From owner-freebsd-bugs Tue Aug 21 9:20: 7 2001 Delivered-To: freebsd-bugs@hub.freebsd.org Received: from freefall.freebsd.org (freefall.freebsd.org [216.136.204.21]) by hub.freebsd.org (Postfix) with ESMTP id CA33137B407 for ; Tue, 21 Aug 2001 09:20:01 -0700 (PDT) (envelope-from gnats@FreeBSD.org) Received: (from gnats@localhost) by freefall.freebsd.org (8.11.4/8.11.4) id f7LGK1s79960; Tue, 21 Aug 2001 09:20:01 -0700 (PDT) (envelope-from gnats) Date: Tue, 21 Aug 2001 09:20:01 -0700 (PDT) Message-Id: <200108211620.f7LGK1s79960@freefall.freebsd.org> To: freebsd-bugs@FreeBSD.org Cc: From: Ian Dowse Subject: Re: kern/29845: 4.4-PRERELEASE crashes under heavy net I/O Reply-To: Ian Dowse Sender: owner-freebsd-bugs@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.org The following reply was made to PR kern/29845; it has been noted by GNATS. From: Ian Dowse To: Warner Losh Cc: Andre Albsmeier , freebsd-gnats-submit@FreeBSD.org Subject: Re: kern/29845: 4.4-PRERELEASE crashes under heavy net I/O Date: Tue, 21 Aug 2001 17:13:26 +0100 In message <200108211539.f7LFdoW65851@harmony.village.org>, Warner Losh writes: >Here's a simple fix you can try. I don't see how this would help, but >if it does, we know what the problem is. Ian suggested this a while >ago, and I'm still not sure how this could be a problem, but if it is >Ian's suggestions are right. No, I was confused when I suggested this to you :-) It is too late when pcic_pci_intr() is called, because at that point a critical section of some network code has already been interrupted. Once a NIC has registered a net interrupt on IRQ X, splimp() should mask IRQ X, but here the pcic code never changes the interrupt mask when a NIC registers its interrupt. e.g. consider some network code that does splimp(): s = splimp(); (critical stuff where no net interrupts should occur) pcic_pci_intr() called s = splhigh(); (this blocks further interrupts) NIC ISR called (messes with splimp-protected state) splx(x); pcic_pci_intr() returns (network code finds its state messed up) splx(s); When the pccard NIC sets up its interrupt, it needs to go through all the mask adjustment behind bus_generic_setup_intr() to ensure that the first splimp() call above actually blocks the pcic interrupts too. That's why I'm suggesting using bus_generic_setup_intr() either within or instead of pcic_pci_setup_intr(). Ian To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-bugs" in the body of the message