From: Remy Noel
get rid of the delete attribute.
We still need to get rid of the context list lock.
Signed-off-by: Remy Noel
---
util/aio-posix.c | 75 ++--
util/aio-win32.c | 43 ++-
2 files changed, 49 insertions(+), 69
From: Remy Noel
It is still used for bottom halves though and to avoid concurent
set_fd_handlers (We could probably decorrelate the two, but
set_fd_handlers are quite rare so it probably isn't worth it).
Signed-off-by: Remy Noel
---
include/block/aio.h | 4 +++-
util/aio-posix.c
From: Remy Noel
We no longer modify existing handlers entries and instead, always insert
those after having properly initialized those.
Also, we do not call aio_epoll_update for deleted handlers as this has
no impact whastoever.
Signed-off-by: Remy Noel
---
util/aio-posix.c | 85
I did some tests and noticed the second and third patch to incur some
performance loss (on a scenario using virtio device)
I will therefore resubmit just the first patch alone.
On 11/16/18 8:02 PM, remy.n...@blade-group.com wrote:
From: Remy Noel
get rid of the delete attribute.
We still
From: Remy Noel
It is possible for an io_poll callback to be concurrently executed along
with an aio_set_fd_handlers. This can cause all sorts of problems, like
a NULL callback or a bad opaque pointer.
This changes set_fd_handlers so that it no longer modify existing handlers
entries and
On 12/10/18 8:05 PM, Stefan Hajnoczi wrote:
On Thu, Dec 06, 2018 at 11:14:23AM +0100, remy.n...@blade-group.com wrote:
+if (is_new) {
+new_node->pfd.fd = fd;
+} else {
+deleted = aio_remove_fd_handler(ctx, node);
+new_node->pfd = node->pfd;
D
From: Remy Noel
It is possible for an io_poll callback to be concurrently executed along
with an aio_set_fd_handlers. This can cause all sorts of problems, like
a NULL callback or a bad opaque pointer.
This changes set_fd_handlers so that it no longer modify existing handlers
entries and
On 12/18/18 6:39 PM, Paolo Bonzini wrote:
On 17/12/18 17:48, remy.n...@blade-group.com wrote:
Also, we do not call aio_epoll_update for deleted handlers as this has
no impact whatsoever.
Why? epoll is used in level-triggered mode, so you do have to remove
the file descriptor...
On 12/18/18 3:14 PM, Stefan Hajnoczi wrote:
Please include a changelog in future patches. For example:
v3:
* Don't drop revents when a handler is modified [Stefan]
That way reviewers know what to look for and which issues you have
addressed.
Sorry, wasn't sure i had to do this in single com
On 12/19/18 8:32 PM, Paolo Bonzini wrote:
You found another bug then. :)
K. Will fix.
From: Remy Noel
Cleaning the events will cause aio_epoll_update to unregister the fd.
Otherwise, the fd is kept registered until it is destroyed.
Signed-off-by: Remy Noel
---
util/aio-posix.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/util/aio-posix.c b/util/aio-posix.c
index
From: Remy Noel
It is possible for an io_poll callback to be concurrently executed along
with an aio_set_fd_handlers. This can cause all sorts of problems, like
a NULL callback or a bad opaque pointer.
This changes set_fd_handlers so that it no longer modify existing handlers
entries and
12 matches
Mail list logo
