If lwIP placed more than 2 pbufs in a TCP segment, the ethernet driver
would fail to send it as it didn't have enough Tx descriptors. The
maximum number of pbufs outstanding for transmit that lwIP keeps is
defined by the TCP_SND_QUEUELEN macro. I modifed the value of
LPC_NUM_BUFF_TXDESCS to take advantage of this lwIP value. The +1
takes into account that LPC_EMAC->TxProduceIndex ==
LPC->TxConsumeIndex is reserved for indicating that the queue is empty
so a full queue uses one less than the maximum count.
tcp_write() would incorrectly byte swap the checksum 1 too many times
when concatenating a pbuf to an existing TCP segment if the number of
bytes in the concatenated data was odd. I hit this issue when I tried
to reproduce a lost segment issue reported by a mbed user in this forum
thread: http://mbed.org/forum/mbed/topic/4354/?page=2#comment-22657
If don't know if this is an issue that anyone cares about. I am also
not sure what the best way to solve it is either. I just thought I
would issue a pull request with this commit to bring the issue to light
and show a possible solution that I have tested on my mbed-1768 device.
Previously the serial_putc() API didn't make any use of the Tx FIFO
since the serial_writable() API it utilizes only returns true when the
FIFO is completely empty. This is due to the fact that the THRE bit of
the UART's LSR (Line Status Register) only goes high when the whole
FIFO is empty.
I noticed this when doing some performance testing with the network
stack. I went from calling printf() to output 3 bytes every 10 seconds
(with packet drop stats) to instead output 4 bytes every 10 seconds.
I thought these should easily fit in the 16 byte FIFO but outputting
one extra byte caused an additional three 550 byte UDP packets to be
dropped. This should only happen if the additional character being
sent to the UART was taking away extra CPU cycles from the network
stack.
My solution is to keep track of the number of bytes that have been
placed in the Tx FIFO since it was last detected as being completely
empty (via the THRE bit). Only once this count hits 16 does the code
then block, waiting for the THRE bit to go high. Each time the THRE
bit does go high, the count is reset to 0 again and it is incremented
for each byte that is loaded into the THR.
Because the LPC81X HAL implementation calls error(), which in turn calls
fprintf(), quite a bit of code is added to the image, which is not a good
idea on such resource constrained targets.
The new RawSerial class is a simple wrapper over the serial HAL that can
be safely used from an interrupt handler.
Interrupt chaining code was removed from InterruptIn, Serial and Ticker
because it caused lots of issues with the RTOS. Interrupt chaining is
still possible using the InterruptManager class.
The test is not automated now, so it can be synced with the official
repository.
Also changed it to look more like the original VodafoneUSBModem test by
adding another thread in the code.
In the future, USBHostConf.h should really reside in the project that
imports USBHost, not inside the USBHost library itself. Doing that now
though might break compatibility with projects that currently import
USBHost, so we need to figure out a better solution.
0xc0170
Bogdan Marinescu
Bogdan Marinescu
Yihui Xiong
Andreas Rebert
Adam Green
pbrier
Emilio Monti
Joe Turner
Joris Aerts
Michael Ammann