D28357.id82928.diff
No OneTemporary
Actions

Size

53 KB

Referenced Files

None

Subscribers

None

D28357.id82928.diff
View Options

	Index: sys/dev/mlx5/mlx5_en/en_rl.h
	===================================================================
	--- sys/dev/mlx5/mlx5_en/en_rl.h
	+++ sys/dev/mlx5/mlx5_en/en_rl.h
	@@ -44,6 +44,9 @@
	#define MLX5E_RL_DEF_SQ_PER_WORKER (12 * 1024) /* software limit */
	#define MLX5E_RL_MAX_SQS (120 * 1024) /* software limit */

	+#define MLX5E_CX4_LIMIT 55000 /* Limit for mlx5 connect-x 4 */
	+#define MLX5E_CX6_LIMIT MLX5E_CX4_LIMIT /* same as 4 for now */
	+
	#define MLX5E_RL_TX_COAL_USEC_DEFAULT 32
	#define MLX5E_RL_TX_COAL_PKTS_DEFAULT 4
	#define MLX5E_RL_TX_COAL_MODE_DEFAULT 0
	@@ -121,6 +124,8 @@
	MLX5E_RL_PARAMS(MLX5E_STATS_VAR)
	u64 table_arg [0];
	MLX5E_RL_TABLE_PARAMS(MLX5E_STATS_VAR)
	+ u64 table_rates[MLX5E_RL_MAX_TX_RATES];
	+ u32 num_rates;
	};

	struct mlx5e_rl_channel_param {
	@@ -166,6 +171,8 @@
	int mlx5e_rl_init(struct mlx5e_priv *priv);
	void mlx5e_rl_cleanup(struct mlx5e_priv *priv);
	void mlx5e_rl_refresh_sq_inline(struct mlx5e_rl_priv_data *rl);
	+void mlx5e_rl_if_query(struct ifnet , struct if_ratelimit_query_results );
	+int mlx5e_rl_if_setup(struct ifnet *, uint64_t, uint32_t);

	if_snd_tag_alloc_t mlx5e_rl_snd_tag_alloc;
	if_snd_tag_modify_t mlx5e_rl_snd_tag_modify;
	Index: sys/dev/mlx5/mlx5_en/mlx5_en_main.c
	===================================================================
	--- sys/dev/mlx5/mlx5_en/mlx5_en_main.c
	+++ sys/dev/mlx5/mlx5_en/mlx5_en_main.c
	@@ -4246,50 +4246,6 @@
	}
	}

	-#ifdef RATELIMIT
	-#define NUM_HDWR_RATES_MLX 13
	-static const uint64_t adapter_rates_mlx[NUM_HDWR_RATES_MLX] = {
	- 135375, /* 1,083,000 */
	- 180500, /* 1,444,000 */
	- 270750, /* 2,166,000 */
	- 361000, /* 2,888,000 */
	- 541500, /* 4,332,000 */
	- 721875, /* 5,775,000 */
	- 1082875, /* 8,663,000 */
	- 1443875, /* 11,551,000 */
	- 2165750, /* 17,326,000 */
	- 2887750, /* 23,102,000 */
	- 4331625, /* 34,653,000 */
	- 5775500, /* 46,204,000 */
	- 8663125 /* 69,305,000 */
	-};
	-
	-static void
	-mlx5e_ratelimit_query(struct ifnet ifp __unused, struct if_ratelimit_query_results q)
	-{
	- /*
	- * This function needs updating by the driver maintainer!
	- * For the MLX card there are currently (ConectX-4?) 13
	- * pre-set rates and others i.e. ConnectX-5, 6, 7??
	- *
	- * This will change based on later adapters
	- * and this code should be updated to look at ifp
	- * and figure out the specific adapter type
	- * settings i.e. how many rates as well
	- * as if they are fixed (as is shown here) or
	- * if they are dynamic (example chelsio t4). Also if there
	- * is a maximum number of flows that the adapter
	- * can handle that too needs to be updated in
	- * the max_flows field.
	- */
	- q->rate_table = adapter_rates_mlx;
	- q->flags = RT_IS_FIXED_TABLE;
	- q->max_flows = 0; /* mlx has no limit */
	- q->number_of_rates = NUM_HDWR_RATES_MLX;
	- q->min_segment_burst = 1;
	-}
	-#endif
	-
	static void
	mlx5e_snd_tag_free(struct m_snd_tag *pmt)
	{
	@@ -4407,7 +4363,8 @@
	ifp->if_snd_tag_modify = mlx5e_snd_tag_modify;
	ifp->if_snd_tag_query = mlx5e_snd_tag_query;
	#ifdef RATELIMIT
	- ifp->if_ratelimit_query = mlx5e_ratelimit_query;
	+ ifp->if_ratelimit_query = mlx5e_rl_if_query;
	+ ifp->if_ratelimit_setup = mlx5e_rl_if_setup;
	#endif
	/* set TSO limits so that we don't have to drop TX packets */
	ifp->if_hw_tsomax = MLX5E_MAX_TX_PAYLOAD_SIZE - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
	Index: sys/dev/mlx5/mlx5_en/mlx5_en_rl.c
	===================================================================
	--- sys/dev/mlx5/mlx5_en/mlx5_en_rl.c
	+++ sys/dev/mlx5/mlx5_en/mlx5_en_rl.c
	@@ -36,8 +36,122 @@
	struct sysctl_oid , const char name, const char *desc);
	static void mlx5e_rl_sysctl_add_stats_u64_oid(struct mlx5e_rl_priv_data *rl, unsigned x,
	struct sysctl_oid node, const char name, const char *desc);
	-static int mlx5e_rl_tx_limit_add(struct mlx5e_rl_priv_data *, uint64_t value);
	-static int mlx5e_rl_tx_limit_clr(struct mlx5e_rl_priv_data *, uint64_t value);
	+static int mlx5e_rl_tx_limit_add(struct mlx5e_rl_priv_data *, uint64_t value, uint32_t index);
	+static int mlx5e_rl_tx_limit_clr(struct mlx5e_rl_priv_data *, uint64_t value, uint32_t index);
	+
	+#define MLX5E_RL_POW_OF_2(arg) ( \
	+ (((uint64_t)((arg) / 3.200000e+01) & 1) ? 4.29496729600000000000e+09 : 1.0) * \
	+ (((uint64_t)((arg) / 1.600000e+01) & 1) ? 6.55360000000000000000e+04 : 1.0) * \
	+ (((uint64_t)((arg) / 8.000000e+00) & 1) ? 2.56000000000000000000e+02 : 1.0) * \
	+ (((uint64_t)((arg) / 4.000000e+00) & 1) ? 1.60000000000000000000e+01 : 1.0) * \
	+ (((uint64_t)((arg) / 2.000000e+00) & 1) ? 4.00000000000000000000e+00 : 1.0) * \
	+ (((uint64_t)((arg) / 1.000000e+00) & 1) ? 2.00000000000000000000e+00 : 1.0) * \
	+ (((uint64_t)((arg) / 5.000000e-01) & 1) ? 1.41421356237309514547e+00 : 1.0) * \
	+ (((uint64_t)((arg) / 2.500000e-01) & 1) ? 1.18920711500272102690e+00 : 1.0) * \
	+ (((uint64_t)((arg) / 1.250000e-01) & 1) ? 1.09050773266525768967e+00 : 1.0) * \
	+ (((uint64_t)((arg) / 6.250000e-02) & 1) ? 1.04427378242741375480e+00 : 1.0) * \
	+ (((uint64_t)((arg) / 3.125000e-02) & 1) ? 1.02189714865411662714e+00 : 1.0) * \
	+ (((uint64_t)((arg) / 1.562500e-02) & 1) ? 1.01088928605170047526e+00 : 1.0) * \
	+ (((uint64_t)((arg) / 7.812500e-03) & 1) ? 1.00542990111280272636e+00 : 1.0) * \
	+ (((uint64_t)((arg) / 3.906250e-03) & 1) ? 1.00271127505020252180e+00 : 1.0) \
	+)
	+
	+#define MLX5E_RL_RATE_MIN 100000.0 /* 100Kbit/s */
	+#define MLX5E_RL_RATE_MAX 10000000000.0 /* 10Gbit/s */
	+#define MLX5E_RL_RATE_DIFF_LOG2 1.66096404744368122408e+01 /* log(max / min) / log(2) */
	+
	+#define MLX5E_RL_RATES_7(m,b,n) \
	+ m(0 + (b), n) \
	+ m(1 + (b), n) \
	+ m(2 + (b), n) \
	+ m(3 + (b), n) \
	+ m(4 + (b), n) \
	+ m(5 + (b), n) \
	+ m(6 + (b), n)
	+
	+#define MLX5E_RL_RATES_15(m,b,n) \
	+ MLX5E_RL_RATES_7(m,0+(b),n) \
	+ MLX5E_RL_RATES_7(m,7+(b),n) \
	+ m(14 + (b), n)
	+
	+#define MLX5E_RL_RATES_31(m,b,n) \
	+ MLX5E_RL_RATES_15(m,0+(b),n) \
	+ MLX5E_RL_RATES_15(m,15+(b),n) \
	+ m(30 + (b), n)
	+
	+#define MLX5E_RL_RATES_63(m,b,n) \
	+ MLX5E_RL_RATES_31(m,0+(b),n) \
	+ MLX5E_RL_RATES_31(m,31+(b),n) \
	+ m(62 + (b), n)
	+
	+#define MLX5E_RL_RATES_127(m,b,n) \
	+ MLX5E_RL_RATES_63(m,0+(b),n) \
	+ MLX5E_RL_RATES_63(m,63+(b),n) \
	+ m(126 + (b), n)
	+
	+#define MLX5E_RL_RATE_ENTRY(b, n) \
	+ (uint64_t)((MLX5E_RL_RATE_MIN * MLX5E_RL_POW_OF_2((b) * \
	+ MLX5E_RL_RATE_DIFF_LOG2 / (n))) / 8.0),
	+
	+#define MLX5E_RL_GENERATE_RATE_TABLE(num) \
	+static const uint64_t mlx5e_rl_rates_##num[num] = { \
	+ MLX5E_RL_RATES_##num(MLX5E_RL_RATE_ENTRY,0,num) \
	+}
	+
	+/* ConnectX-4 */
	+static const uint64_t mlx5e_rl_rates_13[13] = {
	+ 135375, /* 1,083,000 */
	+ 180500, /* 1,444,000 */
	+ 270750, /* 2,166,000 */
	+ 361000, /* 2,888,000 */
	+ 541500, /* 4,332,000 */
	+ 721875, /* 5,775,000 */
	+ 1082875, /* 8,663,000 */
	+ 1443875, /* 11,551,000 */
	+ 2165750, /* 17,326,000 */
	+ 2887750, /* 23,102,000 */
	+ 4331625, /* 34,653,000 */
	+ 5775500, /* 46,204,000 */
	+ 8663125 /* 69,305,000 */
	+};
	+
	+MLX5E_RL_GENERATE_RATE_TABLE(15);
	+MLX5E_RL_GENERATE_RATE_TABLE(31);
	+MLX5E_RL_GENERATE_RATE_TABLE(63);
	+MLX5E_RL_GENERATE_RATE_TABLE(127);
	+
	+void
	+mlx5e_rl_if_query(struct ifnet ifp, struct if_ratelimit_query_results q)
	+{
	+ struct mlx5e_priv *priv = ifp->if_softc;
	+
	+ if (priv->rl.param.num_rates <= 13) {
	+ /* ConnectX-4 */
	+ q->rate_table = priv->rl.param.table_rates;
	+ q->number_of_rates = priv->rl.param.num_rates;
	+ q->flags = RT_IS_FIXED_TABLE;
	+ q->max_flows = MLX5E_CX4_LIMIT;
	+ q->min_segment_burst = priv->rl.param.tx_burst_size;
	+ } else {
	+ /* ConnectX-5 and newer */
	+ q->rate_table = NULL;
	+ q->number_of_rates = priv->rl.param.tx_rates_def;
	+ q->flags = RT_IS_SELECTABLE \| RT_IS_SETUP_REQ;
	+ q->max_flows = MLX5E_CX6_LIMIT;
	+ q->min_segment_burst = priv->rl.param.tx_burst_size;
	+ }
	+}
	+
	+int
	+mlx5e_rl_if_setup(struct ifnet *ifp, uint64_t rate, uint32_t index)
	+{
	+ struct mlx5e_priv *priv = ifp->if_softc;
	+
	+ if (index == -1U)
	+ return (ERANGE);
	+ else
	+ return (mlx5e_rl_tx_limit_add(&priv->rl, 8ULL * rate, index));
	+}

	static void
	mlx5e_rl_build_sq_param(struct mlx5e_rl_priv_data *rl,
	@@ -701,9 +815,12 @@
	*/
	param->tx_allowed_deviation_max = -1ULL /
	param->tx_limit_max;
	+ param->tx_allowed_deviation =
	+ param->tx_allowed_deviation_max;
	+ } else {
	+ /* set default rate deviation */
	+ param->tx_allowed_deviation = 50; /* 5.0% */
	}
	- /* set default rate deviation */
	- param->tx_allowed_deviation = 50; /* 5.0% */

	/* channel parameters */
	param->tx_queue_size = (1 << MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE);
	@@ -725,6 +842,33 @@
	MLX5E_RL_STATS(MLX5E_STATS_DESC)
	};

	+static void
	+mlx5e_rl_setup_default_rates(struct mlx5e_rl_priv_data *rl,
	+ const u64 *src, uint32_t num)
	+{
	+
	+
	+ MPASS(num <= MLX5E_RL_MAX_TX_RATES);
	+
	+ /* check minimum rate */
	+ while (num != 0 && src[0] < (rl->param.tx_limit_min / 8)) {
	+ num--;
	+ src++;
	+ }
	+
	+ /* check maximum rate */
	+ while (num != 0 && src[num - 1] > (rl->param.tx_limit_max / 8))
	+ num--;
	+
	+ /* check we are not exceeding the number of rates supported */
	+ if (num > rl->param.tx_rates_def)
	+ num = rl->param.tx_rates_def;
	+
	+ /* install default rate table */
	+ memcpy(rl->param.table_rates, src, sizeof(src[0]) * num);
	+ rl->param.num_rates = num;
	+}
	+
	int
	mlx5e_rl_init(struct mlx5e_priv *priv)
	{
	@@ -754,6 +898,19 @@
	/* setup default value for parameters */
	mlx5e_rl_set_default_params(&rl->param, priv->mdev);

	+ /* setup rate table */
	+ if (rl->param.tx_rates_def >= 127) {
	+ mlx5e_rl_setup_default_rates(rl, mlx5e_rl_rates_127, 127);
	+ } else if (rl->param.tx_rates_def >= 63) {
	+ mlx5e_rl_setup_default_rates(rl, mlx5e_rl_rates_63, 63);
	+ } else if (rl->param.tx_rates_def >= 31) {
	+ mlx5e_rl_setup_default_rates(rl, mlx5e_rl_rates_31, 31);
	+ } else if (rl->param.tx_rates_def >= 15) {
	+ mlx5e_rl_setup_default_rates(rl, mlx5e_rl_rates_15, 15);
	+ } else {
	+ mlx5e_rl_setup_default_rates(rl, mlx5e_rl_rates_13, 13);
	+ }
	+
	/* update the completion factor */
	mlx5e_rl_sync_tx_completion_fact(rl);

	@@ -805,7 +962,7 @@
	snprintf(buf, sizeof(buf), "dev.mce.%d.rate_limit.tx_rate_add_%d",
	device_get_unit(priv->mdev->pdev->dev.bsddev), (int)i);
	if (TUNABLE_QUAD_FETCH(buf, &j))
	- mlx5e_rl_tx_limit_add(rl, j);
	+ mlx5e_rl_tx_limit_add(rl, j, -1U);
	}

	/* setup rate table sysctls */
	@@ -1252,37 +1409,48 @@
	}
	}
	}
	-
	static int
	-mlx5e_rl_tx_limit_add(struct mlx5e_rl_priv_data *rl, uint64_t value)
	+mlx5e_rl_tx_limit_add(struct mlx5e_rl_priv_data *rl, uint64_t value, uint32_t index)
	{
	unsigned x;
	int error;

	- if (value < 1000 \|\|
	+ if (value < 1000 \|\| (index != -1U && index >= rl->param.tx_rates_def) \|\|
	mlx5_rl_is_in_range(rl->priv->mdev, howmany(value, 1000), 0) == 0)
	- return (EINVAL);
	+ return (ERANGE);

	MLX5E_RL_WLOCK(rl);
	- error = ENOMEM;

	- /* check if rate already exists */
	- for (x = 0; x != rl->param.tx_rates_def; x++) {
	- if (rl->rate_limit_table[x] != value)
	- continue;
	- error = EEXIST;
	- break;
	- }
	+ if (index != -1U) {
	+ error = 0;
	+
	+ /* simply update the rate */
	+ rl->rate_limit_table[index] = value;
	+ if (index < rl->param.num_rates)
	+ rl->param.table_rates[index] = value / 8ULL;
	+ } else {
	+ error = ENOMEM;

	- /* check if there is a free rate entry */
	- if (x == rl->param.tx_rates_def) {
	+ /* check if rate already exists */
	for (x = 0; x != rl->param.tx_rates_def; x++) {
	- if (rl->rate_limit_table[x] != 0)
	+ if (rl->rate_limit_table[x] != value)
	continue;
	- rl->rate_limit_table[x] = value;
	- error = 0;
	+ error = EEXIST;
	break;
	}
	+
	+ /* check if there is a free rate entry */
	+ if (x == rl->param.tx_rates_def) {
	+ for (x = 0; x != rl->param.tx_rates_def; x++) {
	+ if (rl->rate_limit_table[x] != 0)
	+ continue;
	+ rl->rate_limit_table[x] = value;
	+ if (x < rl->param.num_rates)
	+ rl->param.table_rates[x] = value / 8ULL;
	+ error = 0;
	+ break;
	+ }
	+ }
	}
	MLX5E_RL_WUNLOCK(rl);

	@@ -1290,30 +1458,39 @@
	}

	static int
	-mlx5e_rl_tx_limit_clr(struct mlx5e_rl_priv_data *rl, uint64_t value)
	+mlx5e_rl_tx_limit_clr(struct mlx5e_rl_priv_data *rl, uint64_t value, uint32_t index)
	{
	unsigned x;
	int error;

	- if (value == 0)
	- return (EINVAL);
	+ if (value == 0 \|\| (index != -1U && index >= rl->param.tx_rates_def))
	+ return (ERANGE);

	MLX5E_RL_WLOCK(rl);
	-
	- /* check if rate already exists */
	- for (x = 0; x != rl->param.tx_rates_def; x++) {
	- if (rl->rate_limit_table[x] != value)
	- continue;
	+ if (index != -1U) {
	+ error = 0;
	/* free up rate */
	- rl->rate_limit_table[x] = 0;
	- break;
	- }
	+ rl->rate_limit_table[index] = 0;
	+ if (index < rl->param.num_rates)
	+ rl->param.table_rates[index] = 0;
	+ } else {
	+ /* check if rate already exists */
	+ for (x = 0; x != rl->param.tx_rates_def; x++) {
	+ if (rl->rate_limit_table[x] != value)
	+ continue;
	+ /* free up rate */
	+ rl->rate_limit_table[x] = 0;
	+ if (x < rl->param.num_rates)
	+ rl->param.table_rates[x] = 0;
	+ break;
	+ }

	- /* check if there is a free rate entry */
	- if (x == rl->param.tx_rates_def)
	- error = ENOENT;
	- else
	- error = 0;
	+ /* check if there is a free rate entry */
	+ if (x == rl->param.tx_rates_def)
	+ error = ENOENT;
	+ else
	+ error = 0;
	+ }
	MLX5E_RL_WUNLOCK(rl);

	return (error);
	@@ -1479,11 +1656,11 @@
	break;

	case MLX5E_RL_PARAMS_INDEX(tx_limit_add):
	- error = mlx5e_rl_tx_limit_add(rl, value);
	+ error = mlx5e_rl_tx_limit_add(rl, value, -1U);
	break;

	case MLX5E_RL_PARAMS_INDEX(tx_limit_clr):
	- error = mlx5e_rl_tx_limit_clr(rl, value);
	+ error = mlx5e_rl_tx_limit_clr(rl, value, -1U);
	break;

	case MLX5E_RL_PARAMS_INDEX(tx_allowed_deviation):
	Index: sys/net/if_lagg.c
	===================================================================
	--- sys/net/if_lagg.c
	+++ sys/net/if_lagg.c
	@@ -151,6 +151,7 @@
	static int lagg_snd_tag_query(struct m_snd_tag *,
	union if_snd_tag_query_params *);
	static void lagg_snd_tag_free(struct m_snd_tag *);
	+static struct m_snd_tag lagg_next_snd_tag(struct m_snd_tag );
	static void lagg_ratelimit_query(struct ifnet *,
	struct if_ratelimit_query_results *);
	#endif
	@@ -585,6 +586,7 @@
	ifp->if_snd_tag_modify = lagg_snd_tag_modify;
	ifp->if_snd_tag_query = lagg_snd_tag_query;
	ifp->if_snd_tag_free = lagg_snd_tag_free;
	+ ifp->if_next_snd_tag = lagg_next_snd_tag;
	ifp->if_ratelimit_query = lagg_ratelimit_query;
	#endif
	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
	@@ -1834,6 +1836,15 @@
	return (0);
	}

	+static struct m_snd_tag *
	+lagg_next_snd_tag(struct m_snd_tag *mst)
	+{
	+ struct lagg_snd_tag *lst;
	+
	+ lst = mst_to_lst(mst);
	+ return (lst->tag);
	+}
	+
	static int
	lagg_snd_tag_modify(struct m_snd_tag *mst,
	union if_snd_tag_modify_params *params)
	Index: sys/net/if_var.h
	===================================================================
	--- sys/net/if_var.h
	+++ sys/net/if_var.h
	@@ -278,6 +278,7 @@
	typedef int (if_snd_tag_modify_t)(struct m_snd_tag , union if_snd_tag_modify_params );
	typedef int (if_snd_tag_query_t)(struct m_snd_tag , union if_snd_tag_query_params );
	typedef void (if_snd_tag_free_t)(struct m_snd_tag *);
	+typedef struct m_snd_tag (if_next_send_tag_t)(struct m_snd_tag );
	typedef void (if_ratelimit_query_t)(struct ifnet *,
	struct if_ratelimit_query_results *);
	typedef int (if_ratelimit_setup_t)(struct ifnet *, uint64_t, uint32_t);
	@@ -422,6 +423,7 @@
	if_snd_tag_modify_t *if_snd_tag_modify;
	if_snd_tag_query_t *if_snd_tag_query;
	if_snd_tag_free_t *if_snd_tag_free;
	+ if_next_send_tag_t *if_next_snd_tag;
	if_ratelimit_query_t *if_ratelimit_query;
	if_ratelimit_setup_t *if_ratelimit_setup;

	Index: sys/net/if_vlan.c
	===================================================================
	--- sys/net/if_vlan.c
	+++ sys/net/if_vlan.c
	@@ -295,6 +295,9 @@
	static int vlan_snd_tag_query(struct m_snd_tag *,
	union if_snd_tag_query_params *);
	static void vlan_snd_tag_free(struct m_snd_tag *);
	+static struct m_snd_tag vlan_next_snd_tag(struct m_snd_tag );
	+static void vlan_ratelimit_query(struct ifnet *,
	+ struct if_ratelimit_query_results *);
	#endif
	static void vlan_qflush(struct ifnet *ifp);
	static int vlan_setflag(struct ifnet *ifp, int flag, int status,
	@@ -1071,6 +1074,8 @@
	ifp->if_snd_tag_modify = vlan_snd_tag_modify;
	ifp->if_snd_tag_query = vlan_snd_tag_query;
	ifp->if_snd_tag_free = vlan_snd_tag_free;
	+ ifp->if_next_snd_tag = vlan_next_snd_tag;
	+ ifp->if_ratelimit_query = vlan_ratelimit_query;
	#endif
	ifp->if_flags = VLAN_IFFLAGS;
	ether_ifattach(ifp, eaddr);
	@@ -2073,6 +2078,15 @@
	return (0);
	}

	+static struct m_snd_tag *
	+vlan_next_snd_tag(struct m_snd_tag *mst)
	+{
	+ struct vlan_snd_tag *vst;
	+
	+ vst = mst_to_vst(mst);
	+ return (vst->tag);
	+}
	+
	static int
	vlan_snd_tag_modify(struct m_snd_tag *mst,
	union if_snd_tag_modify_params *params)
	@@ -2102,4 +2116,20 @@
	m_snd_tag_rele(vst->tag);
	free(vst, M_VLAN);
	}
	+
	+static void
	+vlan_ratelimit_query(struct ifnet ifp __unused, struct if_ratelimit_query_results q)
	+{
	+ /*
	+ * For vlan, we have an indirect
	+ * interface. The caller needs to
	+ * get a ratelimit tag on the actual
	+ * interface the flow will go on.
	+ */
	+ q->rate_table = NULL;
	+ q->flags = RT_IS_INDIRECT;
	+ q->max_flows = 0;
	+ q->number_of_rates = 0;
	+}
	+
	#endif
	Index: sys/netinet/tcp_log_buf.h
	===================================================================
	--- sys/netinet/tcp_log_buf.h
	+++ sys/netinet/tcp_log_buf.h
	@@ -221,7 +221,7 @@
	BBR_LOG_SETTINGS_CHG, /* Settings changed for loss response 48 */
	BBR_LOG_SRTT_GAIN_EVENT, /* SRTT gaining -- now not used 49 */
	TCP_LOG_REASS, /* Reassembly buffer logging 50 */
	- TCP_HDWR_TLS, /* TCP Hardware TLS logs 51 */
	+ TCP_HDWR_PACE_SIZE, /* TCP pacing size set (rl and rack uses this) 51 */
	BBR_LOG_HDWR_PACE, /* TCP Hardware pacing log 52 */
	BBR_LOG_TSTMP_VAL, /* Temp debug timestamp validation 53 */
	TCP_LOG_CONNEND, /* End of connection 54 */
	Index: sys/netinet/tcp_ratelimit.h
	===================================================================
	--- sys/netinet/tcp_ratelimit.h
	+++ sys/netinet/tcp_ratelimit.h
	@@ -44,6 +44,8 @@
	const struct tcp_rate_set ptbl; / Pointer to parent table */
	struct m_snd_tag tag; / Send tag if needed (chelsio) */
	uint64_t rate; /* Rate we get in Bytes per second (Bps) */
	+ uint64_t using; /* Temporary -- rrs remove */
	+ uint64_t rs_num_enobufs;
	uint32_t time_between; /* Time-Gap between packets at this rate */
	uint32_t flags;
	};
	@@ -99,22 +101,30 @@
	* shows up in your sysctl tree
	* this can be big.
	*/
	+uint64_t inline
	+tcp_hw_highest_rate(const struct tcp_hwrate_limit_table *rle)
	+{
	+ return (rle->ptbl->rs_rlt[rle->ptbl->rs_highest_valid].rate);
	+}
	+
	+uint64_t
	+tcp_hw_highest_rate_ifp(struct ifnet ifp, struct inpcb inp);

	const struct tcp_hwrate_limit_table *
	tcp_set_pacing_rate(struct tcpcb tp, struct ifnet ifp,
	- uint64_t bytes_per_sec, int flags, int *error);
	+ uint64_t bytes_per_sec, int flags, int error, uint64_t lower_rate);

	const struct tcp_hwrate_limit_table *
	tcp_chg_pacing_rate(const struct tcp_hwrate_limit_table *crte,
	struct tcpcb tp, struct ifnet ifp,
	- uint64_t bytes_per_sec, int flags, int *error);
	+ uint64_t bytes_per_sec, int flags, int error, uint64_t lower_rate);
	void
	tcp_rel_pacing_rate(const struct tcp_hwrate_limit_table *crte,
	struct tcpcb *tp);
	#else
	static inline const struct tcp_hwrate_limit_table *
	tcp_set_pacing_rate(struct tcpcb tp, struct ifnet ifp,
	- uint64_t bytes_per_sec, int flags, int *error)
	+ uint64_t bytes_per_sec, int flags, int error, uint64_t lower_rate)
	{
	if (error)
	*error = EOPNOTSUPP;
	@@ -124,7 +134,7 @@
	static inline const struct tcp_hwrate_limit_table *
	tcp_chg_pacing_rate(const struct tcp_hwrate_limit_table *crte,
	struct tcpcb tp, struct ifnet ifp,
	- uint64_t bytes_per_sec, int flags, int *error)
	+ uint64_t bytes_per_sec, int flags, int error, uint64_t lower_rate)
	{
	if (error)
	*error = EOPNOTSUPP;
	@@ -147,8 +157,12 @@
	* delayed ack).
	*/
	uint32_t
	-tcp_get_pacing_burst_size(uint64_t bw, uint32_t segsiz, int can_use_1mss,
	+tcp_get_pacing_burst_size(struct tcpcb *tp, uint64_t bw, uint32_t segsiz, int can_use_1mss,
	const struct tcp_hwrate_limit_table te, int err);

	+
	+void
	+tcp_rl_log_enobuf(const struct tcp_hwrate_limit_table *rte);
	+
	#endif
	#endif
	Index: sys/netinet/tcp_ratelimit.c
	===================================================================
	--- sys/netinet/tcp_ratelimit.c
	+++ sys/netinet/tcp_ratelimit.c
	@@ -57,6 +57,8 @@
	#ifdef INET6
	#include <netinet6/tcp6_var.h>
	#endif
	+#include <netinet/tcp_hpts.h>
	+#include <netinet/tcp_log_buf.h>
	#include <netinet/tcp_ratelimit.h>
	#ifndef USECS_IN_SECOND
	#define USECS_IN_SECOND 1000000
	@@ -154,108 +156,77 @@
	180500, /* 1.44Mpbs - rate 2 common rate */
	375000, /* 3Mbps - rate 3 */
	625000, /* 5Mbps - rate 4 */
	- 875000, /* 7Mbps - rate 5 */
	- 1125000, /* 9Mbps - rate 6 */
	- 1375000, /* 11Mbps - rate 7 */
	- 1625000, /* 13Mbps - rate 8 */
	- 2625000, /* 21Mbps - rate 9 */
	- 3875000, /* 31Mbps - rate 10 */
	- 5125000, /* 41Meg - rate 11 */
	- 12500000, /* 100Mbps - rate 12 */
	- 25000000, /* 200Mbps - rate 13 */
	- 50000000, /* 400Mbps - rate 14 */
	- 63750000, /* 51Mbps - rate 15 */
	+ 1250000, /* 10Mbps - rate 5 */
	+ 1875000, /* 15Mbps - rate 6 */
	+ 2500000, /* 20Mbps - rate 7 */
	+ 3125000, /* 25Mbps - rate 8 */
	+ 3750000, /* 30Mbps - rate 9 */
	+ 4375000, /* 35Mbps - rate 10 */
	+ 5000000, /* 40Meg - rate 11 */
	+ 6250000, /* 50Mbps - rate 12 */
	+ 12500000, /* 100Mbps - rate 13 */
	+ 25000000, /* 200Mbps - rate 14 */
	+ 50000000, /* 400Mbps - rate 15 */
	100000000, /* 800Mbps - rate 16 */
	- 1875000, /* 15Mbps - rate 17 */
	- 2125000, /* 17Mbps - rate 18 */
	- 2375000, /* 19Mbps - rate 19 */
	- 2875000, /* 23Mbps - rate 20 */
	- 3125000, /* 25Mbps - rate 21 */
	- 3375000, /* 27Mbps - rate 22 */
	- 3625000, /* 29Mbps - rate 23 */
	- 4125000, /* 33Mbps - rate 24 */
	- 4375000, /* 35Mbps - rate 25 */
	- 4625000, /* 37Mbps - rate 26 */
	- 4875000, /* 39Mbps - rate 27 */
	- 5375000, /* 43Mbps - rate 28 */
	- 5625000, /* 45Mbps - rate 29 */
	- 5875000, /* 47Mbps - rate 30 */
	- 6125000, /* 49Mbps - rate 31 */
	- 6625000, /* 53Mbps - rate 32 */
	- 6875000, /* 55Mbps - rate 33 */
	- 7125000, /* 57Mbps - rate 34 */
	- 7375000, /* 59Mbps - rate 35 */
	- 7625000, /* 61Mbps - rate 36 */
	- 7875000, /* 63Mbps - rate 37 */
	- 8125000, /* 65Mbps - rate 38 */
	- 8375000, /* 67Mbps - rate 39 */
	- 8625000, /* 69Mbps - rate 40 */
	- 8875000, /* 71Mbps - rate 41 */
	- 9125000, /* 73Mbps - rate 42 */
	- 9375000, /* 75Mbps - rate 43 */
	- 9625000, /* 77Mbps - rate 44 */
	- 9875000, /* 79Mbps - rate 45 */
	- 10125000, /* 81Mbps - rate 46 */
	- 10375000, /* 83Mbps - rate 47 */
	- 10625000, /* 85Mbps - rate 48 */
	- 10875000, /* 87Mbps - rate 49 */
	- 11125000, /* 89Mbps - rate 50 */
	- 11375000, /* 91Mbps - rate 51 */
	- 11625000, /* 93Mbps - rate 52 */
	- 11875000, /* 95Mbps - rate 53 */
	- 13125000, /* 105Mbps - rate 54 */
	- 13750000, /* 110Mbps - rate 55 */
	- 14375000, /* 115Mbps - rate 56 */
	- 15000000, /* 120Mbps - rate 57 */
	- 15625000, /* 125Mbps - rate 58 */
	- 16250000, /* 130Mbps - rate 59 */
	- 16875000, /* 135Mbps - rate 60 */
	- 17500000, /* 140Mbps - rate 61 */
	- 18125000, /* 145Mbps - rate 62 */
	- 18750000, /* 150Mbps - rate 64 */
	- 20000000, /* 160Mbps - rate 65 */
	- 21250000, /* 170Mbps - rate 66 */
	- 22500000, /* 180Mbps - rate 67 */
	- 23750000, /* 190Mbps - rate 68 */
	- 26250000, /* 210Mbps - rate 69 */
	- 27500000, /* 220Mbps - rate 70 */
	- 28750000, /* 230Mbps - rate 71 */
	- 30000000, /* 240Mbps - rate 72 */
	- 31250000, /* 250Mbps - rate 73 */
	- 34375000, /* 275Mbps - rate 74 */
	- 37500000, /* 300Mbps - rate 75 */
	- 40625000, /* 325Mbps - rate 76 */
	- 43750000, /* 350Mbps - rate 77 */
	- 46875000, /* 375Mbps - rate 78 */
	- 53125000, /* 425Mbps - rate 79 */
	- 56250000, /* 450Mbps - rate 80 */
	- 59375000, /* 475Mbps - rate 81 */
	- 62500000, /* 500Mbps - rate 82 */
	- 68750000, /* 550Mbps - rate 83 */
	- 75000000, /* 600Mbps - rate 84 */
	- 81250000, /* 650Mbps - rate 85 */
	- 87500000, /* 700Mbps - rate 86 */
	- 93750000, /* 750Mbps - rate 87 */
	- 106250000, /* 850Mbps - rate 88 */
	- 112500000, /* 900Mbps - rate 89 */
	- 125000000, /* 1Gbps - rate 90 */
	- 156250000, /* 1.25Gps - rate 91 */
	- 187500000, /* 1.5Gps - rate 92 */
	- 218750000, /* 1.75Gps - rate 93 */
	- 250000000, /* 2Gbps - rate 94 */
	- 281250000, /* 2.25Gps - rate 95 */
	- 312500000, /* 2.5Gbps - rate 96 */
	- 343750000, /* 2.75Gbps - rate 97 */
	- 375000000, /* 3Gbps - rate 98 */
	- 500000000, /* 4Gbps - rate 99 */
	- 625000000, /* 5Gbps - rate 100 */
	- 750000000, /* 6Gbps - rate 101 */
	- 875000000, /* 7Gbps - rate 102 */
	- 1000000000, /* 8Gbps - rate 103 */
	- 1125000000, /* 9Gbps - rate 104 */
	- 1250000000, /* 10Gbps - rate 105 */
	- 1875000000, /* 15Gbps - rate 106 */
	- 2500000000 /* 20Gbps - rate 107 */
	+ 5625000, /* 45Mbps - rate 17 */
	+ 6875000, /* 55Mbps - rate 19 */
	+ 7500000, /* 60Mbps - rate 20 */
	+ 8125000, /* 65Mbps - rate 21 */
	+ 8750000, /* 70Mbps - rate 22 */
	+ 9375000, /* 75Mbps - rate 23 */
	+ 10000000, /* 80Mbps - rate 24 */
	+ 10625000, /* 85Mbps - rate 25 */
	+ 11250000, /* 90Mbps - rate 26 */
	+ 11875000, /* 95Mbps - rate 27 */
	+ 12500000, /* 100Mbps - rate 28 */
	+ 13750000, /* 110Mbps - rate 29 */
	+ 15000000, /* 120Mbps - rate 30 */
	+ 16250000, /* 130Mbps - rate 31 */
	+ 17500000, /* 140Mbps - rate 32 */
	+ 18750000, /* 150Mbps - rate 33 */
	+ 20000000, /* 160Mbps - rate 34 */
	+ 21250000, /* 170Mbps - rate 35 */
	+ 22500000, /* 180Mbps - rate 36 */
	+ 23750000, /* 190Mbps - rate 37 */
	+ 26250000, /* 210Mbps - rate 38 */
	+ 27500000, /* 220Mbps - rate 39 */
	+ 28750000, /* 230Mbps - rate 40 */
	+ 30000000, /* 240Mbps - rate 41 */
	+ 31250000, /* 250Mbps - rate 42 */
	+ 34375000, /* 275Mbps - rate 43 */
	+ 37500000, /* 300Mbps - rate 44 */
	+ 40625000, /* 325Mbps - rate 45 */
	+ 43750000, /* 350Mbps - rate 46 */
	+ 46875000, /* 375Mbps - rate 47 */
	+ 53125000, /* 425Mbps - rate 48 */
	+ 56250000, /* 450Mbps - rate 49 */
	+ 59375000, /* 475Mbps - rate 50 */
	+ 62500000, /* 500Mbps - rate 51 */
	+ 68750000, /* 550Mbps - rate 52 */
	+ 75000000, /* 600Mbps - rate 53 */
	+ 81250000, /* 650Mbps - rate 54 */
	+ 87500000, /* 700Mbps - rate 55 */
	+ 93750000, /* 750Mbps - rate 56 */
	+ 106250000, /* 850Mbps - rate 57 */
	+ 112500000, /* 900Mbps - rate 58 */
	+ 125000000, /* 1Gbps - rate 59 */
	+ 156250000, /* 1.25Gps - rate 60 */
	+ 187500000, /* 1.5Gps - rate 61 */
	+ 218750000, /* 1.75Gps - rate 62 */
	+ 250000000, /* 2Gbps - rate 63 */
	+ 281250000, /* 2.25Gps - rate 64 */
	+ 312500000, /* 2.5Gbps - rate 65 */
	+ 343750000, /* 2.75Gbps - rate 66 */
	+ 375000000, /* 3Gbps - rate 67 */
	+ 500000000, /* 4Gbps - rate 68 */
	+ 625000000, /* 5Gbps - rate 69 */
	+ 750000000, /* 6Gbps - rate 70 */
	+ 875000000, /* 7Gbps - rate 71 */
	+ 1000000000, /* 8Gbps - rate 72 */
	+ 1125000000, /* 9Gbps - rate 73 */
	+ 1250000000, /* 10Gbps - rate 74 */
	+ 1875000000, /* 15Gbps - rate 75 */
	+ 2500000000 /* 20Gbps - rate 76 */
	};

	#define MAX_HDWR_RATES (sizeof(desired_rates)/sizeof(uint64_t))
	@@ -283,6 +254,10 @@
	static struct mtx rs_mtx;
	uint32_t rs_number_alive;
	uint32_t rs_number_dead;
	+static uint32_t rs_floor_mss = 0;
	+static uint32_t wait_time_floor = 8000; /* 8 ms */
	+static uint32_t rs_hw_floor_mss = 16;
	+static uint32_t num_of_waits_allowed = 1; /* How many time blocks are we willing to wait */

	SYSCTL_NODE(_net_inet_tcp, OID_AUTO, rl, CTLFLAG_RW \| CTLFLAG_MPSAFE, 0,
	"TCP Ratelimit stats");
	@@ -292,6 +267,20 @@
	SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, dead, CTLFLAG_RW,
	&rs_number_dead, 0,
	"Number of interfaces departing from ratelimiting");
	+SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, floor_mss, CTLFLAG_RW,
	+ &rs_floor_mss, 0,
	+ "Number of MSS that will override the normal minimums (0 means don't enforce)");
	+SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, wait_floor, CTLFLAG_RW,
	+ &wait_time_floor, 2000,
	+ "Has b/w increases what is the wait floor we are willing to wait at the end?");
	+SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, time_blocks, CTLFLAG_RW,
	+ &num_of_waits_allowed, 1,
	+ "How many time blocks on the end should software pacing be willing to wait?");
	+
	+SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, hw_floor_mss, CTLFLAG_RW,
	+ &rs_hw_floor_mss, 16,
	+ "Number of mss that are a minum for hardware pacing?");
	+

	static void
	rl_add_syctl_entries(struct sysctl_oid rl_sysctl_root, struct tcp_rate_set rs)
	@@ -383,6 +372,17 @@
	OID_AUTO, "rate", CTLFLAG_RD,
	&rs->rs_rlt[i].rate, 0,
	"Rate in bytes per second");
	+ SYSCTL_ADD_U64(&rs->sysctl_ctx,
	+ SYSCTL_CHILDREN(rl_rate_num),
	+ OID_AUTO, "using", CTLFLAG_RD,
	+ &rs->rs_rlt[i].using, 0,
	+ "Number of flows using");
	+ SYSCTL_ADD_U64(&rs->sysctl_ctx,
	+ SYSCTL_CHILDREN(rl_rate_num),
	+ OID_AUTO, "enobufs", CTLFLAG_RD,
	+ &rs->rs_rlt[i].rs_num_enobufs, 0,
	+ "Number of enobufs logged on this rate");
	+
	}
	}
	#endif
	@@ -443,10 +443,14 @@
	}

	#ifdef INET
	+#ifdef NETFLIX_STATS
	+extern counter_u64_t rate_limit_new;
	+extern counter_u64_t rate_limit_chg;
	extern counter_u64_t rate_limit_set_ok;
	extern counter_u64_t rate_limit_active;
	extern counter_u64_t rate_limit_alloc_fail;
	#endif
	+#endif

	static int
	rl_attach_txrtlmt(struct ifnet *ifp,
	@@ -466,11 +470,13 @@

	error = m_snd_tag_alloc(ifp, &params, tag);
	#ifdef INET
	+#ifdef NETFLIX_STATS
	if (error == 0) {
	counter_u64_add(rate_limit_set_ok, 1);
	counter_u64_add(rate_limit_active, 1);
	} else if (error != EOPNOTSUPP)
	counter_u64_add(rate_limit_alloc_fail, 1);
	+#endif
	#endif
	return (error);
	}
	@@ -519,6 +525,7 @@
	int i;
	struct if_ratelimit_query_results rl;
	struct sysctl_oid *rl_sysctl_root;
	+ struct epoch_tracker et;
	/*
	* We expect to enter with the
	* mutex locked.
	@@ -562,9 +569,11 @@
	CTLFLAG_RW \| CTLFLAG_MPSAFE, 0,
	"");
	rl_add_syctl_entries(rl_sysctl_root, rs);
	+ NET_EPOCH_ENTER(et);
	mtx_lock(&rs_mtx);
	CK_LIST_INSERT_HEAD(&int_rs, rs, next);
	mtx_unlock(&rs_mtx);
	+ NET_EPOCH_EXIT(et);
	return (rs);
	} else if ((rl.flags & RT_IS_INDIRECT) == RT_IS_INDIRECT) {
	memset(rs, 0, sizeof(struct tcp_rate_set));
	@@ -580,9 +589,11 @@
	CTLFLAG_RW \| CTLFLAG_MPSAFE, 0,
	"");
	rl_add_syctl_entries(rl_sysctl_root, rs);
	+ NET_EPOCH_ENTER(et);
	mtx_lock(&rs_mtx);
	CK_LIST_INSERT_HEAD(&int_rs, rs, next);
	mtx_unlock(&rs_mtx);
	+ NET_EPOCH_EXIT(et);
	return (rs);
	} else if ((rl.flags & RT_IS_FIXED_TABLE) == RT_IS_FIXED_TABLE) {
	/* Mellanox C4 likely */
	@@ -671,6 +682,8 @@
	*/
	rs->rs_rlt[i].ptbl = rs;
	rs->rs_rlt[i].tag = NULL;
	+ rs->rs_rlt[i].using = 0;
	+ rs->rs_rlt[i].rs_num_enobufs = 0;
	/*
	* Calculate the time between.
	*/
	@@ -741,18 +754,24 @@
	CTLFLAG_RW \| CTLFLAG_MPSAFE, 0,
	"");
	rl_add_syctl_entries(rl_sysctl_root, rs);
	+ NET_EPOCH_ENTER(et);
	mtx_lock(&rs_mtx);
	CK_LIST_INSERT_HEAD(&int_rs, rs, next);
	mtx_unlock(&rs_mtx);
	+ NET_EPOCH_EXIT(et);
	return (rs);
	}

	+/*
	+ * For an explanation of why the argument is volatile please
	+ * look at the comments around rt_setup_rate().
	+ */
	static const struct tcp_hwrate_limit_table *
	-tcp_int_find_suitable_rate(const struct tcp_rate_set *rs,
	- uint64_t bytes_per_sec, uint32_t flags)
	+tcp_int_find_suitable_rate(const volatile struct tcp_rate_set *rs,
	+ uint64_t bytes_per_sec, uint32_t flags, uint64_t *lower_rate)
	{
	struct tcp_hwrate_limit_table arte = NULL, rte = NULL;
	- uint64_t mbits_per_sec, ind_calc;
	+ uint64_t mbits_per_sec, ind_calc, previous_rate = 0;
	int i;

	mbits_per_sec = (bytes_per_sec * 8);
	@@ -763,6 +782,7 @@
	* Smaller than 1Meg, only
	* 3 entries can match it.
	*/
	+ previous_rate = 0;
	for(i = rs->rs_lowest_valid; i < 3; i++) {
	if (bytes_per_sec <= rs->rs_rlt[i].rate) {
	rte = &rs->rs_rlt[i];
	@@ -770,6 +790,7 @@
	} else if (rs->rs_rlt[i].flags & HDWRPACE_INITED) {
	arte = &rs->rs_rlt[i];
	}
	+ previous_rate = rs->rs_rlt[i].rate;
	}
	goto done;
	} else if ((mbits_per_sec > RS_ONE_GIGABIT_PERSEC) &&
	@@ -782,6 +803,7 @@
	rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)];
	else
	arte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)];
	+ previous_rate = rs->rs_rlt[(ALL_HARDWARE_RATES-2)].rate;
	goto done;
	}
	/*
	@@ -800,8 +822,11 @@
	ind_calc = ALL_HARDWARE_RATES-1;
	}
	if ((ind_calc >= rs->rs_lowest_valid) &&
	- (ind_calc <= rs->rs_highest_valid))
	- rte = &rs->rs_rlt[ind_calc];
	+ (ind_calc <= rs->rs_highest_valid)) {
	+ rte = &rs->rs_rlt[ind_calc];
	+ if (ind_calc >= 1)
	+ previous_rate = rs->rs_rlt[(ind_calc-1)].rate;
	+ }
	} else if (flags & RS_PACING_EXACT_MATCH) {
	if ((mbits_per_sec < RS_ONE_MEGABIT_PERSEC) &&
	(rs->rs_lowest_valid <= 2)){
	@@ -840,10 +865,16 @@
	for (i=2; i>=rs->rs_lowest_valid; i--) {
	if (bytes_per_sec < rs->rs_rlt[i].rate) {
	rte = &rs->rs_rlt[i];
	+ if (i >= 1) {
	+ previous_rate = rs->rs_rlt[(i-1)].rate;
	+ }
	break;
	} else if ((flags & RS_PACING_GEQ) &&
	(bytes_per_sec == rs->rs_rlt[i].rate)) {
	rte = &rs->rs_rlt[i];
	+ if (i >= 1) {
	+ previous_rate = rs->rs_rlt[(i-1)].rate;
	+ }
	break;
	} else {
	arte = &rs->rs_rlt[i]; /* new alternate */
	@@ -863,6 +894,7 @@
	/* The top rate is an alternative */
	arte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)];
	}
	+ previous_rate = rs->rs_rlt[(ALL_HARDWARE_RATES-2)].rate;
	} else {
	/* Its in our range 1Meg - 1Gig */
	if (flags & RS_PACING_GEQ) {
	@@ -873,6 +905,8 @@
	ind_calc = (ALL_HARDWARE_RATES-1);
	}
	rte = &rs->rs_rlt[ind_calc];
	+ if (ind_calc >= 1)
	+ previous_rate = rs->rs_rlt[(ind_calc-1)].rate;
	}
	goto done;
	}
	@@ -882,8 +916,11 @@
	/* This should not happen */
	ind_calc = ALL_HARDWARE_RATES-1;
	}
	- if (rs->rs_rlt[ind_calc].flags & HDWRPACE_INITED)
	+ if (rs->rs_rlt[ind_calc].flags & HDWRPACE_INITED) {
	rte = &rs->rs_rlt[ind_calc];
	+ if (ind_calc >= 1)
	+ previous_rate = rs->rs_rlt[(ind_calc-1)].rate;
	+ }
	}
	}
	done:
	@@ -893,11 +930,17 @@
	/* We can use the substitute */
	rte = arte;
	}
	+ if (lower_rate)
	+ *lower_rate = previous_rate;
	return (rte);
	}

	+/*
	+ * For an explanation of why the argument is volatile please
	+ * look at the comments around rt_setup_rate().
	+ */
	static const struct tcp_hwrate_limit_table *
	-tcp_find_suitable_rate(const struct tcp_rate_set *rs, uint64_t bytes_per_sec, uint32_t flags)
	+tcp_find_suitable_rate(const volatile struct tcp_rate_set rs, uint64_t bytes_per_sec, uint32_t flags, uint64_t lower_rate)
	{
	/**
	* Hunt the rate table with the restrictions in flags and find a
	@@ -911,6 +954,7 @@
	*/
	int i, matched;
	struct tcp_hwrate_limit_table *rte = NULL;
	+ uint64_t previous_rate = 0;

	if ((rs->rs_flags & RS_INT_TBL) &&
	(rs->rs_rate_cnt >= ALL_HARDWARE_RATES)) {
	@@ -920,7 +964,7 @@
	* from 1Meg - 1000Meg in 1Meg increments.
	* Use an alternate method to "lookup".
	*/
	- return (tcp_int_find_suitable_rate(rs, bytes_per_sec, flags));
	+ return (tcp_int_find_suitable_rate(rs, bytes_per_sec, flags, lower_rate));
	}
	if ((flags & RS_PACING_LT) \|\|
	(flags & RS_PACING_EXACT_MATCH)) {
	@@ -934,13 +978,18 @@
	(bytes_per_sec == rs->rs_rlt[i].rate)) {
	rte = &rs->rs_rlt[i];
	matched = 1;
	+ if (lower_rate != NULL)
	+ *lower_rate = previous_rate;
	break;
	} else if ((flags & RS_PACING_LT) &&
	(bytes_per_sec <= rs->rs_rlt[i].rate)) {
	rte = &rs->rs_rlt[i];
	matched = 1;
	+ if (lower_rate != NULL)
	+ *lower_rate = previous_rate;
	break;
	}
	+ previous_rate = rs->rs_rlt[i].rate;
	if (bytes_per_sec > rs->rs_rlt[i].rate)
	break;
	}
	@@ -979,6 +1028,8 @@
	* We found a table entry that is smaller,
	* stop there will be none greater or equal.
	*/
	+ if (lower_rate != NULL)
	+ *lower_rate = rs->rs_rlt[i].rate;
	break;
	}
	}
	@@ -995,10 +1046,10 @@
	rt_find_real_interface(struct ifnet ifp, struct inpcb inp, int *error)
	{
	struct ifnet *tifp;
	- struct m_snd_tag *tag;
	+ struct m_snd_tag tag, ntag;
	union if_snd_tag_alloc_params params = {
	.rate_limit.hdr.type = IF_SND_TAG_TYPE_RATE_LIMIT,
	- .rate_limit.hdr.flowid = 1,
	+ .rate_limit.hdr.flowid = inp->inp_flowid,
	.rate_limit.hdr.numa_domain = inp->inp_numa_domain,
	.rate_limit.max_rate = COMMON_RATE,
	.rate_limit.flags = M_NOWAIT,
	@@ -1017,38 +1068,92 @@
	*error = err;
	return (NULL);
	}
	- tifp = tag->ifp;
	+ ntag = tag;
	+ while(ntag->ifp->if_next_snd_tag != NULL) {
	+ ntag = ntag->ifp->if_next_snd_tag(ntag);
	+ }
	+ tifp = ntag->ifp;
	m_snd_tag_rele(tag);
	return (tifp);
	}

	+static void
	+rl_increment_using(const struct tcp_hwrate_limit_table *rte)
	+{
	+ struct tcp_hwrate_limit_table *decon_rte;
	+
	+ decon_rte = __DECONST(struct tcp_hwrate_limit_table *, rte);
	+ atomic_add_long(&decon_rte->using, 1);
	+}
	+
	+static void
	+rl_decrement_using(const struct tcp_hwrate_limit_table *rte)
	+{
	+ struct tcp_hwrate_limit_table *decon_rte;
	+
	+ decon_rte = __DECONST(struct tcp_hwrate_limit_table *, rte);
	+ atomic_subtract_long(&decon_rte->using, 1);
	+}
	+
	+void
	+tcp_rl_log_enobuf(const struct tcp_hwrate_limit_table *rte)
	+{
	+ struct tcp_hwrate_limit_table *decon_rte;
	+
	+ decon_rte = __DECONST(struct tcp_hwrate_limit_table *, rte);
	+ atomic_add_long(&decon_rte->rs_num_enobufs, 1);
	+}
	+
	+/*
	+ * Do NOT take the __noinline out of the
	+ * find_rs_for_ifp() function. If you do the inline
	+ * of it for the rt_setup_rate() will show you a
	+ * compiler bug. For some reason the compiler thinks
	+ * the list can never be empty. The consequence of
	+ * this will be a crash when we dereference NULL
	+ * if an ifp is removed just has a hw rate limit
	+ * is attempted. If you are working on the compiler
	+ * and want to "test" this go ahead and take the noinline
	+ * out otherwise let sleeping dogs ly until such time
	+ * as we get a compiler fix 10/2/20 -- RRS
	+ */
	+static __noinline struct tcp_rate_set *
	+find_rs_for_ifp(struct ifnet *ifp)
	+{
	+ struct tcp_rate_set *rs;
	+
	+ CK_LIST_FOREACH(rs, &int_rs, next) {
	+ if ((rs->rs_ifp == ifp) &&
	+ (rs->rs_if_dunit == ifp->if_dunit)) {
	+ /* Ok we found it */
	+ return (rs);
	+ }
	+ }
	+ return (NULL);
	+}
	+
	+
	static const struct tcp_hwrate_limit_table *
	rt_setup_rate(struct inpcb inp, struct ifnet ifp, uint64_t bytes_per_sec,
	- uint32_t flags, int *error)
	+ uint32_t flags, int error, uint64_t lower_rate)
	{
	/* First lets find the interface if it exists */
	const struct tcp_hwrate_limit_table *rte;
	- struct tcp_rate_set *rs;
	+ /*
	+ * So why is rs volatile? This is to defeat a
	+ * compiler bug where in the compiler is convinced
	+ * that rs can never be NULL (which is not true). Because
	+ * of its conviction it nicely optimizes out the if ((rs == NULL
	+ * below which means if you get a NULL back you dereference it.
	+ */
	+ volatile struct tcp_rate_set *rs;
	struct epoch_tracker et;
	+ struct ifnet *oifp = ifp;
	int err;

	NET_EPOCH_ENTER(et);
	use_real_interface:
	- CK_LIST_FOREACH(rs, &int_rs, next) {
	- /*
	- * Note we don't look with the lock since we either see a
	- * new entry or will get one when we try to add it.
	- */
	- if (rs->rs_flags & RS_IS_DEAD) {
	- /* The dead are not looked at */
	- continue;
	- }
	- if ((rs->rs_ifp == ifp) &&
	- (rs->rs_if_dunit == ifp->if_dunit)) {
	- /* Ok we found it */
	- break;
	- }
	- }
	+ rs = find_rs_for_ifp(ifp);
	if ((rs == NULL) \|\|
	(rs->rs_flags & RS_INTF_NO_SUP) \|\|
	(rs->rs_flags & RS_IS_DEAD)) {
	@@ -1063,14 +1168,14 @@
	* might be arguable, but its impossible
	* to tell from the departing case.
	*/
	- if (rs->rs_disable && error)
	+ if (error)
	*error = ENODEV;
	NET_EPOCH_EXIT(et);
	return (NULL);
	}

	if ((rs == NULL) \|\| (rs->rs_disable != 0)) {
	- if (rs->rs_disable && error)
	+ if (error)
	*error = ENOSPC;
	NET_EPOCH_EXIT(et);
	return (NULL);
	@@ -1086,6 +1191,10 @@
	NET_EPOCH_EXIT(et);
	return (NULL);
	}
	+ KASSERT((tifp != ifp),
	+ ("Lookup failure ifp:%p inp:%p rt_find_real_interface() returns the same interface tifp:%p?\n",
	+ ifp, inp, tifp));
	+ ifp = tifp;
	goto use_real_interface;
	}
	if (rs->rs_flow_limit &&
	@@ -1095,9 +1204,9 @@
	NET_EPOCH_EXIT(et);
	return (NULL);
	}
	- rte = tcp_find_suitable_rate(rs, bytes_per_sec, flags);
	+ rte = tcp_find_suitable_rate(rs, bytes_per_sec, flags, lower_rate);
	if (rte) {
	- err = in_pcbattach_txrtlmt(inp, rs->rs_ifp,
	+ err = in_pcbattach_txrtlmt(inp, oifp,
	inp->inp_flowtype,
	inp->inp_flowid,
	rte->rate,
	@@ -1107,6 +1216,13 @@
	if (error)
	*error = err;
	rte = NULL;
	+ } else {
	+ KASSERT((inp->inp_snd_tag != NULL) ,
	+ ("Setup rate has no snd_tag inp:%p rte:%p rate:%lu rs:%p",
	+ inp, rte, rte->rate, rs));
	+#ifdef NETFLIX_STATS
	+ counter_u64_add(rate_limit_new, 1);
	+#endif
	}
	}
	if (rte) {
	@@ -1125,6 +1241,7 @@
	{
	int error;
	struct tcp_rate_set *rs;
	+ struct epoch_tracker et;

	if (((ifp->if_capenable & IFCAP_TXRTLMT) == 0) \|\|
	(link_state != LINK_STATE_UP)) {
	@@ -1134,53 +1251,56 @@
	*/
	return;
	}
	+ NET_EPOCH_ENTER(et);
	mtx_lock(&rs_mtx);
	- CK_LIST_FOREACH(rs, &int_rs, next) {
	- if ((rs->rs_ifp == ifp) &&
	- (rs->rs_if_dunit == ifp->if_dunit)) {
	- /* We already have initialized this guy */
	- mtx_unlock(&rs_mtx);
	- return;
	- }
	+ rs = find_rs_for_ifp(ifp);
	+ if (rs) {
	+ /* We already have initialized this guy */
	+ mtx_unlock(&rs_mtx);
	+ NET_EPOCH_EXIT(et);
	+ return;
	}
	mtx_unlock(&rs_mtx);
	+ NET_EPOCH_EXIT(et);
	rt_setup_new_rs(ifp, &error);
	}

	static void
	tcp_rl_ifnet_departure(void arg __unused, struct ifnet ifp)
	{
	- struct tcp_rate_set rs, nrs;
	+ struct tcp_rate_set *rs;
	+ struct epoch_tracker et;
	int i;

	+ NET_EPOCH_ENTER(et);
	mtx_lock(&rs_mtx);
	- CK_LIST_FOREACH_SAFE(rs, &int_rs, next, nrs) {
	- if ((rs->rs_ifp == ifp) &&
	- (rs->rs_if_dunit == ifp->if_dunit)) {
	- CK_LIST_REMOVE(rs, next);
	- rs_number_alive--;
	- rs->rs_flags \|= RS_IS_DEAD;
	- for (i = 0; i < rs->rs_rate_cnt; i++) {
	- if (rs->rs_rlt[i].flags & HDWRPACE_TAGPRESENT) {
	- in_pcbdetach_tag(rs->rs_rlt[i].tag);
	- rs->rs_rlt[i].tag = NULL;
	- }
	- rs->rs_rlt[i].flags = HDWRPACE_IFPDEPARTED;
	+ rs = find_rs_for_ifp(ifp);
	+ if (rs) {
	+ CK_LIST_REMOVE(rs, next);
	+ rs_number_alive--;
	+ rs->rs_flags \|= RS_IS_DEAD;
	+ for (i = 0; i < rs->rs_rate_cnt; i++) {
	+ if (rs->rs_rlt[i].flags & HDWRPACE_TAGPRESENT) {
	+ in_pcbdetach_tag(rs->rs_rlt[i].tag);
	+ rs->rs_rlt[i].tag = NULL;
	}
	- if (rs->rs_flows_using == 0)
	- rs_defer_destroy(rs);
	- break;
	+ rs->rs_rlt[i].flags = HDWRPACE_IFPDEPARTED;
	}
	+ if (rs->rs_flows_using == 0)
	+ rs_defer_destroy(rs);
	}
	mtx_unlock(&rs_mtx);
	+ NET_EPOCH_EXIT(et);
	}

	static void
	tcp_rl_shutdown(void *arg __unused, int howto __unused)
	{
	struct tcp_rate_set rs, nrs;
	+ struct epoch_tracker et;
	int i;

	+ NET_EPOCH_ENTER(et);
	mtx_lock(&rs_mtx);
	CK_LIST_FOREACH_SAFE(rs, &int_rs, next, nrs) {
	CK_LIST_REMOVE(rs, next);
	@@ -1197,11 +1317,12 @@
	rs_defer_destroy(rs);
	}
	mtx_unlock(&rs_mtx);
	+ NET_EPOCH_EXIT(et);
	}

	const struct tcp_hwrate_limit_table *
	tcp_set_pacing_rate(struct tcpcb tp, struct ifnet ifp,
	- uint64_t bytes_per_sec, int flags, int *error)
	+ uint64_t bytes_per_sec, int flags, int error, uint64_t lower_rate)
	{
	const struct tcp_hwrate_limit_table *rte;
	#ifdef KERN_TLS
	@@ -1233,7 +1354,9 @@
	}
	}
	#endif
	- rte = rt_setup_rate(tp->t_inpcb, ifp, bytes_per_sec, flags, error);
	+ rte = rt_setup_rate(tp->t_inpcb, ifp, bytes_per_sec, flags, error, lower_rate);
	+ if (rte)
	+ rl_increment_using(rte);
	#ifdef KERN_TLS
	if (rte != NULL && tls != NULL && tls->snd_tag != NULL) {
	/*
	@@ -1253,22 +1376,23 @@
	*error = EINVAL;
	rte = NULL;
	}
	- tp->t_pacing_rate = rte->rate;
	- *error = 0;
	+ if (rte != NULL) {
	+ tp->t_pacing_rate = rte->rate;
	+ *error = 0;
	+ }
	return (rte);
	}

	const struct tcp_hwrate_limit_table *
	tcp_chg_pacing_rate(const struct tcp_hwrate_limit_table *crte,
	struct tcpcb tp, struct ifnet ifp,
	- uint64_t bytes_per_sec, int flags, int *error)
	+ uint64_t bytes_per_sec, int flags, int error, uint64_t lower_rate)
	{
	const struct tcp_hwrate_limit_table *nrte;
	const struct tcp_rate_set *rs;
	#ifdef KERN_TLS
	struct ktls_session *tls = NULL;
	#endif
	- int is_indirect = 0;
	int err;

	INP_WLOCK_ASSERT(tp->t_inpcb);
	@@ -1307,41 +1431,13 @@
	if ((rs->rs_flags & RS_IS_DEAD) \|\|
	(crte->flags & HDWRPACE_IFPDEPARTED)) {
	/* Release the rate, and try anew */
	-re_rate:
	+
	tcp_rel_pacing_rate(crte, tp);
	nrte = tcp_set_pacing_rate(tp, ifp,
	- bytes_per_sec, flags, error);
	+ bytes_per_sec, flags, error, lower_rate);
	return (nrte);
	}
	- if ((rs->rs_flags & RT_IS_INDIRECT ) == RT_IS_INDIRECT)
	- is_indirect = 1;
	- else
	- is_indirect = 0;
	- if ((is_indirect == 0) &&
	- ((ifp != rs->rs_ifp) \|\|
	- (ifp->if_dunit != rs->rs_if_dunit))) {
	- /*
	- * Something changed, the user is not pointing to the same
	- * ifp? Maybe a route updated on this guy?
	- */
	- goto re_rate;
	- } else if (is_indirect) {
	- /*
	- * For indirect we have to dig in and find the real interface.
	- */
	- struct ifnet *rifp;
	-
	- rifp = rt_find_real_interface(ifp, tp->t_inpcb, error);
	- if (rifp == NULL) {
	- /* Can't find it? */
	- goto re_rate;
	- }
	- if ((rifp != rs->rs_ifp) \|\|
	- (ifp->if_dunit != rs->rs_if_dunit)) {
	- goto re_rate;
	- }
	- }
	- nrte = tcp_find_suitable_rate(rs, bytes_per_sec, flags);
	+ nrte = tcp_find_suitable_rate(rs, bytes_per_sec, flags, lower_rate);
	if (nrte == crte) {
	/* No change */
	if (error)
	@@ -1350,9 +1446,13 @@
	}
	if (nrte == NULL) {
	/* Release the old rate */
	+ if (error)
	+ *error = ENOENT;
	tcp_rel_pacing_rate(crte, tp);
	return (NULL);
	}
	+ rl_decrement_using(crte);
	+ rl_increment_using(nrte);
	/* Change rates to our new entry */
	#ifdef KERN_TLS
	if (tls != NULL)
	@@ -1361,10 +1461,18 @@
	#endif
	err = in_pcbmodify_txrtlmt(tp->t_inpcb, nrte->rate);
	if (err) {
	+ rl_decrement_using(nrte);
	+ /* Do we still have a snd-tag attached? */
	+ if (tp->t_inpcb->inp_snd_tag)
	+ in_pcbdetach_txrtlmt(tp->t_inpcb);
	if (error)
	*error = err;
	return (NULL);
	}
	+#ifdef NETFLIX_STATS
	+ else
	+ counter_u64_add(rate_limit_chg, 1);
	+#endif
	if (error)
	*error = 0;
	tp->t_pacing_rate = nrte->rate;
	@@ -1387,8 +1495,12 @@
	* in order to release our refcount.
	*/
	rs = __DECONST(struct tcp_rate_set *, crs);
	+ rl_decrement_using(crte);
	pre = atomic_fetchadd_64(&rs->rs_flows_using, -1);
	if (pre == 1) {
	+ struct epoch_tracker et;
	+
	+ NET_EPOCH_ENTER(et);
	mtx_lock(&rs_mtx);
	/*
	* Is it dead?
	@@ -1396,6 +1508,7 @@
	if (rs->rs_flags & RS_IS_DEAD)
	rs_defer_destroy(rs);
	mtx_unlock(&rs_mtx);
	+ NET_EPOCH_EXIT(et);
	}

	/*
	@@ -1412,8 +1525,39 @@
	#define FIVE_HUNDRED_MBPS 62500000 /* 500Mbps in bytes per second */
	#define MAX_MSS_SENT 43 /* 43 mss = 43 x 1500 = 64,500 bytes */

	+static void
	+tcp_log_pacing_size(struct tcpcb *tp, uint64_t bw, uint32_t segsiz, uint32_t new_tso,
	+ uint64_t hw_rate, uint32_t time_between, uint32_t calc_time_between,
	+ uint32_t segs, uint32_t res_div, uint16_t mult, uint8_t mod)
	+{
	+ if (tp->t_logstate != TCP_LOG_STATE_OFF) {
	+ union tcp_log_stackspecific log;
	+ struct timeval tv;
	+ uint32_t cts;
	+
	+ memset(&log, 0, sizeof(log));
	+ cts = tcp_get_usecs(&tv);
	+ log.u_bbr.flex1 = segsiz;
	+ log.u_bbr.flex2 = new_tso;
	+ log.u_bbr.flex3 = time_between;
	+ log.u_bbr.flex4 = calc_time_between;
	+ log.u_bbr.flex5 = segs;
	+ log.u_bbr.flex6 = res_div;
	+ log.u_bbr.flex7 = mult;
	+ log.u_bbr.flex8 = mod;
	+ log.u_bbr.timeStamp = tcp_get_usecs(&tv);
	+ log.u_bbr.cur_del_rate = bw;
	+ log.u_bbr.delRate = hw_rate;
	+ TCP_LOG_EVENTP(tp, NULL,
	+ &tp->t_inpcb->inp_socket->so_rcv,
	+ &tp->t_inpcb->inp_socket->so_snd,
	+ TCP_HDWR_PACE_SIZE, 0,
	+ 0, &log, false, &tv);
	+ }
	+}
	+
	uint32_t
	-tcp_get_pacing_burst_size (uint64_t bw, uint32_t segsiz, int can_use_1mss,
	+tcp_get_pacing_burst_size (struct tcpcb *tp, uint64_t bw, uint32_t segsiz, int can_use_1mss,
	const struct tcp_hwrate_limit_table te, int err)
	{
	/*
	@@ -1439,11 +1583,15 @@
	min_tso_segs = 1;
	else
	min_tso_segs = 2;
	- if (new_tso < min_tso_segs)
	+ if (rs_floor_mss && (new_tso < rs_floor_mss))
	+ new_tso = rs_floor_mss;
	+ else if (new_tso < min_tso_segs)
	new_tso = min_tso_segs;
	if (new_tso > MAX_MSS_SENT)
	new_tso = MAX_MSS_SENT;
	new_tso *= segsiz;
	+ tcp_log_pacing_size(tp, bw, segsiz, new_tso,
	+ 0, 0, 0, 0, 0, 0, 1);
	/*
	* If we are not doing hardware pacing
	* then we are done.
	@@ -1480,25 +1628,33 @@
	* max (43 segments).
	*/
	if (te->rate > FIVE_HUNDRED_MBPS)
	- return (segsiz * MAX_MSS_SENT);
	+ goto max;
	if (te->rate == bw) {
	/* We are pacing at exactly the hdwr rate */
	+max:
	+ tcp_log_pacing_size(tp, bw, segsiz, new_tso,
	+ te->rate, te->time_between, (uint32_t)0,
	+ (segsiz * MAX_MSS_SENT), 0, 0, 3);
	return (segsiz * MAX_MSS_SENT);
	}
	lentim = ETHERNET_SEGMENT_SIZE * USECS_IN_SECOND;
	res = lentim / bw;
	if (res > te->time_between) {
	- uint32_t delta, segs;
	+ uint32_t delta, segs, res_div;

	+ res_div = ((res * num_of_waits_allowed) + wait_time_floor);
	delta = res - te->time_between;
	- segs = (res + delta - 1)/delta;
	- if (te->rate > ONE_HUNDRED_MBPS)
	- segs *= 2;
	+ segs = (res_div + delta - 1)/delta;
	if (segs < min_tso_segs)
	segs = min_tso_segs;
	+ if (segs < rs_hw_floor_mss)
	+ segs = rs_hw_floor_mss;
	if (segs > MAX_MSS_SENT)
	segs = MAX_MSS_SENT;
	segs *= segsiz;
	+ tcp_log_pacing_size(tp, bw, segsiz, new_tso,
	+ te->rate, te->time_between, (uint32_t)res,
	+ segs, res_div, 1, 3);
	if (err)
	*err = 0;
	if (segs < new_tso) {
	@@ -1514,12 +1670,47 @@
	* hardware. Send back the non-hardware
	* rate.
	*/
	+ tcp_log_pacing_size(tp, bw, segsiz, new_tso,
	+ te->rate, te->time_between, (uint32_t)res,
	+ 0, 0, 0, 4);
	if (err)
	*err = -1;
	return (new_tso);
	}
	}

	+uint64_t
	+tcp_hw_highest_rate_ifp(struct ifnet ifp, struct inpcb inp)
	+{
	+ struct epoch_tracker et;
	+ struct tcp_rate_set *rs;
	+ uint64_t rate_ret;
	+
	+ NET_EPOCH_ENTER(et);
	+use_next_interface:
	+ rs = find_rs_for_ifp(ifp);
	+ if (rs == NULL) {
	+ /* This interface does not do ratelimiting */
	+ rate_ret = 0;
	+ } else if (rs->rs_flags & RS_IS_DEFF) {
	+ /* We need to find the real interface */
	+ struct ifnet *tifp;
	+
	+ tifp = rt_find_real_interface(ifp, inp, NULL);
	+ if (tifp == NULL) {
	+ NET_EPOCH_EXIT(et);
	+ return (0);
	+ }
	+ ifp = tifp;
	+ goto use_next_interface;
	+ } else {
	+ /* Lets return the highest rate this guy has */
	+ rate_ret = rs->rs_rlt[rs->rs_highest_valid].rate;
	+ }
	+ NET_EPOCH_EXIT(et);
	+ return(rate_ret);
	+}
	+
	static eventhandler_tag rl_ifnet_departs;
	static eventhandler_tag rl_ifnet_arrives;
	static eventhandler_tag rl_shutdown_start;
	Index: sys/netinet/tcp_stacks/bbr.c
	===================================================================
	--- sys/netinet/tcp_stacks/bbr.c
	+++ sys/netinet/tcp_stacks/bbr.c
	@@ -5551,7 +5551,7 @@
	bbr->rc_inp->inp_route.ro_nh->nh_ifp,
	rate,
	(RS_PACING_GEQ\|RS_PACING_SUB_OK),
	- &error);
	+ &error, NULL);
	if (nrte == NULL) {
	goto lost_rate;
	}
	@@ -14081,7 +14081,7 @@
	inp->inp_route.ro_nh->nh_ifp,
	rate_wanted,
	(RS_PACING_GEQ\|RS_PACING_SUB_OK),
	- &err);
	+ &err, NULL);
	if (bbr->r_ctl.crte) {
	bbr_type_log_hdwr_pacing(bbr,
	bbr->r_ctl.crte->ptbl->rs_ifp,
	Index: sys/netinet/tcp_stacks/rack.c
	===================================================================
	--- sys/netinet/tcp_stacks/rack.c
	+++ sys/netinet/tcp_stacks/rack.c
	@@ -1928,7 +1928,7 @@
	TCP_LOG_EVENTP(tp, NULL,
	&tp->t_inpcb->inp_socket->so_rcv,
	&tp->t_inpcb->inp_socket->so_snd,
	- TCP_HDWR_TLS, 0,
	+ TCP_HDWR_PACE_SIZE, 0,
	0, &log, false, &tv);
	}
	}
	@@ -10289,7 +10289,7 @@
	segsiz = min(ctf_fixed_maxseg(tp),
	rack->r_ctl.rc_pace_min_segs);
	rack->r_ctl.rc_pace_max_segs = tcp_get_pacing_burst_size(
	- bw_est, segsiz, 0,
	+ tp, bw_est, segsiz, 0,
	rack->r_ctl.crte, NULL);
	}
	} else if (rack->rc_always_pace) {
	@@ -11348,7 +11348,7 @@
	/* Use the user mss since we are not exactly matched */
	return (user_max);
	}
	- new_tso = tcp_get_pacing_burst_size(bw, mss, rack_pace_one_seg, rack->r_ctl.crte, NULL);
	+ new_tso = tcp_get_pacing_burst_size(rack->rc_tp, bw, mss, rack_pace_one_seg, rack->r_ctl.crte, NULL);
	if (new_tso > user_max)
	new_tso = user_max;
	return(new_tso);
	@@ -11575,10 +11575,10 @@
	rack->rc_inp->inp_route.ro_nh->nh_ifp,
	rate_wanted,
	RS_PACING_GEQ,
	- &err);
	+ &err, NULL);
	if (rack->r_ctl.crte) {
	rack->rack_hdrw_pacing = 1;
	- rack->r_ctl.rc_pace_max_segs = tcp_get_pacing_burst_size(rate_wanted, segsiz,
	+ rack->r_ctl.rc_pace_max_segs = tcp_get_pacing_burst_size(rack->rc_tp, rate_wanted, segsiz,
	0, rack->r_ctl.crte,
	NULL);
	rack_log_hdwr_pacing(rack, rack->rc_inp->inp_route.ro_nh->nh_ifp,
	@@ -11595,14 +11595,14 @@
	rack->rc_inp->inp_route.ro_nh->nh_ifp,
	rate_wanted,
	RS_PACING_GEQ,
	- &err);
	+ &err, NULL);
	if (nrte == NULL) {
	/* Lost the rate */
	rack->rack_hdrw_pacing = 0;
	rack_set_pace_segments(rack->rc_tp, rack, __LINE__);
	} else if (nrte != rack->r_ctl.crte) {
	rack->r_ctl.crte = nrte;
	- rack->r_ctl.rc_pace_max_segs = tcp_get_pacing_burst_size(rate_wanted,
	+ rack->r_ctl.rc_pace_max_segs = tcp_get_pacing_burst_size(rack->rc_tp, rate_wanted,
	segsiz, 0,
	rack->r_ctl.crte,
	NULL);

File Metadata

Mime Type: text/plain
Expires: Sat, Sep 5, 10:11 PM (17 h, 44 m)
Storage Engine: blob
Storage Format: Raw Data
Storage Handle: 38363527
Default Alt Text: D28357.id82928.diff (53 KB)

D28357.id82928.diffNo OneTemporaryActions

D28357.id82928.diffView Options

File Metadata

Event Timeline

D28357.id82928.diff
No OneTemporary
Actions

D28357.id82928.diff
View Options