* [PATCH rdma-next 0/9] MR cache fixes and refactoring
@ 2020-02-27 12:33 Leon Romanovsky
2020-02-27 12:33 ` [PATCH mlx5-next 1/9] RDMA/mlx5: Move asynchronous mkey creation to mlx5_ib Leon Romanovsky
0 siblings, 1 reply; 5+ messages in thread
From: Leon Romanovsky @ 2020-02-27 12:33 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, Artemy Kovalyov, Eli Cohen, Jack Morgenstein,
linux-rdma, Michael Guralnik, netdev, Or Gerlitz, Saeed Mahameed,
Yishai Hadas
From: Leon Romanovsky <leonro@mellanox.com>
Hi,
This series fixes various corner cases in the mlx5_ib MR
cache implementation, see specific commit messages for more
information.
Thanks
Jason Gunthorpe (8):
RDMA/mlx5: Rename the tracking variables for the MR cache
RDMA/mlx5: Simplify how the MR cache bucket is located
RDMA/mlx5: Always remove MRs from the cache before destroying them
RDMA/mlx5: Fix MR cache size and limit debugfs
RDMA/mlx5: Lock access to ent->available_mrs/limit when doing
queue_work
RDMA/mlx5: Fix locking in MR cache work queue
RDMA/mlx5: Revise how the hysteresis scheme works for cache filling
RDMA/mlx5: Allow MRs to be created in the cache synchronously
Michael Guralnik (1):
RDMA/mlx5: Move asynchronous mkey creation to mlx5_ib
drivers/infiniband/hw/mlx5/mlx5_ib.h | 33 +-
drivers/infiniband/hw/mlx5/mr.c | 642 +++++++++++--------
drivers/infiniband/hw/mlx5/odp.c | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/mr.c | 22 +-
include/linux/mlx5/driver.h | 6 -
5 files changed, 404 insertions(+), 301 deletions(-)
--
2.24.1
^ permalink raw reply [flat|nested] 5+ messages in thread* [PATCH mlx5-next 1/9] RDMA/mlx5: Move asynchronous mkey creation to mlx5_ib 2020-02-27 12:33 [PATCH rdma-next 0/9] MR cache fixes and refactoring Leon Romanovsky @ 2020-02-27 12:33 ` Leon Romanovsky 2020-02-27 19:41 ` Saeed Mahameed 0 siblings, 1 reply; 5+ messages in thread From: Leon Romanovsky @ 2020-02-27 12:33 UTC (permalink / raw) To: Doug Ledford, Jason Gunthorpe Cc: Michael Guralnik, linux-rdma, netdev, Saeed Mahameed From: Michael Guralnik <michaelgur@mellanox.com> As mlx5_ib is the only user of the mlx5_core_create_mkey_cb, move the logic inside mlx5_ib and cleanup the code in mlx5_core. Signed-off-by: Michael Guralnik <michaelgur@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> --- drivers/infiniband/hw/mlx5/mr.c | 25 ++++++++++++++++---- drivers/net/ethernet/mellanox/mlx5/core/mr.c | 22 +++-------------- include/linux/mlx5/driver.h | 6 ----- 3 files changed, 24 insertions(+), 29 deletions(-) diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index 6fa0a83c19de..dea14477a676 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -79,6 +79,25 @@ static bool use_umr_mtt_update(struct mlx5_ib_mr *mr, u64 start, u64 length) length + (start & (MLX5_ADAPTER_PAGE_SIZE - 1)); } +static int create_mkey_cb(struct mlx5_core_dev *dev, struct mlx5_ib_mr *mr, + struct mlx5_async_ctx *async_ctx, u32 *in, int inlen, + mlx5_async_cbk_t callback) +{ + void *mkc; + u8 key; + + spin_lock_irq(&dev->priv.mkey_lock); + key = dev->priv.mkey_key++; + spin_unlock_irq(&dev->priv.mkey_lock); + mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); + + MLX5_SET(create_mkey_in, in, opcode, MLX5_CMD_OP_CREATE_MKEY); + MLX5_SET(mkc, mkc, mkey_7_0, key); + + return mlx5_cmd_exec_cb(async_ctx, in, inlen, mr->out, sizeof(mr->out), + callback, &mr->cb_work); +} + static void reg_mr_callback(int status, struct mlx5_async_work *context) { struct mlx5_ib_mr *mr = @@ -163,10 +182,8 @@ static int add_keys(struct mlx5_ib_dev *dev, int c, int num) spin_lock_irq(&ent->lock); ent->pending++; spin_unlock_irq(&ent->lock); - err = mlx5_core_create_mkey_cb(dev->mdev, &mr->mmkey, - &dev->async_ctx, in, inlen, - mr->out, sizeof(mr->out), - reg_mr_callback, &mr->cb_work); + err = create_mkey_cb(dev->mdev, mr, &dev->async_ctx, in, inlen, + reg_mr_callback); if (err) { spin_lock_irq(&ent->lock); ent->pending--; diff --git a/drivers/net/ethernet/mellanox/mlx5/core/mr.c b/drivers/net/ethernet/mellanox/mlx5/core/mr.c index 42cc3c7ac5b6..83841e4119d7 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/mr.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/mr.c @@ -36,12 +36,9 @@ #include <linux/mlx5/cmd.h> #include "mlx5_core.h" -int mlx5_core_create_mkey_cb(struct mlx5_core_dev *dev, - struct mlx5_core_mkey *mkey, - struct mlx5_async_ctx *async_ctx, u32 *in, - int inlen, u32 *out, int outlen, - mlx5_async_cbk_t callback, - struct mlx5_async_work *context) +int mlx5_core_create_mkey(struct mlx5_core_dev *dev, + struct mlx5_core_mkey *mkey, + u32 *in, int inlen) { u32 lout[MLX5_ST_SZ_DW(create_mkey_out)] = {0}; u32 mkey_index; @@ -57,10 +54,6 @@ int mlx5_core_create_mkey_cb(struct mlx5_core_dev *dev, MLX5_SET(create_mkey_in, in, opcode, MLX5_CMD_OP_CREATE_MKEY); MLX5_SET(mkc, mkc, mkey_7_0, key); - if (callback) - return mlx5_cmd_exec_cb(async_ctx, in, inlen, out, outlen, - callback, context); - err = mlx5_cmd_exec(dev, in, inlen, lout, sizeof(lout)); if (err) return err; @@ -75,15 +68,6 @@ int mlx5_core_create_mkey_cb(struct mlx5_core_dev *dev, mkey_index, key, mkey->key); return 0; } -EXPORT_SYMBOL(mlx5_core_create_mkey_cb); - -int mlx5_core_create_mkey(struct mlx5_core_dev *dev, - struct mlx5_core_mkey *mkey, - u32 *in, int inlen) -{ - return mlx5_core_create_mkey_cb(dev, mkey, NULL, in, inlen, - NULL, 0, NULL, NULL); -} EXPORT_SYMBOL(mlx5_core_create_mkey); int mlx5_core_destroy_mkey(struct mlx5_core_dev *dev, diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h index f2b4225ed650..7225e9ca0f25 100644 --- a/include/linux/mlx5/driver.h +++ b/include/linux/mlx5/driver.h @@ -947,12 +947,6 @@ struct mlx5_cmd_mailbox *mlx5_alloc_cmd_mailbox_chain(struct mlx5_core_dev *dev, gfp_t flags, int npages); void mlx5_free_cmd_mailbox_chain(struct mlx5_core_dev *dev, struct mlx5_cmd_mailbox *head); -int mlx5_core_create_mkey_cb(struct mlx5_core_dev *dev, - struct mlx5_core_mkey *mkey, - struct mlx5_async_ctx *async_ctx, u32 *in, - int inlen, u32 *out, int outlen, - mlx5_async_cbk_t callback, - struct mlx5_async_work *context); int mlx5_core_create_mkey(struct mlx5_core_dev *dev, struct mlx5_core_mkey *mkey, u32 *in, int inlen); -- 2.24.1 ^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [PATCH mlx5-next 1/9] RDMA/mlx5: Move asynchronous mkey creation to mlx5_ib 2020-02-27 12:33 ` [PATCH mlx5-next 1/9] RDMA/mlx5: Move asynchronous mkey creation to mlx5_ib Leon Romanovsky @ 2020-02-27 19:41 ` Saeed Mahameed 2020-02-27 20:00 ` Jason Gunthorpe 0 siblings, 1 reply; 5+ messages in thread From: Saeed Mahameed @ 2020-02-27 19:41 UTC (permalink / raw) To: Jason Gunthorpe, leon@kernel.org, dledford@redhat.com Cc: Michael Guralnik, netdev@vger.kernel.org, linux-rdma@vger.kernel.org On Thu, 2020-02-27 at 14:33 +0200, Leon Romanovsky wrote: > From: Michael Guralnik <michaelgur@mellanox.com> > > As mlx5_ib is the only user of the mlx5_core_create_mkey_cb, move the > logic inside mlx5_ib and cleanup the code in mlx5_core. > I have a WIP series that is moving the whole mr.c to mlx5_ib. https://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux.git/log/?h=topic/mr-relocate > Signed-off-by: Michael Guralnik <michaelgur@mellanox.com> > Signed-off-by: Leon Romanovsky <leonro@mellanox.com> > --- > drivers/infiniband/hw/mlx5/mr.c | 25 ++++++++++++++++ > ---- > drivers/net/ethernet/mellanox/mlx5/core/mr.c | 22 +++-------------- > include/linux/mlx5/driver.h | 6 ----- > 3 files changed, 24 insertions(+), 29 deletions(-) > > diff --git a/drivers/infiniband/hw/mlx5/mr.c > b/drivers/infiniband/hw/mlx5/mr.c > index 6fa0a83c19de..dea14477a676 100644 > --- a/drivers/infiniband/hw/mlx5/mr.c > +++ b/drivers/infiniband/hw/mlx5/mr.c > @@ -79,6 +79,25 @@ static bool use_umr_mtt_update(struct mlx5_ib_mr > *mr, u64 start, u64 length) > length + (start & (MLX5_ADAPTER_PAGE_SIZE - 1)); > } > > +static int create_mkey_cb(struct mlx5_core_dev *dev, struct > mlx5_ib_mr *mr, > + struct mlx5_async_ctx *async_ctx, u32 *in, > int inlen, > + mlx5_async_cbk_t callback) > +{ > + void *mkc; > + u8 key; > + > + spin_lock_irq(&dev->priv.mkey_lock); > + key = dev->priv.mkey_key++; you know i don't like mlx5_ib sniffing around mlx5_core->priv .. this is handled correctly in my series, i can rebase it and make it ready in a couple of days.. let me know if this will be good enough for you. > + spin_unlock_irq(&dev->priv.mkey_lock); > + mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); > + > + MLX5_SET(create_mkey_in, in, opcode, MLX5_CMD_OP_CREATE_MKEY); > + MLX5_SET(mkc, mkc, mkey_7_0, key); > + > + return mlx5_cmd_exec_cb(async_ctx, in, inlen, mr->out, > sizeof(mr->out), > + callback, &mr->cb_work); > +} > + > static void reg_mr_callback(int status, struct mlx5_async_work > *context) > { > struct mlx5_ib_mr *mr = > @@ -163,10 +182,8 @@ static int add_keys(struct mlx5_ib_dev *dev, int > c, int num) > spin_lock_irq(&ent->lock); > ent->pending++; > spin_unlock_irq(&ent->lock); > - err = mlx5_core_create_mkey_cb(dev->mdev, &mr->mmkey, > - &dev->async_ctx, in, > inlen, > - mr->out, sizeof(mr- > >out), > - reg_mr_callback, &mr- > >cb_work); > + err = create_mkey_cb(dev->mdev, mr, &dev->async_ctx, > in, inlen, > + reg_mr_callback); > if (err) { > spin_lock_irq(&ent->lock); > ent->pending--; > diff --git a/drivers/net/ethernet/mellanox/mlx5/core/mr.c > b/drivers/net/ethernet/mellanox/mlx5/core/mr.c > index 42cc3c7ac5b6..83841e4119d7 100644 > --- a/drivers/net/ethernet/mellanox/mlx5/core/mr.c > +++ b/drivers/net/ethernet/mellanox/mlx5/core/mr.c > @@ -36,12 +36,9 @@ > #include <linux/mlx5/cmd.h> > #include "mlx5_core.h" > > -int mlx5_core_create_mkey_cb(struct mlx5_core_dev *dev, > - struct mlx5_core_mkey *mkey, > - struct mlx5_async_ctx *async_ctx, u32 *in, > - int inlen, u32 *out, int outlen, > - mlx5_async_cbk_t callback, > - struct mlx5_async_work *context) > +int mlx5_core_create_mkey(struct mlx5_core_dev *dev, > + struct mlx5_core_mkey *mkey, > + u32 *in, int inlen) > { > u32 lout[MLX5_ST_SZ_DW(create_mkey_out)] = {0}; > u32 mkey_index; > @@ -57,10 +54,6 @@ int mlx5_core_create_mkey_cb(struct mlx5_core_dev > *dev, > MLX5_SET(create_mkey_in, in, opcode, MLX5_CMD_OP_CREATE_MKEY); > MLX5_SET(mkc, mkc, mkey_7_0, key); > > - if (callback) > - return mlx5_cmd_exec_cb(async_ctx, in, inlen, out, > outlen, > - callback, context); > - > err = mlx5_cmd_exec(dev, in, inlen, lout, sizeof(lout)); > if (err) > return err; > @@ -75,15 +68,6 @@ int mlx5_core_create_mkey_cb(struct mlx5_core_dev > *dev, > mkey_index, key, mkey->key); > return 0; > } > -EXPORT_SYMBOL(mlx5_core_create_mkey_cb); > - > -int mlx5_core_create_mkey(struct mlx5_core_dev *dev, > - struct mlx5_core_mkey *mkey, > - u32 *in, int inlen) > -{ > - return mlx5_core_create_mkey_cb(dev, mkey, NULL, in, inlen, > - NULL, 0, NULL, NULL); > -} > EXPORT_SYMBOL(mlx5_core_create_mkey); > > int mlx5_core_destroy_mkey(struct mlx5_core_dev *dev, > diff --git a/include/linux/mlx5/driver.h > b/include/linux/mlx5/driver.h > index f2b4225ed650..7225e9ca0f25 100644 > --- a/include/linux/mlx5/driver.h > +++ b/include/linux/mlx5/driver.h > @@ -947,12 +947,6 @@ struct mlx5_cmd_mailbox > *mlx5_alloc_cmd_mailbox_chain(struct mlx5_core_dev *dev, > gfp_t flags, int > npages); > void mlx5_free_cmd_mailbox_chain(struct mlx5_core_dev *dev, > struct mlx5_cmd_mailbox *head); > -int mlx5_core_create_mkey_cb(struct mlx5_core_dev *dev, > - struct mlx5_core_mkey *mkey, > - struct mlx5_async_ctx *async_ctx, u32 *in, > - int inlen, u32 *out, int outlen, > - mlx5_async_cbk_t callback, > - struct mlx5_async_work *context); > int mlx5_core_create_mkey(struct mlx5_core_dev *dev, > struct mlx5_core_mkey *mkey, > u32 *in, int inlen); ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH mlx5-next 1/9] RDMA/mlx5: Move asynchronous mkey creation to mlx5_ib 2020-02-27 19:41 ` Saeed Mahameed @ 2020-02-27 20:00 ` Jason Gunthorpe 2020-02-27 20:42 ` Saeed Mahameed 0 siblings, 1 reply; 5+ messages in thread From: Jason Gunthorpe @ 2020-02-27 20:00 UTC (permalink / raw) To: Saeed Mahameed Cc: leon@kernel.org, dledford@redhat.com, Michael Guralnik, netdev@vger.kernel.org, linux-rdma@vger.kernel.org On Thu, Feb 27, 2020 at 07:41:24PM +0000, Saeed Mahameed wrote: > On Thu, 2020-02-27 at 14:33 +0200, Leon Romanovsky wrote: > > From: Michael Guralnik <michaelgur@mellanox.com> > > > > As mlx5_ib is the only user of the mlx5_core_create_mkey_cb, move the > > logic inside mlx5_ib and cleanup the code in mlx5_core. > > > > I have a WIP series that is moving the whole mr.c to mlx5_ib. > https://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux.git/log/?h=topic/mr-relocate > > > > Signed-off-by: Michael Guralnik <michaelgur@mellanox.com> > > Signed-off-by: Leon Romanovsky <leonro@mellanox.com> > > drivers/infiniband/hw/mlx5/mr.c | 25 ++++++++++++++++ > > drivers/net/ethernet/mellanox/mlx5/core/mr.c | 22 +++-------------- > > include/linux/mlx5/driver.h | 6 ----- > > 3 files changed, 24 insertions(+), 29 deletions(-) > > > > diff --git a/drivers/infiniband/hw/mlx5/mr.c > > b/drivers/infiniband/hw/mlx5/mr.c > > index 6fa0a83c19de..dea14477a676 100644 > > +++ b/drivers/infiniband/hw/mlx5/mr.c > > @@ -79,6 +79,25 @@ static bool use_umr_mtt_update(struct mlx5_ib_mr > > *mr, u64 start, u64 length) > > length + (start & (MLX5_ADAPTER_PAGE_SIZE - 1)); > > } > > > > +static int create_mkey_cb(struct mlx5_core_dev *dev, struct > > mlx5_ib_mr *mr, > > + struct mlx5_async_ctx *async_ctx, u32 *in, > > int inlen, > > + mlx5_async_cbk_t callback) > > +{ > > + void *mkc; > > + u8 key; > > + > > + spin_lock_irq(&dev->priv.mkey_lock); > > + key = dev->priv.mkey_key++; > > you know i don't like mlx5_ib sniffing around mlx5_core->priv .. > > this is handled correctly in my series, i can rebase it and make it > ready in a couple of days.. let me know if this will be good enough for > you. How about Michael just take the two relevant patches into this series {IB,net}/mlx5: Setup mkey variant before mr create command invocation {IB,net}/mlx5: Assign mkey variant in mlx5_ib only And this partially moves toward your series. It will be more than a few days to rebase and check all the parts of your series I think. Jason ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH mlx5-next 1/9] RDMA/mlx5: Move asynchronous mkey creation to mlx5_ib 2020-02-27 20:00 ` Jason Gunthorpe @ 2020-02-27 20:42 ` Saeed Mahameed 0 siblings, 0 replies; 5+ messages in thread From: Saeed Mahameed @ 2020-02-27 20:42 UTC (permalink / raw) To: Jason Gunthorpe Cc: Michael Guralnik, netdev@vger.kernel.org, linux-rdma@vger.kernel.org, leon@kernel.org, dledford@redhat.com On Thu, 2020-02-27 at 16:00 -0400, Jason Gunthorpe wrote: > On Thu, Feb 27, 2020 at 07:41:24PM +0000, Saeed Mahameed wrote: > > On Thu, 2020-02-27 at 14:33 +0200, Leon Romanovsky wrote: > > > From: Michael Guralnik <michaelgur@mellanox.com> > > > > > > As mlx5_ib is the only user of the mlx5_core_create_mkey_cb, move > > > the > > > logic inside mlx5_ib and cleanup the code in mlx5_core. > > > > > > > I have a WIP series that is moving the whole mr.c to mlx5_ib. > > https://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux.git/log/?h=topic/mr-relocate > > > > > > > Signed-off-by: Michael Guralnik <michaelgur@mellanox.com> > > > Signed-off-by: Leon Romanovsky <leonro@mellanox.com> > > > drivers/infiniband/hw/mlx5/mr.c | 25 > > > ++++++++++++++++ > > > drivers/net/ethernet/mellanox/mlx5/core/mr.c | 22 +++----------- > > > --- > > > include/linux/mlx5/driver.h | 6 ----- > > > 3 files changed, 24 insertions(+), 29 deletions(-) > > > > > > diff --git a/drivers/infiniband/hw/mlx5/mr.c > > > b/drivers/infiniband/hw/mlx5/mr.c > > > index 6fa0a83c19de..dea14477a676 100644 > > > +++ b/drivers/infiniband/hw/mlx5/mr.c > > > @@ -79,6 +79,25 @@ static bool use_umr_mtt_update(struct > > > mlx5_ib_mr > > > *mr, u64 start, u64 length) > > > length + (start & (MLX5_ADAPTER_PAGE_SIZE - 1)); > > > } > > > > > > +static int create_mkey_cb(struct mlx5_core_dev *dev, struct > > > mlx5_ib_mr *mr, > > > + struct mlx5_async_ctx *async_ctx, u32 *in, > > > int inlen, > > > + mlx5_async_cbk_t callback) > > > +{ > > > + void *mkc; > > > + u8 key; > > > + > > > + spin_lock_irq(&dev->priv.mkey_lock); > > > + key = dev->priv.mkey_key++; > > > > you know i don't like mlx5_ib sniffing around mlx5_core->priv .. > > > > this is handled correctly in my series, i can rebase it and make it > > ready in a couple of days.. let me know if this will be good enough > > for > > you. > > How about Michael just take the two relevant patches into this series > > {IB,net}/mlx5: Setup mkey variant before mr create command invocation > {IB,net}/mlx5: Assign mkey variant in mlx5_ib only > Also, IB/mlx5: Replace spinlock protected write with atomic var it is a good one :) > And this partially moves toward your series. It will be more than a > few days to rebase and check all the parts of your series I think. Okay. -Saeed. ^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2020-02-27 20:42 UTC | newest] Thread overview: 5+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2020-02-27 12:33 [PATCH rdma-next 0/9] MR cache fixes and refactoring Leon Romanovsky 2020-02-27 12:33 ` [PATCH mlx5-next 1/9] RDMA/mlx5: Move asynchronous mkey creation to mlx5_ib Leon Romanovsky 2020-02-27 19:41 ` Saeed Mahameed 2020-02-27 20:00 ` Jason Gunthorpe 2020-02-27 20:42 ` Saeed Mahameed
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).