[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <e3f0e78a-3e3d-aae7-2588-599184dee534@arm.com>
Date: Mon, 10 Sep 2018 14:26:30 +0100
From: Robin Murphy <robin.murphy@....com>
To: Christoph Hellwig <hch@....de>, Meelis Roos <mroos@...ux.ee>
Cc: iommu@...ts.linux-foundation.org,
Linux Kernel list <linux-kernel@...r.kernel.org>
Subject: Re: 4.19-rc3: IOMMU PT warnings: WARNING: CPU: 0 PID: 1 at
mm/page_alloc.c:4065 __alloc_pages_slowpath+0xb7b/0xcd0
On 10/09/18 14:23, Christoph Hellwig wrote:
> On Mon, Sep 10, 2018 at 11:16:12AM +0300, Meelis Roos wrote:
>> In 4.19-rc1, I found that to keep Intel integrated graphics working when
>> VT-d is on, I need to have IOMMU passthough on or GPU init would fail
>> with DMAR error and the screen would display static gibberish. So I
>> turned IOMMU PT on by default in kernel config. It worked, also in rc2.
>>
>> Now with rc3, it still works but I get some warning splats in dmesg
>> about IOMMU passthrough:
>
> Any chance you could bisect this down to a specific commit? For
> a start it might be worth if it is reproducable with:
>
> git://git.infradead.org/users/hch/dma-mapping.git dma-mapping-4.19-2
>
> which were the only dma-mapping changes post -rc2.
Given the __alloc_pages_slowpath() warning, this looks like it's
probably stemming from the oversized PASID table issue which a few other
folks have hit too. I think this patch is the most up-to-date fix:
https://www.mail-archive.com/iommu@lists.linux-foundation.org/msg25639.html
Robin.
Powered by blists - more mailing lists