Re: [PATCH v2] PCI/IDE: Fix duplicate stream symlink names for TSM class devices

[<dan.j.williams@intel.com>]

Xu Yilun wrote:
> On Thu, Jan 22, 2026 at 05:07:01PM -0800, dan.j.williams@intel.com wrote:
> > dan.j.williams@ wrote:
> > [..]
> > > However, after seeing Jonathan's feedback and noticing that he missed
> > > that 'H' 'R' and 'E' are documented in the host bridge ABI I think it
> > > would be better to simplify this to just a link back to the host bridge.
> > > 
> > >     /sys/class/tsm/tsmN/pciDDDD:BB => /sys/devices/pciDDDD:BB
> > > 
> > > That achieves the same result and is easier to document as "When a TSM
> > > has a established any IDE stream it links to the host bridge. When the
> > > last stream is removed the link is removed." It achieves the goal of
> > > letting an admin do "ls /sys/class/tsm/tsmN/*/stream*" to get a survey
> > > of all consumed stream resources in the system.
> > 
> > In fact it does not even need to be dynamic. At tsm_register() time when
> > @pci_ops is provided, link all host bridges. Unlink them at unregister
> 
> I'm sort of supporting dynamic. My DUT has 40 host bridges registered,
> most of them has nothing to do with TSM/IDE, so I'm afraid if it is
> overkill to list them all, and bury the real TSM capable bridge in the
> noise.
> 
> And if TSM always list all bridges then why we need these symlinks, we
> can just:
> 
>   ls -d /sys/devices/pci*\:*/stream*
> 
> 
> I assume the annoying part of dynamic is we need to refcount, which IMHO
> unnecessarily complex and you are trying to avoid, is it?

I am ok with a simple xarray of registered host bridges that gets
cleaned up when the last stream leaves.

The end goal is "ls /sys/class/tsm/tsmN/*/stream*" gives valuable signal to
the user, and yes 40 host bridges of noise should be avoided.