[Spice-devel] [PATCH spice-server v2 00/14] char device flow control v2

[Hans de Goede]

Series looks good to me, ack series!

Regards,

Hans


On 07/03/2012 01:33 PM, Yonit Halperin wrote:
> Changes from v1:
> Fixes for the comments Hans and Alon made. Mainly for patch 3:
>
> * use do_flow_control flag, instead of ~0 for infinite tokens
> * avoid code duplication in spice_char_device_send_to_client_tokens_set/add
> * fix bug in num_self_token counting when the cur_write_buf was associated with
>    a client that was destroyed.
> * moved the recursion tracking in read_from_device to the device state, instead of using
>    a static variable
> * fix ref counting for the char device state
> * add some asserts
> * removed a routine that wasn't used: spice_char_device_client_num_tokens_get
>
> ----------------------------
> Hi,
>
> This patch series extends SpiceCharDeviceState to manage the character devices flow control:
> wakeup and reading from the device; writing to the device; start/stop the device; server/client tokens
> management, including partial support for multiple clients.
> The agent, spicevmc, and smartcard, were modified to use SpiceCharDeviceState.
>
> Functional differences from the previous implementation:
> (1) Api for starting and stopping the device: when the device is stopped, incoming messages from the client are kept in the
> server till the device is started (or migrated - not implemented yet).
> (2) vdagent: not ignoring tokens received from the client; generalize token management for multiple clients.
> (3) smartcard, spicevmc:
>      (a) support partial writes; while this might not be necessary for usbredir, it may be required for future devices.
>      (b) smartcard and spicevmc channels can be easily extended to support tokening.
>
> The patches serve also as a preliminary work for supporting seamless migration by:
> (1) having start/stop api: we need to stop writing to the guest when savevm starts
> (2) having write queues for all character devices and not just the agent:
>      we need to save data that reached the server after the guest was stopped.
> (3) having tokens support: in addition to the ongoing flow control it provides,
>      it also allows us to set an upper limit to the amount of data we might need to migrate.
>
> The series also contain some bug fixes.
>
> Regards,
> Yonit.
>
> Yonit Halperin (14):
>    smartcard: fix calc of remaining data size when reading more than one
>      msg from the device
>    smartcard: fix an error message
>    char_device: Introducing shared flow control code for char devices.
>    agent: Fix tokens handling in main_channel
>    agent: remove save/restore migration data code
>    agent: employ SpiceCharDeviceState for managing reading from the
>      device
>    agent: use SpiceCharDeviceWriteBuffer for agent data from the client
>    agent: employ SpiceCharDeviceState for writing to the device
>    spicevmc: employ SpiceCharDeviceState for managing reading from the
>      guest device
>    spicevmc: use SpiceCharDeviceState for writing to the guest device
>    smartcard: use SpiceCharDeviceState for managing reading from the
>      device
>    smartcard: creating SmartCardChannelClient type
>    smartcard: use SpiceCharDeviceState for writing to the guest device
>    char_device: move SpiceCharDeviceState from the header
>
>   server/Makefile.am    |    1 +
>   server/char_device.c  |  775 +++++++++++++++++++++++++++++++++++++++++++++
>   server/char_device.h  |  183 +++++++++++-
>   server/main_channel.c |   94 +++----
>   server/main_channel.h |    6 +-
>   server/reds.c         |  842 ++++++++++++++++++-------------------------------
>   server/reds.h         |    5 +-
>   server/smartcard.c    |  367 ++++++++++++++++------
>   server/smartcard.h    |    6 +-
>   server/spicevmc.c     |  193 ++++++++----
>   10 files changed, 1713 insertions(+), 759 deletions(-)
>   create mode 100644 server/char_device.c
>