[Spice-devel] spice ssl live migration
Hi all! I am using spice and ovirt(a old version). when I use remote-viewer with correct ssl certifacate to connect the vm , and then live migrate the vm to another host , remote-viewer is dead .I must reconnect the vm . Can I use remote-viewer continuous when the vm is migrating? How to configure the ssl certificate ? Thank you . ___ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
Re: [Spice-devel] win 7 terminal recognize usb as libusbx device when exiting usb redirect
On 10/22/2014 04:40 AM, 陈晴 wrote: hi, all: environment: terminal: win 7 guest: win 7 usb redirect: usb clerk (provided by Cody Chan) usb device: kinston data traveller 8G teclast 16G (use correctly) usb harddisk (use correctly) following is my steps to use kinston usb storage: 1.connect usb to terminal, recognize as: mass storage device 2.use virt-viewer to connect to guest, usb device is successfully redirected 3.disconnect from guest 4.back to terminal, usb device recognize as universal serial bus device whose driver is libusbx, and oemxx.inf generated in windows/system32/inf (something happened in this step make win 7 to choose libusbx as its driver) Hi, Seems like a usbclerk problem == not removing the WinUSB driver upon client distconnect - What are the versions of usbclerk, virt-viewer ? - Anything interesting in usbclerk's log (try at c:\windows\temp\usbclerk.log) ? - Older versions of usbclerk did not remove the driver upon client disconnect, can you try un-redirect the usb device (from the remote-viewer menu) before disconnecting the client ? Uri ___ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
[Spice-devel] Fwd: a question about Xspice
Hi, Forwarding this to you guys from the xorg-devel list. Regards, Hans Forwarded Message Subject: a question about Xspice Date: Tue, 21 Oct 2014 17:30:54 +0800 (CST) From: cynthia cynthia_...@163.com To: xorg-de...@lists.x.org, x...@lists.x.org HI All, I am trying to build Xspice which is included in guest QXL driver, but one question confuses me a lot. As the Xspice's description in Readme, i get that Xspice is a X server and spice server in one, running in guest os, but our team has developed a set of windows virtual desktop in which the spiceserver is deployed in HOST os linked with QEMU, and i think that is coordinated with the spice protocol., So, could anyone explain it for me if spiceserver is deployed in guest os or host os? I will be very very appreciated~ Best Regards Cynthia ___ xorg-de...@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel ___ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
Re: [Spice-devel] Fwd: a question about Xspice
On 10/22/2014 11:37 AM, Hans de Goede wrote: Hi, Forwarding this to you guys from the xorg-devel list. Regards, Hans Forwarded Message Subject: a question about Xspice Date: Tue, 21 Oct 2014 17:30:54 +0800 (CST) From: cynthia cynthia_...@163.com To: xorg-de...@lists.x.org, x...@lists.x.org HI All, I am trying to build Xspice which is included in guest QXL driver, but one question confuses me a lot. As the Xspice's description in Readme, i get that Xspice is a X server and spice server in one, running in guest os, but our team has developed a set of windows virtual desktop in which the spiceserver is deployed in HOST os linked with QEMU, and i think that is coordinated with the spice protocol., So, could anyone explain it for me if spiceserver is deployed in guest os or host os? I will be very very appreciated~ For your virtual machines spice server is running in the host, i.e. in each qemu process. For Xspice the spice server is Xspice - it's a different configuration, there is no virtual machine, no host/guest distinction. Wherever Xspice is running, there the server is. Best Regards Cynthia ___ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel ___ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
Re: [Spice-devel] Fwd: a question about Xspice
On 10/22/2014 11:37 AM, Hans de Goede wrote: Hi, Forwarding this to you guys from the xorg-devel list. Regards, Hans Forwarded Message Subject: a question about Xspice Date: Tue, 21 Oct 2014 17:30:54 +0800 (CST) From: cynthia cynthia_...@163.com To: xorg-de...@lists.x.org, x...@lists.x.org HI All, I am trying to build Xspice which is included in guest QXL driver, but one question confuses me a lot. As the Xspice's description in Readme, i get that Xspice is a X server and spice server in one, running in guest os, but our team has developed a set of windows virtual desktop in which the spiceserver is deployed in HOST os linked with QEMU, and i think that is coordinated with the spice protocol., So, could anyone explain it for me if spiceserver is deployed in guest os or host os? I will be very very appreciated~ Hi Cynthia, Xspice was developed with the (Linux) server/host in mind, not the guest. Running on a server, it enable remote access to a physical machine using Spice. Of course, it can instead run on the guest, but that was not its purpose. We already had a solution for the guest that uses Spice, as you've mentioned. Hope that helps, Uri. ___ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
Re: [Spice-devel] spice ssl live migration
Hi, On St, 2014-10-22 at 12:03 +0800, yao xu wrote: Hi all! I am using spice and ovirt(a old version). when I use remote-viewer with correct ssl certifacate to connect the vm , and then live migrate the vm to another host , The certificates for both hosts have to be issued by the same CA. remote-viewer is dead .I must reconnect the vm . Can I use remote-viewer continuous when the vm is migrating? How to configure the ssl certificate ? Yes, you can. Configuration may be different based on the way you launch the VMs: e.g. when you use oVirt/RHEV, it will take care of correct TLS setup automatically. Libvirt can do most of the tasks for you but you still have to set up certificates correctly yourself and when you use plain qemu, you have to feed correct configuration even to qemu monitor. So: how do you manage your VMs? David Thank you . ___ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel ___ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
Re: [Spice-devel] [PATCH xf86-video-qxl v2] Revise the XSpice audio processing to avoid the use of pthreads.
Hey,
There are a bunch of calloc/malloc throughout the patch with no check
for a NULL return, would be nice to add that.
This also adds inotify monitoring, do we need to add a check for it to
configure.ac? (eg a check for sys/inotify.h)
One small note below for a typo, looks good to me apart from this.
Christophe
On Tue, Oct 21, 2014 at 04:00:08PM -0500, Jeremy White wrote:
The initial implementation used a separate thread to drive the audio
playback channel. But if you have adaptive streaming turned on,
you will eventually get a update_client_playback_latency message on the
display channel (which in the Xspice case is being driven by the main,
Xorg, thread).
After enough time you would get a thread collision and bad things
would result. I saw this manifest as infinite spin loops in snd_send_data.
This patch eliminates the use of threading altogether, making everything
run in the main Xorg thread using watches and timers, eliminating the
possibility of thread collision.
Signed-off-by: Jeremy White jwh...@codeweavers.com
---
src/qxl.h|3 +-
src/spiceqxl_audio.c | 564
+++---
2 files changed, 352 insertions(+), 215 deletions(-)
diff --git a/src/qxl.h b/src/qxl.h
index fa9b13f..603faca 100644
--- a/src/qxl.h
+++ b/src/qxl.h
@@ -334,8 +334,6 @@ struct _qxl_screen_t
/* XSpice specific, dragged from the Device */
QXLReleaseInfo *last_release;
-pthread_t audio_thread;
-
uint32_t cmdflags;
uint32_t oom_running;
uint32_t num_free_res; /* is having a release ring effective
@@ -353,6 +351,7 @@ struct _qxl_screen_t
} guest_primary;
char playback_fifo_dir[PATH_MAX];
+void *playback_opaque;
#endif /* XSPICE */
uint32_t deferred_fps;
diff --git a/src/spiceqxl_audio.c b/src/spiceqxl_audio.c
index 02859ee..9012f27 100644
--- a/src/spiceqxl_audio.c
+++ b/src/spiceqxl_audio.c
@@ -20,6 +20,10 @@
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
+/* XSpice based audio feed; reads from files (presumably fifos) in a
configured directory,
+ and mixes their raw data on to the spice playback channel. */
+
+
#ifdef HAVE_CONFIG_H
#include config.h
#endif
@@ -32,283 +36,375 @@
#include sys/time.h
#include unistd.h
#include dirent.h
+#include sys/inotify.h
+
+/* mplayer + pulse will write data to the fifo as fast as we can read it.
+ So we need to pace both how quickly we consume the data and how
quickly
+ we feed the data in to Spice. We will read ahead (up to
READ_BUFFER_PERIODS),
+ and feed ahead into the Spice server (up to FEED_BUFFER_PERIODS).
+*/
+
+#define PERIOD_MS10
+#define READ_BUFFER_PERIODS 2
+#define FEED_BUFFER_PERIODS 8
-#define BUFFER_PERIODS 10
-#define PERIOD_MS 10
#define MAX_FIFOS 16
+struct fifo_data {
+char *buffer;
+int size;
+int len;
+int add_to;
+int fd;
+SpiceWatch *watch;
+};
+
struct audio_data {
-int fifo_fds[MAX_FIFOS];
-ino_t inodes[MAX_FIFOS];
-uint32_t valid_bytes, write_offs;
-char *buffer, *spice_buffer;
-int period_frames;
-uint32_t spice_write_offs, spice_buffer_bytes;
-uint32_t frame_bytes, period_bytes, fed, buffer_bytes;
-struct timeval last_read_time;
+struct fifo_data fifos[MAX_FIFOS];
+uint32_t *spice_buffer;
+int spice_buffer_bytes;
+int period_bytes;
+struct timeval fed_through_time;
+int remainder;
+int fifo_count;
+SpiceTimer *wall_timer;
+int wall_timer_live;
+int dir_watch;
+int fifo_dir_watch;
+SpiceWatch *fifo_dir_qxl_watch;
};
-static ssize_t
-read_from_fifos (struct audio_data *data)
+/* We maintain a ring buffer for each file we are reading from;
+ these helper functions faciliate adding data to the buffer,
Typo here, 'facilitate'
+ and removing it. */
+static inline void fifo_data_added(struct fifo_data *f, int n)
{
-size_t to_read_bytes = min(data-period_bytes, data-buffer_bytes -
data-write_offs);
-int i;
-ssize_t max_readed = 0;
-int16_t *out_buf = (int16_t*)(data-buffer + data-write_offs), *buf;
+f-add_to = (f-add_to + n) % f-size;
+f-len += n;
+}
-buf = malloc(to_read_bytes);
-if (!buf)
-{
-ErrorF(playback: malloc failed: %s\n, strerror(errno));
-return 0;
+static inline int fifo_read(struct fifo_data *f)
+{
+int rc;
+int len = min(f-size - f-len, f-size - f-add_to);
+rc = read(f-fd, f-buffer + f-add_to, len);
+if (rc 0)
+fifo_data_added(f, rc);
+
+if (rc 0 rc == len f-size - f-len 0) {
+rc = read(f-fd, f-buffer + f-add_to, f-size - f-len);
+if (rc 0)
+fifo_data_added(f, rc);
}
-memset(out_buf, 0, to_read_bytes);
-
-for (i = 0; i
[Spice-devel] [PATCH] drmmode: Fix compilation warning
One recent change introduced a gcc compilation warning:
CC qxl_drmmode.lo
qxl_drmmode.c: In function 'drmmode_set_mode_major':
qxl_drmmode.c:251:9: error: ISO C90 forbids mixed declarations and code
[-Werror=declaration-after-statement]
CursorPtr cursor = xf86_config-cursor;
This commit moves the CursorPtr declaration to the top of the
drmmode_set_mode_major function in order to fix that warning.
---
src/qxl_drmmode.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/src/qxl_drmmode.c b/src/qxl_drmmode.c
index e9d6436..f8d5369 100644
--- a/src/qxl_drmmode.c
+++ b/src/qxl_drmmode.c
@@ -132,6 +132,7 @@ drmmode_set_mode_major(xf86CrtcPtr crtc, DisplayModePtr
mode,
Rotation rotation, int x, int y)
{
ScrnInfoPtr pScrn = crtc-scrn;
+ CursorPtr cursor;
xf86CrtcConfigPtr xf86_config = XF86_CRTC_CONFIG_PTR(crtc-scrn);
drmmode_crtc_private_ptr drmmode_crtc = crtc-driver_private;
drmmode_ptr drmmode = drmmode_crtc-drmmode;
@@ -248,7 +249,7 @@ done:
crtc-active = TRUE;
#endif
-CursorPtr cursor = xf86_config-cursor;
+cursor = xf86_config-cursor;
if (cursor)
drmmode_show_cursor(crtc);
--
2.1.0
___
Spice-devel mailing list
Spice-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/spice-devel
Re: [Spice-devel] [PATCH] drmmode: Fix compilation warning
ack,
- Original Message -
One recent change introduced a gcc compilation warning:
CC qxl_drmmode.lo
qxl_drmmode.c: In function 'drmmode_set_mode_major':
qxl_drmmode.c:251:9: error: ISO C90 forbids mixed declarations and code
[-Werror=declaration-after-statement]
CursorPtr cursor = xf86_config-cursor;
This commit moves the CursorPtr declaration to the top of the
drmmode_set_mode_major function in order to fix that warning.
---
src/qxl_drmmode.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/src/qxl_drmmode.c b/src/qxl_drmmode.c
index e9d6436..f8d5369 100644
--- a/src/qxl_drmmode.c
+++ b/src/qxl_drmmode.c
@@ -132,6 +132,7 @@ drmmode_set_mode_major(xf86CrtcPtr crtc, DisplayModePtr
mode,
Rotation rotation, int x, int y)
{
ScrnInfoPtr pScrn = crtc-scrn;
+ CursorPtr cursor;
xf86CrtcConfigPtr xf86_config = XF86_CRTC_CONFIG_PTR(crtc-scrn);
drmmode_crtc_private_ptr drmmode_crtc = crtc-driver_private;
drmmode_ptr drmmode = drmmode_crtc-drmmode;
@@ -248,7 +249,7 @@ done:
crtc-active = TRUE;
#endif
-CursorPtr cursor = xf86_config-cursor;
+cursor = xf86_config-cursor;
if (cursor)
drmmode_show_cursor(crtc);
--
2.1.0
___
Spice-devel mailing list
Spice-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/spice-devel
___
Spice-devel mailing list
Spice-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/spice-devel
Re: [Spice-devel] [PATCH] drmmode: Fix compilation warning
On Wed, Oct 22, 2014 at 10:45:47AM -0400, Marc-André Lureau wrote: ack, Fwiw, I don't think I have write access to qxl git repository (or maybe I'm having the same issue as the one you are having) Christophe pgpU5B67bD9OF.pgp Description: PGP signature ___ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
[Spice-devel] [spice-gtk] Fix SndCodec leak in playback_handle_start()
An audio SpiceChannel can get several times the 'start' message during
its lifetime. If received multiple times (for example when rebooting a
VM), this would lead to a SndCodec leak as reported by valgrind:
==7749== 27,036 (48 direct, 26,988 indirect) bytes in 1 blocks are definitely
lost in loss record 11,418 of 11,440
==7749==at 0x4A08946: calloc (in
/usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
==7749==by 0x5772A9A: spice_malloc0 (mem.c:109)
==7749==by 0x5772D42: spice_malloc0_n (mem.c:173)
==7749==by 0x57FB3E2: snd_codec_create (snd_codec.c:267)
==7749==by 0x573774C: playback_handle_start (channel-playback.c:380)
==7749==by 0x5721E44: spice_channel_handle_msg (spice-channel.c:2859)
==7749==by 0x571ECB3: spice_channel_recv_msg (spice-channel.c:1877)
==7749==by 0x571F402: spice_channel_iterate_read (spice-channel.c:2114)
==7749==by 0x571F614: spice_channel_iterate (spice-channel.c:2152)
==7749==by 0x5720411: spice_channel_coroutine (spice-channel.c:2429)
==7749==by 0x575CD3A: coroutine_trampoline (coroutine_ucontext.c:63)
==7749==by 0x575C9F4: continuation_trampoline (continuation.c:55)
==7749==by 0x3898E47FEF: ??? (in /usr/lib64/libc-2.20.so)
==7749==by 0xAED547666C92C2FF: ???
==7749==by 0xFFEFFF92F: ???
---
gtk/channel-playback.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/gtk/channel-playback.c b/gtk/channel-playback.c
index e5b9658..40e01ba 100644
--- a/gtk/channel-playback.c
+++ b/gtk/channel-playback.c
@@ -375,6 +375,7 @@ static void playback_handle_start(SpiceChannel *channel,
SpiceMsgIn *in)
c-codec = NULL;
if (c-mode != SPICE_AUDIO_DATA_MODE_RAW) {
+snd_codec_destroy(c-codec);
if (snd_codec_create(c-codec, c-mode, start-frequency,
SND_CODEC_DECODE) != SND_CODEC_OK) {
g_warning(create decoder failed);
return;
--
2.1.0
___
Spice-devel mailing list
Spice-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/spice-devel
Re: [Spice-devel] [spice-gtk] Fix SndCodec leak in playback_handle_start()
ack
- Original Message -
An audio SpiceChannel can get several times the 'start' message during
its lifetime. If received multiple times (for example when rebooting a
VM), this would lead to a SndCodec leak as reported by valgrind:
==7749== 27,036 (48 direct, 26,988 indirect) bytes in 1 blocks are definitely
lost in loss record 11,418 of 11,440
==7749==at 0x4A08946: calloc (in
/usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
==7749==by 0x5772A9A: spice_malloc0 (mem.c:109)
==7749==by 0x5772D42: spice_malloc0_n (mem.c:173)
==7749==by 0x57FB3E2: snd_codec_create (snd_codec.c:267)
==7749==by 0x573774C: playback_handle_start (channel-playback.c:380)
==7749==by 0x5721E44: spice_channel_handle_msg (spice-channel.c:2859)
==7749==by 0x571ECB3: spice_channel_recv_msg (spice-channel.c:1877)
==7749==by 0x571F402: spice_channel_iterate_read (spice-channel.c:2114)
==7749==by 0x571F614: spice_channel_iterate (spice-channel.c:2152)
==7749==by 0x5720411: spice_channel_coroutine (spice-channel.c:2429)
==7749==by 0x575CD3A: coroutine_trampoline (coroutine_ucontext.c:63)
==7749==by 0x575C9F4: continuation_trampoline (continuation.c:55)
==7749==by 0x3898E47FEF: ??? (in /usr/lib64/libc-2.20.so)
==7749==by 0xAED547666C92C2FF: ???
==7749==by 0xFFEFFF92F: ???
---
gtk/channel-playback.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/gtk/channel-playback.c b/gtk/channel-playback.c
index e5b9658..40e01ba 100644
--- a/gtk/channel-playback.c
+++ b/gtk/channel-playback.c
@@ -375,6 +375,7 @@ static void playback_handle_start(SpiceChannel *channel,
SpiceMsgIn *in)
c-codec = NULL;
if (c-mode != SPICE_AUDIO_DATA_MODE_RAW) {
+snd_codec_destroy(c-codec);
if (snd_codec_create(c-codec, c-mode, start-frequency,
SND_CODEC_DECODE) != SND_CODEC_OK) {
g_warning(create decoder failed);
return;
--
2.1.0
___
Spice-devel mailing list
Spice-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/spice-devel
___
Spice-devel mailing list
Spice-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/spice-devel
Re: [Spice-devel] [PATCH xf86-video-qxl v2] Revise the XSpice audio processing to avoid the use of pthreads.
On 10/22/2014 09:38 AM, Christophe Fergeau wrote: Hey, There are a bunch of calloc/malloc throughout the patch with no check for a NULL return, would be nice to add that. Hmm. I've seen this argued both ways. One side argues that if an allocation is never expected to fail, that it's best to just trigger the resulting crash as a diagnostic. Trying to 'handle' an error where there is no reasonable response can just mask the problem, or so goes the rationale. And these are all small allocs that really should never fail. The code base is unclear; it seems that allocs are sometimes checked, and sometimes not. It's easy to add; if you feel it's the preferred Spice standard, I'll add the checks. This also adds inotify monitoring, do we need to add a check for it to configure.ac? (eg a check for sys/inotify.h) I don't see how it could hurt; I'll add a check tot he XSpice section. One small note below for a typo, looks good to me apart from this. Great, thanks. Cheers, Jeremy ___ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
Re: [Spice-devel] Fwd: a question about Xspice
At 2014-10-22 19:26:20, Uri Lublin u...@redhat.com wrote: On 10/22/2014 11:37 AM, Hans de Goede wrote: Hi, Forwarding this to you guys from the xorg-devel list. Regards, Hans Forwarded Message Subject: a question about Xspice Date: Tue, 21 Oct 2014 17:30:54 +0800 (CST) From: cynthia cynthia_...@163.com To: xorg-de...@lists.x.org, x...@lists.x.org HI All, I am trying to build Xspice which is included in guest QXL driver, but one question confuses me a lot. As the Xspice's description in Readme, i get that Xspice is a X server and spice server in one, running in guest os, but our team has developed a set of windows virtual desktop in which the spiceserver is deployed in HOST os linked with QEMU, and i think that is coordinated with the spice protocol., So, could anyone explain it for me if spiceserver is deployed in guest os or host os? I will be very very appreciated~ Hi Cynthia, Xspice was developed with the (Linux) server/host in mind, not the guest. Running on a server, it enable remote access to a physical machine using Spice. Of course, it can instead run on the guest, but that was not its purpose. We already had a solution for the guest that uses Spice, as you've mentioned. Dear Uri, Thank you so so much for your mail , it's very important. I started to learn Xspice because Our team are trying to find a solution to Linux virtual desktop. Today I created a Fedora guest OS and found there are already pre-installed vdagent and qxl driver in it. My spice client can connect to the server and logged into the guest, the most important part is the qxl dirver in guest is working, I think the Linux virtual desktop works, but i didn't do any Xspice thing, i use the host used to hosting Windows guest. In my opinion spice client will be able to access to physical machine when an spice server package is installed, but what do we use Xspice for? and what will setting environment variables for X server in the host and enabling video, mouse and keyboard driver(libspiceqxl_drv.so) help for? It's a little long mail, many questions in my head. Anyway ,thank you Uri again:) B R Cynthia___ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
Re: [Spice-devel] [PATCH xf86-video-qxl v2] Revise the XSpice audio processing to avoid the use of pthreads.
Forgot to mention that there are more error situations in this function where it looks like data would be leaked. Sure, added a few more checks. v3 with configure checking, the typo fix, and those checks inbound shortly. Cheers, Jeremy ___ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
[Spice-devel] [PATCH xf86-video-qxl v3] Revise the XSpice audio processing to avoid the use of pthreads.
The initial implementation used a separate thread to drive the audio
playback channel. But if you have adaptive streaming turned on,
you will eventually get a update_client_playback_latency message on the
display channel (which in the Xspice case is being driven by the main,
Xorg, thread).
After enough time you would get a thread collision and bad things
would result. I saw this manifest as infinite spin loops in snd_send_data.
This patch eliminates the use of threading altogether, making everything
run in the main Xorg thread using watches and timers, eliminating the
possibility of thread collision.
Signed-off-by: Jeremy White jwh...@codeweavers.com
---
configure.ac |2 +
src/qxl.h|3 +-
src/spiceqxl_audio.c | 581 +++---
3 files changed, 368 insertions(+), 218 deletions(-)
diff --git a/configure.ac b/configure.ac
index 3ca32b6..119340a 100644
--- a/configure.ac
+++ b/configure.ac
@@ -126,6 +126,8 @@ AC_ARG_ENABLE(xspice,
])
if test x$enable_xspice = xyes; then
+AC_CHECK_HEADERS(sys/inotify.h)
+AC_CHECK_FUNCS(inotify_init1)
PKG_CHECK_MODULES([SPICE], [spice-server = 0.6.3],
[
AC_SUBST(SPICE_CFLAGS)
diff --git a/src/qxl.h b/src/qxl.h
index fa9b13f..603faca 100644
--- a/src/qxl.h
+++ b/src/qxl.h
@@ -334,8 +334,6 @@ struct _qxl_screen_t
/* XSpice specific, dragged from the Device */
QXLReleaseInfo *last_release;
-pthread_t audio_thread;
-
uint32_t cmdflags;
uint32_t oom_running;
uint32_t num_free_res; /* is having a release ring effective
@@ -353,6 +351,7 @@ struct _qxl_screen_t
} guest_primary;
char playback_fifo_dir[PATH_MAX];
+void *playback_opaque;
#endif /* XSPICE */
uint32_t deferred_fps;
diff --git a/src/spiceqxl_audio.c b/src/spiceqxl_audio.c
index 02859ee..086b943 100644
--- a/src/spiceqxl_audio.c
+++ b/src/spiceqxl_audio.c
@@ -20,6 +20,10 @@
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
+/* XSpice based audio feed; reads from files (presumably fifos) in a
configured directory,
+ and mixes their raw data on to the spice playback channel. */
+
+
#ifdef HAVE_CONFIG_H
#include config.h
#endif
@@ -32,283 +36,379 @@
#include sys/time.h
#include unistd.h
#include dirent.h
+#if defined(HAVE_SYS_INOTIFY_H)
+#include sys/inotify.h
+#endif
+
+/* mplayer + pulse will write data to the fifo as fast as we can read it.
+ So we need to pace both how quickly we consume the data and how quickly
+ we feed the data in to Spice. We will read ahead (up to
READ_BUFFER_PERIODS),
+ and feed ahead into the Spice server (up to FEED_BUFFER_PERIODS).
+*/
+
+#define PERIOD_MS10
+#define READ_BUFFER_PERIODS 2
+#define FEED_BUFFER_PERIODS 8
-#define BUFFER_PERIODS 10
-#define PERIOD_MS 10
#define MAX_FIFOS 16
+struct fifo_data {
+char *buffer;
+int size;
+int len;
+int add_to;
+int fd;
+SpiceWatch *watch;
+};
+
struct audio_data {
-int fifo_fds[MAX_FIFOS];
-ino_t inodes[MAX_FIFOS];
-uint32_t valid_bytes, write_offs;
-char *buffer, *spice_buffer;
-int period_frames;
-uint32_t spice_write_offs, spice_buffer_bytes;
-uint32_t frame_bytes, period_bytes, fed, buffer_bytes;
-struct timeval last_read_time;
+struct fifo_data fifos[MAX_FIFOS];
+uint32_t *spice_buffer;
+int spice_buffer_bytes;
+int period_bytes;
+struct timeval fed_through_time;
+int remainder;
+int fifo_count;
+SpiceTimer *wall_timer;
+int wall_timer_live;
+int dir_watch;
+int fifo_dir_watch;
+SpiceWatch *fifo_dir_qxl_watch;
};
-static ssize_t
-read_from_fifos (struct audio_data *data)
+/* We maintain a ring buffer for each file we are reading from;
+ these helper functions facilitate adding data to the buffer,
+ and removing it. */
+static inline void fifo_data_added(struct fifo_data *f, int n)
{
-size_t to_read_bytes = min(data-period_bytes, data-buffer_bytes -
data-write_offs);
-int i;
-ssize_t max_readed = 0;
-int16_t *out_buf = (int16_t*)(data-buffer + data-write_offs), *buf;
+f-add_to = (f-add_to + n) % f-size;
+f-len += n;
+}
-buf = malloc(to_read_bytes);
-if (!buf)
-{
-ErrorF(playback: malloc failed: %s\n, strerror(errno));
-return 0;
+static inline int fifo_read(struct fifo_data *f)
+{
+int rc;
+int len = min(f-size - f-len, f-size - f-add_to);
+rc = read(f-fd, f-buffer + f-add_to, len);
+if (rc 0)
+fifo_data_added(f, rc);
+
+if (rc 0 rc == len f-size - f-len 0) {
+rc = read(f-fd, f-buffer + f-add_to, f-size - f-len);
+if (rc 0)
+fifo_data_added(f, rc);
}
-memset(out_buf, 0, to_read_bytes);
-
-for (i = 0; i MAX_FIFOS; ++i)
-{
-unsigned int s;
-ssize_t readed;
-
-if (data-fifo_fds[i] 0)
-
