Changes from 1.1.93 to 1.1.94
These changes are archived on ftp.crynwr.com:pub/kchanges
and http://www.crynwr.com/kchanges.
I cannot answer questions about Linux -- I merely summarize the kernel
patches after reading them. I try to make them useful to everyone,
but kernel hackers should take them with a grain of salt and read the
patches themselves. The patch file is patch94.gz
- More Alpha changes.
- i386 threading changed to not clone. Instead a different syscall is
used to clone.
- Ptrace, shared memory, and proc fs code added to look for the page
middle.
- Sony CDU-535 driver cleaned up.
- Wavelan driver cleaned up, processor-independent delays added.
- When shrinking filesystem buffers, we didn't look at enough buffers.
- Add a new mount option for ISO9660 fs -- "unhide" hidden files.
- Removed unnecessary free list handling in lock handler.
- Added unsigned long casts to ntohl() and ntohs() (portability).
- Swapping code changed.
- Stop destroying dead sockets when we bind a new one.
- Don't initialize the TCP retransmit timer twice.
- Serial handler changes:
- rely on the line dispcline to prevent echo wars.
- IEXTEN does not need to be enabled in order for IXANY to be active.
- Change the interrupt poll time from 60 seconds to 10 seconds,
configurable with the #define RS_STROBE_TIME.
- Add provisions for a new type of interrupt service routine, which
better supports multiple serial ports on a single IRQ.
- Add support to detect the StarTech 16650 chip. Treat it as a 16450
for now, because of its FIFO bugs.
- Support close_wait in the serial driver (see below).
Close wait:
This is helpful for slow devices (like serial
plotters) so that their outputs don't get flushed upon
device close. This has to be configurable because
normally we don't want ports to be hung up for long
periods of time during a close when they are not
connected to a device, or the device is powered off.
The default is to wait 30 seconds; in the case of a
very slow device, the close_wait timeout should be
lengthed. If it is set to 0, the kernel will wait
forever for all of the data to be transmitted.