On Sun, Feb 28, 2016 at 10:56 AM, Alexander Duyck
<alexander.du...@gmail.com> wrote:
> On Sat, Feb 27, 2016 at 12:30 AM, Alexander Duyck
> <alexander.du...@gmail.com> wrote:
>>> +{
>>> + asm("lea 40f(, %[slen], 4), %%r11\n\t"
>>> + "clc\n\t"
>>> + "jmpq *%%r11\n\t"
>>> + "adcq 7*8(%[src]),%[res]\n\t"
>>> + "adcq 6*8(%[src]),%[res]\n\t"
>>> + "adcq 5*8(%[src]),%[res]\n\t"
>>> + "adcq 4*8(%[src]),%[res]\n\t"
>>> + "adcq 3*8(%[src]),%[res]\n\t"
>>> + "adcq 2*8(%[src]),%[res]\n\t"
>>> + "adcq 1*8(%[src]),%[res]\n\t"
>>> + "adcq 0*8(%[src]),%[res]\n\t"
>>> + "nop\n\t"
>>> + "40: adcq $0,%[res]"
>>> + : [res] "=r" (sum)
>>> + : [src] "r" (buff),
>>> + [slen] "r" (-((unsigned long)(len >> 3))), "[res]"
>>> (sum)
>>> + : "r11");
>>> +
>>
>> With this patch I cannot mix/match different length checksums without
>> things failing. In perf the jmpq in the loop above seems to be set to
>> a fixed value so perhaps it is something in how the compiler is
>> interpreting the inline assembler.
>
> The perf thing was a red herring. Turns out the code is working
> correctly there.
>
> I actually found the root cause. The problem is in add32_with_carry3.
>
Thanks for the follow-up. btw are you trying to build csum_partial in
userspace for testing, or was this all in kernel?
>> +static inline unsigned int add32_with_carry3(unsigned int a, unsigned int b,
>> + unsigned int c)
>> +{
>> + asm("addl %2,%0\n\t"
>> + "adcl %3,%0\n\t"
>> + "adcl $0,%0"
>> + : "=r" (a)
>> + : "" (a), "rm" (b), "rm" (c));
>> +
>> + return a;
>> +}
>> +
>
> You need to set the 'a' input variable attribute to "0" instead of ""
> and then things work for me correctly.
>
> - Alex
