* [PATCH net-next] net: stmmac: xgmac: Report the VLAN tag type from the descriptor
@ 2026-11-18 16:21 Maxime Chevallier
2026-11-18 20:09 ` Nicolai Buchwitz
2026-11-19 17:09 ` netdev-bot+sashiko
0 siblings, 2 replies; 4+ messages in thread
From: Maxime Chevallier @ 2026-11-18 16:21 UTC (permalink / raw)
To: Andrew Lunn, Jakub Kicinski, davem, Eric Dumazet, Paolo Abeni,
Simon Horman, Maxime Coquelin, Alexandre Torgue, Russell King,
Lorenzo Bianconi, Nicolai Buchwitz, Ovidiu Panait, Joseph Steel,
Muhammad Nazim Amirul Nazle Asmade
Cc: Maxime Chevallier, thomas.petazzoni, Alexis Lothoré,
netdev, linux-kernel, linux-arm-kernel
Stmmac's double VLAN handling hasn't been working great for a while,
it's been recently reworked for dwmac4 by Ovidiu Panait. In that work,
it was found that the tag stripping hardcodes the 8021Q tag type for
XGMAC while stripping the tag, while the IP actually reports the outer
tag type in the descriptor.
The tag type can be one of 6 different values (note inner/outer is
reversed in the macro names):
- XGMAC_ET_LT_VLAN_STAG single S-tag
- XGMAC_ET_LT_VLAN_CTAG single C-tag
- XGMAC_ET_LT_DVLAN_CTAG_CTAG outer C-tag, inner C-tag
- XGMAC_ET_LT_DVLAN_STAG_STAG outer S-tag, inner S-tag
- XGMAC_ET_LT_DVLAN_CTAG_STAG outer S-tag, inner C-tag
- XGMAC_ET_LT_DVLAN_STAG_CTAG outer C-tag, inner S-tag
Report that tag type through a dedicated desc ops and report it to the
stack.
We keep the 8021Q fallback when we don't have the ops, as we don't
set NETIF_F_HW_VLAN_STAG_FILTER on the cores that don't support that.
This has been tested on AgileX5 with the vlan.py selftest.
Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
---
.../net/ethernet/stmicro/stmmac/dwxgmac2_descs.c | 13 +++++++++++++
drivers/net/ethernet/stmicro/stmmac/hwif.h | 2 ++
drivers/net/ethernet/stmicro/stmmac/stmmac_vlan.c | 6 +++++-
3 files changed, 20 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
index 6719ac6e395b..71ce2bc9b5e1 100644
--- a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
+++ b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
@@ -76,6 +76,18 @@ static bool dwxgmac2_wrback_get_rx_vlan_valid(struct dma_desc *p)
et_lt <= XGMAC_ET_LT_DVLAN_STAG_CTAG;
}
+static u16 dwxgmac2_wrback_get_rx_vlan_proto(struct dma_desc *p)
+{
+ switch (FIELD_GET(XGMAC_RDES3_ET_LT, le32_to_cpu(p->des3))) {
+ case XGMAC_ET_LT_VLAN_STAG:
+ case XGMAC_ET_LT_DVLAN_STAG_STAG:
+ case XGMAC_ET_LT_DVLAN_CTAG_STAG:
+ return ETH_P_8021AD;
+ default:
+ return ETH_P_8021Q;
+ }
+}
+
static int dwxgmac2_get_rx_frame_len(struct dma_desc *p, int rx_coe)
{
return (le32_to_cpu(p->des3) & XGMAC_RDES3_PL);
@@ -352,6 +364,7 @@ const struct stmmac_desc_ops dwxgmac210_desc_ops = {
.set_rx_owner = dwxgmac2_set_rx_owner,
.get_rx_vlan_tci = dwxgmac2_wrback_get_rx_vlan_tci,
.get_rx_vlan_valid = dwxgmac2_wrback_get_rx_vlan_valid,
+ .get_rx_vlan_proto = dwxgmac2_wrback_get_rx_vlan_proto,
.get_rx_frame_len = dwxgmac2_get_rx_frame_len,
.enable_tx_timestamp = dwxgmac2_enable_tx_timestamp,
.get_tx_timestamp_status = dwxgmac2_get_tx_timestamp_status,
diff --git a/drivers/net/ethernet/stmicro/stmmac/hwif.h b/drivers/net/ethernet/stmicro/stmmac/hwif.h
index 8ae95e91a36a..047bda5cbd92 100644
--- a/drivers/net/ethernet/stmicro/stmmac/hwif.h
+++ b/drivers/net/ethernet/stmicro/stmmac/hwif.h
@@ -60,6 +60,8 @@ struct stmmac_desc_ops {
u16 (*get_rx_vlan_tci)(struct dma_desc *p);
/* Get the valid status of descriptor */
bool (*get_rx_vlan_valid)(struct dma_desc *p);
+ /* Get the vlan tag proto from the descriptor */
+ u16 (*get_rx_vlan_proto)(struct dma_desc *p);
/* Return the transmit status looking at the TDES1 */
int (*tx_status)(struct stmmac_extra_stats *x,
struct dma_desc *p, void __iomem *ioaddr);
diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_vlan.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_vlan.c
index fb9aad748cb3..bf23d82e9f38 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_vlan.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_vlan.c
@@ -209,8 +209,12 @@ static void vlan_rx_hw(struct mac_device_info *hw,
{
if (hw->desc->get_rx_vlan_valid(rx_desc)) {
u16 vid = hw->desc->get_rx_vlan_tci(rx_desc);
+ u16 proto = ETH_P_8021Q;
- __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
+ if (hw->desc->get_rx_vlan_proto)
+ proto = hw->desc->get_rx_vlan_proto(rx_desc);
+
+ __vlan_hwaccel_put_tag(skb, htons(proto), vid);
}
}
--
2.56.1
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH net-next] net: stmmac: xgmac: Report the VLAN tag type from the descriptor
2026-11-18 16:21 [PATCH net-next] net: stmmac: xgmac: Report the VLAN tag type from the descriptor Maxime Chevallier
@ 2026-11-18 20:09 ` Nicolai Buchwitz
2026-11-19 7:36 ` Maxime Chevallier
2026-11-19 17:09 ` netdev-bot+sashiko
1 sibling, 1 reply; 4+ messages in thread
From: Nicolai Buchwitz @ 2026-11-18 20:09 UTC (permalink / raw)
To: Maxime Chevallier
Cc: Andrew Lunn, Jakub Kicinski, davem, Eric Dumazet, Paolo Abeni,
Simon Horman, Maxime Coquelin, Alexandre Torgue, Russell King,
Lorenzo Bianconi, Ovidiu Panait, Joseph Steel,
Muhammad Nazim Amirul Nazle Asmade, thomas.petazzoni,
Alexis Lothoré,
netdev, linux-kernel, linux-arm-kernel
Hi Maxime
On 18.11.2026 17:21, Maxime Chevallier wrote:
> Stmmac's double VLAN handling hasn't been working great for a while,
> it's been recently reworked for dwmac4 by Ovidiu Panait. In that work,
> it was found that the tag stripping hardcodes the 8021Q tag type for
> XGMAC while stripping the tag, while the IP actually reports the outer
> tag type in the descriptor.
>
> The tag type can be one of 6 different values (note inner/outer is
> reversed in the macro names):
>
> - XGMAC_ET_LT_VLAN_STAG single S-tag
> - XGMAC_ET_LT_VLAN_CTAG single C-tag
> - XGMAC_ET_LT_DVLAN_CTAG_CTAG outer C-tag, inner C-tag
> - XGMAC_ET_LT_DVLAN_STAG_STAG outer S-tag, inner S-tag
> - XGMAC_ET_LT_DVLAN_CTAG_STAG outer S-tag, inner C-tag
> - XGMAC_ET_LT_DVLAN_STAG_CTAG outer C-tag, inner S-tag
>
> Report that tag type through a dedicated desc ops and report it to the
> stack.
>
> We keep the 8021Q fallback when we don't have the ops, as we don't
> set NETIF_F_HW_VLAN_STAG_FILTER on the cores that don't support that.
>
> This has been tested on AgileX5 with the vlan.py selftest.
>
> Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Isn't this a fix too? I mean xgmac advertises STAG_RX but still
reports S-tags as 802.1Q. Then again nobody seems to have hit it
so far, so net-next is probably fine.
Anyway:
Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de>
Thanks,
Nicolai
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH net-next] net: stmmac: xgmac: Report the VLAN tag type from the descriptor
2026-11-18 20:09 ` Nicolai Buchwitz
@ 2026-11-19 7:36 ` Maxime Chevallier
0 siblings, 0 replies; 4+ messages in thread
From: Maxime Chevallier @ 2026-11-19 7:36 UTC (permalink / raw)
To: Nicolai Buchwitz
Cc: Andrew Lunn, Jakub Kicinski, davem, Eric Dumazet, Paolo Abeni,
Simon Horman, Maxime Coquelin, Alexandre Torgue, Russell King,
Lorenzo Bianconi, Ovidiu Panait, Joseph Steel,
Muhammad Nazim Amirul Nazle Asmade, thomas.petazzoni,
Alexis Lothoré,
netdev, linux-kernel, linux-arm-kernel
HI,
On 11/18/26 21:09, Nicolai Buchwitz wrote:
> Hi Maxime
>
> On 18.11.2026 17:21, Maxime Chevallier wrote:
>> Stmmac's double VLAN handling hasn't been working great for a while,
>> it's been recently reworked for dwmac4 by Ovidiu Panait. In that work,
>> it was found that the tag stripping hardcodes the 8021Q tag type for
>> XGMAC while stripping the tag, while the IP actually reports the outer
>> tag type in the descriptor.
>>
>> The tag type can be one of 6 different values (note inner/outer is
>> reversed in the macro names):
>>
>> - XGMAC_ET_LT_VLAN_STAG single S-tag
>> - XGMAC_ET_LT_VLAN_CTAG single C-tag
>> - XGMAC_ET_LT_DVLAN_CTAG_CTAG outer C-tag, inner C-tag
>> - XGMAC_ET_LT_DVLAN_STAG_STAG outer S-tag, inner S-tag
>> - XGMAC_ET_LT_DVLAN_CTAG_STAG outer S-tag, inner C-tag
>> - XGMAC_ET_LT_DVLAN_STAG_CTAG outer C-tag, inner S-tag
>>
>> Report that tag type through a dedicated desc ops and report it to the
>> stack.
>>
>> We keep the 8021Q fallback when we don't have the ops, as we don't
>> set NETIF_F_HW_VLAN_STAG_FILTER on the cores that don't support that.
>>
>> This has been tested on AgileX5 with the vlan.py selftest.
>>
>> Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
>
> Isn't this a fix too? I mean xgmac advertises STAG_RX but still
> reports S-tags as 802.1Q. Then again nobody seems to have hit it
> so far, so net-next is probably fine.
Yeah not only was this discovered while working on something else, it
seems no-one noticed and it's always been broken... I wanted to avoid
spamming net too much, both options are fine for me TBH,
Thanks for the review :)
Maxime
>
> Anyway:
>
> Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de>
>
> Thanks,
> Nicolai
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH net-next] net: stmmac: xgmac: Report the VLAN tag type from the descriptor
2026-11-18 16:21 [PATCH net-next] net: stmmac: xgmac: Report the VLAN tag type from the descriptor Maxime Chevallier
2026-11-18 20:09 ` Nicolai Buchwitz
@ 2026-11-19 17:09 ` netdev-bot+sashiko
1 sibling, 0 replies; 4+ messages in thread
From: netdev-bot+sashiko @ 2026-11-19 17:09 UTC (permalink / raw)
To: maxime.chevallier
Cc: andrew+netdev, kuba, davem, edumazet, pabeni, horms,
mcoquelin.stm32, alexandre.torgue, linux, lorenzo.bianconi, nb,
ovidiu.panait.rb, recv.jo, muhammad.nazim.amirul.nazle.asmade,
thomas.petazzoni, alexis.lothore, netdev, linux-kernel,
linux-arm-kernel
Thank you for your contribution! Sashiko AI review found 2 potential
issue(s) to consider.
Critical: 0 · High: 0 · Medium: 1 · Low: 1
If you believe none of the issues are legitimate, please reply to each
issue of Medium severity or higher and explain why the AI is wrong.
If any of them are legitimate and you plan to respin, please let patchwork
know by including "pw-bot: cr" as a separate line at the end of your reply
(one such reply per series is enough).
- [Medium] This patch is a bug fix but carries no Fixes: tag, and the
subject reads like a new feature ("Report the VLAN tag type from the…
- [Low] The ET/LT macro names in dwxgmac2.h disagree with how the new
dwxgmac2_wrback_get_rx_vlan_proto() reads them.
--
net: stmmac: xgmac: Report the VLAN tag type from the descriptor
This adds a get_rx_vlan_proto() desc op for XGMAC. The op decodes the
RDES3 ET_LT field and reports either ETH_P_8021AD or ETH_P_8021Q as the
stripped outer tag type. vlan_rx_hw() now uses that op when it exists and
keeps ETH_P_8021Q as the fallback.
> diff --git a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
> index 6719ac6e395bc..71ce2bc9b5e11 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
> @@ -76,6 +76,18 @@ static bool dwxgmac2_wrback_get_rx_vlan_valid(struct dma_desc *p)
> et_lt <= XGMAC_ET_LT_DVLAN_STAG_CTAG;
> }
>
> +static u16 dwxgmac2_wrback_get_rx_vlan_proto(struct dma_desc *p)
> +{
> + switch (FIELD_GET(XGMAC_RDES3_ET_LT, le32_to_cpu(p->des3))) {
> + case XGMAC_ET_LT_VLAN_STAG:
> + case XGMAC_ET_LT_DVLAN_STAG_STAG:
> + case XGMAC_ET_LT_DVLAN_CTAG_STAG:
> + return ETH_P_8021AD;
> + default:
> + return ETH_P_8021Q;
> + }
> +}
[Severity: Low]
This isn't a bug, but the macro names in dwxgmac2.h don't match how they
are used here. The only comment on them is:
drivers/net/ethernet/stmicro/stmmac/dwxgmac2.h:
/* Error Type or L2 Type(ET/LT) Field Number */
#define XGMAC_ET_LT_VLAN_STAG 8
#define XGMAC_ET_LT_VLAN_CTAG 9
#define XGMAC_ET_LT_DVLAN_CTAG_CTAG 10
#define XGMAC_ET_LT_DVLAN_STAG_STAG 11
#define XGMAC_ET_LT_DVLAN_CTAG_STAG 12
#define XGMAC_ET_LT_DVLAN_STAG_CTAG 13
Read outer-first, XGMAC_ET_LT_DVLAN_CTAG_STAG means an outer C-tag, yet
dwxgmac2_wrback_get_rx_vlan_proto() returns ETH_P_8021AD for it. Likewise
XGMAC_ET_LT_DVLAN_STAG_CTAG falls through to ETH_P_8021Q.
The commit message says "note inner/outer is reversed in the macro
names", but nothing in the tree records that. Could the macros be renamed
(for example XGMAC_ET_LT_DVLAN_OUTER_STAG_INNER_CTAG), or could a comment
in dwxgmac2.h give the tag order? Otherwise a future user of these
constants could easily invert the mapping.
If the names do follow the databook's outer-first order, the mapping for
values 12 and 13 would be inverted. That can't be checked without the
databook, so it would help to confirm which reading is correct.
[ ... ]
> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_vlan.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_vlan.c
> index fb9aad748cb34..bf23d82e9f383 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_vlan.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_vlan.c
> @@ -209,8 +209,12 @@ static void vlan_rx_hw(struct mac_device_info *hw,
> {
> if (hw->desc->get_rx_vlan_valid(rx_desc)) {
> u16 vid = hw->desc->get_rx_vlan_tci(rx_desc);
> + u16 proto = ETH_P_8021Q;
>
> - __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
> + if (hw->desc->get_rx_vlan_proto)
> + proto = hw->desc->get_rx_vlan_proto(rx_desc);
> +
> + __vlan_hwaccel_put_tag(skb, htons(proto), vid);
[Severity: Medium]
Should this patch carry a Fixes: tag? It looks like it fixes a
user-visible regression on XGMAC.
On XGMAC, __stmmac_dvr_probe() advertises NETIF_F_HW_VLAN_STAG_RX and
enables MAC-level stripping by default:
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c:__stmmac_dvr_probe() {
...
if (priv->plat->core_type == DWMAC_CORE_XGMAC)
ndev->features |= NETIF_F_HW_VLAN_STAG_RX;
if (dwmac_is_xmac(priv->plat->core_type)) {
ndev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
priv->hw->hw_vlan_en = true;
}
...
}
stmmac_fix_features() also keeps STAG_RX tied to CTAG_RX on XGMAC.
Because of this, every frame with a stripped 802.1ad outer tag went
through the old vlan_rx_hw() and reached the stack labelled as
ETH_P_8021Q. 802.1ad VLAN uppers received no traffic and QinQ between
hosts failed.
The sibling commit da17990421d7 ("net: stmmac: Disable S-Tag processing
on dwmac4") describes this exact failure and carries a Fixes: tag. Its
message says the dwxgmac2 side can be fixed using the RDES3 ET_LT field,
which is what this patch does.
Without a Fixes: tag, stable trees that pick up the dwmac4 half would
still mislabel S-tags on XGMAC. Would this be a suitable tag?
Fixes: 534df0c1724b ("net: stmmac: dwxgmac2: Add support for HW-accelerated VLAN stripping")
That commit started XGMAC hardware stripping with the hardcoded
ETH_P_8021Q proto.
The subject also reads like a new feature. The commit message only says
double VLAN handling "hasn't been working great for a while". Could it
state the actual failure, that S-tagged and QinQ frames are passed up as
802.1Q?
> }
> }
--
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261118162136.2782405-1-maxime.chevallier%40bootlin.com
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2026-11-19 17:09 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-18 16:21 [PATCH net-next] net: stmmac: xgmac: Report the VLAN tag type from the descriptor Maxime Chevallier
2026-11-18 20:09 ` Nicolai Buchwitz
2026-11-19 7:36 ` Maxime Chevallier
2026-11-19 17:09 ` netdev-bot+sashiko
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®