Re: [PATCH v11 00/12] linux-user: User support for AArch64 BTI

[Richard Henderson]

On 10/20/20 1:09 PM, Peter Maydell wrote:
> On Fri, 16 Oct 2020 at 19:42, Richard Henderson
> <richard.henderson@linaro.org> wrote:
>>
>> The kernel abi for this was merged in v5.8, just as the qemu 5.1
>> merge window was closing, so this slipped to the next dev cycle.
>>
>> Changes from v10:
>>   * Include Phil's plug of interp_name memory leak.
>>   * Convert error reporting to Error api.
>>   * Mirror the kernel's code structure for parsing notes
>>     (though Error means that it's not exactly the same).
>>   * Split aarch64 stuff from basic note parsing patch.
>>
>> Changes from v9:
>>   * Split what is now patch 7 into 3 more (pmm).
>>   * All prerequisites are now upstream.
> 
> Unfortunately the bti-2 test dumps core (x86-64 Linux host):

This is odd.  Works on my laptop,

  TEST    pauth-1 on aarch64
  TEST    pauth-2 on aarch64
  TEST    pauth-4 on aarch64
  TEST    pauth-5 on aarch64
  TEST    bti-1 on aarch64
  TEST    bti-2 on aarch64
  TEST    semihosting on aarch64

However, this crashes on an aarch64 host.  Looking at the trace, the compiler
generated bad code -- the difference of two symbols resulted in 0.  So we never
copied the test code to the mmaped page.

I assume your x86_64 build is using the docker images for cross-compile?  I
wonder if my laptop is picking up a local cross-compiler instead.  And then the
aarch64 host compiler has the same problem as the docker version?


r~