0. Quark SOC and Galileo Architecture (ESP – Fall 2014)
Computer Science & Engineering Department Arizona State University Tempe, AZ 85287
Dr. Yann-Hang Lee (480)

1. Current Processor Design
Moore’s law continues to hold true, transistor counts doubling every 18 months
But can no longer rely upon increasing clock rates and instruction-level parallelism to meet computing performance demands
Semiconductor device fabrication process
65 nm – 2006, 45 nm – 2008, 32 nm – 2010, and 22 nm – 2012
How to best exploit ever-increasing on-chip transistor counts?
Multi- & many-core (MC) devices are new technology wave
exploiting explicit parallelism in the new devices
Size and Power constraints

2. Intel Processors X86 32/64 architecture Processors for Differences
486 – first pipelined x86 design
Pentium – the first x86 superscalar CPU
Processors for
Server (Xeon), desktop (Core i3/i5/i7), mobile (Core i3/i5/i7), and embedded (Atom)
All of them support hypervisor (VM)
Differences
CPUs, memory, and interconnection bandwidth
reliability (quality of dies) and form factor
power and thermal requirements
Uses available clock cycles and power, not to push up higher clock speeds and energy needs

3. Galileo Board 400MHz Quark SoC 256MB DDR3 Ethernet USB Host Port
MicroSD Support
I2C, SPI Support
PCI Express Mini Cards
Serial Connectivity
GPIO
Linux on Board
Source:

4. Intel Quark SoC X1000. SOC – Chip size, power and pins CPU core (x86)
cache, internal memory (flash, SRAM)
IO interfaces and external buses
interconnection or switches
misc (clock, JTAG)
Chip size, power and pins
32nm process in 1st Quark
one-fifth the size and
one-tenth the power of
low-end Atom chip
393 solder balls on 15mm2
5 power rails (3.3V, 1.8V,
1.5V, 1.05V, 1.0V)

5. Pins in Quark In Galileo schematics
Example: High Speed UART Interface, SIU1_RDX SIU1_TXD
Six different power states
S0 – the system is completely powered ON and fully operational
S5 – the system is completely powered OFF
S1, S2, S3 and S4 – sleeping states, the system appears OFF because of low power consumption and retains enough of the hardware context to return to the working state
In Galileo schematics
Default Buffer State
Signal Name
Dir
Term
Power
Type
S4/S5 S3
Reset
Enter S0
SIU0_RXD I
20k(H)
3.3V
CMOS3.3
Off
Pull-up
SIU0_TXD O -
VOH

6. Quark Core Internal Architecture
32-bit RISC integer core
Single cycle execution
Instruction pipelining
Floating-point unit
Cache with cache consistency support (16-Kbyte for both data and instructions)
Memory management unit

7. 486 Pipeline

8. IO Expander and GPIO Multiplexing
CY8C9540A – I2C interfaced expander
with 40 I/O data pins (ports 0-5) independently configurable as inputs, outputs, bi-directional input/outputs, or PWM outputs
To configure a pin
an I2C control message to the chip which
includes a register address

9. X86 ISA Data Representations
Little-endian byte ordering in memory
Words, doublewords, and quadwords do not need to be aligned in memory on natural boundaries.
2 memory accesses for an unaligned memory access
aligned accesses require only one
Unsigned integer, signed (two's complement)
FP, string of bits, bytes, .. etc.
SIMD packed data
Pointer
Near
Far (logical)

10. Memory Model
Flat memory model – a single, continuous linear address space of 232 bytes
Segmented model – a logical address consisting of a segment selector and an offset
Real-address mode – for 8086,
16 segments of 64K
Linear address space 
(paging) physical space

11. Modes of Operation Protected mode (32 bits address)
native mode (Windows, Linux), full features, separate memory
virtual-8086 mode
Real-address mode (20 bits address)
the programming environment of the Intel 8086 processor with extensions
native MS-DOS
System management mode
power management, system security, diagnostics
IA-32e (Intel 64 architecture)
Compatibility mode – similar to 32-bit protected mode
64-bit mode –
16 64-bit general purpose registers
default address size is 64 bits and its default operand size is 32 bits.

12. Programmer’s model

13. Protected Mode Memory Management
Use segment descriptor to protect memory accesses
Each program has a descriptor table to map segments
allow shared segments
Memory access checks
Limit, type, privilege level checks.
Restrictions of addressable domain,
procedure entry-points,
and instruction set.

14. Virtual Memory and Paging
uses disk as part of the memory, thus allowing sum of all programs can be larger than physical memory
Only part of a program must be kept in memory, while the remaining parts are kept on disk.
The memory used by the program is divided into small units called pages (4096-byte).
OS maintains page directory and page tables
Page translation: CPU converts the linear address into a physical address
Page fault: occurs when a needed page is not in memory, and the CPU interrupts the program
Virtual memory manager (VMM) – OS utility that manages the loading and unloading of pages

15. Page Translation
A linear address is divided into a page directory field, page table field, and page frame offset.
The CPU uses all three to calculate the physical address.

16. Interrupt and Exception
an asynchronous event that is typically triggered by an I/O device.
Exception
a synchronous event that is generated when the processor detects one or more predefined conditions while executing an instruction.
three classes of exceptions: faults, traps, and aborts.
18 predefined interrupts and exceptions and 224 user defined interrupts
Access handler procedures through entries in the interrupt descriptor table (IDT)
A call to a handler procedure is similar to a procedure call to another protection level

17. Interrupt and Exception
Interrupt vector references
an interrupt gate (interrupt enable (IF) flag in the EFLAGS register is cleared)
a trap gate
Gate contains
access rights information
segment selector for the code segment of the handler procedure
an offset into the code segment to entry point of the handler procedure

18. Interrupt and APIC Interrupt in 8086 APIC Two pins: NMI and INTR
Interrupt Acknowledge Cycle to
fetch the interrupt vector number
from 8259
APIC
In Pentium and P6 processors
Receives interrupts and send to core for handling
APIC bus: bi-directional data signals (APICD[1:0]) and clock (APICCLK)
Inter-processor interrupt messages for multi-processor systems
static and dynamic (based on the priority of executing tasks) distribution

19. Interrupt Handling IO APIC delivers interrupt message to local APIC
Programmable vector number for each interrupt source
Implied priority based on vector number
local APIC determines when to service the interrupt relative to the other activities of the
processor
priority = vector / 16
Locate gate from IDT
Far call to the handler
(SS, ESP), EFLAGS, CS, EIP, and Error
code are saved in stack

20. Hardware Initialization and Reset
Reset processor state
EIP=0000FFF0H, CS=F000H(segment) and FFFF0000H (base)
Disable paging, cache, and in real-address mode
Execute the first instruction at physical address FFFFFFF0H.
The EPROM containing the software initialization code or BIOS should be located at the upper memory space (including this address)
Run in real-mode, invalidate the TLBs, set up a GDT for selector 0x08 (code) and 0x10 (data), switch to protected mode
Start other components on motherboard (FPU, APIC, southbridge, etc.)

21. Typical x86 System Architecture
Processor
Host Bus (PSB) 100/133/200MHz 64-bit
HubLink Bus
PCI Bus 33 MHz 32-bit
AGP Bus
System Memory
Audio
USB
LAN
IDE
Keybrd
Mouse
Floppy
Serial
Parallel
Clock Gen
Host Clock
PCI Clock
USB Clock
Hublink Clock
LPC Bus
SM Bus
CNR
SIO
South Bridge (ICH)
North Bridge (MCH)
FWH
Chipset
North Bridge
South Bridge
Firmware Hub
Various chipsets available from Intel to meet performance requirements
FSB, DMI/Hub interface
System control hub (SCH) – GMCH and ICH are merged into one chip

22. Host Bridge in Quark
A central hub that routes transactions to and from Quark CPU core, DRAM controller, and other functional blocks.
CPU core  PCI devices
via MMIO and IO accesses