Re: [PATCH 2/2] drm/xe: Release runtime pm for error path of xe_devcoredump_read()

[Matthew Brost <matthew.brost@intel.com>]

On Thu, Jul 03, 2025 at 06:20:27AM +0000, Shuicheng Lin wrote:
> xe_pm_runtime_put() is missed to be called for the error path in
> xe_devcoredump_read().
> Add function description comments for xe_devcoredump_read() to help
> understand it.
> 
> Fixes: c4a2e5f865b7 ("drm/xe: Add devcoredump chunking")
> Cc: Matthew Brost <matthew.brost@intel.com>
> Signed-off-by: Shuicheng Lin <shuicheng.lin@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_devcoredump.c | 32 +++++++++++++++++++++++------
>  1 file changed, 26 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_devcoredump.c b/drivers/gpu/drm/xe/xe_devcoredump.c
> index 94625010abc4..701ffe6c8264 100644
> --- a/drivers/gpu/drm/xe/xe_devcoredump.c
> +++ b/drivers/gpu/drm/xe/xe_devcoredump.c
> @@ -171,14 +171,29 @@ static void xe_devcoredump_snapshot_free(struct xe_devcoredump_snapshot *ss)
>  
>  #define XE_DEVCOREDUMP_CHUNK_MAX	(SZ_512M + SZ_1G)
>  
> +/**
> + * xe_devcoredump_read - Read data from the Xe device coredump snapshot

The preferred style is:

s/xe_devcoredump_read/xe_devcoredump_read()


> + * @buffer: Destination buffer to copy the coredump data into
> + * @offset: Offset in the coredump data to start reading from
> + * @count: Number of bytes to read
> + * @data: Pointer to the xe_devcoredump structure
> + * @datalen: Length of the data (unused)
> + *
> + * Reads a chunk of the coredump snapshot data into the provided buffer,
> + * handling chunked reads for large coredumps and ensuring proper locking.

A little a more descriptive, would be:

Reads a chunk of the coredump snapshot data into the provided buffer. If the
devcoredump is smaller than 1.5 GB, it is read directly from a pre-written
buffer. For larger devcoredumps, the pre-written buffer must be periodically
repopulated from the snapshot state due to kmalloc size limitations.

> + *
> + * Return: Number of bytes copied on success, or a negative error code on failure.
> + */
>  static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,
>  				   size_t count, void *data, size_t datalen)
>  {
>  	struct xe_devcoredump *coredump = data;
>  	struct xe_devcoredump_snapshot *ss;
> -	ssize_t byte_copied;
> +	ssize_t byte_copied = 0;
>  	u32 chunk_offset;
>  	ssize_t new_chunk_position;
> +	bool pm_acquired = false;
> +	int ret = 0;
>  
>  	if (!coredump)
>  		return -ENODEV;
> @@ -188,19 +203,23 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,
>  	/* Ensure delayed work is captured before continuing */
>  	flush_work(&ss->work);
>  
> -	if (ss->read.size > XE_DEVCOREDUMP_CHUNK_MAX)
> +	if (ss->read.size > XE_DEVCOREDUMP_CHUNK_MAX) {
>  		xe_pm_runtime_get(gt_to_xe(ss->gt));
> +		pm_acquired = true;
> +	}

I'd write this like:

pm_acquired = ss->read.size > XE_DEVCOREDUMP_CHUNK_MAX;
if (pm_acquired)
	xe_pm_runtime_get(gt_to_xe(ss->gt));

>  
>  	mutex_lock(&coredump->lock);
>  
>  	if (!ss->read.buffer) {
>  		mutex_unlock(&coredump->lock);
> -		return -ENODEV;
> +		ret = -ENODEV;
> +		goto pm_put;
>  	}
>  
>  	if (offset >= ss->read.size) {
>  		mutex_unlock(&coredump->lock);
> -		return 0;
> +		ret = 0;

You don't need to set ret = 0 here as it is initialized to zero above.

Nits aside, patch functionally looks correct.

Matt

> +		goto pm_put;
>  	}
>  
>  	new_chunk_position = div_u64_rem(offset,
> @@ -223,10 +242,11 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,
>  
>  	mutex_unlock(&coredump->lock);
>  
> -	if (ss->read.size > XE_DEVCOREDUMP_CHUNK_MAX)
> +pm_put:
> +	if (pm_acquired)
>  		xe_pm_runtime_put(gt_to_xe(ss->gt));
>  
> -	return byte_copied;
> +	return byte_copied ? byte_copied : ret;
>  }
>  
>  static void xe_devcoredump_free(void *data)
> -- 
> 2.49.0
>