On Mon, Dec 18, 2023 at 12:22 AM Chengming Zhou
<[email protected]> wrote:
>
> Since the introduce of reusing the dstmem in the load path, it seems
> confusing that we are now using acomp_ctx->dstmem and acomp_ctx->mutex
> now for purposes other than what the naming suggests.
>
> Yosry suggested removing these two fields from acomp_ctx, and directly
> using zswap_dstmem and zswap_mutex in both the load and store paths,
> rename them, and add proper comments above their definitions that they
> are for generic percpu buffering on the load and store paths.
>
> So this patch remove dstmem and mutex from acomp_ctx, and rename the
> zswap_dstmem to zswap_buffer, using the percpu mutex and buffer on
> the load and store paths. And refactor out __zswap_store() to only
> include the compress & store, since I found zswap_store() is too long.
I am not sure refactoring out __zswap_store() is useful to be honest,
but I am not objecting to it, it mirrors __zswap_load() in a sense.
However, if you want to do so, please do it in a separate patch from
renaming the percpu buffers and mutex. This will make reviewing easier
(and make my Suggested-by correctly scoped).
Also, any reason why raw_smp_processor_id() is used here instead of
smp_processor_id()?
>
> Suggested-by: Yosry Ahmed <[email protected]>
> Signed-off-by: Chengming Zhou <[email protected]>
> ---
> mm/zswap.c | 193 +++++++++++++++++++++++++++++++++----------------------------
> 1 file changed, 104 insertions(+), 89 deletions(-)
>
> diff --git a/mm/zswap.c b/mm/zswap.c
> index 2c349fd88904..b7449294ec3a 100644
> --- a/mm/zswap.c
> +++ b/mm/zswap.c
> @@ -166,8 +166,6 @@ struct crypto_acomp_ctx {
> struct crypto_acomp *acomp;
> struct acomp_req *req;
> struct crypto_wait wait;
> - u8 *dstmem;
> - struct mutex *mutex;
> };
>
> /*
> @@ -694,7 +692,7 @@ static void zswap_alloc_shrinker(struct zswap_pool *pool)
> /*********************************
> * per-cpu code
> **********************************/
> -static DEFINE_PER_CPU(u8 *, zswap_dstmem);
> +static DEFINE_PER_CPU(u8 *, zswap_buffer);
> /*
> * If users dynamically change the zpool type and compressor at runtime, i.e.
> * zswap is running, zswap can have more than one zpool on one cpu, but they
> @@ -702,39 +700,39 @@ static DEFINE_PER_CPU(u8 *, zswap_dstmem);
> */
> static DEFINE_PER_CPU(struct mutex *, zswap_mutex);
>
> -static int zswap_dstmem_prepare(unsigned int cpu)
> +static int zswap_buffer_prepare(unsigned int cpu)
> {
> struct mutex *mutex;
> - u8 *dst;
> + u8 *buf;
>
> - dst = kmalloc_node(PAGE_SIZE, GFP_KERNEL, cpu_to_node(cpu));
> - if (!dst)
> + buf = kmalloc_node(PAGE_SIZE, GFP_KERNEL, cpu_to_node(cpu));
> + if (!buf)
> return -ENOMEM;
>
> mutex = kmalloc_node(sizeof(*mutex), GFP_KERNEL, cpu_to_node(cpu));
> if (!mutex) {
> - kfree(dst);
> + kfree(buf);
> return -ENOMEM;
> }
>
> mutex_init(mutex);
> - per_cpu(zswap_dstmem, cpu) = dst;
> + per_cpu(zswap_buffer, cpu) = buf;
> per_cpu(zswap_mutex, cpu) = mutex;
> return 0;
> }
>
> -static int zswap_dstmem_dead(unsigned int cpu)
> +static int zswap_buffer_dead(unsigned int cpu)
> {
> struct mutex *mutex;
> - u8 *dst;
> + u8 *buf;
>
> mutex = per_cpu(zswap_mutex, cpu);
> kfree(mutex);
> per_cpu(zswap_mutex, cpu) = NULL;
>
> - dst = per_cpu(zswap_dstmem, cpu);
> - kfree(dst);
> - per_cpu(zswap_dstmem, cpu) = NULL;
> + buf = per_cpu(zswap_buffer, cpu);
> + kfree(buf);
> + per_cpu(zswap_buffer, cpu) = NULL;
>
> return 0;
> }
> @@ -772,9 +770,6 @@ static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
> acomp_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
> crypto_req_done, &acomp_ctx->wait);
>
> - acomp_ctx->mutex = per_cpu(zswap_mutex, cpu);
> - acomp_ctx->dstmem = per_cpu(zswap_dstmem, cpu);
> -
> return 0;
> }
>
> @@ -1392,20 +1387,98 @@ static int zswap_enabled_param_set(const char *val,
> return ret;
> }
>
> +static int __zswap_store(struct zswap_entry *entry, struct page *page)
> +{
> + struct scatterlist input, output;
> + struct crypto_acomp_ctx *acomp_ctx;
> + struct zpool *zpool;
> + unsigned long handle;
> + unsigned int dlen;
> + u8 *buf, *dst;
> + gfp_t gfp;
> + int ret;
> + int cpu;
> + struct mutex *mutex;
> +
> + cpu = raw_smp_processor_id();
> + mutex = per_cpu(zswap_mutex, cpu);
> + mutex_lock(mutex);
> +
> + acomp_ctx = per_cpu_ptr(entry->pool->acomp_ctx, cpu);
> + buf = per_cpu(zswap_buffer, cpu);
> +
> + sg_init_table(&input, 1);
> + sg_set_page(&input, page, PAGE_SIZE, 0);
> + sg_init_one(&output, buf, PAGE_SIZE);
> + acomp_request_set_params(acomp_ctx->req, &input, &output, PAGE_SIZE, PAGE_SIZE);
> + /*
> + * it maybe looks a little bit silly that we send an asynchronous request,
> + * then wait for its completion synchronously. This makes the process look
> + * synchronous in fact.
> + * Theoretically, acomp supports users send multiple acomp requests in one
> + * acomp instance, then get those requests done simultaneously. but in this
> + * case, zswap actually does store and load page by page, there is no
> + * existing method to send the second page before the first page is done
> + * in one thread doing zwap.
> + * but in different threads running on different cpu, we have different
> + * acomp instance, so multiple threads can do (de)compression in parallel.
> + */
> + ret = crypto_wait_req(crypto_acomp_compress(acomp_ctx->req), &acomp_ctx->wait);
> + dlen = acomp_ctx->req->dlen;
> +
> + if (ret) {
> + zswap_reject_compress_fail++;
> + goto unlock;
> + }
> +
> + /* store */
> + zpool = zswap_find_zpool(entry);
> + gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
> + if (zpool_malloc_support_movable(zpool))
> + gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
> + ret = zpool_malloc(zpool, dlen, gfp, &handle);
> + if (ret == -ENOSPC) {
> + zswap_reject_compress_poor++;
> + goto unlock;
> + }
> + if (ret) {
> + zswap_reject_alloc_fail++;
> + goto unlock;
> + }
> + dst = zpool_map_handle(zpool, handle, ZPOOL_MM_WO);
> + memcpy(dst, buf, dlen);
> + zpool_unmap_handle(zpool, handle);
> + mutex_unlock(mutex);
> +
> + entry->handle = handle;
> + entry->length = dlen;
> + return 0;
> +
> +unlock:
> + mutex_unlock(mutex);
> + return ret;
> +}
> +
> static void __zswap_load(struct zswap_entry *entry, struct page *page)
> {
> struct zpool *zpool = zswap_find_zpool(entry);
> struct scatterlist input, output;
> struct crypto_acomp_ctx *acomp_ctx;
> - u8 *src;
> + u8 *src, *buf;
> + int cpu;
> + struct mutex *mutex;
> +
> + cpu = raw_smp_processor_id();
> + mutex = per_cpu(zswap_mutex, cpu);
> + mutex_lock(mutex);
>
> - acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
> - mutex_lock(acomp_ctx->mutex);
> + acomp_ctx = per_cpu_ptr(entry->pool->acomp_ctx, cpu);
>
> src = zpool_map_handle(zpool, entry->handle, ZPOOL_MM_RO);
> if (!zpool_can_sleep_mapped(zpool)) {
> - memcpy(acomp_ctx->dstmem, src, entry->length);
> - src = acomp_ctx->dstmem;
> + buf = per_cpu(zswap_buffer, cpu);
> + memcpy(buf, src, entry->length);
> + src = buf;
> zpool_unmap_handle(zpool, entry->handle);
> }
>
> @@ -1415,7 +1488,7 @@ static void __zswap_load(struct zswap_entry *entry, struct page *page)
> acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, PAGE_SIZE);
> BUG_ON(crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait));
> BUG_ON(acomp_ctx->req->dlen != PAGE_SIZE);
> - mutex_unlock(acomp_ctx->mutex);
> + mutex_unlock(mutex);
>
> if (zpool_can_sleep_mapped(zpool))
> zpool_unmap_handle(zpool, entry->handle);
> @@ -1539,18 +1612,11 @@ bool zswap_store(struct folio *folio)
> struct page *page = &folio->page;
> struct zswap_tree *tree = zswap_trees[type];
> struct zswap_entry *entry, *dupentry;
> - struct scatterlist input, output;
> - struct crypto_acomp_ctx *acomp_ctx;
> struct obj_cgroup *objcg = NULL;
> struct mem_cgroup *memcg = NULL;
> struct zswap_pool *pool;
> - struct zpool *zpool;
> - unsigned int dlen = PAGE_SIZE;
> - unsigned long handle, value;
> - char *buf;
> - u8 *src, *dst;
> - gfp_t gfp;
> - int ret;
> + u8 *src;
> + unsigned long value;
>
> VM_WARN_ON_ONCE(!folio_test_locked(folio));
> VM_WARN_ON_ONCE(!folio_test_swapcache(folio));
> @@ -1635,60 +1701,11 @@ bool zswap_store(struct folio *folio)
> mem_cgroup_put(memcg);
> }
>
> - /* compress */
> - acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
> -
> - mutex_lock(acomp_ctx->mutex);
> -
> - dst = acomp_ctx->dstmem;
> - sg_init_table(&input, 1);
> - sg_set_page(&input, page, PAGE_SIZE, 0);
> -
> - sg_init_one(&output, dst, PAGE_SIZE);
> - acomp_request_set_params(acomp_ctx->req, &input, &output, PAGE_SIZE, dlen);
> - /*
> - * it maybe looks a little bit silly that we send an asynchronous request,
> - * then wait for its completion synchronously. This makes the process look
> - * synchronous in fact.
> - * Theoretically, acomp supports users send multiple acomp requests in one
> - * acomp instance, then get those requests done simultaneously. but in this
> - * case, zswap actually does store and load page by page, there is no
> - * existing method to send the second page before the first page is done
> - * in one thread doing zwap.
> - * but in different threads running on different cpu, we have different
> - * acomp instance, so multiple threads can do (de)compression in parallel.
> - */
> - ret = crypto_wait_req(crypto_acomp_compress(acomp_ctx->req), &acomp_ctx->wait);
> - dlen = acomp_ctx->req->dlen;
> -
> - if (ret) {
> - zswap_reject_compress_fail++;
> - goto put_dstmem;
> - }
> -
> - /* store */
> - zpool = zswap_find_zpool(entry);
> - gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
> - if (zpool_malloc_support_movable(zpool))
> - gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
> - ret = zpool_malloc(zpool, dlen, gfp, &handle);
> - if (ret == -ENOSPC) {
> - zswap_reject_compress_poor++;
> - goto put_dstmem;
> - }
> - if (ret) {
> - zswap_reject_alloc_fail++;
> - goto put_dstmem;
> - }
> - buf = zpool_map_handle(zpool, handle, ZPOOL_MM_WO);
> - memcpy(buf, dst, dlen);
> - zpool_unmap_handle(zpool, handle);
> - mutex_unlock(acomp_ctx->mutex);
> + if (__zswap_store(entry, page))
> + goto put_pool;
>
> /* populate entry */
> entry->swpentry = swp_entry(type, offset);
> - entry->handle = handle;
> - entry->length = dlen;
>
> insert_entry:
> entry->objcg = objcg;
> @@ -1725,8 +1742,6 @@ bool zswap_store(struct folio *folio)
>
> return true;
>
> -put_dstmem:
> - mutex_unlock(acomp_ctx->mutex);
> put_pool:
> zswap_pool_put(entry->pool);
> freepage:
> @@ -1902,10 +1917,10 @@ static int zswap_setup(void)
> }
>
> ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
> - zswap_dstmem_prepare, zswap_dstmem_dead);
> + zswap_buffer_prepare, zswap_buffer_dead);
> if (ret) {
> - pr_err("dstmem alloc failed\n");
> - goto dstmem_fail;
> + pr_err("buffer alloc failed\n");
> + goto buffer_fail;
> }
>
> ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
> @@ -1940,7 +1955,7 @@ static int zswap_setup(void)
> zswap_pool_destroy(pool);
> hp_fail:
> cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
> -dstmem_fail:
> +buffer_fail:
> kmem_cache_destroy(zswap_entry_cache);
> cache_fail:
> /* if built-in, we aren't unloaded on failure; don't allow use */
>
> --
> b4 0.10.1
On 2023/12/18 17:37, Yosry Ahmed wrote:
> On Mon, Dec 18, 2023 at 12:22 AM Chengming Zhou
> <[email protected]> wrote:
>>
>> Since the introduce of reusing the dstmem in the load path, it seems
>> confusing that we are now using acomp_ctx->dstmem and acomp_ctx->mutex
>> now for purposes other than what the naming suggests.
>>
>> Yosry suggested removing these two fields from acomp_ctx, and directly
>> using zswap_dstmem and zswap_mutex in both the load and store paths,
>> rename them, and add proper comments above their definitions that they
>> are for generic percpu buffering on the load and store paths.
>>
>> So this patch remove dstmem and mutex from acomp_ctx, and rename the
>> zswap_dstmem to zswap_buffer, using the percpu mutex and buffer on
>> the load and store paths. And refactor out __zswap_store() to only
>> include the compress & store, since I found zswap_store() is too long.
>
> I am not sure refactoring out __zswap_store() is useful to be honest,
> but I am not objecting to it, it mirrors __zswap_load() in a sense.
Yes, it mirrors __zswap_load() and only includes compress and store.
And it makes easy for me to only concentrate on __zswap_store/load()
when renaming the percpu buffers and mutex. But if anyone has objection,
I can drop it.
> However, if you want to do so, please do it in a separate patch from
> renaming the percpu buffers and mutex. This will make reviewing easier
> (and make my Suggested-by correctly scoped).
Right, will do.
>
> Also, any reason why raw_smp_processor_id() is used here instead of
> smp_processor_id()?
>
Here we don't need the CPU id stable, since we only need to pick one
CPU and use the mutex to serialize.
And from the comments below in <include/linux/smp.h>, WARN would happen
if we use smp_processor_id() here without other helpers.
* The CPU id is stable when:
*
* - IRQs are disabled;
* - preemption is disabled;
* - the task is CPU affine.
* When CONFIG_DEBUG_PREEMPT; we verify these assumption and WARN
* when smp_processor_id() is used when the CPU id is not stable.
Thanks!
On 2023/12/18 17:37, Yosry Ahmed wrote:
> On Mon, Dec 18, 2023 at 12:22 AM Chengming Zhou
> <[email protected]> wrote:
>>
>> Since the introduce of reusing the dstmem in the load path, it seems
>> confusing that we are now using acomp_ctx->dstmem and acomp_ctx->mutex
>> now for purposes other than what the naming suggests.
>>
>> Yosry suggested removing these two fields from acomp_ctx, and directly
>> using zswap_dstmem and zswap_mutex in both the load and store paths,
>> rename them, and add proper comments above their definitions that they
>> are for generic percpu buffering on the load and store paths.
>>
>> So this patch remove dstmem and mutex from acomp_ctx, and rename the
>> zswap_dstmem to zswap_buffer, using the percpu mutex and buffer on
>> the load and store paths. And refactor out __zswap_store() to only
>> include the compress & store, since I found zswap_store() is too long.
>
> I am not sure refactoring out __zswap_store() is useful to be honest,
> but I am not objecting to it, it mirrors __zswap_load() in a sense.
> However, if you want to do so, please do it in a separate patch from
> renaming the percpu buffers and mutex. This will make reviewing easier
> (and make my Suggested-by correctly scoped).
>
After thinking twice, I dropped the __zswap_store() part. Right, it's
not much useful.
Thanks.