January 2007 - Darwin-kernel

Re: /System/Library/Extensions vs /Library/StartupItems
by Kevin Elliott 31 Jan '07

31 Jan '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com At 7:34 -0500 on 1/31/07, Michael Cashwell wrote: Kernel extensions matching static hardware (built-in or on a non-hot-plug bus like PCI) should load before global daemons but I'm not aware of any explicit synchronization. Indeed, even that not really true for anything but the simplest case. It's true that the driver for a particular piece of static hardware will load very early (and probably before StartupItem launching starts), but the behavior of anything above that driver is completely up to the driver and what ever other factors may interfere/interact with the rest of the stack loading. The only general statement that can be made is that the kext loading process starts before the StartupItem process does. Anything more than that and your making assumption that don't generalize. -- ______________________________________________________ Arguing with an engineer is like wrestling with a pig in mud. After a while, you realize the pig is enjoying it. ______________________________________________________ Kevin Elliott <mailto:kelliott@mac.com> AIM/iChatAV: kelliott(a)mac.com (video chat available) ______________________________________________________ _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com

1 0

Re: /System/Library/Extensions vs /Library/StartupItems
by subodh singh 31 Jan '07

31 Jan '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=beta; h=received:message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; b=PlFjhE249x8WvaiPS7W47JXoVyymMK4pNPS7saIbKf3KhktieLAb6us8MGAQvOUwbqDamhvnpd/NGgfNLCA/k64zjBE8DGeYxui1CmycdSAUc7c99++emWz+ZjAk4xqRJxG+m942Kv4TZONWkUJZxoJ9Pw2Wi06IGACY+7CC5hA= Hello Stephane, First of all / System/Library/Extensions are loaded in which Drivers are loaded after that /Library/StartupItems are called which contains call for applications , and when Applications satrted on the call of applications drivers are start to work. Thanks Subodh Which one is supposed to be loaded first at Startup: /System/Library/Extensions or /Library/StartupItems ? AFAIK, it's /System/Library/Extensions but maybe this is just a empirical rule. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/subodh79%40gmail.com This email sent to subodh79(a)gmail.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… On 1/31/07, Stephane Sudre <ssudre(a)intego.com> wrote: This email sent to site_archiver(a)lists.apple.com

1 0

Re: /System/Library/Extensions vs /Library/StartupItems
by Michael Cashwell 31 Jan '07

31 Jan '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Jan 31, 2007, at 6:43 AM, Stephane Sudre wrote: Which one is supposed to be loaded first at Startup: /System/Library/Extensions or /Library/StartupItems ? HTH. -Mike _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… AFAIK, it's /System/Library/Extensions but maybe this is just a empirical rule. I don't think there's any rule or guarantee here. The former holds kernel extensions and their matching process (and thus loading) does start very early. But I don't think there's any mechanism in place to hold off starting daemon processes in /L/S/ until that's "finished." Part of the reason is that there is no "finished." The matching process is on-going and hardware can appear via PC-Card, PCI Express, USB or Firewire at any time and cause kernel extensions to load long after /L/S is done. Kernel extensions matching static hardware (built-in or on a non-hot- plug bus like PCI) should load before global daemons but I'm not aware of any explicit synchronization. This email sent to site_archiver(a)lists.apple.com

1 0

/System/Library/Extensions vs /Library/StartupItems
by Stephane Sudre 31 Jan '07

31 Jan '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Which one is supposed to be loaded first at Startup: /System/Library/Extensions or /Library/StartupItems ? _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… AFAIK, it's /System/Library/Extensions but maybe this is just a empirical rule. This email sent to site_archiver(a)lists.apple.com

1 0

Re: mbuf_dup/mbuf_freem with mbufs containing aux data
by Bhavesh Davda 30 Jan '07

30 Jan '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=beta; h=received:message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; b=oxfu2er5lyAlJuLMwEKmPINgwsH7szt0mWt/ibXbsgOa75v6UIxg/jCZ4nIcI8wKx8wabaifPO35BrHZsrZSFNGdTTvAXWNFePcLok4NOoJapnfvSzuu6BL7uWSXF7HYpPLGFHkH1nSiQPIQf1YDt5qfv5gjPO/2H7rHHJrQpH4= > How do I check if the new mbuf resulted from mbuf_dup() has auxiliary > data at all? I couldn't see any accessors in kpi_mbuf. There is mbuf_aux_find() but it requires you to specify the data type for the query to return a match. I thought that was part of the un-exported private kpi_mbuf. > If there were, then I would have simply queried if there was aux data > attached in the new mbuf, and NULL'ed the aux pointer in the new mbuf, > because my clients for the duplicated mbuf don't care about that aux > data. Have you tried using mbuf_copym() instead of mbuf_dup()? For the former case, the auxiliary mbuf and any packet tags get migrated from the original to the new one. Hmmm.. That's an interesting idea. Are there any performance trade-offs between mbuf_dup versus mbuf_copym? I'll give mbuf_copym a try. Thanks - Bhavesh _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com

1 0

Re: mbuf_dup/mbuf_freem with mbufs containing aux data
by Adi Masputra 30 Jan '07

30 Jan '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Jan 29, 2007, at 6:57 PM, Bhavesh Davda wrote: How do I check if the new mbuf resulted from mbuf_dup() has auxiliary data at all? I couldn't see any accessors in kpi_mbuf. There is mbuf_aux_find() but it requires you to specify the data type for the query to return a match. If there were, then I would have simply queried if there was aux data attached in the new mbuf, and NULL'ed the aux pointer in the new mbuf, because my clients for the duplicated mbuf don't care about that aux data. Have you tried using mbuf_copym() instead of mbuf_dup()? For the former case, the auxiliary mbuf and any packet tags get migrated from the original to the new one. Adi Thanks! -- Bhavesh P. Davda _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com smime.p7s

1 0

Re: mbuf_dup/mbuf_freem with mbufs containing aux data
by Bhavesh Davda 30 Jan '07

30 Jan '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=beta; h=received:message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; b=FJ8Gb9kQBmwP5kPWDsk306slxy7ZKMkePDzAVK3JjBloRg6st6J8PvcUoTOjDmkPByNMmc2PxNTXhkGMreA/8jErtrHYP8p9x1eaXQW+SEGO/S4pXXNclYsUsaRzwREWHuSwASrVbyozzb0Uy+EchAK5XX4eMIt/0FYlR+/5LrY= Hi Adi, > What's my workaround? You may want your code to first check if the new mbuf resulted from mbuf_dup() has the same auxiliary mbuf attached to it, and if so manually copy the auxiliary mbuf and attach it to the new mbuf using mbuf_aux_add(). That way, when this bug is fixed your code would still work. How do I check if the new mbuf resulted from mbuf_dup() has auxiliary data at all? I couldn't see any accessors in kpi_mbuf. If there were, then I would have simply queried if there was aux data attached in the new mbuf, and NULL'ed the aux pointer in the new mbuf, because my clients for the duplicated mbuf don't care about that aux data. Thanks! -- Bhavesh P. Davda _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… Adi This email sent to site_archiver(a)lists.apple.com

1 0

Re: mbuf_dup/mbuf_freem with mbufs containing aux data
by Adi Masputra 30 Jan '07

30 Jan '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Bhavesh, On Jan 29, 2007, at 5:45 PM, Bhavesh Davda wrote: I noticed when my interface filter output method gets called for IPV6 packets (in particular ICMP6 echo replies), that the mbuf contains an aux data pointer in its m_pkthdr. Browsing the xnu-792.13.8 code, it looks like when you mbuf_dup() an mbuf, the pkthdr of the new mbuf is a copy of the pkthdr of the source mbuf, which means that the new mbuf's pkthdr's aux field points to the same piece of data as the original mbuf. Then when mbuf_freem() is called, it unconditionally frees this aux data if there was any on the mbuf chain that you're freeing. Without reference counting the aux field, this is really bad. "freeing free mbufs" panics ensue from this. This is a bug indeed. You may want your code to first check if the new mbuf resulted from mbuf_dup() has the same auxiliary mbuf attached to it, and if so manually copy the auxiliary mbuf and attach it to the new mbuf using mbuf_aux_add(). That way, when this bug is fixed your code would still work. Adi Thanks! -- Bhavesh P. Davda _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/adi%40apple.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com What's my workaround? This email sent to adi(a)apple.com smime.p7s

1 0

mbuf_dup/mbuf_freem with mbufs containing aux data
by Bhavesh Davda 30 Jan '07

30 Jan '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=beta; h=received:message-id:date:from:to:subject:mime-version:content-type:content-transfer-encoding:content-disposition; b=SFqdj6ML5nQypomwgZHKnz0zNWJCTXgZKYTzlAVuZST9ZRW+/4UjFmo/qxOKUwtSNx3HDTYO8qiKg0YbXc+3oKYq9adTrDtFyLFpV5KFgVIjOXKtw/L79ZR8TK1ILLcQGeRtsE/X7do91KrNkqRQh6/IfrGxOkmY/62v0RcSKM8= I noticed when my interface filter output method gets called for IPV6 packets (in particular ICMP6 echo replies), that the mbuf contains an aux data pointer in its m_pkthdr. Browsing the xnu-792.13.8 code, it looks like when you mbuf_dup() an mbuf, the pkthdr of the new mbuf is a copy of the pkthdr of the source mbuf, which means that the new mbuf's pkthdr's aux field points to the same piece of data as the original mbuf. Then when mbuf_freem() is called, it unconditionally frees this aux data if there was any on the mbuf chain that you're freeing. Without reference counting the aux field, this is really bad. "freeing free mbufs" panics ensue from this. What's my workaround? Thanks! -- Bhavesh P. Davda _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com

1 0

DMA Performance / IODMACommand / IOSubMemoryDescriptor / Segments
by devmaillists 29 Jan '07

29 Jan '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com

1 0