diff options
Diffstat (limited to 'target-mips')
-rw-r--r-- | target-mips/TODO | 27 |
1 files changed, 26 insertions, 1 deletions
diff --git a/target-mips/TODO b/target-mips/TODO index 2d207cf900..bc287f63a5 100644 --- a/target-mips/TODO +++ b/target-mips/TODO @@ -3,7 +3,32 @@ Unsolved issues/bugs in the mips/mipsel backend General ------- -- [ls][dw][lr] report broken (aligned) BadVAddr +- Unimplemented ASEs: + - MIPS16 + - MDMX + - SmartMIPS + - DSP r1 + - DSP r2 +- MT ASE only partially implemented and not functional +- Shadow register support only partially implemented, + lacks set switching on interrupt/exception. +- 34K ITC not implemented. +- A general lack of documentation, especially for technical internals. + Existing documentation is x86-centric. +- Reverse endianness bit not implemented +- The TLB emulation is very inefficient: + Qemu's softmmu implements a x86-style MMU, with separate entries + for read/write/execute, a TLB index which is just a modulo of the + virtual address, and a set of TLBs for each user/kernel/supervisor + MMU mode. + MIPS has a single entry for read/write/execute and only one MMU mode. + But it is fully associative with randomized entry indices, and uses + up to 256 ASID tags as additional matching criterion (which roughly + equates to 256 MMU modes). It also has a global flag which causes + entries to match regardless of ASID. + To cope with these differences, Qemu currently flushes the TLB at + each ASID change. Using the MMU modes to implement ASIDs hinges on + implementing the global bit efficiently. MIPS64 ------ |