On Thu, Dec 24, 2020 at 1:16 PM Tian Tao <tiantao6@xxxxxxxxxxxxx> wrote:
>
> add a flag to zpool, named is "can_sleep_mapped", and have it set true
> for zbud/z3fold, set false for zsmalloc. Then zswap could go the current
> path if the flag is true; and if it's false, copy data from src to a
> temporary buffer, then unmap the handle, take the mutex, process the
> buffer instead of src to avoid sleeping function called from atomic
> context.
>
> Signed-off-by: Tian Tao <tiantao6@xxxxxxxxxxxxx>
> ---
> include/linux/zpool.h | 2 ++
> mm/z3fold.c | 1 +
> mm/zbud.c | 1 +
> mm/zpool.c | 13 +++++++++++++
> mm/zsmalloc.c | 1 +
> mm/zswap.c | 23 +++++++++++++++++++++--
> 6 files changed, 39 insertions(+), 2 deletions(-)
>
> diff --git a/include/linux/zpool.h b/include/linux/zpool.h
> index 51bf430..a354f4fe 100644
> --- a/include/linux/zpool.h
> +++ b/include/linux/zpool.h
> @@ -82,6 +82,7 @@ u64 zpool_get_total_size(struct zpool *pool);
> */
> struct zpool_driver {
> char *type;
> + bool sleep_mapped;
Could you please add this somewhat further down, close to map / unmap
callbacks, and add a description of the field above, too?
> struct module *owner;
> atomic_t refcount;
> struct list_head list;
> @@ -112,5 +113,6 @@ void zpool_register_driver(struct zpool_driver *driver);
> int zpool_unregister_driver(struct zpool_driver *driver);
>
> bool zpool_evictable(struct zpool *pool);
> +bool zpool_can_sleep_mapped(struct zpool *pool);
>
> #endif
> diff --git a/mm/z3fold.c b/mm/z3fold.c
> index dacb0d7..234b46f 100644
> --- a/mm/z3fold.c
> +++ b/mm/z3fold.c
> @@ -1778,6 +1778,7 @@ static u64 z3fold_zpool_total_size(void *pool)
>
> static struct zpool_driver z3fold_zpool_driver = {
> .type = "z3fold",
> + .sleep_mapped = true,
> .owner = THIS_MODULE,
> .create = z3fold_zpool_create,
> .destroy = z3fold_zpool_destroy,
> diff --git a/mm/zbud.c b/mm/zbud.c
> index c49966e..7ec5f27 100644
> --- a/mm/zbud.c
> +++ b/mm/zbud.c
> @@ -203,6 +203,7 @@ static u64 zbud_zpool_total_size(void *pool)
>
> static struct zpool_driver zbud_zpool_driver = {
> .type = "zbud",
> + .sleep_mapped = true,
> .owner = THIS_MODULE,
> .create = zbud_zpool_create,
> .destroy = zbud_zpool_destroy,
> diff --git a/mm/zpool.c b/mm/zpool.c
> index 3744a2d..5ed7120 100644
> --- a/mm/zpool.c
> +++ b/mm/zpool.c
> @@ -23,6 +23,7 @@ struct zpool {
> void *pool;
> const struct zpool_ops *ops;
> bool evictable;
> + bool can_sleep_mapped;
>
> struct list_head list;
> };
> @@ -183,6 +184,7 @@ struct zpool *zpool_create_pool(const char *type, const char *name, gfp_t gfp,
> zpool->pool = driver->create(name, gfp, ops, zpool);
> zpool->ops = ops;
> zpool->evictable = driver->shrink && ops && ops->evict;
> + zpool->can_sleep_mapped = driver->sleep_mapped;
>
> if (!zpool->pool) {
> pr_err("couldn't create %s pool\n", type);
> @@ -393,6 +395,17 @@ bool zpool_evictable(struct zpool *zpool)
> return zpool->evictable;
> }
>
> +/**
> + * zpool_can_sleep_mapped - Test if zpool can sleep when do mapped.
> + * @zpool: The zpool to test
> + *
> + * Returns: true if zpool can sleep; false otherwise.
> + */
> +bool zpool_can_sleep_mapped(struct zpool *zpool)
> +{
> + return zpool->can_sleep_mapped;
> +}
> +
> MODULE_LICENSE("GPL");
> MODULE_AUTHOR("Dan Streetman <ddstreet@xxxxxxxx>");
> MODULE_DESCRIPTION("Common API for compressed memory storage");
> diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c
> index 7289f50..a991d3c 100644
> --- a/mm/zsmalloc.c
> +++ b/mm/zsmalloc.c
> @@ -440,6 +440,7 @@ static u64 zs_zpool_total_size(void *pool)
>
> static struct zpool_driver zs_zpool_driver = {
> .type = "zsmalloc",
> + .sleep_mapped = false,
You don't have to add this, it will be false if not explicitly specified.
> .owner = THIS_MODULE,
> .create = zs_zpool_create,
> .destroy = zs_zpool_destroy,
> diff --git a/mm/zswap.c b/mm/zswap.c
> index 182f6ad..51b033a 100644
> --- a/mm/zswap.c
> +++ b/mm/zswap.c
> @@ -1235,7 +1235,7 @@ static int zswap_frontswap_load(unsigned type, pgoff_t offset,
> struct zswap_entry *entry;
> struct scatterlist input, output;
> struct crypto_acomp_ctx *acomp_ctx;
> - u8 *src, *dst;
> + u8 *src, *dst, *tmp;
> unsigned int dlen;
> int ret;
>
> @@ -1256,12 +1256,27 @@ static int zswap_frontswap_load(unsigned type, pgoff_t offset,
> goto freeentry;
> }
>
> + if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
> +
> + tmp = kzalloc(entry->length, GFP_KERNEL);
Do we really need kzalloc here? the buffer will be written to in a few
moments anyway.
Also, we allocate a small buffer for a short while, and this is likely
to be a performance critical path, so GFP_ATOMIC looks more
appropriate.
> + if (!tmp)
> + return -ENOMEM;
You can't just return here, you need to call zswap_entry_put() first.
Since this is obviously not the final version of your patch, could you
please split it in 2 the next time? I'd prefer setting flags to
specific backends in a separate patch.
Best regards,
Vitaly
> + }
> +
> /* decompress */
> dlen = PAGE_SIZE;
> src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
> if (zpool_evictable(entry->pool->zpool))
> src += sizeof(struct zswap_header);
>
> + if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
> +
> + memcpy(tmp, src, entry->length);
> + src = tmp;
> +
> + zpool_unmap_handle(entry->pool->zpool, entry->handle);
> + }
> +
> acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
> mutex_lock(acomp_ctx->mutex);
> sg_init_one(&input, src, entry->length);
> @@ -1271,7 +1286,11 @@ static int zswap_frontswap_load(unsigned type, pgoff_t offset,
> ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
> mutex_unlock(acomp_ctx->mutex);
>
> - zpool_unmap_handle(entry->pool->zpool, entry->handle);
> + if (zpool_can_sleep_mapped(entry->pool->zpool))
> + zpool_unmap_handle(entry->pool->zpool, entry->handle);
> + else
> + kfree(tmp);
> +
> BUG_ON(ret);
>
> freeentry:
> --
> 2.7.4
>
