diff --git a/flake.lock b/flake.lock
index d355ba6..ff735d5 100644
--- a/flake.lock
+++ b/flake.lock
@@ -3,19 +3,19 @@
     "artiq": {
       "inputs": {
         "artiq-comtools": "artiq-comtools",
-        "mozilla-overlay": "mozilla-overlay",
         "nixpkgs": "nixpkgs",
+        "rust-overlay": "rust-overlay",
         "sipyco": "sipyco",
         "src-migen": "src-migen",
         "src-misoc": "src-misoc",
         "src-pythonparser": "src-pythonparser"
       },
       "locked": {
-        "lastModified": 1717638354,
-        "narHash": "sha256-eyI8OsOrn/j8ChbCpyFpS5VXBW8xSNGGIboFGQj/d4I=",
+        "lastModified": 1720514025,
+        "narHash": "sha256-eU9Lux5olHTxRCwSiApKSPEdNhrlZ2OJ9ZqSofqp9Uk=",
         "ref": "refs/heads/master",
-        "rev": "ebc1e3fb767d6c0eac6eac20c3afeaba2ab70d1a",
-        "revCount": 8833,
+        "rev": "7a50afd9a96be8e6510d5c952808f39f0b74202a",
+        "revCount": 8887,
         "type": "git",
         "url": "https://github.com/m-labs/artiq.git"
       },
@@ -100,29 +100,13 @@
         "type": "github"
       }
     },
-    "mozilla-overlay_3": {
-      "flake": false,
-      "locked": {
-        "lastModified": 1704373101,
-        "narHash": "sha256-+gi59LRWRQmwROrmE1E2b3mtocwueCQqZ60CwLG+gbg=",
-        "owner": "mozilla",
-        "repo": "nixpkgs-mozilla",
-        "rev": "9b11a87c0cc54e308fa83aac5b4ee1816d5418a2",
-        "type": "github"
-      },
-      "original": {
-        "owner": "mozilla",
-        "repo": "nixpkgs-mozilla",
-        "type": "github"
-      }
-    },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1717281328,
-        "narHash": "sha256-evZPzpf59oNcDUXxh2GHcxHkTEG4fjae2ytWP85jXRo=",
+        "lastModified": 1720386169,
+        "narHash": "sha256-NGKVY4PjzwAa4upkGtAMz1npHGoRzWotlSnVlqI40mo=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "b3b2b28c1daa04fe2ae47c21bb76fd226eac4ca1",
+        "rev": "194846768975b7ad2c4988bdb82572c00222c0d7",
         "type": "github"
       },
       "original": {
@@ -135,10 +119,31 @@
     "root": {
       "inputs": {
         "artiq": "artiq",
-        "mozilla-overlay": "mozilla-overlay_2",
+        "mozilla-overlay": "mozilla-overlay",
         "zynq-rs": "zynq-rs"
       }
     },
+    "rust-overlay": {
+      "inputs": {
+        "nixpkgs": [
+          "artiq",
+          "nixpkgs"
+        ]
+      },
+      "locked": {
+        "lastModified": 1720491570,
+        "narHash": "sha256-PHS2BcQ9kxBpu9GKlDg3uAlrX/ahQOoAiVmwGl6BjD4=",
+        "owner": "oxalica",
+        "repo": "rust-overlay",
+        "rev": "b970af40fdc4bd80fd764796c5f97c15e2b564eb",
+        "type": "github"
+      },
+      "original": {
+        "owner": "oxalica",
+        "repo": "rust-overlay",
+        "type": "github"
+      }
+    },
     "sipyco": {
       "inputs": {
         "nixpkgs": [
@@ -163,11 +168,11 @@
     "src-migen": {
       "flake": false,
       "locked": {
-        "lastModified": 1715484909,
-        "narHash": "sha256-4DCHBUBfc/VA+7NW2Hr0+JP4NnKPru2uVJyZjCCk0Ws=",
+        "lastModified": 1720332047,
+        "narHash": "sha256-FdYVEHVtXHrzPhBqpXOTo9uHQAtuCsDPmAPY8JrfHOY=",
         "owner": "m-labs",
         "repo": "migen",
-        "rev": "4790bb577681a8c3a8d226bc196a4e5deb39e4df",
+        "rev": "60739a161e64630ce7ba62d1a5bac1252b66c3b9",
         "type": "github"
       },
       "original": {
@@ -227,18 +232,18 @@
     },
     "zynq-rs": {
       "inputs": {
-        "mozilla-overlay": "mozilla-overlay_3",
+        "mozilla-overlay": "mozilla-overlay_2",
         "nixpkgs": [
           "artiq",
           "nixpkgs"
         ]
       },
       "locked": {
-        "lastModified": 1716889070,
-        "narHash": "sha256-w+M5NfukmcMTt73+u6Cmltjn4qpQH8bcoTGskB2IzBE=",
+        "lastModified": 1717654016,
+        "narHash": "sha256-y/c0EZNDNlxb/yLy/D23X9PLoiQ8I9mXAA0zsVOy2t8=",
         "ref": "refs/heads/master",
-        "rev": "51b8111e799ebe90ca231ae4d02f84ede32632d5",
-        "revCount": 646,
+        "rev": "0efbbe39fe643c03f15e29c208bff3247a987766",
+        "revCount": 647,
         "type": "git",
         "url": "https://git.m-labs.hk/m-labs/zynq-rs"
       },
diff --git a/src/gateware/kasli_soc.py b/src/gateware/kasli_soc.py
index 45183ad..98d6f41 100755
--- a/src/gateware/kasli_soc.py
+++ b/src/gateware/kasli_soc.py
@@ -560,7 +560,10 @@ class GenericSatellite(SoCCore):
         self.submodules.local_io = SyncRTIO(
             self.rtio_tsc, self.rtio_channels, lane_count=description["sed_lanes"]
         )
-        self.comb += self.drtiosat.async_errors.eq(self.local_io.async_errors)
+        self.comb += [
+            self.drtiosat.async_errors.eq(self.local_io.async_errors),
+            self.local_io.sed_spread_enable.eq(self.drtiosat.sed_spread_enable.storage)
+        ]
 
         self.submodules.cri_con = rtio.CRIInterconnectShared(
             [self.drtiosat.cri, self.rtio_dma.cri, self.rtio.cri],
diff --git a/src/gateware/zc706.py b/src/gateware/zc706.py
index 92c9208..395d6c6 100755
--- a/src/gateware/zc706.py
+++ b/src/gateware/zc706.py
@@ -487,6 +487,10 @@ class _SatelliteBase(SoCCore):
         self.csr_devices.append("rtio_dma")
 
         self.submodules.local_io = SyncRTIO(self.rtio_tsc, rtio_channels)
+        self.comb += [
+            self.drtiosat.async_errors.eq(self.local_io.async_errors),
+            self.local_io.sed_spread_enable.eq(self.drtiosat.sed_spread_enable.storage)
+        ]
         self.submodules.cri_con = rtio.CRIInterconnectShared(
             [self.drtiosat.cri, self.rtio_dma.cri, self.rtio.cri],
             [self.local_io.cri] + self.drtio_cri,
diff --git a/src/libksupport/src/kernel/api.rs b/src/libksupport/src/kernel/api.rs
index a25d979..836e81e 100644
--- a/src/libksupport/src/kernel/api.rs
+++ b/src/libksupport/src/kernel/api.rs
@@ -303,6 +303,7 @@ pub fn resolve(required: &[u8]) -> Option<u32> {
         api_libm_f64f64f64!(nextafter),
         api_libm_f64f64f64!(pow),
         api_libm_f64f64!(round),
+        api_libm_f64f64!(rint),
         api_libm_f64f64!(sin),
         api_libm_f64f64!(sinh),
         api_libm_f64f64!(sqrt),
diff --git a/src/runtime/src/comms.rs b/src/runtime/src/comms.rs
index 7747850..3ee3201 100644
--- a/src/runtime/src/comms.rs
+++ b/src/runtime/src/comms.rs
@@ -771,7 +771,7 @@ pub fn main(timer: GlobalTimer, cfg: Config) {
     #[cfg(has_drtio_routing)]
     drtio_routing::interconnect_disable_all();
 
-    rtio_mgt::startup(&aux_mutex, &drtio_routing_table, &up_destinations, timer);
+    rtio_mgt::startup(&aux_mutex, &drtio_routing_table, &up_destinations, &cfg, timer);
     ksupport::setup_device_map(&cfg);
 
     analyzer::start(&aux_mutex, &drtio_routing_table, &up_destinations, timer);
diff --git a/src/runtime/src/rtio_mgt.rs b/src/runtime/src/rtio_mgt.rs
index 60e2fbb..8644e90 100644
--- a/src/runtime/src/rtio_mgt.rs
+++ b/src/runtime/src/rtio_mgt.rs
@@ -3,7 +3,9 @@ use core::cell::RefCell;
 
 use libboard_artiq::{drtio_routing, drtio_routing::RoutingTable, pl::csr};
 use libboard_zynq::timer::GlobalTimer;
+use libconfig::Config;
 use libcortex_a9::mutex::Mutex;
+use log::{info, warn};
 
 #[cfg(has_drtio)]
 pub mod drtio {
@@ -898,12 +900,36 @@ pub mod drtio {
     pub fn reset(_aux_mutex: Rc<Mutex<bool>>, _routing_table: &RoutingTable, mut _timer: GlobalTimer) {}
 }
 
+fn toggle_sed_spread(val: u8) {
+    unsafe {
+        csr::rtio_core::sed_spread_enable_write(val);
+    }
+}
+
+fn setup_sed_spread(cfg: &Config) {
+    if let Ok(spread_enable) = cfg.read_str("sed_spread_enable") {
+        match spread_enable.as_ref() {
+            "1" => toggle_sed_spread(1),
+            "0" => toggle_sed_spread(0),
+            _ => {
+                warn!("sed_spread_enable value not supported (only 1, 0 allowed), disabling by default");
+                toggle_sed_spread(0)
+            }
+        };
+    } else {
+        info!("SED spreading disabled by default");
+        toggle_sed_spread(0);
+    }
+}
+
 pub fn startup(
     aux_mutex: &Rc<Mutex<bool>>,
     routing_table: &Rc<RefCell<RoutingTable>>,
     up_destinations: &Rc<RefCell<[bool; drtio_routing::DEST_COUNT]>>,
+    cfg: &Config,
     timer: GlobalTimer,
 ) {
+    setup_sed_spread(cfg);
     drtio::startup(aux_mutex, routing_table, up_destinations, timer);
     unsafe {
         csr::rtio_core::reset_phy_write(1);
diff --git a/src/satman/src/main.rs b/src/satman/src/main.rs
index bb41fda..422b260 100644
--- a/src/satman/src/main.rs
+++ b/src/satman/src/main.rs
@@ -12,6 +12,7 @@ extern crate io;
 extern crate ksupport;
 extern crate libboard_artiq;
 extern crate libboard_zynq;
+extern crate libconfig;
 extern crate libcortex_a9;
 extern crate libregister;
 extern crate libsupport_zynq;
@@ -38,6 +39,7 @@ use libboard_artiq::{drtio_routing, drtioaux,
 #[cfg(feature = "target_kasli_soc")]
 use libboard_zynq::error_led::ErrorLED;
 use libboard_zynq::{i2c::I2c, print, println, time::Milliseconds, timer::GlobalTimer};
+use libconfig::Config;
 use libcortex_a9::{l2c::enable_l2_cache, regs::MPIDR};
 use libregister::RegisterR;
 use libsupport_zynq::{exception_vectors, ram};
@@ -81,15 +83,37 @@ fn drtiosat_tsc_loaded() -> bool {
     }
 }
 
+fn toggle_sed_spread(val: u8) {
+    unsafe {
+        csr::drtiosat::sed_spread_enable_write(val);
+    }
+}
+
 #[cfg(has_drtio_routing)]
 macro_rules! forward {
-    ($routing_table:expr, $destination:expr, $rank:expr, $repeaters:expr, $packet:expr, $timer:expr) => {{
+    (
+        $router:expr,
+        $routing_table:expr,
+        $destination:expr,
+        $rank:expr,
+        $self_destination:expr,
+        $repeaters:expr,
+        $packet:expr,
+        $timer:expr
+    ) => {{
         let hop = $routing_table.0[$destination as usize][$rank as usize];
         if hop != 0 {
             let repno = (hop - 1) as usize;
             if repno < $repeaters.len() {
                 if $packet.expects_response() {
-                    return $repeaters[repno].aux_forward($packet, $timer);
+                    return $repeaters[repno].aux_forward(
+                        $packet,
+                        $router,
+                        $routing_table,
+                        $rank,
+                        $self_destination,
+                        $timer,
+                    );
                 } else {
                     return $repeaters[repno].aux_send($packet);
                 }
@@ -102,7 +126,16 @@ macro_rules! forward {
 
 #[cfg(not(has_drtio_routing))]
 macro_rules! forward {
-    ($routing_table:expr, $destination:expr, $rank:expr, $repeaters:expr, $packet:expr, $timer:expr) => {};
+    (
+        $router:expr,
+        $routing_table:expr,
+        $destination:expr,
+        $rank:expr,
+        $self_destination:expr,
+        $repeaters:expr,
+        $packet:expr,
+        $timer:expr
+    ) => {};
 }
 
 fn process_aux_packet(
@@ -183,6 +216,10 @@ fn process_aux_packet(
                             &drtioaux::Packet::DestinationStatusRequest {
                                 destination: destination,
                             },
+                            router,
+                            _routing_table,
+                            *rank,
+                            *self_destination,
                             timer,
                         ) {
                             Ok(()) => (),
@@ -244,7 +281,16 @@ fn process_aux_packet(
             channel,
             probe,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             let value;
             #[cfg(has_rtio_moninj)]
             unsafe {
@@ -266,7 +312,16 @@ fn process_aux_packet(
             overrd,
             value,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             #[cfg(has_rtio_moninj)]
             unsafe {
                 csr::rtio_moninj::inj_chan_sel_write(channel as _);
@@ -280,7 +335,16 @@ fn process_aux_packet(
             channel,
             overrd,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             let value;
             #[cfg(has_rtio_moninj)]
             unsafe {
@@ -299,7 +363,16 @@ fn process_aux_packet(
             destination: _destination,
             busno: _busno,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             let succeeded = i2c.start().is_ok();
             drtioaux::send(0, &drtioaux::Packet::I2cBasicReply { succeeded: succeeded })
         }
@@ -307,7 +380,16 @@ fn process_aux_packet(
             destination: _destination,
             busno: _busno,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             let succeeded = i2c.restart().is_ok();
             drtioaux::send(0, &drtioaux::Packet::I2cBasicReply { succeeded: succeeded })
         }
@@ -315,7 +397,16 @@ fn process_aux_packet(
             destination: _destination,
             busno: _busno,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             let succeeded = i2c.stop().is_ok();
             drtioaux::send(0, &drtioaux::Packet::I2cBasicReply { succeeded: succeeded })
         }
@@ -324,7 +415,16 @@ fn process_aux_packet(
             busno: _busno,
             data,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             match i2c.write(data) {
                 Ok(ack) => drtioaux::send(
                     0,
@@ -347,7 +447,16 @@ fn process_aux_packet(
             busno: _busno,
             ack,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             match i2c.read(ack) {
                 Ok(data) => drtioaux::send(
                     0,
@@ -371,7 +480,16 @@ fn process_aux_packet(
             address,
             mask,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             let ch = match mask {
                 //decode from mainline, PCA9548-centric API
                 0x00 => None,
@@ -397,7 +515,16 @@ fn process_aux_packet(
             div: _div,
             cs: _cs,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             // todo: reimplement when/if SPI is available
             //let succeeded = spi::set_config(busno, flags, length, div, cs).is_ok();
             drtioaux::send(0, &drtioaux::Packet::SpiBasicReply { succeeded: false })
@@ -407,7 +534,16 @@ fn process_aux_packet(
             busno: _busno,
             data: _data,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             // todo: reimplement when/if SPI is available
             //let succeeded = spi::write(busno, data).is_ok();
             drtioaux::send(0, &drtioaux::Packet::SpiBasicReply { succeeded: false })
@@ -416,7 +552,16 @@ fn process_aux_packet(
             destination: _destination,
             busno: _busno,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             // todo: reimplement when/if SPI is available
             // match spi::read(busno) {
             //     Ok(data) => drtioaux::send(0,
@@ -436,7 +581,16 @@ fn process_aux_packet(
         drtioaux::Packet::AnalyzerHeaderRequest {
             destination: _destination,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             let header = analyzer.get_header();
             drtioaux::send(
                 0,
@@ -450,7 +604,16 @@ fn process_aux_packet(
         drtioaux::Packet::AnalyzerDataRequest {
             destination: _destination,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             let mut data_slice: [u8; SAT_PAYLOAD_MAX_SIZE] = [0; SAT_PAYLOAD_MAX_SIZE];
             let meta = analyzer.get_data(&mut data_slice);
             drtioaux::send(
@@ -471,7 +634,16 @@ fn process_aux_packet(
             length,
             trace,
         } => {
-            forward!(_routing_table, destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             *self_destination = destination;
             let succeeded = dma_manager.add(source, id, status, &trace, length as usize).is_ok();
             router.send(
@@ -492,7 +664,16 @@ fn process_aux_packet(
             id,
             succeeded,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             dma_manager.ack_upload(
                 kernel_manager,
                 source,
@@ -510,7 +691,16 @@ fn process_aux_packet(
             destination: _destination,
             id,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             let succeeded = dma_manager.erase(source, id).is_ok();
             router.send(
                 drtioaux::Packet::DmaRemoveTraceReply {
@@ -526,7 +716,16 @@ fn process_aux_packet(
             destination: _destination,
             succeeded: _,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             Ok(())
         }
         drtioaux::Packet::DmaPlaybackRequest {
@@ -535,7 +734,16 @@ fn process_aux_packet(
             id,
             timestamp,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             let succeeded = if !kernel_manager.running() {
                 dma_manager.playback(source, id, timestamp).is_ok()
             } else {
@@ -555,7 +763,16 @@ fn process_aux_packet(
             destination: _destination,
             succeeded,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             if !succeeded {
                 kernel_manager.ddma_nack();
             }
@@ -569,7 +786,16 @@ fn process_aux_packet(
             channel,
             timestamp,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             dma_manager.remote_finished(kernel_manager, id, error, channel, timestamp);
             Ok(())
         }
@@ -581,7 +807,16 @@ fn process_aux_packet(
             length,
             data,
         } => {
-            forward!(_routing_table, destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             *self_destination = destination;
             let succeeded = kernel_manager.add(id, status, &data, length as usize).is_ok();
             drtioaux::send(0, &drtioaux::Packet::SubkernelAddDataReply { succeeded: succeeded })
@@ -592,7 +827,16 @@ fn process_aux_packet(
             id,
             run,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             let mut succeeded = kernel_manager.load(id).is_ok();
             // allow preloading a kernel with delayed run
             if run {
@@ -617,7 +861,16 @@ fn process_aux_packet(
             destination: _destination,
             succeeded,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             // received if local subkernel started another, remote subkernel
             kernel_manager.subkernel_load_run_reply(succeeded);
             Ok(())
@@ -628,14 +881,32 @@ fn process_aux_packet(
             with_exception,
             exception_src,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             kernel_manager.remote_subkernel_finished(id, with_exception, exception_src);
             Ok(())
         }
         drtioaux::Packet::SubkernelExceptionRequest {
             destination: _destination,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             let mut data_slice: [u8; SAT_PAYLOAD_MAX_SIZE] = [0; SAT_PAYLOAD_MAX_SIZE];
             let meta = kernel_manager.exception_get_slice(&mut data_slice);
             drtioaux::send(
@@ -655,7 +926,16 @@ fn process_aux_packet(
             length,
             data,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             kernel_manager.message_handle_incoming(status, id, length as usize, &data);
             router.send(
                 drtioaux::Packet::SubkernelMessageAck { destination: source },
@@ -667,7 +947,16 @@ fn process_aux_packet(
         drtioaux::Packet::SubkernelMessageAck {
             destination: _destination,
         } => {
-            forward!(_routing_table, _destination, *rank, _repeaters, &packet, timer);
+            forward!(
+                router,
+                _routing_table,
+                _destination,
+                *rank,
+                *self_destination,
+                _repeaters,
+                &packet,
+                timer
+            );
             if kernel_manager.message_ack_slice() {
                 let mut data_slice: [u8; MASTER_PAYLOAD_MAX_SIZE] = [0; MASTER_PAYLOAD_MAX_SIZE];
                 if let Some(meta) = kernel_manager.message_get_slice(&mut data_slice) {
@@ -920,6 +1209,28 @@ pub extern "C" fn main_core0() -> i32 {
     #[cfg(has_si549)]
     si549::helper_setup(&mut timer, &SI549_SETTINGS).expect("cannot initialize helper Si549");
 
+    let cfg = match Config::new() {
+        Ok(cfg) => cfg,
+        Err(err) => {
+            warn!("config initialization failed: {}", err);
+            Config::new_dummy()
+        }
+    };
+
+    if let Ok(spread_enable) = cfg.read_str("sed_spread_enable") {
+        match spread_enable.as_ref() {
+            "1" => toggle_sed_spread(1),
+            "0" => toggle_sed_spread(0),
+            _ => {
+                warn!("sed_spread_enable value not supported (only 1, 0 allowed), disabling by default");
+                toggle_sed_spread(0)
+            }
+        };
+    } else {
+        info!("SED spreading disabled by default");
+        toggle_sed_spread(0);
+    }
+
     #[cfg(has_drtio_routing)]
     let mut repeaters = [repeater::Repeater::default(); csr::DRTIOREP.len()];
     #[cfg(not(has_drtio_routing))]
diff --git a/src/satman/src/repeater.rs b/src/satman/src/repeater.rs
index 9ebc9b7..7834007 100644
--- a/src/satman/src/repeater.rs
+++ b/src/satman/src/repeater.rs
@@ -208,10 +208,37 @@ impl Repeater {
         }
     }
 
-    pub fn aux_forward(&self, request: &drtioaux::Packet, timer: &mut GlobalTimer) -> Result<(), drtioaux::Error> {
+    pub fn aux_forward(
+        &self,
+        request: &drtioaux::Packet,
+        router: &mut Router,
+        routing_table: &drtio_routing::RoutingTable,
+        rank: u8,
+        self_destination: u8,
+        timer: &mut GlobalTimer,
+    ) -> Result<(), drtioaux::Error> {
         self.aux_send(request)?;
-        let reply = self.recv_aux_timeout(200, timer)?;
-        drtioaux::send(0, &reply).unwrap();
+        loop {
+            let reply = self.recv_aux_timeout(200, timer)?;
+            match reply {
+                // async/locally requested packets to be consumed or routed
+                // these may come while a packet would be forwarded
+                drtioaux::Packet::DmaPlaybackStatus { .. }
+                | drtioaux::Packet::SubkernelFinished { .. }
+                | drtioaux::Packet::SubkernelMessage { .. }
+                | drtioaux::Packet::SubkernelMessageAck { .. }
+                | drtioaux::Packet::SubkernelLoadRunReply { .. }
+                | drtioaux::Packet::SubkernelException { .. }
+                | drtioaux::Packet::DmaAddTraceReply { .. }
+                | drtioaux::Packet::DmaPlaybackReply { .. } => {
+                    router.route(reply, routing_table, rank, self_destination);
+                }
+                _ => {
+                    drtioaux::send(0, &reply).unwrap();
+                    break;
+                }
+            }
+        }
         Ok(())
     }