RFR (S) 8217879: hs_err should print more instructions in hex dump

[Stefan Karlsson]

While looking at this I also considered if this:

  delta = max_delta;
  address low = pc - delta;
  while (delta > 0 && pc > low && !is_readable_pointer(low)) {
    delta >>= 1;
    low = pc - delta;
  }

  delta = max_delta;
  address high = pc + delta;
  while (delta > 0 && pc < high && !is_readable_pointer(high)) {
    delta >>= 1;
    high = pc + delta;
  }

could be changed to:

  delta = max_delta;
while (delta > 0 && !is_readable_pointer(pc - delta)) {
    delta >>= 1;
  }
  address low = pc - delta;

  delta = max_delta;
while (delta > 0 && !is_readable_pointer(pc + delta)) {
    delta >>= 1;
  }
address high = pc + delta;

to simplify the loops.

I realize that this removes the underflow/overflow checks, but I think 
the compiler might actually reduce them to true, because of the 
undefined behavior of underflowing/overflowing:

  address low = pc - delta;
while (delta > 0 && pc > low ...) {

Since the compiler can assume that 'pc - delta' doesn't underflow it can 
transform 'pc > low' to 'pc > pc - delta' and then to '0 > 0 - delta', 
which is always true when delta is positive.

StefanK

On 2019-01-29 12:36, Aleksey Shipilev wrote:
> On 1/29/19 12:21 PM, Stefan Karlsson wrote:
>> For this to work as you intend it, max_delta must not be set to page size. If max_delta is set to
>> page size, the following would end up with low in Page 0 and high in Page 2:
>>
>> +--------+
>> | Page 0 |   readable
>> +--------+ < 'pc' beginning of Page 1
>> | Page 1|   unreadable
>> +--------+
>> | Page 2 |   readable
>> +--------+
>>
>> Therefore it seems a bit off to cap with the page size in:
>> MIN2(256, min_page_size());
>>
>> Given that you use 256 today, this isn't really a problem, but maybe future proof this a bit and use
>> min_page_size / 2 to guarantee that either low or high ends up in Page 1?
> Right, thanks. To handle that corner case, let's do /2 indeed, updated webrev in-place.
>
> -Aleksey
>