Fix for #2204

[furbrain]

This PR provides a fix for issue #2204. It fixes a moderate bug which results in hard segfaults when you repeatedly initialise a display. The original context for this is a display which is powered down prior to the processor going in to light sleep

[tannewt]

Please explain what the bug is. It isn't immediately obvious to me.

[furbrain]

I will use line numbers from the original source code
gc_never_free will crash the fourth time it is called.

• First time it is called the permanent_pointers item of mp_state is set to null.
  The while block on line 1017 is skipped and next_block is allocated - a set of four pointers
  The if statement on line 1029 evaluates as true and line 1030 executes, correctly storing the address of next_block in permanent_pointers
  the pointer passed into the function is stored in index [1] of current_reference_block
• second and third time around it loads the previously allocated block from permanent_pointers into current_reference_block correctly just places the pointer into current_reference_block[2] and [3]
• Fourth time around current_reference_block looks like this:
  [0] = NULL
  [1] = ptr1 (first pointer added)
  [2] = ptr2
  [3] = ptr 3
  As there is nowhere to store the new pointer current_reference_block is set to current_reference_block[0] - which is NULL - indicating that we need to create a new block
  However, on line 1032 we set current_reference_block[0] to the new block.
  But at this point current_reference_block is NULL
  So line 1032 tries to set the memory at 0x00000000 to the address of next_block - which segfaults

[furbrain]

My fix keeps a record of the previous current_record_block so we can assign the next block into it correctly.

[akoebbe]

@furbrain Thanks for this change. Just wanted to confirm here that this change resolves #7974 . The only possible side effect is that auto-reload crashes, though pressing reset does bring it back. Again, I can't confirm that this fix has anything to do with that, but just something to keep an eye out for.

[tannewt]

Thanks for the fix!

[tannewt]

(and the thorough explanation.)

[dhalbert]

@furbrain I am closing all the issues you think are related. Let us know if you think not.

[dhalbert]

@furbrain - I marked them as "needs re-test" instead.