diff mbox

[v9] xen: add para-virtual sound interface header files

Message ID 1479919089-20174-2-git-send-email-andr2000@gmail.com (mailing list archive)
State New, archived
Headers show

Commit Message

Oleksandr Andrushchenko Nov. 23, 2016, 4:38 p.m. UTC
This is the ABI for the two halves of a para-virtualized
sound driver to communicate with each to other.

Signed-off-by: Oleksandr Andrushchenko <andr2000@gmail.com>
Signed-off-by: Oleksandr Grytsov <al1img@gmail.com>
Signed-off-by: Oleksandr Dmytryshyn <oleksandr.dmytryshyn@globallogic.com>
Signed-off-by: Iurii Konovalenko <iurii.konovalenko@globallogic.com>

---
Changes since v1:
 * removed __attribute__((__packed__)) from all structures definitions

Changes since v2:
 * removed all C structures
 * added protocol description between frontend and backend drivers

Changes since v3:
 * fixed some typos
 * renamed XENSND_PCM_FORMAT_FLOAT_** to XENSND_PCM_FORMAT_F32_**
 * renamed XENSND_PCM_FORMAT_FLOAT64_** to XENSND_PCM_FORMAT_F64_**
 * added 'id' field to the request and response packets
 * renamed 'stream_id' to 'stream' in the packets description
 * renamed 'pcm_data_rate' to 'pcm_rate' in the packets description
 * renamed 'pcm_stream_type' to 'pcm_type' in the packets description
 * removed 'stream_id' field from the response packets

Changes since v4:
 * renamed 'stream_id' back to the to 'stream' in the packets description
 * moved 'id' field to the upper position in the response packets

Changes since v5:
 * Slightly reworked request/response packets
 * Size of the request/response packet is changed to the 64 bytes
 * Now parameters for the XENSND_OP_SET_VOLUME/XENSND_OP_GET_VOLUME are
   passed via shared page
 * Added parameters for the XenBus nodes (now each stream can be mapped
   to the defined sound device in the backend using those parameters)
 * Added XenBus state diagrams description

Changes since v6:
 * Reworked streams description  in the Backend XenBus Nodes

Changes since v7:
 * re-worked backend device parameters to be more generic and flexible
 * extended frontend device parameters
 * slightly updated state machine description added mute/unmute commands
 * added constants for XenStore configuration strings
   (fields, PCM formats etc.)
 * changed request/response structure size from 64 octets to 16
 * introduced dynamic buffer allocation instead of
   static XENSND_MAX_PAGES_PER_REQUEST
 * re-worked open request to allow dynamic buffer allocation
 * re-worked read/write/volume requests, so they don't pass grefs:
   buffer from the open request is used for these operations to pass data
 * specified type of the volume value to be a signed value in steps
   of 0.001 dBm, while 0 being 0dBm.
 * added Linux include file with structure definitions

Changes since v8:
 * changed frontend-id to frontend_id
 * single sound card support, configured with bunch of
   devices/streams
 * clarifucation made on sample rates and formats expressed as
   decimals w/o any particular ordering
 * put description of migration/disconnection state
 * replaced __attribute__((packed)) to __packed
 * changed padding of ring structures to 64 to fit cache line
 * removeed #ifdef __KERNEL
 * explicitly stated which indices in XenStore configuration
   are contiguous
 * added description to what frontend's defaults are
 * made names of virtual card/devices optional
 * removed PCM_FORMAT_SPECIAL
 * changed volume units from dBm to dB
---
 include/xen/interface/io/sndif.h       | 609 +++++++++++++++++++++++++++++++++
 include/xen/interface/io/sndif_linux.h | 112 ++++++
 2 files changed, 721 insertions(+)
 create mode 100644 include/xen/interface/io/sndif.h
 create mode 100644 include/xen/interface/io/sndif_linux.h

Comments

Jan Beulich Nov. 24, 2016, 10:58 a.m. UTC | #1
>>> On 23.11.16 at 17:38, <andr2000@gmail.com> wrote:
>  include/xen/interface/io/sndif.h       | 609 +++++++++++++++++++++++++++++++++
>  include/xen/interface/io/sndif_linux.h | 112 ++++++
>  2 files changed, 721 insertions(+)
>  create mode 100644 include/xen/interface/io/sndif.h
>  create mode 100644 include/xen/interface/io/sndif_linux.h

Mind explaining what this second file is doing here? I don't think
we want any Linux specific in our public headers. Was this perhaps
meant to instead be include/xen/sendif.h in the Linux tree?

Jan
Oleksandr Andrushchenko Nov. 24, 2016, 11:49 a.m. UTC | #2
1.  As per previous discussions on this there was a request that protocol files
must be platform agnostic meaning that no specific compiler/platform code
allowed. For that reason we have created 2 separate files: protocol itself
+ companion Linux specific file with structure definitions, packaging etc.

2. My intention was that these files go into Linux kernel tree and the
corresponding
Xen headers are all under include/xen/interface/io folder.

Best regards,
Oleksandr Andrushchenko
Julien Grall Nov. 24, 2016, 12:13 p.m. UTC | #3
Hi Oleksandr,

On 23/11/16 16:38, Oleksandr Andrushchenko wrote:
> diff --git a/include/xen/interface/io/sndif_linux.h b/include/xen/interface/io/sndif_linux.h
> new file mode 100644
> index 0000000..2802504
> --- /dev/null
> +++ b/include/xen/interface/io/sndif_linux.h
> @@ -0,0 +1,112 @@
> +/******************************************************************************
> + * sndif_linux.h
> + *
> + * Unified sound-device I/O interface for Xen guest OSes.
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a copy
> + * of this software and associated documentation files (the "Software"), to
> + * deal in the Software without restriction, including without limitation the
> + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
> + * sell copies of the Software, and to permit persons to whom the Software is
> + * furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice shall be included in
> + * all copies or substantial portions of the Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
> + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
> + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
> + * DEALINGS IN THE SOFTWARE.
> + *
> + * Copyright (C) 2016 EPAM Systems Inc.
> + *
> + * Authors: Oleksandr Andrushchenko <andr2000@gmail.com>
> + *          Oleksandr Grytsov <al1img@gmail.com>
> + */
> +
> +#ifndef __XEN_PUBLIC_IO_XENSND_LINUX_H__
> +#define __XEN_PUBLIC_IO_XENSND_LINUX_H__
> +
> +#include <xen/interface/io/ring.h>
> +#include <xen/interface/io/sndif.h>
> +#include <xen/interface/grant_table.h>
> +
> +struct xensnd_open_req {
> +	uint32_t pcm_rate;
> +	uint8_t pcm_format;
> +	uint8_t pcm_channels;
> +	/* in Hz */

It is not clear to me to which field this comment applies. Is it 
pcm_channels?

Regards,
Julien Grall Nov. 24, 2016, 12:15 p.m. UTC | #4
Hi Oleksandr,

On 24/11/16 11:49, Oleksandr Andrushchenko wrote:
> 1.  As per previous discussions on this there was a request that protocol files
> must be platform agnostic meaning that no specific compiler/platform code
> allowed. For that reason we have created 2 separate files: protocol itself
> + companion Linux specific file with structure definitions, packaging etc.

Any people who wants to use the PV driver for other OS than Linux will 
have to do it. So, may I ask why don't you write a platform agnostic header?

This would also make easier to share the PV driver code between 
different OS.

Regards,
Jan Beulich Nov. 24, 2016, 12:15 p.m. UTC | #5
>>> On 24.11.16 at 12:49, <andr2000@gmail.com> wrote:
> 1.  As per previous discussions on this there was a request that protocol files
> must be platform agnostic meaning that no specific compiler/platform code
> allowed. For that reason we have created 2 separate files: protocol itself
> + companion Linux specific file with structure definitions, packaging etc.
> 
> 2. My intention was that these files go into Linux kernel tree and the
> corresponding
> Xen headers are all under include/xen/interface/io folder.

You still talk about "Xen headers", i.e. plural, so I don't think your
reply addresses my earlier question. IOW I continue to think there
should be only one header in the patch here.

Jan
Oleksandr Andrushchenko Nov. 24, 2016, 12:35 p.m. UTC | #6
> It is not clear to me to which field this comment applies. Is it
> pcm_channels?
It is for pcm_rate, fixed, thanks
Oleksandr Andrushchenko Nov. 24, 2016, 12:52 p.m. UTC | #7
> You still talk about "Xen headers", i.e. plural, so I don't think your
> reply addresses my earlier question. IOW I continue to think there
> should be only one header in the patch here.

Ok, makes sense. I will create 2 patches: one for sndif.h which
is a generic protocol (platform agnostic) and sndif_linux.h which is
Linux specific.
Does this address your concern?
Oleksandr Andrushchenko Nov. 24, 2016, 12:53 p.m. UTC | #8
> Any people who wants to use the PV driver for other OS than Linux will have
> to do it. So, may I ask why don't you write a platform agnostic header?
>
> This would also make easier to share the PV driver code between different
> OS.
There are 2 files: sndif.h which is platform agnostic and sndif_linux.h
which is Linux platform specific (this is exactly the file which every platform
will need to define, e.g. sndif_windows.h).
In my case I need it to at least define Linux specific __packed
attribute which is not supported by Win AFAIK.
Julien Grall Nov. 24, 2016, 1 p.m. UTC | #9
Hello Oleksandr,

On 24/11/16 12:53, Oleksandr Andrushchenko wrote:
>> Any people who wants to use the PV driver for other OS than Linux will have
>> to do it. So, may I ask why don't you write a platform agnostic header?
>>
>> This would also make easier to share the PV driver code between different
>> OS.
> There are 2 files: sndif.h which is platform agnostic and sndif_linux.h

I don't consider sndif.h been agnostic. It is just a bunch of array 
definition. Each OS would have to implement their own headers in order 
to easily access command fields.

> which is Linux platform specific (this is exactly the file which every platform
> will need to define, e.g. sndif_windows.h).
> In my case I need it to at least define Linux specific __packed
> attribute which is not supported by Win AFAIK.

All the other PV protocols are able to define structure that are 
platform agnostic. I would like to understand why you cannot do this here.

Not mentioning that using __packed is usually a bad idea because it can 
mess-up with the natural alignment of the fields and makes the 
programmer's life more miserable (see an example in drm PV drivers).

Regards,
Jan Beulich Nov. 24, 2016, 1:02 p.m. UTC | #10
>>> On 24.11.16 at 13:52, <andr2000@gmail.com> wrote:
>>  You still talk about "Xen headers", i.e. plural, so I don't think your
>> reply addresses my earlier question. IOW I continue to think there
>> should be only one header in the patch here.
> 
> Ok, makes sense. I will create 2 patches: one for sndif.h which
> is a generic protocol (platform agnostic) and sndif_linux.h which is
> Linux specific.
> Does this address your concern?

Well - see my reply to your respective DRM patch. It may well be
that the same issue exists here, in which case you rather want to
fold the two files.

Jan
Jan Beulich Nov. 24, 2016, 1:04 p.m. UTC | #11
>>> On 24.11.16 at 13:53, <andr2000@gmail.com> wrote:
>>  Any people who wants to use the PV driver for other OS than Linux will have
>> to do it. So, may I ask why don't you write a platform agnostic header?
>>
>> This would also make easier to share the PV driver code between different
>> OS.
> There are 2 files: sndif.h which is platform agnostic and sndif_linux.h
> which is Linux platform specific (this is exactly the file which every 
> platform
> will need to define, e.g. sndif_windows.h).

But that is the very problem: There shouldn't be an assumption that
both side of the communication sit on the same OS, i.e. you need
one fully spelled out header, and that header ...

> In my case I need it to at least define Linux specific __packed
> attribute which is not supported by Win AFAIK.

... needs to avoid using platform specific constructs (or making
other assumptions on the platform). I can only re-iterate: Please
follow the model other PV protocols already present (without
repeating their mistakes, if at all possible).

Jan
Artem Mygaiev Nov. 24, 2016, 1:18 p.m. UTC | #12
On 24.11.16 15:04, Jan Beulich wrote:
>> In my case I need it to at least define Linux specific __packed
>> attribute which is not supported by Win AFAIK.
> ... needs to avoid using platform specific constructs (or making
> other assumptions on the platform). I can only re-iterate: Please
> follow the model other PV protocols already present (without
> repeating their mistakes, if at all possible).
But if we don't pack (or enforce some specific alignment) data
structures used between domains they may have different alignment on
different domains, no?
Jan Beulich Nov. 24, 2016, 1:31 p.m. UTC | #13
>>> On 24.11.16 at 14:18, <artem_mygaiev@epam.com> wrote:
> On 24.11.16 15:04, Jan Beulich wrote:
>>> In my case I need it to at least define Linux specific __packed
>>> attribute which is not supported by Win AFAIK.
>> ... needs to avoid using platform specific constructs (or making
>> other assumptions on the platform). I can only re-iterate: Please
>> follow the model other PV protocols already present (without
>> repeating their mistakes, if at all possible).
> But if we don't pack (or enforce some specific alignment) data
> structures used between domains they may have different alignment on
> different domains, no?

Let me repeat what I've said before: Please take a look at existing
interface headers. By properly placing fields and adding explicit
padding where needed, you can avoid such layout differences.

Jan
Oleksandr Andrushchenko Nov. 24, 2016, 2:28 p.m. UTC | #14
> I don't consider sndif.h been agnostic. It is just a bunch of array
> definition.
>
I am not sure I'm following you here.
sndif.h doesn't have any platform specific definitions, but sndif_linux.h does.
Both don't define any methods, but only structures (and __packed as of now).
>Each OS would have to implement their own headers in order to
> easily access command fields.
This is what I did for Linux, the result is sndif_linux.h
> All the other PV protocols are able to define structure that are platform
> agnostic. I would like to understand why you cannot do this here.
>
> Not mentioning that using __packed is usually a bad idea because it can
> mess-up with the natural alignment of the fields and makes the programmer's
> life more miserable (see an example in drm PV drivers).
I need some more time for that
Artem Mygaiev Nov. 24, 2016, 2:58 p.m. UTC | #15
On 24.11.16 15:31, Jan Beulich wrote:
>>>> On 24.11.16 at 14:18, <artem_mygaiev@epam.com> wrote:
>> On 24.11.16 15:04, Jan Beulich wrote:
>>>> In my case I need it to at least define Linux specific __packed
>>>> attribute which is not supported by Win AFAIK.
>>> ... needs to avoid using platform specific constructs (or making
>>> other assumptions on the platform). I can only re-iterate: Please
>>> follow the model other PV protocols already present (without
>>> repeating their mistakes, if at all possible).
>> But if we don't pack (or enforce some specific alignment) data
>> structures used between domains they may have different alignment on
>> different domains, no?
> Let me repeat what I've said before: Please take a look at existing
> interface headers. By properly placing fields and adding explicit
> padding where needed, you can avoid such layout differences.

Don't get me wrong - I have checked the i/f headers before. My concern
is very generic: C standard defines alignment as an
*implementation-defined* integer, so current implementation is based
under two assumptions on the guest platforms: a) guest compiler does not
have "packing" enabled by default and b) alignment follows typical
values (1 byte for char, 2 bytes for short, ...). While this is true in
most cases, with some embedded RTOS this may become an issue.

This is not critical/important ATM, and completely disconnected from the
purpose of the patch, so I agree it has to be rewritten according to
current practices.
Julien Grall Nov. 24, 2016, 3:09 p.m. UTC | #16
Hi Artem,

On 24/11/16 14:58, Artem Mygaiev wrote:
> On 24.11.16 15:31, Jan Beulich wrote:
>>>>> On 24.11.16 at 14:18, <artem_mygaiev@epam.com> wrote:
>>> On 24.11.16 15:04, Jan Beulich wrote:
>>>>> In my case I need it to at least define Linux specific __packed
>>>>> attribute which is not supported by Win AFAIK.
>>>> ... needs to avoid using platform specific constructs (or making
>>>> other assumptions on the platform). I can only re-iterate: Please
>>>> follow the model other PV protocols already present (without
>>>> repeating their mistakes, if at all possible).
>>> But if we don't pack (or enforce some specific alignment) data
>>> structures used between domains they may have different alignment on
>>> different domains, no?
>> Let me repeat what I've said before: Please take a look at existing
>> interface headers. By properly placing fields and adding explicit
>> padding where needed, you can avoid such layout differences.
>
> Don't get me wrong - I have checked the i/f headers before. My concern
> is very generic: C standard defines alignment as an
> *implementation-defined* integer, so current implementation is based
> under two assumptions on the guest platforms: a) guest compiler does not
> have "packing" enabled by default and b) alignment follows typical
> values (1 byte for char, 2 bytes for short, ...). While this is true in
> most cases, with some embedded RTOS this may become an issue.

I am not sure why you mentioned embedded RTOS, this is specific to the 
compiler and not an OS.

ARM provides a specification (see AAPCS [1] for ARM32 and AAPCS64 [2]) 
that explain the alignment of the structure and other stuff. Any 
compiler should implement for interoperability. If not, then you are 
already in trouble to issue hypercall to Xen when using standard headers.

x86  has fairly similar requirements as they are all sensible all not 
all specific to the ARM platform.

So it is possible to write header in the way which will accommodate all 
the architecture.

>
> This is not critical/important ATM, and completely disconnected from the
> purpose of the patch, so I agree it has to be rewritten according to
> current practices.
>

[1] 
http://infocenter.arm.com/help/topic/com.arm.doc.ihi0042f/IHI0042F_aapcs.pdf
[2] 
http://infocenter.arm.com/help/topic/com.arm.doc.ihi0055b/IHI0055B_aapcs64.pdf
Jan Beulich Nov. 24, 2016, 3:11 p.m. UTC | #17
>>> On 24.11.16 at 15:58, <artem_mygaiev@epam.com> wrote:
> On 24.11.16 15:31, Jan Beulich wrote:
>>>>> On 24.11.16 at 14:18, <artem_mygaiev@epam.com> wrote:
>>> On 24.11.16 15:04, Jan Beulich wrote:
>>>>> In my case I need it to at least define Linux specific __packed
>>>>> attribute which is not supported by Win AFAIK.
>>>> ... needs to avoid using platform specific constructs (or making
>>>> other assumptions on the platform). I can only re-iterate: Please
>>>> follow the model other PV protocols already present (without
>>>> repeating their mistakes, if at all possible).
>>> But if we don't pack (or enforce some specific alignment) data
>>> structures used between domains they may have different alignment on
>>> different domains, no?
>> Let me repeat what I've said before: Please take a look at existing
>> interface headers. By properly placing fields and adding explicit
>> padding where needed, you can avoid such layout differences.
> 
> Don't get me wrong - I have checked the i/f headers before. My concern
> is very generic: C standard defines alignment as an
> *implementation-defined* integer, so current implementation is based
> under two assumptions on the guest platforms: a) guest compiler does not
> have "packing" enabled by default and b) alignment follows typical
> values (1 byte for char, 2 bytes for short, ...). While this is true in
> most cases, with some embedded RTOS this may become an issue.

This being true, I think the way existing interfaces have their
header written, there are relatively little extra assumptions: I
think all we require is that types don't have an alignment larger
than their size. In particular point a) can be taken care of by
adding explicit padding fields (as is the case for existing headers).

Jan
Oleksandr Andrushchenko Nov. 24, 2016, 3:29 p.m. UTC | #18
> This is not critical/important ATM, and completely disconnected from the
> purpose of the patch, so I agree it has to be rewritten according to
> current practices.
Agree, will resend new patch v10 soon

Thank you all
Artem Mygaiev Nov. 24, 2016, 3:50 p.m. UTC | #19
On 24.11.16 17:09, Julien Grall wrote:
> Hi Artem,
>
> On 24/11/16 14:58, Artem Mygaiev wrote:
>> On 24.11.16 15:31, Jan Beulich wrote:
>>>>>> On 24.11.16 at 14:18, <artem_mygaiev@epam.com> wrote:
>>>> On 24.11.16 15:04, Jan Beulich wrote:
>>>>>> In my case I need it to at least define Linux specific __packed
>>>>>> attribute which is not supported by Win AFAIK.
>>>>> ... needs to avoid using platform specific constructs (or making
>>>>> other assumptions on the platform). I can only re-iterate: Please
>>>>> follow the model other PV protocols already present (without
>>>>> repeating their mistakes, if at all possible).
>>>> But if we don't pack (or enforce some specific alignment) data
>>>> structures used between domains they may have different alignment on
>>>> different domains, no?
>>> Let me repeat what I've said before: Please take a look at existing
>>> interface headers. By properly placing fields and adding explicit
>>> padding where needed, you can avoid such layout differences.
>>
>> Don't get me wrong - I have checked the i/f headers before. My concern
>> is very generic: C standard defines alignment as an
>> *implementation-defined* integer, so current implementation is based
>> under two assumptions on the guest platforms: a) guest compiler does not
>> have "packing" enabled by default and b) alignment follows typical
>> values (1 byte for char, 2 bytes for short, ...). While this is true in
>> most cases, with some embedded RTOS this may become an issue.
>
> I am not sure why you mentioned embedded RTOS, this is specific to the
> compiler and not an OS.
>
Indeed, my statement is not quite correct. It is just that there are
proprietary RTOSes bundled with IDEs with own or customized tools,
compilers, etc.

> ARM provides a specification (see AAPCS [1] for ARM32 and AAPCS64 [2])
> that explain the alignment of the structure and other stuff. Any
> compiler should implement for interoperability. If not, then you are
> already in trouble to issue hypercall to Xen when using standard headers.
>
> x86  has fairly similar requirements as they are all sensible all not
> all specific to the ARM platform.
>
> So it is possible to write header in the way which will accommodate
> all the architecture.

Understood.
>
>>
>> This is not critical/important ATM, and completely disconnected from the
>> purpose of the patch, so I agree it has to be rewritten according to
>> current practices.
>>
>
> [1]
> http://infocenter.arm.com/help/topic/com.arm.doc.ihi0042f/IHI0042F_aapcs.pdf
> [2]
> http://infocenter.arm.com/help/topic/com.arm.doc.ihi0055b/IHI0055B_aapcs64.pdf
>
Julien Grall Nov. 24, 2016, 4:08 p.m. UTC | #20
On 24/11/16 15:50, Artem Mygaiev wrote:
>
>
> On 24.11.16 17:09, Julien Grall wrote:
>> Hi Artem,
>>
>> On 24/11/16 14:58, Artem Mygaiev wrote:
>>> On 24.11.16 15:31, Jan Beulich wrote:
>>>>>>> On 24.11.16 at 14:18, <artem_mygaiev@epam.com> wrote:
>>>>> On 24.11.16 15:04, Jan Beulich wrote:
>>>>>>> In my case I need it to at least define Linux specific __packed
>>>>>>> attribute which is not supported by Win AFAIK.
>>>>>> ... needs to avoid using platform specific constructs (or making
>>>>>> other assumptions on the platform). I can only re-iterate: Please
>>>>>> follow the model other PV protocols already present (without
>>>>>> repeating their mistakes, if at all possible).
>>>>> But if we don't pack (or enforce some specific alignment) data
>>>>> structures used between domains they may have different alignment on
>>>>> different domains, no?
>>>> Let me repeat what I've said before: Please take a look at existing
>>>> interface headers. By properly placing fields and adding explicit
>>>> padding where needed, you can avoid such layout differences.
>>>
>>> Don't get me wrong - I have checked the i/f headers before. My concern
>>> is very generic: C standard defines alignment as an
>>> *implementation-defined* integer, so current implementation is based
>>> under two assumptions on the guest platforms: a) guest compiler does not
>>> have "packing" enabled by default and b) alignment follows typical
>>> values (1 byte for char, 2 bytes for short, ...). While this is true in
>>> most cases, with some embedded RTOS this may become an issue.
>>
>> I am not sure why you mentioned embedded RTOS, this is specific to the
>> compiler and not an OS.
>>
> Indeed, my statement is not quite correct. It is just that there are
> proprietary RTOSes bundled with IDEs with own or customized tools,
> compilers, etc.

And their own ABI? It sounds crazy to me that someone may decide to use 
padding even when field are naturally aligned.

Regards,
Artem Mygaiev Nov. 24, 2016, 4:29 p.m. UTC | #21
On 24.11.16 18:08, Julien Grall wrote:
> On 24/11/16 15:50, Artem Mygaiev wrote:
>> On 24.11.16 17:09, Julien Grall wrote:
>>> Hi Artem,
>>>
>>> On 24/11/16 14:58, Artem Mygaiev wrote:
>>>> On 24.11.16 15:31, Jan Beulich wrote:
>>>>>>>> On 24.11.16 at 14:18, <artem_mygaiev@epam.com> wrote:
>>>>>> On 24.11.16 15:04, Jan Beulich wrote:
>>>>>>>> In my case I need it to at least define Linux specific __packed
>>>>>>>> attribute which is not supported by Win AFAIK.
>>>>>>> ... needs to avoid using platform specific constructs (or making
>>>>>>> other assumptions on the platform). I can only re-iterate: Please
>>>>>>> follow the model other PV protocols already present (without
>>>>>>> repeating their mistakes, if at all possible).
>>>>>> But if we don't pack (or enforce some specific alignment) data
>>>>>> structures used between domains they may have different alignment on
>>>>>> different domains, no?
>>>>> Let me repeat what I've said before: Please take a look at existing
>>>>> interface headers. By properly placing fields and adding explicit
>>>>> padding where needed, you can avoid such layout differences.
>>>>
>>>> Don't get me wrong - I have checked the i/f headers before. My concern
>>>> is very generic: C standard defines alignment as an
>>>> *implementation-defined* integer, so current implementation is based
>>>> under two assumptions on the guest platforms: a) guest compiler
>>>> does not
>>>> have "packing" enabled by default and b) alignment follows typical
>>>> values (1 byte for char, 2 bytes for short, ...). While this is
>>>> true in
>>>> most cases, with some embedded RTOS this may become an issue.
>>>
>>> I am not sure why you mentioned embedded RTOS, this is specific to the
>>> compiler and not an OS.
>>>
>> Indeed, my statement is not quite correct. It is just that there are
>> proprietary RTOSes bundled with IDEs with own or customized tools,
>> compilers, etc.
>
> And their own ABI? It sounds crazy to me that someone may decide to
> use padding even when field are naturally aligned.
No, no, that would be overkill... I am just talking about system-wide
alignment configuration for performance/size balance (reduce padding,
not increase).
diff mbox

Patch

diff --git a/include/xen/interface/io/sndif.h b/include/xen/interface/io/sndif.h
new file mode 100644
index 0000000..06079c7
--- /dev/null
+++ b/include/xen/interface/io/sndif.h
@@ -0,0 +1,609 @@ 
+/******************************************************************************
+ * sndif.h
+ *
+ * Unified sound-device I/O interface for Xen guest OSes.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Copyright (C) 2013-2015 GlobalLogic Inc.
+ * Copyright (C) 2016 EPAM Systems Inc.
+ *
+ * Authors: Oleksandr Andrushchenko <andr2000@gmail.com>
+ *          Oleksandr Grytsov <al1img@gmail.com>
+ *          Oleksandr Dmytryshyn <oleksandr.dmytryshyn@globallogic.com>
+ *          Iurii Konovalenko <iurii.konovalenko@globallogic.com>
+ */
+
+#ifndef __XEN_PUBLIC_IO_XENSND_H__
+#define __XEN_PUBLIC_IO_XENSND_H__
+
+/*
+ * Front->back notifications: When enqueuing a new request, sending a
+ * notification can be made conditional on req_event (i.e., the generic
+ * hold-off mechanism provided by the ring macros). Backends must set
+ * req_event appropriately (e.g., using RING_FINAL_CHECK_FOR_REQUESTS()).
+ *
+ * Back->front notifications: When enqueuing a new response, sending a
+ * notification can be made conditional on rsp_event (i.e., the generic
+ * hold-off mechanism provided by the ring macros). Frontends must set
+ * rsp_event appropriately (e.g., using RING_FINAL_CHECK_FOR_RESPONSES()).
+ */
+
+/*
+ * Feature and Parameter Negotiation
+ * =================================
+ * The two halves of a Para-virtual sound card driver utilize nodes within the
+ * XenStore to communicate capabilities and to negotiate operating parameters.
+ * This section enumerates these nodes which reside in the respective front and
+ * backend portions of the XenStore, following the XenBus convention.
+ *
+ * All data in the XenStore is stored as strings.  Nodes specifying numeric
+ * values are encoded in decimal.  Integer value ranges listed below are
+ * expressed as fixed sized integer types capable of storing the conversion
+ * of a properly formated node string, without loss of information.
+ *
+ *****************************************************************************
+ *                            Backend XenBus Nodes
+ *****************************************************************************
+ *
+ *-------------------------------- Addressing ---------------------------------
+ *
+ * Indices used to address frontends, driver instances, cards,
+ * devices and streams.
+ *
+ * frontend_id
+ *      Values:         <uint>
+ *
+ *      Domain ID of the sound frontend.
+ *
+ * drv_idx
+ *      Values:         <uint>
+ *
+ *      Zero based contiguous index of the virtualized sound driver instance in
+ *      this domain. Multiple PV drivers are allowed in the domain
+ *      at the same time.
+ *
+ * dev_id
+ *      Values:         <uint>
+ *
+ *      Unique device ID.
+ *      Doesn't have to be zero based and/or to be contiguous.
+ *
+ * stream_idx
+ *      Values:         <uint>
+ *
+ *      Zero based contiguous index of the stream of the device.
+ *
+ * Example for the frontend running in domain 5, instance of the driver
+ * in the front is 0 (single or first PV driver), device id 2,
+ * first stream (0):
+ * /local/domain/<frontend_id>/device/vsnd/<drv_idx>/
+ *         device/<dev_id>/stream/<stream_idx>/type = "p"
+ * /local/domain/5/device/vsnd/0/device/2/stream/0/type = "p"
+ *
+ *------------------------------- PCM settings --------------------------------
+ *
+ * Every virtualized sound frontend has set of devices and streams, each
+ * is individually configured. Part of the PCM configuration can be defined at
+ * higher level and be fully or partially re-used by the underlying layers.
+ * These configuration values are:
+ *  o number of channels (min/max)
+ *  o supported sample rates
+ *  o supported sample formats.
+ * E.g. one can define these values for the whole driver, device or stream.
+ * Every underlying layer in turn can re-define some or all of them to better
+ * fit its needs. For example, driver may define number of channels to be
+ * in [1; 8] range, and some particular stream may be limited to [1; 2] only.
+ * The rule is that the underlying layer must be a subset of the upper layer
+ * range.
+ *
+ * Note: if any of the values are not defined then PV driver should use
+ * its default values instead.
+ *
+ * channels-min
+ *      Values:         <uint>
+ *
+ *      The minimum amount of channels that is supported.
+ *      Must be at least 1. If not defined then use frontend's default.
+ *
+ * channels-max
+ *      Values:         <uint>
+ *
+ *      The maximum amount of channels that is supported.
+ *      Must be at least <channels-min>. If not defined then use frontend's
+ *      default.
+ *
+ * sample-rates
+ *      Values:         <list of uints>
+ *
+ *      List of supported sample rates separated by XENSND_LIST_SEPARATOR.
+ *      If not defined then use frontend's default. Sample rates are expressed
+ *      as a list of decimal values w/o any ordering requirement.
+ *
+ * sample-formats
+ *      Values:         <list of XENSND_PCM_FORMAT_XXX_STR>
+ *
+ *      List of supported sample formats separated by XENSND_LIST_SEPARATOR.
+ *      If not defined then use frontend's default.
+ *
+ * buffer-size
+ *      Values:         <uint>
+ *
+ *      The maximum size in octets of the buffer to allocate per stream.
+ *
+ * Example configuration:
+ *
+ * Driver configuration used by all streams:
+ * /local/domain/5/device/vsnd/0/sample-formats = "s8;u8;s16_le;s16_be"
+ * Stream overrides sample rates supported:
+ * /local/domain/5/device/vsnd/0/device/2/stream/0/sample-rates =
+ *        "8000;22050;44100;48000"
+ *
+ *----------------------- Virtual sound card settings --------------------------
+ * short-name
+ *      Values:         <char[32]>
+ *
+ *      Short name of the virtual sound card. Optional.
+ *
+ * long-name
+ *      Values:         <char[80]>
+ *
+ *      Long name of the virtual sound card. Optional.
+ *
+ * For example,
+ * /local/domain/5/device/vsnd/0/short-name = "Virtual audio"
+ * /local/domain/5/device/vsnd/0/long-name =
+ *         "Virtual audio at center stack"
+ *
+ *----------------------------- Device settings --------------------------------
+ * name
+ *      Values:         <char[80]>
+ *
+ *      Name of the sound device within the virtual sound card. Optional.
+ *
+ * For example,
+ * /local/domain/5/device/vsnd/0/device/0/name = "General analog"
+ *
+ *----------------------------- Stream settings -------------------------------
+ *
+ * type
+ *      Values:         "p", "c"
+ *
+ *      Stream type: "p" - playback stream, "c" - capture stream
+ *
+ *      If both capture and playback are needed then two streams need to be
+ *      defined under the same device. For example,
+ *      /local/domain/5/device/vsnd/0/device/0/stream/0/type = "p"
+ *      /local/domain/5/device/vsnd/0/device/0/stream/1/type = "c"
+ *
+ *****************************************************************************
+ *                            Frontend XenBus Nodes
+ *****************************************************************************
+ *
+ *----------------------- Request Transport Parameters -----------------------
+ *
+ * These are per stream.
+ *
+ * event-channel
+ *      Values:         <uint>
+ *
+ *      The identifier of the Xen event channel used to signal activity
+ *      in the ring buffer.
+ *
+ * ring-ref
+ *      Values:         <uint>
+ *
+ *      The Xen grant reference granting permission for the backend to map
+ *      a sole page in a single page sized ring buffer.
+ *
+ * index
+ *      Values:         <uint>
+ *
+ *      After stream initialization it is assigned a unique ID (within the front
+ *      driver), so every stream of the frontend can be identified by the
+ *      backend by this ID. This is not equal to stream_idx as the later is
+ *      zero based within a device, but this index is contiguous within the
+ *      driver.
+ */
+
+/*
+ * STATE DIAGRAMS
+ *
+ *****************************************************************************
+ *                                   Startup                                 *
+ *****************************************************************************
+ *
+ * Tool stack creates front and back state nodes with initial state
+ * XenbusStateInitialising.
+ * Tool stack creates and sets up frontend sound configuration nodes per domain.
+ *
+ * Front                                Back
+ * =================================    =====================================
+ * XenbusStateInitialising              XenbusStateInitialising
+ *                                       o Query backend device identification
+ *                                         data.
+ *                                       o Open and validate backend device.
+ *                                                      |
+ *                                                      |
+ *                                                      V
+ *                                      XenbusStateInitWait
+ *
+ * o Query frontend configuration
+ * o Allocate and initialize
+ *   event channels per configured
+ *   playback/capture stream.
+ * o Publish transport parameters
+ *   that will be in effect during
+ *   this connection.
+ *              |
+ *              |
+ *              V
+ * XenbusStateInitialised
+ *
+ *                                       o Query frontend transport parameters.
+ *                                       o Connect to the event channels.
+ *                                                      |
+ *                                                      |
+ *                                                      V
+ *                                      XenbusStateConnected
+ *
+ *  o Create and initialize OS
+ *  virtual sound device instances
+ *  as per configuration.
+ *              |
+ *              |
+ *              V
+ * XenbusStateConnected
+ *
+ *                                      XenbusStateUnknown
+ *                                      XenbusStateClosed
+ *                                      XenbusStateClosing
+ * o Remove virtual sound device
+ * o Remove event channels
+ *              |
+ *              |
+ *              V
+ * XenbusStateClosed
+ *
+ */
+
+/*
+ * PCM FORMATS
+ *
+ * XENSND_PCM_FORMAT_<format>[_<endian>]
+ *
+ * format: <S/U/F><bits> or <name>
+ *     S - signed, U - unsigned, F - float
+ *     bits - 8, 16, 24, 32
+ *     name - MU_LAW, GSM, etc.
+ *
+ * endian: <LE/BE>, may be absent
+ *     LE - Little endian, BE - Big endian
+ */
+#define XENSND_PCM_FORMAT_S8            0
+#define XENSND_PCM_FORMAT_U8            1
+#define XENSND_PCM_FORMAT_S16_LE        2
+#define XENSND_PCM_FORMAT_S16_BE        3
+#define XENSND_PCM_FORMAT_U16_LE        4
+#define XENSND_PCM_FORMAT_U16_BE        5
+#define XENSND_PCM_FORMAT_S24_LE        6
+#define XENSND_PCM_FORMAT_S24_BE        7
+#define XENSND_PCM_FORMAT_U24_LE        8
+#define XENSND_PCM_FORMAT_U24_BE        9
+#define XENSND_PCM_FORMAT_S32_LE        10
+#define XENSND_PCM_FORMAT_S32_BE        11
+#define XENSND_PCM_FORMAT_U32_LE        12
+#define XENSND_PCM_FORMAT_U32_BE        13
+#define XENSND_PCM_FORMAT_F32_LE        14 /* 4-byte float, IEEE-754 32-bit, */
+#define XENSND_PCM_FORMAT_F32_BE        15 /* range -1.0 to 1.0              */
+#define XENSND_PCM_FORMAT_F64_LE        16 /* 8-byte float, IEEE-754 64-bit, */
+#define XENSND_PCM_FORMAT_F64_BE        17 /* range -1.0 to 1.0              */
+#define XENSND_PCM_FORMAT_IEC958_SUBFRAME_LE 18
+#define XENSND_PCM_FORMAT_IEC958_SUBFRAME_BE 19
+#define XENSND_PCM_FORMAT_MU_LAW        20
+#define XENSND_PCM_FORMAT_A_LAW         21
+#define XENSND_PCM_FORMAT_IMA_ADPCM     22
+#define XENSND_PCM_FORMAT_MPEG          23
+#define XENSND_PCM_FORMAT_GSM           24
+
+/*
+ * REQUEST CODES.
+ */
+#define XENSND_OP_OPEN                  0
+#define XENSND_OP_CLOSE                 1
+#define XENSND_OP_READ                  2
+#define XENSND_OP_WRITE                 3
+#define XENSND_OP_SET_VOLUME            4
+#define XENSND_OP_GET_VOLUME            5
+#define XENSND_OP_MUTE                  6
+#define XENSND_OP_UNMUTE                7
+
+/*
+ * XENSTORE FIELD AND PATH NAME STRINGS, HELPERS.
+ */
+#define XENSND_DRIVER_NAME                   "vsnd"
+
+#define XENSND_LIST_SEPARATOR                ";"
+/* Path entries */
+#define XENSND_PATH_DEVICE                   "device"
+#define XENSND_PATH_STREAM                   "stream"
+/* Field names */
+#define XENSND_FIELD_VCARD_SHORT_NAME        "short-name"
+#define XENSND_FIELD_VCARD_LONG_NAME         "long-name"
+#define XENSND_FIELD_RING_REF                "ring-ref"
+#define XENSND_FIELD_EVT_CHNL                "event-channel"
+#define XENSND_FIELD_DEVICE_NAME             "name"
+#define XENSND_FIELD_TYPE                    "type"
+#define XENSND_FIELD_STREAM_INDEX            "index"
+#define XENSND_FIELD_CHANNELS_MIN            "channels-min"
+#define XENSND_FIELD_CHANNELS_MAX            "channels-max"
+#define XENSND_FIELD_SAMPLE_RATES            "sample-rates"
+#define XENSND_FIELD_SAMPLE_FORMATS          "sample-formats"
+#define XENSND_FIELD_BUFFER_SIZE             "buffer-size"
+
+/* Stream type field values. */
+#define XENSND_STREAM_TYPE_PLAYBACK          "p"
+#define XENSND_STREAM_TYPE_CAPTURE           "c"
+/* Sample rate max string length */
+#define XENSND_SAMPLE_RATE_MAX_LEN            6
+/* Sample format field values */
+#define XENSND_SAMPLE_FORMAT_MAX_LEN         24
+
+#define XENSND_PCM_FORMAT_S8_STR                 "s8"
+#define XENSND_PCM_FORMAT_U8_STR                 "u8"
+#define XENSND_PCM_FORMAT_S16_LE_STR             "s16_le"
+#define XENSND_PCM_FORMAT_S16_BE_STR             "s16_be"
+#define XENSND_PCM_FORMAT_U16_LE_STR             "u16_le"
+#define XENSND_PCM_FORMAT_U16_BE_STR             "u16_be"
+#define XENSND_PCM_FORMAT_S24_LE_STR             "s24_le"
+#define XENSND_PCM_FORMAT_S24_BE_STR             "s24_be"
+#define XENSND_PCM_FORMAT_U24_LE_STR             "u24_le"
+#define XENSND_PCM_FORMAT_U24_BE_STR             "u24_be"
+#define XENSND_PCM_FORMAT_S32_LE_STR             "s32_le"
+#define XENSND_PCM_FORMAT_S32_BE_STR             "s32_be"
+#define XENSND_PCM_FORMAT_U32_LE_STR             "u32_le"
+#define XENSND_PCM_FORMAT_U32_BE_STR             "u32_be"
+#define XENSND_PCM_FORMAT_F32_LE_STR             "float_le"
+#define XENSND_PCM_FORMAT_F32_BE_STR             "float_be"
+#define XENSND_PCM_FORMAT_F64_LE_STR             "float64_le"
+#define XENSND_PCM_FORMAT_F64_BE_STR             "float64_be"
+#define XENSND_PCM_FORMAT_IEC958_SUBFRAME_LE_STR "iec958_subframe_le"
+#define XENSND_PCM_FORMAT_IEC958_SUBFRAME_BE_STR "iec958_subframe_be"
+#define XENSND_PCM_FORMAT_MU_LAW_STR             "mu_law"
+#define XENSND_PCM_FORMAT_A_LAW_STR              "a_law"
+#define XENSND_PCM_FORMAT_IMA_ADPCM_STR          "ima_adpcm"
+#define XENSND_PCM_FORMAT_MPEG_STR               "mpeg"
+#define XENSND_PCM_FORMAT_GSM_STR                "gsm"
+
+/*
+ * STATUS RETURN CODES.
+ */
+ /* Operation failed for some unspecified reason (e. g. -EIO). */
+#define XENSND_RSP_ERROR                 (-1)
+ /* Operation completed successfully. */
+#define XENSND_RSP_OKAY                  0
+
+/*
+ * Description of the protocol between frontend and backend driver.
+ *
+ * The two halves of a Para-virtual sound driver communicates with
+ * each other using a shared page and an event channel.
+ * Shared page contains a ring with request/response packets.
+ *
+ *
+ * All request packets have the same length (64 octets)
+ *
+ *
+ * Request open - open a PCM stream for playback or capture:
+ *          0                 1                  2                3        octet
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                 id                |    operation    |     stream_idx  |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                                pcm_rate                               |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |  pcm_format     |  pcm_channels   |             reserved              |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                         gref_directory_start                          |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/|
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                               reserved                                |
+ * +-----------------+-----------------+-----------------+-----------------+
+ *
+ * id - uint16_t, private guest value, echoed in response
+ * operation - uint8_t, XENSND_OP_OPEN
+ * stream_idx - uint8_t, index of the stream ("streams_idx" XenStore entry
+ *   of the stream)
+ * pcm_rate - uint32_t, stream data rate, Hz
+ * pcm_format - uint8_t, XENSND_PCM_FORMAT_XXX value
+ * pcm_channels - uint8_t, number of channels of this stream
+ * gref_directory_start - grant_ref_t, a reference to the first shared page
+ *   describing shared buffer references. At least one page exists. If shared
+ *   buffer size exceeds what can be addressed by this single page, then
+ *   reference to the next page must be supplied (gref_dir_next_page below
+ *   is not NULL)
+ *
+ * Shared page for XENSND_OP_OPEN buffer descriptor (gref_directory in the
+ *   request) employs a list of pages, describing all pages of the shared data
+ *   buffer:
+ *          0                 1                  2                3        octet
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                          gref_dir_next_page                           |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                                num_grefs                              |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                                gref[0]                                |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                                gref[1]                                |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/|
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                                gref[N -1]                             |
+ * +-----------------+-----------------+-----------------+-----------------+
+ *
+ * gref_dir_next_page - grant_ref_t, reference to the next page describing
+ *   page directory
+ * num_grefs - number of grefs in this page
+ * gref[i] - grant_ref_t, reference to a shared page of the buffer
+ *   allocated at XENSND_OP_OPEN
+ *
+ * Request close - close an opened pcm stream:
+ *          0                 1                  2                3        octet
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                 id                |    operation    |     stream_idx  |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                               reserved                                |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/|
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                               reserved                                |
+ * +-----------------+-----------------+-----------------+-----------------+
+ *
+ * id - uint16_t, private guest value, echoed in response
+ * operation - uint8_t, XENSND_OP_CLOSE
+ * stream_idx - uint8_t, index of the stream ("streams_idx" XenStore entry
+ *   of the stream)
+ *
+ *
+ * Request read/write - used for read (for capture) or write (for playback):
+ *          0                 1                  2                3        octet
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                 id                |    operation    |     stream_idx  |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                                offset                                 |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                                length                                 |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                               reserved                                |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/|
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                               reserved                                |
+ * +-----------------+-----------------+-----------------+-----------------+
+ *
+ * id - uint16_t, private guest value, echoed in response
+ * operation - uint8_t, XENSND_OP_READ/XENSND_OP_WRITE
+ * stream_idx - uint8_t, index of the stream ("streams_idx" XenStore entry
+ *   of the stream)
+ * offset - uint32_t, read or write data offset within the shared buffer
+ *   passed with XENSND_OP_OPEN request
+ * length - uint32_t, read or write data length
+ *
+ *
+ * Request set/get volume - set/get channels' volume of the stream given:
+ *          0                 1                  2                3        octet
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                 id                |    operation    |     stream_idx  |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                               reserved                                |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/|
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                               reserved                                |
+ * +-----------------+-----------------+-----------------+-----------------+
+ *
+ * id - uint16_t, private guest value, echoed in response
+ * operation - uint8_t, XENSND_OP_SET_VOLUME or XENSND_OP_GET_VOLUME
+ * stream_idx - uint8_t, index of the stream ("streams_idx" XenStore entry
+ *   of the stream)
+ * Buffer passed with XENSND_OP_OPEN is used to exchange volume
+ * values:
+ *
+ *          0                 1                  2                3        octet
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                               channel[0]                              |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                               channel[1]                              |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * +/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/|
+ * +-----------------+-----------------+-----------------+-----------------+
+ *                  channel[XENSND_OP_OPEN.pcm_channels - 1]               |
+ * +-----------------+-----------------+-----------------+-----------------+
+ *
+ * channel[i] - sint32_t, volume of i-th channel
+ * Volume is expressed as a signed value in steps of 0.001 dB,
+ * while 0 being 0 dB.
+ *
+ *
+ * Request mute/unmute - mute/unmute stream:
+ *          0                 1                  2                3        octet
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                 id                |    operation    |     stream_idx  |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                               reserved                                |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/|
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                               reserved                                |
+ * +-----------------+-----------------+-----------------+-----------------+
+ *
+ * id - uint16_t, private guest value, echoed in response
+ * operation - uint8_t, XENSND_OP_MUTE/XENSND_OP_UNMUTE
+ * stream_idx - uint8_t, index of the stream ("streams_idx" XenStore entry
+ *   of the stream)
+ * Buffer passed with XENSND_OP_OPEN is used to exchange mute/unmute
+ * values:
+ *
+ *          0                 1                  2                3        octet
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |   channel[0]    |   channel[1]    |   channel[2]    |   channel[3]    |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * +/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/|
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |   channel[i]    |   channel[i+1]  |   channel[i+2]  |   channel[i+3]  |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * +/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/|
+ * +-----------------+-----------------+-----------------+-----------------+
+ *
+ * channel[i] - uint8_t, non-zero if i-th channel needs to be muted/unmuted
+ * Number of channels passed is equal to XENSND_OP_OPEN request pcm_channels
+ * field
+ *
+ *
+ * All response packets have the same length (64 bytes)
+ *
+ * Response for all requests:
+ *          0                 1                  2                3        octet
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                 id                |    operation    |     stream_idx  |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |      status     |                      reserved                       |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                              reserved                                 |
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/|
+ * +-----------------+-----------------+-----------------+-----------------+
+ * |                               reserved                                |
+ * +-----------------+-----------------+-----------------+-----------------+
+ *
+ * id - uint16_t, copied from the request
+ * stream_idx - uint8_t, copied from request
+ * operation - uint8_t, XENSND_OP_XXX - copied from request
+ * status - int8_t, response status (XENSND_RSP_???)
+ */
+
+struct xensnd_request {
+	uint8_t raw[64];
+};
+
+struct xensnd_response {
+	uint8_t raw[64];
+};
+
+#endif /* __XEN_PUBLIC_IO_XENSND_H__ */
diff --git a/include/xen/interface/io/sndif_linux.h b/include/xen/interface/io/sndif_linux.h
new file mode 100644
index 0000000..2802504
--- /dev/null
+++ b/include/xen/interface/io/sndif_linux.h
@@ -0,0 +1,112 @@ 
+/******************************************************************************
+ * sndif_linux.h
+ *
+ * Unified sound-device I/O interface for Xen guest OSes.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Copyright (C) 2016 EPAM Systems Inc.
+ *
+ * Authors: Oleksandr Andrushchenko <andr2000@gmail.com>
+ *          Oleksandr Grytsov <al1img@gmail.com>
+ */
+
+#ifndef __XEN_PUBLIC_IO_XENSND_LINUX_H__
+#define __XEN_PUBLIC_IO_XENSND_LINUX_H__
+
+#include <xen/interface/io/ring.h>
+#include <xen/interface/io/sndif.h>
+#include <xen/interface/grant_table.h>
+
+struct xensnd_open_req {
+	uint32_t pcm_rate;
+	uint8_t pcm_format;
+	uint8_t pcm_channels;
+	/* in Hz */
+	uint16_t __reserved0;
+	grant_ref_t gref_directory_start;
+} __packed;
+
+struct xensnd_page_directory {
+	grant_ref_t gref_dir_next_page;
+	uint32_t num_grefs;
+	grant_ref_t gref[0];
+} __packed;
+
+struct xensnd_close_req {
+} __packed;
+
+struct xensnd_write_req {
+	uint32_t offset;
+	uint32_t len;
+} __packed;
+
+struct xensnd_read_req {
+	uint32_t offset;
+	uint32_t len;
+} __packed;
+
+struct xensnd_get_vol_req {
+} __packed;
+
+struct xensnd_set_vol_req {
+} __packed;
+
+struct xensnd_mute_req {
+} __packed;
+
+struct xensnd_unmute_req {
+} __packed;
+
+struct xensnd_req {
+	union {
+		struct xensnd_request raw;
+		struct {
+			uint16_t id;
+			uint8_t operation;
+			uint8_t stream_idx;
+			union {
+				struct xensnd_open_req open;
+				struct xensnd_close_req close;
+				struct xensnd_write_req write;
+				struct xensnd_read_req read;
+				struct xensnd_get_vol_req get_vol;
+				struct xensnd_set_vol_req set_vol;
+				struct xensnd_mute_req mute;
+				struct xensnd_unmute_req unmute;
+			} op;
+		} data;
+	} u;
+};
+
+struct xensnd_resp {
+	union {
+		struct xensnd_response raw;
+		struct {
+			uint16_t id;
+			uint8_t operation;
+			uint8_t stream_idx;
+			int8_t status;
+		} data;
+	} u;
+};
+
+DEFINE_RING_TYPES(xen_sndif, struct xensnd_req, struct xensnd_resp);
+
+#endif /* __XEN_PUBLIC_IO_XENSND_LINUX_H__ */