When our netstack is replying packets when processing RX, the replied
packets will be sent without considering of TX quota because we don't
want to drop them. Now we may delay them by pushing them to a queue and
send them later, which can always keep quota working.
Signed-off-by: Zhe Weng <wengzhe@xiaomi.com>
Some drivers may set too big quota by accident and consume all of our
buffers, so we add a check when registering devices.
Signed-off-by: Zhe Weng <wengzhe@xiaomi.com>
Commit 8a63d29c removed `devif_iob_send` from `udp_sendto_buffered`
workflow, `devif_iob_send` drops too big packet. Now we still need a
place to check the packet length, otherwise a packet larger than MTU
may be sent to the net driver.
In case of similar problem happens somewhere else, this commit also
adds a check in `netdev_upperhalf`, and count these cases into
`NETDEV_TXERRORS`.
Signed-off-by: Zhe Weng <wengzhe@xiaomi.com>
Some driver like virtio-net can offload fragmented IOBs, so an interface
to support this feature is needed.
Signed-off-by: Zhe Weng <wengzhe@xiaomi.com>
Take the idea from Linux's iw_handler array and esp32c3_wlan's wlan_ops_s, and make it a common logic of upper-half driver.
Signed-off-by: Zhe Weng <wengzhe@xiaomi.com>
There is no need to use global spinlock to protect netdev
specific data counters. Allocate per-netdev specific spinlock
to get better locking granularity.
Move C/C++ atomic support checking to compiler.h
Signed-off-by: Petro Karashchenko <petro.karashchenko@gmail.com>
Before atomic is ready on every platform, we need a way to protect
quota, previously it's protected by net_lock, but this makes netpkt_xxx
interfaces unable to be called under interrupt context, so changed to
spin lock. Also, we can easily change them into atomic under this
implementation.
Signed-off-by: Zhe Weng <wengzhe@xiaomi.com>
