Re: [PATCH v2 11/12] xenctx: Dump registers via hvm info if available.

[Don Slutz <dslutz@verizon.com>]

On 11/07/13 03:38, Jan Beulich wrote:
>>>> On 06.11.13 at 21:08, Don Slutz <dslutz@verizon.com> wrote:
>> @@ -931,7 +1174,44 @@ static void dump_ctx(int vcpu, guest_word_t mem_addr, guest_word_t stk_addr)
>>   #endif
>>       } else {
>>           if (!stk_addr) {
>> -            print_ctx(&ctx);
>> +#if defined(__i386__) || defined(__x86_64__)
>> +            if (xenctx.dominfo.hvm && ctxt_word_size == 8) {
>> +                if (guest_word_size == 4) {
>> +                    if ((((uint32_t)ctx.x64.user_regs.eip) != cpuctx.rip) ||
>> +                        (((uint32_t)ctx.x64.user_regs.esp) != cpuctx.rsp) ||
>> +                        (((uint32_t)ctx.x64.ctrlreg[3]) != cpuctx.cr3)) {
>> +                        fprintf(stderr, "Regs mismatch ip=%llx vs %llx sp=%llx vs %llx cr3=%llx vs %llx\n",
>> +                                (long long)((uint32_t)ctx.x64.user_regs.eip),
>> +                                (long long)cpuctx.rip,
>> +                                (long long)((uint32_t)ctx.x64.user_regs.esp),
>> +                                (long long)cpuctx.rsp,
>> +                                (long long)((uint32_t)ctx.x64.ctrlreg[3]),
>> +                                (long long)cpuctx.cr3);
>> +                        fprintf(stdout, "=============Regs mismatch start=============\n");
>> +                        print_ctx(&ctx);
>> +                        fprintf(stdout, "=============Regs mismatch end=============\n");
>> +                    }
>> +                } else {
>> +                    if ((ctx.x64.user_regs.eip != cpuctx.rip) ||
>> +                        (ctx.x64.user_regs.esp != cpuctx.rsp) ||
>> +                        (ctx.x64.ctrlreg[3] != cpuctx.cr3)) {
>> +                        fprintf(stderr, "Regs mismatch ip=%llx vs %llx sp=%llx vs %llx cr3=%llx vs %llx\n",
>> +                                (long long)ctx.x64.user_regs.eip,
>> +                                (long long)cpuctx.rip,
>> +                                (long long)ctx.x64.user_regs.esp,
>> +                                (long long)cpuctx.rsp,
>> +                                (long long)ctx.x64.ctrlreg[3],
>> +                                (long long)cpuctx.cr3);
>> +                        fprintf(stdout, "=============Regs mismatch start=============\n");
>> +                        print_ctx(&ctx);
>> +                        fprintf(stdout, "=============Regs mismatch end=============\n");
>> +                    }
>> +                }
>> +                print_cpuctx(&cpuctx);
>> +            }
>> +            else
>> +#endif
>> +                print_ctx(&ctx);
> Apart from Andrew's comments, which I agree with - most of the
> additions above clearly don't belong here: This is not a diagnostic
> utility.
Fine with me, I will drop this part. I added during the time (~2010-2011) when I was looking at a DomU that was crashed.  What I remember about this DomU was that vCPU 1 was offline and vCPU 0 was the crash cause and vCPU 2 & 3 where in the code to go offline. The routine xc_domain_hvm_getcontext_partial() was returning vCPU 2's data when asked for vCPU 1 (via instance == 1).  This is the key reason I did the code in patch #12 (basically a way to call xc_domain_hvm_getcontext_partial() and xc_vcpu_getcontext() with different args).  At the time I was not working on full xen code, just changing xenctx the help me out as needed.  Looking at the code now, I do not understand how this could happen.
     -Don Slutz
> Jan
>