Thread (1 message) 1 message, 1 author, 2017-06-01

Re: [PATCH 14/22] gpu: host1x: Forbid relocation address shifting in the firewall

From: Dmitry Osipenko <hidden>
Date: 2017-06-01 19:15:14
Also in: dri-devel

On 01.06.2017 21:51, Mikko Perttunen wrote:
On 06/01/2017 09:44 PM, Dmitry Osipenko wrote:
quoted
On 01.06.2017 21:37, Dmitry Osipenko wrote:
quoted
On 01.06.2017 20:39, Mikko Perttunen wrote:
quoted
Reviewed-by: Mikko Perttunen <redacted>

On 05/23/2017 03:14 AM, Dmitry Osipenko wrote:
quoted
Incorrectly shifted relocation address will cause a lower memory corruption
and likely a hang on a write or a read of an arbitrary data in case of IOMMU
absent. As of now there is no use for the address shifting (at least on
Tegra20) and adding a proper shifting / sizes validation is much more work.
Perhaps change to "As of now there is no use for the address shifting on
Tegra20" if you post another revision.
I'll post a new revision of the series after getting comments to the all
patches, to not churn the ML. Thank you very much for the reviews!
However, given your previous comments to this patch, I'll probably add a bypass
of the shit checking in case of IOMMU presence.
The IOMMU presence checking probably wouldn't be enough. Better to check the
Host1x version instead, to not break the non-IOMMU case on modern Tegras.
I don't think that's needed - the firewall will deny pretty much all VIC
submissions due to is_addr_reg not being implemented so it cannot reasonably be
used on modern Tegras anyway.
Either firewall should be completely avoided on newer Tegras or it should
perform at least some checks and not break the newer Tegras.

-- 
Dmitry
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help