[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20240502124103.GA3341011@nvidia.com>
Date: Thu, 2 May 2024 09:41:03 -0300
From: Jason Gunthorpe <jgg@...dia.com>
To: Nicolin Chen <nicolinc@...dia.com>
Cc: will@...nel.org, robin.murphy@....com, joro@...tes.org,
thierry.reding@...il.com, vdumpa@...dia.com, jonathanh@...dia.com,
linux-kernel@...r.kernel.org, iommu@...ts.linux.dev,
linux-arm-kernel@...ts.infradead.org, linux-tegra@...r.kernel.org
Subject: Re: [PATCH v6 5/6] iommu/arm-smmu-v3: Add in-kernel support for
NVIDIA Tegra241 (Grace) CMDQV
On Wed, May 01, 2024 at 10:43:39AM -0700, Nicolin Chen wrote:
> > It doesn't fix any real race, I'm not sure what this is supposed to be
> > doing. The cmdq becomes broken and you get an ISR, so before the ISR
> > it will still post but get stuck, during the ISR it will avoid
> > posting, and after it will go back to posting?
> >
> > Why? Just always post to the Q and let the ISR fix it?
>
> Yes, we could do so. I was thinking of the worst case by giving
> the guest OS a chance to continue (though in a slower mode), if
> something unrecoverable happens to the VINTF/VCMDQ part.
Does that happn? The stuck vcmdq will have stuck entries on it no
matter what, can we actually fully recover from that? Ie re-issue the
commands on another queue?
> > So just don't use it. There is no value if the places where it should
> > work automatically are not functioning.
>
> I thought devm could work when rmmod too, not only when the probe
> fails..
It is limited to cases when the probing driver of the passed struct
device unbinds, including probe failure.
Jason
Powered by blists - more mailing lists