| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Handle lazy fpu context switches correctly for ptrace(2).
- Don't read $fcr31 on context switch, it's pure bloat.
- New processes get $fcr31 initialized to no exceptions.
- First beginnings of floating point support kernel code. For now we
only can handle cvt.w.s and cvt.w.d instructions where the source
register is a NaN, infinity or denorm. This is good enough to get
Mozilla up.
- Don't send SIGFPE on every floating point instruction we don't know
how to handle, rather just complain.
- Cleanup headerfile inclusions in the HPC3 code.
- Fix the definition of PAGE_NONE.
|
| |
|
|
| |
fbmem (framebuffer device)
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
user would expect it to work, like under IRIX with the exception
that the machine doesn't freeze solid as long as the button is
pressed.
o Realtime clock driver (CONFIG_SGI_DS1286) for the Indy. Includes
/proc/rtc interface which tells you everything about your RTC which
you never wanted to know. You have to configure this driver into
your kernel or you won't be able to set your RTC.
o Some cleanup of the RTC interfaces in Linux, RTC does no longer mean
PC-style RTC. The userland interface definitions are now in
<linux/rtc.h>. We still need some more cleanup until we can enable
several RTC drivers.
o Virtual aliasing fixes for ptrace(2).
o Add ptrace(2) interface definitions needed to build GDB to
<asm/ptrace.h>.
o Revert to some older routines, the newer optimized ones are buggy.
o On panic the Indy LED will be blinking fast crying to the admin
passing by [music from Rocky Horror show] ``fix me, fix me, fix
me''.
o On panic the power button will only reboot the machine, not power it
down. That's probably more what people want it to do.
o Remove some old files.
o Did I tell you that CVS is buggy?
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
o Cleanup the machine dependencies of floppy and rtc. The driver for
the Dallas thingy in the Indy is still missing.
o Handle allocation of zero'd pages correct for R4000SC / R4400SC.
o Page colouring shit to match the virtual and physical colour of all
mapped pages. This tends to produce extreme fragmentation problems,
so it's deactivated for now. Users of R4000SC / R4400SC may re-enable
the code in arch/mips/mm/init.c by removing the definition of
CONF_GIVE_A_SHIT_ABOUT_COLOURS. Should get them somewhat further -
but don't shake to hard ...
o Fixed ptrace(2)-ing of syscalls, strace is now working again.
o Fix the interrupt forwarding from the keyboard driver to the psaux
driver, PS/2 mice are now working on the Indy. The fix is somewhat
broken as it prevents generic kernels for Indy and machines which handle
things different.
o Things I can't remember.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
| |
o Fix ancient bug in the ELF loader making ldd crash.
o Fix ancient bug in the keyboard code for SGI, SNI and Jazz.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
o New clear_user.
o Memcpy now efficiently copies the (src^dest)&3 != 0.
o Memmove new correctly deals with overlaps
o Rewrite csum_partial in assembler.
o Rewrte csum_partial_from_user in assembler.
o __copy_user is now integrated with memcpy.
o get_user now returns a zero value on error.
o copy_from_user now clears the destination buffer on error.
o strncpy_user now has a more efficient caller routine.
o strlen_user now has a more efficient caller routines and is faster.
o The unaligned handler is now much cleaner. It's now also save from
interrupt. Some more esotheric bugs fixed as well.
o Don't export bcopy anymore, it's now a inline function.
o Delete ancient junk from the first days of Linux/MIPS.
o Delete dead code in indy_sc.c.
o Including the IDE driver doesn't crash an Indy anymore.
o Eleminate active_ds. We now use current_ds directly in the thread
structure which is faster and threadsafe. Saves almost 2kb on the
kernel.
o Serial console should work again.
|
| |
|
|
|
|
|
|
|
| |
o First round of bugfixes for the SC/MC CPUs.
o FPU context switch fixes.
o Lazy context switches.
o Faster syscalls.
o Removed dead code.
o Shitloads of other things I forgot ...
|
| |
|
|
|
|
|
|
|
|
| |
Pentium :-)
o Little bit smarter handling of unimplemented exceptions.
o Fix FPU context switches.
o Fix reboot / halt. Powerdown in software still doesn't work.
o Fix the fix for handling of return values of interrupted syscalls.
o Handling of the Indy second level cache now works as spec'ed. Purely
cosmentic, this was not causing any problems.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
o Just count missed heartbeats on Indys but don't print a messages. This
was a bug by itself.
o Fix a crash in exit_mmap().
o Fix scanmem().
o Fix clear_active_bh(), a ~ was missing causing weak performance.
o Remove experimental cache instructions from stackframe.h. It wasn't
improving performance as I was hoping.
o Shrink the size of the exception handler routines in uaccess.h.
o Cleanup, remove dead code in various files.
|
| |
|
|
|
| |
o Divide L1 cache sizes by 1024 before printing, makes the numbers a
bit more credible ...
|
| | |
|
| |
|
|
|
|
|
| |
o New faster syscalls for 2.1.x, too
o Upgrade to 2.1.89.
Don't try to run this. It's flaky as hell. But feel free to debug ...
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
o Fix interrupt handler to not loose cycles
o Warn about lost interrupts.
o Fix calibration of the R4000 CPU timer.
Still to do:
o General cleanup of the timer code
o Microsecond timers for the Indy
o The calculated values for r4k_offset are still off by ~480 from the
theoretical values. That means we're going to loose about 46 s per
day. Are the crystals that bad or is there still a bug hidden
somewhere? Maybe an option to set r4k_offset to a user supplied
value might help?
o Handle lost interrupts more gently so we don't loose time.
o Handle 100MHz R4000 count / compare bug.
|
| | |
|
| |
|
|
|
|
| |
routine by one third. I hope I fixed the problems with the second
level cache of R4000SC / R4400SC - I can't test it since I only have
other MIPS variants.
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
The new signal code with exception of the code for the rt signals.
The definitions in <asm/siginfo.h> and <asm/ucontext.h> are currently
just stolen from the Alpha and will need to be overhauled.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
slap slap slap.
actually return the value of vmalloc.
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
information. This needs to be fixed at least to report the proper
graphics capabilities on the machine. Optimally, we should scatter
the right calls to the inventory all over the kernel source.
|
| |
|
|
|
|
|
|
| |
inode.c: - Formatting, minor change to start supporting indirect extents.
dir.c: - Prevent access to an indirect directory from hosing the system.
*.h: - Add forgotten include files.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
On some MIPS boxes the firmware doesn't do that for us.
- We still had two unprotected loads in the sys_sigrestore(2). Use
__get_user().
- Handle QED-style L1 caches != 16kb per cache correctly.
- Protect the cacheflush instructions for the signal trampoline
just like the loads in __get_user(). Otherwise the following
code will result in a nice "Can not handle kernel paging request"
message:
#include <signal.h>
static void hurz(void)
{
}
main()
{
signal(SIGSEGV, hurz);
/* Chainsaw the stack pointer ... */
asm("move $29,%0" : :"r"(0x70000000));
*(int *) 8 = 0;
}
With the fix applied I still get the error message. The cause of
this problem is that gas produces an __ex_table section with wrong
contents. Oh well, how good that I have a nice working post 2.8.1
binutils version at home in good ol' Germany ...
|
| |
|
|
| |
my script to do that broke :-(
|
| |
|
|
| |
CVS archive.
|
| | |
|
| | |
|
| | |
|
