[Lldb-commits] [PATCH] D16916: [LLDB][MIPS] Fix TestExpressionInSyscall.py for MIPS

[Zachary Turner via lldb-commits]

On Thu, Feb 4, 2016 at 11:32 PM Bhushan Attarde via lldb-commits <
lldb-commits at lists.llvm.org> wrote:

> bhushan created this revision.
> bhushan added a reviewer: clayborg.
> bhushan added subscribers: lldb-commits, nitesh.jain, mohit.bhakkad,
> sagar, jaydeep.
> bhushan set the repository for this revision to rL LLVM.
>
> This test (TestExpressionInSyscall.py) checks if we are able to evaluate
> expressions when the inferior is blocked in a syscall.
>
> As a part of expression evaluation LLDB checks for memory allocation on
> target (by executing mmap).
> So we setup call to mmap by setting argument registers and PC.
> Now the process is stopped in the syscall and when it continue to allocate
> memory, the system call is restarted.
>
> In MIPS, to restart a syscall, kernel decreases the PC by 4 so the
> resulting PC now points to mmap-4
> and also register R7 that provides 'flags' argument to mmap gets clobbered
> to 0 and hence mmap fails.
>
> A fix to this issue is to postpone the syscall restart until the
> expression is evaluated.
> In MIPS, register R0 controls syscall restart. This patch writes 0 into
> register R0 when preparing call to mmap.
> This setting avoids a syscall restart and prevents automatic decrement of
> the PC so that expression can be evaluated correctly.
>
> Once the expression completes the registers are restored and program
> resumes the interrupted syscall when the continue command is issued.
>
> This fixes TestExpressionInSyscall.py and solves bug 23659 for MIPS.
>
> Repository:
>   rL LLVM
>
> http://reviews.llvm.org/D16916
>
> Files:
>   source/Plugins/ABI/SysV-mips/ABISysV_mips.cpp
>   source/Plugins/ABI/SysV-mips64/ABISysV_mips64.cpp
>
> Index: source/Plugins/ABI/SysV-mips64/ABISysV_mips64.cpp
> ===================================================================
> --- source/Plugins/ABI/SysV-mips64/ABISysV_mips64.cpp
> +++ source/Plugins/ABI/SysV-mips64/ABISysV_mips64.cpp
> @@ -207,6 +207,17 @@
>      const RegisterInfo *sp_reg_info = reg_ctx->GetRegisterInfo
> (eRegisterKindGeneric, LLDB_REGNUM_GENERIC_SP);
>      const RegisterInfo *ra_reg_info = reg_ctx->GetRegisterInfo
> (eRegisterKindGeneric, LLDB_REGNUM_GENERIC_RA);
>      const RegisterInfo *r25_info = reg_ctx->GetRegisterInfoByName("r25",
> 0);
> +    const RegisterInfo *r0_info = reg_ctx->GetRegisterInfoByName("zero",
> 0);
> +
> +    if (log)
> +    log->Printf("Writing R0: 0x%" PRIx64, (uint64_t)0);
>
Your indentation is wrong here.  Please use clang-format on this patch.


> +
> +    /* Write r0 with 0, in case we are stopped in syscall,
> +     * such setting prevents automatic decrement of the PC.
> +     * This clears the bug 23659 for MIPS.
> +    */
> +    if (!reg_ctx->WriteRegisterFromUnsigned (r0_info, (uint64_t)0))
> +        return false;
>
>      if (log)
>      log->Printf("Writing SP: 0x%" PRIx64, (uint64_t)sp);
> Index: source/Plugins/ABI/SysV-mips/ABISysV_mips.cpp
> ===================================================================
> --- source/Plugins/ABI/SysV-mips/ABISysV_mips.cpp
> +++ source/Plugins/ABI/SysV-mips/ABISysV_mips.cpp
> @@ -242,6 +242,17 @@
>      const RegisterInfo *sp_reg_info = reg_ctx->GetRegisterInfo
> (eRegisterKindGeneric, LLDB_REGNUM_GENERIC_SP);
>      const RegisterInfo *ra_reg_info = reg_ctx->GetRegisterInfo
> (eRegisterKindGeneric, LLDB_REGNUM_GENERIC_RA);
>      const RegisterInfo *r25_info = reg_ctx->GetRegisterInfoByName("r25",
> 0);
> +    const RegisterInfo *r0_info = reg_ctx->GetRegisterInfoByName("zero",
> 0);
> +
> +    if (log)
> +    log->Printf("Writing R0: 0x%" PRIx64, (uint64_t)0);
>
Also wrong here.


> +
> +    /* Write r0 with 0, in case we are stopped in syscall,
> +     * such setting prevents automatic decrement of the PC.
> +     * This clears the bug 23659 for MIPS.
> +    */
> +    if (!reg_ctx->WriteRegisterFromUnsigned (r0_info, (uint64_t)0))
> +        return false;
>
>      if (log)
>      log->Printf("Writing SP: 0x%" PRIx64, (uint64_t)sp);
>
>
> _______________________________________________
> lldb-commits mailing list
> lldb-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-commits/attachments/20160205/5ca73c6e/attachment.html>