Historically, the loopback driver used the largest packet size of all enabled link layer protocols. This permitted packets to be forward via the loopbak device with no major loss of performance. However, in experimenting with configurations where no other link layer protocols were enabled, this means the loopback packet size was set to the smallest possible size, to the SLIP minimum of 296 bytes. This resulted in terrible loopback performance.
This commit adds an option to increase the loopback packet size with the option CONFIG_NET_LOOPBACK_PACKETSIZE.
The loopback driver packet buffer should be quite large. The larger the loopback packet buffer, the better will be TCP performance of the loopback transfers. The Linux loopback device historically used packet buffers of size 16Kb, but that was increased in recent Linux versions to 64Kb. Those sizes may be excessive for resource constrained MCUs, however.
The network still enforces the lower limit that is the maximum packet size of all enabled link layer protocols. But this new option permits the loopback packet size to be increased from that.
* net/Kconfig: Adds CONFIG_NET_LOOPBACK_PKTSIZE option
* include/nuttx/net/netconfig.h: Assures that the packet size that is used is at least as large as the largest packet size of other link layer protocols.
* drivers/net/loopback.c: Use that larger packet size.
* boards/sim/sim/sim/configs/tcploop/defconfig: Set the loopback packet size to 1500
Call xxx_timer_initialize from clock subsystem to make timer ready for use as soon as possiblei and revert the workaround:
commit 0863e771a9
Author: Gregory Nutt <gnutt@nuttx.org>
Date: Fri Apr 26 07:24:57 2019 -0600
Revert "sched/clock/clock_initialize.c: clock_inittime() needs to be done with CONFIG_SCHED_TICKLESS and clock_initialize should skip clock_inittime() for external RTC case since the RTC isn't ready yet."
This reverts commit 2bc709d4b9.
Commit 2bc709d4b9 was intended to handle the case where up_timer_gettime may not start from zero case. However, this change has the side-effect of breaking every implementation of tickless mode: After this change the tickless timer structures are used before they are initialized in clock_inittime(). Initialization happens later when up_initialize is called() when arm_timer_initialize().
Since the tickless mode timer is very special, one solution might be to
1. Rename xxx_timer_initialize to up_timer_initialize
2 Move up_timer_initialize to include/nuttx/arch.h
3. Call it from clock subsystem instead up_initialize
Basically, this change make timer initialization almost same as rtc initialization(up_rtc_initialize).
For now, however, we just need to revert the change.
* include: Introduce elf64.h and elf.h
Added elf64.h for 64bit ELF support and moved common definitions
from elf32.h to elf.h. Also introduced Elf_xxx to be used in
common libraries such as binfmt.
* binfmt, include, modlib, module: Add support for ELF64
Elf_xxx must be used instead of Elf32_xxx to support ELF64.
To use ELF64, CONFIG_ELF_64BIT must be enabled.
* binfmt, modlib: Add support for relocate address
* arch: risc-v: Add include/elf.h
* libs: machine: Add risc-v related files.
NOTE: Currently only supports ELF64
* boards: maix-bit: Add elf and posix_spawn configurations
* boards: maix-bit: Add support for module configuration
If SMP is enabled this function will return the number of the CPU that the thread is running on. This is non-standard but follows GLIBC if __GNU_SOURCE is enabled. The returned CPU number is, however, worthless since it returns the CPU number of the CPU that was executing the task when the function was called. The application can never know the true CPU number of the CPU tht it is running on since that value is volatile and change change at any time.
1. Remove the unused and unimplemented ipv6_chksum declaration
2. Update NET_ARCH_CHKSUM description to align with the implementation
3. Declare all checksum function prototype regardless CONFIG_NET_ARCH_CHKSUM
4. Remove the CONFIG_NET_ARCH_CHKSUM guard for tcp_ipv[4|6]_chksum
Gregory Nutt <gnutt@nuttx.org>
Run all .c and .h files modified in this PR through nxstyle and correct all coding standard problems.
Xiang Xiao <xiaoxiang@xiaomi.com>
Remove TIME_EXTENDED option to more conform C standard
Note: the code/data size increment is small
Author: Alan Carvalho de Assis <acassis@gmail.com>
Run nxstyle against .c and .h files and fix it
Author: Juha Niskanen <juha.niskanen@haltian.com>
Fix typos and some incorrect comments
Author: Gregory Nutt <gnutt@nuttx.org>
Run all .c and .h files in last PR through tools/nxstyle and correct all coding standard complaints.
Author: Oleg <ev.mipt@gmail.com>
* Fix CAN driver to work with internal OS interfaces. Store internal file pointer in f_priv and use it then to distinct readers (See issue #111)
* Store reader pointer in f_priv instead of filep
* Don't need in filep in can_reader_s
1.Consolidate absolute to relative timeout conversion into one place(_net_timedwait)
2.Drive the wait timeout logic by net_timedwait instead of devif_timer
This patch help us remove devif_timer(period tick) to save the power in the future.
Change-Id: I534748a5d767ca6da8a7843c3c2f993ed9ea77d4
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
Here is the email loop talk about why it is better to remove the option:
https://groups.google.com/forum/#!topic/nuttx/AaNkS7oU6R0
Change-Id: Ib66c037752149ad4b2787ef447f966c77aa12aad
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
* Simplify EINTR/ECANCEL error handling
1. Add semaphore uninterruptible wait function
2 .Replace semaphore wait loop with a single uninterruptible wait
3. Replace all sem_xxx to nxsem_xxx
* Unify the void cast usage
1. Remove void cast for function because many place ignore the returned value witout cast
2. Replace void cast for variable with UNUSED macro
Author: Gregory Nutt <gnutt@nuttx.org>
Run all .h and .c files modified in last PR through nxstyle.
Author: Xiang Xiao <xiaoxiang@xiaomi.com>
Net cleanup (#17)
* Fix the semaphore usage issue found in tcp/udp
1. The count semaphore need disable priority inheritance
2. Loop again if net_lockedwait return -EINTR
3. Call nxsem_trywait to avoid the race condition
4. Call nxsem_post instead of sem_post
* Put the work notifier into free list to avoid the heap fragment in the long run. Since the allocation strategy is encapsulated internally, we can even refine the implementation later.
* Network stack shouldn't allocate memory in the poll implementation to avoid the heap fragment in the long run, other modification include:
1. Select MM_IOB automatically since ICMP[v6] socket can't work without the read ahead buffer
2. Remove the net lock since xxx_callback_free already do the same thing
3. TCP/UDP poll should work even the read ahead buffer isn't enabled at all
* Add NET_ prefix for UDP_NOTIFIER and TCP_NOTIFIER option to align with other UDP/TCP option convention
* Remove the unused _SF_[IDLE|ACCEPT|SEND|RECV|MASK] flags since there are code to set/clear these flags, but nobody check them.
Squashed commit of the following:
Author: Gregory Nutt <gnutt@nuttx.org>
Ran nxstyle against many of the affected files. But this job was too big for today. Many of the network drivers under arch are highly non-compiant and generate many, many faults from nxstyle. Those will have to be visited again another day.
Author: Xiang Xiao <xiaoxiang@xiaomi.com>
This effects all network drivers as well as timing related portions of net/: devif_poll_tcp_timer shouldn't be skipped in the multiple card case. devif_timer will be called multiple time in one period if the multiple card exist, the elapsed time calculated for the first callback is right, but the flowing callback in the same period is wrong(very short) because the global variable g_polltimer is used in the calculation. So let's pass the delay time to devif_timer and remove g_polltimer.
Squashed commit of the following:
Author: Gregory Nutt <gnutt@nuttx.org>
net/: Trivial changes to PR from review. Biggest difference: type of s_error changed to int16_t to save a byte or two.
Author: liuhuahang <liuhuahang@xiaomi.com>
Implement SO_ERROR for getsockopt()
