Commit 8f3d65c166 ("net/smc: fix wait on already cleared link")
introduced link refcounting to avoid waits on already cleared links.
This patch extents and improves the refcounting to cover all
remaining possible cases for this kind of error situation.
Fixes: 15e1b99aad ("net/smc: no WR buffer wait for terminating link group")
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ioana Ciornei says:
====================
net: enetc: add support for software TSO
This series adds support for driver level TSO in the enetc driver.
Ever since the ENETC MDIO erratum workaround is in place, the Tx path is
incurring a penalty (enetc_lock_mdio/enetc_unlock_mdio) for each skb to
be sent out. On top of this, ENETC does not support Tx checksum
offloading. This means that software TSO would help performance just by
the fact that one single mdio lock/unlock sequence would cover multiple
packets sent. On the other hand, checksum needs to be computed in
software since the controller cannot handle it.
This is why, beside using the usual tso_build_hdr()/tso_build_data()
this specific implementation also has to compute the checksum, both IP
and L4, for each resulted segment.
Even with that, the performance improvement of a TCP flow running on a
single A72@1.3GHz of the LS1028A SoC (2.5Gbit/s port) is the following:
before: 1.63 Gbits/sec
after: 2.34 Gbits/sec
Changes in v2:
- declare NETIF_F_HW_CSUM instead of NETIF_F_IP_CSUM in 1/2
- add support for TSO over IPv6 (NETIF_F_TSO6 and csum compute) in 2/2
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds support for driver level TSO in the enetc driver using
the TSO API.
Beside using the usual tso_build_hdr(), tso_build_data() this specific
implementation also has to compute the checksum, both IP and L4, for
each resulted segment. This is because the ENETC controller does not
support Tx checksum offload which is needed in order to take advantage
of TSO.
With the workaround for the ENETC MDIO erratum in place the Tx path of
the driver is forced to lock/unlock for each skb sent. This is why, even
though we are computing the checksum by hand we see the following
improvement in TCP termination on the LS1028A SoC, on a single A72 core
running at 1.3GHz:
before: 1.63 Gbits/sec
after: 2.34 Gbits/sec
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is just a preparation patch for software TSO in the enetc driver.
Unfortunately, ENETC does not support Tx checksum offload which would
normally render TSO, even software, impossible.
Declare NETIF_F_HW_CSUM as part of the feature set and do it at driver
level using skb_csum_hwoffload_help() so that we can move forward and
also add support for TSO in the next patch.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ido Schimmel says:
====================
selftests: forwarding: Add ip6gre tests
This patchset adds forwarding selftests for ip6gre. The tests can be run
with veth pairs or with physical loopbacks.
Patch #1 adds a new config option to determine if 'skip_sw' / 'skip_hw'
flags are used when installing tc filters. By default, it is not set
which means the flags are not used. 'skip_sw' is useful to ensure
traffic is forwarded by the hardware data path.
Patch #2 adds a new helper function.
Patches #3-#4 add the forwarding selftests.
Patch #5 adds a mlxsw-specific selftest to validate correct behavior of
the 'decap_error' trap with IPv6 underlay.
Patches #6-#8 align the corresponding IPv4 underlay test to the IPv6
one.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
As part of adding same test for GRE tunnel with IPv6 underlay, missing
bytes for key were found.
mausezahn does not fill zeros between two colons, so send them
explicitly. For example, use "00:00:00:E9:" instead of ":E9:"
Signed-off-by: Amit Cohen <amcohen@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As part of adding same test for GRE tunnel with IPv6 underlay, an
optional improvement was found - call ipip_payload_get from
ecn_payload_get, so do not duplicate the code which creates the payload.
Signed-off-by: Amit Cohen <amcohen@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As part of adding same test for GRE tunnel with IPv6 underlay, wrong
alignments were found, fix them.
Signed-off-by: Amit Cohen <amcohen@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
IPv6 underlay support was added, add test to check that "decap_error" trap
is triggered under the right conditions and that devlink counters increase.
Signed-off-by: Amit Cohen <amcohen@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add tests that check IPv6-in-IPv6, IPv4-in-IPv6 and MTU change of GRE
tunnel. The tests use hierarchical model - the tunnel is bound to a device
in a different VRF.
These tests can be run with TC_FLAG=skip_sw, so then they will verify
that packets go through hardware as part of enacp and decap phases.
Signed-off-by: Amit Cohen <amcohen@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add tests that check IPv6-in-IPv6, IPv4-in-IPv6 and MTU change of GRE
tunnel. The tests use flat model - overlay and underlay share the same VRF.
These tests can be run with TC_FLAG=skip_sw, so then they will verify
that packets go through hardware as part of enacp and decap phases.
Signed-off-by: Amit Cohen <amcohen@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add function that checks that at least X packets hit the tc rule.
There are cases that it is not possible to catch only the interesting
packets, so then, it is possible to send many packets and verify that at
least this amount of packets hit the rule.
This function will be used in the next patch for general tc rule that
can be used to test both software and hardware.
Signed-off-by: Amit Cohen <amcohen@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add TC_FLAG value to tests topology.
This flag supposed to be skip_sw/skip_hw which means do not filter by
software/hardware.
This can be useful for adding tests to forwarding directory, and be able
to verify that packets go through the hardware.
When the flag is not set or set to 'skip_hw', tests can still be executed
with veth pairs.
Signed-off-by: Amit Cohen <amcohen@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Merge branch 'stmmac-regression-fix'
Herve Codina says:
====================
net: stmmac: fix regression on SPEAr3xx SOC
The ethernet driver used on old SPEAr3xx soc was previously supported on old
kernel. Some regressions were introduced during the different updates leading
to a broken driver for this soc.
This series fixes these regressions and brings back ethernet on SPEAr3xx.
Tested on a SPEAr320 board.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
On SPEAr3xx, ethernet driver is not compatible with the SPEAr600
one.
Indeed, SPEAr3xx uses an earlier version of this IP (v3.40) and
needs some driver tuning compare to SPEAr600.
The v3.40 IP support was added to stmmac driver and this patch
fixes this issue and use the correct compatible string for
SPEAr3xx
Signed-off-by: Herve Codina <herve.codina@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
dwmac 3.40a is an old ip version that can be found on SPEAr3xx soc.
Signed-off-by: Herve Codina <herve.codina@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
dwmac 3.40a is an old ip version that can be found on SPEAr3xx soc.
Signed-off-by: Herve Codina <herve.codina@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some old IPs do not provide the hardware feature register.
On these IPs, this register is read 0x00000000.
In old driver version, this feature was handled but a regression came
with the commit f10a6a3541 ("stmmac: rework get_hw_feature function").
Indeed, this commit removes the return value in dma->get_hw_feature().
This return value was used to indicate the validity of retrieved
information and used later on in stmmac_hw_init() to override
priv->plat data if this hardware feature were valid.
This patch restores the return code in ->get_hw_feature() in order
to indicate the hardware feature validity and override priv->plat
data only if this hardware feature is valid.
Fixes: f10a6a3541 ("stmmac: rework get_hw_feature function")
Signed-off-by: Herve Codina <herve.codina@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Reuse the timeval compat code from core/sock to handle 32-bit and
64-bit timeval structures. Also introduce a new socket option define
to allow using y2038 safe timeval under 32-bit.
The existing behavior of sock_set_timeout and vsock's timeout setter
differ when the time value is out of bounds. vsocks current behavior
is retained at the expense of not being able to share the full
implementation.
This allows the LTP test vsock01 to pass under 32-bit compat mode.
Fixes: fe0c72f3db ("socket: move compat timeout handling into sock.c")
Signed-off-by: Richard Palethorpe <rpalethorpe@suse.com>
Cc: Richard Palethorpe <rpalethorpe@richiejp.com>
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for sharing the implementation of sock_get_timeout.
Signed-off-by: Richard Palethorpe <rpalethorpe@suse.com>
Cc: Richard Palethorpe <rpalethorpe@richiejp.com>
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Stephen reported that ath11k was failing to build on m68k and xtensa:
In file included from <command-line>:0:0:
In function 'ath11k_peer_assoc_h_smps',
inlined from 'ath11k_peer_assoc_prepare' at drivers/net/wireless/ath/ath11k/mac.c:2362:2:
include/linux/compiler_types.h:317:38: error: call to '__compiletime_assert_650' declared with attribute error: FIELD_GET: type of reg too small for mask
_compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
^
include/linux/compiler_types.h:298:4: note: in definition of macro '__compiletime_assert'
prefix ## suffix(); \
^
include/linux/compiler_types.h:317:2: note: in expansion of macro '_compiletime_assert'
_compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
^
include/linux/build_bug.h:39:37: note: in expansion of macro 'compiletime_assert'
#define BUILD_BUG_ON_MSG(cond, msg) compiletime_assert(!(cond), msg)
^
include/linux/bitfield.h:52:3: note: in expansion of macro 'BUILD_BUG_ON_MSG'
BUILD_BUG_ON_MSG((_mask) > (typeof(_reg))~0ull, \
^
include/linux/bitfield.h:108:3: note: in expansion of macro '__BF_FIELD_CHECK'
__BF_FIELD_CHECK(_mask, _reg, 0U, "FIELD_GET: "); \
^
drivers/net/wireless/ath/ath11k/mac.c:2079:10: note: in expansion of macro 'FIELD_GET'
smps = FIELD_GET(IEEE80211_HE_6GHZ_CAP_SM_PS,
Fix the issue by using le16_get_bits() to specify the size explicitly.
Fixes: 6f4d70308e ("ath11k: support SMPS configuration for 6 GHz")
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Tested-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
dp-aux-bus.yaml says we can list an eDP panel as a child of
an eDP controller node to represent the fact that the panel
is connected to the controller's DP AUX bus.
Let's add it to the ps8640 bindings.
Signed-off-by: Philip Chen <philipchen@chromium.org>
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Stephen Boyd <swboyd@chromium.org>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20210929173343.v2.2.I6050ba184b24d887e92692a72ee3054d643d6091@changeid
Add devicetree binding for Qualcomm PCIe Endpoint controller used in
platforms like SDX55. The Endpoint controller is based on the DesignWare
core with Qualcomm-specific wrappers.
Link: https://lore.kernel.org/r/20210920065946.15090-2-manivannan.sadhasivam@linaro.org
Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Rob Herring <robh@kernel.org>
This patch introduces a new inline function to convert a timeval to
usec.
This function will be used also in the next patch.
Signed-off-by: Riccardo Mancini <rickyman7@gmail.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/b95035ec4a125355be8ea843f7275c4580da6398.1629490974.git.rickyman7@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Remove includes that are not needed, to speed up (re)compilation.
Most of these are relics from splitting the driver in a host and a
common part.
[bhelgaas: use driver tag analogous to rcar-ep]
Link: https://lore.kernel.org/r/54bed9a0e6991490ddb2b07e5abfaf40a7a62928.1633090577.git.geert+renesas@glider.be
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
This patch adds a new function in util/mmap.c to duplicate a mmap_cpu_mask.
This new function will be used in patches in the workqueue patchkit.
Signed-off-by: Riccardo Mancini <rickyman7@gmail.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/8943a548ef7a3dd3e015095afad7e9a8b2154c05.1629490974.git.rickyman7@gmail.com
[ bitmap_alloc() was renamed to bitmap_zalloc() ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Remove includes that are not needed, to speed up (re)compilation. Include
<linux/pm_runtime.h>, which is needed, and was included implicitly through
<linux/phy/phy.h> before.
Most of these are relics from splitting the driver in a host and a common
part and adding endpoint support.
[bhelgaas: use driver tag consistent with cadence-ep, designware-ep]
Link: https://lore.kernel.org/r/7c708841a2bf84f85b14a963271c3e99c8ba38a5.1633090444.git.geert+renesas@glider.be
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
Since 7074674e73 ("perf cpumap: Maintain cpumaps ordered and
without dups") perf_cpu_map elements are sorted in ascending order.
This patch improves perf_cpu_map__idx() by using a binary search.
Signed-off-by: Riccardo Mancini <rickyman7@gmail.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/f1543c15797169c21e8b205a4a6751159180580d.1629490974.git.rickyman7@gmail.com
[ Removed 'else' after if + return, declared variables where needed ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Due to deadlocks in the networking subsystem spotted 12 years ago[1],
a workaround was put in place[2] to avoid taking the rtnl lock when it
was not available and restarting the syscall (back to VFS, letting
userspace spin). The following construction is found a lot in the net
sysfs and sysctl code:
if (!rtnl_trylock())
return restart_syscall();
This can be problematic when multiple userspace threads use such
interfaces in a short period, making them to spin a lot. This happens
for example when adding and moving virtual interfaces: userspace
programs listening on events, such as systemd-udevd and NetworkManager,
do trigger actions reading files in sysfs. It gets worse when a lot of
virtual interfaces are created concurrently, say when creating
containers at boot time.
Returning early without hitting the above pattern when the syscall will
fail eventually does make things better. While it is not a fix for the
issue, it does ease things.
[1] https://lore.kernel.org/netdev/49A4D5D5.5090602@trash.net/https://lore.kernel.org/netdev/m14oyhis31.fsf@fess.ebiederm.org/
and https://lore.kernel.org/netdev/20090226084924.16cb3e08@nehalam/
[2] Rightfully, those deadlocks are *hard* to solve.
Signed-off-by: Antoine Tenart <atenart@kernel.org>
Reviewed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
PHYLIB device drivers must match by either numerical PHY ID or by their
.match_phy_device method. Matching by DT is not permitted.
Link: https://lore.kernel.org/r/2b1dc053-8c9a-e3e4-b450-eecdfca3fe16@gmail.com
Tested-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
On the MDIO bus, we have PHYLIB devices and drivers, and we have non-
PHYLIB devices and drivers. PHYLIB devices are MDIO devices that are
wrapped with a struct phy_device.
Trying to bind a MDIO device with a PHYLIB driver results in out-of-
bounds accesses as we attempt to access struct phy_device members. So,
let's prevent this by ensuring that the type of the MDIO device
(indicated by the MDIO_DEVICE_FLAG_PHY flag) matches the type of the
MDIO driver (indicated by the MDIO_DEVICE_IS_PHY flag.)
Link: https://lore.kernel.org/r/2b1dc053-8c9a-e3e4-b450-eecdfca3fe16@gmail.com
Tested-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
leaf classes of ETS qdiscs are served in strict priority or deficit round
robin (DRR), depending on the value of 'nstrict'. Since this value can be
changed while traffic is running, we need to be sure that the active list
of DRR classes can be updated at any time, so:
1) call INIT_LIST_HEAD(&alist) on all leaf classes in .init(), before the
first packet hits any of them.
2) ensure that 'alist' is not overwritten with zeros when a leaf class is
no more strict priority nor DRR (i.e. array elements beyond 'nbands').
Link: https://lore.kernel.org/netdev/YS%2FoZ+f0Nr8eQkzH@dcaratti.users.ipa.redhat.com
Suggested-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Eliminate the following coccicheck warning:
./drivers/net/dsa/rtl8366rb.c:1348:2-3: Unneeded semicolon
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jakub Kicinski says:
====================
net: add a helpers for loading netdev->dev_addr from platform
Similarly to recently added device_get_ethdev_address()
and of_get_ethdev_address() create a helper for drivers
loading mac addr from platform data.
nvmem_get_mac_address() does not have driver callers
so instead of adding a helper there unexport it.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the new platform_get_ethdev_address() helper for the cases
where dev->dev_addr is passed in directly as the destination.
@@
expression dev, net;
@@
- eth_platform_get_mac_address(dev, net->dev_addr)
+ platform_get_ethdev_address(dev, net)
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 406f42fa0d ("net-next: When a bond have a massive amount
of VLANs...") introduced a rbtree for faster Ethernet address look
up. To maintain netdev->dev_addr in this tree we need to make all
the writes to it got through appropriate helpers.
There is a handful of drivers which pass netdev->dev_addr as
the destination buffer to eth_platform_get_mac_address().
Add a helper which takes a dev pointer instead, so it can call
an appropriate helper.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
nvmem_get_mac_address() is only called from of_net.c
we don't need the export.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tony Nguyen says:
====================
100GbE Intel Wired LAN Driver Updates 2021-10-07
Michal Swiatkowski says:
The following patch series introduces basic switchdev model
support in ice driver. Implement the following blocks of
switchdev framework:
- VF port representors creation
- control plane VSI definition
- exception path (a. k. a. "slow-path") - to allow a virtual
switch or linux bridge to receive any packet that doesn't match
any hw filter
- link state management of virtual ports
- query virtual port statistics
Hardware offload support in switchdev mode is out of scope of
this patchset. Devlink interface is used to toggle between
switchdev and legacy (the default) modes of the driver.
---
Note: This series includes the use enum ice_status, however, we have
patches in our queue to remove it from the driver [1]. We are working
through the patches that precede the removal series.
[1] https://patchwork.ozlabs.org/project/intel-wired-lan/list/?series=265957
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
This comment likely refers to the obsolete DLPAR workflow where some
resource state transitions were driven more directly from user space
utilities, but it also seems to contradict itself: "Change isolate state to
Isolate [...]" is at odds with the preceding sentences, and it does not
relate at all to the code that follows.
Remove it to prevent confusion.
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210927201933.76786-5-nathanl@linux.ibm.com
The core DLPAR code supports two actions (add and remove) and three
subtypes of action:
* By DRC index: the action is attempted on a single specified resource.
This is the usual case for processors.
* By indexed count: the action is attempted on a range of resources
beginning at the specified index. This is implemented only by the memory
DLPAR code.
* By count: the lower layer (CPU or memory) is responsible for locating the
specified number of resources to which the action can be applied.
I cannot find any evidence of the "by count" subtype being used by drmgr or
qemu for processors. And when I try to exercise this code, the add case
does not work:
$ ppc64_cpu --smt ; nproc
SMT=8
24
$ printf "cpu remove count 2" > /sys/kernel/dlpar
$ nproc
8
$ printf "cpu add count 2" > /sys/kernel/dlpar
-bash: printf: write error: Invalid argument
$ dmesg | tail -2
pseries-hotplug-cpu: Failed to find enough CPUs (1 of 2) to add
dlpar: Could not handle DLPAR request "cpu add count 2"
$ nproc
8
$ drmgr -c cpu -a -q 2 # this uses the by-index method
Validating CPU DLPAR capability...yes.
CPU 1
CPU 17
$ nproc
24
This is because find_drc_info_cpus_to_add() does not increment drc_index
appropriately during its search.
This is not hard to fix. But the _by_count() functions also have the
property that they attempt to roll back all prior operations if the entire
request cannot be satisfied, even though the rollback itself can encounter
errors. It's not possible to provide transaction-like behavior at this
level, and it's undesirable to have code that can only pretend to do that.
Any users of these functions cannot know what the state of the system is in
the error case. And the error paths are, to my knowledge, impossible to
test without adding custom error injection code.
Summary:
* This code has not worked reliably since its introduction.
* There is no evidence that it is used.
* It contains questionable rollback behaviors in error paths which are
difficult to test.
So let's remove it.
Fixes: ac71380071 ("powerpc/pseries: Add CPU dlpar remove functionality")
Fixes: 90edf184b9 ("powerpc/pseries: Add CPU dlpar add functionality")
Fixes: b015f6bc95 ("powerpc/pseries: Add cpu DLPAR support for drc-info property")
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Tested-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210927201933.76786-4-nathanl@linux.ibm.com
If, due to bugs elsewhere, we get into unregister_cpu_online() with a CPU
that isn't marked hotpluggable, we can emit a warning and return an
appropriate error instead of crashing.
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210927201933.76786-3-nathanl@linux.ibm.com
On pseries, cache nodes in the device tree can be added and removed by the
CPU DLPAR code as well as the partition migration (mobility) code. PowerVM
partitions in dedicated processor mode typically have L2 and L3 cache
nodes.
The CPU DLPAR code has the following shortcomings:
* Cache nodes returned as siblings of a new CPU node by
ibm,configure-connector are silently discarded; only the CPU node is
added to the device tree.
* Cache nodes which become unreferenced in the processor removal path are
not removed from the device tree. This can lead to duplicate nodes when
the post-migration device tree update code replaces cache nodes.
This is long-standing behavior. Presumably it has gone mostly unnoticed
because the two bugs have the property of obscuring each other in common
simple scenarios (e.g. remove a CPU and add it back). Likely you'd notice
only if you cared to inspect the device tree or the sysfs cacheinfo
information.
Booted with two processors:
$ pwd
/sys/firmware/devicetree/base/cpus
$ ls -1d */
l2-cache@2010/
l2-cache@2011/
l3-cache@3110/
l3-cache@3111/
PowerPC,POWER9@0/
PowerPC,POWER9@8/
$ lsprop */l2-cache
l2-cache@2010/l2-cache
00003110 (12560)
l2-cache@2011/l2-cache
00003111 (12561)
PowerPC,POWER9@0/l2-cache
00002010 (8208)
PowerPC,POWER9@8/l2-cache
00002011 (8209)
$ ls /sys/devices/system/cpu/cpu0/cache/
index0 index1 index2 index3
After DLPAR-adding PowerPC,POWER9@10, we see that its associated cache
nodes are absent, its threads' L2+L3 cacheinfo is unpopulated, and it is
missing a cache level in its sched domain hierarchy:
$ ls -1d */
l2-cache@2010/
l2-cache@2011/
l3-cache@3110/
l3-cache@3111/
PowerPC,POWER9@0/
PowerPC,POWER9@10/
PowerPC,POWER9@8/
$ lsprop PowerPC\,POWER9@10/l2-cache
PowerPC,POWER9@10/l2-cache
00002012 (8210)
$ ls /sys/devices/system/cpu/cpu16/cache/
index0 index1
$ grep . /sys/kernel/debug/sched/domains/cpu{0,8,16}/domain*/name
/sys/kernel/debug/sched/domains/cpu0/domain0/name:SMT
/sys/kernel/debug/sched/domains/cpu0/domain1/name:CACHE
/sys/kernel/debug/sched/domains/cpu0/domain2/name:DIE
/sys/kernel/debug/sched/domains/cpu8/domain0/name:SMT
/sys/kernel/debug/sched/domains/cpu8/domain1/name:CACHE
/sys/kernel/debug/sched/domains/cpu8/domain2/name:DIE
/sys/kernel/debug/sched/domains/cpu16/domain0/name:SMT
/sys/kernel/debug/sched/domains/cpu16/domain1/name:DIE
When removing PowerPC,POWER9@8, we see that its cache nodes are left
behind:
$ ls -1d */
l2-cache@2010/
l2-cache@2011/
l3-cache@3110/
l3-cache@3111/
PowerPC,POWER9@0/
When DLPAR is combined with VM migration, we can get duplicate nodes. E.g.
removing one processor, then migrating, adding a processor, and then
migrating again can result in warnings from the OF core during
post-migration device tree updates:
Duplicate name in cpus, renamed to "l2-cache@2011#1"
Duplicate name in cpus, renamed to "l3-cache@3111#1"
and nodes with duplicated phandles in the tree, making lookup behavior
unpredictable:
$ lsprop l[23]-cache@*/ibm,phandle
l2-cache@2010/ibm,phandle
00002010 (8208)
l2-cache@2011#1/ibm,phandle
00002011 (8209)
l2-cache@2011/ibm,phandle
00002011 (8209)
l3-cache@3110/ibm,phandle
00003110 (12560)
l3-cache@3111#1/ibm,phandle
00003111 (12561)
l3-cache@3111/ibm,phandle
00003111 (12561)
Address these issues by:
* Correctly processing siblings of the node returned from
dlpar_configure_connector().
* Removing cache nodes in the CPU remove path when it can be determined
that they are not associated with other CPUs or caches.
Use the of_changeset API in both cases, which allows us to keep the error
handling in this code from becoming more complex while ensuring that the
device tree cannot become inconsistent.
Fixes: ac71380071 ("powerpc/pseries: Add CPU dlpar remove functionality")
Fixes: 90edf184b9 ("powerpc/pseries: Add CPU dlpar add functionality")
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Tested-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210927201933.76786-2-nathanl@linux.ibm.com