RE: [PATCH v2 5/9] NTB: Alter Scratchpads API to support multi-ports devices

From: Allen Hubbe
Date: Mon Dec 12 2016 - 23:13:52 EST


From: Serge Semin
> Even though there is no any real NTB hardware, which would have both more
> than two ports and Scratchpad registers, it is logically correct to have
> Scratchpad API accepting a peer port index as well. Intel/AMD drivers utilize
> Primary and Secondary topology to split Scratchpad between connected root
> devices. Since port-index API introduced, Intel/AMD NTB hadrware drivers can

s/hadrware/hardware/

> use device port to determine which Scratchpad registers actually belong to
> local and peer devices. The same approach can be used if some potential
> hardware in future will be multi-port and have some set of Scratchpads.
> Here are the brief of changes in the API:
> ntb_spad_count() - return number of Scratchpad per each port
> ntb_peer_spad_addr(pidx, sidx) - address of Scratchpad register of the
> peer device with pidx-index
> ntb_peer_spad_read(pidx, sidx) - read specified Scratchpad register of the
> peer with pidx-index
> ntb_peer_spad_write(pidx, sidx) - write data to Scratchpad register of the
> peer with pidx-index
>
> Since there is hardware which doesn't support Scratchpad registers, the
> corresponding API methods are now made optional.

The api change looks good. See the comment to simplify ntb_tool.

> Signed-off-by: Serge Semin <fancer.lancer@xxxxxxxxx>
>
> ---
> drivers/ntb/hw/amd/ntb_hw_amd.c | 14 +++----
> drivers/ntb/hw/intel/ntb_hw_intel.c | 14 +++----
> drivers/ntb/ntb_transport.c | 17 ++++-----
> drivers/ntb/test/ntb_perf.c | 6 +--
> drivers/ntb/test/ntb_pingpong.c | 8 +++-
> drivers/ntb/test/ntb_tool.c | 45 +++++++++++++++++-----
> include/linux/ntb.h | 76 +++++++++++++++++++++++--------------
> 7 files changed, 115 insertions(+), 65 deletions(-)
>
> diff --git a/drivers/ntb/hw/amd/ntb_hw_amd.c b/drivers/ntb/hw/amd/ntb_hw_amd.c
> index 74fe9b8..a2596ad 100644
> --- a/drivers/ntb/hw/amd/ntb_hw_amd.c
> +++ b/drivers/ntb/hw/amd/ntb_hw_amd.c
> @@ -476,30 +476,30 @@ static int amd_ntb_spad_write(struct ntb_dev *ntb,
> return 0;
> }
>
> -static u32 amd_ntb_peer_spad_read(struct ntb_dev *ntb, int idx)
> +static u32 amd_ntb_peer_spad_read(struct ntb_dev *ntb, int pidx, int sidx)
> {
> struct amd_ntb_dev *ndev = ntb_ndev(ntb);
> void __iomem *mmio = ndev->self_mmio;
> u32 offset;
>
> - if (idx < 0 || idx >= ndev->spad_count)
> + if (sidx < 0 || sidx >= ndev->spad_count)
> return -EINVAL;
>
> - offset = ndev->peer_spad + (idx << 2);
> + offset = ndev->peer_spad + (sidx << 2);
> return readl(mmio + AMD_SPAD_OFFSET + offset);
> }
>
> -static int amd_ntb_peer_spad_write(struct ntb_dev *ntb,
> - int idx, u32 val)
> +static int amd_ntb_peer_spad_write(struct ntb_dev *ntb, int pidx,
> + int sidx, u32 val)
> {
> struct amd_ntb_dev *ndev = ntb_ndev(ntb);
> void __iomem *mmio = ndev->self_mmio;
> u32 offset;
>
> - if (idx < 0 || idx >= ndev->spad_count)
> + if (sidx < 0 || sidx >= ndev->spad_count)
> return -EINVAL;
>
> - offset = ndev->peer_spad + (idx << 2);
> + offset = ndev->peer_spad + (sidx << 2);
> writel(val, mmio + AMD_SPAD_OFFSET + offset);
>
> return 0;
> diff --git a/drivers/ntb/hw/intel/ntb_hw_intel.c b/drivers/ntb/hw/intel/ntb_hw_intel.c
> index 5a57d9e..471b0ba 100644
> --- a/drivers/ntb/hw/intel/ntb_hw_intel.c
> +++ b/drivers/ntb/hw/intel/ntb_hw_intel.c
> @@ -1452,30 +1452,30 @@ static int intel_ntb_spad_write(struct ntb_dev *ntb,
> ndev->self_reg->spad);
> }
>
> -static int intel_ntb_peer_spad_addr(struct ntb_dev *ntb, int idx,
> +static int intel_ntb_peer_spad_addr(struct ntb_dev *ntb, int pidx, int sidx,
> phys_addr_t *spad_addr)
> {
> struct intel_ntb_dev *ndev = ntb_ndev(ntb);
>
> - return ndev_spad_addr(ndev, idx, spad_addr, ndev->peer_addr,
> + return ndev_spad_addr(ndev, sidx, spad_addr, ndev->peer_addr,
> ndev->peer_reg->spad);
> }
>
> -static u32 intel_ntb_peer_spad_read(struct ntb_dev *ntb, int idx)
> +static u32 intel_ntb_peer_spad_read(struct ntb_dev *ntb, int pidx, int sidx)
> {
> struct intel_ntb_dev *ndev = ntb_ndev(ntb);
>
> - return ndev_spad_read(ndev, idx,
> + return ndev_spad_read(ndev, sidx,
> ndev->peer_mmio +
> ndev->peer_reg->spad);
> }
>
> -static int intel_ntb_peer_spad_write(struct ntb_dev *ntb,
> - int idx, u32 val)
> +static int intel_ntb_peer_spad_write(struct ntb_dev *ntb, int pidx,
> + int sidx, u32 val)
> {
> struct intel_ntb_dev *ndev = ntb_ndev(ntb);
>
> - return ndev_spad_write(ndev, idx, val,
> + return ndev_spad_write(ndev, sidx, val,
> ndev->peer_mmio +
> ndev->peer_reg->spad);
> }
> diff --git a/drivers/ntb/ntb_transport.c b/drivers/ntb/ntb_transport.c
> index cb4f99889..b2475f4 100644
> --- a/drivers/ntb/ntb_transport.c
> +++ b/drivers/ntb/ntb_transport.c
> @@ -875,17 +875,17 @@ static void ntb_transport_link_work(struct work_struct *work)
> size = max_mw_size;
>
> spad = MW0_SZ_HIGH + (i * 2);
> - ntb_peer_spad_write(ndev, spad, upper_32_bits(size));
> + ntb_peer_spad_write(ndev, PIDX, spad, upper_32_bits(size));
>
> spad = MW0_SZ_LOW + (i * 2);
> - ntb_peer_spad_write(ndev, spad, lower_32_bits(size));
> + ntb_peer_spad_write(ndev, PIDX, spad, lower_32_bits(size));
> }
>
> - ntb_peer_spad_write(ndev, NUM_MWS, nt->mw_count);
> + ntb_peer_spad_write(ndev, PIDX, NUM_MWS, nt->mw_count);
>
> - ntb_peer_spad_write(ndev, NUM_QPS, nt->qp_count);
> + ntb_peer_spad_write(ndev, PIDX, NUM_QPS, nt->qp_count);
>
> - ntb_peer_spad_write(ndev, VERSION, NTB_TRANSPORT_VERSION);
> + ntb_peer_spad_write(ndev, PIDX, VERSION, NTB_TRANSPORT_VERSION);
>
> /* Query the remote side for its info */
> val = ntb_spad_read(ndev, VERSION);
> @@ -961,10 +961,10 @@ static void ntb_qp_link_work(struct work_struct *work)
>
> val = ntb_spad_read(nt->ndev, QP_LINKS);
>
> - ntb_peer_spad_write(nt->ndev, QP_LINKS, val | BIT(qp->qp_num));
> + ntb_peer_spad_write(nt->ndev, PIDX, QP_LINKS, val | BIT(qp->qp_num));
>
> /* query remote spad for qp ready bits */
> - ntb_peer_spad_read(nt->ndev, QP_LINKS);
> + ntb_peer_spad_read(nt->ndev, PIDX, QP_LINKS);
> dev_dbg_ratelimited(&pdev->dev, "Remote QP link status = %x\n", val);
>
> /* See if the remote side is up */
> @@ -2141,8 +2141,7 @@ void ntb_transport_link_down(struct ntb_transport_qp *qp)
>
> val = ntb_spad_read(qp->ndev, QP_LINKS);
>
> - ntb_peer_spad_write(qp->ndev, QP_LINKS,
> - val & ~BIT(qp->qp_num));
> + ntb_peer_spad_write(qp->ndev, PIDX, QP_LINKS, val & ~BIT(qp->qp_num));
>
> if (qp->link_is_up)
> ntb_send_link_down(qp);
> diff --git a/drivers/ntb/test/ntb_perf.c b/drivers/ntb/test/ntb_perf.c
> index 3efb5b5..99f1522 100644
> --- a/drivers/ntb/test/ntb_perf.c
> +++ b/drivers/ntb/test/ntb_perf.c
> @@ -516,9 +516,9 @@ static void perf_link_work(struct work_struct *work)
> if (max_mw_size && size > max_mw_size)
> size = max_mw_size;
>
> - ntb_peer_spad_write(ndev, MW_SZ_HIGH, upper_32_bits(size));
> - ntb_peer_spad_write(ndev, MW_SZ_LOW, lower_32_bits(size));
> - ntb_peer_spad_write(ndev, VERSION, PERF_VERSION);
> + ntb_peer_spad_write(ndev, PIDX, MW_SZ_HIGH, upper_32_bits(size));
> + ntb_peer_spad_write(ndev, PIDX, MW_SZ_LOW, lower_32_bits(size));
> + ntb_peer_spad_write(ndev, PIDX, VERSION, PERF_VERSION);
>
> /* now read what peer wrote */
> val = ntb_spad_read(ndev, VERSION);
> diff --git a/drivers/ntb/test/ntb_pingpong.c b/drivers/ntb/test/ntb_pingpong.c
> index 6dd7582..4ee5c14 100644
> --- a/drivers/ntb/test/ntb_pingpong.c
> +++ b/drivers/ntb/test/ntb_pingpong.c
> @@ -138,7 +138,7 @@ static void pp_ping(unsigned long ctx)
> "Ping bits %#llx read %#x write %#x\n",
> db_bits, spad_rd, spad_wr);
>
> - ntb_peer_spad_write(pp->ntb, 0, spad_wr);
> + ntb_peer_spad_write(pp->ntb, PIDX, 0, spad_wr);
> ntb_peer_db_set(pp->ntb, db_bits);
> ntb_db_clear_mask(pp->ntb, db_mask);
>
> @@ -225,6 +225,12 @@ static int pp_probe(struct ntb_client *client,
> }
> }
>
> + if (ntb_spad_count(ntb) < 1) {
> + dev_dbg(&ntb->dev, "no enough scratchpads\n");
> + rc = -EINVAL;
> + goto err_pp;
> + }
> +
> if (ntb_spad_is_unsafe(ntb)) {
> dev_dbg(&ntb->dev, "scratchpad is unsafe\n");
> if (!unsafe) {
> diff --git a/drivers/ntb/test/ntb_tool.c b/drivers/ntb/test/ntb_tool.c
> index 7aa6018..0f57b2e 100644
> --- a/drivers/ntb/test/ntb_tool.c
> +++ b/drivers/ntb/test/ntb_tool.c
> @@ -264,14 +264,17 @@ static ssize_t tool_dbfn_write(struct tool_ctx *tc,
>
> static ssize_t tool_spadfn_read(struct tool_ctx *tc, char __user *ubuf,
> size_t size, loff_t *offp,
> - u32 (*spad_read_fn)(struct ntb_dev *, int))
> + u32 (*spad_read_fn)(struct ntb_dev *, int),
> + u32 (*spad_peer_read_fn)(struct ntb_dev *, int,
> + int))
> {
> size_t buf_size;
> char *buf;
> ssize_t pos, rc;
> int i, spad_count;
> + u32 data;
>
> - if (!spad_read_fn)
> + if (!spad_read_fn && !spad_peer_read_fn)
> return -EINVAL;
>
> spad_count = ntb_spad_count(tc->ntb);
> @@ -290,8 +293,12 @@ static ssize_t tool_spadfn_read(struct tool_ctx *tc, char __user
> *ubuf,
> pos = 0;
>
> for (i = 0; i < spad_count; ++i) {
> + if (spad_read_fn)
> + data = spad_read_fn(tc->ntb, i);
> + else
> + data = spad_peer_read_fn(tc->ntb, PIDX, i);

As long as we are just supporting pidx zero for now, the changes in ntb_tool can be simplified.

static u32 ntb_tool_peer_spad_read(struct ntb_dev *ntb, int sidx)
{
return ntb_peer_spad_read(ntb, PIDX, sidx);
}

Then, just pass ntb_tool_peer_spad_read in place of tc->ntb->ops.peer_spad_read.

Similar for peer_spad_write.

> pos += scnprintf(buf + pos, buf_size - pos, "%d\t%#x\n",
> - i, spad_read_fn(tc->ntb, i));
> + i, data);
> }
>
> rc = simple_read_from_buffer(ubuf, size, offp, buf, pos);
> @@ -305,7 +312,9 @@ static ssize_t tool_spadfn_write(struct tool_ctx *tc,
> const char __user *ubuf,
> size_t size, loff_t *offp,
> int (*spad_write_fn)(struct ntb_dev *,
> - int, u32))
> + int, u32),
> + int (*spad_peer_write_fn)(struct ntb_dev *,
> + int, int, u32))
> {
> int spad_idx;
> u32 spad_val;
> @@ -313,7 +322,7 @@ static ssize_t tool_spadfn_write(struct tool_ctx *tc,
> int pos, n;
> ssize_t rc;
>
> - if (!spad_write_fn) {
> + if (!spad_write_fn || !spad_peer_write_fn) {
> dev_dbg(&tc->ntb->dev, "no spad write fn\n");
> return -EINVAL;
> }
> @@ -333,7 +342,11 @@ static ssize_t tool_spadfn_write(struct tool_ctx *tc,
> n = sscanf(buf_ptr, "%d %i%n", &spad_idx, &spad_val, &pos);
> while (n == 2) {
> buf_ptr += pos;
> - rc = spad_write_fn(tc->ntb, spad_idx, spad_val);
> + if (spad_write_fn)
> + rc = spad_write_fn(tc->ntb, spad_idx, spad_val);
> + else
> + rc = spad_peer_write_fn(tc->ntb, PIDX, spad_idx,
> + spad_val);
> if (rc)
> break;
>
> @@ -446,7 +459,7 @@ static ssize_t tool_spad_read(struct file *filep, char __user *ubuf,
> struct tool_ctx *tc = filep->private_data;
>
> return tool_spadfn_read(tc, ubuf, size, offp,
> - tc->ntb->ops->spad_read);
> + tc->ntb->ops->spad_read, NULL);
> }
>
> static ssize_t tool_spad_write(struct file *filep, const char __user *ubuf,
> @@ -455,7 +468,7 @@ static ssize_t tool_spad_write(struct file *filep, const char __user
> *ubuf,
> struct tool_ctx *tc = filep->private_data;
>
> return tool_spadfn_write(tc, ubuf, size, offp,
> - tc->ntb->ops->spad_write);
> + tc->ntb->ops->spad_write, NULL);
> }
>
> static TOOL_FOPS_RDWR(tool_spad_fops,
> @@ -467,7 +480,7 @@ static ssize_t tool_peer_spad_read(struct file *filep, char __user
> *ubuf,
> {
> struct tool_ctx *tc = filep->private_data;
>
> - return tool_spadfn_read(tc, ubuf, size, offp,
> + return tool_spadfn_read(tc, ubuf, size, offp, NULL,
> tc->ntb->ops->peer_spad_read);
> }
>
> @@ -476,7 +489,7 @@ static ssize_t tool_peer_spad_write(struct file *filep, const char
> __user *ubuf,
> {
> struct tool_ctx *tc = filep->private_data;
>
> - return tool_spadfn_write(tc, ubuf, size, offp,
> + return tool_spadfn_write(tc, ubuf, size, offp, NULL,
> tc->ntb->ops->peer_spad_write);
> }
>
> @@ -935,6 +948,18 @@ static int tool_probe(struct ntb_client *self, struct ntb_dev *ntb)
> if (ntb_peer_port_count(ntb) != 1)
> dev_warn(&ntb->dev, "multi-port NTB is unsupported\n");
>
> + if (ntb_spad_count(ntb) < 1) {
> + dev_dbg(&ntb->dev, "no enough scratchpads\n");
> + rc = -EINVAL;
> + goto err_tc;
> + }
> +
> + if (!ntb->ops->mw_set_trans) {
> + dev_dbg(&ntb->dev, "need inbound MW based NTB API\n");
> + rc = -EINVAL;
> + goto err_tc;
> + }
> +
> tc = kzalloc(sizeof(*tc), GFP_KERNEL);
> if (!tc) {
> rc = -ENOMEM;
> diff --git a/include/linux/ntb.h b/include/linux/ntb.h
> index fb78663..a6bf15d 100644
> --- a/include/linux/ntb.h
> +++ b/include/linux/ntb.h
> @@ -274,13 +274,14 @@ struct ntb_dev_ops {
> int (*spad_is_unsafe)(struct ntb_dev *ntb);
> int (*spad_count)(struct ntb_dev *ntb);
>
> - u32 (*spad_read)(struct ntb_dev *ntb, int idx);
> - int (*spad_write)(struct ntb_dev *ntb, int idx, u32 val);
> + u32 (*spad_read)(struct ntb_dev *ntb, int sidx);
> + int (*spad_write)(struct ntb_dev *ntb, int sidx, u32 val);
>
> - int (*peer_spad_addr)(struct ntb_dev *ntb, int idx,
> + int (*peer_spad_addr)(struct ntb_dev *ntb, int pidx, int sidx,
> phys_addr_t *spad_addr);
> - u32 (*peer_spad_read)(struct ntb_dev *ntb, int idx);
> - int (*peer_spad_write)(struct ntb_dev *ntb, int idx, u32 val);
> + u32 (*peer_spad_read)(struct ntb_dev *ntb, int pidx, int sidx);
> + int (*peer_spad_write)(struct ntb_dev *ntb, int pidx, int sidx,
> + u32 val);
> };
>
> static inline int ntb_dev_ops_is_valid(const struct ntb_dev_ops *ops)
> @@ -322,13 +323,12 @@ static inline int ntb_dev_ops_is_valid(const struct ntb_dev_ops
> *ops)
> /* ops->peer_db_read_mask && */
> /* ops->peer_db_set_mask && */
> /* ops->peer_db_clear_mask && */
> - /* ops->spad_is_unsafe && */
> - ops->spad_count &&
> - ops->spad_read &&
> - ops->spad_write &&
> - /* ops->peer_spad_addr && */
> - /* ops->peer_spad_read && */
> - ops->peer_spad_write &&
> + /* !ops->spad_is_unsafe == !ops->spad_count && */
> + !ops->spad_read == !ops->spad_count &&
> + !ops->spad_write == !ops->spad_count &&
> + /* !ops->peer_spad_addr == !ops->spad_count && */
> + /* !ops->peer_spad_read == !ops->spad_count && */
> + !ops->peer_spad_write == !ops->spad_count &&
> 1;
> }
>
> @@ -1087,51 +1087,62 @@ static inline int ntb_spad_is_unsafe(struct ntb_dev *ntb)
> }
>
> /**
> - * ntb_mw_count() - get the number of scratchpads
> + * ntb_spad_count() - get the number of scratchpads
> * @ntb: NTB device context.
> *
> * Hardware and topology may support a different number of scratchpads.
> + * Although it must be the same for all ports per NTB device.
> *
> * Return: the number of scratchpads.
> */
> static inline int ntb_spad_count(struct ntb_dev *ntb)
> {
> + if (!ntb->ops->spad_count)
> + return 0;
> +
> return ntb->ops->spad_count(ntb);
> }
>
> /**
> * ntb_spad_read() - read the local scratchpad register
> * @ntb: NTB device context.
> - * @idx: Scratchpad index.
> + * @sidx: Scratchpad index.
> *
> * Read the local scratchpad register, and return the value.
> *
> * Return: The value of the local scratchpad register.
> */
> -static inline u32 ntb_spad_read(struct ntb_dev *ntb, int idx)
> +static inline u32 ntb_spad_read(struct ntb_dev *ntb, int sidx)
> {
> - return ntb->ops->spad_read(ntb, idx);
> + if (!ntb->ops->spad_read)
> + return ~(u32)0;
> +
> + return ntb->ops->spad_read(ntb, sidx);
> }
>
> /**
> * ntb_spad_write() - write the local scratchpad register
> * @ntb: NTB device context.
> - * @idx: Scratchpad index.
> + * @sidx: Scratchpad index.
> * @val: Scratchpad value.
> *
> * Write the value to the local scratchpad register.
> *
> * Return: Zero on success, otherwise an error number.
> */
> -static inline int ntb_spad_write(struct ntb_dev *ntb, int idx, u32 val)
> +static inline int ntb_spad_write(struct ntb_dev *ntb, int sidx, u32 val)
> {
> - return ntb->ops->spad_write(ntb, idx, val);
> + if (!ntb->ops->spad_write)
> + return -EINVAL;
> +
> + return ntb->ops->spad_write(ntb, sidx, val);
> }
>
> /**
> * ntb_peer_spad_addr() - address of the peer scratchpad register
> * @ntb: NTB device context.
> - * @idx: Scratchpad index.
> + * @pidx: Port index of peer device.
> + * @sidx: Scratchpad index.
> * @spad_addr: OUT - The address of the peer scratchpad register.
> *
> * Return the address of the peer doorbell register. This may be used, for
> @@ -1139,42 +1150,51 @@ static inline int ntb_spad_write(struct ntb_dev *ntb, int idx, u32
> val)
> *
> * Return: Zero on success, otherwise an error number.
> */
> -static inline int ntb_peer_spad_addr(struct ntb_dev *ntb, int idx,
> +static inline int ntb_peer_spad_addr(struct ntb_dev *ntb, int pidx, int sidx,
> phys_addr_t *spad_addr)
> {
> if (!ntb->ops->peer_spad_addr)
> return -EINVAL;
>
> - return ntb->ops->peer_spad_addr(ntb, idx, spad_addr);
> + return ntb->ops->peer_spad_addr(ntb, pidx, sidx, spad_addr);
> }
>
> /**
> * ntb_peer_spad_read() - read the peer scratchpad register
> * @ntb: NTB device context.
> - * @idx: Scratchpad index.
> + * @pidx: Port index of peer device.
> + * @sidx: Scratchpad index.
> *
> * Read the peer scratchpad register, and return the value.
> *
> * Return: The value of the local scratchpad register.
> */
> -static inline u32 ntb_peer_spad_read(struct ntb_dev *ntb, int idx)
> +static inline u32 ntb_peer_spad_read(struct ntb_dev *ntb, int pidx, int sidx)
> {
> - return ntb->ops->peer_spad_read(ntb, idx);
> + if (!ntb->ops->peer_spad_read)
> + return ~(u32)0;
> +
> + return ntb->ops->peer_spad_read(ntb, pidx, sidx);
> }
>
> /**
> * ntb_peer_spad_write() - write the peer scratchpad register
> * @ntb: NTB device context.
> - * @idx: Scratchpad index.
> + * @pidx: Port index of peer device.
> + * @sidx: Scratchpad index.
> * @val: Scratchpad value.
> *
> * Write the value to the peer scratchpad register.
> *
> * Return: Zero on success, otherwise an error number.
> */
> -static inline int ntb_peer_spad_write(struct ntb_dev *ntb, int idx, u32 val)
> +static inline int ntb_peer_spad_write(struct ntb_dev *ntb, int pidx, int sidx,
> + u32 val)
> {
> - return ntb->ops->peer_spad_write(ntb, idx, val);
> + if (!ntb->ops->peer_spad_write)
> + return -EINVAL;
> +
> + return ntb->ops->peer_spad_write(ntb, pidx, sidx, val);
> }
>
> #endif
> --
> 2.6.6