The for loop checks whether cur_section is NULL on every iteration, but
we know it can never be NULL as there is another check towards the
bottom of the loop body. Refactor to avoid this unnecessary check.
Also, increment the variable i inline for clarity
Addresses-Coverity: 1496984 ("Null pointer dereferences)
Suggested-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Alex Dewar <alex.dewar90@gmail.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Link: https://lore.kernel.org/r/20200916165748.20927-1-alex.dewar90@gmail.com
Saeed Mahameed says:
====================
mlx5-updates-2020-11-03
This series includes updates to mlx5 software steering component.
1) Few improvements in the DR area, such as removing unneeded checks,
renaming to better general names, refactor in some places, etc.
2) Software steering (DR) Memory management improvements
This patch series contains SW Steering memory management improvements:
using buddy allocator instead of an existing bucket allocator, and
several other optimizations.
The buddy system is a memory allocation and management algorithm
that manages memory in power of two increments.
The algorithm is well-known and well-described, such as here:
https://en.wikipedia.org/wiki/Buddy_memory_allocation
Linux uses this algorithm for managing and allocating physical pages,
as described here:
https://www.kernel.org/doc/gorman/html/understand/understand009.html
In our case, although the algorithm in principal is similar to the
Linux physical page allocator, the "building blocks" and the circumstances
are different: in SW steering, buddy allocator doesn't really allocates
a memory, but rather manages ICM (Interconnect Context Memory) that was
previously allocated and registered.
The ICM memory that is used in SW steering is always power
of 2 (order), so buddy system is a good fit for this.
Patches in this series:
[PATH 4] net/mlx5: DR, Add buddy allocator utilities
This patch adds a modified implementation of a well-known buddy allocator,
adjusted for SW steering needs: the algorithm in principal is similar to
the Linux physical page allocator, but in our case buddy allocator doesn't
really allocate a memory, but rather manages ICM memory that was previously
allocated and registered.
[PATH 5] net/mlx5: DR, Handle ICM memory via buddy allocation instead of bucket management
This patch changes ICM management of SW steering to use buddy-system mechanism
Instead of the previous bucket management.
[PATH 6] net/mlx5: DR, Sync chunks only during free
This patch makes syncing happen only when freeing memory chunks.
[PATH 7] net/mlx5: DR, ICM memory pools sync optimization
This patch adds tracking of pool's "hot" memory and makes the
check whether steering sync is required much shorter and faster.
[PATH 8] net/mlx5: DR, Free buddy ICM memory if it is unused
This patch adds tracking buddy's used ICM memory,
and frees the buddy if all its memory becomes unused.
3) Misc code cleanups
* tag 'mlx5-updates-2020-11-03' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux:
net: mlx5: Replace in_irq() usage
net/mlx5: Cleanup kernel-doc warnings
net/mlx4: Cleanup kernel-doc warnings
net/mlx5e: Validate stop_room size upon user input
net/mlx5: DR, Free unused buddy ICM memory
net/mlx5: DR, ICM memory pools sync optimization
net/mlx5: DR, Sync chunks only during free
net/mlx5: DR, Handle ICM memory via buddy allocation instead of buckets
net/mlx5: DR, Add buddy allocator utilities
net/mlx5: DR, Rename matcher functions to be more HW agnostic
net/mlx5: DR, Rename builders HW specific names
net/mlx5: DR, Remove unused member of action struct
====================
Link: https://lore.kernel.org/r/20201105201242.21716-1-saeedm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Support ECM mode based on cdc_ether with relative mii functions,
when CONFIG_USB_RTL8152 is not set, or the device is not supported
by r8152 driver.
Both r8152 and r8153_ecm would check the return value of
rtl8152_get_version() in porbe(). If rtl8152_get_version()
return none zero value, the r8152 is used for the device
with vendor mode. Otherwise, the r8153_ecm is used for the
device with ECM mode.
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Link: https://lore.kernel.org/r/1394712342-15778-392-Taiwan-albertk@realtek.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.
This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.
Cc: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Cc: Willy Liu <willy.liu@realtek.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.
Cc: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Cc: Willy Liu <willy.liu@realtek.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
It seems there are cases where the interrupts are handled by another
entity (ie an IRQ controller embedded inside the PHY) and do not need
any other interraction from phylib. For this kind of PHYs, like the
RTL8366RB, add the genphy_handle_interrupt_no_ack() function which just
triggers the link state machine.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.
This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.
This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.
This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.
Cc: Michael Walle <michael@walle.cc>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.
Cc: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Tested-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.
This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.
Cc: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.
Cc: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.
This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.
Cc: Antoine Tenart <atenart@kernel.org>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Tested-by: Vladimir Oltean <olteanv@gmail.com> # VSC8514
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.
Also, remove the .did_interrupt() callback since it's not anymore used.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Tested-by: Vladimir Oltean <olteanv@gmail.com> # VSC8514
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
According to the comment describing the phy_mac_interrupt() function, it
it intended to be used by MAC drivers which have noticed a link change
thus its use in the mscc PHY driver is improper and, most probably, was
added just because phy_trigger_machine() was not exported.
Now that we have acces to trigger the link state machine, use directly
the phy_trigger_machine() function to notify a link change detected by
the PHY driver.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Tested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.
This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.
Cc: Oleksij Rempel <o.rempel@pengutronix.de>
Cc: Michael Walle <michael@walle.cc>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Tested-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.
Cc: Oleksij Rempel <o.rempel@pengutronix.de>
Cc: Michael Walle <michael@walle.cc>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Tested-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The switch has two controllable I/Os which are usually connected to LEDs. This
is useful to immediately visually see the PTP status.
These provide two signals:
* is_gm
This LED can be activated if the current device is the grand master in that
PTP domain.
* sync_good
This LED can be activated if the current device is in sync with the network
time.
Expose these via the LED framework to be controlled via user space
e.g. linuxptp.
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The switch has the ability to take hardware generated time stamps per port for
PTPv2 event messages in Rx and Tx direction. That is useful for achieving needed
time synchronization precision for TSN devices/switches. So add support for it.
There are two directions:
* RX
The switch has a single register per port to capture a timestamp. That
mechanism is not used due to correlation problems. If the software processing
is too slow and a PTPv2 event message is received before the previous one has
been processed, false timestamps will be captured. Therefore, the switch can
do "inline" timestamping which means it can insert the nanoseconds part of
the timestamp directly into the PTPv2 event message. The reserved field (4
bytes) is leveraged for that. This might not be in accordance with (older)
PTP standards, but is the only way to get reliable results.
* TX
In Tx direction there is no correlation problem, because the software and the
driver has to ensure that only one event message is "on the fly". However,
the switch provides also a mechanism to check whether a timestamp is
lost. That can only happen when a timestamp is read and at this point another
message is timestamped. So, that lost bit is checked just in case to indicate
to the user that the driver or the software is somewhat buggy.
Signed-off-by: Kamil Alkhouri <kamil.alkhouri@hs-offenburg.de>
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The switch has internal PTP hardware clocks. Add support for it. There are three
clocks:
* Synchronized
* Syntonized
* Free running
Currently the synchronized clock is exported to user space which is a good
default for the beginning. The free running clock might be exported later
e.g. for implementing 802.1AS-2011/2020 Time Aware Bridges (TAB). The switch
also supports cross time stamping for that purpose.
The implementation adds support setting/getting the time as well as offset and
frequency adjustments. However, the clock only holds a partial timeofday
timestamp. This is why we track the seconds completely in software (see overflow
work and last_ts).
Furthermore, add the PTP multicast addresses into the FDB to forward that
packages only to the CPU port where they are processed by a PTP program.
Signed-off-by: Kamil Alkhouri <kamil.alkhouri@hs-offenburg.de>
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a basic DSA driver for Hirschmann Hellcreek switches. Those switches are
implementing features needed for Time Sensitive Networking (TSN) such as support
for the Time Precision Protocol and various shapers like the Time Aware Shaper.
This driver includes basic support for networking:
* VLAN handling
* FDB handling
* Port statistics
* STP
* Phylink
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
rq->xdp_prog is RCU-protected and should be accessed only with
rcu_access_pointer for the NULL check in mlx5e_poll_rx_cq.
rq->xdp_prog may change on the fly only from one non-NULL value to
another non-NULL value, so the checks in mlx5e_xdp_handle and
mlx5e_poll_rx_cq will have the same result during one NAPI cycle,
meaning that no additional synchronization is needed.
Fixes: fe45386a20 ("net/mlx5e: Use RCU to protect rq->xdp_prog")
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
During driver reload, perform firmware tear-down which results in
firmware losing the configured VXLAN ports. These ports are still
available in the driver's database. Fix this by cleaning up driver's
VXLAN database in the nic unload flow, before firmware tear-down. With
that, minimize mlx5_vxlan_destroy() to remove only what was added in
mlx5_vxlan_create() and warn on leftover UDP ports.
Fixes: 18a2b7f969 ("net/mlx5: convert to new udp_tunnel infrastructure")
Signed-off-by: Aya Levin <ayal@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
When E-switch vport is disabled, querying its hardware address is
unsupported.
Avoid setting extack error log message in such case.
Fixes: f099fde16d ("net/mlx5: E-switch, Support querying port function mac address")
Signed-off-by: Parav Pandit <parav@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
When a rule is duplicated, the refcount of the rule is increased so only
the second deletion of the rule should cause destruction of the FTE.
Currently, the FTE will be destroyed in the first deletion of rule since
the modify_mask will be 0.
Fix it and call to destroy FTE only if all the rules (FTE's children)
have been removed.
Fixes: 718ce4d601 ("net/mlx5: Consolidate update FTE for all removal changes")
Signed-off-by: Maor Gottlieb <maorg@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
async_icosq_lock may be taken from softirq and non-softirq contexts. It
requires protection with spin_lock_bh, otherwise a softirq may be
triggered in the middle of the critical section, and it may deadlock if
it tries to take the same lock. This patch fixes such a scenario by
using spin_lock_bh to disable softirqs on that CPU while inside the
critical section.
Fixes: 8d94b590f1 ("net/mlx5e: Turn XSK ICOSQ into a general asynchronous one")
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
In functions mlx5e_route_lookup_ipv{4|6}() route_dev can be arbitrary net
device and not necessary mlx5 eswitch port representor. As such, in order
to ensure that route_dev is not destroyed concurrent the code needs either
explicitly take reference to the device before releasing reference to
rtable instance or ensure that caller holds rtnl lock. First approach is
chosen as a fix since rtnl lock dependency was intentionally removed from
mlx5 TC layer.
To prevent unprotected usage of route_dev in encap code take a reference to
the device before releasing rt. Don't save direct pointer to the device in
mlx5_encap_entry structure and use ifindex instead. Modify users of
route_dev pointer to properly obtain the net device instance from its
ifindex.
Fixes: 61086f3910 ("net/mlx5e: Protect encap hash table with mutex")
Fixes: 6707f74be8 ("net/mlx5e: Update hw flows when encap source mac changed")
Signed-off-by: Vlad Buslov <vladbu@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Modify header actions are allocated during parse tc actions and only
freed during the flow creation, however, on error flow the allocated
memory is wrongly unfreed.
Fix this by calling dealloc_mod_hdr_actions in __mlx5e_add_fdb_flow
and mlx5e_add_nic_flow error flow.
Fixes: d7e75a325c ("net/mlx5e: Add offloading of E-Switch TC pedit (header re-write) actions")
Fixes: 2f4fe4cab0 ("net/mlx5e: Add offloading of NIC TC pedit (header re-write) actions")
Signed-off-by: Maor Dickman <maord@nvidia.com>
Reviewed-by: Paul Blakey <paulb@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
mlx5_eq_async_int() uses in_irq() to decide whether eq::lock needs to be
acquired and released with spin_[un]lock() or the irq saving/restoring
variants.
The usage of in_*() in drivers is phased out and Linus clearly requested
that code which changes behaviour depending on context should either be
seperated or the context be conveyed in an argument passed by the caller,
which usually knows the context.
mlx5_eq_async_int() knows the context via the action argument already so
using it for the lock variant decision is a straight forward replacement
for in_irq().
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
$ git ls-files *.[ch] | egrep drivers/net/ethernet/mellanox/ | \
xargs scripts/kernel-doc -none
drivers/net/ethernet/mellanox/mlx5/core/fpga/sdk.h:57:
warning: Enum value 'MLX5_FPGA_ACCESS_TYPE_I2C' not described ...
drivers/net/ethernet/mellanox/mlx5/core/fpga/sdk.h:57:
warning: Enum value 'MLX5_FPGA_ACCESS_TYPE_DONTCARE' not described ...
drivers/net/ethernet/mellanox/mlx5/core/fpga/sdk.h:118:
warning: Function parameter or member 'cb_arg' not described ...
drivers/net/ethernet/mellanox/mlx5/core/fpga/sdk.h:160:
warning: Function parameter or member 'conn' not described ...
drivers/net/ethernet/mellanox/mlx5/core/fpga/sdk.h:160:
warning: Excess function parameter 'fdev' description ...
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Reported-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Stop room is a space that may be taken by WQEs in the SQ during a packet
transmit. It is used to check if next packet has enough room in the SQ.
Stop room guarantees this packet can be served and if not, the queue is
stopped, so no more packets are passed to the driver until it's ready.
Currently, stop_room size is calculated and validated upon tx queues
allocation. This makes it impossible to know if user provided valid
input for certain parameters when interface is down.
Instead, store stop_room in mlx5e_sq_param and create
mlx5e_validate_params(), to validate its fields upon user input even
when the interface is down.
Signed-off-by: Vladyslav Tarasiuk <vladyslavt@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Track buddy's used ICM memory, and free it if all
of the buddy's memory bacame unused.
Do this only for STEs.
MODIFY_ACTION buddies are much smaller, so in case there
is a large amount of modify_header actions, which result
in large amount of MODIFY_ACTION buddies, doing this
cleanup during sync will result in performance hit while
not freeing significant amount of memory.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Track the pool's hot ICM memory when freeing/allocating
chunk, so that when checking if the sync is required, just
check if the pool hot memory has reached the sync threshold.
Signed-off-by: Hamdan Igbaria <hamdani@nvidia.com>
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
When freeing chunks, we want to sync the steering
so that all the "hot" memory will be written to ICM
and all the chunks that are in the hot_list will be
actually destroyed.
When allocating from the pool, we don't have a need
to sync the steering, as we're not freeing anything,
and sync might just hurt the performance in terms of
flow-per-second offloaded.
Signed-off-by: Erez Shitrit <erezsh@nvidia.com>
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Till now in order to manage the ICM memory we used bucket
mechanism, which kept a bucket per specified size (sizes were
between 1 block to 2^21 blocks).
Now changing that with buddy-system mechanism, which gives us much
more flexible way to manage the ICM memory.
Its biggest advantage over the bucket is by using the same ICM memory
area for all the sizes of blocks, which reduces the memory consumption.
Signed-off-by: Erez Shitrit <erezsh@nvidia.com>
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Add implementation of SW Steering variation of buddy allocator.
The buddy system for ICM memory uses 2 main data structures:
- Bitmap per order, that keeps the current state of allocated
blocks for this order
- Indicator for the number of available blocks per each order
Signed-off-by: Erez Shitrit <erezsh@nvidia.com>
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Remove flex parser from the matcher function names since
the matcher should not be aware of such HW specific details.
Signed-off-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
We will support multiple STE versions.
The existing naming is not suitable for newer versions.
Removed the HW specific details and renamed with a more
general names.
Signed-off-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
