virtio.h: add memory operation for virtio device #541
Conversation
CV-Bowen
force-pushed
the
virtio-buff
branch
3 times, most recently
from
3c591de
to
d31e41a
Compare
lib/remoteproc/remoteproc_virtio.c
Outdated
| return; | ||
|
|
||
| rpvdev->free_buf(rpvdev, buf); | ||
| } |
There was a problem hiding this comment.
I think you should call the public functions from virtio top layer interface and let it to redirect to vdev specific alloc/free function instead of invoking the dispatch functions directly.
There was a problem hiding this comment.
Thanks for your suggestion, I'm a little confused, my motivation is that different platforms may have different memory management strategies for remoteproc transport layer, so I introduce function rproc_virtio_set_mm_callback() to let the platform set their own share memory allocate and free functions.
Do you mean we should add a memory operations to the virtio device and then we can make different virtio devices implement their own share memory management strategies?
|
This pull request has been marked as a stale pull request because it has been open (more than) 45 days with no activity. |
CV-Bowen
force-pushed
the
virtio-buff
branch
2 times, most recently
from
58d5fac
to
f58cce3
Compare
CV-Bowen
changed the title
|
Hi @CV-Bowen, Your proposal seems reasonable. For your information, in the meantime, there is a working group initiated by ARM around a new Virtio transport called virtio-msg. This group aims to propose a new Virtio transport that should support virtualization and AMP systems. One objective is also to address memory allocation challenges. My current concern with your PR is that Integrating your work now would imply that we probably have two APIs to support or deprecate this one in the short/middle term. We have to discuss with the OpenAMP team to determine the strategy we want. |
|
@arnopo Thank you for your reply and waiting for your decisions. |
We discussed the PR in the OpenAMP meeting. The decision is to postpone this PR to the next release to give more time to verify that this approach will also fit the other virtio transports. |
when will the next release out? it's always better to merge this patch immediately after the release branch? so the community could try this change as soon as possible. |
That seems reasonable to me , need first reviews from @edmooring and/or @tnmysh |
Buffer management is different for different transport layer: For MMIO transport layer, the buffer can direclty malloced from the gust os heap beacase the hypervisor can access all the memmory own by guest os. For remoteproc transpor layer, the buffer should be malloced from the share memory region to make sure the remote core can access this buffer too. So add memory ops in virtio device to make different transport/device can implement their own share memory management. Signed-off-by: Bowen Wang <[email protected]>
|
Let's move forward with this new API. One objective for the next release is to ensure that it also addresses the needs of virtio-mmio and virtio-msg. |
|
I have updated the commit message to fix some typos. commit sha1 is now 4ace354 |
Buffer management is different for different transport layer:
For MMIO transport layer, the buffer can direclty malloced from
the gust os heap beacase the hypervisor can access all the memmory own
by guest os.
For remoteproc transpor layer, the buffer should be malloced from
the share memory region to make sure the remote core can access this
buffer too.
So add memory ops in virtio device to make different transport/device can
implement their own share memory management.