| Message ID | 1490701787-15205-1-git-send-email-loic.pallardy@st.com |
|---|---|
| Headers | show |
| Series | virtio_rpmsg: make rpmsg channel configurable | expand |
On 03/28/2017 01:49 PM, Loic Pallardy wrote: > This goal of this series is to offer more flexibility for virtio_rpmsg > communication link configuration. > > It should be possible to tune rpmsg buffer size per Rpmsg link, to provide > selected configuration to coprocessor, to take into account coprocessor > specificities like memory region. > > This series introduces an optional virtio rpmsg configuration structure, which > will be managed as an extension of struct fw_rsc_vdev present in coprocessor firmware > resource table. > Structure access is possible thanks to virtio get and set API. > This structure contains the different information to tune the link between the > host and the coprocessor. > > Patch "rpmsg: virtio_rpmsg_bus: fix sg_set_buf() when addr is not a valid > kernel address" has the same goal as Wendy's RFC [1]. I don't have tested > Wendy's series with level driver buffer allocation. > > Last patch "rpmsg: virtio_rpmsg: set buffer configuration to virtio" has a > dependency with remoteproc subdevice boot sequence as coprocessor resource > table must be updated before coprocessor start. See [2] > > [1] http://www.spinics.net/lists/linux-remoteproc/msg00852.html > [2] http://www.spinics.net/lists/linux-remoteproc/msg00826.html > > --- > Changes from v1: > - No more dependency with [2]. Coprocessor firmware should wait for first kick > indicating virtio rpmsg link ready before accessing resource table information. > - Correct issues reported by Suman > - Fix warnings found by checkpatch with --strict option > > Changes from v2: > - Fix warnings reported by kbuild test robot > > Changes from v3: > After discussions with Bjorn, virtual address sharing between low level driver > and virtio_rpmsg could be replaced by associated a dedicated memory region to > vdev and changing virtio_rpmsg allocation from dev->parent->parent to > dev->parent, i.e. vdev and not rproc device. > The advantage of the solution is that there will be only few changes in > virtio_rpmsg layer and it will be common with virtio_console driver. > These will be part of a dedicated patch series. > So v4 drops previous patch 5 "rpmsg: virtio_rpmsg: don't allocate buffer if > provided by low level driver" dealing with buffer previously allocated and va > fields from struct virtio_rpmsg_cfg. > In addition, patch series has been rebased on kernel 4.11-rc4. > Hi Bjorn, Any time to review this series? Regards, Loic > > Loic Pallardy (5): > rpmsg: virtio_rpmsg: set rpmsg_buf_size customizable > rpmsg: virtio_rpmsg_bus: fix sg_set_buf() when addr is not a valid > kernel address > include: virtio_rpmsg: add virtio rpmsg configuration structure > rpmsg: virtio_rpmsg: get buffer configuration from virtio > rpmsg: virtio_rpmsg: set buffer configuration to virtio > > drivers/rpmsg/virtio_rpmsg_bus.c | 123 +++++++++++++++++++++++++++++++++---- > include/linux/rpmsg/virtio_rpmsg.h | 30 +++++++++ > 2 files changed, 141 insertions(+), 12 deletions(-) > create mode 100644 include/linux/rpmsg/virtio_rpmsg.h >
This goal of this series is to offer more flexibility for virtio_rpmsg communication link configuration. It should be possible to tune rpmsg buffer size per Rpmsg link, to provide selected configuration to coprocessor, to take into account coprocessor specificities like memory region. This series introduces an optional virtio rpmsg configuration structure, which will be managed as an extension of struct fw_rsc_vdev present in coprocessor firmware resource table. Structure access is possible thanks to virtio get and set API. This structure contains the different information to tune the link between the host and the coprocessor. Patch "rpmsg: virtio_rpmsg_bus: fix sg_set_buf() when addr is not a valid kernel address" has the same goal as Wendy's RFC [1]. I don't have tested Wendy's series with level driver buffer allocation. Last patch "rpmsg: virtio_rpmsg: set buffer configuration to virtio" has a dependency with remoteproc subdevice boot sequence as coprocessor resource table must be updated before coprocessor start. See [2] [1] http://www.spinics.net/lists/linux-remoteproc/msg00852.html [2] http://www.spinics.net/lists/linux-remoteproc/msg00826.html --- Changes from v1: - No more dependency with [2]. Coprocessor firmware should wait for first kick indicating virtio rpmsg link ready before accessing resource table information. - Correct issues reported by Suman - Fix warnings found by checkpatch with --strict option Changes from v2: - Fix warnings reported by kbuild test robot Changes from v3: After discussions with Bjorn, virtual address sharing between low level driver and virtio_rpmsg could be replaced by associated a dedicated memory region to vdev and changing virtio_rpmsg allocation from dev->parent->parent to dev->parent, i.e. vdev and not rproc device. The advantage of the solution is that there will be only few changes in virtio_rpmsg layer and it will be common with virtio_console driver. These will be part of a dedicated patch series. So v4 drops previous patch 5 "rpmsg: virtio_rpmsg: don't allocate buffer if provided by low level driver" dealing with buffer previously allocated and va fields from struct virtio_rpmsg_cfg. In addition, patch series has been rebased on kernel 4.11-rc4. Loic Pallardy (5): rpmsg: virtio_rpmsg: set rpmsg_buf_size customizable rpmsg: virtio_rpmsg_bus: fix sg_set_buf() when addr is not a valid kernel address include: virtio_rpmsg: add virtio rpmsg configuration structure rpmsg: virtio_rpmsg: get buffer configuration from virtio rpmsg: virtio_rpmsg: set buffer configuration to virtio drivers/rpmsg/virtio_rpmsg_bus.c | 123 +++++++++++++++++++++++++++++++++---- include/linux/rpmsg/virtio_rpmsg.h | 30 +++++++++ 2 files changed, 141 insertions(+), 12 deletions(-) create mode 100644 include/linux/rpmsg/virtio_rpmsg.h -- 1.9.1