Mork wrote: > The previous fixup was incomplete, and the offsets for the > queue and crc_error cpu_tag bitfields were still wrong on > RTL839x. > > Fixes: 545c6113c93b ("realtek: fix RTL838x receive tag decoding") > Suggested-by: Jan Hoffmann <jan@3e8.eu> > Signed-off-by: Bjørn Mork <bjorn@mork.no> > --- > Jan pointed put that I missed two fields on the RTL839x. Sloppy. Fix > those as well. > > This time even without complaints from checkpatch :-) > > But build tested only. I don't have any 839x and I don't know how > to verify these features anyway. +CC Andreas, Hiroshi, Luiz, Markus as rtl839x users. For dsa_tag::queue, it looks like the (received) value is only ever used in log statements. dsa_tag::crc_error is used to set sk_buff::ip_summed to either CHECKSUM_NONE or CHECKSUM_UNNECESSARY. AFAICT invalid crc_error values could lead to invalid packets being marked as valid (unlikely on a good connection), or redundant checksum calculations. I think crc_error is the only one that might be causing issues, but probably not as bad as the issue that was fixed with commit 545c6113c93b ("realtek: fix RTL838x receive tag decoding"). This patch is backed by the documentation, so if nobody is set up to check if it fixes anything, it should be merged anyway. Best, Sander
On Fri, 2022-09-09 at 08:28 +0200, Bjørn Mork wrote: > The previous fixup was incomplete, and the offsets for the > queue and crc_error cpu_tag bitfields were still wrong on > RTL839x. > > Fixes: 545c6113c93b ("realtek: fix RTL838x receive tag decoding") > Suggested-by: Jan Hoffmann <jan@3e8.eu> > Signed-off-by: Bjørn Mork <bjorn@mork.no> > --- I've gone ahead and just merged the patch. Up to the next fix! Best, Sander
> > The previous fixup was incomplete, and the offsets for the > > queue and crc_error cpu_tag bitfields were still wrong on > > RTL839x. > > > > Fixes: 545c6113c93b ("realtek: fix RTL838x receive tag decoding") > > Suggested-by: Jan Hoffmann <jan@3e8.eu> > > Signed-off-by: Bjørn Mork <bjorn@mork.no> > > --- > > Jan pointed put that I missed two fields on the RTL839x. Sloppy. Fix > > those as well. > > > > This time even without complaints from checkpatch :-) > > > > But build tested only. I don't have any 839x and I don't know how > > to verify these features anyway. > > +CC Andreas, Hiroshi, Luiz, Markus as rtl839x users. I was hoping this would fix the missing vlan tag for traffic that goes through the switch. I just tested https://github.com/openwrt/openwrt/pull/10643 rebased on current master and the issue is the same: tagged vlan to tagged vlan forwarded by the switch is sent untagged while tagged traffic that goes to/from the CPU works as expected. It does not seem to be related to the received tag. It looks like the tag is added correctly by the CPU but the switch is not configured correctly to add a tag for traffic to a specific vlan. I would bet it is the internal switch vlan setup, not the linux side. BTW, it might be time to merge that PR as, AFAIK, this vlan issue affects the SoC, not that specific device. Regards, Luiz
diff --git a/target/linux/realtek/files-5.10/drivers/net/ethernet/rtl838x_eth.c b/target/linux/realtek/files-5.10/drivers/net/ethernet/rtl838x_eth.c index d9ade6552698..e96c5a7216f8 100644 --- a/target/linux/realtek/files-5.10/drivers/net/ethernet/rtl838x_eth.c +++ b/target/linux/realtek/files-5.10/drivers/net/ethernet/rtl838x_eth.c @@ -282,9 +282,9 @@ bool rtl839x_decode_tag(struct p_hdr *h, struct dsa_tag *t) { /* cpu_tag[0] is reserved. Fields are off-by-one */ t->reason = h->cpu_tag[5] & 0x1f; - t->queue = (h->cpu_tag[3] & 0xe000) >> 13; + t->queue = (h->cpu_tag[4] & 0xe000) >> 13; t->port = h->cpu_tag[1] & 0x3f; - t->crc_error = h->cpu_tag[3] & BIT(2); + t->crc_error = h->cpu_tag[4] & BIT(6); pr_debug("Reason: %d\n", t->reason); if ((t->reason >= 7 && t->reason <= 13) || // NIC_RX_REASON_RMA
The previous fixup was incomplete, and the offsets for the queue and crc_error cpu_tag bitfields were still wrong on RTL839x. Fixes: 545c6113c93b ("realtek: fix RTL838x receive tag decoding") Suggested-by: Jan Hoffmann <jan@3e8.eu> Signed-off-by: Bjørn Mork <bjorn@mork.no> --- Jan pointed put that I missed two fields on the RTL839x. Sloppy. Fix those as well. This time even without complaints from checkpatch :-) But build tested only. I don't have any 839x and I don't know how to verify these features anyway. Bjørn .../realtek/files-5.10/drivers/net/ethernet/rtl838x_eth.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)