Skip to contentSkip to conversation

Fire SIGIO when IO requests are completed

RFC PATCH In review Anthony Liguori 4 participants 7 messages Oct 6, 2008 – Oct 7, 2008

lore
Share & tools
lore

Anthony Liguori

author
#0 lore raw

RFC PATCH 1/1 Fire SIGIO when IO requests are completed

Hi Blue Swirl,

Can you test the attached patch with sparc-softmmu? Better yet, do you have an image, or a recipe for an image that I could use to try and reproduce the problem? The sparc test code on the QEMU website just uses a ram disk so the regression isn't evident with it.

Regards,

Anthony Liguori

block-raw-posix.c +6 -0
index c0404fa..9b136ee 100644
@@ -591,6 +591,12 @@ static int posix_aio_init(void)
 
     fcntl(s->fd, F_SETFL, O_NONBLOCK);
 
+    {
+	extern void enable_sigio_timer(int fd);
+
+	enable_sigio_timer(s->fd);
+    }
+
     qemu_aio_set_fd_handler(s->fd, posix_aio_read, NULL, posix_aio_flush, s);
 
 #if defined(__linux__)
qemu-tool.c +4 -0
index 63e2056..c37fdea 100644
@@ -26,6 +26,10 @@ struct QEMUBH
     void *opaque;
 };
 
+void enable_sigio_timer(int fd)
+{
+}
+
 void term_printf(const char *fmt, ...)
 {
 }
vl.c +3 -3
index c94fdc0..8f66a90 100644
@@ -1307,7 +1307,7 @@ static void host_alarm_handler(int host_signum)
         last_clock = ti;
     }
 #endif
-    if (alarm_has_dynticks(alarm_timer) ||
+    if (1 || alarm_has_dynticks(alarm_timer) ||
         (!use_icount &&
             qemu_timer_expired(active_timers[QEMU_TIMER_VIRTUAL],
                                qemu_get_clock(vm_clock))) ||

@@ -1383,7 +1383,7 @@ static uint64_t qemu_next_deadline_dyntick(void)
 
 #define RTC_FREQ 1024
 
-static void enable_sigio_timer(int fd)
+void enable_sigio_timer(int fd)
 {
     struct sigaction act;
 

@@ -1393,7 +1393,7 @@ static void enable_sigio_timer(int fd)
     act.sa_handler = host_alarm_handler;
 
     sigaction(SIGIO, &act, NULL);
-    fcntl(fd, F_SETFL, O_ASYNC);
+    fcntl(fd, F_SETFL, O_ASYNC | fcntl(fd, F_GETFL));
     fcntl(fd, F_SETOWN, getpid());
 }
#1 lore raw
Quoted 6 lines · Anthony Liguori

Hi Blue Swirl,

Can you test the attached patch with sparc-softmmu? Better yet, do you
have an image, or a recipe for an image that I could use to try and reproduce the problem? The sparc test code on the QEMU website just uses a ram disk so the regression isn't evident with it.

For quick testing, Splack 8 installer boots in a few seconds:
ftp://ftp.splack.org/pub/splack/splack-8.0-iso/splack-24oct2k3.iso

If you want to do a full installation, I'd suggest installing using Debian 3.1 and then upgrade to 4.0.

#2 lore raw

Am 06.10.2008 um 20:32 schrieb Blue Swirl:

Quoted 2 lines

If you want to do a full installation, I'd suggest installing using Debian 3.1 and then upgrade to 4.0.

Installing 4.0r0 used to work fine, is there a regression?

Andreas

#3 lore raw
Quoted 9 lines · Andreas Färber

Am 06.10.2008 um 20:32 schrieb Blue Swirl:

Quoted 3 lines

If you want to do a full installation, I'd suggest installing using Debian 3.1 and then upgrade to 4.0.

Installing 4.0r0 used to work fine, is there a regression?

I get a crash at 7% of install (or package verification or whatever), it hasn't ever worked.

#4 lore raw
In reply to Anthony Liguori
Quoted 6 lines · Anthony Liguori

Hi Blue Swirl,

Can you test the attached patch with sparc-softmmu? Better yet, do you
have an image, or a recipe for an image that I could use to try and reproduce the problem? The sparc test code on the QEMU website just uses a ram disk so the regression isn't evident with it.

Your patch fixes the slowness, great!

Please put the prototype of enable_sigio_timer to a .h file somewhere, I get a warning with -Wmissing-prototypes.

author
#5 lore raw
Quoted 4 lines · Blue Swirl
Quoted 2 lines · Anthony Liguori

Hi Blue Swirl,

Quoted 8 lines
Quoted 5 lines

Can you test the attached patch with sparc-softmmu? Better yet, do you
have an image, or a recipe for an image that I could use to try and reproduce the problem? The sparc test code on the QEMU website just uses a ram disk so the regression isn't evident with it.

Your patch fixes the slowness, great!

Okay, I'll code up a proper fix then. That patch was just to confirm the problem.

Thanks,

Anthony Liguori

Quoted 3 lines

Please put the prototype of enable_sigio_timer to a .h file somewhere, I get a warning with -Wmissing-prototypes.

#6 lore raw
In reply to Anthony Liguori
Quoted 6 lines · Anthony Liguori

Hi Blue Swirl,

Can you test the attached patch with sparc-softmmu? Better yet, do you have an image, or a recipe for an image that I could use to try and reproduce the problem? The sparc test code on the QEMU website just uses a ram disk so the regression isn't evident with it.

Your patch works perfectly with -clock unix, but the problem is still present here with -clock dynticks.

If you want to test the problem, you can download an image file from
http://people.debian.org/~aurel32/qemu

BTW I wonder if we should also fire SIGIO on the network device.

Aurelien

Signature · 4 lines
  .''`.  Aurelien Jarno	            | GPG: 1024D/F1BCDB73
 : :' :  Debian developer           | Electrical Engineer
 `. `'   [email protected]         | [email protected]
   `-    people.debian.org/~aurel32 | www.aurel32.net