Re: [PATCH 05/24] compatfd is included before, and it is compiled unconditionally
On 10/01/2009 06:50 PM, Anthony Liguori wrote: Avi Kivity wrote: On 10/01/2009 04:23 PM, Anthony Liguori wrote: Juan Quintela wrote: Discused with Anthony about it. signalfd is complicated for qemu upstream (too difficult to use properly), It's not an issue of being difficult. To emulate signalfd, we need to create a thread that writes to a pipe from a signal handler. The problem is that a write() can return a partial result and following the partial result, we can end up getting an EAGAIN. We have no way to queue signals beyond that point and we have no sane way to deal with partial writes. pipe buffers are multiples of of the signalfd size. As long as we read and write signalfd-sized blocks, we won't get partial writes. It's true that depending on an implementation detail is bad practice, but this is emulation code, and if helps simplifying everything else, I think it's fine to use it. That's a pretty hairy detail to rely upon.. Well, it's a posix detail, as I quoted below. I'm not in love with it but it should work. Instead, how we do this in upstream QEMU is that we install a signal handler and write one byte to the fd. If we get EAGAIN, that's fine because all we care about is that at least one byte exists in the fd's buffer. This requires that we use an fd-per-signal which means we end up with a different model than signalfd. The reason to use signalfd over what we do in upstream QEMU is that signalfd can allow us to mask the signals which means less EINTRs. I don't think that's a huge advantage and the inability to do backwards compatibility in a sane way means that emulated signalfd is not workable. signalfd is several microseconds faster than signals + pipes. Do we have so much performance we can throw some of it away? Do we have any indication that this difference is actually observable? This seems like very premature optimization. Multiply the signal rate by "a few microseconds", if you get more than 0.1% cpu it's worthwhile in my opinion. The code is localized, and signalfd is a better interface than signals. The same is generally true for eventfd. eventfd emulation will also never get partial writes. But you cannot emulate eventfd faithfully because eventfd is supposed to be additive. If you write 1 50x to eventfd, you should be able to read a set of integers that add up to 50. If you hit EAGAIN in a signal handler, you have no way of handling that. We never rely on the count anyway. You can simply ignore EAGAIN. As I said earlier, the better thing to do is have a higher level interface that has a subset of the behavior of eventfd/signalfd that we can emulate correctly. Sure, but it's more work. Copying an existing interface is easier. It's not like there's no other work in qemu left to be done. -- Do not meddle in the internals of kernels, for they are subtle and quick to panic. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 05/24] compatfd is included before, and it is compiled unconditionally
Avi Kivity wrote: On 10/01/2009 04:23 PM, Anthony Liguori wrote: Juan Quintela wrote: Discused with Anthony about it. signalfd is complicated for qemu upstream (too difficult to use properly), It's not an issue of being difficult. To emulate signalfd, we need to create a thread that writes to a pipe from a signal handler. The problem is that a write() can return a partial result and following the partial result, we can end up getting an EAGAIN. We have no way to queue signals beyond that point and we have no sane way to deal with partial writes. pipe buffers are multiples of of the signalfd size. As long as we read and write signalfd-sized blocks, we won't get partial writes. It's true that depending on an implementation detail is bad practice, but this is emulation code, and if helps simplifying everything else, I think it's fine to use it. That's a pretty hairy detail to rely upon.. Instead, how we do this in upstream QEMU is that we install a signal handler and write one byte to the fd. If we get EAGAIN, that's fine because all we care about is that at least one byte exists in the fd's buffer. This requires that we use an fd-per-signal which means we end up with a different model than signalfd. The reason to use signalfd over what we do in upstream QEMU is that signalfd can allow us to mask the signals which means less EINTRs. I don't think that's a huge advantage and the inability to do backwards compatibility in a sane way means that emulated signalfd is not workable. signalfd is several microseconds faster than signals + pipes. Do we have so much performance we can throw some of it away? Do we have any indication that this difference is actually observable? This seems like very premature optimization. The same is generally true for eventfd. eventfd emulation will also never get partial writes. But you cannot emulate eventfd faithfully because eventfd is supposed to be additive. If you write 1 50x to eventfd, you should be able to read a set of integers that add up to 50. If you hit EAGAIN in a signal handler, you have no way of handling that. As I said earlier, the better thing to do is have a higher level interface that has a subset of the behavior of eventfd/signalfd that we can emulate correctly. Regards, Anthony Liguori -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 05/24] compatfd is included before, and it is compiled unconditionally
On 10/01/2009 04:23 PM, Anthony Liguori wrote: Juan Quintela wrote: Discused with Anthony about it. signalfd is complicated for qemu upstream (too difficult to use properly), It's not an issue of being difficult. To emulate signalfd, we need to create a thread that writes to a pipe from a signal handler. The problem is that a write() can return a partial result and following the partial result, we can end up getting an EAGAIN. We have no way to queue signals beyond that point and we have no sane way to deal with partial writes. pipe buffers are multiples of of the signalfd size. As long as we read and write signalfd-sized blocks, we won't get partial writes. It's true that depending on an implementation detail is bad practice, but this is emulation code, and if helps simplifying everything else, I think it's fine to use it. hmm, pipe(7) says writes smaller than the pipe buffer size are atomic: O_NONBLOCK enabled, n <= PIPE_BUF If there is room to write n bytes to the pipe, then write(2) succeeds immediately, writing all n bytes; otherwise write(2) fails, with errno set to EAGAIN. so it seems this practice has been blessed by posix. Instead, how we do this in upstream QEMU is that we install a signal handler and write one byte to the fd. If we get EAGAIN, that's fine because all we care about is that at least one byte exists in the fd's buffer. This requires that we use an fd-per-signal which means we end up with a different model than signalfd. The reason to use signalfd over what we do in upstream QEMU is that signalfd can allow us to mask the signals which means less EINTRs. I don't think that's a huge advantage and the inability to do backwards compatibility in a sane way means that emulated signalfd is not workable. signalfd is several microseconds faster than signals + pipes. Do we have so much performance we can throw some of it away? The same is generally true for eventfd. eventfd emulation will also never get partial writes. -- Do not meddle in the internals of kernels, for they are subtle and quick to panic. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 05/24] compatfd is included before, and it is compiled unconditionally
Juan Quintela wrote: Discused with Anthony about it. signalfd is complicated for qemu upstream (too difficult to use properly), It's not an issue of being difficult. To emulate signalfd, we need to create a thread that writes to a pipe from a signal handler. The problem is that a write() can return a partial result and following the partial result, we can end up getting an EAGAIN. We have no way to queue signals beyond that point and we have no sane way to deal with partial writes. Instead, how we do this in upstream QEMU is that we install a signal handler and write one byte to the fd. If we get EAGAIN, that's fine because all we care about is that at least one byte exists in the fd's buffer. This requires that we use an fd-per-signal which means we end up with a different model than signalfd. The reason to use signalfd over what we do in upstream QEMU is that signalfd can allow us to mask the signals which means less EINTRs. I don't think that's a huge advantage and the inability to do backwards compatibility in a sane way means that emulated signalfd is not workable. We could possibly introduce a higher level interface that only required one fd per signal and that had a function that drained the signals from the fd without returning any special information. The same is generally true for eventfd. Regards, Anthony Liguori -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 05/24] compatfd is included before, and it is compiled unconditionally
Christoph Hellwig wrote: On Thu, Oct 01, 2009 at 01:58:10PM +0200, Juan Quintela wrote: Discused with Anthony about it. signalfd is complicated for qemu upstream (too difficult to use properly), and eventfd ... The current eventfd emulation is worse than the pipe code that it substitutes. His suggestion here was to create a new abstraction with an API like: push_notify() pop_notify() and then you can implement it with eventfd() pipes/whatever. What was missing for you of compatfd: qemu_eventfd/qemu_signalfd? Do a push_notify()/pop_notify() work for you? I don't desperately want to use it myself anyway. I just want to get rid of the highly annoyind spurious differences in the AIO code due to use of compatfd. I would be perfectly fine with just killing this use of eventfd in qemu-kvm. That's what I'd suggest. The use of eventfd in qemu-kvm is wrong because the compat function is not implemented correctly. Regards, Anthony Liguori -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 05/24] compatfd is included before, and it is compiled unconditionally
On Thu, Oct 01, 2009 at 01:58:10PM +0200, Juan Quintela wrote: > Discused with Anthony about it. signalfd is complicated for qemu > upstream (too difficult to use properly), and eventfd ... The current > eventfd emulation is worse than the pipe code that it substitutes. > > His suggestion here was to create a new abstraction with an API like: > > push_notify() > > pop_notify() > > and then you can implement it with eventfd() pipes/whatever. > > What was missing for you of compatfd: qemu_eventfd/qemu_signalfd? > Do a push_notify()/pop_notify() work for you? I don't desperately want to use it myself anyway. I just want to get rid of the highly annoyind spurious differences in the AIO code due to use of compatfd. I would be perfectly fine with just killing this use of eventfd in qemu-kvm. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 05/24] compatfd is included before, and it is compiled unconditionally
Christoph Hellwig wrote: > On Tue, Sep 22, 2009 at 03:25:13PM +0200, Juan Quintela wrote: >> Christoph Hellwig wrote: >> > Btw, what's the state of getting compatfd upstream? It's a pretty >> > annoying difference between qemu upstream and qemu-kvm. >> >> I haven't tried. I can try to send a patch. Do you have any use case >> that will help the cause? > > Well, the eventfd compat is used in the thread pool AIO code. I don't > know what difference it makes, but I really hate this code beeing > different in both trees. I want to see compatfd used either in both or > none. Discused with Anthony about it. signalfd is complicated for qemu upstream (too difficult to use properly), and eventfd ... The current eventfd emulation is worse than the pipe code that it substitutes. His suggestion here was to create a new abstraction with an API like: push_notify() pop_notify() and then you can implement it with eventfd() pipes/whatever. What was missing for you of compatfd: qemu_eventfd/qemu_signalfd? Do a push_notify()/pop_notify() work for you? Later, Juan. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 05/24] compatfd is included before, and it is compiled unconditionally
On Tue, Sep 22, 2009 at 03:25:13PM +0200, Juan Quintela wrote: > Christoph Hellwig wrote: > > Btw, what's the state of getting compatfd upstream? It's a pretty > > annoying difference between qemu upstream and qemu-kvm. > > I haven't tried. I can try to send a patch. Do you have any use case > that will help the cause? Well, the eventfd compat is used in the thread pool AIO code. I don't know what difference it makes, but I really hate this code beeing different in both trees. I want to see compatfd used either in both or none. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 05/24] compatfd is included before, and it is compiled unconditionally
Christoph Hellwig wrote: > Btw, what's the state of getting compatfd upstream? It's a pretty > annoying difference between qemu upstream and qemu-kvm. I haven't tried. I can try to send a patch. Do you have any use case that will help the cause? Later, Juan. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 05/24] compatfd is included before, and it is compiled unconditionally
Btw, what's the state of getting compatfd upstream? It's a pretty annoying difference between qemu upstream and qemu-kvm. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html