The link status was previously not returned via the value parameter, as required
by the netdev_driver_t API. As a result, e.g. the `ifconfig` shell command
showed garbage.
Using the TER bit in the TX descriptors when only using a single descriptor for
sending triggered a hardware bug. Thus, stop using the TER bit and store the
currently active TX descriptor in RAM instead.
The stm32_eth driver was build on top of the internal API periph_eth, which
was unused anywhere. (Additionally, with two obscure exceptions, no functions
where declared in headers, making them pretty hard to use anyway.)
The separation of the driver into two layers incurs overhead, but does not
result in cleaner structure or reuse of code. Thus, this artificial separation
was dropped.
The Ethernet DMA is capable of collecting a frame from multiple chunks, just
like the send function of the netdev interface passes. The send function was
rewritten to just set up the Ethernet DMA up to collect the outgoing frame
while sending. As a result, the send function blocks until the frame is
sent to keep control over the buffers.
This frees 6 KiB of RAM previously used for TX buffers.
1. Move buffer configuration from boards to cpu/stm32
2. Allow overwriting buffer configuration
- If the default configuration ever needs touching, this will be due to a
use case and should be done by the application rather than the board
3. Reduce default RX buffer size
- Now that handling of frames split up into multiple DMA descriptors works,
we can make use of this
Note: With the significantly smaller RX buffers the driver will now perform
much worse when receiving data at maximum throughput. But as long as frames
are small (which is to be expected for IoT or boarder gateway scenarios) the
performance should not be affected.
If any incoming frame is bigger than a single DMA buffer, the Ethernet DMA will
split the content and use multiple DMA buffers instead. But only the DMA
descriptor of the last Ethernet frame segment will contain the frame length.
Previously, the frame length calculation, reassembly of the frame, and the
freeing of DMA descriptors was completely broken and only worked in case the
received frame was small enough to fit into one DMA buffer. This is now fixed,
so that smaller DMA buffers can safely be used now.
Additionally the interface was simplified: Previously two receive flavors were
implemented, with only one ever being used. None of those function was
public due to missing declarations in headers. The unused interface was
dropped and the remaining was streamlined to better fit the use case.
- Added missing wait for TX flush
- Grouped access to the same registers of the Ethernet PHY to reduce accesses.
(The compiler won't optimize accesses to `volatile`, as defined in the C
standard.)
- Add missing `volatile` to DMA descriptor, as memory is also accessed by the
DMA without knowledge of the compiler
- Dropped `__attribute__((packed))` from DMA descriptor
- The DMA descriptor fields need to be aligned on word boundries to
properly function
- The compiler can now more efficiently access the fields (safes ~300 B ROM)
- Moved the DMA descriptor struct and the flags to `periph_cpu.h`
- This allows Doxygen documentation being build for it
- Those types and fields are needed for a future PTP implementation
- Renamed DMA descriptor flags
- They now reflect to which field in the DMA descriptor they refer to, so
that confusion is avoided
- Added documentation to the DMA descriptor and the corresponding flags
