Claude review: accel/neutron: Add driver for NXP Neutron NPU

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

Patch Review

Several issues:

**`of_match_ptr()` usage:**
```c
.of_match_table	= of_match_ptr(neutron_match_table),
```
This will cause a warning about unused `neutron_match_table` when `CONFIG_OF` is disabled, and the driver won't match. Since the driver `depends on ARCH_MXC` which implies OF, it's not a runtime problem, but the kernel convention is to not use `of_match_ptr()` for this reason. Just use the table directly.

**`dma_set_mask_and_coherent` return value not checked:**
```c
dma_set_mask_and_coherent(dev, DMA_BIT_MASK(48));
```
The return value should be checked and propagated.

**Firmware loading security:** The ELF parser in `neutron_load_firmware()` lacks bounds checking:
```c
ehdr = (struct elf32_hdr *)fw->data;
```
There's no check that `fw->size >= sizeof(struct elf32_hdr)`, or that `ehdr->e_phoff + ehdr->e_phnum * sizeof(struct elf32_phdr)` doesn't exceed `fw->size`, or that `seg->p_offset + seg->p_filesz` doesn't exceed `fw->size`. A malformed firmware file could cause out-of-bounds reads. This is a security issue even though firmware loading is privileged.

**`neutron_stop()` ignores `readl_poll_timeout` return value:**
```c
readl_poll_timeout(NEUTRON_REG(ndev, RESETCTRL),
		   resetctrl, !(resetctrl & RESETCTRL_ZVRUN),
		   100, 100 * USEC_PER_MSEC);
```
If stop times out, the driver silently continues. At minimum add a dev_warn.

**IRQ handler:** The threaded IRQ handler unconditionally returns `IRQ_HANDLED` without checking if the interrupt was actually for this device. There's no hardirq handler to check and clear the interrupt source atomically. Consider adding a primary handler that checks the interrupt status register before waking the thread.

---
Generated by Claude Code Patch Reviewer