That's good to hear
Mike.
Unfortunately
I can not share your excitement. I recompiled npf.sys from Winpcap 3.0 alpha 2
(Sept.19) with VC++ 6.0 and tried it on multiple Dell Latitude 2.0 Mhz (both
win2K/sp6, server and XP/sp1pro ). I have Sender and Receiver both using
npf.sys. Sender sends using NPF_BufferedSend. Receiver crashes unpredictably,
mostly with 0x0A: IRQL_NOT_LESS_OR_EQUAL. Depending on a particular box,
combination of driver's buffer, read timeout, and minimum to copy sizes,
receiver may run very long time (I've seen 68 hours) with very heavy load (70-90
MBps), or may crash immediately after starting receiving, or somewhere in the
middle of the test. I am trying to find the bug using driver verifier - no
success so far.
Igor
-----Original Message-----
From: Michael Montague [mailto:[EMAIL PROTECTED]]
Sent: Friday, October 04, 2002 4:02 AM
To: [EMAIL PROTECTED]
Subject: [WinPcap-users] SMP Safe npf.sysOn a whim I installed the Windows 2000 DDK, downloaded the winpcap source code and went to work on making a version of the driver that was safe for SMP machines. (All my dev boxes are dual processor machines, and using winpcap means my custom packet sniffers will work on windows and linux.)This is the first time I've ever dealt with device drivers in my life. On the other hand multithreading and low level code are things I am familiar with. That's a disclaimer for the faint at heart.I am currently running four copies of windump on a Dual P4 Xeon 1.7Mhz Dell Precision Workstation running Windows 2000 SP3. I'm going to keep them running for at least 24 hours. If they don't blue screen my machine by that time I'd love to find a way to get more people to test this. Also, how would I get these changes merged back into the main branch?For reference the source code I used was Winpcap 3.0 alpha 2. The only changes I made were wrapping all accesses to global variables with spin locks.- Mike
