| Message ID | 1468323255-20454-1-git-send-email-caoj.fnst@cn.fujitsu.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 07/12/2016 05:34 AM, Cao jin wrote: > The current judegement of caller is meaningless, make it useful. Is this something you can trigger? If so, what command line? If not, how did you find it? Spelled 'judgment', not 'judegement'; but rather than just fixing the typo, it might read better as: The lone caller of aio_context_setup() only checks whether an error was set, but aio_context_setup() is not setting an error. > > Signed-off-by: Cao jin <caoj.fnst@cn.fujitsu.com> > --- > aio-posix.c | 2 ++ > 1 file changed, 2 insertions(+) > > diff --git a/aio-posix.c b/aio-posix.c > index 6006122..8b0deb7 100644 > --- a/aio-posix.c > +++ b/aio-posix.c > @@ -18,6 +18,7 @@ > #include "block/block.h" > #include "qemu/queue.h" > #include "qemu/sockets.h" > +#include "qapi/error.h" > #ifdef CONFIG_EPOLL_CREATE1 > #include <sys/epoll.h> > #endif > @@ -491,6 +492,7 @@ void aio_context_setup(AioContext *ctx, Error **errp) > assert(!ctx->epollfd); > ctx->epollfd = epoll_create1(EPOLL_CLOEXEC); > if (ctx->epollfd == -1) { > + error_setg_errno(errp, errno, "Failed to create epoll instance"); > ctx->epoll_available = false; > } else { > ctx->epoll_available = true; However, I'm not even sure your patch is right. The mere fact that we have ctx->epoll_available makes it sound like we WANT to be able to gracefully fall back when epoll is not available, rather than raising an error message.
On Tue, 07/12 09:41, Eric Blake wrote: > On 07/12/2016 05:34 AM, Cao jin wrote: > > The current judegement of caller is meaningless, make it useful. > > Is this something you can trigger? If so, what command line? If not, > how did you find it? > > Spelled 'judgment', not 'judegement'; but rather than just fixing the > typo, it might read better as: > > The lone caller of aio_context_setup() only checks whether an error was > set, but aio_context_setup() is not setting an error. > > > > > Signed-off-by: Cao jin <caoj.fnst@cn.fujitsu.com> > > --- > > aio-posix.c | 2 ++ > > 1 file changed, 2 insertions(+) > > > > diff --git a/aio-posix.c b/aio-posix.c > > index 6006122..8b0deb7 100644 > > --- a/aio-posix.c > > +++ b/aio-posix.c > > @@ -18,6 +18,7 @@ > > #include "block/block.h" > > #include "qemu/queue.h" > > #include "qemu/sockets.h" > > +#include "qapi/error.h" > > #ifdef CONFIG_EPOLL_CREATE1 > > #include <sys/epoll.h> > > #endif > > @@ -491,6 +492,7 @@ void aio_context_setup(AioContext *ctx, Error **errp) > > assert(!ctx->epollfd); > > ctx->epollfd = epoll_create1(EPOLL_CLOEXEC); > > if (ctx->epollfd == -1) { > > + error_setg_errno(errp, errno, "Failed to create epoll instance"); > > ctx->epoll_available = false; > > } else { > > ctx->epoll_available = true; > > However, I'm not even sure your patch is right. The mere fact that we > have ctx->epoll_available makes it sound like we WANT to be able to > gracefully fall back when epoll is not available, rather than raising an > error message. Yes. This patch doesn't make sense to me. What can be done is 1) remove errp parameter as it's not used; 2) fprintf(stderr, ...) or error_report(....) here if epoll_create1 failed. Fam
On 07/12/2016 11:41 PM, Eric Blake wrote: > On 07/12/2016 05:34 AM, Cao jin wrote: >> The current judegement of caller is meaningless, make it useful. > > Is this something you can trigger? If so, what command line? If not, > how did you find it? > No, I didn't find it by "trigger" it. I just a very carefully code reader when I learn it:) > Spelled 'judgment', not 'judegement'; but rather than just fixing the > typo, it might read better as: > > The lone caller of aio_context_setup() only checks whether an error was > set, but aio_context_setup() is not setting an error. > Ok >> >> Signed-off-by: Cao jin <caoj.fnst@cn.fujitsu.com> >> --- >> aio-posix.c | 2 ++ >> 1 file changed, 2 insertions(+) >> >> diff --git a/aio-posix.c b/aio-posix.c >> index 6006122..8b0deb7 100644 >> --- a/aio-posix.c >> +++ b/aio-posix.c >> @@ -18,6 +18,7 @@ >> #include "block/block.h" >> #include "qemu/queue.h" >> #include "qemu/sockets.h" >> +#include "qapi/error.h" >> #ifdef CONFIG_EPOLL_CREATE1 >> #include <sys/epoll.h> >> #endif >> @@ -491,6 +492,7 @@ void aio_context_setup(AioContext *ctx, Error **errp) >> assert(!ctx->epollfd); >> ctx->epollfd = epoll_create1(EPOLL_CLOEXEC); >> if (ctx->epollfd == -1) { >> + error_setg_errno(errp, errno, "Failed to create epoll instance"); >> ctx->epoll_available = false; >> } else { >> ctx->epoll_available = true; > > However, I'm not even sure your patch is right. The mere fact that we > have ctx->epoll_available makes it sound like we WANT to be able to > gracefully fall back when epoll is not available, rather than raising an > error message. >
On 07/13/2016 09:33 AM, Fam Zheng wrote: > On Tue, 07/12 09:41, Eric Blake wrote: >> On 07/12/2016 05:34 AM, Cao jin wrote: >>> The current judegement of caller is meaningless, make it useful. >> >> Is this something you can trigger? If so, what command line? If not, >> how did you find it? >> >> Spelled 'judgment', not 'judegement'; but rather than just fixing the >> typo, it might read better as: >> >> The lone caller of aio_context_setup() only checks whether an error was >> set, but aio_context_setup() is not setting an error. >> >>> >>> Signed-off-by: Cao jin <caoj.fnst@cn.fujitsu.com> >>> --- >>> aio-posix.c | 2 ++ >>> 1 file changed, 2 insertions(+) >>> >>> diff --git a/aio-posix.c b/aio-posix.c >>> index 6006122..8b0deb7 100644 >>> --- a/aio-posix.c >>> +++ b/aio-posix.c >>> @@ -18,6 +18,7 @@ >>> #include "block/block.h" >>> #include "qemu/queue.h" >>> #include "qemu/sockets.h" >>> +#include "qapi/error.h" >>> #ifdef CONFIG_EPOLL_CREATE1 >>> #include <sys/epoll.h> >>> #endif >>> @@ -491,6 +492,7 @@ void aio_context_setup(AioContext *ctx, Error **errp) >>> assert(!ctx->epollfd); >>> ctx->epollfd = epoll_create1(EPOLL_CLOEXEC); >>> if (ctx->epollfd == -1) { >>> + error_setg_errno(errp, errno, "Failed to create epoll instance"); >>> ctx->epoll_available = false; >>> } else { >>> ctx->epoll_available = true; >> >> However, I'm not even sure your patch is right. The mere fact that we >> have ctx->epoll_available makes it sound like we WANT to be able to >> gracefully fall back when epoll is not available, rather than raising an >> error message. > > Yes. This patch doesn't make sense to me. What can be done is 1) remove errp > parameter as it's not used; 2) fprintf(stderr, ...) or error_report(....) here > if epoll_create1 failed. > > Fam > Actually, my first thought is 1), and I also checked how ctx->epoll_available is used(seems I still didn't read it carefully). But when I see how event_notifier_init()'s failure(in aio_context_new()) is handled, I think maybe keeping consistence is better. ok, I will choose 1)
diff --git a/aio-posix.c b/aio-posix.c index 6006122..8b0deb7 100644 --- a/aio-posix.c +++ b/aio-posix.c @@ -18,6 +18,7 @@ #include "block/block.h" #include "qemu/queue.h" #include "qemu/sockets.h" +#include "qapi/error.h" #ifdef CONFIG_EPOLL_CREATE1 #include <sys/epoll.h> #endif @@ -491,6 +492,7 @@ void aio_context_setup(AioContext *ctx, Error **errp) assert(!ctx->epollfd); ctx->epollfd = epoll_create1(EPOLL_CLOEXEC); if (ctx->epollfd == -1) { + error_setg_errno(errp, errno, "Failed to create epoll instance"); ctx->epoll_available = false; } else { ctx->epoll_available = true;
The current judegement of caller is meaningless, make it useful. Signed-off-by: Cao jin <caoj.fnst@cn.fujitsu.com> --- aio-posix.c | 2 ++ 1 file changed, 2 insertions(+)