linux-kernel - Re: [PATCH] PCI: Check for PCIe downtraining conditions

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

Message-ID: <c4a57eef-8e5a-af00-ba77-1f6b1ccb5672@gmail.com>
Date:   Thu, 31 May 2018 16:52:02 -0500
From:   "Alex G." <mr.nuke.me@...il.com>
To:     Sinan Kaya <okaya@...eaurora.org>, Alex_Gagniuc@...lteam.com,
        bhelgaas@...gle.com
Cc:     Austin.Bolen@...l.com, Shyam.Iyer@...l.com, keith.busch@...el.com,
        linux-pci@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] PCI: Check for PCIe downtraining conditions



On 05/31/2018 12:27 PM, Alex G. wrote:
> On 05/31/2018 12:11 PM, Sinan Kaya wrote:
>> On 5/31/2018 12:49 PM, Alex G. wrote:
>>>> 	bw_cap = pcie_bandwidth_capable(dev, &speed_cap, &width_cap);
>>>> 	bw_avail = pcie_bandwidth_available(dev, &limiting_dev, &speed, &width, *parent*);
>>> That's confusing. I'd expect _capable() and _available() to be
>>> symmetrical. They either both look at one link only, or both go down to
>>> the root port. Though it seems _capable() is link-local, and
>>> _available() is down to root port.
>>>
>>
>> As you know, link speed is a qualification of two devices speed capability.
>> Both speed and width parameters get negotiated by two devices during TS1 and TS2
>> ordered set exchange. 
>>
>> You need to see what your link partner can support in available function() vs.
>> what this device can do in bandwidth() function.
> 
> I see. I'm not sure I can use pcie_print_link_status() without some
> major refactoring. I need to look at capability of device and it
> downstream port. There's no point complaining that an x16 device is
> running at x4 when the port is only x4 capable.
> 
> Let me think some more on this.

I did some thinking, and I don't like using the same message for both
point-to-point links and full-tree links. From a userspace point of view
having an arbitrary terminator in the tree parsing is confusing. So
we're better of not modifying the behavior here.

On the other hand, just printing the link status on probe might be good
enough. If we're downtraining, but there's a slower link somewhere else,
it won't matter much that we're downtrained. Just calling the unmodified
pcie_print_link_status() will most of the time find the exact segment
that's also downtrained. So let's do this in v2

Alex

> Alex
>