Skip to content

Fix downstream endpoint recovery #175

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
faizana-nvidia wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Fix downstream endpoint recovery #175

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
fcb4202
mctpd: return the queried UUID to the caller of query_get_peer_uuid()
faizana-nvidia Sep 2, 2026
7509c16
mctpd: fix EID recovery for bridged (downstream) endpoints
faizana-nvidia Sep 2, 2026
f50de80
tests: add downstream endpoint recovery tests
faizana-nvidia Aug 27, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
201 changes: 148 additions & 53 deletions src/mctpd.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -224,6 +224,9 @@ struct peer {
uint8_t pool_size;
uint8_t pool_start;

// Next hop toward the bus owner for downstream peer
struct peer *gateway;

struct {
sd_event_source **sources;
} bridge_ep_poll;
Expand Down Expand Up @@ -2954,7 +2957,7 @@ static int query_get_peer_uuid_by_phys(struct ctx *ctx, const dest_phys *dest,
return rc;
}

static int query_get_peer_uuid(struct peer *peer)
static int query_get_peer_uuid(struct peer *peer, uint8_t uuid[16])
Comment thread
jk-ozlabs marked this conversation as resolved.
{
struct mctp_ctrl_resp_get_uuid *resp = NULL;
struct mctp_ctrl_cmd_get_uuid req;
Expand Down Expand Up @@ -2985,10 +2988,7 @@ static int query_get_peer_uuid(struct peer *peer)
goto out;

resp = cmd.resp;
rc = peer_set_uuid(peer, resp->uuid);
if (rc < 0)
goto out;
rc = 0;
memcpy(uuid, resp->uuid, sizeof(resp->uuid));

out:
mctp_ctrl_cmd_free(&cmd);
Expand Down Expand Up @@ -3458,10 +3458,19 @@ static int query_peer_properties(struct peer *peer)
}
}

rc = query_get_peer_uuid(peer);
if (rc < 0 && peer->ctx->verbose) {
errno = -rc;
warn("Error getting UUID for %s", peer_tostr(peer));
{
uint8_t uuid[16];
memset(uuid, 0, sizeof(uuid));
rc = query_get_peer_uuid(peer, uuid);
if (rc < 0) {
if (peer->ctx->verbose) {
errno = -rc;
warn("Error getting UUID for %s",
peer_tostr(peer));
}
} else {
rc = peer_set_uuid(peer, uuid);
}
}

// TODO: emit property changed? Though currently they are all const.
Expand Down Expand Up @@ -3751,35 +3760,48 @@ static int peer_endpoint_recover(sd_event_source *s, uint64_t usec,
* response reporting the current EID. This is the test recommended by 8.17.6
* of DSP0236 v1.3.1.
*/
rc = query_get_endpoint_id(ctx, &peer->phys, &peer->recovery.eid,
&peer->recovery.endpoint_type,
&peer->recovery.medium_spec, /*peer=*/NULL,
/*retry=*/false);
if (rc < 0) {
goto reschedule;
}
if (peer->gateway) {
/*
* For downstream endpoint recovery :
*
* If EID probe fails, check if the bridges are still responsive by walking up the bridge chain
* diagnosing loss of connectivity.
* In case we've got a response, verify UUID to detect device exchange at the same pool EID.
* for non matching UUID, remove old peer and publish the new peer with the same EID.
*/

@jk-ozlabs jk-ozlabs Sep 1, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Minor things:

  • please keep the wrapping consistent with the rest of the file
  • diagnosting → diagnosting

not clear on what "point of exchange" is supposed to indicate though, perhaps "diagnosing loss of connectivity" ?

@faizana-nvidia faizana-nvidia Sep 1, 2026

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ack, will do the wrapping.

not clear on what "point of exchange" is supposed to indicate though, perhaps "diagnosing loss of connectivity" ?

yeah this sounds better, I'll update


/*
* If we've got a response there are two scenarios:
*
* 1. The device responds with the EID that we expect it to have
* 2. The device responds with an unexpected EID, e.g. 0
*
* For scenario 1 we're done as the device is responsive and has the expected
* address. For scenario 2, we may not yet consider the EID assignment as
* expired, so check the UUID for a match. If the UUID matches we reassign the
* expected EID to the device. If the UUID does not match we allocate a new
* EID for the exchanged device, given it is responsive.
*/
if (peer->recovery.eid != peer->eid) {
static const uint8_t nil_uuid[16] = { 0 };
bool uuid_matches_peer = false;
bool uuid_matches_nil = false;
uint8_t uuid[16] = { 0 };
mctp_eid_t new_eid;

rc = query_get_peer_uuid_by_phys(ctx, &peer->phys, uuid);
if (!rc && peer->uuid) {
rc = query_get_endpoint_id(ctx, &peer->phys,
&peer->recovery.eid,
&peer->recovery.endpoint_type,
&peer->recovery.medium_spec, peer,
/*retry=*/false);
if (rc < 0) {
struct peer *gw = peer->gateway;
while (gw) {
int gw_rc = query_get_endpoint_id(
ctx, &gw->phys, &peer->recovery.eid,
&peer->recovery.endpoint_type,
&peer->recovery.medium_spec,
gw->gateway ? gw : NULL,
/*retry=*/false);
if (gw_rc < 0)
warnx("Recovery: EID %d unreachable; bridge EID %d failed to respond",
peer->eid, gw->eid);
gw = gw->gateway;
}
goto reschedule;
}

rc = query_get_peer_uuid(peer, uuid);
if (rc < 0)
goto reschedule;

if (peer->uuid) {
static_assert(sizeof(uuid) == sizeof(nil_uuid),
"Unsynchronized UUID sizes");
uuid_matches_peer =
Expand All @@ -3788,43 +3810,114 @@ static int peer_endpoint_recover(sd_event_source *s, uint64_t usec,
memcmp(uuid, nil_uuid, sizeof(uuid)) == 0;
}

if (rc || !uuid_matches_peer ||
if (!uuid_matches_peer ||
(uuid_matches_nil && !MCTPD_RECOVER_NIL_UUID)) {
/* It's not known to be the same device, allocate a new EID */
/*
* The bridge has re-assigned this EID to a new device.
* Remove the old peer and register the new one with the
* same EID.
*/
dest_phys phys = peer->phys;
mctp_eid_t eid = peer->eid;
uint32_t net = peer->net;
struct peer *gateway = peer->gateway;
struct peer *new_peer;

assert(sd_event_source_get_enabled(
peer->recovery.source, NULL) == 0);
remove_peer(peer);
/*
* The representation of the old peer is now gone. Set up the new peer,
* after which we immediately return as there's no old peer state left to
* maintain.
*/
return endpoint_assign_eid(ctx, NULL, &phys, &peer, 0,
false);
rc = add_peer(ctx, &phys, eid, net, &new_peer,
/*allow_bridged=*/true);
if (rc < 0)
return rc;

new_peer->gateway = gateway;
return setup_added_peer(new_peer);
}

/* Confirmation of the same device, apply its already allocated EID */
rc = endpoint_send_set_endpoint_id(peer, &new_eid, NULL);
} else {
// Directly-connected endpoint.
rc = query_get_endpoint_id(ctx, &peer->phys,
&peer->recovery.eid,
&peer->recovery.endpoint_type,
&peer->recovery.medium_spec,
/*peer=*/NULL, /*retry=*/false);
if (rc < 0) {
goto reschedule;
}

if (new_eid != peer->eid) {
rc = change_peer_eid(peer, new_eid);
if (rc < 0) {
goto reclaim;
/*
* If we've got a response there are two scenarios:
*
* 1. The device responds with the EID that we expect it to have
* 2. The device responds with an unexpected EID, e.g. 0
*
* For scenario 1 we're done as the device is responsive and has the expected
* address. For scenario 2, we may not yet consider the EID assignment as
* expired, so check the UUID for a match. If the UUID matches we reassign the
* expected EID to the device. If the UUID does not match we allocate a new
* EID for the exchanged device, given it is responsive.
*/

if (peer->recovery.eid != peer->eid) {
static const uint8_t nil_uuid[16] = { 0 };
bool uuid_matches_peer = false;
bool uuid_matches_nil = false;
uint8_t uuid[16] = { 0 };
mctp_eid_t new_eid;

rc = query_get_peer_uuid_by_phys(ctx, &peer->phys,
uuid);
if (!rc && peer->uuid) {
static_assert(sizeof(uuid) == sizeof(nil_uuid),
"Unsynchronized UUID sizes");
uuid_matches_peer = memcmp(uuid, peer->uuid,
sizeof(uuid)) == 0;
uuid_matches_nil = memcmp(uuid, nil_uuid,
sizeof(uuid)) == 0;
}

if (rc || !uuid_matches_peer ||
(uuid_matches_nil && !MCTPD_RECOVER_NIL_UUID)) {
/* It's not known to be the same device, allocate a new EID */
dest_phys phys = peer->phys;

assert(sd_event_source_get_enabled(
peer->recovery.source, NULL) ==
0);
remove_peer(peer);
/*
* The representation of the old peer is now
* gone. Set up the new peer, after which we
* immediately return as there's no old peer
* state left to maintain.
*/
return endpoint_assign_eid(ctx, NULL, &phys,
&peer, 0, false);
}
/* change_peer_eid() leaves the peer unpublished; the
* object path moves with the EID. The properties were
* queried when the peer was first set up, so
* publishing here re-registers the full interface
* set. */
rc = publish_peer(peer);

/* Confirmation of the same device, apply its already allocated EID */
rc = endpoint_send_set_endpoint_id(peer, &new_eid,
NULL);
if (rc < 0) {
goto reschedule;
}

if (new_eid != peer->eid) {
rc = change_peer_eid(peer, new_eid);
if (rc < 0) {
goto reclaim;
}
/* change_peer_eid() leaves the peer unpublished; the
* object path moves with the EID. The properties were
* queried when the peer was first set up, so
* publishing here re-registers the full interface
* set. */
rc = publish_peer(peer);
if (rc < 0) {
goto reschedule;
}
}
}
}

Expand Down Expand Up @@ -6225,6 +6318,8 @@ static int peer_endpoint_poll(sd_event_source *s, uint64_t usec, void *userdata)
&peer, true);
if (rc < 0)
goto exit;
// TODO: Need to add Query HOP logic to update immediate bridge peer's gateway chain
peer->gateway = bridge;
}

rc = setup_added_peer(peer);
Expand Down
Loading
Loading