From: Jonathan Cameron <Jonathan.Cameron@Huawei.com>
To: Ben Widawsky <ben.widawsky@intel.com>
Cc: <linux-cxl@vger.kernel.org>, <patches@lists.linux.dev>,
Alison Schofield <alison.schofield@intel.com>,
Dan Williams <dan.j.williams@intel.com>,
"Ira Weiny" <ira.weiny@intel.com>,
Vishal Verma <vishal.l.verma@intel.com>,
"Bjorn Helgaas" <helgaas@kernel.org>, <nvdimm@lists.linux.dev>,
<linux-pci@vger.kernel.org>
Subject: Re: [PATCH v3 11/14] cxl/region: Add support for single switch level
Date: Fri, 18 Feb 2022 18:23:48 +0000 [thread overview]
Message-ID: <20220218182348.00002226@Huawei.com> (raw)
In-Reply-To: <20220215161014.00001c68@Huawei.com>
On Tue, 15 Feb 2022 16:10:14 +0000
Jonathan Cameron <Jonathan.Cameron@Huawei.com> wrote:
> On Thu, 27 Jan 2022 16:27:04 -0800
> Ben Widawsky <ben.widawsky@intel.com> wrote:
>
> > CXL switches have HDM decoders just like host bridges and endpoints.
> > Their programming works in a similar fashion.
> >
> > The spec does not prohibit multiple levels of switches, however, those
> > are not implemented at this time.
> >
> > Signed-off-by: Ben Widawsky <ben.widawsky@intel.com>
> Hi Ben,
>
> I'm still hammering away at trying to bring up qemu switch emulation.
> Even though I know you are reworking this, seems only sensible to point
> out issues when I hit them. If no longer relevant you can ignore them!
>
> With these bits and a few other minor tweaks the decoders now look
> to be right - I just need to wire up the QEMU side so that
> I don't get hardware exceptions on actually reading and writing once
> the region is bound :)
QEMU emulation of a very basic CXL switch at:
https://gitlab.com/jic23/qemu/-/commits/cxl-v7-draft-for-test
Hope that is helpful for testing next version of this.
I've been testing with a 4 port switch and 4 type 3 devices
and with a few hacks/fixes as mentioned in review feedback it all
works (well devmem2 gives me what I expect. :)
Jonathan
>
> Thanks,
>
> J
> > ---
> > drivers/cxl/cxl.h | 5 ++++
> > drivers/cxl/region.c | 61 ++++++++++++++++++++++++++++++++++++++++++--
> > 2 files changed, 64 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
> > index 8ace6cca0776..d70d8c85d05f 100644
> > --- a/drivers/cxl/cxl.h
> > +++ b/drivers/cxl/cxl.h
> > @@ -96,6 +96,11 @@ static inline u8 cxl_to_ig(u16 g)
> > return ilog2(g) - 8;
> > }
> >
> > +static inline int cxl_to_ways(u8 ways)
> > +{
> > + return 1 << ways;
> > +}
> > +
> > static inline bool cxl_is_interleave_ways_valid(int iw)
> > {
> > switch (iw) {
> > diff --git a/drivers/cxl/region.c b/drivers/cxl/region.c
> > index b8982be13bfe..f748060733dd 100644
> > --- a/drivers/cxl/region.c
> > +++ b/drivers/cxl/region.c
> > @@ -359,6 +359,23 @@ static bool has_switch(const struct cxl_region *cxlr)
> > return false;
> > }
> >
> > +static bool has_multi_switch(const struct cxl_region *cxlr)
> > +{
> > + struct cxl_memdev *ep;
> > + int i;
> > +
> > + for_each_cxl_endpoint(ep, cxlr, i)
> > + if (ep->port->depth > 3)
> > + return true;
> > +
> > + return false;
> > +}
> > +
> > +static struct cxl_port *get_switch(struct cxl_memdev *ep)
> > +{
> > + return to_cxl_port(ep->port->dev.parent);
> > +}
> > +
> > static struct cxl_decoder *get_decoder(struct cxl_region *cxlr,
> > struct cxl_port *p)
> > {
> > @@ -409,6 +426,8 @@ static bool region_hb_rp_config_valid(struct cxl_region *cxlr,
> > const struct cxl_decoder *rootd,
> > bool state_update)
> > {
> > + const int region_ig = cxl_to_ig(cxlr->config.interleave_granularity);
> > + const int region_eniw = cxl_to_eniw(cxlr->config.interleave_ways);
> > const int num_root_ports = get_num_root_ports(cxlr);
> > struct cxl_port *hbs[CXL_DECODER_MAX_INTERLEAVE];
> > struct cxl_decoder *cxld, *c;
> > @@ -416,8 +435,12 @@ static bool region_hb_rp_config_valid(struct cxl_region *cxlr,
> >
> > hb_count = get_unique_hostbridges(cxlr, hbs);
> >
> > - /* TODO: Switch support */
> > - if (has_switch(cxlr))
> > + /* TODO: support multiple levels of switches */
> > + if (has_multi_switch(cxlr))
> > + return false;
> > +
> > + /* TODO: x3 interleave for switches is hard. */
> > + if (has_switch(cxlr) && !is_power_of_2(region_ways(cxlr)))
> > return false;
> >
> > /*
> > @@ -470,8 +493,14 @@ static bool region_hb_rp_config_valid(struct cxl_region *cxlr,
> > list_for_each_entry(rp, &hb->dports, list) {
> > struct cxl_memdev *ep;
> > int port_grouping = -1;
> > + int target_ndx;
> As things currently stand, with a switch connected below a single port
> of a host bridge (4 type 3 off the switch) this will program the HB
> decoder to have 4 targets, all routed to the switch USP.
>
> There is an argument that this is correct but its not what I'd expect.
> I'd expect to see 1 target only. It's not a problem for small cases, but
> with enough rp and switches we can run out of targets.
>
> >
> > for_each_cxl_endpoint_hb(ep, cxlr, hb, idx) {
> > + struct cxl_decoder *switch_cxld;
> > + struct cxl_dport *target;
> > + struct cxl_port *switch_port;
> > + bool found = false;
> > +
> > if (get_rp(ep) != rp)
> > continue;
> >
> > @@ -499,6 +528,34 @@ static bool region_hb_rp_config_valid(struct cxl_region *cxlr,
> >
> > cxld->interleave_ways++;
> > cxld->target[port_grouping] = get_rp(ep);
> > +
> > + /*
> > + * At least one switch is connected here if the endpoint
> > + * has a depth > 2
> > + */
> > + if (ep->port->depth == 2)
> > + continue;
> > +
> > + /* Check the staged list to see if this
> > + * port has already been added
> > + */
> > + switch_port = get_switch(ep);
> > + list_for_each_entry(switch_cxld, &cxlr->staged_list, region_link) {
> > + if (to_cxl_port(switch_cxld->dev.parent) == switch_port)
> > + found = true;
>
> break;
>
> > + }
> > +
> > + if (found) {
> > + target = cxl_find_dport_by_dev(switch_port, ep->dev.parent->parent);
> > + switch_cxld->target[target_ndx++] = target;
> > + continue;
> > + }
> > +
> > + target_ndx = 0;
> > +
> > + switch_cxld = get_decoder(cxlr, switch_port);
> > + switch_cxld->interleave_ways++;
> > + switch_cxld->interleave_granularity = cxl_to_ways(region_ig + region_eniw);
>
> I'm not following this. Perhaps comment on why this particular maths? I was assuming the switch
> interleave granularity would that of the region as the switch is last level of decode.
>
> Need to do the equivalent here of what you do in the if (found) or the first target is missed.
> Also need to be updating interleave_ways only in the found path, not here (as the default is 1)
>
> > }
> > }
> > }
>
next prev parent reply other threads:[~2022-02-18 18:23 UTC|newest]
Thread overview: 70+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-01-28 0:26 [PATCH v3 00/14] CXL Region driver Ben Widawsky
2022-01-28 0:26 ` [PATCH v3 01/14] cxl/region: Add region creation ABI Ben Widawsky
2022-01-28 18:14 ` Dan Williams
2022-01-28 18:59 ` Dan Williams
2022-02-02 18:26 ` Ben Widawsky
2022-02-02 18:28 ` Ben Widawsky
2022-02-02 18:48 ` Ben Widawsky
2022-02-02 19:00 ` Dan Williams
2022-02-02 19:02 ` Ben Widawsky
2022-02-02 19:15 ` Dan Williams
2022-02-01 22:42 ` Ben Widawsky
2022-02-01 15:53 ` Jonathan Cameron
2022-02-17 17:10 ` [PATCH v4 " Ben Widawsky
2022-02-17 17:19 ` [PATCH v5 01/15] " Ben Widawsky
2022-02-17 17:33 ` Ben Widawsky
2022-02-17 17:58 ` Dan Williams
2022-02-17 18:58 ` Ben Widawsky
2022-02-17 20:26 ` Dan Williams
2022-02-17 22:22 ` Ben Widawsky
2022-02-17 23:32 ` Dan Williams
2022-02-18 16:41 ` Ben Widawsky
2022-01-28 0:26 ` [PATCH v3 02/14] cxl/region: Introduce concept of region configuration Ben Widawsky
2022-01-29 0:25 ` Dan Williams
2022-02-01 14:59 ` Ben Widawsky
2022-02-03 5:06 ` Dan Williams
2022-02-01 23:11 ` Ben Widawsky
2022-02-03 17:48 ` Dan Williams
2022-02-03 22:23 ` Ben Widawsky
2022-02-03 23:27 ` Dan Williams
2022-02-04 0:19 ` Ben Widawsky
2022-02-04 2:45 ` Dan Williams
2022-02-17 18:36 ` Ben Widawsky
2022-02-17 19:57 ` Dan Williams
2022-02-17 20:20 ` Ben Widawsky
2022-02-17 21:12 ` Dan Williams
2022-02-23 21:49 ` Ben Widawsky
2022-02-23 22:24 ` Dan Williams
2022-02-23 22:31 ` Ben Widawsky
2022-02-23 22:42 ` Dan Williams
2022-01-28 0:26 ` [PATCH v3 03/14] cxl/mem: Cache port created by the mem dev Ben Widawsky
2022-02-17 1:20 ` Dan Williams
2022-01-28 0:26 ` [PATCH v3 04/14] cxl/region: Introduce a cxl_region driver Ben Widawsky
2022-02-01 16:21 ` Jonathan Cameron
2022-02-17 6:04 ` Dan Williams
2022-01-28 0:26 ` [PATCH v3 05/14] cxl/acpi: Handle address space allocation Ben Widawsky
2022-02-18 19:17 ` Dan Williams
2022-01-28 0:26 ` [PATCH v3 06/14] cxl/region: Address " Ben Widawsky
2022-02-18 19:51 ` Dan Williams
2022-01-28 0:27 ` [PATCH v3 07/14] cxl/region: Implement XHB verification Ben Widawsky
2022-02-18 20:23 ` Dan Williams
2022-01-28 0:27 ` [PATCH v3 08/14] cxl/region: HB port config verification Ben Widawsky
2022-02-14 16:20 ` Jonathan Cameron
2022-02-14 17:51 ` Ben Widawsky
2022-02-14 18:09 ` Jonathan Cameron
2022-02-15 16:35 ` Jonathan Cameron
2022-02-18 21:04 ` Dan Williams
2022-01-28 0:27 ` [PATCH v3 09/14] cxl/region: Add infrastructure for decoder programming Ben Widawsky
2022-02-01 18:16 ` Jonathan Cameron
2022-02-18 21:53 ` Dan Williams
2022-01-28 0:27 ` [PATCH v3 10/14] cxl/region: Collect host bridge decoders Ben Widawsky
2022-02-01 18:21 ` Jonathan Cameron
2022-02-18 23:42 ` Dan Williams
2022-01-28 0:27 ` [PATCH v3 11/14] cxl/region: Add support for single switch level Ben Widawsky
2022-02-01 18:26 ` Jonathan Cameron
2022-02-15 16:10 ` Jonathan Cameron
2022-02-18 18:23 ` Jonathan Cameron [this message]
2022-01-28 0:27 ` [PATCH v3 12/14] cxl: Program decoders for regions Ben Widawsky
2022-02-24 0:08 ` Dan Williams
2022-01-28 0:27 ` [PATCH v3 13/14] cxl/pmem: Convert nvdimm bridge API to use dev Ben Widawsky
2022-01-28 0:27 ` [PATCH v3 14/14] cxl/region: Create an nd_region Ben Widawsky
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=20220218182348.00002226@Huawei.com \
--to=jonathan.cameron@huawei.com \
--cc=alison.schofield@intel.com \
--cc=ben.widawsky@intel.com \
--cc=dan.j.williams@intel.com \
--cc=helgaas@kernel.org \
--cc=ira.weiny@intel.com \
--cc=linux-cxl@vger.kernel.org \
--cc=linux-pci@vger.kernel.org \
--cc=nvdimm@lists.linux.dev \
--cc=patches@lists.linux.dev \
--cc=vishal.l.verma@intel.com \
/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).