So, I think the problem here may be that the game tries to free old (already freed) pointers in a way that works on the PSP, but not in PPSSPP. The reason is we currently have different allocation semantics.

If it affects this game (assuming I'm right), it can affect others...

The PSP maintains a freelist and uses 8-byte blocks for allocation. Here's how it basically looks:

struct VplBlock {
    VplBlock *next;
    u32 sizeDiv8;
};

struct VplHeader {
    void *start;
    void *startAgain;
    void *startPlusSeven;
    u32 totalSizeMinus8; // or offset of last block?
    u32 allocatedDiv8;
    VplBlock *nextFreeBlock;
    VplBlock bottomBlock;
};

The VplHeader is found at the start of the vpl (the first 0x20 bytes which are not allocatable.) Going from the bottom upward, it's easy enough to list all blocks using sizeDiv8. The nextFreeBlock is usually at the top and goes downward using a simple linked list of only free blocks. next on a block that has been allocated points to startPlusSeven - maybe it's +7 to cause an alignment fault? Note also that the "size" (totalSizeMinus8) is not divided by 8.

Currently, we always allocate from the top down. This means we more eagerly reuse blocks than a freelist algorithm would. This may fragment memory better (providing more contiguous free blocks), depending on usage, but is also slower.

So I guess we could try changing the allocation strategy.

I've also tried to implement SMALLEST, and found bugs in our current allocation prioritization (namely, the most common method, FIFO, was treated as SMALLEST.) But further looks at logs to this game revealed that it only allocates from one thread, so it probably doesn't matter.

-[Unknown]

N.O.V.A. Mini causing crash in intro #3485

Description

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions