Configuring traffic classes via new hardware offload mechanism in tc/mqprio
Subsystems (4)
v3 cover Configuring traffic classes via new hardware offload mechanism in tc/mqprio
The following series introduces a new hardware offload mode in tc/mqprio where the TCs, the queue configurations and bandwidth rate limits are offloaded to the hardware. The existing mqprio framework is extended to configure the queue counts and layout and also added support for rate limiting. This is achieved through new netlink attributes for the 'mode' option which takes values such as 'dcb' (default) and 'channel' and a 'shaper' option for QoS attributes such as bandwidth rate limits in hw mode 1. Legacy devices can fall back to the existing setup supporting hw mode 1 without these additional options where only the TCs are offloaded and then the 'mode' and 'shaper' options defaults to DCB support. The i40e driver enables the new mqprio hardware offload mechanism factoring the TCs, queue configuration and bandwidth rates by creating HW channel VSIs.
In this new mode, the priority to traffic class mapping and the user specified queue ranges are used to configure the traffic class when the 'mode' option is set to 'channel'. This is achieved by creating HW channels(VSI). A new channel is created for each of the traffic class configuration offloaded via mqprio framework except for the first TC (TC0) which is for the main VSI. TC0 for the main VSI is also reconfigured as per user provided queue parameters. Finally, bandwidth rate limits are set on these traffic classes through the shaper attribute by sending these rates in addition to the number of TCs and the queue configurations.
Example:
# tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\
queues 4 at 0 4 at 4 hw 1 mode channel shaper bw_rlimit\
min_rate 1Gbit 2Gbit max_rate 4Gbit 5GbitTo dump the bandwidth rates:
# tc qdisc show dev eth0
qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0
queues:(0:3) (4:7)
mode:channel
shaper:bw_rlimit min_rate:1Gbit 2Gbit max_rate:4Gbit 5Gbit
---
Amritha Nambiar (6):
mqprio: Introduce new hardware offload mode and shaper in mqprio
i40e: Add macro for PF reset bit
i40e: Add infrastructure for queue channel support
i40e: Enable 'channel' mode in mqprio for TC configs
i40e: Refactor VF BW rate limiting
i40e: Add support setting TC max bandwidth rates
drivers/net/ethernet/intel/i40e/i40e.h | 44 + drivers/net/ethernet/intel/i40e/i40e_debugfs.c | 3 drivers/net/ethernet/intel/i40e/i40e_ethtool.c | 8 drivers/net/ethernet/intel/i40e/i40e_main.c | 1463 +++++++++++++++++--- drivers/net/ethernet/intel/i40e/i40e_txrx.h | 2 drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 50 - include/net/pkt_cls.h | 9 include/uapi/linux/pkt_sched.h | 32 net/sched/sch_mqprio.c | 183 ++- 9 files changed, 1551 insertions(+), 243 deletions(-)
--
RFC PATCH v3 1/6 mqprio: Introduce new hardware offload mode and shaper in mqprio
The offload types currently supported in mqprio are 0 (no offload) and 1 (offload only TCs) by setting these values for the 'hw' option. If offloads are supported by setting the 'hw' option to 1, the default offload mode is 'dcb' where only the TC values are offloaded to the device. This patch introduces a new hardware offload mode called 'channel' with 'hw' set to 1 in mqprio which makes full use of the mqprio options, the TCs, the queue configurations and the QoS parameters for the TCs. This is achieved through a new netlink attribute for the 'mode' option which takes values such as 'dcb' (default) and 'channel'. The 'channel' mode also supports QoS attributes for traffic class such as minimum and maximum values for bandwidth rate limits.
This patch enables configuring additional HW shaper attributes associated with a traffic class. Currently the shaper for bandwidth rate limiting is supported which takes options such as minimum and maximum bandwidth rates and are offloaded to the hardware in the 'channel' mode. The min and max limits for bandwidth rates are provided by the user along with the the TCs and the queue configurations when creating the mqprio qdisc. The interface can be extended to support new HW shapers in future through the 'shaper' attribute.
Introduces a new datastructure 'tc_mqprio_qopt_offload' for offloading mqprio queue options and use this to be shared between the kernel and device driver. This contains a copy of the exisiting datastructure for mqprio queue options. This new datastructure can be extended when adding new attributes for traffic class such as mode, shaper, shaper parameters (bandwidth rate limits). The existing datastructure for mqprio queue options will be shared between the kernel and userspace.
Example:
# tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\ queues 4 at 0 4 at 4 hw 1 mode channel shaper bw_rlimit\ min_rate 1Gbit 2Gbit max_rate 4Gbit 5Gbit
To dump the bandwidth rates:
# tc qdisc show dev eth0
qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0
queues:(0:3) (4:7)
mode:channel
shaper:bw_rlimit min_rate:1Gbit 2Gbit max_rate:4Gbit 5Gbit
v3 : Removed supporting new offloads through value 2 for 'hw' option, introduced new netlink based options for offload mode and HW shaper. v2 : Jiri's changes accepted upstream removes the struct tc_to_netdev. Clean up the full offload related changes added to mqprio_init() and mqprio_destroy() to rebase on these changes.
--- include/net/pkt_cls.h | 9 ++ include/uapi/linux/pkt_sched.h | 32 +++++++ net/sched/sch_mqprio.c | 183 ++++++++++++++++++++++++++++++++++++++-- 3 files changed, 215 insertions(+), 9 deletions(-)
include/net/pkt_cls.h +9 -0
@@ -546,6 +546,15 @@ struct tc_cls_bpf_offload {u32 gen_flags; }; +struct tc_mqprio_qopt_offload { +/* struct tc_mqprio_qopt must always be the first element */ +struct tc_mqprio_qopt qopt; +u16 mode; +u16 shaper; +u32 flags; +u64 min_rate[TC_QOPT_MAX_QUEUE ]; +u64 max_rate[TC_QOPT_MAX_QUEUE ]; +};/* This structure holds cookie structure that is passed from user * to the kernel for actions and classifiers
include/uapi/linux/pkt_sched.h +32 -0
@@ -625,6 +625,22 @@ enum {#define TC_MQPRIO_HW_OFFLOAD_MAX (__TC_MQPRIO_HW_OFFLOAD_MAX - 1) +enum { +TC_MQPRIO_MODE_DCB , +TC_MQPRIO_MODE_CHANNEL , + __TC_MQPRIO_MODE_MAX +}; + +#define __TC_MQPRIO_MODE_MAX (__TC_MQPRIO_MODE_MAX - 1) + +enum { +TC_MQPRIO_SHAPER_DCB , +TC_MQPRIO_SHAPER_BW_RATE ,/* Add new shapers below */ + __TC_MQPRIO_SHAPER_MAX +}; + +#define __TC_MQPRIO_SHAPER_MAX (__TC_MQPRIO_SHAPER_MAX - 1) +struct tc_mqprio_qopt {__u8 num_tc;__u8 prio_tc_map[TC_QOPT_BITMASK +1 ];
@@ -633,6 +649,22 @@ struct tc_mqprio_qopt {__u16 offset[TC_QOPT_MAX_QUEUE ]; }; +#define TC_MQPRIO_F_MODE 0x1 +#define TC_MQPRIO_F_SHAPER 0x2 +#define TC_MQPRIO_F_MIN_RATE 0x4 +#define TC_MQPRIO_F_MAX_RATE 0x8 + +enum { +TCA_MQPRIO_UNSPEC , +TCA_MQPRIO_MODE , +TCA_MQPRIO_SHAPER , +TCA_MQPRIO_MIN_RATE64 , +TCA_MQPRIO_MAX_RATE64 , + __TCA_MQPRIO_MAX, +}; + +#define TCA_MQPRIO_MAX (__TCA_MQPRIO_MAX - 1) +/* SFB */ enum {
net/sched/sch_mqprio.c +174 -9
248 lines not on this page · raw message
Bowers
RFC PATCH v3 1/6 mqprio: Introduce new hardware offload mode and shaper in mqprio
Quoted 60 lines
-----Original Message-----
From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of Amritha Nambiar
Sent: Thursday, September 7, 2017 4:00 AM
To: intel-wired-lan at lists.osuosl.org; Kirsher, Jeffrey T <[email protected]>
Cc: netdev at vger.kernel.org
Subject: [Intel-wired-lan] [RFC PATCH v3 1/6] mqprio: Introduce new hardware offload mode and shaper in mqprioThe offload types currently supported in mqprio are 0 (no offload) and 1 (offload only TCs) by setting these values for the 'hw' option. If offloads are supported by setting the 'hw' option to 1, the default offload mode is 'dcb' where only the TC values are offloaded to the device. This patch introduces a new hardware offload mode called 'channel' with 'hw' set to 1 in mqprio which makes full use of the mqprio options, the TCs, the queue configurations and the QoS parameters for the TCs. This is achieved through a new netlink attribute for the 'mode' option which takes values such as 'dcb' (default) and 'channel'.
The 'channel' mode also supports QoS attributes for traffic class such as minimum and maximum values for bandwidth rate limits.This patch enables configuring additional HW shaper attributes associated with a traffic class. Currently the shaper for bandwidth rate limiting is supported which takes options such as minimum and maximum bandwidth rates and are offloaded to the hardware in the 'channel' mode. The min and max limits for bandwidth rates are provided by the user along with the the TCs and the queue configurations when creating the mqprio qdisc. The interface can be extended to support new HW shapers in future through the 'shaper'
attribute.Introduces a new datastructure '
tc_mqprio_qopt_offload' for offloading mqprio queue options and use this to be shared between the kernel and device driver. This contains a copy of the exisiting datastructure for mqprio queue options. This new datastructure can be extended when adding new attributes for traffic class such as mode, shaper, shaper parameters (bandwidth rate limits). The existing datastructure for mqprio queue options will be shared between the kernel and userspace.Example:
# tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\ queues 4 at 0 4 at 4 hw 1 mode channel shaper bw_rlimit\ min_rate 1Gbit 2Gbit max_rate 4Gbit 5GbitTo dump the bandwidth rates:
# tc qdisc show dev eth0qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0
queues:(0:3) (4:7)
mode:channel
shaper:bw_rlimit min_rate:1Gbit 2Gbit max_rate:4Gbit 5Gbitv3 : Removed supporting new offloads through value 2 for 'hw' option, introduced new netlink based options for offload mode and HW shaper. v2 : Jiri's changes accepted upstream removes the struct
tc_to_netdev. Clean up the full offload related changes added tomqprio_init()andmqprio_destroy()to rebase on these changes.
Quoted 7 lines
Signed-off-by Amritha Nambiar <[email protected]>--- include/net/pkt_cls.h | 9 ++ include/uapi/linux/pkt_sched.h | 32 +++++++ net/sched/sch_mqprio.c | 183 ++++++++++++++++++++++++++++++++++++++-- 3 files changed, 215 insertions(+), 9 deletions(-)
RFC PATCH v3 2/6 i40e: Add macro for PF reset bit
In reply to Amritha NambiarIntroduce a macro for the bit setting the PF reset flag and update its usages. This makes it easier to use this flag in functions to be introduced in future without encountering checkpatch issues related to alignment and line over 80 characters.
--- drivers/net/ethernet/intel/i40e/i40e.h | 2 ++ drivers/net/ethernet/intel/i40e/i40e_debugfs.c | 3 +-- drivers/net/ethernet/intel/i40e/i40e_main.c | 9 ++++----- drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 5 ++--- 4 files changed, 9 insertions(+), 10 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +2 -0
@@ -157,6 +157,8 @@ enum i40e_state_t { __I40E_STATE_SIZE__, }; +#define I40E_PF_RESET_FLAG BIT_ULL(__I40E_PF_RESET_REQUESTED) +/* VSI state flags */ enum i40e_vsi_state_t { __I40E_VSI_DOWN,
drivers/net/ethernet/intel/i40e/i40e_debugfs.c +1 -2
@@ -798,8 +798,7 @@ static ssize_t i40e_dbg_command_write(struct file *filp,*/ if (!(pf->flags &I40E_FLAG_VEB_MODE_ENABLED )) { pf->flags |=I40E_FLAG_VEB_MODE_ENABLED ; -i40e_do_reset_safe (pf, -BIT_ULL (__I40E_PF_RESET_REQUESTED)); +i40e_do_reset_safe (pf,I40E_PF_RESET_FLAG ); } vsi =i40e_vsi_setup (pf,I40E_VSI_VMDQ2 , vsi_seid,0 );
drivers/net/ethernet/intel/i40e/i40e_main.c +4 -5
@@ -5753,7 +5753,7 @@ int i40e_vsi_open(struct i40e_vsi *vsi) err_setup_tx:i40e_vsi_free_tx_resources (vsi);if (vsi == pf->vsi[pf->lan_vsi]) -i40e_do_reset (pf,BIT_ULL (__I40E_PF_RESET_REQUESTED), true); +i40e_do_reset (pf,I40E_PF_RESET_FLAG , true);return err; }
@@ -5881,7 +5881,7 @@ void i40e_do_reset(struct i40e_pf *pf, u32 reset_flags, bool lock_acquired)wr32 (&pf->hw,I40E_GLGEN_RTRIG , val);i40e_flush (&pf->hw); - }else if (reset_flags &BIT_ULL (__I40E_PF_RESET_REQUESTED)) { + }else if (reset_flags &I40E_PF_RESET_FLAG ) {/* Request a PF Reset *
@@ -9229,7 +9229,7 @@ static int i40e_set_features(struct net_device *netdev, need_reset =i40e_set_ntuple (pf, features);if (need_reset) -i40e_do_reset (pf,BIT_ULL (__I40E_PF_RESET_REQUESTED), true); +i40e_do_reset (pf,I40E_PF_RESET_FLAG , true);return 0 ; }
@@ -9481,8 +9481,7 @@ static int i40e_ndo_bridge_setlink(struct net_device *dev, pf->flags |=I40E_FLAG_VEB_MODE_ENABLED ; else pf->flags &= ~I40E_FLAG_VEB_MODE_ENABLED; -i40e_do_reset (pf,BIT_ULL (__I40E_PF_RESET_REQUESTED), - true); +i40e_do_reset (pf,I40E_PF_RESET_FLAG , true);break ; } }
drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c +2 -3
@@ -1424,8 +1424,7 @@ int i40e_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)if (num_vfs) {if (!(pf->flags &I40E_FLAG_VEB_MODE_ENABLED )) { pf->flags |=I40E_FLAG_VEB_MODE_ENABLED ; -i40e_do_reset_safe (pf, -BIT_ULL (__I40E_PF_RESET_REQUESTED)); +i40e_do_reset_safe (pf,I40E_PF_RESET_FLAG ); }return i40e_pci_sriov_enable (pdev, num_vfs); }
@@ -1433,7 +1432,7 @@ int i40e_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)if (!pci_vfs_assigned (pf->pdev)) {i40e_free_vfs (pf); pf->flags &= ~I40E_FLAG_VEB_MODE_ENABLED ; -i40e_do_reset_safe (pf,BIT_ULL (__I40E_PF_RESET_REQUESTED)); +i40e_do_reset_safe (pf,I40E_PF_RESET_FLAG ); }else {dev_warn (&pdev->dev,"Unable to free VFs because some are assigned to VMs.\n" );return -EINVAL ;
Bowers
RFC PATCH v3 2/6 i40e: Add macro for PF reset bit
Quoted 14 lines
-----Original Message-----
From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of Amritha Nambiar
Sent: Thursday, September 7, 2017 4:00 AM
To: intel-wired-lan at lists.osuosl.org; Kirsher, Jeffrey T <[email protected]>
Cc: netdev at vger.kernel.org
Subject: [Intel-wired-lan] [RFC PATCH v3 2/6] i40e: Add macro for PF reset bitIntroduce a macro for the bit setting the PF reset flag and update its usages. This makes it easier to use this flag in functions to be introduced in future without encountering checkpatch issues related to alignment and line over 80 characters.
Quoted 7 lines
Signed-off-by Amritha Nambiar <[email protected]>--- drivers/net/ethernet/intel/i40e/i40e.h | 2 ++ drivers/net/ethernet/intel/i40e/i40e_debugfs.c | 3 +-- drivers/net/ethernet/intel/i40e/i40e_main.c | 9 ++++----- drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 5 ++--- 4 files changed, 9 insertions(+), 10 deletions(-)
RFC PATCH v3 3/6 i40e: Add infrastructure for queue channel support
In reply to Amritha NambiarThis patch sets up the infrastructure for offloading TCs and queue configurations to the hardware by creating HW channels(VSI). A new channel is created for each of the traffic class configuration offloaded via mqprio framework except for the first TC (TC0). TC0 for the main VSI is also reconfigured as per user provided queue parameters. Queue counts that are not power-of-2 are handled by reconfiguring RSS by reprogramming LUTs using the queue count value. This patch also handles configuring the TX rings for the channels, setting up the RX queue map for channel.
Also, the channels so created are removed and all the queue configuration is set to default when the qdisc is detached from the root of the device.
v3: Addressed Shannon's feedback removing unwanted code, variables, atomic operations (since setup_tc is protected with rtnl_lock held by stack) and other minor clean up.
--- drivers/net/ethernet/intel/i40e/i40e.h | 32 + drivers/net/ethernet/intel/i40e/i40e_main.c | 718 +++++++++++++++++++++++++++ drivers/net/ethernet/intel/i40e/i40e_txrx.h | 2 3 files changed, 743 insertions(+), 9 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +32 -0
@@ -87,6 +87,7 @@#define I40E_AQ_LEN 256#define I40E_AQ_WORK_LIMIT 66/* max number of VFs + a little */ #define I40E_MAX_USER_PRIORITY 8 +#define I40E_MAX_QUEUES_PER_CH 64#define I40E_DEFAULT_TRAFFIC_CLASS BIT(0)#define I40E_DEFAULT_MSG_ENABLE 4#define I40E_QUEUE_WAIT_RETRY_LIMIT 10
@@ -340,6 +341,23 @@ struct i40e_flex_pit {u8 pit_index; }; +struct i40e_channel { +struct list_head list; +bool initialized; +u8 type; +u16 vsi_number;/* Assigned VSI number from AQ 'Add VSI' response */ +u16 stat_counter_idx; +u16 base_queue; +u16 num_queue_pairs;/* Requested by user */ +u16 seid; + +u8 enabled_tc; +struct i40e_aqc_vsi_properties_data info; + +/* track this channel belongs to which VSI */ +struct i40e_vsi *parent_vsi; +}; +/* struct that defines the Ethernet device */ struct i40e_pf {struct pci_dev *pdev;
@@ -456,6 +474,7 @@ struct i40e_pf {#define I40E_FLAG_CLIENT_RESET BIT(26)#define I40E_FLAG_LINK_DOWN_ON_CLOSE_ENABLED BIT(27)#define I40E_FLAG_SOURCE_PRUNING_DISABLED BIT(28) +#define I40E_FLAG_TC_MQPRIO BIT(29)struct i40e_client_instance *cinst;bool stat_offsets_loaded;
@@ -536,6 +555,8 @@ struct i40e_pf {u32 ioremap_len;u32 fd_inv;u16 phy_led_val; + +u16 override_q_count; };/**
@@ -700,6 +721,15 @@ struct i40e_vsi {bool current_isup;/* Sync 'link up' logging */ enum i40e_aq_link_speed current_speed;/* Sync link speed logging */ +/* channel specific fields */ +u16 cnt_q_avail;/* num of queues available for channel usage */ +u16 orig_rss_size; +u16 current_rss_size; + +u16 next_base_queue;/* next queue to be used for channel setup */ + +struct list_head ch_list; +void *priv;/* client driver data reference. */ /* VSI specific handlers */
@@ -1007,4 +1037,6 @@ static inline bool i40e_enabled_xdp_vsi(struct i40e_vsi *vsi) {return !!vsi->xdp_prog; } + +int i40e_create_queue_channel (struct i40e_vsi *vsi,struct i40e_channel *ch);#endif /* _I40E_H_ */
drivers/net/ethernet/intel/i40e/i40e_main.c +709 -9
769 lines not on this page · raw message
drivers/net/ethernet/intel/i40e/i40e_txrx.h +2 -0
@@ -424,6 +424,8 @@ struct i40e_ring {* i40e_clean_rx_ring_irq() is called * for this ring. */ + +struct i40e_channel *ch; }____cacheline_internodealigned_in_smp ;static inline bool ring_uses_build_skb (struct i40e_ring *ring)
Bowers
RFC PATCH v3 3/6 i40e: Add infrastructure for queue channel support
Quoted 27 lines
-----Original Message-----
From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of Amritha Nambiar
Sent: Thursday, September 7, 2017 4:00 AM
To: intel-wired-lan at lists.osuosl.org; Kirsher, Jeffrey T <[email protected]>
Cc: netdev at vger.kernel.org
Subject: [Intel-wired-lan] [RFC PATCH v3 3/6] i40e: Add infrastructure for queue channel supportThis patch sets up the infrastructure for offloading TCs and queue configurations to the hardware by creating HW channels(VSI).
A new channel is created for each of the traffic class configuration offloaded via mqprio framework except for the first TC (TC0). TC0 for the main VSI is also reconfigured as per user provided queue parameters. Queue counts that are not power-of-2 are handled by reconfiguring RSS by reprogramming LUTs using the queue count value.
This patch also handles configuring the TX rings for the channels, setting up the RX queue map for channel.Also, the channels so created are removed and all the queue configuration is set to default when the qdisc is detached from the root of the device.
v3: Addressed Shannon's feedback removing unwanted code, variables, atomic operations (since
setup_tcis protected withrtnl_lockheld by stack) and other minor clean up.
Quoted 8 lines
Signed-off-by Amritha Nambiar <[email protected]>Signed-off-by Kiran Patil <[email protected]>--- drivers/net/ethernet/intel/i40e/i40e.h | 32 + drivers/net/ethernet/intel/i40e/i40e_main.c | 718 +++++++++++++++++++++++++++ drivers/net/ethernet/intel/i40e/i40e_txrx.h | 2 3 files changed, 743 insertions(+), 9 deletions(-)
RFC PATCH v3 4/6 i40e: Enable 'channel' mode in mqprio for TC configs
In reply to Amritha NambiarThe i40e driver is modified to enable the new mqprio hardware offload mode and factor the TCs and queue configuration by creating channel VSIs. In this mode, the priority to traffic class mapping and the user specified queue ranges are used to configure the traffic classes by setting the mode option to 'channel'.
Example:
# tc qdisc add dev eth0 root mqprio num_tc 4\ map 0 0 0 0 1 2 2 3 queues 2 at 0 2 at 2 1 at 4 1 at 5\ hw 1 mode channel
# tc qdisc show dev eth0
qdisc mqprio 8038: root tc 4 map 0 0 0 0 1 2 2 3 0 0 0 0 0 0 0 0
queues:(0:1) (2:3) (4:4) (5:5)
mode:channel
shaper:dcbThe HW channels created are removed and all the queue configuration is set to default when the qdisc is detached from the root of the device.
# tc qdisc del dev eth0 root
This patch also disables setting up channels via ethtool (ethtool -L) when the TCs are configured using mqprio scheduler.
The patch also limits setting ethtool Rx flow hash indirection (ethtool -X eth0 equal N) to max queues configured via mqprio. The Rx flow hash indirection input through ethtool should be validated so that it is within in the queue range configured via tc/mqprio. The bound checking is achieved by reporting the current rss size to the kernel when queues are configured via mqprio.
Example:
# tc qdisc add dev eth0 root mqprio num_tc 4\ map 0 0 0 1 0 2 3 0 queues 2 at 0 4 at 2 8 at 6 11 at 14\ hw 1 mode channel
# ethtool -X eth0 equal 12 Cannot set RX flow hash configuration: Invalid argument
v3: Changes to incorporate new mqprio mode option. Minor clean up of setup_tc error handling based on Shannon's comments.
v2: Clean up __i40e_setup_tc() and i40e_setup_tc() to work with Jiri's changes to the ndo_setup_tc interface which now takes a type and the type_data for the offload. No need to disable ATR in MQPRIO mode.
--- drivers/net/ethernet/intel/i40e/i40e.h | 3 drivers/net/ethernet/intel/i40e/i40e_ethtool.c | 8 drivers/net/ethernet/intel/i40e/i40e_main.c | 457 ++++++++++++++++++------ 3 files changed, 362 insertions(+), 106 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +3 -0
@@ -54,6 +54,7 @@#include <linux/clocksource.h> #include <linux/net_tstamp.h> #include <linux/ptp_clock_kernel.h> +#include <net/pkt_cls.h> #include "i40e_type.h" #include "i40e_prototype.h" #include "i40e_client.h"
@@ -700,6 +701,7 @@ struct i40e_vsi {enum i40e_vsi_type type;/* VSI type, e.g., LAN, FCoE, etc */ s16 vf_id;/* Virtual function ID for SRIOV VSIs */ +struct tc_mqprio_qopt_offload mqprio_qopt;/* queue parameters */ struct i40e_tc_configuration tc_config;struct i40e_aqc_vsi_properties_data info;
@@ -725,6 +727,7 @@ struct i40e_vsi {u16 cnt_q_avail;/* num of queues available for channel usage */ u16 orig_rss_size;u16 current_rss_size; +bool reconfig_rss;u16 next_base_queue;/* next queue to be used for channel setup */
drivers/net/ethernet/intel/i40e/i40e_ethtool.c +7 -1
@@ -2652,7 +2652,7 @@ static int i40e_get_rxnfc(struct net_device *netdev, struct ethtool_rxnfc *cmd,switch (cmd->cmd) {case ETHTOOL_GRXRINGS : - cmd->data = vsi->num_queue_pairs; + cmd->data = vsi->rss_size; ret =0 ;break ;case ETHTOOL_GRXFH :
@@ -3897,6 +3897,12 @@ static int i40e_set_channels(struct net_device *dev,if (vsi->type !=I40E_VSI_MAIN )return -EINVAL ; +/* We do not support setting channels via ethtool when TCs are +* configured through mqprio +*/ +if (pf->flags &I40E_FLAG_TC_MQPRIO ) +return -EINVAL ; +/* verify they are not requesting separate vectors */ if (!count || ch->rx_count || ch->tx_count)return -EINVAL ;
drivers/net/ethernet/intel/i40e/i40e_main.c +352 -105
594 lines not on this page · raw message
Bowers
RFC PATCH v3 4/6 i40e: Enable 'channel' mode in mqprio for TC configs
Quoted 58 lines
-----Original Message-----
From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of Amritha Nambiar
Sent: Thursday, September 7, 2017 4:00 AM
To: intel-wired-lan at lists.osuosl.org; Kirsher, Jeffrey T <[email protected]>
Cc: netdev at vger.kernel.org
Subject: [Intel-wired-lan] [RFC PATCH v3 4/6] i40e: Enable 'channel' mode in mqprio for TC configsThe i40e driver is modified to enable the new mqprio hardware offload mode and factor the TCs and queue configuration by creating channel VSIs. In this mode, the priority to traffic class mapping and the user specified queue ranges are used to configure the traffic classes by setting the mode option to 'channel'.
Example:
# tc qdisc add dev eth0 root mqprio num_tc 4\ map 0 0 0 0 1 2 2 3 queues 2 at 0 2 at 2 1 at 4 1 at 5\ hw 1 mode channel# tc qdisc show dev eth0qdisc mqprio 8038: root tc 4 map 0 0 0 0 1 2 2 3 0 0 0 0 0 0 0 0
queues:(0:1) (2:3) (4:4) (5:5) mode:channel shaper:dcbThe HW channels created are removed and all the queue configuration is set to default when the qdisc is detached from the root of the device.
# tc qdisc del dev eth0 rootThis patch also disables setting up channels via ethtool (ethtool -L) when the TCs are configured using mqprio scheduler.
The patch also limits setting ethtool Rx flow hash indirection (ethtool -X eth0 equal N) to max queues configured via mqprio.
The Rx flow hash indirection input through ethtool should be validated so that it is within in the queue range configured via tc/mqprio. The bound checking is achieved by reporting the current rss size to the kernel when queues are configured via mqprio.Example:
# tc qdisc add dev eth0 root mqprio num_tc 4\ map 0 0 0 1 0 2 3 0 queues 2 at 0 4 at 2 8 at 6 11 at 14\ hw 1 mode channel# ethtool -X eth0 equal 12 Cannot set RX flow hash configuration: Invalid argumentv3: Changes to incorporate new mqprio mode option. Minor clean up of
setup_tcerror handling based on Shannon's comments.
v2: Clean up__i40e_setup_tc()andi40e_setup_tc()to work with Jiri's changes to thendo_setup_tcinterface which now takes a type and thetype_datafor the offload.
No need to disable ATR in MQPRIO mode.
Quoted 7 lines
Signed-off-by Amritha Nambiar <[email protected]>--- drivers/net/ethernet/intel/i40e/i40e.h | 3 drivers/net/ethernet/intel/i40e/i40e_ethtool.c | 8 drivers/net/ethernet/intel/i40e/i40e_main.c | 457 ++++++++++++++++++------ 3 files changed, 362 insertions(+), 106 deletions(-)
RFC PATCH v3 5/6 i40e: Refactor VF BW rate limiting
In reply to Amritha NambiarThis patch refactors the BW rate limiting for Tx traffic on the VF to be reused in the next patch for rate limiting Tx traffic for the VSIs on the PF as well.
v3: Minor fixes, clean up log messages based on Shannon's comments.
--- drivers/net/ethernet/intel/i40e/i40e.h | 5 ++ drivers/net/ethernet/intel/i40e/i40e_main.c | 64 ++++++++++++++++++++ drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 45 +------------- 3 files changed, 71 insertions(+), 43 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +5 -0
@@ -128,6 +128,10 @@/* default to trying for four seconds */ #define I40E_TRY_LINK_TIMEOUT (4 * HZ) +/* BW rate limiting */ +#define I40E_BW_CREDIT_DIVISOR 50/* 50Mbps per BW credit */ +#define I40E_MAX_BW_INACTIVE_ACCUM 4/* accumulate 4 credits max */ +/* driver state flags */ enum i40e_state_t { __I40E_TESTING,
@@ -1042,4 +1046,5 @@ static inline bool i40e_enabled_xdp_vsi(struct i40e_vsi *vsi) }int i40e_create_queue_channel (struct i40e_vsi *vsi,struct i40e_channel *ch); +int i40e_set_bw_limit (struct i40e_vsi *vsi,u16 seid,u64 max_tx_rate);#endif /* _I40E_H_ */
drivers/net/ethernet/intel/i40e/i40e_main.c +64 -0
70 lines not on this page · raw message
drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c +2 -43
65 lines not on this page · raw message
Bowers
RFC PATCH v3 5/6 i40e: Refactor VF BW rate limiting
Quoted 15 lines
-----Original Message-----
From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of Amritha Nambiar
Sent: Thursday, September 7, 2017 4:00 AM
To: intel-wired-lan at lists.osuosl.org; Kirsher, Jeffrey T <[email protected]>
Cc: netdev at vger.kernel.org
Subject: [Intel-wired-lan] [RFC PATCH v3 5/6] i40e: Refactor VF BW rate limitingThis patch refactors the BW rate limiting for Tx traffic on the VF to be reused in the next patch for rate limiting Tx traffic for the VSIs on the PF as well.
v3: Minor fixes, clean up log messages based on Shannon's comments.
Quoted 7 lines
Signed-off-by Amritha Nambiar <[email protected]>--- drivers/net/ethernet/intel/i40e/i40e.h | 5 ++ drivers/net/ethernet/intel/i40e/i40e_main.c | 64 ++++++++++++++++++++ drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 45 +------------- 3 files changed, 71 insertions(+), 43 deletions(-)
RFC PATCH v3 6/6 i40e: Add support setting TC max bandwidth rates
In reply to Amritha NambiarThis patch enables setting up maximum Tx rates for the traffic classes in i40e. The maximum rate is offloaded to the hardware through the mqprio framework by specifying the mode option as 'channel' and shaper option as 'bw_rlimit' and is configured for the VSI. Configuring minimum Tx rate limit is not supported in the device. The minimum usable value for Tx rate is 50Mbps.
Example:
# tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\ queues 4 at 0 4 at 4 hw 1 mode channel shaper bw_rlimit\ max_rate 4Gbit 5Gbit
To dump the bandwidth rates:
# tc qdisc show dev eth0
qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0
queues:(0:3) (4:7)
mode:channel
shaper:bw_rlimit max_rate:4Gbit 5Gbit
--- drivers/net/ethernet/intel/i40e/i40e.h | 2 + drivers/net/ethernet/intel/i40e/i40e_main.c | 101 +++++++++++++++++++++++++-- 2 files changed, 94 insertions(+), 9 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +2 -0
@@ -359,6 +359,8 @@ struct i40e_channel {u8 enabled_tc;struct i40e_aqc_vsi_properties_data info; +u64 max_tx_rate; +/* track this channel belongs to which VSI */ struct i40e_vsi *parent_vsi; };
drivers/net/ethernet/intel/i40e/i40e_main.c +92 -9
192 lines not on this page · raw message
Bowers
RFC PATCH v3 6/6 i40e: Add support setting TC max bandwidth rates
Quoted 30 lines
-----Original Message-----
From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of Amritha Nambiar
Sent: Thursday, September 7, 2017 4:01 AM
To: intel-wired-lan at lists.osuosl.org; Kirsher, Jeffrey T <[email protected]>
Cc: netdev at vger.kernel.org
Subject: [Intel-wired-lan] [RFC PATCH v3 6/6] i40e: Add support setting TC max bandwidth ratesThis patch enables setting up maximum Tx rates for the traffic classes in i40e. The maximum rate is offloaded to the hardware through the mqprio framework by specifying the mode option as 'channel' and shaper option as '
bw_rlimit' and is configured for the VSI. Configuring minimum Tx rate limit is not supported in the device. The minimum usable value for Tx rate is 50Mbps.Example:
# tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\ queues 4 at 0 4 at 4 hw 1 mode channel shaper bw_rlimit\ max_rate 4Gbit 5GbitTo dump the bandwidth rates:
# tc qdisc show dev eth0qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0
queues:(0:3) (4:7)
mode:channel
shaper:bw_rlimit max_rate:4Gbit 5Gbit
Quoted 6 lines
Signed-off-by Amritha Nambiar <[email protected]>--- drivers/net/ethernet/intel/i40e/i40e.h | 2 + drivers/net/ethernet/intel/i40e/i40e_main.c | 101 +++++++++++++++++++++++++-- 2 files changed, 94 insertions(+), 9 deletions(-)
Quoted 42 lines · Amritha Nambiar
The following series introduces a new hardware offload mode in tc/mqprio where the TCs, the queue configurations and bandwidth rate limits are offloaded to the hardware. The existing mqprio framework is extended to configure the queue counts and layout and also added support for rate limiting. This is achieved through new netlink attributes for the 'mode' option which takes values such as 'dcb' (default) and 'channel' and a 'shaper' option for QoS attributes such as bandwidth rate limits in hw mode 1. Legacy devices can fall back to the existing setup supporting hw mode 1 without these additional options where only the TCs are offloaded and then the 'mode' and 'shaper' options defaults to DCB support. The i40e driver enables the new mqprio hardware offload mechanism factoring the TCs, queue configuration and bandwidth rates by creating HW channel VSIs.
In this new mode, the priority to traffic class mapping and the user specified queue ranges are used to configure the traffic class when the 'mode' option is set to 'channel'. This is achieved by creating HW channels(VSI). A new channel is created for each of the traffic class configuration offloaded via mqprio framework except for the first TC (TC0) which is for the main VSI. TC0 for the main VSI is also reconfigured as per user provided queue parameters. Finally, bandwidth rate limits are set on these traffic classes through the shaper attribute by sending these rates in addition to the number of TCs and the queue configurations.
Example:
# tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\ queues 4@0 4@4 hw 1 mode channel shaper bw_rlimit\ min_rate 1Gbit 2Gbit max_rate 4Gbit 5GbitTo dump the bandwidth rates:
# tc qdisc show dev eth0
qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0
queues:(0:3) (4:7)
mode:channel
shaper:bw_rlimit min_rate:1Gbit 2Gbit max_rate:4Gbit 5Gbit---
Quoted 8 lines
Amritha Nambiar (6):
mqprio: Introduce new hardware offload mode and shaper in mqprio
i40e: Add macro for PF reset bit
i40e: Add infrastructure for queue channel support
i40e: Enable 'channel' mode in mqprio for TC configs
i40e: Refactor VF BW rate limiting
i40e: Add support setting TC max bandwidth rates
It would be nice to know what has changed since the last review, either summarized here or in the individual patch files. This helps in knowing how much attention should be given to this new set of patches, and encourages further review. I don't remember seeing any responses to my previous comments, and it looks like not all of them were acted upon.
sln
Quoted 18 lines
drivers/net/ethernet/intel/i40e/i40e.h | 44 + drivers/net/ethernet/intel/i40e/i40e_debugfs.c | 3 drivers/net/ethernet/intel/i40e/i40e_ethtool.c | 8 drivers/net/ethernet/intel/i40e/i40e_main.c | 1463 +++++++++++++++++--- drivers/net/ethernet/intel/i40e/i40e_txrx.h | 2 drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 50 - include/net/pkt_cls.h | 9 include/uapi/linux/pkt_sched.h | 32 net/sched/sch_mqprio.c | 183 ++- 9 files changed, 1551 insertions(+), 243 deletions(-)--
_______________________________________________
Intel-wired-lan mailing list
[email protected]
https://lists.osuosl.org/mailman/listinfo/intel-wired-lan
Quoted 52 lines · Shannon Nelson
Quoted 42 lines · Amritha Nambiar
The following series introduces a new hardware offload mode in tc/mqprio where the TCs, the queue configurations and bandwidth rate limits are offloaded to the hardware. The existing mqprio framework is extended to configure the queue counts and layout and also added support for rate limiting. This is achieved through new netlink attributes for the 'mode' option which takes values such as 'dcb' (default) and 'channel' and a 'shaper' option for QoS attributes such as bandwidth rate limits in hw mode 1. Legacy devices can fall back to the existing setup supporting hw mode 1 without these additional options where only the TCs are offloaded and then the 'mode' and 'shaper' options defaults to DCB support. The i40e driver enables the new mqprio hardware offload mechanism factoring the TCs, queue configuration and bandwidth rates by creating HW channel VSIs.
In this new mode, the priority to traffic class mapping and the user specified queue ranges are used to configure the traffic class when the 'mode' option is set to 'channel'. This is achieved by creating HW channels(VSI). A new channel is created for each of the traffic class configuration offloaded via mqprio framework except for the first TC (TC0) which is for the main VSI. TC0 for the main VSI is also reconfigured as per user provided queue parameters. Finally, bandwidth rate limits are set on these traffic classes through the shaper attribute by sending these rates in addition to the number of TCs and the queue configurations.
Example:
# tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\ queues 4 at 0 4 at 4 hw 1 mode channel shaper bw_rlimit\ min_rate 1Gbit 2Gbit max_rate 4Gbit 5GbitTo dump the bandwidth rates:
# tc qdisc show dev eth0
qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0
queues:(0:3) (4:7)
mode:channel
shaper:bw_rlimit min_rate:1Gbit 2Gbit max_rate:4Gbit 5Gbit---
Quoted 8 lines
Amritha Nambiar (6):
mqprio: Introduce new hardware offload mode and shaper in mqprio
i40e: Add macro for PF reset bit
i40e: Add infrastructure for queue channel support
i40e: Enable 'channel' mode in mqprio for TC configs
i40e: Refactor VF BW rate limiting
i40e: Add support setting TC max bandwidth rates
Quoted 5 lines
It would be nice to know what has changed since the last review, either summarized here or in the individual patch files. This helps in knowing how much attention should be given to this new set of patches, and encourages further review. I don't remember seeing any responses to my previous comments, and it looks like not all of them were acted upon.
For all those patch files that have changed since the last revision, I have captured all the new changes in the section titled "v3: " of each patch file. Also, I had replied to most of your comments that they'll be fixed in v3 and the one in the mqprio patch that error handling was already being done and these responses can be verified in patchwork. I missed to respond to your comment regarding supporting macvlan offloads through these channels, that will not be a part of this series and will be looked upon at a later time in a subsequent patch series.
Quoted 21 lines
sln
Quoted 18 lines
drivers/net/ethernet/intel/i40e/i40e.h | 44 + drivers/net/ethernet/intel/i40e/i40e_debugfs.c | 3 drivers/net/ethernet/intel/i40e/i40e_ethtool.c | 8 drivers/net/ethernet/intel/i40e/i40e_main.c | 1463 +++++++++++++++++--- drivers/net/ethernet/intel/i40e/i40e_txrx.h | 2 drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 50 - include/net/pkt_cls.h | 9 include/uapi/linux/pkt_sched.h | 32 net/sched/sch_mqprio.c | 183 ++- 9 files changed, 1551 insertions(+), 243 deletions(-)--
_______________________________________________
Intel-wired-lan mailing list
Intel-wired-lan at osuosl.org
https://lists.osuosl.org/mailman/listinfo/intel-wired-lan
Quoted 1 line · Amritha
On 9/7/2017 9:45 AM, Shannon Nelson wrote:
[...]
Quoted 3 lines
Quoted 3 lines
It would be nice to know what has changed since the last review, either summarized here or in the individual patch files.
[...]
Quoted 4 lines
For all those patch files that have changed since the last revision, I have captured all the new changes in the section titled "v3: " of each patch file.
Mea culpa.
ugh.
sln
Quoted 8 lines · Amritha Nambiar
The following series introduces a new hardware offload mode in tc/mqprio where the TCs, the queue configurations and bandwidth rate limits are offloaded to the hardware. The existing mqprio framework is extended to configure the queue counts and layout and also added support for rate limiting. This is achieved through new netlink attributes for the 'mode' option which takes values such as 'dcb' (default) and 'channel' and a 'shaper' option for QoS attributes such as bandwidth rate limits in hw mode 1.
So "dcb" defines a default priorities to queue mapping?
Quoted 3 lines
Legacy devices can fall back to the existing setup supporting hw mode 1 without these additional options where only the TCs are offloaded and then the 'mode' and 'shaper' options defaults to DCB support.
That's the last part that confuses me, see below.
Quoted 3 lines
The i40e driver enables the new mqprio hardware offload mechanism factoring the TCs, queue configuration and bandwidth rates by creating HW channel VSIs.
I am really confused by what you call hw_mode 1, as I understand it there are really 3 different modes:
- legacy: you don't define any traffic class mapping, but you can still chain this scheduler with a match + action (like what Documentation/networking/multiqueue.txt) you can optionally also add "shaper" arguments, but there should not be any default DCB queue mapping either?
- dcb: a default mapping for traffic classes to queues is defined, optional "shaper" arguments
- channel: (maybe calling that "custom_tc_map" would be clearer?) where you express the exact traffic classes to queue mapping and optional "shaper" arguments
I think that's what you are doing, but I just got confused by the cover letter.
Quoted 12 lines
In this new mode, the priority to traffic class mapping and the user specified queue ranges are used to configure the traffic class when the 'mode' option is set to 'channel'. This is achieved by creating HW channels(VSI). A new channel is created for each of the traffic class configuration offloaded via mqprio framework except for the first TC (TC0) which is for the main VSI. TC0 for the main VSI is also reconfigured as per user provided queue parameters. Finally, bandwidth rate limits are set on these traffic classes through the shaper attribute by sending these rates in addition to the number of TCs and the queue configurations.
Quoted 3 lines
Example:
# tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\ queues 4 at 0 4 at 4 hw 1 mode channel shaper bw_rlimit\
Do you see a case where you can declare a different number of traffic classes say 4 and map them onto just 2 hardware queues? If not, it seems a tiny bit redundant to have to specify both the map and the queue mapping should be sufficient, right?
Quoted 10 lines
min_rate1Gbit 2Gbitmax_rate4Gbit 5GbitTo dump the bandwidth rates:
# tc qdisc show dev eth0
qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0
queues:(0:3) (4:7)
mode:channel
shaper:bw_rlimit min_rate:1Gbit 2Gbit max_rate:4Gbit 5Gbit
I am not well versed into tc, but being able to specify "shaper" arguments has actually value outside of just the multiq scheduler and it could probably be an action on its own?
Quoted 25 lines
---
Amritha Nambiar (6):
mqprio: Introduce new hardware offload mode and shaper in mqprio
i40e: Add macro for PF reset bit
i40e: Add infrastructure for queue channel support
i40e: Enable 'channel' mode in mqprio for TC configs
i40e: Refactor VF BW rate limiting
i40e: Add support setting TC max bandwidth ratesdrivers/net/ethernet/intel/i40e/i40e.h | 44 + drivers/net/ethernet/intel/i40e/i40e_debugfs.c | 3 drivers/net/ethernet/intel/i40e/i40e_ethtool.c | 8 drivers/net/ethernet/intel/i40e/i40e_main.c | 1463 +++++++++++++++++--- drivers/net/ethernet/intel/i40e/i40e_txrx.h | 2 drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 50 - include/net/pkt_cls.h | 9 include/uapi/linux/pkt_sched.h | 32 net/sched/sch_mqprio.c | 183 ++- 9 files changed, 1551 insertions(+), 243 deletions(-)--
Signature · 1 line
Florian
Quoted 11 lines · Florian Fainelli
Quoted 8 lines · Amritha Nambiar
The following series introduces a new hardware offload mode in tc/mqprio where the TCs, the queue configurations and bandwidth rate limits are offloaded to the hardware. The existing mqprio framework is extended to configure the queue counts and layout and
also added support for rate limiting. This is achieved through new netlink attributes for the 'mode' option which takes values such
as 'dcb' (default) and 'channel' and a 'shaper' option for QoS attributes such as bandwidth rate limits in hw mode 1.So "dcb" defines a default priorities to queue mapping?
In the default offload implementation, only the basic hw offload was supported factoring only the 'number of TCs' values. The rest of the queue configuration was being ignored. This is the legacy behavior with hw mode set to 1.
Example:
# tc qdisc add dev eth0 root mqprio num_tc 4 map 0 0 0 0 1 1 1 1 queues 4 at 0 4 at 4 hw 1 I just named this default behavior to 'dcb' while introducing a new offload mechanism.
Quoted 7 lines
Quoted 5 lines
Legacy devices can fall back to the existing setup supporting hw mode 1 without these additional options where only the TCs are offloaded and then the 'mode' and 'shaper' options defaults to DCB
support.That's the last part that confuses me, see below.
As I introduced new options for 'mode' and 'shaper', I set the defaults for these options to 'dcb' so existing offloaders can continue to work without supporting these new options. Patch 1 has a detailed description on how this is done.
Quoted 7 lines
Quoted 4 lines
The i40e driver enables the new mqprio hardware offload mechanism factoring the TCs, queue configuration and bandwidth rates by creating HW channel VSIs.
I am really confused by what you call
hw_mode1, as I understand it there are really 3 different modes:
There are actually 2 modes now with 'hw' option set to 1, legacy/dcb and channel.
Quoted 9 lines
- legacy: you don't define any traffic class mapping, but you can still chain this scheduler with a match + action (like what
Documentation/networking/multiqueue.txt) you can optionally also add "shaper" arguments, but there should not be any default DCB queue mapping either?- dcb: a default mapping for traffic classes to queues is defined, optional "shaper" arguments
The legacy mode now becomes the dcb mode. In this mode, although the TC values, the queue configurations, prio-tc-mapping are all offloaded to the device, the existing implementation in current drivers support only a basic hw offload factoring only the TC values.
Examples:
# ... num_tc 2 map 0 0 0 0 1 1 1 1 queues 4 at 0 4 at 4 hw 1 # ... num_tc 2 map 0 0 0 0 1 1 1 1 queues 4 at 0 4 at 4 hw 1 mode dcb # ... num_tc 2 map 0 0 0 0 1 1 1 1 queues 4 at 0 4 at 4 hw 1 mode dcb\ shaper dcb
Quoted 4 lines
- channel: (maybe calling that "
custom_tc_map" would be clearer?) where you express the exact traffic classes to queue mapping and optional "shaper" arguments
In the channel mode, a full hw offload is supported, the TC values, the queue configurations and additionally QoS attributes (optional) are all used in the new implementation.
Examples:
# ... num_tc 2 map 0 0 0 0 1 1 1 1 queues 4 at 0 4 at 4 hw 1 mode channel # ... num_tc 2 map 0 0 0 0 1 1 1 1 queues 4 at 0 4 at 4 hw 1 mode channel\ shaper bw_rlimit max_rate 4Gbit 5Gbit
Quoted 16 lines
I think that's what you are doing, but I just got confused by the cover letter.
Quoted 12 lines
In this new mode, the priority to traffic class mapping and the user specified queue ranges are used to configure the traffic class when the 'mode' option is set to 'channel'. This is achieved by creating HW channels(VSI). A new channel is created for each of the traffic class configuration offloaded via mqprio framework except for the first TC (TC0) which is for the main VSI. TC0 for the main VSI is also reconfigured as per user provided queue parameters. Finally, bandwidth rate limits are set on these traffic classes through the shaper attribute by sending these rates in addition to the number of TCs and the queue configurations.
Quoted 7 lines
Quoted 2 lines
Example: # tc qdisc add dev eth0 root mqprio
num_tc2 map 0 0 0 0 1 1 1 1\ queues 4 at 0 4 at 4 hw 1 mode channel shaper bw_rlimit\Do you see a case where you can declare a different number of traffic classes say 4 and map them onto just 2 hardware queues? If not, it seems a tiny bit redundant to have to specify both the map and the queue mapping should be sufficient, right?
This will be subjected to validation of the user input and will be treated as invalid configuration. The 'map' specifies the mapping between user priorities and traffic classes while the queue mapping is the queue layout specifying the queue count and offsets.
Quoted 7 lines
Quoted 7 lines
min_rate1Gbit 2Gbitmax_rate4Gbit 5GbitTo dump the bandwidth rates:
# tc qdisc show dev eth0
Quoted 7 lines
Quoted 3 lines
qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0
queues:(0:3) (4:7) mode:channel shaper:bw_rlimit min_rate:1Gbit
2Gbit max_rate:4Gbit 5GbitI am not well versed into tc, but being able to specify "shaper" arguments has actually value outside of just the multiq scheduler and it could probably be an action on its own?
The mqprio scheduler already supports configuring the traffic classes and enabling support for HW shapers was just a matter to extending it with new netlink based attributes.
Quoted 26 lines
Quoted 24 lines
---
Amritha Nambiar (6): mqprio: Introduce new hardware offload mode and shaper in mqprio i40e: Add macro for PF reset bit i40e: Add infrastructure for queue channel support i40e: Enable 'channel' mode in mqprio for TC configs i40e: Refactor VF BW rate limiting
i40e: Add support setting TC max bandwidth ratesdrivers/net/ethernet/intel/i40e/i40e.h | 44 + drivers/net/ethernet/intel/i40e/i40e_debugfs.c | 3 drivers/net/ethernet/intel/i40e/i40e_ethtool.c | 8 drivers/net/ethernet/intel/i40e/i40e_main.c | 1463 +++++++++++++++++--- drivers/net/ethernet/intel/i40e/i40e_txrx.h | 2 drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 50 - include/net/pkt_cls.h | 9 include/uapi/linux/pkt_sched.h | 32 net/sched/sch_mqprio.c | 183 ++- 9files changed, 1551 insertions(+), 243 deletions(-)
--