Committed on the Free edition of March Hare Software CVSNT Server.
Upgrade to CVS Suite for more features and support:
http://march-hare.com/cvsnt/


git-svn-id: http://moon:8086/svn/vhdl/trunk@684 cc03376c-175c-47c8-b038-4cd826a8556b
This commit is contained in:
2009-11-24 12:06:21 +00:00
parent 58316c2141
commit f33f341adb
+11 -2
View File
@@ -23,7 +23,7 @@ Changes in release 12
- gpio_wb.vhd: added 32 bit timer with interrupt and auto reload
- added debugger in libsys. New files are mipsdis.c/h and dbg.c/h
- exception handlers can now be registered to any exception type
- bugfix: cop0 registers were not written during exception commit. Cop registers now behave like general registers.
- bugfix mips_pipeline/mips_cop: cop0 registers were not written during exception commit. Cop registers now behave like general registers.
- added pin 'eb' to mips_top.vhd to select CPU-endianess at reset time (eb = 0 : little-endian, eb = 1 : big-endian)
- Status register bit 'RE'= 1 (bit 25) reverses endianess in user-mode
- Status register bit 'TS' (TLB Shutdown, bit 21) is tied high to indicate the absence of an MMU.
@@ -36,7 +36,7 @@ Changes in release 12
Drawback of using NMI/RST: Unitialized/unreseted periphery may cause trouble (e.g. firing unwanted interrupts), but I think you know that already.
- Bootloader: Startup.S now invalidates I/D-Caches at boot.
- uart_wb.vhd: fixed crappy tx-empty interrupt, Added Tx-IRQ acknowledge.
- Bugfix: register bypass with same target register, used for unaligned loads (LWL, LWR), was incorrect.
- Bugfix in mips_pipeline: register bypass with same target register, used for unaligned loads (LWL, LWR), was incorrect.
Newlib's MIPS-optimized version of memcpy() uses this for unaligned copy which revealed the bug.
Symptom: Consecutive load instructions without nops using same target register (which is legal) like
lw $1, 0(mem)
@@ -45,6 +45,15 @@ Changes in release 12
lwl $1, 0(mem)
lwr $1, 1(mem)
or similar, lead to incorrect result. Now it's fixed.
- Bugfix mips_biu: instructions uncached LW => cached LW (with cache-miss) was buggy.
Symptom:
...
1.: lw $a0, 32($s0) # Uncached load from memory mapped I/O
2.: lw $a2, 4($v0) # Cached load and cache-miss.
# $a0 receives same value as $a2.
# CPU skips writing $a0 due to busy flag of D-Cache,
# which is already active in the previous cycle
...
-----------------------------------------------------------------
Changes in release 11