diff --git a/src/kernal/iec/iec_tx_byte.s b/src/kernal/iec/iec_tx_byte.s
index 014dca9e..ef0efffd 100644
--- a/src/kernal/iec/iec_tx_byte.s
+++ b/src/kernal/iec/iec_tx_byte.s
@@ -62,6 +62,14 @@ iec_tx_byte:
 	jsr kernalstatus_DEVICE_NOT_FOUND
 	jmp iec_return_DEVICE_NOT_FOUND
 @1:
+	; Timing is critical from here on - execute on disabled IRQs, from before
+	; CLK is released: an IRQ between the release and the EOI wait would hide
+	; the listener's EOI acknowledgement, and the transfer would hang
+	; The best practice would be to do PHP first, but it seems this is not
+	; what the original ROM does; furthermore, pushing additional bytes to
+	; stack can wreck autostart software loading at $0100
+	sei
+
 	; Notify all devices that we are going to send a byte
 	; and it is going to be a data byte (released ATN)
 	jsr iec_release_atn_clk_data
@@ -72,11 +80,7 @@ iec_tx_byte:
 
 iec_tx_common:
 
-	; Timing is critical here - execute on disabled IRQs
-	; The best practice would be to do PHP first, but it seems this is not
-	; what the original ROM does; furthermore, pushing additional bytes to
-	; stack can wreck autostart software loading at $0100
-	sei
+	; Entered with IRQs disabled, by iec_tx_byte and iec_tx_command
 
 	; Wait till all receivers are ready, they should all release DATA
 	jsr iec_wait_for_data_release
diff --git a/src/kernal/iec/iec_tx_command.s b/src/kernal/iec/iec_tx_command.s
index 3e4a223d..c8b19295 100644
--- a/src/kernal/iec/iec_tx_command.s
+++ b/src/kernal/iec/iec_tx_command.s
@@ -32,7 +32,10 @@ iec_tx_command:
 	; (can be busy processing something), we have to wait till they are all
 	; ready (or bored with DOS attack...)
 	; Release back CLK, keep DATA released - to ask devices for status.
+	; From here on IRQs are disabled, as in iec_tx_byte: a listener takes a
+	; talker that waits too long after the release for EOI
 
+	sei
 	jsr iec_release_clk_data
 
 	; Common part of iec_txbyte and iec_tx_common - waits for devices