linux-kernel - Re: [PATCH] iommu/amd: Fix event counter availability check

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:   Fri, 26 Feb 2021 14:55:16 -0700
From:   Shuah Khan <skhan@...uxfoundation.org>
To:     Paul Menzel <pmenzel@...gen.mpg.de>,
        Alexander Monakov <amonakov@...ras.ru>,
        Suravee Suthikulpanit <suravee.suthikulpanit@....com>
Cc:     iommu@...ts.linux-foundation.org, linux-kernel@...r.kernel.org,
        David Coe <david.coe@...e.co.uk>,
        Joerg Roedel <joro@...tes.org>,
        "Tj (Elloe Linux)" <ml.linux@...oe.vision>,
        Shuah Khan <skhan@...uxfoundation.org>
Subject: Re: [PATCH] iommu/amd: Fix event counter availability check

On 2/26/21 2:44 PM, Paul Menzel wrote:
> [cc: +suravee, +jörg]
> 
> Dear Alex, dear Shuah, dear Suravee, dear Jörg,
> 
> 
> Am 03.06.20 um 08:54 schrieb Alexander Monakov:
>> On Tue, 2 Jun 2020, Shuah Khan wrote:
>>
>>> I changed the logic to read config to get max banks and counters
>>> before checking if counters are writable and tried writing to all.
>>> The result is the same and all of them aren't writable. However,
>>> when disable the writable check and assume they are, I can run
>> [snip]
>>
>> This is similar to what I did. I also noticed that counters can
>> be successfully used with perf if the initial check is ignored.
>> I was considering sending a patch to remove the check and adjust
>> the event counting logic to use counters as read-only, but after
>> a bit more investigation I've noticed how late pci_enable_device
>> is done, and came up with this patch. It's a path of less resistance:
>> I'd expect maintainers to be more averse to removing the check
>> rather than fixing it so it works as intended (even though I think
>> the check should not be there in the first place).
>>
>> However:
>>
>> The ability to modify the counters is needed only for sampling the
>> events (getting an interrupt when a counter overflows). There's no
>> code to do that for these AMD IOMMU counters. A solution I would
>> prefer is to not write to those counters at all. It would simplify or
>> even remove a bunch of code. I can submit a corresponding patch if
>> there's general agreement this path is ok.
>>
>> What do you think?
> 
> I like this idea. Suravee, Jörg, what do you think?
> 
> Commit 6778ff5b21b (iommu/amd: Fix performance counter initialization) 
> delays the boot up to 100 ms, which is over 20 % on fast systems, and 
> also just a workaround, and does not seem to work always. The delay is 
> also not mentioned in the commit message.
> 
>

Sounds good to me. I can test this out on my system.

thanks,
-- Shuah