This patch series allow qemu-system-loongarch64 to emulate a LoongArch32
machine. A mode enum is added to CPUArchState to select LA32 or LA64 at
runtime. A new CPU model is added for loongarch32. Initial GDB support
is added.

Changes since v2:

- Fix typo in previous commit
- Fix VPPN width in TLBEHI/TLBREHI

Changes since v1:

- No longer create a separate qemu-system-loongarch32 executable, but
  allow user to run loongarch32 emulation using qemu-system-loongarch64
- Add loongarch32 cpu support for virt machine

Full changes:

Jiajie Chen (6):
  target/loongarch: Add loongarch32 mode for loongarch64-softmmu
  target/loongarch: Add loongarch32 cpu la132
  target/loongarch: Add GDB support for loongarch32 mode
  target/loongarch: Support LoongArch32 TLB entry
  target/loongarch: Support LoongArch32 DMW
  target/loongarch: Support LoongArch32 VPPN

 configs/targets/loongarch64-softmmu.mak |  2 +-
 gdb-xml/loongarch-base32.xml            | 45 ++++++++++++++++
 hw/loongarch/virt.c                     |  5 --
 target/loongarch/cpu-csr.h              | 22 ++++----
 target/loongarch/cpu.c                  | 51 +++++++++++++++++-
 target/loongarch/cpu.h                  | 18 +++++++
 target/loongarch/gdbstub.c              | 32 ++++++++---
 target/loongarch/tlb_helper.c           | 71 ++++++++++++++++++++-----
 8 files changed, 209 insertions(+), 37 deletions(-)
 create mode 100644 gdb-xml/loongarch-base32.xml

-- 
2.39.2

On 8/7/23 02:44, Jiajie Chen wrote:
> This patch series allow qemu-system-loongarch64 to emulate a LoongArch32
> machine. A mode enum is added to CPUArchState to select LA32 or LA64 at
> runtime. A new CPU model is added for loongarch32. Initial GDB support
> is added.
> 
> Changes since v2:
> 
> - Fix typo in previous commit
> - Fix VPPN width in TLBEHI/TLBREHI
> 
> Changes since v1:
> 
> - No longer create a separate qemu-system-loongarch32 executable, but
>    allow user to run loongarch32 emulation using qemu-system-loongarch64
> - Add loongarch32 cpu support for virt machine
> 
> Full changes:
> 
> Jiajie Chen (6):
>    target/loongarch: Add loongarch32 mode for loongarch64-softmmu
>    target/loongarch: Add loongarch32 cpu la132
>    target/loongarch: Add GDB support for loongarch32 mode
>    target/loongarch: Support LoongArch32 TLB entry
>    target/loongarch: Support LoongArch32 DMW
>    target/loongarch: Support LoongArch32 VPPN

There are changes missing for the translator.

All of the doubleword (D) instructions must be rejected in LA32 mode.

Virtual addresses must be zero-extended from 32 bits in 32-bit addressing mode.

I see a note about VA32L1/VA32L2/VA32L3 in CSR.MISC enabling 32-bit address mode for LA64. 
  You would want to implement this with a HW_FLAGS bit which indicates to the translator 
whether 32-bit addressing is enabled in the current mode.  This would always be true for 
LA32, and from MISC for the current priv level for LA64.

There are changes to BL, JIRL and PCADD* to sign-extend in 32-bit address mode.

What happened to the PWCH adjustment from v1?


r~

On 2023/8/7 23:40, Richard Henderson wrote:
> On 8/7/23 02:44, Jiajie Chen wrote:
>> This patch series allow qemu-system-loongarch64 to emulate a LoongArch32
>> machine. A mode enum is added to CPUArchState to select LA32 or LA64 at
>> runtime. A new CPU model is added for loongarch32. Initial GDB support
>> is added.
>>
>> Changes since v2:
>>
>> - Fix typo in previous commit
>> - Fix VPPN width in TLBEHI/TLBREHI
>>
>> Changes since v1:
>>
>> - No longer create a separate qemu-system-loongarch32 executable, but
>>    allow user to run loongarch32 emulation using qemu-system-loongarch64
>> - Add loongarch32 cpu support for virt machine
>>
>> Full changes:
>>
>> Jiajie Chen (6):
>>    target/loongarch: Add loongarch32 mode for loongarch64-softmmu
>>    target/loongarch: Add loongarch32 cpu la132
>>    target/loongarch: Add GDB support for loongarch32 mode
>>    target/loongarch: Support LoongArch32 TLB entry
>>    target/loongarch: Support LoongArch32 DMW
>>    target/loongarch: Support LoongArch32 VPPN
>
> There are changes missing for the translator.
>
> All of the doubleword (D) instructions must be rejected in LA32 mode.
I was trying to do this, but I failed to figure out how to read the 
current cpucfg when translating instructions to TCP ops. This problem 
applies to the mode-specific behavior below: VA32L1, BL, JIRL, PCADD*, 
PWCH etc.
>
> Virtual addresses must be zero-extended from 32 bits in 32-bit 
> addressing mode.
It is done in the TLB change.
>
> I see a note about VA32L1/VA32L2/VA32L3 in CSR.MISC enabling 32-bit 
> address mode for LA64.  You would want to implement this with a 
> HW_FLAGS bit which indicates to the translator whether 32-bit 
> addressing is enabled in the current mode.  This would always be true 
> for LA32, and from MISC for the current priv level for LA64.
>
> There are changes to BL, JIRL and PCADD* to sign-extend in 32-bit 
> address mode.
>
> What happened to the PWCH adjustment from v1?
>
>
> r~

On 8/7/23 08:43, Jiajie Chen wrote:
>> There are changes missing for the translator.
>>
>> All of the doubleword (D) instructions must be rejected in LA32 mode.
> I was trying to do this, but I failed to figure out how to read the current cpucfg when 
> translating instructions to TCP ops. This problem applies to the mode-specific behavior 
> below: VA32L1, BL, JIRL, PCADD*, PWCH etc.

Because the CPUCFG values are constant, you may read them in 
loongarch_tr_init_disas_context and save the value in DisasContext.  This is sufficient 
for LA32 vs LA64.

But virtual address width should be in HW_FLAGS and may be read from ctx->base.tb->flags. 
You may wish to simplify usage throughout the translation routines by extracting a 'bool 
va32' in loongarch_tr_init_disas_context.


r~

On 2023/8/7 23:56, Richard Henderson wrote:
> On 8/7/23 08:43, Jiajie Chen wrote:
>>> There are changes missing for the translator.
>>>
>>> All of the doubleword (D) instructions must be rejected in LA32 mode.
>> I was trying to do this, but I failed to figure out how to read the 
>> current cpucfg when translating instructions to TCP ops. This problem 
>> applies to the mode-specific behavior below: VA32L1, BL, JIRL, 
>> PCADD*, PWCH etc.
>
> Because the CPUCFG values are constant, you may read them in 
> loongarch_tr_init_disas_context and save the value in DisasContext.  
> This is sufficient for LA32 vs LA64.
>
> But virtual address width should be in HW_FLAGS and may be read from 
> ctx->base.tb->flags. You may wish to simplify usage throughout the 
> translation routines by extracting a 'bool va32' in 
> loongarch_tr_init_disas_context.
Thank you very much, I will rewrite the code to use this approach.
>
>
> r~