Re: [PATCH] PCI: Introduce Cray ClusterStor E1000 NVMe slot LED driver

[Bjorn Helgaas <helgaas@kernel.org>]

On Thu, Sep 05, 2024 at 08:19:25AM +0200, Mariusz Tkaczyk wrote:
> On Tue, 3 Sep 2024 17:18:20 -0500
> Bjorn Helgaas <helgaas@kernel.org> wrote:
> 
> > On Tue, Aug 27, 2024 at 02:03:48PM -0700, Tony Hutter wrote:
> > > Add driver to control the NVMe slot LEDs on the Cray ClusterStor E1000.
> > > The driver provides hotplug attention status callbacks for the 24 NVMe
> > > slots on the E1000.  This allows users to access the E1000's locate and
> > > fault LEDs via the normal /sys/bus/pci/slots/<slot>/attention sysfs
> > > entries.  This driver uses IPMI to communicate with the E1000 controller to
> > > toggle the LEDs.  
> > 
> > I hope/assume the interface is the same as one of the others, i.e.,
> > the existing one added for NVMe behind VMD by
> > https://git.kernel.org/linus/576243b3f9ea ("PCI: pciehp: Allow
> > exclusive userspace control of indicators") or the new one for NPEM
> > and the _DSM at
> > https://lore.kernel.org/linux-pci/20240814122900.13525-3-mariusz.tkaczyk@linux.intel.com/
> > 
> > I suppose we intend that the ledmon utility will be able to drive
> > these LEDs?  Whatever the user, we should try to minimize the number
> > of different interfaces for this functionality.
> 
> Ledmon won't support it, at least not in current form. Ledmon
> support for pciehp attention is limited to VMD, i.e. first we must
> find VMD driver then we are looking for slot/attention.  I'm not
> familiar with any attempt to add support for this in ledmon.
> 
> From the end user perspective, I don't like pciehp/attention because
> we are refereeing to pciehp driver not pcieport and to determine
> proper slot we need to do additional matching by slot/address. I
> would be simpler.
> https://github.com/intel/ledmon/blob/main/src/lib/vmdssd.c#L100

All I'm trying to say is that NPEM, the related _DSM, and the VMD
special case are all ways to control NVMe slot IDs.  This Cray thing
is another.  We already have two user interfaces (the NPEM/_DSM one
and the VMD one), and I'd like to avoid adding a third.