Multiple port support for virtio-console

Multiple port support for virtio-console

[Amit Shah]

Hello all,

Here is a new iteration of the patch series that implements a
transport for guest and host communications.

I've tested for compatibility (old qemu & new kernel, new qemu & old
kernel, new qemu & new kernel) and it all works fine*.

There are a few items on my todo list but this works well.

New since last send:
- live migration support**
- write path in the guest driver can now be called from irq context

TODO:
- Convert all config writes to little endian in qemu / convert from
little endian to host endian in guest
- Address a few FIXMEs spread in the code
- Introduce a watermark to stop a rogue host process flooding guest
with data

Conditions:
* Heavy IO on an hvc port and a non-hvc port causes memory corruption
  each time, same place. It could be realted to locking but I doubt
  that. This still has to be sorted out

** qemu now caches buffers till a char device is opened or till an
   entire write request is not received in userspace. These buffers
   aren't yet migrated.

Please review.
	Amit

Re: Multiple Port Support for virtio-console

[Amit Shah]

On (Wed) Sep 09 2009 [13:41:59], Amit Shah wrote:
> 
> Hello all,
> 
> Here is a new iteration of the patch series that implements a
> transport for guest and host communications.
> 
> I've tested for compatibility (old qemu & new kernel, new qemu & old
> kernel, new qemu & new kernel) and it all works fine*. Migration works
> with the patch Juan just posted.
> 
> There are a few items on my todo list but this works well.
> 
> New since last send:
> - migrate per-port buffers that are queued up
> - in-qemu api for open/close/read/write of ports.
>   - the read() is a callback that's invoked when complete data
>     corresponding to one write() request on the guest is available.
> - removed comments from virtio_console.c that are no longer relevant.
> - address review comments by Juan
> 
> TODO:
> - Convert all config writes to little endian in qemu / convert from
>   little endian to host endian in guest
> - Address a few FIXMEs spread in the code
> - Introduce a watermark to stop a rogue host process flooding guest
>   with data
> 
> Conditions:
> * Heavy IO on an hvc port and a non-hvc port causes memory corruption
>   each time, same place. It could be realted to locking but I doubt
>   that. This still has to be sorted out

I can't reproduce this badness anymore with the linux-next kernel in the
guest.

So I request the maintainers to please review this patchset for and
consider for inclusion.

Thanks,
		Amit

Re: Multiple Port Support for virtio-console

[Amit Shah]

On (Wed) Sep 09 2009 [13:41:59], Amit Shah wrote:
> 
> Hello all,
> 
> Here is a new iteration of the patch series that implements a
> transport for guest and host communications.
> 
> I've tested for compatibility (old qemu & new kernel, new qemu & old
> kernel, new qemu & new kernel) and it all works fine*. Migration works
> with the patch Juan just posted.
> 
> There are a few items on my todo list but this works well.
> 
> New since last send:
> - migrate per-port buffers that are queued up
> - in-qemu api for open/close/read/write of ports.
>   - the read() is a callback that's invoked when complete data
>     corresponding to one write() request on the guest is available.
> - removed comments from virtio_console.c that are no longer relevant.
> - address review comments by Juan
> 
> TODO:
> - Convert all config writes to little endian in qemu / convert from
>   little endian to host endian in guest
> - Address a few FIXMEs spread in the code
> - Introduce a watermark to stop a rogue host process flooding guest
>   with data
> 
> Conditions:
> * Heavy IO on an hvc port and a non-hvc port causes memory corruption
>   each time, same place. It could be realted to locking but I doubt
>   that. This still has to be sorted out

I can't reproduce this badness anymore with the linux-next kernel in the
guest.

So I request the maintainers to please review this patchset for and
consider for inclusion.

Thanks,
		Amit

Multiple Port Support for virtio-console

[Amit Shah]

Hello all,

Here is a new iteration of the patch series that implements a
transport for guest and host communications.

I've tested for compatibility (old qemu & new kernel, new qemu & old
kernel, new qemu & new kernel) and it all works fine*. Migration works
with the patch Juan just posted.

There are a few items on my todo list but this works well.

New since last send:
- migrate per-port buffers that are queued up
- in-qemu api for open/close/read/write of ports.
  - the read() is a callback that's invoked when complete data
    corresponding to one write() request on the guest is available.
- removed comments from virtio_console.c that are no longer relevant.
- address review comments by Juan

TODO:
- Convert all config writes to little endian in qemu / convert from
  little endian to host endian in guest
- Address a few FIXMEs spread in the code
- Introduce a watermark to stop a rogue host process flooding guest
  with data

Conditions:
* Heavy IO on an hvc port and a non-hvc port causes memory corruption
  each time, same place. It could be realted to locking but I doubt
  that. This still has to be sorted out

Please review.
	Amit

Multiple Port Support for virtio-console

[Amit Shah]

Hello all,

Here is a new iteration of the patch series that implements a
transport for guest and host communications.

I've tested for compatibility (old qemu & new kernel, new qemu & old
kernel, new qemu & new kernel) and it all works fine*. Migration works
with the patch Juan just posted.

There are a few items on my todo list but this works well.

New since last send:
- migrate per-port buffers that are queued up
- in-qemu api for open/close/read/write of ports.
  - the read() is a callback that's invoked when complete data
    corresponding to one write() request on the guest is available.
- removed comments from virtio_console.c that are no longer relevant.
- address review comments by Juan

TODO:
- Convert all config writes to little endian in qemu / convert from
  little endian to host endian in guest
- Address a few FIXMEs spread in the code
- Introduce a watermark to stop a rogue host process flooding guest
  with data

Conditions:
* Heavy IO on an hvc port and a non-hvc port causes memory corruption
  each time, same place. It could be realted to locking but I doubt
  that. This still has to be sorted out

Please review.
	Amit

Multiple port support for virtio-console

[Amit Shah]

Hello all,

Here is a new iteration of the patch series that implements a
transport for guest and host communications.

I've tested for compatibility (old qemu & new kernel, new qemu & old
kernel, new qemu & new kernel) and it all works fine*.

There are a few items on my todo list but this works well.

New since last send:
- live migration support**
- write path in the guest driver can now be called from irq context

TODO:
- Convert all config writes to little endian in qemu / convert from
little endian to host endian in guest
- Address a few FIXMEs spread in the code
- Introduce a watermark to stop a rogue host process flooding guest
with data

Conditions:
* Heavy IO on an hvc port and a non-hvc port causes memory corruption
  each time, same place. It could be realted to locking but I doubt
  that. This still has to be sorted out

** qemu now caches buffers till a char device is opened or till an
   entire write request is not received in userspace. These buffers
   aren't yet migrated.

Please review.
	Amit