Re: [Qemu-devel] [PATCH] arm eabi TLS
Paul Brook wrote: - It would be good to limit the changes in the CPU emulation code to handle the TLS. For example, on MIPS, the TLS register must not be stored in the CPU state. Same for ARM. I disagree. The TLS register is part of the CPU state. On many machines (including ARMv6 CPUs) it's an actual CPU register. I'm fairly sure the same is true for recent MIPS revisions. If some CPUs implement it in hardware, then I agree. Fabrice.
[Qemu-devel] [PPC] segfault with crnand instruction
Hi everybody,
I get a segfault with qemu-ppc-softmmu, Source downloaded yesterday, 12/12/O7.
After dump my program, I see this instruction fails at nip
4d d6 d1 c2 crnand 4*cr3+eq,4*cr5+eq,4*cr6+eq
in function
translate.c: static inline int gen_intermediate_code_internal
l 6218
/* Is opcode *REALLY* valid ? */
if (unlikely(handler->handler == &gen_invalid)) {
handler->handler is not valid.
Could you confirm ?
Regards,
Claude
Re: [Qemu-devel] [PATCH] Fix compilation with Cygwin
I just want to confirm that the fix works also for MinGW. At first I dind't like the word "type" and would have preferred something like "if_type", but I realized that drives[i].type is just as readable as drives[i].if_type. - Original Message From: Hervé Poussineau <[EMAIL PROTECTED]> To: qemu-devel@nongnu.org Sent: Thursday, 13 December, 2007 9:42:07 AM Subject: [Qemu-devel] [PATCH] Fix compilation with Cygwin Hi, Cygwin compiler doesn't seem to like variables/structure members which are named "interface". This patch renames them to "type". Hervé __ Sent from Yahoo! Mail - a smarter inbox http://uk.mail.yahoo.com
Re: [Qemu-devel] [PATCH] arm eabi TLS
On Thu, Dec 13, 2007 at 01:21:03AM +, Paul Brook wrote: > I disagree. The TLS register is part of the CPU state. On many machines > (including ARMv6 CPUs) it's an actual CPU register. I'm fairly sure the same > is true for recent MIPS revisions. That's correct, though I don't know if there is silicon to match yet. -- Daniel Jacobowitz CodeSourcery
Re: [Qemu-devel][Patch] Windows build broken
Here is a patch using WIN32_LEAN_AND_MEAN which tries not to break dsound (and vl.c). I could only test it using cross compilation from Linux, so please send comments. Patch description: * add WIN32_LEAN_AND_MEAN to all inclusions of windows.h * remove windows.h from vl.c (it is already included otherwise) * add mmsystem.h to vl.c and dsoundaudio.c (they need it and got it from windows.h) * fixed indentation in vl.c (wrong indentation was caused by an earlier patch from me) It will be difficult to avoid using variable names like "interface" in the future, so I still think using a minimal set of includes is the better (and faster) solution. Anyway, Windows users need a working solution, so it would be nice if one of these patches could be applied to CVS HEAD. Thank you. Stefan JonY schrieb: > Stefan Weil wrote: >> basetyps.h is included by windows.h / rpc.h. QEMU does not need it, so >> you can avoid it like this: >> >> #define WIN32_LEAN_AND_MEAN >> #include >> >> WIN32_LEAN_AND_MEAN reduces the number of includes in windows.h >> and increases compilation speed. And you don't have to rename >> variables like interface :-) >> >> Regards, >> Stefan > > Hi, > > I would prefer Eduardo's patch, defining WIN32_LEAN_AND_MEAN breaks > dsound. > > Thanks. > > > Index: qemu_socket.h === RCS file: /sources/qemu/qemu/qemu_socket.h,v retrieving revision 1.2 diff -u -r1.2 qemu_socket.h --- qemu_socket.h 21 Dec 2006 19:46:43 - 1.2 +++ qemu_socket.h 13 Dec 2007 20:09:04 - @@ -3,7 +3,7 @@ #define QEMU_SOCKET_H #ifdef _WIN32 - +#define WIN32_LEAN_AND_MEAN #include #include #include Index: osdep.c === RCS file: /sources/qemu/qemu/osdep.c,v retrieving revision 1.20 diff -u -r1.20 osdep.c --- osdep.c 18 Nov 2007 01:44:35 - 1.20 +++ osdep.c 13 Dec 2007 20:09:05 - @@ -37,6 +37,7 @@ #include "sysemu.h" #ifdef _WIN32 +#define WIN32_LEAN_AND_MEAN #include #elif defined(_BSD) #include Index: vl.c === RCS file: /sources/qemu/qemu/vl.c,v retrieving revision 1.379 diff -u -r1.379 vl.c --- vl.c 11 Dec 2007 23:23:51 - 1.379 +++ vl.c 13 Dec 2007 20:09:07 - @@ -109,7 +109,7 @@ #ifdef _WIN32 #include #include -#include +#include #define getopt_long_only getopt_long #define memalign(align, size) malloc(size) #endif @@ -8937,7 +8937,7 @@ s->down_script[0]) launch_script(s->down_script, ifname, s->fd); } -} +} } #endif return 0; Index: tap-win32.c === RCS file: /sources/qemu/qemu/tap-win32.c,v retrieving revision 1.7 diff -u -r1.7 tap-win32.c --- tap-win32.c 17 Nov 2007 17:14:38 - 1.7 +++ tap-win32.c 13 Dec 2007 20:09:07 - @@ -30,6 +30,7 @@ #include "net.h" #include "sysemu.h" #include +#define WIN32_LEAN_AND_MEAN #include /* NOTE: PCIBus is redefined in winddk.h */ Index: qemu-common.h === RCS file: /sources/qemu/qemu/qemu-common.h,v retrieving revision 1.3 diff -u -r1.3 qemu-common.h --- qemu-common.h 19 Nov 2007 00:38:33 - 1.3 +++ qemu-common.h 13 Dec 2007 20:09:07 - @@ -28,6 +28,7 @@ #endif #ifdef _WIN32 +#define WIN32_LEAN_AND_MEAN #include #define fsync _commit #define lseek _lseeki64 Index: exec.c === RCS file: /sources/qemu/qemu/exec.c,v retrieving revision 1.119 diff -u -r1.119 exec.c --- exec.c 12 Dec 2007 01:16:22 - 1.119 +++ exec.c 13 Dec 2007 20:09:08 - @@ -19,6 +19,7 @@ */ #include "config.h" #ifdef _WIN32 +#define WIN32_LEAN_AND_MEAN #include #else #include Index: qemu-img.c === RCS file: /sources/qemu/qemu/qemu-img.c,v retrieving revision 1.25 diff -u -r1.25 qemu-img.c --- qemu-img.c 11 Nov 2007 03:33:13 - 1.25 +++ qemu-img.c 13 Dec 2007 20:09:09 - @@ -26,6 +26,7 @@ #include #ifdef _WIN32 +#define WIN32_LEAN_AND_MEAN #include #endif Index: kqemu.c === RCS file: /sources/qemu/qemu/kqemu.c,v retrieving revision 1.19 diff -u -r1.19 kqemu.c --- kqemu.c 17 Sep 2007 08:09:44 - 1.19 +++ kqemu.c 13 Dec 2007 20:09:09 - @@ -19,6 +19,7 @@ */ #include "config.h" #ifdef _WIN32 +#define WIN32_LEAN_AND_MEAN #include #include #else Index: audio/dsoundaudio.c === RCS file: /sources/qemu/qemu/audio/dsoundaudio.c,v retrieving revision 1.4 diff -u -r1.4 dsoundaudio.c --- audio/dsoundaudio.c 17 Nov 2007 17:35:54 - 1.4 +++ audio/dsoundaudio.c 13 Dec 2007 20:09:09 - @@ -32,7 +32,9 @@ #define AUDIO_CAP "dsound" #include "audio_int.h" +#define WIN32_LEAN_AN
[Qemu-devel] multiple VM on one tap device
Hello all. Is it possible run several VM connected with one tap device? I tried to run two VM with allready exist tap0 but when I started second I've got message: "warning: could not configure /dev/net/tun: no virtual network emulation Could not initialize device 'tap' ". I did following under root: -bash-3.00# tunctl -b -t tap0 -bash-3.00# ifconfig tap0 192.168.4.1 netmask 255.255.255.0 up -bash-3.00# ifconfig tap0 tap0 Link encap:Ethernet HWaddr 00:FF:A1:81:05:B0 inet addr:192.168.4.1 Bcast:192.168.4.255 Mask:255.255.255.0 UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:0 errors:0 dropped:0 overruns:0 frame:0 TX packets:0 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:500 RX bytes:0 (0.0 b) TX bytes:0 (0.0 b) Then started VM with following parameters: 1. -net nic,macaddr=00:1a:4b:34:74:43 -net tap,vlan=0,ifname=tap0,host=sv3,script=no 2. -net nic,macaddr=00:1a:4b:34:74:44 -net tap,vlan=0,ifname=tap0,host=sv4,script=no I did it under linux kernel 2.6.23.9. What does it mean fd in -net tap,fd ? File discriptor of /dev/net/tun or tap device.
[Qemu-devel] [PATCH] Reduce redundant timer rearming
The dynticks code in qemu sets and gets timers very often. These are the system calls (strace -c) of qemu/kvm running an idle Linux kernel at 250Hz for 10 seconds: % time seconds usecs/call callserrors syscall -- --- --- - - 99.670.014391 1 23248 3187 ioctl 0.330.47 0 43643 clock_gettime 0.000.00 0 6375 gettimeofday 0.000.00 0 6708 select 0.000.00 0 3187 rt_sigaction 0.000.00 0 9562 6375 rt_sigtimedwait 0.000.00 0 10311 timer_settime 0.000.00 0 12750 timer_gettime -- --- --- - - 100.000.014438115784 9562 total The qemu_rearm_alarm_timer() function looks at vm_clock as well as rt_clock timers, but is called from qemu_run_timers() which looks at just one queue. When an rt_clock timer has expired, the vm_clock iteration will rearm with MIN_TIMER_REARM_US. This is not needed, since the timer in question will be removed right away when qemu_run_timers() is run on the rt_clock queue. Moving the rearm call to after the two calls of qemu_run_timers() helps a lot: % time seconds usecs/call callserrors syscall -- --- --- - - 99.740.012590 1 22766 2706 ioctl 0.260.33 0 27044 clock_gettime 0.000.00 0 5413 gettimeofday 0.000.00 0 5745 select 0.000.00 0 2706 rt_sigaction 0.000.00 0 8119 5413 rt_sigtimedwait 0.000.00 0 5215 timer_settime 0.000.00 0 5413 timer_gettime -- --- --- - - 100.000.012623 82421 8119 total Patch (based on kvm-userspace git) attached. Cheers, Anders. diff --git a/qemu/vl.c b/qemu/vl.c index 80ceb2f..2cd580d 100644 --- a/qemu/vl.c +++ b/qemu/vl.c @@ -1063,7 +1063,6 @@ static void qemu_run_timers(QEMUTimer **ptimer_head, int64_t current_time) /* run the callback (the timer list can be modified) */ ts->cb(ts->opaque); } -qemu_rearm_alarm_timer(alarm_timer); } int64_t qemu_get_clock(QEMUClock *clock) @@ -7216,6 +7215,8 @@ void main_loop_wait(int timeout) qemu_run_timers(&active_timers[QEMU_TIMER_REALTIME], qemu_get_clock(rt_clock)); +qemu_rearm_alarm_timer(alarm_timer); + /* Check bottom-halves last in case any of the earlier events triggered them. */ qemu_bh_poll();
Re: [Qemu-devel] Mac OS X build failure (qemu-img)
Am 12.12.2007 um 13:08 schrieb Alexander Graf: Does it work if you specify "-framework CoreFoundation -framework IOKit" in your LDFLAGS? That's what was needed on my Leopard x86_64 build. If I configure with LDFLAGS: LDFLAGS="-framework CoreFoundation -framework IOKit" ./configure -- prefix=/Users/andreas/Q/latest --target-list=sparc-softmmu --enable- cocoa it now ends like this: gcc-3.3 -Wall -O2 -g -fno-strict-aliasing -I. -I.. -I/Users/andreas/Q/ myqemu/target-sparc -I/Users/andreas/Q/myqemu -D__powerpc__ - D_GNU_SOURCE -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -I/Users/ andreas/Q/myqemu/fpu -I/Users/andreas/Q/myqemu/slirp -mdynamic-no- pic -c -o cocoa.o /Users/andreas/Q/myqemu/cocoa.m In file included from /System/Library/Frameworks/Foundation.framework/ Headers/Foundation.h:15, from /System/Library/Frameworks/Cocoa.framework/ Headers/Cocoa.h:12, from /Users/andreas/Q/myqemu/cocoa.m:38: /System/Library/Frameworks/Foundation.framework/Headers/NSArray.h:62: error: parse error before "__attribute__" /System/Library/Frameworks/Foundation.framework/Headers/NSArray.h:66: error: parse error before "__attribute__" In file included from /System/Library/Frameworks/Foundation.framework/ Headers/NSAttributedString.h:6, from /System/Library/Frameworks/Foundation.framework/ Headers/Foundation.h:16, from /System/Library/Frameworks/Cocoa.framework/ Headers/Cocoa.h:12, from /Users/andreas/Q/myqemu/cocoa.m:38: /System/Library/Frameworks/Foundation.framework/Headers/NSDictionary.h: 45: error: parse error before "__attribute__" /System/Library/Frameworks/Foundation.framework/Headers/NSDictionary.h: 50: error: parse error before "__attribute__" In file included from /System/Library/Frameworks/Foundation.framework/ Headers/NSHashTable.h:5, from /System/Library/Frameworks/Foundation.framework/ Headers/Foundation.h:43, from /System/Library/Frameworks/Cocoa.framework/ Headers/Cocoa.h:12, from /Users/andreas/Q/myqemu/cocoa.m:38: /System/Library/Frameworks/Foundation.framework/Headers/ NSPointerFunctions.h:59: error: parse error before '@' token In file included from /System/Library/Frameworks/Foundation.framework/ Headers/NSKeyValueCoding.h:9, from /System/Library/Frameworks/Foundation.framework/ Headers/Foundation.h:49, from /System/Library/Frameworks/Cocoa.framework/ Headers/Cocoa.h:12, from /Users/andreas/Q/myqemu/cocoa.m:38: /System/Library/Frameworks/Foundation.framework/Headers/NSSet.h:45: error: parse error before "__attribute__" /System/Library/Frameworks/Foundation.framework/Headers/NSSet.h:50: error: parse error before "__attribute__" In file included from /System/Library/Frameworks/AppKit.framework/ Headers/AppKit.h:19, from /System/Library/Frameworks/Cocoa.framework/ Headers/Cocoa.h:13, from /Users/andreas/Q/myqemu/cocoa.m:38: /System/Library/Frameworks/AppKit.framework/Headers/NSApplication.h: 68: error: parse error before "__weak" In file included from /System/Library/Frameworks/AppKit.framework/ Headers/AppKit.h:78, from /System/Library/Frameworks/Cocoa.framework/ Headers/Cocoa.h:13, from /Users/andreas/Q/myqemu/cocoa.m:38: /System/Library/Frameworks/AppKit.framework/Headers/NSPrintPanel.h:57: error: parse error before '@' token In file included from /System/Library/Frameworks/AppKit.framework/ Headers/AppKit.h:95, from /System/Library/Frameworks/Cocoa.framework/ Headers/Cocoa.h:13, from /Users/andreas/Q/myqemu/cocoa.m:38: /System/Library/Frameworks/AppKit.framework/Headers/NSTrackingArea.h: 44: error: parse error before "__weak" In file included from /System/Library/Frameworks/AppKit.framework/ Headers/NSTextView.h:16, from /System/Library/Frameworks/AppKit.framework/ Headers/AppKit.h:112, from /System/Library/Frameworks/Cocoa.framework/ Headers/Cocoa.h:13, from /Users/andreas/Q/myqemu/cocoa.m:38: /System/Library/Frameworks/AppKit.framework/Headers/ NSTextInputClient.h:17: error: parse error before '@' token /System/Library/Frameworks/AppKit.framework/Headers/ NSTextInputClient.h:62: error: parse error before '@' token In file included from /System/Library/Frameworks/AppKit.framework/ Headers/NSObjectController.h:10, from /System/Library/Frameworks/AppKit.framework/ Headers/AppKit.h:158, from /System/Library/Frameworks/Cocoa.framework/ Headers/Cocoa.h:13, from /Users/andreas/Q/myqemu/cocoa.m:38: /System/Library/Frameworks/CoreData.framework/Headers/NSFetchRequest.h: 24: error: parse error before "__weak" In file included from /System/Library/Frameworks/CoreData.framework/ Headers/NSAttributeDescription.h:10, from /System/Library/Frameworks/Ap
Re: [Qemu-devel] [PATCH 2] Reduce redundant timer rearming
(resending with the subscribed sender address)
> % time seconds usecs/call callserrors syscall
> -- --- --- - -
> 99.740.012590 1 22766 2706 ioctl
>0.260.33 0 27044 clock_gettime
>0.000.00 0 5413 gettimeofday
>0.000.00 0 5745 select
>0.000.00 0 2706 rt_sigaction
>0.000.00 0 8119 5413 rt_sigtimedwait
>0.000.00 0 5215 timer_settime
>0.000.00 0 5413 timer_gettime
> -- --- --- - -
> 100.000.012623 82421 8119 total
Here is another patch that implements a dirty bit for the timer lists.
The bit is set when timers are modified or expire. Rearming clears the
bit (and only happens if it is set).
Numbers improve to this:
% time seconds usecs/call callserrors syscall
-- --- --- - -
100.000.13 0 21813 clock_gettime
0.000.00 0 22789 2729 ioctl
0.000.00 0 5459 gettimeofday
0.000.00 0 5790 select
0.000.00 0 2729 rt_sigaction
0.000.00 0 8188 5459 rt_sigtimedwait
0.000.00 0 2729 timer_settime
0.000.00 0 2752 timer_gettime
-- --- --- - -
100.000.13 72249 8188 total
I am not so sure about this one, so comments are appreciated.
Best regards,
Anders.
diff --git a/qemu/vl.c b/qemu/vl.c
index 2cd580d..924736e 100644
--- a/qemu/vl.c
+++ b/qemu/vl.c
@@ -816,6 +816,7 @@ struct qemu_alarm_timer {
};
#define ALARM_FLAG_DYNTICKS 0x1
+#define ALARM_FLAG_MODIFIED 0x2
static inline int alarm_has_dynticks(struct qemu_alarm_timer *t)
{
@@ -827,6 +828,11 @@ static void qemu_rearm_alarm_timer(struct qemu_alarm_timer *t)
if (!alarm_has_dynticks(t))
return;
+if (!(t->flags & ALARM_FLAG_MODIFIED))
+return;
+
+t->flags &= ~(ALARM_FLAG_MODIFIED);
+
t->rearm(t);
}
@@ -989,6 +995,8 @@ void qemu_del_timer(QEMUTimer *ts)
{
QEMUTimer **pt, *t;
+alarm_timer->flags |= ALARM_FLAG_MODIFIED;
+
/* NOTE: this code must be signal safe because
qemu_timer_expired() can be called from a signal. */
pt = &active_timers[ts->clock->type];
@@ -1182,6 +1190,7 @@ static void host_alarm_handler(int host_signum)
#endif
CPUState *env = cpu_single_env;
if (env) {
+alarm_timer->flags |= ALARM_FLAG_MODIFIED;
/* stop the currently executing cpu because a timer occured */
cpu_interrupt(env, CPU_INTERRUPT_EXIT);
#ifdef USE_KQEMU
Re: [Qemu-devel] [PATCH 3] Reduce redundant timer rearming
The VGA refresh (a 33 Hz timer) is running even with -nographics,
apparently doing nothing. The vga_hw_update() call ends up just
returning, because the display depth is zero.
This patch removes the dummy refresh handler, and thus the GUI refresh
timer.
With an idle dyntick Linux guest in qemu/kvm, there are now only two
timers per second (those are from the rtc). Another 10 seconds strace:
% time seconds usecs/call callserrors syscall
-- --- --- - -
100.000.12 0 376020 ioctl
0.000.00 041 select
0.000.00 020 rt_sigaction
0.000.00 06141 rt_sigtimedwait
0.000.00 020 timer_settime
0.000.00 020 timer_gettime
0.000.00 0 162 clock_gettime
-- --- --- - -
100.000.12 408461 total
Cheers,
Anders.
diff --git a/qemu/vl.c b/qemu/vl.c
index 30c9537..8d67314 100644
--- a/qemu/vl.c
+++ b/qemu/vl.c
@@ -4902,13 +4902,6 @@ static void dumb_resize(DisplayState *ds, int w, int h)
{
}
-static void dumb_refresh(DisplayState *ds)
-{
-#if defined(CONFIG_SDL)
-vga_hw_update();
-#endif
-}
-
static void dumb_display_init(DisplayState *ds)
{
ds->data = NULL;
@@ -4916,7 +4909,7 @@ static void dumb_display_init(DisplayState *ds)
ds->depth = 0;
ds->dpy_update = dumb_update;
ds->dpy_resize = dumb_resize;
-ds->dpy_refresh = dumb_refresh;
+ds->dpy_refresh = NULL;
}
/***/
Re: [Qemu-devel][Patch] Windows build broken
Stefan Weil wrote: Here is a patch using WIN32_LEAN_AND_MEAN which tries not to break dsound (and vl.c). I could only test it using cross compilation from Linux, so please send comments. Patch description: * add WIN32_LEAN_AND_MEAN to all inclusions of windows.h * remove windows.h from vl.c (it is already included otherwise) * add mmsystem.h to vl.c and dsoundaudio.c (they need it and got it from windows.h) * fixed indentation in vl.c (wrong indentation was caused by an earlier patch from me) It will be difficult to avoid using variable names like "interface" in the future, so I still think using a minimal set of includes is the better (and faster) solution. Anyway, Windows users need a working solution, so it would be nice if one of these patches could be applied to CVS HEAD. Thank you. Stefan JonY schrieb: Stefan Weil wrote: basetyps.h is included by windows.h / rpc.h. QEMU does not need it, so you can avoid it like this: #define WIN32_LEAN_AND_MEAN #include WIN32_LEAN_AND_MEAN reduces the number of includes in windows.h and increases compilation speed. And you don't have to rename variables like interface :-) Regards, Stefan Hi, I would prefer Eduardo's patch, defining WIN32_LEAN_AND_MEAN breaks dsound. Thanks. Ok, this patch works as intended. Thanks for the patch! Slightly off topic, but still on Windows builds broken. For some odd reason, dyngen-exec.h is always conflicting with stdint.h(?!!) for me. Can somebody explain why the typedefs are in dyngen-exec.h? Mingw does not have ffs, localtime_r, gmtime_r, which breaks when compiling some files. I propose adding the following for *_r functions on mingw hosts. #define localtime_r(x, y) memcpy(y, localtime(x), sizeof(y)) #define gmtime_r(x, y) memcpy(y, gmtime(x), sizeof(y)) There are many implementations of ffs(), but I'm unsure what it does :(
