October 2006 - Darwin-kernel

Re: Determining disk activity per volume
by Michael Smith 11 Oct '06

11 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Oct 10, 2006, at 10:19 PM, Arvind Dalvi wrote: how about developing I/O Media Filter ? "An I/O Media Object acts as a channel for all I/O that goes to the storage underlying it" = 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.… http://lists.apple.com/archives/Darwin-development/2002/May/ msg00382.html Again, this puts you on the hot path for potentially thousands of operations a second, to update a UI at a fraction of that rate. Unless you are trying to gather statistics that aren't otherwise available, not a good idea. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Re: Determining disk activity per volume
by Arvind Dalvi 11 Oct '06

11 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; b=R6qhpgtT9icdRjqAdZb9C/28ylzLnhER2l3/v4XDEwNpHeEhSzMhy9hbygtPm8mvolDUh30nra0fPW72lXIJIp+zMfPjejxXuSYGJ9cwjOFJcM/YV65c2fE8pg6mOvariPp5eFZ3O6PJ43YS+u4u4bxHRCaDCbEn0xyTpElqumE= Rick, how about developing I/O Media Filter ? "An I/O Media Object acts as a channel for all I/O that goes to the storage underlying it" http://lists.apple.com/archives/Darwin-development/2002/May/msg00382.html a.d. _______________________________________________ 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: passing data to kernel driver from a file.
by John Dalgliesh 09 Oct '06

09 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com I have written a kernel resident driver.I need to pass some data to the driver which is stored in a file.How can this be accomplised.Came to know that file operations are not supported in kernel.Read about the UserClient but still not clear.Please suggest me how to pass data to the driver. No worries. {P^/ John On Mon, 9 Oct 2006, sanjay yaragatti wrote: Hi, Hi, please don't cross-post to multiple lists. If you don't get an answer from one list after a few days it's OK to resend your question to another. It's not OK to spam both at the same time like this. OK, here we go again. File access in the kernel is discouraged (as I have learnt the hard way :). Even if you are accessing it through a UserClient it still imports userland dependencies into the kernel. So the first question is: why? What are the contents of this file? Is it like a firmware blob that gets loaded once and discarded, or is it a settings file? Something else? Also: when? At boot time or later? Note that if you're trying to access a resource file within your kext's directory, then I believe that there are even cases where it won't be available (e.g. when netbooted) so you need to worry about that too. If the file you want is just a firmware blob, then a good way around loading a file is to have a kext in your plugins directory with the blob compiled-in. That way you stay within the kext system and it can even be unloaded when you're done with it. If however the file is a settings file you will need to go the UserClient route. I have done this before and now have a daemon plus a kext that provides access to files through a nice C++ class. It can even return a .plist file as an OSDictionary. You're welcome to use it or just study it and write your own if you prefer. Thanks in Advance.. (You can find that daemon+kext in my CVS repository, instructions to access it are under 'Development' here: <http://www.defyne.org/dvb/>. Check out Driver/UDMUtils/UDMResourceFile.hpp for the interface. The drivers package on that site installs it all for you if you don't want to build it yourself.) _______________________________________________ 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

passing data to kernel driver from a file.
by sanjay yaragatti 09 Oct '06

09 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=s1024; d=yahoo.co.in; h=Message-ID:Received:Date:From:Subject:To:Cc:MIME-Version:Content-Type:Content-Transfer-Encoding; b=mpMbmgoa/LoEpdDNT2zNs7YdC/NT34FRjpYmBl4K/IRzDcFYRd4Rd4Ct6Ke4bEUYDfb8r4JeHVtxcIUv/YlkEHxK+VGtLHVfNUNkwzev3Pnwrc7+zAwCtbxTN6uF1SSEa8B9Q8Rl9gBeDDsAtc0I5vAR8S2bNWSUTcFYmJhAocw= ; Hi, I have written a kernel resident driver.I need to pass some data to the driver which is stored in a file.How can this be accomplised.Came to know that file operations are not supported in kernel.Read about the UserClient but still not clear.Please suggest me how to pass data to the driver. Thanks in Advance.. __________________________________________________________ Yahoo! India Answers: Share what you know. Learn something new http://in.answers.yahoo.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

Re: Determining disk activity per volume
by Michael Smith 07 Oct '06

07 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Oct 7, 2006, at 01:59 , Michael Smith wrote: Perhaps you could explain what you mean by "a write has begun or ended"? = 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.… On Oct 7, 2006, at 10:51 AM, Rick Mann wrote: No, it will work very badly. Please look at the iostat code I referenced yesterday. If you don't follow my advice, please at least let us know the name of your product so that I can avoid it like the plague. Perhaps you can elaborate? I did take a look at iostat, and since it wasn't immediately obvious what it was doing, and I saw the Kauth solution at about the same time, I chose that instead. Maybe if you can explain why that's such a bad approach, I'll reconsider. The kauth hooks were designed for code that needs to be on the hot path; either because it is actively participating in the process of authorising filesystem operations, or because it hooks into an external queueing mechanism that requires precise notification of operations. Your application does not participate in authorisation, and it does not require precise notification. From reading your messages, it's clear that you have little or no understanding of the consequences of injecting code on the hot path: I believe the approach I'm looking at now, Kauth, will miss requests satisfied by the cache. Kauth sits on the application side of the I/O stack, but it only covers authorisable operations. These have no direct relationship with disk I/O. I looked at iostat's code again. I see that it asks the OS for device statistics, but I don't see how that helps me to implement a disk activity indicator. I want to display, within a humanly- imperceptible delay period, an indication that a read or write had begun or ended. My Mac workstation here at home has a disk subsystem capable of around 15,000 IO/sec. What sort of "notification" do you intend to display to the user that each of those I/Os has "begun" or "ended"? This email sent to site_archiver(a)lists.apple.com

1 0

Re: Determining disk activity per volume
by Rick Mann 07 Oct '06

07 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Oct 7, 2006, at 01:59 , Michael Smith wrote: -- Rick _______________________________________________ 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.… No, it will work very badly. Please look at the iostat code I referenced yesterday. If you don't follow my advice, please at least let us know the name of your product so that I can avoid it like the plague. I looked at iostat's code again. I see that it asks the OS for device statistics, but I don't see how that helps me to implement a disk activity indicator. I want to display, within a humanly-imperceptible delay period, an indication that a read or write had begun or ended. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Determining disk activity per volume
by Rick Mann 07 Oct '06

07 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Oct 7, 2006, at 01:59 , Michael Smith wrote: -- Rick _______________________________________________ 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.… No, it will work very badly. Please look at the iostat code I referenced yesterday. If you don't follow my advice, please at least let us know the name of your product so that I can avoid it like the plague. Perhaps you can elaborate? I did take a look at iostat, and since it wasn't immediately obvious what it was doing, and I saw the Kauth solution at about the same time, I chose that instead. Maybe if you can explain why that's such a bad approach, I'll reconsider. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Determining disk activity per volume
by Rick Mann 07 Oct '06

07 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Oct 7, 2006, at 01:29 , Arvind Dalvi wrote: the below options should help you if you need to watch for file i/o, but then will the hooking be implemented before the VNODE/VFS file cache ? -- Rick _______________________________________________ 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.… do you want to watch for disk i/o or file i/o ? Ideally, disk I/O. If a cache hit satisfies a request, and I miss that event, that's fine. I believe the approach I'm looking at now, Kauth, will miss requests satisfied by the cache. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Determining disk activity per volume
by Michael Smith 07 Oct '06

07 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Oct 6, 2006, at 16:14 , Thulli wrote: = 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.… On Oct 6, 2006, at 6:37 PM, Rick Mann wrote: Option 1. If you are interested only in read and write system calls and don't want to do polling, take a look at this. http://developer.apple.com/technotes/tn2005/tn2127.html. You need to write a kauth hook and user land client process. If the kauth hook kext could provide a device interface, then the user land client can block on a "read(/dev/yourKauthDevice)" and process what you read with no polling. Oooh! This might work very well. No, it will work very badly. Please look at the iostat code I referenced yesterday. If you don't follow my advice, please at least let us know the name of your product so that I can avoid it like the plague. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Re: Determining disk activity per volume
by Arvind Dalvi 07 Oct '06

07 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; b=WYzffd2YEM/G91RdO1MBwq7GodWfPcAqO94tYsTkdOAVpXRuP1u9qs2LQW6mfldjGrj7bcSnO/kbCMWmzkKKCylQgdugANt1im/x38JtUtR7Si+rpZ4+1D/NNWjxztv+oBUG9L7wP/kAzGpwku/aoHhH0MecT9dEMiIMDNym+Gs= Rick, do you want to watch for disk i/o or file i/o ? the below options should help you if you need to watch for file i/o, but then will the hooking be implemented before the VNODE/VFS file cache ? a.d. On Oct 6, 2006, at 16:14 , Thulli wrote: > Option 1. If you are interested only in read and write system calls > and don't want to do polling, take a look at this. > http://developer.apple.com/technotes/tn2005/tn2127.html. You need > to write a kauth hook and user land client process. If the kauth > hook kext could provide a device interface, then the user land > client can block on a "read(/dev/yourKauthDevice)" and process what > you read with no polling. Oooh! This might work very well. > Option 2. If you don't want to write a kext and still achieve > results with NO polling, then take a look at this > http://developer.apple.com/documentation/Performance/Conceptual/ > FileSystem/Articles/TrackingChanges.html I think the latency here is very high, isn't it (hundreds of milliseconds to seconds)? -- Rick _______________________________________________ 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 10/7/06, Rick Mann wrote: This email sent to site_archiver(a)lists.apple.com

1 0