Re: [PATCH] PCI/ASPM: Enable ASPM for links under VMD domain

From: Christoph Hellwig <hch@infradead.org>
To: Kai Heng Feng <kai.heng.feng@canonical.com>
Cc: Christoph Hellwig <hch@infradead.org>,
	Bjorn Helgaas <bhelgaas@google.com>,
	jonathan.derrick@intel.com, Mario.Limonciello@dell.com,
	Heiner Kallweit <hkallweit1@gmail.com>,
	Mika Westerberg <mika.westerberg@linux.intel.com>,
	"Rafael J. Wysocki" <rafael.j.wysocki@intel.com>,
	Xiongfeng Wang <wangxiongfeng2@huawei.com>,
	Krzysztof Wilczynski <kw@linux.com>,
	"open list:PCI SUBSYSTEM" <linux-pci@vger.kernel.org>,
	open list <linux-kernel@vger.kernel.org>,
	Dan Williams <dan.j.williams@intel.com>,
	"Huffman, Amber" <amber.huffman@intel.com>
Subject: Re: [PATCH] PCI/ASPM: Enable ASPM for links under VMD domain
Date: Tue, 25 Aug 2020 07:56:34 +0100	[thread overview]
Message-ID: <20200825065634.GA2691@infradead.org> (raw)
In-Reply-To: <08080FC7-861B-472A-BD7D-02D33926677F@canonical.com>

On Tue, Aug 25, 2020 at 02:39:55PM +0800, Kai Heng Feng wrote:
> Hi Christoph,
> 
> > On Aug 25, 2020, at 2:23 PM, Christoph Hellwig <hch@infradead.org> wrote:
> > 
> > On Fri, Aug 21, 2020 at 08:32:20PM +0800, Kai-Heng Feng wrote:
> >> New Intel laptops with VMD cannot reach deeper power saving state,
> >> renders very short battery time.
> > 
> > So what about just disabling VMD given how bloody pointless it is?
> > Hasn't anyone learned from the AHCI remapping debacle?
> > 
> > I'm really pissed at all this pointless crap intel comes up with just
> > to make life hard for absolutely no gain.  Is it so hard to just leave
> > a NVMe device as a standard NVMe device instead of f*^&ing everything
> > up in the chipset to make OS support a pain and I/O slower than by
> > doing nothing?
> 
> From what I can see from the hardwares at my hand, VMD only enables a PCI domain and PCI bridges behind it.
> 
> NVMe works as a regular NVMe under those bridges. No magic remapping happens here.

It definitively is less bad than the AHCI remapping, that is for sure.

But it still requires:

 - a new OS driver just to mak the PCIe device show up
 - indirections in the irq handling
 - indirections in the DMA handling
 - hacks for ASPSM
 - hacks for X (there were a few more)

while adding absolutely no value.  Basically we have to add a large
chunk of kernel code just to undo silicone/firmware Intel added to their
platform to make things complicated.  I mean it is their platform and if
they want a "make things complicated" option that is fine, but it should
not be on by default.