diff --git a/content/notes/23-IO-systems.md b/content/notes/23-IO-systems.md
index 84041ce02..13c785337 100644
--- a/content/notes/23-IO-systems.md
+++ b/content/notes/23-IO-systems.md
@@ -8,13 +8,12 @@ tags:
 
 interactive between device drivers and IO devices
 
-# Hardware
+Hardware
 - IO device linked to machine via Port
 - link is a set of wires called a Bus
 - as each end of the link is a device controller (basically a process)
 - ![diagram|400](https://i.imgur.com/qnmTxrb.png)
-
-![more complete diagram of PCI bus|400](https://i.imgur.com/fbJqpwr.png)
+- ![more complete diagram of PCI bus|400](https://i.imgur.com/fbJqpwr.png)
 
 device controller registers
 - each contoller has registers for holding signals
@@ -26,6 +25,32 @@ adressesing devices
 - CPU can issue IO instructions to particular addresses
 - communication can be 1 byte at a time for via direct memory access
 
-# Kernel interface
+polling
+- CPU cycles waiting for busy bit to be clear is known as polling (busy waiting)
+- waste CPU time
+- if CPU is polling for input from devices too, then all devices have to be monitored - this is big waste of CPU time
 
-# Performance
+interupts
+- used for communcation with IO devices: prefered over polling
+- CPU hardware contains wire calld interrupt request line
+- if cpu senses this line after each instruction it executes
+- if intr sensed, CPU saves its current location and jumps to address in the interrupt vector indicated by the signal
+- maskable/nonmaskable
+	- deal with different priorities of interrupt
+	- maskable: device generated, can be diabled when OS is processing a more urgent intr
+	- nonmaskable: for error messges, never disabled
+
+Direct memory access
+- sending large amounts of data to IO devices
+- special purpose process DMA controller used
+- CPU writes a DMA command block into memory, specifying which bytes to transfer, and where to put them
+- DMA then controller can then execute concurrently
+- after transfer is finsished DMA controller interrupts CPU
+
+IO performance
+- reduce context switches
+- reduce data copies in memory (normamly two: user mem)
+- reduce number of interrupts (e.g., using polling)
+- increase concurrency (e.g., DMA)
+- move data processing to hardware (e.g., network interface card)
+- balance CPU, memory system, bus, IO performance
\ No newline at end of file