[feat] add nvme support #205
Conversation
ZR233
commented
Nov 27, 2024
ZR233
commented
- add aarch64 dcache clean and invalid funcs.
- add nvme driver.
ZR233
force-pushed
the
phytium_pi
branch
5 times, most recently
from
4aee9a5
to
bd39c2b
Compare
| @@ -57,6 +57,9 @@ pub mod irq; | |||
| #[cfg(feature = "paging")] | |||
| pub mod paging; | |||
|
|
|||
| #[cfg(feature = "alloc")] | |||
| mod dma; | |||
There was a problem hiding this comment.
this is extremely strange, why the dma mod is decorated under #[cfg(feature = "alloc")]
There was a problem hiding this comment.
Dma-api use normal alloc api to alloc Vec.
There was a problem hiding this comment.
Dma-api use normal alloc api to alloc Vec.
I just do not get it, generally, the size of dma region is known and fixed, why a Vec is needed since it's usually related to dynamically arrays with unknown size.
There was a problem hiding this comment.
dma use global allocator.
There was a problem hiding this comment.
maybe we can delete this now?
| @@ -0,0 +1,26 @@ | |||
| use dma_api::Impl; | |||
There was a problem hiding this comment.
I think the Impl trait exposed by dma-api crate is strange.
I believe that you want to use this trait for the translation from IOVA to actual PA
There was a problem hiding this comment.
maybe you can refer to the DMA addr related trait defined in virtio-drivers?
https://github.com/rcore-os/virtio-drivers/blob/master/src/hal.rs
There was a problem hiding this comment.
This crate is not just about converting virtual addresses into physical addresses，but also got dma direction usage by map func, when read from data, crate will invlid cache, when write, it will flush back from cache to memory, so that we can use normal type memory for dma, and direct use upstream &[u8] without copy.
There was a problem hiding this comment.
When use &mut [u8] from os, map to dma slice type, and when give back &[u8] to os, umap to normal memory. If use like virtio-drivers, we need to copy &[u8] to no cache type memory for dma and copy back to &[u8] for send to os.
There was a problem hiding this comment.
Sorry, I need to learn how dma_api works before offering any more valuable suggestions.
kind of busy lately, so this may take a little bit longer
ZR233
force-pushed
the
phytium_pi
branch
3 times, most recently
from
c55cdad
to
3f0576d
Compare
| @@ -57,6 +57,9 @@ pub mod irq; | |||
| #[cfg(feature = "paging")] | |||
| pub mod paging; | |||
|
|
|||
| #[cfg(feature = "alloc")] | |||
| mod dma; | |||
There was a problem hiding this comment.
maybe we can delete this now?
| @@ -0,0 +1,26 @@ | |||
| use dma_api::Impl; | |||
There was a problem hiding this comment.
Sorry, I need to learn how dma_api works before offering any more valuable suggestions.
kind of busy lately, so this may take a little bit longer
| } | ||
| } | ||
|
|
||
| dma_api::set_impl!(DmaImpl); |
There was a problem hiding this comment.
BTW, this is kind of strange, too.
There was a problem hiding this comment.
It a good way for singleton trait. It is widely used in embedded-hal group, like critical-section
