Claude review: accel/amdxdna: Fix out-of-bounds memset in command slot handling

[Claude Code Review Bot <claude-review@example.com>]

Patch Review

The fix is applied consistently across all four affected functions: `aie2_cmdlist_fill_npu_cf`, `aie2_cmdlist_fill_npu_dpu`, `aie2_cmdlist_fill_npu_preempt`, and `aie2_cmdlist_fill_npu_elf`.

The pattern in each case is the same. Taking `aie2_cmdlist_fill_npu_cf` as representative:

> -	memset(npu_slot, 0, sizeof(*npu_slot));
>  	cmd = amdxdna_cmd_get_payload(cmd_bo, &cmd_len);
>  	if (*size < sizeof(*npu_slot) + cmd_len)
>  		return -EINVAL;
>
> +	memset(npu_slot, 0, sizeof(*npu_slot));
>  	npu_slot->cu_idx = amdxdna_cmd_get_cu_idx(cmd_bo);

The `slot` pointer comes from `cmdbuf_abo->mem.kva + offset` where the available space is `cmdbuf_abo->mem.size - offset`. When multiple commands are chained, `offset` advances with each slot, and the remaining space can become smaller than `sizeof(struct cmd_chain_slot_npu)` (which is 56+ bytes with the packed struct header fields). Writing `sizeof(*npu_slot)` bytes via memset before checking that there is room would corrupt whatever follows the buffer.

Moving the memset after the size validation is the correct fix. After the check passes, `*size >= sizeof(*npu_slot) + payload_size`, so the memset of `sizeof(*npu_slot)` bytes is guaranteed to be in bounds.

One minor observation on `aie2_cmdlist_fill_npu_elf`: unlike the other three functions, this one does not set `npu_slot->cu_idx` and instead immediately sets `npu_slot->type`. The memset still correctly appears after the size check, so the fix is consistent. This just means `cu_idx` will be left at zero from the memset, which appears intentional since ELF-type slots apparently don't need a CU index.

The commit message is accurate, the Fixes tag correctly identifies the commit that introduced the regression, and the change is minimal and correct. No issues found.

---
Generated by Claude Code Patch Reviewer