• Locked Locked
  • Replies 3 replies
  • Subscribers 119 subscribers
  • Views 5219 views
  • Users 0 members are here
2025 survey

We are running a survey to help us improve the experience for all of our members. If you see the survey appear, please take the time to tell us about your experience if you can.

Options
Related
How was your experience today?
This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

AXI 3/AXI 4 Exclusive Access

Sai Karthik Madabhushi
Note: This was originally posted on 14th June 2011 at http://forums.arm.com

Hi,

I have a query  regarding Exclusive Access in AXI 3/ AXI 4. The spec does NOT say anything about this behavior, hence I need a clarification.

1. Can there be two Exclusive Reads monitoring the same address region?

   ARID = 0    ARADDR = 0x004   ARSIZE = 0   ARLEN = 2
   ARID = 3    ARADDR = 0x004   ARSIZE = 1   ARLEN = 1

** Lets assume the above ID's have come sequentially in subsequent cycles and NO exclusive write has come.

2. If YES, then How should the behavior of Exclusive Write be?
a. Exclusive Write of which ID should be entertained?
        b. Should we stop monitoring the OTHER ID when the exclusive write for ONE ID arrives?

3. In case of a successful Exclusive Write, is there a compulsion for the slave to send EXOKAY?
    a. Can it also send SLVERR, or DECERR?

Regards,
Sai Karthik
  • Note: This was originally posted on 15th June 2011 at http://forums.arm.com

    HI,

    This question is from the perspective of a CHECKER for AXI.
    The Spec says that " The Slave returns an EXOKAY response to the master".
    The thing I want to understand is,

    1. The slave MUST send an EXOKAY response
           OR

        Is there a chance that the slave can also send SLVERR response, due to ANY internal condition.

    2. Is it a protocol violation if the slave sends SLVERR instead of EXOKAY?

    Regards,
    Sai Karthik
  • Note: This was originally posted on 14th June 2011 at http://forums.arm.com


    I have a query  regarding Exclusive Access in AXI 3/ AXI 4. The spec does NOT say anything about this behavior, hence I need a clarification.

    AXI is only the carrier of the respective signals. Exclusive access is defined in the ARM ARM.


    1. Can there be two Exclusive Reads monitoring the same address region?

       ARID = 0    ARADDR = 0x004   ARSIZE = 0   ARLEN = 2
       ARID = 3    ARADDR = 0x004   ARSIZE = 1   ARLEN = 1

    Yes.

    2. If YES, then How should the behavior of Exclusive Write be?
    a. Exclusive Write of which ID should be entertained?
            b. Should we stop monitoring the OTHER ID when the exclusive write for ONE ID arrives?

    a. Either. Exclusive write is based on a first come first served strategy. The whole idea of exclusive access was to prevent bus contention caused by one master holding on to resources for a generous amount of time. If a specific order were required, perhaps even deadlocks could occur.
    b. Yes. All monitors of the same address (range) will have to be cleared.

    3. In case of a successful Exclusive Write, is there a compulsion for the slave to send EXOKAY?
        a. Can it also send SLVERR, or DECERR?

    This question is strange. Why would it want to return anything but EXOKAY for a successful exclusive write?

    Kindly,
    Marcus
  • Note: This was originally posted on 15th June 2011 at http://forums.arm.com


    1. The slave MUST send an EXOKAY response

    The slave will only send EXOKAY, if the transfer succeeded to indicate
    • The slave actually supports exclusive access
    • An exclusive write access has updated memory

    Is there a chance that the slave can also send SLVERR response, due to ANY internal condition.
    2. Is it a protocol violation if the slave sends SLVERR instead of EXOKAY?

    Indeed, the spec is not explicit and I see why there might be a doubt. My understanding is that the error responses SLVERR and DECERR may be returned for any transaction, both read and write. Otherwise we'd have to issue a normal access (r/w) to a semaphore first, to verify whether there isn't any error, followed by an exclusive read.

    Kindly
    Marcus