[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20241202163309.05603e96@kernel.org>
Date: Mon, 2 Dec 2024 16:33:09 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: Furong Xu <0x1207@...il.com>
Cc: Jon Hunter <jonathanh@...dia.com>, netdev@...r.kernel.org,
linux-stm32@...md-mailman.stormreply.com,
linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
Alexandre Torgue <alexandre.torgue@...s.st.com>, Jose Abreu
<joabreu@...opsys.com>, "David S. Miller" <davem@...emloft.net>, Eric
Dumazet <edumazet@...gle.com>, Paolo Abeni <pabeni@...hat.com>, Maxime
Coquelin <mcoquelin.stm32@...il.com>, xfr@...look.com, Suraj Jaiswal
<quic_jsuraj@...cinc.com>, Thierry Reding <treding@...dia.com>,
"linux-tegra@...r.kernel.org" <linux-tegra@...r.kernel.org>
Subject: Re: [PATCH net v1] net: stmmac: TSO: Fix unbalanced DMA map/unmap
for non-paged SKB data
On Thu, 28 Nov 2024 14:45:01 +0800 Furong Xu wrote:
> > Let me know if you need any more information.
>
> [ 149.986210] dwc-eth-dwmac 2490000.ethernet eth0: Tx DMA map failed
> and
> [ 245.571688] dwc-eth-dwmac 2490000.ethernet eth0: Tx DMA map failed
> [ 245.575349] dwc-eth-dwmac 2490000.ethernet eth0: Tx DMA map failed
> are reported by stmmac_xmit() obviously, but not stmmac_tso_xmit().
>
> And these crashes are caused by "Tx DMA map failed", as you can see that
> current driver code does not handle this kind of failure so well. It is clear
> that we need to figure out why Tx DMA map failed.
>
> This patch corrects the sequence and timing of DMA unmap by waiting all
> DMA transmit descriptors to be closed by DMA engine for one DMA map in
> stmmac_tso_xmit(), it never leaks DMA addresses and never introduces
> other side effect.
>
> "Tx DMA map failed" is a weird failure, and I cannot reproduce this failure
> on my device with DWMAC CORE 5.10a(Synopsys ID: 0x51) and DWXGMAC CORE 3.20a.
Let me repeat Jon's question - is there any info or test you need from
Jon to make progress with a fix?
If Jon's board worked before and doesn't work with this patch we will
need *a* fix, if no fix is provided our only choice is revert.
Powered by blists - more mailing lists