| Message ID | 1462236478-61645-3-git-send-email-chrisfriedt@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
[ccing somebody else who ran into this, since I've figured out why.] On 3 May 2016 at 01:47, Christopher Friedt <chrisfriedt@gmail.com> wrote: > The file ivshmem.c unconditionally references event_notifier_init_fd() > in util/event_notifier-posix.c, even if CONFIG_EVENTFD is not defined. Yes, but ivshmem.c is only built if CONFIG_IVSHMEM, and CONFIG_IVSHMEM is set (in default-configs/pci.mak) to CONFIG_EVENTFD. So if CONFIG_EVENTFD is not defined then we should never build ivshmem.o. The problem here is that you've run into a bug in QEMU's makefiles, where a change in an included .mak file in default-configs fails to cause the config-devices.mak file to be rebuilt. In commit 330b583 we changed pci.mak so that CONFIG_IVSHMEM is set to CONFIG_EVENTFD rather than CONFIG_POSIX, so if your config-devices.mak predates that commit then it will have incorrectly not been recreated and so your build will still try to build ivshmem.c. You can check whether this is true by looking at (for instance) x86_64-softmmu/config-devices.mak in your build tree: if it has a line "CONFIG_IVSHMEM=$(CONFIG_POSIX)" in it then it's the out of date version. You should be able to fix this by deleting */config-devices.mak from your build tree (or by blowing away the build tree entirely and recreating it.) Then try rebuilding -- ivshmem.c should not be compiled. We really must track down this dependency bug, it is very confusing when it bites people. thanks -- PMM
On Mon, May 2, 2016 at 9:01 PM, Peter Maydell <peter.maydell@linaro.org> wrote: > [ccing somebody else who ran into this, since I've figured out why.] > > On 3 May 2016 at 01:47, Christopher Friedt <chrisfriedt@gmail.com> wrote: >> The file ivshmem.c unconditionally references event_notifier_init_fd() >> in util/event_notifier-posix.c, even if CONFIG_EVENTFD is not defined. > > Yes, but ivshmem.c is only built if CONFIG_IVSHMEM, and > CONFIG_IVSHMEM is set (in default-configs/pci.mak) to CONFIG_EVENTFD. > So if CONFIG_EVENTFD is not defined then we should never build ivshmem.o. > > The problem here is that you've run into a bug in QEMU's makefiles, That would explain things. > where a change in an included .mak file in default-configs fails > to cause the config-devices.mak file to be rebuilt. In commit > 330b583 we changed pci.mak so that CONFIG_IVSHMEM is set to > CONFIG_EVENTFD rather than CONFIG_POSIX, so if your config-devices.mak > predates that commit then it will have incorrectly not been recreated > and so your build will still try to build ivshmem.c. > > You can check whether this is true by looking at > (for instance) x86_64-softmmu/config-devices.mak in your build > tree: if it has a line "CONFIG_IVSHMEM=$(CONFIG_POSIX)" in it > then it's the out of date version. $ grep "CONFIG_IVSHMEM" x86_64-softmmu/config-devices.mak CONFIG_IVSHMEM=$(CONFIG_EVENTFD) I have been re-configuring on every build, just to eliminate that potential error source. > You should be able to fix this by deleting */config-devices.mak > from your build tree (or by blowing away the build tree entirely > and recreating it.) Then try rebuilding -- ivshmem.c should not > be compiled. I'll give that a go anyway.
On 3 May 2016 at 02:14, Christopher Friedt <chrisfriedt@gmail.com> wrote: > On Mon, May 2, 2016 at 9:01 PM, Peter Maydell <peter.maydell@linaro.org> wrote: >> [ccing somebody else who ran into this, since I've figured out why.] >> >> On 3 May 2016 at 01:47, Christopher Friedt <chrisfriedt@gmail.com> wrote: >>> The file ivshmem.c unconditionally references event_notifier_init_fd() >>> in util/event_notifier-posix.c, even if CONFIG_EVENTFD is not defined. >> >> Yes, but ivshmem.c is only built if CONFIG_IVSHMEM, and >> CONFIG_IVSHMEM is set (in default-configs/pci.mak) to CONFIG_EVENTFD. >> So if CONFIG_EVENTFD is not defined then we should never build ivshmem.o. >> >> The problem here is that you've run into a bug in QEMU's makefiles, > > That would explain things. > >> where a change in an included .mak file in default-configs fails >> to cause the config-devices.mak file to be rebuilt. In commit >> 330b583 we changed pci.mak so that CONFIG_IVSHMEM is set to >> CONFIG_EVENTFD rather than CONFIG_POSIX, so if your config-devices.mak >> predates that commit then it will have incorrectly not been recreated >> and so your build will still try to build ivshmem.c. >> >> You can check whether this is true by looking at >> (for instance) x86_64-softmmu/config-devices.mak in your build >> tree: if it has a line "CONFIG_IVSHMEM=$(CONFIG_POSIX)" in it >> then it's the out of date version. > > $ grep "CONFIG_IVSHMEM" x86_64-softmmu/config-devices.mak > CONFIG_IVSHMEM=$(CONFIG_EVENTFD) Hmm. (Is it set the same for every config-devices.mak for every target you're trying to build?) Next things to check: is CONFIG_EVENTFD definitely not set in config-host.mak (it shouldn't be)? hw/misc/Makefile.objs ought to say "obj-$(CONFIG_IVSHMEM) += ivshmem.o", and that should mean that we don't try to build it... thanks -- PMM
On Mon, May 2, 2016 at 9:18 PM, Peter Maydell <peter.maydell@linaro.org> wrote: > On 3 May 2016 at 02:14, Christopher Friedt <chrisfriedt@gmail.com> wrote: >> On Mon, May 2, 2016 at 9:01 PM, Peter Maydell <peter.maydell@linaro.org> wrote: >>> [ccing somebody else who ran into this, since I've figured out why.] >>> >>> On 3 May 2016 at 01:47, Christopher Friedt <chrisfriedt@gmail.com> wrote: >>>> The file ivshmem.c unconditionally references event_notifier_init_fd() >>>> in util/event_notifier-posix.c, even if CONFIG_EVENTFD is not defined. >>> >>> Yes, but ivshmem.c is only built if CONFIG_IVSHMEM, and >>> CONFIG_IVSHMEM is set (in default-configs/pci.mak) to CONFIG_EVENTFD. >>> So if CONFIG_EVENTFD is not defined then we should never build ivshmem.o. >>> >>> The problem here is that you've run into a bug in QEMU's makefiles, >> >> That would explain things. > Hmm. (Is it set the same for every config-devices.mak for > every target you're trying to build?) Ack - too late! I manually removed all of the *-softmmu/ directories after doing a make clean, because they were still hanging around. Rebuilding worked fine without PATCH 2/2. Thanks!
On 3 May 2016 at 02:25, Christopher Friedt <chrisfriedt@gmail.com> wrote: > Ack - too late! I manually removed all of the *-softmmu/ directories > after doing a make clean, because they were still hanging around. > Rebuilding worked fine without PATCH 2/2. Oh well. If blowing away the build dirs fixed it then it seems pretty likely that the makefile bug is what you were running into. thanks -- PMM
diff --git a/util/event_notifier-posix.c b/util/event_notifier-posix.c index c1f0d79..c9bb34d 100644 --- a/util/event_notifier-posix.c +++ b/util/event_notifier-posix.c @@ -21,7 +21,6 @@ #include <sys/eventfd.h> #endif -#ifdef CONFIG_EVENTFD /* * Initialize @e with existing file descriptor @fd. * @fd must be a genuine eventfd object, emulation with pipe won't do. @@ -31,7 +30,6 @@ void event_notifier_init_fd(EventNotifier *e, int fd) e->rfd = fd; e->wfd = fd; } -#endif int event_notifier_init(EventNotifier *e, int active) {
The file ivshmem.c unconditionally references event_notifier_init_fd() in util/event_notifier-posix.c, even if CONFIG_EVENTFD is not defined. On platforms where CONFIG_POSIX is defined, but CONFIG_EVENTFD is not defined, that results in an undefined symbol referenced from ivshmem.c and the link fails. That applies to Mac OS X, but possibly other BSD-based distros. Note: there is nothing specific to eventfd inside and event_notifier_init() also fails unconditionally if CONFIG_EVENTFD is not defined. Signed-off-by: Christopher Friedt <chrisfriedt@gmail.com> --- util/event_notifier-posix.c | 2 -- 1 file changed, 2 deletions(-)