I have a question regarding the recvtiming() function in bus.cc. Please refer
to the following code snippet:
if (dest == Packet::Broadcast) {
port = findPort(pkt->getAddr(), pkt->getSrc());
if (timingSnoop(pkt, port ? port : interfaces[pkt->getSrc()])) {
bool success;
pkt->flags |= SNOOP_COMMIT;
success = timingSnoop(pkt, port ? port : .....
Does the destination field in packet is set to Broadcast each time a cache miss
occurs? If not, under what scenario is the destination of packet a Broadcast.
Looks like timingsnoop() is being called on events other than a cache miss !
Also, any ideas how I can track the information that the current packet needs
to occupy the bus because of an earlier cache miss.
---- Original message ----
>Date: Sun, 18 Nov 2007 16:23:00 -0800
>From: "Steve Reinhardt" <[EMAIL PROTECTED]>
>Subject: Re: [m5-users] Changing Bus Latency for Packet
>To: "M5 users mailing list" <[email protected]>
>
>Just to be clear: ticks are just ticks, the smallest discrete unit of
>time. They aren't processor cycles. It may just so happen that your
>processor cycle time is 1 tick, but that's not a given. Particularly
>in FS mode we use 1 tick == 1ps, and we've been trying to move there
>in SE mode too.
>
>The 'clock' field in the bus object is the number of ticks in a bus
>clock. You should not be adding values that are not a multiple of
>clock to tickNextIdle, as then the bus may start processing events at
>times that are not bus clock edges... this won't necessarily break
>things but it doesn't make sense from a timing perspective.
>
>I don't see any reason why your code change would lead to negative
>numbers; you should run under gdb to figure that kind of thing out.
>
>Steve
>
>On Nov 18, 2007 3:52 PM, Shoaib Akram <[EMAIL PROTECTED]> wrote:
>> The -ve number was an issue with something else ! However, if I add
>> busCycles, the traces are what is expected. But if I insert processor cycles
>> e.g., tickNextIdle+=78, the traces are not in accordance with what is
>> expected and even the total numebr of cycles are less than without adding
>> extra cycles. I am looking for the right way to insert processor cycles
>> (ticks rather than numCycles) for a certain packet?
>>
>>
>> ---- Original message ----
>> >Date: Sun, 18 Nov 2007 16:47:23 -0600 (CST)
>> >From: Shoaib Akram <[EMAIL PROTECTED]>
>> >Subject: [m5-users] Changing Bus Latency for Packet
>> >To: [email protected]
>> >
>> >I want to have extra bus cycles added for certain type of packets. If I add
>> >a busCycle, tickNextIdle+=clock, something anomalous happens. At one point
>> >after first 4000 cycles, the tickNextIdle gets calculated to a negative
>> >number, which the following trace on bus shows: Please suggest whats the
>> >right way to add busCycles. If I add processor cycles (tickNextIdle+=100),
>> >I dont have negative numbers but the traces show results not expected !
>> >
>> >1001: system.membus: recvTiming: packet src 1 dest -1 addr 0x140 cmd ReadReq
>> >1001: system.membus: found addr 0x140 on device 0
>> >1001: system.membus: The bus is now occupied from tick 1001 to 4000
>> >1002: system.membus: recvTiming: packet src 0 dest 1 addr 0x140 cmd ReadResp
>> >1002: Event_10318838: Waiting on retry
>> >4000: system.membus: Received a retry
>> >4000: system.membus: Sending a retry
>> >4000: system.membus: recvTiming: packet src 0 dest 1 addr 0x140 cmd ReadResp
>> >4000: system.membus: The bus is now occupied from tick 4000 to -243651592
>> >
>> >_______________________________________________
>> >m5-users mailing list
>> >[email protected]
>> >http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>> _______________________________________________
>> m5-users mailing list
>> [email protected]
>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>
>_______________________________________________
>m5-users mailing list
>[email protected]
>http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
