Class 0 granted
(highest priority)
Class 1 granted
Class 2 granted
Class 1 granted
Class 2 granted
C0-C1-C2
C0-C2-C1
SMS_INTERCLASS_ARBITER[7:0] CLASS1PRIO field:
defines the number M of transactions dedicated to class 1.
SMS_INTERCLASS_ARBITER[23:16] CLASS2PRIO field:
defines the number N of transactions dedicated to class 2.
PWM arbitration
sdrc-026
Public Version
SDRAM Controller (SDRC) Subsystem
www.ti.com
10.2.6.2.2.2 Priority Between Groups
The SMS.SMS_CLASS_ARBITERi[7:6] HIGHPRIOVECTOR field defines the priority between groups in a
class.
10.2.6.2.2.3 Priority Between Classes
Class 0 always has the highest priority. Then, a weight is programmed within the arbitration of Class 1 and
Class 2:
•
The SMS.
[23:16] CLASS1PRIO field specifies the number (M) of
transactions dedicated to Class 1.
•
The SMS.
[7:0] CLASS2PRIO field specifies the number (N) of
transactions dedicated to Class 2.
M and N parameters are used to set a PWM arbitration. For instance, two schemes appear: at time t,
Class 1 is serviced for M cycles, and is directly followed by service to Class 2 for N cycles. If Class 2 is
serviced first for N cycles, then Class 1 is granted for M cycles. Arbitration is given more importance;
Class 1/2 is favored over Class 2/1. As shown in
, two arbitration schemes can appear:
•
Class 0 is serviced first, followed by Class 1, and then Class 2 (C0 - C1 - C2).
•
Class 0 is serviced first, followed by Class 2, and then Class 1 (C0 - C2 - C1).
Figure 10-69. Priority Between Classes
10.2.6.2.3 Arbitration Granularity
Arbitration is the mechanism that give more or less importance to incoming requests depending on the
initiator origin. Granularity influences the boundary of a transaction because it imposes the following
questions:
•
How many requests to service?
•
How long to track the current transaction?
•
When to make the next arbitration decision, or when does the next transaction boundary occur?
10.2.6.2.3.1 Idle Cycle
The idle gives more granularity to the arbitration; it lets the arbiter wait one idle cycle before moving the
arbitration grant to another thread.
Within a burst (see
for an example):
2288
Memory Subsystem
SWPU177N – December 2009 – Revised November 2010
Copyright © 2009–2010, Texas Instruments Incorporated
