qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Dan Williams <dan.j.williams@intel.com>
To: Jonathan Cameron <Jonathan.Cameron@huawei.com>,
	Dan Williams <dan.j.williams@intel.com>
Cc: <qemu-devel@nongnu.org>, <linux-cxl@vger.kernel.org>
Subject: Re: [PATCH] hw/acpi/cxl: Drop device-memory support from CFMWS entries
Date: Wed, 22 Mar 2023 15:10:53 -0700	[thread overview]
Message-ID: <641b7ced3bc1f_269929449@dwillia2-xfh.jf.intel.com.notmuch> (raw)
In-Reply-To: <20230321120147.00003ea7@Huawei.com>

Jonathan Cameron wrote:
> On Mon, 20 Mar 2023 23:08:31 -0700
> Dan Williams <dan.j.williams@intel.com> wrote:
> 
> > While it was a reasonable idea to specify no window restricitions at the
> > outset of the CXL emulation support, it turns out that in practice a
> > platform will never follow the QEMU example of specifying simultaneous
> > support for HDM-H and HDM-D[B] in a single window.
> > 
> > HDM-D mandates extra bus cycles for host/device bias protocol, and HDM-DB
> > mandates extra bus cycles for back-invalidate protocol, so hardware must
> > be explicitly prepared for device-memory unlike host-only memory
> > (HDM-H).
> > 
> > In preparation for the kernel dropping support for windows that do not
> > select between device and host-only memory, move QEMU exclusively to
> > declaring host-only windows.
> > 
> > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> Hi Dan,
> 
> Can we have some spec references? I think the Protocol tables in
> appendix C would work for that - but more specific examples called
> out from them would be good.

I presume our messages crossed in the ether:

https://lore.kernel.org/linux-cxl/641a018ed7fb8_269929455@dwillia2-xfh.jf.intel.com.notmuch/

...but yes, if I was still committed to this change, which I am not at
this point, some references from Appendix C and "3.3.11 Forward Progress
and Ordering Rules would be appropriate".

> I'm also not totally convinced it isn't a host implementation detail
> - key here is that host bridge's are still part of the host so can
> do impdef stuff as long as they look correct to CXL side and to
> host side.
> 
> Time for some wild implementation conjecturing.
> 
> Imagine a host that has host bridges of above average smartness.
> Those host bridges have HDM decoders (this doesn't work if not)
> 
> Host is using a single HPA window for HDM-D[B] and HDM-H.
> The host bridge knows the target is HDM-H - it can get that form
> the HDM decoder Target Type bit etc.  The HB can send (to the
> rest of the Host) whatever replies are necessary / fill in extra
> info to make it look like HDM-D[B] to the host interconnect protocol.
> 
> (after some fun with a white board we think you can make this efficient
>  by potentially making the Host bridge HDM decoder setup visible to
>  other parts of the host - relatively easy give lots of time allowed
>  for a decoder commit).
> 
> Why would you do this?  Limited HPA space availability on the host
> and wanting to be very flexible about use of the CXL windows.

Limited HPA space and fewer decode rules at the root level indeed sounds
compelling.

> Obviously this is all moot if there is a constraint we can point to
> in a specification.

At this time I don't have such a reference.

> BTW. I'm carrying a patch (it's on the gitlab tree) that I haven't
> posted yet that lets you configure this restriction at runtime as
> a similar potential host implementation restriction occurs for
> PMEM vs Volatile.  That is also needed to exercise the fun corners of
> QTG etc.

Sounds good.


  reply	other threads:[~2023-03-22 23:32 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-03-21  6:08 [PATCH] hw/acpi/cxl: Drop device-memory support from CFMWS entries Dan Williams
2023-03-21 12:01 ` Jonathan Cameron via
2023-03-22 22:10   ` Dan Williams [this message]
2023-03-21 19:12 ` Dan Williams

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=641b7ced3bc1f_269929449@dwillia2-xfh.jf.intel.com.notmuch \
    --to=dan.j.williams@intel.com \
    --cc=Jonathan.Cameron@huawei.com \
    --cc=linux-cxl@vger.kernel.org \
    --cc=qemu-devel@nongnu.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).