]>
xenbits.xensource.com Git - osstest/rumprun.git/log
Antti Kantee [Sat, 14 Nov 2015 13:31:42 +0000 (13:31 +0000)]
build-rr: abspath PLATFORMDIR
Antti Kantee [Fri, 13 Nov 2015 14:44:08 +0000 (14:44 +0000)]
rumprun-bake: fix grammar in usage and reword slightly
Antti Kantee [Fri, 13 Nov 2015 14:28:01 +0000 (14:28 +0000)]
build-rr: catch errors from buildrump.sh probe
Antti Kantee [Fri, 13 Nov 2015 13:58:07 +0000 (13:58 +0000)]
build-rr: run setvars also for install
Makes "./build-rr hw install" work. Without this fix the
default object directory would not get set.
Antti Kantee [Fri, 13 Nov 2015 12:41:03 +0000 (12:41 +0000)]
build-rr: add -d to synopsis
Antti Kantee [Wed, 11 Nov 2015 15:28:08 +0000 (15:28 +0000)]
rumprun launcher: fail if -D is used with iso
Antti Kantee [Thu, 5 Nov 2015 23:13:00 +0000 (23:13 +0000)]
app-tools cc: fail for -print-multiarch
Some packages (e.g. python) will do the wrong thing if the
compiler pretends to support multiarch.
with thanks to @rday
Antti Kantee [Thu, 5 Nov 2015 23:11:06 +0000 (23:11 +0000)]
app-tools cc: use basename $0 for error, not whole $0
Antti Kantee [Thu, 5 Nov 2015 21:34:49 +0000 (21:34 +0000)]
build-rr: write "config" which includes both mk/sh configs
Antti Kantee [Thu, 5 Nov 2015 16:59:02 +0000 (16:59 +0000)]
Run "make includes" so that unwind.h gets installed again.
Fixes rumprun-packages/38
Antti Kantee [Thu, 5 Nov 2015 16:33:16 +0000 (16:33 +0000)]
x86: ack insr interrupts on PIC.
Partial fix for #59 in the sense that the guest no longer
stops responding.
Still no idea why repeating the problem is only possible on
amd64 with serial console, so further exploration is necessary.
per tip from @mato
Antti Kantee [Wed, 4 Nov 2015 23:11:10 +0000 (23:11 +0000)]
g/c leftover local prototypes
no functional change
Antti Kantee [Wed, 4 Nov 2015 15:49:12 +0000 (15:49 +0000)]
build-rr: update usage
Antti Kantee [Wed, 4 Nov 2015 13:50:02 +0000 (13:50 +0000)]
rumprun: fix -W t,inet,dhcp under set -u
Antti Kantee [Wed, 4 Nov 2015 13:49:13 +0000 (13:49 +0000)]
rumprun: adjust copypasteware in error messages
Antti Kantee [Wed, 4 Nov 2015 13:35:31 +0000 (13:35 +0000)]
rumprun launcher: set -eu
Tried to go through a bunch of different scenarios, but some
fallout is possible, of course.
Antti Kantee [Wed, 4 Nov 2015 12:48:51 +0000 (12:48 +0000)]
prune unnecessary header
Antti Kantee [Wed, 4 Nov 2015 10:55:27 +0000 (10:55 +0000)]
Merge pull request #58 from gaffo/patch-1
Correct error output, $image_path is blank / not a real variable
Martin Lucina [Wed, 4 Nov 2015 10:33:03 +0000 (11:33 +0100)]
toolchain.cmake.in: Fix CMAKE_FIND_ROOT_PATH
Mike Gaffney [Wed, 4 Nov 2015 08:07:27 +0000 (00:07 -0800)]
Correct error output, $image_path is blank / not a real variable
When running, error doesn't say what it's missing, it is actually empty string in this case due to the wrong variable name being used.
Antti Kantee [Wed, 4 Nov 2015 02:06:08 +0000 (02:06 +0000)]
memalloc: nicer alloc stats print
Antti Kantee [Wed, 4 Nov 2015 01:58:14 +0000 (01:58 +0000)]
memalloc: ctassert malloc header size
Antti Kantee [Wed, 4 Nov 2015 01:52:25 +0000 (01:52 +0000)]
reduce malloc header overhead from 16 bytes to 8 bytes.
As a side effect, alignment is limited to 2GB, but I think
that's enough, given that you'd have to have a guest with >4GB
memory for larger alignments to be possible.
Antti Kantee [Wed, 4 Nov 2015 00:08:15 +0000 (00:08 +0000)]
memalloc: use __builtin_clzl() instead of for loop.
Antti Kantee [Tue, 3 Nov 2015 23:29:57 +0000 (23:29 +0000)]
memalloc: NBUCKETS -> LOCALBUCKETS
(other buckets elsewhere)
Antti Kantee [Tue, 3 Nov 2015 23:26:00 +0000 (23:26 +0000)]
memalloc: a bunch of improvements
* go directly to page allocator for large allocations
* make the test hit both small and large allocs
* code simplifications
Antti Kantee [Tue, 3 Nov 2015 21:44:51 +0000 (21:44 +0000)]
memalloc: avoid list ping-pong when allocating from core
Antti Kantee [Tue, 3 Nov 2015 21:28:09 +0000 (21:28 +0000)]
memalloc: use structs, not an union for alloc blocks
Antti Kantee [Tue, 3 Nov 2015 20:11:27 +0000 (20:11 +0000)]
pgalloc: fix comment
no fuctional change
Antti Kantee [Tue, 3 Nov 2015 19:48:33 +0000 (19:48 +0000)]
carveandlink_freechunk -> freechunk_link
more descriptional name
no functional change
Antti Kantee [Tue, 3 Nov 2015 19:42:17 +0000 (19:42 +0000)]
rumpuser_mem: use page allocator for large allocations
Antti Kantee [Tue, 3 Nov 2015 18:17:08 +0000 (18:17 +0000)]
pgalloc: use builtin_ctz/clz instead of for loops.
Antti Kantee [Tue, 3 Nov 2015 17:11:47 +0000 (17:11 +0000)]
Add bmk_pgalloc_align()
Can be used to allocate chunks with alignment > size.
Antti Kantee [Mon, 2 Nov 2015 23:52:22 +0000 (23:52 +0000)]
memalloc: g/c RCHECK never-conditionally-compiled checks
Antti Kantee [Mon, 2 Nov 2015 23:51:18 +0000 (23:51 +0000)]
use constant width types
Antti Kantee [Mon, 2 Nov 2015 23:07:58 +0000 (23:07 +0000)]
memalloc: remove unnecessary pagesize variable
Antti Kantee [Mon, 2 Nov 2015 22:52:44 +0000 (22:52 +0000)]
some comment massage
Antti Kantee [Mon, 2 Nov 2015 22:25:03 +0000 (22:25 +0000)]
make the memalloc tester run on top of bmk
Antti Kantee [Mon, 2 Nov 2015 14:44:49 +0000 (14:44 +0000)]
get install-time vars from objdir
Antti Kantee [Mon, 2 Nov 2015 13:27:37 +0000 (13:27 +0000)]
Make it clear that die produces an ERROR.
Antti Kantee [Mon, 2 Nov 2015 13:26:41 +0000 (13:26 +0000)]
Throw nice error if user tries to install non-existent build
Antti Kantee [Mon, 2 Nov 2015 13:06:30 +0000 (13:06 +0000)]
Build malloc stats by default.
Antti Kantee [Sun, 1 Nov 2015 23:46:18 +0000 (23:46 +0000)]
add hw_virtio_scsi configuration
Antti Kantee [Sun, 1 Nov 2015 23:45:12 +0000 (23:45 +0000)]
update buildrump.sh and src-netbsd
Antti Kantee [Sun, 1 Nov 2015 22:11:47 +0000 (22:11 +0000)]
g/c old way of specifying PCI support on platform/board
Antti Kantee [Sat, 31 Oct 2015 12:23:21 +0000 (12:23 +0000)]
Disable "stray interrupt" print
Doesn't convey anything useful in its current state.
Antti Kantee [Sat, 31 Oct 2015 12:19:26 +0000 (12:19 +0000)]
When building with -s, default destdir to rumprun-extsrc.
Mirrors what object dirs do.
Antti Kantee [Fri, 30 Oct 2015 20:49:02 +0000 (20:49 +0000)]
Merge pull request #57 from mattgray/master
rumprun launcher: trim od output to support OS X
Matt Gray [Fri, 30 Oct 2015 19:23:48 +0000 (19:23 +0000)]
AUTHORS updated
Matt Gray [Fri, 30 Oct 2015 20:00:53 +0000 (20:00 +0000)]
rumprun launcher: OS X support: use printf to fix output of od
Antti Kantee [Fri, 30 Oct 2015 01:43:19 +0000 (01:43 +0000)]
checkprevbuilds is now necessary only for xen
Antti Kantee [Fri, 30 Oct 2015 01:40:43 +0000 (01:40 +0000)]
platform/hw: create header links into objdir
hw platform now builds completely into objdir, meaning multiple
parallel builds with different options are possible from one
working directory.
Antti Kantee [Thu, 29 Oct 2015 19:46:37 +0000 (19:46 +0000)]
oops, revert buildrump.sh revision pointing to local repo
Antti Kantee [Thu, 29 Oct 2015 18:20:36 +0000 (18:20 +0000)]
rumprun launcher: remove use or readlink(1)
As reported by @mattgray on irc, the one available on OS X is
incompatible, and since resolving the canonical path doesn't
seem to be absolutely necessary, let's simply opt to not do it
at all.
Antti Kantee [Thu, 29 Oct 2015 16:40:04 +0000 (16:40 +0000)]
build-rr: set -eu
Both were previously set, but now they're set in the same place.
No functional change intended.
Antti Kantee [Thu, 29 Oct 2015 01:24:34 +0000 (01:24 +0000)]
Some archs: fix SRCS/ASMS to be those instead of .o
Antti Kantee [Thu, 29 Oct 2015 01:15:14 +0000 (01:15 +0000)]
Build hw platform into $RROBJ.
Antti Kantee [Wed, 28 Oct 2015 22:05:01 +0000 (22:05 +0000)]
set --sysroot for compiler wrapper
Makes cc -print-sysroot work. That said, why do we still
need -isystem and -L in our specs? (yes, I tried without them,
lossage generated)
Antti Kantee [Wed, 28 Oct 2015 18:27:26 +0000 (18:27 +0000)]
makepseudolinkstubs: Use ${NM}, not nm
Antti Kantee [Wed, 28 Oct 2015 17:46:46 +0000 (17:46 +0000)]
Fix include path for non-BUILDRR case.
Antti Kantee [Wed, 28 Oct 2015 17:35:26 +0000 (17:35 +0000)]
update buildrump.sh
Antti Kantee [Wed, 28 Oct 2015 17:25:39 +0000 (17:25 +0000)]
Check if qemu supports -no-kvm
Some really old qemus might not.
Antti Kantee [Wed, 28 Oct 2015 13:10:34 +0000 (13:10 +0000)]
fix error branch pastos in previous
Antti Kantee [Wed, 28 Oct 2015 13:07:42 +0000 (13:07 +0000)]
Use ${READELF:=readelf} instead of hardcoding objdump.
Use of readelf also mirrors rumprun-bake, so you need only
one binary dumping tool now (useful in env where the GNU'ish
toolchain isn't standard).
Antti Kantee [Wed, 28 Oct 2015 12:48:12 +0000 (12:48 +0000)]
xen ldscripts: include .text.* in text (like on hw)
Might fix problems reported here:
https://www.freelists.org/post/rumpkernel-users/Issues-linking-executable-for-Rumprun-with-additional-library,4
Antti Kantee [Wed, 28 Oct 2015 12:13:09 +0000 (12:13 +0000)]
Use destdir/rumprun-$MACHINE_GNU_ARCH$ as a GNU-like prefix.
Antti Kantee [Wed, 28 Oct 2015 01:35:55 +0000 (01:35 +0000)]
Put rumptools into the object directory.
Antti Kantee [Wed, 28 Oct 2015 01:34:32 +0000 (01:34 +0000)]
Update buildrump.sh and use new "probe" method.
The new probe method doesn't require any directories
to be created.
Antti Kantee [Tue, 27 Oct 2015 23:41:38 +0000 (23:41 +0000)]
Use MACHINE_GNU_ARCH, not MACHINE_ARCH
That matches the terms (and make variables!) with the NetBSD
Makefiles and avoids mismatch lossage.
Antti Kantee [Tue, 27 Oct 2015 23:02:21 +0000 (23:02 +0000)]
platform/hw: need to use i386 as MACHINE_ARCH here
Antti Kantee [Tue, 27 Oct 2015 22:45:45 +0000 (22:45 +0000)]
build-rr: don't override correct MACHINE_ARCH.
The lossage stems from the fact that NetBSD uses i386 as
MACHINE, i386 as MACHINE_ARCH, and i486 as MACHINE_GNU_ARCH.
We want to follow the MACHINE_GNU_ARCH since it actually makes sense.
Antti Kantee [Tue, 27 Oct 2015 22:23:46 +0000 (22:23 +0000)]
Ok, maybe the platform Makefiles aren't ready for -j
Antti Kantee [Tue, 27 Oct 2015 22:08:50 +0000 (22:08 +0000)]
build-rr: run platform make with -j
Antti Kantee [Tue, 27 Oct 2015 22:05:08 +0000 (22:05 +0000)]
Install specs to lib instead of share root.
specs are now split into arch and machine specs.
Antti Kantee [Tue, 27 Oct 2015 21:21:23 +0000 (21:21 +0000)]
build-rr: simplify MACHINE_ARCH handling.
Antti Kantee [Tue, 27 Oct 2015 18:37:57 +0000 (18:37 +0000)]
Install brprintmetainfo, it's used by cc
Fixes all kinds of problems in not detecting compile failures.
Problem reported by meena on irc.
Wei Liu [Tue, 27 Oct 2015 17:54:19 +0000 (17:54 +0000)]
xenbus: add two more mb()s
Signed-off-by: Wei Liu <liuw@liuw.name>
Antti Kantee [Tue, 27 Oct 2015 17:38:05 +0000 (17:38 +0000)]
rumprun-bake: change regex to a form which OS X grep groks
Antti Kantee [Tue, 27 Oct 2015 15:52:05 +0000 (15:52 +0000)]
Make it rumprun[-MACHINE_ARCH[-PLATFORM]]
That way MACHINE_ARCH (if it exists in the particular name),
is always in the same spot, and the same holds for PLATFORM.
Antti Kantee [Tue, 27 Oct 2015 15:25:47 +0000 (15:25 +0000)]
create the bmk.ldscript link into objdir
Antti Kantee [Tue, 27 Oct 2015 15:15:57 +0000 (15:15 +0000)]
Assign intermediate variables in Makefile
no functional change
Antti Kantee [Tue, 27 Oct 2015 15:12:25 +0000 (15:12 +0000)]
call it pseudolinkstubs.o, install into lib/rumprun-arch
Antti Kantee [Tue, 27 Oct 2015 15:09:38 +0000 (15:09 +0000)]
call it rumprun.o, install into rumprun-platform-arch
Antti Kantee [Tue, 27 Oct 2015 15:07:49 +0000 (15:07 +0000)]
Fix -I for BUILDRR=false case.
Antti Kantee [Tue, 27 Oct 2015 14:05:51 +0000 (14:05 +0000)]
Update header location to re-fix kernonly test
Antti Kantee [Tue, 27 Oct 2015 13:53:53 +0000 (13:53 +0000)]
Install headers into destdir/include/rumprun.
That way they don't conflict with anything already
in destdir/include.
Antti Kantee [Tue, 27 Oct 2015 13:37:47 +0000 (13:37 +0000)]
Need separate BRLIBEXECINST and BRLIBEXECDEST
Antti Kantee [Tue, 27 Oct 2015 13:17:36 +0000 (13:17 +0000)]
libexecdir is in RRDEST, not INSTALLDIR.
(latter varies depending on if it's a manual "make" or not)
Antti Kantee [Mon, 26 Oct 2015 16:59:25 +0000 (16:59 +0000)]
fix to previous: completely remove previnst check
Antti Kantee [Mon, 26 Oct 2015 16:11:48 +0000 (16:11 +0000)]
build-rr: remove previnst check, not needed anymore.
Antti Kantee [Mon, 26 Oct 2015 16:09:17 +0000 (16:09 +0000)]
g/c debugging aid that slipped into the public tree
Antti Kantee [Mon, 26 Oct 2015 15:55:42 +0000 (15:55 +0000)]
pseudolinkstubs.o needs MACHINE_ARCH in the name.
Antti Kantee [Mon, 26 Oct 2015 15:53:09 +0000 (15:53 +0000)]
kernonly test: change -L to match new reality.
Antti Kantee [Mon, 26 Oct 2015 02:55:16 +0000 (02:55 +0000)]
Install libs to non-conflicting subdirs.
Platform-dependent libs go into lib/rumprun-platform-machine_arch
and platform-independent libs go into lib/rumprun-machine_arch.
(all libs are arch-dependent, naturally)
Allows xen and hw and various machine_arch installations
to coexist.
Antti Kantee [Mon, 26 Oct 2015 15:02:48 +0000 (15:02 +0000)]
Turn duplicate address detection off.
Otherwise the network is unavailable for several seconds
after boot. There's really no good solution here, but this
one is probably the best, i.e. hoping that things don't go wrong.
nb. turning DAD off cannot be implemented properly with our
current src-netbsd because the dad_count sysctl node is missing.
The current hack will prevent from linking a non-kernonly image
without networking, but I don't think anyone is using that
currently anyway. To be fixed properly after I next update
src-netbsd.
Fixes issue #56.
Antti Kantee [Mon, 26 Oct 2015 14:44:48 +0000 (14:44 +0000)]
Add "-extsrc" to default object directory name for -s builds.
Wei Liu [Mon, 26 Oct 2015 09:56:07 +0000 (09:56 +0000)]
xenbus: notify the other end when necessary
The xenbus thread didn't send notification to other end when it expected
more data or consumed responses, which led to stalling the ring from
time to time.
Fix this by sending notification to the other end at the right places.
Signed-off-by: Wei Liu <liuw@liuw.name>
Antti Kantee [Mon, 26 Oct 2015 00:57:52 +0000 (00:57 +0000)]
Default kernonly objdir to end with "-kernonly"
Antti Kantee [Mon, 26 Oct 2015 00:55:46 +0000 (00:55 +0000)]
build-rr: nuke prevbuild platform from error message
Mirrors how the platform became a non-issue with (partial)
objdir build support.
Antti Kantee [Sun, 25 Oct 2015 16:53:32 +0000 (16:53 +0000)]
pull in new src-netbsd (and buildrump.sh)