[PATCH] net: atlantic: AQC113: Initialize struct hw_atl2_priv fully

[PATCH] net: atlantic: AQC113: Initialize struct hw_atl2_priv fully

[Janne Grunau]

Leaving parts of struct hw_atl2_priv uninitialized breaks data
transmission on AQC113 NICs integrated in Apple silicon desktop Macs.
Commit 6c1ca112a6e4 ("net: atlantic: fix AQC113 HW init: ART, L2 filter
slot, MAC address") replaced a memset of the whole struct with memsets
of a few fields and initialisation of l3l4_filters[].l[34]_index to -1.
This leaves other fields uninitialized.
Restoring the full struct memset while keeping the initialization of
the l3/l4 filter indices fixes the regression on Apple silicon systems.

Fixes: 6c1ca112a6e4 ("net: atlantic: fix AQC113 HW init: ART, L2 filter slot, MAC address")
Cc: stable@vger.kernel.org
Signed-off-by: Janne Grunau <j@jannau.net>
---
Hej,

commit 6c1ca112a6e4 ("net: atlantic: fix AQC113 HW init: ART, L2 filter
slot, MAC address") broke data transmission on Apple silicon systems
with integrated AQC113. The cause appears to leaving parts of struct
hw_atl2_priv uninitialized. Since the most sensible fix for the issue is
restoring the full struct memset I did not investigate which value in
which field causes the issue.
The fix restores ethernet data transmission on an Apple M1 Max Mac
Studio.
---
 drivers/net/ethernet/aquantia/atlantic/hw_atl2/hw_atl2.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/aquantia/atlantic/hw_atl2/hw_atl2.c b/drivers/net/ethernet/aquantia/atlantic/hw_atl2/hw_atl2.c
index 25dd7150aaaf..5d6e81fa3bae 100644
--- a/drivers/net/ethernet/aquantia/atlantic/hw_atl2/hw_atl2.c
+++ b/drivers/net/ethernet/aquantia/atlantic/hw_atl2/hw_atl2.c
@@ -145,11 +145,7 @@ static int hw_atl2_hw_reset(struct aq_hw_s *self)
 	if (err)
 		return err;
 
-	memset(&priv->last_stats, 0, sizeof(priv->last_stats));
-	memset(priv->l3_v4_filters, 0, sizeof(priv->l3_v4_filters));
-	memset(priv->l3_v6_filters, 0, sizeof(priv->l3_v6_filters));
-	memset(priv->l4_filters, 0, sizeof(priv->l4_filters));
-	memset(priv->etype_policy, 0, sizeof(priv->etype_policy));
+	memset(priv, 0, sizeof(*priv));
 	for (i = 0; i < HW_ATL2_RPF_L3L4_FILTERS; i++) {
 		priv->l3l4_filters[i].l3_index = -1;
 		priv->l3l4_filters[i].l4_index = -1;

---
base-commit: c03c1a14507dfaf8b856c2608f9ac216021fc222
change-id: 20261024-net-atlantic-aq113-apple-regression-e40f40bd57de

Best regards,
-- 
Janne Grunau <j@jannau.net>

Re: [PATCH] net: atlantic: AQC113: Initialize struct hw_atl2_priv fully

[netdev-bot+sashiko]

Thank you for your contribution! Sashiko AI review found 1 potential
issue(s) to consider.

Critical: 0 · High: 1 · Medium: 0 · Low: 0

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).

- [High] The patch puts back `memset(priv, 0, sizeof(*priv))` in
  hw_atl2_hw_reset() (hw_atl2.c:148).

--

net: atlantic: AQC113: Initialize struct hw_atl2_priv fully

This patch puts back the full memset of struct hw_atl2_priv in
hw_atl2_hw_reset() and still sets l3l4_filters[].l3_index and l4_index
to -1. Its aim is to fix data transmission on AQC113 NICs in Apple
silicon desktop Macs, which broke after commit 6c1ca112a6e4.

> diff --git a/drivers/net/ethernet/aquantia/atlantic/hw_atl2/hw_atl2.c b/drivers/net/ethernet/aquantia/atlantic/hw_atl2/hw_atl2.c
> index 25dd7150aaaf2..5d6e81fa3baee 100644
> --- a/drivers/net/ethernet/aquantia/atlantic/hw_atl2/hw_atl2.c
> +++ b/drivers/net/ethernet/aquantia/atlantic/hw_atl2/hw_atl2.c
> @@ -145,11 +145,7 @@ static int hw_atl2_hw_reset(struct aq_hw_s *self)
>  	if (err)
>  		return err;
>  
> -	memset(&priv->last_stats, 0, sizeof(priv->last_stats));
> -	memset(priv->l3_v4_filters, 0, sizeof(priv->l3_v4_filters));
> -	memset(priv->l3_v6_filters, 0, sizeof(priv->l3_v6_filters));
> -	memset(priv->l4_filters, 0, sizeof(priv->l4_filters));
> -	memset(priv->etype_policy, 0, sizeof(priv->etype_policy));
> +	memset(priv, 0, sizeof(*priv));

[Severity: High]
Doesn't this memset clear the firmware filter caps that
hw_atl2_utils_soft_reset() has just read?

This is the call chain just before the memset:

hw_atl2_hw_reset()
  hw_atl2_utils_soft_reset()
    self->aq_fw_ops->init()   /* aq_a2_fw_init() */
      hw_atl2_utils_get_filter_caps()

hw_atl2_utils_get_filter_caps() is the only place that sets these fields:

	priv->art_base_index = filter_caps.rslv_tbl_base_index * 8;
	priv->art_count = filter_caps.rslv_tbl_count * 8;
	...
	priv->l2_filters_base_index = filter_caps.l2_filters_base_index;
	...

It also sets the etype, vlan, l3_v4, l3_v6 and l4 base indices and
counts.

Nothing reads the caps again after hw_atl2_hw_reset(), so hw_init()
would run with all of them set to zero. This happens on every reset, so
wouldn't it affect every AQC113 and not only the Apple systems?

With the caps zeroed, several paths behave differently.

In hw_atl2_hw_init_new_rx_filters(), art_first_sec and art_last_sec
are both 0. art_mask is then 0 and falls back to 0xFFFF. The driver's
ART records get written starting at index 0, even though the comment
there says:

	/* FW reserves the beginning of ART, thus all driver entries must
	 * start from the offset specified in FW caps.
	 */

hw_atl2_hw_init_rx_path(), hw_atl2_hw_mac_addr_set() and
hw_atl2_hw_multicast_list_set() would start from L2 slot 0 instead of
the slot the firmware assigned.

hw_atl2_hw_fl2_set() calls hw_atl2_filter_tag_get() with
priv->etype_filter_tag_top == 0. Both loops are then empty, so every
ethertype filter install returns -ENOSPC.

hw_atl2_new_fl3l4_find_l3() and hw_atl2_new_fl3l4_find_l4() would see
first == last == 0. Every ethtool ntuple L3/L4 filter install through
hw_atl2_hw_fl3l4_set() would then also fail with -ENOSPC.

There is also a question about the commit message:

> This leaves other fields uninitialized.

Is this accurate? aq_pci_probe() allocates priv with kzalloc(). The
removed memsets already cleared last_stats, the l3/l4 filter arrays and
etype_policy. The only fields they did not touch seem to be the
firmware caps fields, which had just been refreshed, plus
l3l4_filters[].ipv6.

Could the full memset be fixing the Apple systems by turning off the
firmware caps handling that 6c1ca112a6e4 added (the ART section mask
and the firmware L2 slot)? Nothing that was actually uninitialized
seems to be getting initialized.

The patch is tagged for stable. Would it be better to keep the caps?
Possible ways are clearing priv before hw_atl2_utils_soft_reset(),
clearing only up to art_base_index, or saving and restoring the caps.
It might also help to find which firmware value or init step breaks the
Apple hardware.

>  	for (i = 0; i < HW_ATL2_RPF_L3L4_FILTERS; i++) {
>  		priv->l3l4_filters[i].l3_index = -1;
>  		priv->l3l4_filters[i].l4_index = -1;

-- 
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261024-net-atlantic-aq113-apple-regression-v1-1-b8305415a8b1%40jannau.net