Hi, thanks. Looks mostly good but some comments below.
On Mon, Jul 4, 2016 at 7:26 PM, Justin Cinkelj
wrote:
> Signed-off-by: Justin Cinkelj
> ---
> Makefile | 1 +
> core/osv_c_wrappers.cc | 39 +++
> include/osv/osv_c_wrappers.h
- they all are supposed to have C linkage, but the the "extern C" is
almost always in the header file,
not the .c file. Maybe the prototype is not exactly identical? I'll take a
look at your patch.
> BR Justin
>
> On 07/04/2016 06:54 PM, Nadav Har'El wrote:
>
> Hi,
On Tue, Jul 5, 2016 at 9:56 AM, Justin Cinkelj
wrote:
> Signed-off-by: Justin Cinkelj
> ---
> Makefile | 1 +
> core/osv_c_wrappers.cc | 33 +
> include/osv/osv_c_wrappers.h | 27 +++
> 3 files changed, 61 insert
enough for applications
which want to use libquadmath (for example) for calculations but do not
intend to use its printf extensions.
Refs issue #765 (this issue should not be closed until we have a full
implementation of these functions, or make a decision that we never will).
Signed-off-by: Nadav
Previously, pthread_setaffinity_np() only supported working on the current
thread. With this patch we use the new sched::thread::pin(thread*, cpu*)
to support pinning of other threads as well.
Fixes #520.
Signed-off-by: Nadav Har'El
---
libc/pthread.cc | 7 ++-
1 file changed, 2 inser
Add tests for pin()ing this thread or a different thread.
I attempted to capture some of the edge cases, such as pin() on a running,
thread, a sleeping thread, or a thread which is often in migrate-disable
state.
Signed-off-by: Nadav Har'El
---
tests/tst-pin.cc
race
with a wake() of the same thread, resulting in the thread remaining on
the old cpu's wakeup queues, and we better ignore it there.
Signed-off-by: Nadav Har'El
---
core/sched.cc | 3 +++
1 file changed, 3 insertions(+)
diff --git a/core/sched.cc b/core/sched.cc
index 7a4afeb..0507d
()ed before, will cause pin() to
never return. In the future, we should separate the temporary migrate_disable()
and permanent pinning, and only wait while the former is in effect.
Signed-off-by: Nadav Har'El
---
include/osv/sched.hh | 20 +++
core/sched.cc
Allow running export_runtime() twice without ill-effect.
A following patch will allow migrating sleeping threads, and migrating it again
before it got to run is possible.
Signed-off-by: Nadav Har'El
---
core/sched.cc | 6 --
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a
enum, which do have conversion to
int. Put this enum inside a namespace so we can still use names like
init_prio::sched and don't need to change any of the callers.
Fixes #769
Signed-off-by: Nadav Har'El
---
include/osv/prio.hh | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
in about an illegal
conversion of an integer to a pointer (!?) in a huge macro expansion
resulting in a line with over 130,000 characters (!). It wasn't easy
to find the needle in this haystack, but eventually I found it.
Signed-off-by: Nadav Har'El
---
bsd/sys/xen/interface/io/ring.h | 2 +-
ide_drive's constructor contained an obvious typo, not setting the
_dev member as intended but rather copying it from itself (!).
Older compilers happily accepted this error, but gcc 6 refused to
accept this, and found a bug.
Signed-off-by: Nadav Har'El
---
drivers/ide.cc | 2 +-
1 fi
.
Signed-off-by: Nadav Har'El
---
core/mmu.cc | 1 +
core/sched.cc | 1 +
2 files changed, 2 insertions(+)
diff --git a/core/mmu.cc b/core/mmu.cc
index 6a31ce7..f929412 100644
--- a/core/mmu.cc
+++ b/core/mmu.cc
@@ -28,6 +28,7 @@
#include "dump.hh"
#include
#include
+#include
Signed-off-by: Nadav Har'El
---
core/mempool.cc | 1 -
1 file changed, 1 deletion(-)
diff --git a/core/mempool.cc b/core/mempool.cc
index 0fbd857..50c938f 100644
--- a/core/mempool.cc
+++ b/core/mempool.cc
@@ -11,7 +11,6 @@
#include
#include
#include
-#include
#include
#in
Gcc 6 does not like an unsized array in a union. Fortunately, there
was no real reason for this trickery in the first place - a
traditional cast to char* works just as well.
Signed-off-by: Nadav Har'El
---
tools/cpiod/cpio.cc | 18 +++---
1 file changed, 7 insertions(+
names
with different arguments.
Signed-off-by: Nadav Har'El
---
include/osv/bitset-iter.hh | 29 -
1 file changed, 4 insertions(+), 25 deletions(-)
diff --git a/include/osv/bitset-iter.hh b/include/osv/bitset-iter.hh
index 524328d..3bcca25 100644
--- a/include/osv/b
ret_cast's pthread_t, whatever it is,
into a pointer. Additionally, we add a static_assert to verify that these
two types (the pointer and pthread_t) really have the same length, as
intended.
Fixes #770.
Signed-off-by: Nadav Har'El
---
libc/pthread.cc| 2 ++
generate bad code. We need to stop both, and this is what this patch does.
Signed-off-by: Nadav Har'El
---
Makefile | 15 +++
1 file changed, 11 insertions(+), 4 deletions(-)
diff --git a/Makefile b/Makefile
index feb6d54..63427bf 100644
--- a/Makefile
+++ b/Makefile
@@ -465,7 +
An enum class can be explicitly converted to the underlying int,
using a static_cast, but this conversion does not happen
implicitly, and gcc 6 is even stricter about this than the
previous versions of the compiler.
Signed-off-by: Nadav Har'El
---
arch/x64/apic.hh | 6 +++---
1 file chang
Needed on gcc 6 for one of the files
Signed-off-by: Nadav Har'El
---
Makefile | 2 ++
1 file changed, 2 insertions(+)
diff --git a/Makefile b/Makefile
index 63427bf..530e9a5 100644
--- a/Makefile
+++ b/Makefile
@@ -502,6 +502,8 @@ $(out)/bsd/sys/netinet/in_rmx.o:
COMMON+=-fno-strict-ali
we didn't.
--
Nadav Har'El
n...@scylladb.com
On Thu, Jul 21, 2016 at 12:12 AM, Nadav Har'El wrote:
> The "&" character was strangely missing in the original code in the
> familiar idiom for calculating the offset of a field in a structure.
> I don't kn
to figure
out its default header search path, and look for those with "c++" in the
name. Usually this will include /usr/include/c++/ and also
/usr/include/c++//.
Fixes #768
Signed-off-by: Nadav Har'El
---
Makefile | 8
1 file changed, 8 insertions(+)
diff --git a/Ma
ted as errors
> Makefile:372: recipe for target 'build/release.x64/loader.o' failed
> make: *** [build/release.x64/loader.o] Error 1
> [benoit@scylla osv]$ ^C
> [benoit@scylla osv]$
>
>
> 2016-07-22 20:32 GMT+02:00 Nadav Har'El :
>
>> Starting with
On Sun, Jul 24, 2016 at 12:33 PM, Gleb Natapov wrote:
> On Sun, Jul 24, 2016 at 11:27:02AM +0200, Benoît Canet wrote:
> > ZFS is also used on ubuntu.
> >
> So what? It does not changes the fact that writing using write() system
> call does not go through out write page cache on OSv. It does go th
On Sun, Jul 24, 2016 at 1:06 PM, Gleb Natapov wrote:
>
> > So it appears that unless you're using mmap(), OSv does *not* have any
> > writeback buffer, not even for ZFS filesystems: If you do a write()
> system
> > call to write 10 bytes, you will create a 10-byte disk operation; If
> > understan
On Sun, Jul 24, 2016 at 1:28 PM, Benoît Canet
wrote:
>
> But, Benoit, is it possible that ZFS *is* trying to coalesce writes, just
>> somehow misconfigured to aim at the wrong size? What I mean is, if you do
>> write() operations of 10 bytes, will you see 10-byte operations, or are
>> they still
race
with a wake() of the same thread, resulting in the thread remaining on
the old cpu's wakeup queues, and we better ignore it there.
Signed-off-by: Nadav Har'El
---
core/sched.cc | 3 +++
1 file changed, 3 insertions(+)
diff --git a/core/sched.cc b/core/sched.cc
index 7a4afeb..0507d
Allow running export_runtime() twice without ill-effect.
A following patch will allow migrating sleeping threads, and migrating it
again before it got to run is possible.
Signed-off-by: Nadav Har'El
---
core/sched.cc | 6 --
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a
not be migrated, as code surrounded
by migrate_disable() cannot handle being migrated in the middle. So pin()
will sleep a bit, in a loop, until the thread becomes migratable again.
Re-pinning of an already pin()ed thread is supported, though.
Signed-off-by: Nadav Har'El
---
include/osv/sch
Previously, pthread_setaffinity_np() only supported working on the current
thread. With this patch we use the new sched::thread::pin(thread*, cpu*)
to support pinning of other threads as well.
Fixes #520.
Signed-off-by: Nadav Har'El
---
libc/pthread.cc | 7 ++-
1 file changed, 2 inser
.
Signed-off-by: Nadav Har'El
---
tests/tst-pin.cc | 198 +
modules/tests/Makefile | 2 +-
2 files changed, 199 insertions(+), 1 deletion(-)
create mode 100644 tests/tst-pin.cc
diff --git a/tests/tst-pin.cc b/tests/tst-pin.cc
new file
"_pinned" boolean flag should be replaced by a bitmask
of CPUs which this thread is allowed to be on, so that a thread can be
pinned to a set of CPUs instead of just one.
Signed-off-by: Nadav Har'El
---
include/osv/sched.hh | 7 +++
core/sched.cc| 16 ++--
2 fi
On Fri, Jul 29, 2016 at 10:58 AM, Wind Yu wrote:
> Hi everyone,
> I want to compare the network performance of OSV and Linux. I really want
> to know how to do the performance test. I don't know if I get the meaning.
>
What do you mean "the meaning"?
I assume that to compare the performance of
On Tue, Aug 2, 2016 at 12:51 PM, wrote:
> Hi Team,
> I'm trying to build an OSv application. I have successfully build the same
> using capstan. But when I ran the application it prints the following logs
> and exits.
>
> ~/bin/capstan run
> Created instance: test
> OSv v0.24
> eth0: 192.168.122.
On Wed, Aug 3, 2016 at 12:54 PM, wrote:
> Hi Team ,
> I'm trying to port a Java Project to OSv.
> PFB some doubts,
> how can I pre create a directory structure in OSv, before starting my OSv
> App, so that the App can use this directory for storing data and writing
> logs. ?
>
Sounds like all yo
On Mon, Jul 4, 2016 at 10:27 AM, Justin Cinkelj
wrote:
> Signed-off-by: Justin Cinkelj
> ---
> libc/pthread.cc | 53 -
> 1 file changed, 28 insertions(+), 25 deletions(-)
>
> diff --git a/libc/pthread.cc b/libc/pthread.cc
> index 308b220..6088
,
though: pthread_getaffinity_np() follows pthread tradition and returns
errors directly, while sched_getaffinity() follows Unix tradition and
returns errors in errno.
Fixes #776.
Signed-off-by: Nadav Har'El
---
libc/pthread.cc | 49 +++--
1 file change
migration
lock - so let's start using it.
Refs #776.
Signed-off-by: Nadav Har'El
---
include/osv/sched.hh | 1 +
libc/pthread.cc | 13 +
2 files changed, 6 insertions(+), 8 deletions(-)
diff --git a/include/osv/sched.hh b/include/osv/sched.hh
index 2ff1d00..261bcfd 10
of them will call. Note how the return values differ,
though: pthread_setaffinity_np() follows pthread tradition and returns
errors directly, while sched_setaffinity() follows Unix tradition and
returns errors in errno.
Refs #776.
Signed-off-by: Nadav Har'El
---
libc/pthread.cc
if cpusetsize is low, our code incorrectly looked beyond the end of
the array, found random bits and usually complained that more than one
is set. Use the correct set size, as specified by the user.
Fixes #771.
Signed-off-by: Nadav Har'El
---
libc/pthread.cc | 2 +-
1 file changed, 1 inse
code :-)
This patch also adds a test for this bug, which passes on Linux, failed
on OSv before this patch, and now passes on OSv too.
Fixes #775
Signed-off-by: Nadav Har'El
---
bsd/sys/kern/uipc_syscalls_wrap.cc | 8 +++
tests/tst-uio.cc
I observed that the same bug did NOT exist in sendmsg(), and decided to go
with the same fix we already had there (deeper inside the code).
So please disregard this patch and I'll send a new one.
--
Nadav Har'El
n...@scylladb.com
On Wed, Aug 3, 2016 at 8:01 PM, Nadav Har'El
commits: by making a
temporary copy of the iovec array before calling the internal code which
messes it up.
This patch also adds a test for sendmsg() and recvmsg(), which passes on
Linux, failed on OSv before this patch, and now passes on OSv too.
Fixes #775
Signed-off-by: Nadav Har'El
---
bs
quot;auio" structure and puts the pointer there, and since it's temporary we
don't need to restore it.
>
> On Wed, Aug 3, 2016 at 7:27 PM, Nadav Har'El wrote:
>
>> The recvmsg() function had a bug where the internal implementation
>> modified
>> the i
On Fri, Aug 5, 2016 at 1:22 PM, Krishnaprasad A S wrote:
> How can I mount a local directory in OSv ?
>
What do you mean "a local directory"? Do you mean mount a second virtual
disk with a ZFS partition?
That should be doable. You can see the code in mount_zfs_rootfs() as an
example of code whi
On Fri, Aug 5, 2016 at 11:49 AM, Justin Cinkelj
wrote:
> This 3 patches work OK for me.
>
> Only one minor complain - once thread affinity is set to a single CPU, it
> cannot be reset back to "all CPUs" (cpuset=0x0F, on 4 CPU VM). As Open MPI
> only reports warning, I can ignore this :).
>
Yes,
FYI, these tests have been failing for a long time on Jenkins, it's just
that a "scripts/build" bug hid this failure (see
https://github.com/cloudius-systems/osv/issues/779)
On Sun, Aug 7, 2016 at 10:46 PM, wrote:
> + exec ./scripts/test.py
> TEST javaOK (2.766
an also mount the ZFS filesystem from the image to your
host and inspect it.
We need to improve the wikis in this area (or maybe there are, and I'm just
not aware of them).
> On Sun, Aug 7, 2016 at 7:37 PM, Nadav Har'El wrote:
>
>>
>> On Fri, Aug 5, 2016 at 1:22 PM, K
__sched_cpucount() function, which is why we need it.
So shouldn't we perhaps just do the same? Make CPU_COUNT and CPU_COUNT_S
macros (or inline functions) which calls this __sched_cpucount() function,
instead of replicating it?
--
Nadav Har'El
n...@scylladb.com
On Mon, Aug 8, 2016 at 6:52 AM, Just
On Mon, Aug 8, 2016 at 6:00 PM, Waldek Kozaczuk
wrote:
> Given that there is a single address space and single process in OSv I
> would imagine it is possible for an application thread to access (read or
> write) any memory due to a bug or intentionally and corrupt kernel/system
> memory structur
The file libc/network/getifaddrs.c was not used - the Makefile was using
the one from musl/. And it's not working anyway (issue #783).
Signed-off-by: Nadav Har'El
---
libc/network/getifaddrs.c | 180 --
1 file changed, 180 deletions(-)
d
.
Musl's if_nameindex() code calls this ioctl with increasing ifc_len starting
with 0, so we need this patch for it not to fail on the first iteration of
its loop.
Signed-off-by: Nadav Har'El
---
bsd/sys/compat/linux/linux_ioctl.cc | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
di
ose to create a UNIX domain socket, which is not currently supported
in OSv (see issue #355). Just changing the code to use AF_INET instead
of AF_UNIX makes it work correctly.
Signed-off-by: Nadav Har'El
---
Makefile| 2 +-
libc/network
l about it,
and so should we, to be compatible with Linux.
With this patch (and the previous patch to another ioctl), the
getifaddrs() now works.
Fixes #783.
Signed-off-by: Nadav Har'El
---
bsd/sys/netinet/in.cc | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/b
Add support for setup.py installing the prerequisite packages on Fedora 24.
It appears that the same list of Fedora 23 works also on Fedora 24.
Signed-off-by: Nadav Har'El
---
scripts/setup.py | 9 -
1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/scripts/setup.py b/sc
Thanks Justin.
I've been away from civilization for a week (a vacation plus a
malfunctioning laptop) so hopefully I'll be able to commit your patch
tomorrow.
The patch looks good and correct..
--
Nadav Har'El
n...@scylladb.com
On Thu, Aug 18, 2016 at 3:30 PM, Justin Cinkelj
wr
On Thu, Aug 18, 2016 at 2:41 PM, Benoît Canet
wrote:
>
> For Makefile they are a well know UNIX mess.
> One of the official documentation is here: https://www.gnu.org/
> software/make/manual/make.html but it's not easy to grok.
> You will probably end up doing like everyone and make progress by c
Hi Benoit,
On Wed, Aug 17, 2016 at 9:54 PM, Benoit Canet <
benoit.canet.cont...@gmail.com> wrote:
> This fix a bug in OSv musl that has been fixed in upstream musl
> by using netlink. Since OSv does not implement netlink add
> a function in if.cc in order to copy and interface mac address
> given
now is to modify scripts/test.py to look for the tests actually
installed on the image - not any file in build/release/tests.
--
Nadav Har'El
n...@scylladb.com
On Sun, Aug 21, 2016 at 12:29 PM, wrote:
> See <http://jenkins.cloudius-systems.com:8080/job/osv-build/1108/change
On Wed, Aug 17, 2016 at 9:54 PM, Benoit Canet <
benoit.canet.cont...@gmail.com> wrote:
> Signed-of-by: BenoÃt Canet
> ---
> bsd/porting/networking.cc | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/bsd/porting/networking.cc b/bsd/porting/networking.cc
> index 1af0ecc..92
Hi Justin,
On Tue, Aug 16, 2016 at 9:50 AM, Justin Cinkelj
wrote:
> pthread_getid is only a convience function to convert pthread_t to
> corresponding thread ID. It is added here to avoid manually casting
> pthread_t to osv::pthread at other places.
>
> Signed-off-by: Justin Cinkelj
> ---
> i
Hi Justin,
Unless I'm misunderstanding something, I think there is a race in your
code: osv_execve() spawns a new thread to run a function
thread_run_app_in_namespace(), and this will eventually set the value
pointed by pid_t *thread_id. But that happens some time in the future, and
at the time
On Sun, Aug 21, 2016 at 5:22 PM, Nadav Har'El wrote:
> Hi Justin,
>
>
>
> Unless I'm misunderstanding something, I think there is a race in your
> code: osv_execve() spawns a new thread to run a function
> thread_run_app_in_namespace(), and this will eventually s
uses this new application::run_and_join().
The osv_execve() suffered from the unnecessary thread in that the actual
application code does not run in the thread id returned by it, but rather
in a child thread. After this patch, it should return the right thread id.
Signed-off-by: Nadav Har'E
On Mon, Aug 22, 2016 at 12:31 PM, Justin Cinkelj
wrote:
>
>
> On 08/21/2016 04:38 PM, Nadav Har'El wrote:
>
>
>
> I've been thinking about this some more, and why you had to change your
> osv_execve() implementation - which already returned a thread id, but th
> Maybe with_all_app_threads gets confused, as it seems to return thread ID
> of threads belonging to orted.so, not to mpi_app.so.
> I will try to modify tst-execve/tst-execve-payload to show that, without
> using open mpi.
>
> BR Justin
>
> On 08/22/2016 02:09 PM, Nadav Ha
efore returning.
Signed-off-by: Nadav Har'El
---
core/osv_execve.cc | 11 ---
1 file changed, 8 insertions(+), 3 deletions(-)
diff --git a/core/osv_execve.cc b/core/osv_execve.cc
index 5bd1d54..a7594b3 100644
--- a/core/osv_execve.cc
+++ b/core/osv_execve.cc
@@ -18,7 +18,8 @@
On Tue, Aug 23, 2016 at 12:32 AM, Nadav Har'El wrote:
>
> On Mon, Aug 22, 2016 at 6:51 PM, Justin Cinkelj
> wrote:
>
>> Seems that here and there, osv_execve returns with thread_id still at 0,
>> instead of containing new thread id.
>> So a little race.
>&
-off-by: Nadav Har'El
---
tests/tst-run.cc | 54 ++
modules/tests/Makefile | 2 +-
2 files changed, 55 insertions(+), 1 deletion(-)
create mode 100644 tests/tst-run.cc
diff --git a/tests/tst-run.cc b/tests/tst-run.cc
new file mode 1
On Tue, Aug 23, 2016 at 1:03 AM, Nadav Har'El wrote:
>
> On Tue, Aug 23, 2016 at 12:32 AM, Nadav Har'El wrote:
>
>>
>> On Mon, Aug 22, 2016 at 6:51 PM, Justin Cinkelj
>> wrote:
>>
>>> Seems that here and there, osv_execve returns with thread
On Tue, Aug 23, 2016 at 3:28 PM, Justin Cinkelj
wrote:
>
> @nadav
> With second RFC patch, osv_execve dosn't return thread_id==0 any more.
>
>
> (code after RFC-2 pathc) The application::start_and_join() line
> sched::thread::current()->set_app_runtime(runtime()); has to finish, so
> that with_al
r only know use case.
Signed-off-by: Nadav Har'El
---
include/osv/app.hh | 7 +--
core/app.cc| 11 ---
core/osv_execve.cc | 23 ---
3 files changed, 25 insertions(+), 16 deletions(-)
diff --git a/include/osv/app.hh b/include/osv/app.hh
index dbdf
I just sent another RFC patch which should cause osv_execve to wait until
app_runtime() is the new application's.
I haven't really tested it much beyond the usual "make check" - can you
please try it out?
Thanks,
Nadav.
--
Nadav Har'El
n...@scylladb.com
On Wed, Aug 24
Hi Benoit (and Pekka), thanks. Here are some comments and questions:
--
Nadav Har'El
n...@scylladb.com
On Thu, Aug 25, 2016 at 1:07 PM, Benoit Canet <
benoit.canet.cont...@gmail.com> wrote:
> Enable "fast system calls" via the 'syscall' instruction on OS
On Thu, Aug 25, 2016 at 1:07 PM, Benoit Canet <
benoit.canet.cont...@gmail.com> wrote:
> The commented one will require manual work.
>
> Signed-off-by: Benoît Canet
> ---
> linux.cc | 56
> syscalls.h | 141 ++
>
ed on
http://wiki.lazarus.freepascal.org/Free_Pascal_on_OSv
but I wanted to not need the special linking instructions.
Signed-off-by: Nadav Har'El
---
fpc-example/module.py | 3 +++
fpc-example/Makefile | 11 +++
fpc-example/README| 1 +
fpc-example/fpc
fies errno, which it shouldn't. I think you need to save errno before
calling syscall(), and restore it after the call.
Some more minor comments below.
--
Nadav Har'El
n...@scylladb.com
On Sun, Aug 28, 2016 at 10:17 AM, Benoit Canet wrote:
> Enable "fast system calls"
s>},
> _wakeup_link = {_helper = 0xa2776188, _next =
> 0xa56baf80}, static _s_idgen = 370, static _s_reaper =
> 0x827df040,
> stat_switches = {_counter = {> = {_M_i
> = 571}, }}, stat_preemptions = {
> ---Type to continue, or q to quit---
> _
Hi Benoit, sorry for all these iterations, but there are some important
issues below (especialy the last one).
On Sun, Aug 28, 2016 at 1:11 PM, Benoit Canet <
benoit.canet.cont...@gmail.com> wrote:
> Enable "fast system calls" via the 'syscall' instruction on OSv. The
> instruction is used by Go
By the way, a question:
When running with "-V" (--verbose), do you see a warning about a stubbed
"pthread_key_delete" function?
--
Nadav Har'El
n...@scylladb.com
On Sun, Aug 28, 2016 at 12:42 PM, Nadav Har'El wrote:
>
> On Fri, Aug 26, 2016 at 2:55 PM,
d too!). It's a hack, but should be good
enough for use cases like osv_execve() which anyway uses a fresh thread
for running the application.
Signed-off-by: Nadav Har'El
---
libc/pthread.hh | 4
core/app.cc | 9 +
libc/pthread.cc | 38 ++--
Hi Justin, I hate this patch, but can you please check if it fixes your
crash-during-exit problem that you reported yesterday?
--
Nadav Har'El
n...@scylladb.com
On Sun, Aug 28, 2016 at 3:29 PM, Nadav Har'El wrote:
> The change to osv::run() (and friends) to reuse the calling
-great side of not re-using keys is that our current implementation
is limited to 128 keys so we can run out of them. But this patch doesn't make
this any worse than before.
Signed-off-by: Nadav Har'El
---
libc/pthread.cc | 15 +--
1 file changed, 13 insertions(+), 2 deletion
Looks good. Just some nitpicks about the stack stuff and debugging stuff
below.
On Sun, Aug 28, 2016 at 4:01 PM, Benoit Canet <
benoit.canet.cont...@gmail.com> wrote:
> Enable "fast system calls" via the 'syscall' instruction on OSv. The
> instruction is used by Go programs on Linux/x86-64 for sy
On Sun, Aug 28, 2016 at 4:26 PM, Benoît Canet
wrote:
>
> I opted for a signal frame because linux does it for the syscall
> instruction handler.
>
Ok. It's a bit strange, but I'll commit this for now and we can fix later
if we want.
--
You received this message because you are subscribed to t
o get this test to actually work :-)
Nadav.
--
Nadav Har'El
n...@scylladb.com
On Sun, Aug 28, 2016 at 5:38 PM, Benoit Canet <
benoit.canet.cont...@gmail.com> wrote:
> This C syscall example is just doing exit(42)
>
> Signed-off-by: Benoît Canet
> ---
> syscall/Makefil
compilation, on both single vcpu (run.py -c1) and many (-c4).
With debug build, the patch still fails - I have other fixes to make it
work.
Refs #791.
Signed-off-by: Nadav Har'El
---
arch/x64/arch-cpu.hh | 2 ++
arch/x64/processor.hh | 2 ++
arch/x64/arch-setup.cc | 6 +++---
3 files chang
Not saving and restoring the rbp register causes tst-syscall to crash in
the debug build.
Refs #791.
Signed-off-by: Nadav Har'El
---
arch/x64/entry.S | 2 ++
1 file changed, 2 insertions(+)
diff --git a/arch/x64/entry.S b/arch/x64/entry.S
index 0526aa7..c2417b8 100644
--- a/arch/x64/en
On Mon, Aug 29, 2016 at 1:47 PM, Benoît Canet
wrote:
>
> If you look as the assembly dump of loader.elf you will see save saving
> rbp and restoring it is usually the first and last thing function do.
>
> It marks the begining of a frame.
>
Yes, that is the old-style "frame pointers" way of figu
nctions as well, but this
one bothered me.
Signed-off-by: Nadav Har'El
---
arch/x64/entry.S | 1 +
1 file changed, 1 insertion(+)
diff --git a/arch/x64/entry.S b/arch/x64/entry.S
index 0526aa7..25f3cba 100644
--- a/arch/x64/entry.S
+++ b/arch/x64/entry.S
@@ -255,3 +255,4 @@ syscall_entry
this patch also code to align the stack to 16 bytes before
calling the C function. We use a nice trick to do that without using up
another register.
Signed-off-by: Nadav Har'El
---
arch/x64/entry.S | 28
1 file changed, 28 insertions(+)
diff --git a/arch/x64/ent
gnal, when it
can resemble a norma function call?
Would you have any objection that you or I clean up all these duplicate
saved stuff in syscall_entry(), and *not* emulate a signal frame?
All the CFI stuff in this function is broken anyway, and will need to be
reworked.
--
Nadav Har'El
n...@s
It seems I committed the first version of this patch. I don't know how the
second version differed. If it did, please send an incremental patch.
--
Nadav Har'El
n...@scylladb.com
On Mon, Aug 29, 2016 at 4:10 PM, Benoit Canet <
benoit.canet.cont...@gmail.com> wrote:
> This
sary.
Signed-off-by: Nadav Har'El
---
arch/x64/entry.S | 6 ++
1 file changed, 6 insertions(+)
diff --git a/arch/x64/entry.S b/arch/x64/entry.S
index 48a0a71..cdd2c61 100644
--- a/arch/x64/entry.S
+++ b/arch/x64/entry.S
@@ -166,6 +166,10 @@ syscall_entry:
.cfi_startproc simple
only then try to
skip the red zone - by that time we will have already overwritten the
caller's precious red-zone data.
>
> On Mon, Aug 29, 2016 at 4:14 PM, Nadav Har'El wrote:
>
>> The AMD64 ABI allows a function to use 128 bytes beyond the bottom of the
>> stack
On Mon, Aug 29, 2016 at 5:57 PM, Avi Kivity wrote:
>
>
> On 08/29/2016 05:14 PM, Nadav Har'El wrote:
>
>> The AMD64 ABI allows a function to use 128 bytes beyond the bottom of the
>> stack without updating the stack pointer - as an optimization known as the
>>
the registers saved on the stack, so I took
this opportunity to remove some of the useless things we saved or
saved multiple times.
Signed-off-by: Nadav Har'El
---
arch/x64/entry.S | 104 +--
1 file changed, 40 insertions(+), 64 deletions(
On Mon, Aug 29, 2016 at 8:12 PM, Avi Kivity wrote:
>
>
> On 08/29/2016 08:10 PM, Benoît Canet wrote:
>
>
>
> On Mon, Aug 29, 2016 at 7:08 PM, Avi Kivity wrote:
>
>> diff --git a/arch/x64/entry.S b/arch/x64/entry.S
>>
>>> index e265a30..e3be0bc 100644
>>> --- a/arch/x64/entry.S
>>> +++ b/arch/x64
No functional changes.
Signed-off-by: Nadav Har'El
---
arch/x64/entry.S | 166 ---
1 file changed, 86 insertions(+), 80 deletions(-)
diff --git a/arch/x64/entry.S b/arch/x64/entry.S
index 48a0a71..e265a30 100644
--- a/arch/x64/entry.S
out of the registers saved on the stack, so I took
this opportunity to remove some of the useless things we saved (like rax,
which will be overwritten by a return value anyway) or saved multiple times
(rcx, rsp).
Signed-off-by: Nadav Har'El
--
1001 - 1100 of 1373 matches
Mail list logo
