* [RFC PATCH] cxl/region: Calculate peer distance only for reused targets
@ 2026-10-15 9:40 Richard Cheng
2026-10-16 18:23 ` Jonathan Cameron
2026-10-17 0:30 ` Alison Schofield
0 siblings, 2 replies; 3+ messages in thread
From: Richard Cheng @ 2026-10-15 9:40 UTC (permalink / raw)
To: dave, jic23, dave.jiang, alison.schofield, vishal.l.verma
Cc: iweiny, ming.li, icheng, gourry, rrichter, linux-cxl, linux-kernel
This patch depends on Alison Schofield's v3 "cxl/region: Refactor
cxl_port_setup_targets()" series [1], this is inspired from that serie
as well, since her serie contains no functional change, I pull this one
out as a standalone patch.
cxl_switch_port_setup_targets() calculates peer distance before
searching the existing targets. When the endpoint needs a new target,
the calculated distance is never used.
Move cxl_port_peer_distance() into the matching-target branch,
immediately before check_last_peer(). This avoids unnecessary ancestor
walks and per-port region lookups when adding distinct targets.
[1]:
https://lore.kernel.org/linux-cxl/cover.1791860305.git.alison.schofield@intel.com/
Signed-off-by: Richard Cheng <icheng@nvidia.com>
---
I don't have available HW for multi-way interleaving setup, so I can't
exercise the case optimized here. I would appreciate if someone is
williing to validate the actual benefit of this.
Best regards,
Richard Cheng.
---
drivers/cxl/core/region.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
index 50f4830c225d..efd173c57dad 100644
--- a/drivers/cxl/core/region.c
+++ b/drivers/cxl/core/region.c
@@ -1666,10 +1666,10 @@ static int cxl_switch_port_setup_targets(struct cxl_port *port, struct cxl_regio
cxlsd = to_cxl_switch_decoder(&cxld->dev);
if (cxl_rr->nr_targets_set) {
- int distance = cxl_port_peer_distance(port, cxlr);
-
for (int i = 0; i < cxl_rr->nr_targets_set; i++)
if (ep->dport == cxlsd->target[i]) {
+ int distance = cxl_port_peer_distance(port, cxlr);
+
rc = check_last_peer(cxled, ep, cxl_rr,
distance);
if (rc)
--
2.43.0
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [RFC PATCH] cxl/region: Calculate peer distance only for reused targets
2026-10-15 9:40 [RFC PATCH] cxl/region: Calculate peer distance only for reused targets Richard Cheng
@ 2026-10-16 18:23 ` Jonathan Cameron
2026-10-17 0:30 ` Alison Schofield
1 sibling, 0 replies; 3+ messages in thread
From: Jonathan Cameron @ 2026-10-16 18:23 UTC (permalink / raw)
To: Richard Cheng
Cc: dave, dave.jiang, alison.schofield, vishal.l.verma, iweiny,
ming.li, gourry, rrichter, linux-cxl, linux-kernel
On Thu, 15 Oct 2026 17:40:59 +0800
Richard Cheng <icheng@nvidia.com> wrote:
> This patch depends on Alison Schofield's v3 "cxl/region: Refactor
> cxl_port_setup_targets()" series [1], this is inspired from that serie
series is a an odd word in English. Singular and plural are the same.
> as well, since her serie contains no functional change, I pull this one
> out as a standalone patch.
>
> cxl_switch_port_setup_targets() calculates peer distance before
> searching the existing targets. When the endpoint needs a new target,
> the calculated distance is never used.
>
> Move cxl_port_peer_distance() into the matching-target branch,
> immediately before check_last_peer(). This avoids unnecessary ancestor
> walks and per-port region lookups when adding distinct targets.
>
> [1]:
> https://lore.kernel.org/linux-cxl/cover.1791860305.git.alison.schofield@intel.com/
Use a link tag.
> Signed-off-by: Richard Cheng <icheng@nvidia.com>
> ---
> I don't have available HW for multi-way interleaving setup, so I can't
> exercise the case optimized here. I would appreciate if someone is
> williing to validate the actual benefit of this.
Maybe give it a spin on qemu. But it looks like it's just a code reorder
effectively so It would be very odd if it is a problem.
I'd just put this as a comment on Alison's series as I don't think
that has merged yet.
Seems like a nice little optimization.
>
> Best regards,
> Richard Cheng.
> ---
> drivers/cxl/core/region.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
> index 50f4830c225d..efd173c57dad 100644
> --- a/drivers/cxl/core/region.c
> +++ b/drivers/cxl/core/region.c
> @@ -1666,10 +1666,10 @@ static int cxl_switch_port_setup_targets(struct cxl_port *port, struct cxl_regio
>
> cxlsd = to_cxl_switch_decoder(&cxld->dev);
> if (cxl_rr->nr_targets_set) {
> - int distance = cxl_port_peer_distance(port, cxlr);
> -
> for (int i = 0; i < cxl_rr->nr_targets_set; i++)
> if (ep->dport == cxlsd->target[i]) {
> + int distance = cxl_port_peer_distance(port, cxlr);
> +
> rc = check_last_peer(cxled, ep, cxl_rr,
> distance);
> if (rc)
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [RFC PATCH] cxl/region: Calculate peer distance only for reused targets
2026-10-15 9:40 [RFC PATCH] cxl/region: Calculate peer distance only for reused targets Richard Cheng
2026-10-16 18:23 ` Jonathan Cameron
@ 2026-10-17 0:30 ` Alison Schofield
1 sibling, 0 replies; 3+ messages in thread
From: Alison Schofield @ 2026-10-17 0:30 UTC (permalink / raw)
To: Richard Cheng
Cc: dave, jic23, dave.jiang, vishal.l.verma, iweiny, ming.li, gourry,
rrichter, linux-cxl, linux-kernel
On Thu, Oct 15, 2026 at 05:40:59PM +0800, Richard Cheng wrote:
> This patch depends on Alison Schofield's v3 "cxl/region: Refactor
> cxl_port_setup_targets()" series [1], this is inspired from that serie
> as well, since her serie contains no functional change, I pull this one
> out as a standalone patch.
>
> cxl_switch_port_setup_targets() calculates peer distance before
> searching the existing targets. When the endpoint needs a new target,
> the calculated distance is never used.
>
> Move cxl_port_peer_distance() into the matching-target branch,
> immediately before check_last_peer(). This avoids unnecessary ancestor
> walks and per-port region lookups when adding distinct targets.
>
> [1]:
> https://lore.kernel.org/linux-cxl/cover.1791860305.git.alison.schofield@intel.com/
> Signed-off-by: Richard Cheng <icheng@nvidia.com>
> ---
> I don't have available HW for multi-way interleaving setup, so I can't
> exercise the case optimized here. I would appreciate if someone is
> williing to validate the actual benefit of this.
Nice patch! I've picked it up into v4 of the refactor set.
I tested it with using the cxl unit test cxl-region-sysfs.sh and this
at the end of cxl_port_peer_distance():
dev_dbg(&cxlr->dev, "%s:%s: peer distance %d\n",
dev_name(port->uport_dev), dev_name(&port->dev), distance);
The x8 region went from 10 calls to 4. The eliminated calls were for
new downstream targets and the remaining calls were for reused
targets where check_last_peer() needs the distance.
The x2 auto region also went from 2 calls to 1.
I tested the peer validation path by swapping endpoints at positions
0 and 2 in cxl-region-sysfs.sh. The mismatched peer was detected and
the region rejected.
-- Alison
>
> Best regards,
> Richard Cheng.
> ---
> drivers/cxl/core/region.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
> index 50f4830c225d..efd173c57dad 100644
> --- a/drivers/cxl/core/region.c
> +++ b/drivers/cxl/core/region.c
> @@ -1666,10 +1666,10 @@ static int cxl_switch_port_setup_targets(struct cxl_port *port, struct cxl_regio
>
> cxlsd = to_cxl_switch_decoder(&cxld->dev);
> if (cxl_rr->nr_targets_set) {
> - int distance = cxl_port_peer_distance(port, cxlr);
> -
> for (int i = 0; i < cxl_rr->nr_targets_set; i++)
> if (ep->dport == cxlsd->target[i]) {
> + int distance = cxl_port_peer_distance(port, cxlr);
> +
> rc = check_last_peer(cxled, ep, cxl_rr,
> distance);
> if (rc)
> --
> 2.43.0
>
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-10-17 0:30 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-10-15 9:40 [RFC PATCH] cxl/region: Calculate peer distance only for reused targets Richard Cheng
2026-10-16 18:23 ` Jonathan Cameron
2026-10-17 0:30 ` Alison Schofield
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®