linux-kernel - Re: [PATCH] iommu: Print default strict or lazy mode at init time

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Tue, 1 Jun 2021 18:44:00 +0100
From:   John Garry <john.garry@...wei.com>
To:     Robin Murphy <robin.murphy@....com>, <joro@...tes.org>,
        <will@...nel.org>
CC:     <iommu@...ts.linux-foundation.org>, <linux-kernel@...r.kernel.org>,
        <hch@....de>
Subject: Re: [PATCH] iommu: Print default strict or lazy mode at init time

>>
>> pr_info("DMA domain default TLB invalidation policy: %s mode %s\n",
>>                iommu_dma_strict ? "strict" : "lazy",
>>                 (iommu_cmd_line & IOMMU_CMD_LINE_STRICT) ?
>>                         "(set via kernel command line)" : "");
>>
>> I think it's worth mentioning "default" somewhere, as not all IOMMUs 
>> or devices will use lazy mode even if it's default.
> 
> But that's part of what I think is misleading - I boot and see that the 
> default is something, so I reboot with iommu.strict to explicitly set it 
> the other way, but now that's the default... huh?
> 
> The way I see it, we're saying what the current IOMMU API policy is - 
> the value of iommu_dma_strict at any given time is fact - but we're not 
> necessarily saying how widely that policy is enforced. We similarly 
> report the type for default domains from global policy even though that 
> may also be overridden per-group by drivers and/or userspace later; 

> we 
> don't say it's the *default* default domain type.

I think that is this is the behavior a user would understand from that 
message.

However from a glance at the intel IOMMU driver, it seems possible to 
change default domain type after iommu_subsys_init().

> 
> However, having now debugged the AMD issue from another thread, I think 
> doing this at subsys_initcall is in fact going to be too early to be 
> meaningful, since it ignores drivers' ability to change the global 
> policy :(

A user may still learn the IOMMU group domain type from sysfs. There is 
no such thing for TLB invalidation mode - how about add a file for this? 
It would be useful.

Thanks,
John