[Openib-windows] Debug prints in free version

[Tzachi Dar]

I guess that we all agree that the debug level should be "unprintable by
default, but to give some way to turn them on at run time". What I'm
suggesting is that this way will be to copy the debug versions of this
files. Do you fill that there is a need to make this simpler (for example
instead of asking the user to copy files we might ask him to make a change
in the registry or so).
 
About Fab suggestion to print into the event log. We have to choose
carefully what we print into the debug log (for example, errors on shutdown,
might be printed using debug prints, but not to the event log).
 
Thanks
Tzachi
 
-----Original Message-----
From: Leonid Keller 
Sent: Tuesday, August 09, 2005 2:23 PM
To: Yossi Leybovich; Tzachi Dar; 'Fab Tillier'
Cc: 'openib-windows at openib.org'
Subject: RE: [Openib-windows] Debug prints in free version
 
i agree with Yossy.
To my mind, errors have to be printed in release version also.
A compromise is to make them unprintable by default, but to give some way to
turn them on at run time.
-----Original Message-----
From: Yossi Leybovich [mailto:sleybo at mellanox.co.il]
Sent: Monday, August 08, 2005 3:42 PM
To: Tzachi Dar; 'Fab Tillier'
Cc: 'openib-windows at openib.org'
Subject: RE: [Openib-windows] Debug prints in free version
CB
-----Original Message-----
From: Tzachi Dar 
Sent: Monday, August 08, 2005 3:15 PM
To: Yossi Leybovich; Fab Tillier
Cc: openib-windows at openib.org
Subject: RE: [Openib-windows] Debug prints in free version
Printing (even only in the case of errors) will hurt performance for the
following reasons:
 
Today, these macros are being removed out at compile time. If they are there
in release, the check that is made if to print or not will still be
executed. 
[YL] not if you use them just in case of error (i.e if (status != IB_SUCCESS
) AL_TRACE(AL_DBG_ERROR......) and that the way most\all of the current code
do
UI also think that this theway VAPI\Gen2 behave.
Having these functions in our code also means bigger foot print of our
functions and therefore another hit in performance (for example more cache
misses).
[YL] That right but I think that if you check it, it want heart the
performance that much but you help the user a lot to define his mistakes or
give us log to debug our bugs
Please also note that your idea is not enough since the IPOIB code is using
the macro IPOIB_TRACE which should also be changed to exist in free also.
(There is a list of macros that should be changed).
[YL] AFAIK all Macros map to CL_PRINT so you just need to change in one
place on all the ULPs behave the same.
We might do a separate build that is free but still has the prints, but
debug messages should be omitted from the binaries that we give to our
customers.
[YL] I still think that error massages should free version it want help the
customer if will force him to reinsatll another version of the driver.
Thanks
Tzachi
 
-----Original Message-----
From: Yossi Leybovich [mailto:sleybo at mellanox.co.il] 
Sent: Monday, August 08, 2005 3:05 PM
To: Fab Tillier
Cc: openib-windows at openib.org
Subject: [Openib-windows] Debug prints in free version
 
Fab 
I think it will make it easy to all of us if the stack will generate printf
in case of error. 
In the current defines in order to debug any kernel level component you need
to replace the sys file to the debug version 
It will be better if : 
CL_PRINT in case of ERROR_DBG should use cl_msg_out so it will generate
print even in free version 
It should not heart the data flow (we print just in case of error) 
What do you think ? 
Yossi 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ofw/attachments/20050809/d695262a/attachment.html>