[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <20190730.102638.2201830580799063447.davem@davemloft.net>
Date: Tue, 30 Jul 2019 10:26:38 -0700 (PDT)
From: David Miller <davem@...emloft.net>
To: Jose.Abreu@...opsys.com
Cc: netdev@...r.kernel.org, Joao.Pinto@...opsys.com,
peppe.cavallaro@...com, alexandre.torgue@...com,
mcoquelin.stm32@...il.com,
linux-stm32@...md-mailman.stormreply.com,
linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
jonathanh@...dia.com
Subject: Re: [PATCH net] net: stmmac: Sync RX Buffer upon allocation
From: Jose Abreu <Jose.Abreu@...opsys.com>
Date: Tue, 30 Jul 2019 15:57:16 +0200
> With recent changes that introduced support for Page Pool in stmmac, Jon
> reported that NFS boot was no longer working on an ARM64 based platform
> that had the IP behind an IOMMU.
>
> As Page Pool API does not guarantee DMA syncing because of the use of
> DMA_ATTR_SKIP_CPU_SYNC flag, we have to explicit sync the whole buffer upon
> re-allocation because we are always re-using same pages.
>
> In fact, ARM64 code invalidates the DMA area upon two situations [1]:
> - sync_single_for_cpu(): Invalidates if direction != DMA_TO_DEVICE
> - sync_single_for_device(): Invalidates if direction == DMA_FROM_DEVICE
>
> So, as we must invalidate both the current RX buffer and the newly allocated
> buffer we propose this fix.
>
> [1] arch/arm64/mm/cache.S
>
> Reported-by: Jon Hunter <jonathanh@...dia.com>
> Tested-by: Jon Hunter <jonathanh@...dia.com>
> Fixes: 2af6106ae949 ("net: stmmac: Introducing support for Page Pool")
> Signed-off-by: Jose Abreu <joabreu@...opsys.com>
Applied.
Powered by blists - more mailing lists