linux-kernel - Re: [PATCH 0/1] usb: xhci: Queue URB_ZERO

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]

Message-ID: <20250922101610.0102e1a1.michal.pecio@gmail.com>
Date: Mon, 22 Sep 2025 10:16:10 +0200
From: Michal Pecio <michal.pecio@...il.com>
To: Mathias Nyman <mathias.nyman@...ux.intel.com>
Cc: linux-usb@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 0/1] usb: xhci: Queue URB_ZERO_PACKET as one TD

On Wed, 10 Sep 2025 01:57:39 +0300, Mathias Nyman wrote:
> On 9.9.2025 20.38, Michal Pecio wrote:
> > But this is not what this patch is about - the trick is to use an
> > *unchained* TRB, which is a separate TD from HW's perspective, and
> > to count it as part of the same TD from the driver's perspective.  
> 
> Ok, I see.
> The whole TD without completion flag does worry me a bit.
> 
> We need to make sure stop/stald mid TD cases work, and  urb length is
> set correctly.

I came up with a potential problem case for clearing IOC:

1. all data of the first TD are sent out sucessfully
2. no completion is generated because no IOC
3. ring stops before advancing to the zero-length TD
4. we only get FSE (Stopped - Length Invalid)

See xHCI 4.6.9:
     Table 4-2: Stop Endpoint Command TRB Handling
       2nd row: Stopped on TD boundary

Current event handler doesn't expect this to happen and actual length
will be reported incorrectly. This would be easy to fix.

But there is also the 0.96 spec where FSE was optional (xHCI G.2), so
on some HCs (like NEC uPD720200) we won't get any event whatsoever and
the almost fully completed URB will seem to have transferred no data.

(This assumes that any HC would stop in this manner rather than advance
to the zero-length TD atomically after previous TD completion and stop
normally in the zero-length TD. So not sure if it's a real problem and
the condition seems hard to trigger for testing purposes.)

Control URBs have the same problem - FSE isn't handled very well and
old HCs would seem to need IOC on the data stage to ensure correct
actual length of cancelled URBs, if anyone cares.

Regards,
Michal