RE: [openib-general] Some Missing Features from mthca/user MAD ac cess

Eitan Zahavi Mon, 10 Jan 2005 12:10:23 -0800

Title: RE: [openib-general] Some Missing Features from mthca/user MAD access

I'm sorry if I did not explain myself very well.

I did not mean to say another application will respond to SMInfo.

Nor I claimed that SMInfo should be used for anything else but SM to SM discovery.

I only referred to the idea that one can set the SM bit by inspecting the mask of registrations to mad filtering requests.

I assume that one can register by supplying a mask. Now - you probably do not have any control over the mask an arbitrary application might use. So depending on it not to set the SMInfo bit (even though it will not process the mads) is too risky in my mind.

However, from your response it seems like only one client is allowed to register as the receiver for incoming requests of specific attribute. Is this correct? If this is the case then a careless application can block the SM from obtaining any SMInfo or other queries.

Eitan Zahavi

Design Technology Director

Mellanox Technologies LTD

Tel:+972-4-9097208
Fax:+972-4-9593245

P.O. Box 586 Yokneam 20692 ISRAEL

-----Original Message-----
From: shaharf [mailto:[EMAIL PROTECTED]
Sent: Monday, January 10, 2005 6:56 PM
To: Eitan Zahavi; Michael S. Tsirkin; Roland Dreier
Cc: [email protected]
Subject: RE: [openib-general] Some Missing Features from mthca/user MAD access

Eitan, the main (and only) purpose of the IS_SM bit is for SM to SM coordination. As a matter of fact it means that when discavering a port with this bit you have to query it with sminfo to see if he is the master or maybe it should be the master. There is not other use. Having an application other then the SM respond to the sminfo would not work in the current scheme. As a matter of fact it maybe catastrophic - meaning it may lead to multiple masters SM in the subnet!!!

I really don't understand what resources you refer to. I think I know the OpenSM pretty well but maybe I have new things to learn. Please enlighten me.

Ref-counting too is useless. There should be only one SM on the port. I see not reason to change that.

Shahar

From: Eitan Zahavi [mailto:[EMAIL PROTECTED]
Sent: Monday, January 10, 2005 6:43 PM
To: shaharf; Eitan Zahavi; Michael S. Tsirkin; Roland Dreier
Cc: [email protected]
Subject: RE: [openib-general] Some Missing Features from mthca/user MAD access

Shahar> Yes, but when you want to respond to attributes you have to specify a mask. Using that mask you can register to any attribute set you want.

Yes, but if the application (that is not an SM) will use a mask that has the sminfo set - it will be considered an SM.

I would prefer having the SMBit stay on when the SM dies then having spurious SMBit transitions.

It will take more resources from the SM when these bits will start to change.

Also you will need to start ref-counting on the port since several apps can share it and they will not obey the rule for not masking the sminfo if they are not SMs.