Hi Dominique,
Could you help paste the test result of before-after-applied this patch in
comment? And please see my comments below.
On 2018/8/9 22:33, Dominique Martinet wrote:
> From: Dominique Martinet <dominique.martinet@xxxxxx>
>
> Having a specific cache for the fcall allocations helps speed up
> allocations a bit, especially in case of non-"round" msizes.
>
> The caches will automatically be merged if there are multiple caches
> of items with the same size so we do not need to try to share a cache
> between different clients of the same size.
>
> Since the msize is negotiated with the server, only allocate the cache
> after that negotiation has happened - previous allocations or
> allocations of different sizes (e.g. zero-copy fcall) are made with
> kmalloc directly.
>
> Signed-off-by: Dominique Martinet <dominique.martinet@xxxxxx>
> Cc: Matthew Wilcox <willy@xxxxxxxxxxxxx>
> Cc: Greg Kurz <groug@xxxxxxxx>
> Cc: Jun Piao <piaojun@xxxxxxxxxx>
> ---
> v2:
> - Add a pointer to the cache in p9_fcall to make sure a buffer
> allocated with kmalloc gets freed with kfree and vice-versa
> This could have been smaller with a bool but this spares having
> to look at the client so looked a bit cleaner, I'm expecting this
> patch will need a v3 one way or another so I went for the bolder
> approach - please say if you think a smaller item is better ; I *think*
> nothing relies on this being ordered the same way as the data on the
> wire (struct isn't packed anyway) so we can move id after tag and add
> another u8 to not have any overhead
>
> - added likely() to cache existence check in allocation, but nothing
> for msize check or free because of zc request being of different size
>
> v3:
> - defined the userdata region in the cache, as read or write calls can
> access the buffer directly (lead to warnings with HARDENED_USERCOPY=y)
>
>
>
> I didn't get any comment on v2 for this patch, but I'm still not fully
> happy with this in all honesty.
>
> Part of me believes we might be better off always allocating from the
> cache even for zero-copy headers, it's a waste of space but the buffers
> are there and being reused constantly for non-zc calls, and mixing
> kmallocs in could lead to these buffers being really freed and
> reallocated all the time instead of reusing the slab buffers
> continuously.
>
> That would let me move the likely() for the fast path, with the only
> exception being the TVERSION initial call on mount, for which I still
> don't have any nice idea on how to handle, using a different size in
> p9_client_rpc or prepare_req if the type is TVERSION and I'm really not
> happy with that...
>
>
> include/net/9p/9p.h | 4 ++++
> include/net/9p/client.h | 1 +
> net/9p/client.c | 40 +++++++++++++++++++++++++++++++++++-----
> 3 files changed, 40 insertions(+), 5 deletions(-)
>
> diff --git a/include/net/9p/9p.h b/include/net/9p/9p.h
> index e23896116d9a..645266b74652 100644
> --- a/include/net/9p/9p.h
> +++ b/include/net/9p/9p.h
> @@ -336,6 +336,9 @@ enum p9_qid_t {
> #define P9_NOFID (u32)(~0)
> #define P9_MAXWELEM 16
>
> +/* Minimal header size: len + id + tag */
Here should be 'size + id + tag'.
> +#define P9_HDRSZ 7
> +
> /* ample room for Twrite/Rread header */
> #define P9_IOHDRSZ 24
>
> @@ -558,6 +561,7 @@ struct p9_fcall {
> size_t offset;
> size_t capacity;
>
> + struct kmem_cache *cache;
> u8 *sdata;
> };
>
> diff --git a/include/net/9p/client.h b/include/net/9p/client.h
> index c2671d40bb6b..735f3979d559 100644
> --- a/include/net/9p/client.h
> +++ b/include/net/9p/client.h
> @@ -123,6 +123,7 @@ struct p9_client {
> struct p9_trans_module *trans_mod;
> enum p9_trans_status status;
> void *trans;
> + struct kmem_cache *fcall_cache;
>
> union {
> struct {
> diff --git a/net/9p/client.c b/net/9p/client.c
> index ed78751aee7c..6c57ab1294d7 100644
> --- a/net/9p/client.c
> +++ b/net/9p/client.c
> @@ -192,6 +192,9 @@ static int parse_opts(char *opts, struct p9_client *clnt)
> goto free_and_return;
> }
>
> + if (clnt->trans_mod) {
> + pr_warn("Had trans %s\n", clnt->trans_mod->name);
> + }
> v9fs_put_trans(clnt->trans_mod);
> clnt->trans_mod = v9fs_get_trans_by_name(s);
> if (clnt->trans_mod == NULL) {
> @@ -231,9 +234,16 @@ static int parse_opts(char *opts, struct p9_client *clnt)
> return ret;
> }
>
> -static int p9_fcall_init(struct p9_fcall *fc, int alloc_msize)
> +static int p9_fcall_init(struct p9_client *c, struct p9_fcall *fc,
> + int alloc_msize)
> {
> - fc->sdata = kmalloc(alloc_msize, GFP_NOFS);
> + if (likely(c->fcall_cache) && alloc_msize == c->msize) {
> + fc->sdata = kmem_cache_alloc(c->fcall_cache, GFP_NOFS);
> + fc->cache = c->fcall_cache;
> + } else {
> + fc->sdata = kmalloc(alloc_msize, GFP_NOFS);
> + fc->cache = NULL;
> + }
> if (!fc->sdata)
> return -ENOMEM;
> fc->capacity = alloc_msize;
> @@ -242,7 +252,16 @@ static int p9_fcall_init(struct p9_fcall *fc, int alloc_msize)
>
> void p9_fcall_fini(struct p9_fcall *fc)
> {
> - kfree(fc->sdata);
> + /* sdata can be NULL for interrupted requests in trans_rdma,
> + * and kmem_cache_free does not do NULL-check for us
> + */
> + if (unlikely(!fc->sdata))
> + return;
> +
> + if (fc->cache)
> + kmem_cache_free(fc->cache, fc->sdata);
> + else
> + kfree(fc->sdata);
> }
> EXPORT_SYMBOL(p9_fcall_fini);
>
> @@ -267,9 +286,9 @@ p9_tag_alloc(struct p9_client *c, int8_t type, unsigned int max_size)
> if (!req)
> return NULL;
>
> - if (p9_fcall_init(&req->tc, alloc_msize))
> + if (p9_fcall_init(c, &req->tc, alloc_msize))
> goto free_req;
> - if (p9_fcall_init(&req->rc, alloc_msize))
> + if (p9_fcall_init(c, &req->rc, alloc_msize))
> goto free;
>
> p9pdu_reset(&req->tc);
> @@ -951,6 +970,7 @@ struct p9_client *p9_client_create(const char *dev_name, char *options)
>
> clnt->trans_mod = NULL;
> clnt->trans = NULL;
> + clnt->fcall_cache = NULL;
>
> client_id = utsname()->nodename;
> memcpy(clnt->name, client_id, strlen(client_id) + 1);
> @@ -987,6 +1007,15 @@ struct p9_client *p9_client_create(const char *dev_name, char *options)
> if (err)
> goto close_trans;
>
> + /* P9_HDRSZ + 4 is the smallest packet header we can have that is
> + * followed by data accessed from userspace by read
> + */
> + clnt->fcall_cache =
> + kmem_cache_create_usercopy("9p-fcall-cache", clnt->msize,
> + 0, 0, P9_HDRSZ + 4,
> + clnt->msize - (P9_HDRSZ + 4),
> + NULL);
> +
> return clnt;
>
> close_trans:
> @@ -1018,6 +1047,7 @@ void p9_client_destroy(struct p9_client *clnt)
>
> p9_tag_cleanup(clnt);
>
> + kmem_cache_destroy(clnt->fcall_cache);
I'm afraid that we should check NULL for clnt->fcall_cache.
Thanks,
Jun
> kfree(clnt);
> }
> EXPORT_SYMBOL(p9_client_destroy);
>
