Tilelink spec query: 9.4. TL-UL and TL-UH Messages on Channel A and D

[Asutosh21]

Hi,

We have few doubts on Tilleink spec.

9.4. TL-UL and TL-UH Messages on Channel A and D

In addition to new messages, TL-C specifies implied permission transfers for these existing messages from TL-UL and TL-UH.

Get implicitly Cap the permissions as None (Invalid).
{PutFullData, PutPartialData, ArithmeticData, LogicalData} implicitly Cap the permissions as not Read+Write [Trunk] or [Tip], i.e. as None [Invalid] or Read [Branch].

From above section of the spec, we understand if a data block is already acquired and the TL-UH command is tried from one Core (i.e. M1) then it affects the permission of the Cache attached to M1.

Queries:

1. Suppose we have Acquired a block, then send Get for that block. As per my understanding if a data block is Acquired, with at least with Read permission and assuming this access is not released then that Master/Core should not get Read miss and so should not send Get command further. So in which case this scenario might arise? Please confirm whether my understanding correct?
2. With sending Transfer Access command from Master to Slave, does the permission automatically updated or any additional Probe command needed? Because as specified in spec the permission is capped.
3. Does it impact other Cores as well? If yes how would slave conveys? Does it send Probe command?
4. If we send PutFullData after a Acquire message with Read+Write permission, then the permission can be reduced to None or Read only. How we decide which permission to achieve? Who decides it whether Master or Slave?

Regards,

Asutosh

[Dolu1990]

Hi ^^

This is a good question.
I kind have the same issue as you, a few things doesn't seems to be specified.
May have more luck on https://lists.chipsalliance.org/g/interconnects-wg ?

[Dolu1990]

Ahh it isn't the same question.
The question on the google group is related to aquire BtoT / probe toN interaction.
This github issue seems more related to the interraction of block ownership / (get+put) emited from a given master agent.

[TommyMurphyTM1234]

Ahh it isn't the same question. The question on the google group is related to aquire BtoT / probe toN interaction. This github issue seems more related to the interraction of block ownership / (get+put) emited from a given master agent.

Apologies, I made a mistake and deleted my earlier post here to avoid confusion, but may have actually caused it! ☹️