Re: [PATCH 1/2] mtd: rawnand: nandsim: use a NOFS scope for cache file I/O

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] mtd: nandsim: IO reclaim self-deadlock due to memalloc_nofs_save()
--

commit 6c4530163aa059b5f98e17ba82a9c2b06c3f48bb
Author: Tal Zussman <tz2294@columbia.edu>

mtd: rawnand: nandsim: use a NOFS scope for cache file I/O

This commit replaces memalloc_noreclaim_save() with memalloc_nofs_save() to
prevent allocations during nandsim cache file operations from recursing into
filesystem reclaim.

> diff --git a/drivers/mtd/nand/raw/nandsim.c b/drivers/mtd/nand/raw/nandsim.c
> index fe968037f75a2..4559df2cfa3cc 100644
> --- a/drivers/mtd/nand/raw/nandsim.c
> +++ b/drivers/mtd/nand/raw/nandsim.c
[ ... ]
> @@ -1342,14 +1342,14 @@ static ssize_t ns_read_file(struct nandsim *ns, struct file *file, void *buf,
>  {
>  	ssize_t tx;
>  	int err;
> -	unsigned int noreclaim_flag;
> +	unsigned int nofs_flag;
>  
>  	err = ns_get_pages(ns, file, count, pos);
>  	if (err)
>  		return err;
> -	noreclaim_flag = memalloc_noreclaim_save();
> +	nofs_flag = memalloc_nofs_save();
>  	tx = kernel_read(file, buf, count, &pos);
> -	memalloc_noreclaim_restore(noreclaim_flag);

[Severity: High]
Can this cause an IO reclaim self-deadlock?

Nandsim executes MTD operations while holding the NAND subsystem device
lock (chip->lock). While memalloc_nofs_save() strips the filesystem reclaim
flags, it leaves direct IO reclaim enabled.

If memory allocations within kernel_read() trigger direct reclaim, the
shrinker might attempt to write out dirty pages belonging to this same
device (for example, if used as a swap device or via mtdblock).

Writing out these pages would route through the block layer back to the MTD
layer, which will attempt to acquire chip->lock again, resulting in a
self-deadlock.

Should this use memalloc_noio_save() instead to safely prevent reclaim
from issuing any block IO while the MTD lock is held?

>  	ns_put_pages(ns);
>  	return tx;
>  }
> @@ -1359,14 +1359,14 @@ static ssize_t ns_write_file(struct nandsim *ns, struct file *file, void *buf,
>  {
>  	ssize_t tx;
>  	int err;
> -	unsigned int noreclaim_flag;
> +	unsigned int nofs_flag;
>  
>  	err = ns_get_pages(ns, file, count, pos);
>  	if (err)
>  		return err;
> -	noreclaim_flag = memalloc_noreclaim_save();
> +	nofs_flag = memalloc_nofs_save();
>  	tx = kernel_write(file, buf, count, &pos);
> -	memalloc_noreclaim_restore(noreclaim_flag);

[Severity: High]
Does kernel_write() face the same IO reclaim self-deadlock risk here?

Allocations within kernel_write() could also trigger direct IO reclaim
while the chip lock is held. Using memalloc_noio_save() would protect this
path as well.

>  	ns_put_pages(ns);
>  	return tx;
>  }

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261014-nandsim-nofs-v1-0-47e49f37e387@columbia.edu?part=1

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/