|WARNING| dma/cnxk: fix crash on secondary process cleanup

[dpdklab at iol.unh.edu]

Test-Label: ai-code-review
Test-Status: WARNING
https://dpdk.org/patch/165025

_AI Code Review_

=== Patch Review: series.patch (via Claude) ===
Review date: 2026-06-05

# DPDK Patch Review: dma/cnxk: fix crash on secondary process cleanup

## Summary
This patch fixes a crash in secondary processes by preventing hardware re-initialization that overwrites shared device state with process-local pointers.

---

## Errors

### 1. Secondary process attach incomplete
**Location**: After the early return at line 620

**Issue**: The patch adds an early return for secondary processes without attaching to the shared device data. The commit message states "attach to the shared device data and return", but no attachment actually occurs.

**Why it matters**: Secondary processes need to attach to the device data structure allocated by the primary process to function correctly. Without attachment, the secondary process cannot use the DMA device.

**Suggested fix**:
```c
if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
	/* Attach to existing shared device data */
	dmadev->data = rte_dma_pmd_get_named_dev_data(pci_dev->device.name);
	if (dmadev->data == NULL)
		return -ENODEV;
	return 0;
}
```
Or verify that `dmadev->fp_obj->dev_private` assignment (line 617) already provides the necessary attachment, and update the commit message to match the actual behavior.

---

## Warnings

### 1. Missing NULL check on dmadev->fp_obj
**Location**: Line 617

**Issue**: `dmadev->fp_obj` is dereferenced without a NULL check. While this may be guaranteed by the framework, defensive programming would verify it.

**Suggested fix**:
```c
if (dmadev->fp_obj == NULL)
	return -ENOMEM;
dmadev->fp_obj->dev_private = dpivf;
```

### 2. Commit message accuracy
**Location**: Commit message

**Issue**: The commit message claims the fix "attach[es] to the shared device data and return[s]", but the code only returns early without any explicit attachment logic visible in the patch context.

**Suggested fix**: Either:
- Add the attachment code as shown in Error #1, or
- Revise the commit message to accurately describe what the code does (e.g., "Skip hardware initialization in secondary processes, relying on device data already set by framework")

---

## Info

### 1. Code placement
The early return is placed after `dmadev->dev_ops` assignment. Consider whether `dev_ops` should also be skipped for secondary processes or if it needs to be set for them to issue device operations. The current placement suggests the ops table is needed, which may be correct.

### 2. Verification needed
The fix assumes that:
1. The framework has already allocated and attached `dmadev->fp_obj->dev_private` before `cnxk_dmadev_probe()` is called in secondary processes
2. `rdpi` fields below the early return are only needed in the primary process

Both assumptions should be verified against the DPDK DMA framework behavior.

---

## Positive Notes

- The fix correctly identifies the root cause: secondary processes should not perform hardware initialization
- The early return placement prevents overwriting `rdpi->pci_dev` 
- The process type check uses the correct DPDK API (`rte_eal_process_type()`)
- Includes appropriate `Fixes:` tag and `Cc: stable at dpdk.org`

---

## Recommendation

Address Error #1 by either adding explicit device data attachment or clarifying in the commit message that attachment is handled by the framework. The current mismatch between commit message and code will confuse future maintainers.