From: Sasha Levin <sashal@kernel.org>
To: stable@vger.kernel.org
Cc: Sasha Levin <sashal@kernel.org>,
	Eddie Phillips <eddiephillips@google.com>,
	Jordan Rhee <jordanrhee@google.com>,
	Harshitha Ramamurthy <hramamurthy@google.com>,
	Przemek Kitszel <przemyslaw.kitszel@intel.com>,
	Jakub Kicinski <kuba@kernel.org>
Subject: Re: [PATCH 6.1.y] gve: fix Rx queue stall on alloc failure
Date: Fri, 31 Jul 2026 21:40:04 -0400
Message-ID: <20260731211002.stable-0003@kernel.org>
In-Reply-To: <20260731154449.2755799-1-eddiephillips@google.com>
References: <2026072934-rule-library-9954@gregkh> <20260731154449.2755799-1-eddiephillips@google.com>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit

On Fri, Jul 31, 2026 at 08:44:49AM -0700, Eddie Phillips wrote:
> (cherry picked from commit b65352a1bac64442ad95e64f385b40ccb9f1b0db)
> ---
>  drivers/net/ethernet/google/gve/gve.h | 2 ++
>  1 file changed, 2 insertions(+)

This 6.1.y version only contains the gve.h hunk: the <linux/timer.h>
include and the "struct timer_list starvation_timer" member. The
starvation timer callback, the timer_setup(), the mod_timer() threshold
check in gve_rx_post_buffers_dqo() and the timer_shutdown_sync() teardown
are all missing, so it applies cleanly but is a no-op that only adds a
dead struct member - the queue stall stays unfixed. (The new member is
also indented with spaces rather than a tab.) Could you send a v2 with
the complete backport?

While you're at it, could you send a 6.6.y backport too? 6.6 failed the
same way and still doesn't carry b65352a1bac6 ("gve: fix Rx queue stall
on alloc failure") at all. 6.6 and 6.1 have the same helpers as 5.15, so
the shape of your 5.15.y patch (timer_setup() in gve_rx_alloc_ring_dqo(),
timer_shutdown_sync() in gve_rx_free_ring_dqo() plus gve_remove_napi())
should carry over with little change.

Your 5.15.y patch itself looks good, but I'm holding it until the 6.6 and
6.1 backports are ready - otherwise the older tree ships the fix while
the newer ones stay broken. Once all three are in hand I'll queue them
together, newest tree first.

-- 
Thanks,
Sasha