April 2015 - Darwin-kernel - lists.apple.com

XPC service crash
by Arjun SM 27 Apr '15

27 Apr '15

_______________________________________________ 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: https://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Debuggin XNU build
by Regis Duchesne 21 Apr '15

21 Apr '15

On a normal Mac, pressing the power button (there is also a key combo to do the same) will drop you to the debugger. Which is why ways to send an NMI to the VM were mentioned earlier in this thread. You will need to set bit 0x4 in debug= in boot-args: DB_NMI 0x04 Drop into debugger on NMI (Command–Power, Command-Option-Control-Shift-Escape, or interrupt switch). — hpreg > On Apr 21, 2015, at 6:22 PM, Prokash Sinha <prokash(a)garlic.com> wrote: > > First, I’m using lldb, since gdb is not being supported anymore. In the debugger, hitting Cntl+C would not work. AFAIRC, gdb does not work that way either when do the kernel debugging ( I could be wrong), but in freebsd it does not work. So in kgdb, the kludge is to make a kernel variable like Kernel.Debug field to have a trap when we toggle or try to write from the target. And that trap is captured in the debugger. > > There got to be a way to Break in to the running kernel directly ( either with key combination on the host debugger) or toggling some variable — I just don’t know. > > Prokash > > On Apr 21, 2015, at 5:43 AM, Regis Duchesne <hpreg(a)vmware.com> wrote: > >>> one more question (though different ) is that - Is there any kernel variable that we can toggle on the target using sysctl to trigger a break into the debugger. Or if there is any key-combination that would do that. This is when kernel debugger connection is fine, and I let the target to boot completely, then I want to break in. Assume I don’t have any component running where I can set BP before letting the target go on running. >> >> I can think of two ways to do this: >> >> 1) Use nvram to add flag 0x1 to debug= in boot-args. >> >> DB_HALT >> 0x01 >> Halt at boot-time and wait for debugger attach (gdb). >> >> Then after the OS X guest boots, it will wait for your debugger to connect. Then just run "c" (continue) and your guest will run normally until it either hits a breakpoint, or you hit Ctrl-C at the gdb prompt. >> >> 2) Write a trivial kernel module which simply calls Debugger(<some string>); in its start method. Then load that module in the guest every time you need to drop to the debugger. >> >> Cheers, >> -- >> hpreg > _______________________________________________ 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: https://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Debuggin XNU build
by Manu . 21 Apr '15

21 Apr '15

_______________________________________________ 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: https://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Debuggin XNU build
by Prokash Sinha 21 Apr '15

21 Apr '15

First, I’m using lldb, since gdb is not being supported anymore. In the debugger, hitting Cntl+C would not work. AFAIRC, gdb does not work that way either when do the kernel debugging ( I could be wrong), but in freebsd it does not work. So in kgdb, the kludge is to make a kernel variable like Kernel.Debug field to have a trap when we toggle or try to write from the target. And that trap is captured in the debugger. There got to be a way to Break in to the running kernel directly ( either with key combination on the host debugger) or toggling some variable — I just don’t know. Prokash On Apr 21, 2015, at 5:43 AM, Regis Duchesne <hpreg(a)vmware.com> wrote: >> one more question (though different ) is that - Is there any kernel variable that we can toggle on the target using sysctl to trigger a break into the debugger. Or if there is any key-combination that would do that. This is when kernel debugger connection is fine, and I let the target to boot completely, then I want to break in. Assume I don’t have any component running where I can set BP before letting the target go on running. > > I can think of two ways to do this: > > 1) Use nvram to add flag 0x1 to debug= in boot-args. > > DB_HALT > 0x01 > Halt at boot-time and wait for debugger attach (gdb). > > Then after the OS X guest boots, it will wait for your debugger to connect. Then just run "c" (continue) and your guest will run normally until it either hits a breakpoint, or you hit Ctrl-C at the gdb prompt. > > 2) Write a trivial kernel module which simply calls Debugger(<some string>); in its start method. Then load that module in the guest every time you need to drop to the debugger. > > Cheers, > -- > hpreg _______________________________________________ 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: https://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Debuggin XNU build
by Regis Duchesne 21 Apr '15

21 Apr '15

> one more question (though different ) is that - Is there any kernel variable that we can toggle on the target using sysctl to trigger a break into the debugger. Or if there is any key-combination that would do that. This is when kernel debugger connection is fine, and I let the target to boot completely, then I want to break in. Assume I don’t have any component running where I can set BP before letting the target go on running. I can think of two ways to do this: 1) Use nvram to add flag 0x1 to debug= in boot-args. DB_HALT 0x01 Halt at boot-time and wait for debugger attach (gdb). Then after the OS X guest boots, it will wait for your debugger to connect. Then just run "c" (continue) and your guest will run normally until it either hits a breakpoint, or you hit Ctrl-C at the gdb prompt. 2) Write a trivial kernel module which simply calls Debugger(<some string>); in its start method. Then load that module in the guest every time you need to drop to the debugger. Cheers, -- hpreg _______________________________________________ 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: https://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists… This email sent to site_archiver(a)lists.apple.com

1 0

Re: What is or are the Best APIs for Thread creation etc in kext ?
by Quinn "The Eskimo!" 17 Apr '15

17 Apr '15

On 9 Apr 2015, at 03:47, Prokash Sinha <prokash(a)garlic.com> wrote: > Is pthread interface valid alternative? No, it's not supported in the kernel. I generally point folks at the "thread call" KPIs, <kern/thread_call.h>. Share and Enjoy -- Quinn "The Eskimo!" <http://www.apple.com/developer/> Apple Developer Relations, Developer Technical Support, Core OS/Hardware _______________________________________________ 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: https://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Payload limitation on kern_ctl_reg
by Prokash Sinha 16 Apr '15

16 Apr '15

Inside the messaging it is checking for MLEN ( i.e. right around 200 bytes ). No that may not be practical, it could be a hot spot sometime. I need to test out the normal socket path, instead of the cntl-socket-event-path (due to its ~200 bytes limitation ), before I think about mapping pages/section/region etc. As per Apple’s Kernel Programming doc, the socket interface could use COW and mapping when the data is large, so I’m trying to see if that gives me some clue… Thanks, Prokash On Apr 16, 2015, at 11:13 AM, Peter Moody <pmoody(a)google.com> wrote: > > can't you loop over your data until you've sent it all? > > something like (not tested): > > for (int i = 0; i < sizeof(msg);) { > size_t space; > > if (ctl_getenqueuespace(ctl_ref, unit, &space) != KERN_SUCCESS) > goto bail; > // Send the smaller of what's left of msg and how much space is available > // in the queue. > space = (sizeof(msg) - i) < space ? sizeof(msg) - i : space; > > if (ctl_enqueuedata(ctl_ref, unit, (void*)(&msg) + i, space, CTL_DATA_NOWAKEUP) != > KERN_SUCCESS) > goto bail; > i += space; > } > bail: > ctl_enqueuedata(ctl_ref, unit, NULL, 1, CTL_DATA_EOR); > > When I was playing with this, I seem to recall seeing a queue size of > something like 480 bytes .. ? I might be mis-remembering that. > > On Wed, Apr 15 2015 at 15:01, Prokash Sinha wrote: >> Folks, >> >> Is there a a limit to the data payload for this type control/socket interface ? >> >> The kext does provide the ctl, app connects to the ctl using socket interface from the user space. >> >> Kext is the data provider. I’ve a choice of tacking a pointer inside the event structure, but that is a kernel vm address without proper mapping. So I >> can stick data in a way, that the event structure will have it…. >> >> So when the event is posted using lev_msg_post(&msg), where msg is struct lev_msg, the msg.dv[0].data_length is where the data length is filled out, >> and msg.dv[0].data is the address of data. >> >> It seems like the maximum amount of data that can be placed here is 160 bytes or little more. >> >> Is there any restriction for this? >> >> My requirement is between 2K to 3K max, less than a page, so may be I should be able to shovel a page at the most in any event. >> >> Otherwise I will have resort to kernel to use memory mapping. >> >> Any suggestion ? >> >> Thanks, >> Prokash >> _______________________________________________ >> 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: >> https://lists.apple.com/mailman/options/darwin-kernel/pmoody%40google.com >> >> This email sent to pmoody(a)google.com > > -- > peter moody _______________________________________________ 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: https://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Payload limitation on kern_ctl_reg
by Peter Moody 16 Apr '15

16 Apr '15

can't you loop over your data until you've sent it all? something like (not tested): for (int i = 0; i < sizeof(msg);) { size_t space; if (ctl_getenqueuespace(ctl_ref, unit, &space) != KERN_SUCCESS) goto bail; // Send the smaller of what's left of msg and how much space is available // in the queue. space = (sizeof(msg) - i) < space ? sizeof(msg) - i : space; if (ctl_enqueuedata(ctl_ref, unit, (void*)(&msg) + i, space, CTL_DATA_NOWAKEUP) != KERN_SUCCESS) goto bail; i += space; } bail: ctl_enqueuedata(ctl_ref, unit, NULL, 1, CTL_DATA_EOR); When I was playing with this, I seem to recall seeing a queue size of something like 480 bytes .. ? I might be mis-remembering that. On Wed, Apr 15 2015 at 15:01, Prokash Sinha wrote: > Folks, > > Is there a a limit to the data payload for this type control/socket interface ? > > The kext does provide the ctl, app connects to the ctl using socket interface from the user space. > > Kext is the data provider. I’ve a choice of tacking a pointer inside the event structure, but that is a kernel vm address without proper mapping. So I > can stick data in a way, that the event structure will have it…. > > So when the event is posted using lev_msg_post(&msg), where msg is struct lev_msg, the msg.dv[0].data_length is where the data length is filled out, > and msg.dv[0].data is the address of data. > > It seems like the maximum amount of data that can be placed here is 160 bytes or little more. > > Is there any restriction for this? > > My requirement is between 2K to 3K max, less than a page, so may be I should be able to shovel a page at the most in any event. > > Otherwise I will have resort to kernel to use memory mapping. > > Any suggestion ? > > Thanks, > Prokash > _______________________________________________ > 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: > https://lists.apple.com/mailman/options/darwin-kernel/pmoody%40google.com > > This email sent to pmoody(a)google.com -- peter moody _______________________________________________ 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: https://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists… This email sent to site_archiver(a)lists.apple.com

1 0

Payload limitation on kern_ctl_reg
by Prokash Sinha 15 Apr '15

15 Apr '15

Folks, Is there a a limit to the data payload for this type control/socket interface ? The kext does provide the ctl, app connects to the ctl using socket interface from the user space. Kext is the data provider. I’ve a choice of tacking a pointer inside the event structure, but that is a kernel vm address without proper mapping. So I can stick data in a way, that the event structure will have it…. So when the event is posted using lev_msg_post(&msg), where msg is struct lev_msg, the msg.dv[0].data_length is where the data length is filled out, and msg.dv[0].data is the address of data. It seems like the maximum amount of data that can be placed here is 160 bytes or little more. Is there any restriction for this? My requirement is between 2K to 3K max, less than a page, so may be I should be able to shovel a page at the most in any event. Otherwise I will have resort to kernel to use memory mapping. Any suggestion ? Thanks, Prokash _______________________________________________ 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: https://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists… This email sent to site_archiver(a)lists.apple.com

1 0

What is or are the Best APIs for Thread creation etc in kext ?
by Prokash Sinha 09 Apr '15

09 Apr '15

Folks, I’m trying to find out what alternatives do we have in the kernel module for thread creation and management ? I see the mach thread management APIs, which is more involved the IOKit related API. But IOKit APIs are deprecated. Yes, I can disable the warnings, but that is not quite the way to go. Is pthread interface valid alternative? Or is it supported in the kernel module ? Thanks, Prokash _______________________________________________ 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: https://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists… This email sent to site_archiver(a)lists.apple.com

1 0