Eric Dumazet says:
====================
tcp: avoid sending too small packets
tcp_sendmsg() cooks 'large' skbs, that are later split
if needed from tcp_write_xmit().
After a split, the leftover skb size is smaller than the optimal
size, and this causes a performance drop.
In this series, tcp_grow_skb() helper is added to shift
payload from the second skb in the write queue to the first
skb to always send optimal sized skbs.
This increases TSO efficiency, and decreases number of ACK
packets.
====================
Link: https://lore.kernel.org/r/20240418214600.1291486-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
While investigating TCP performance, I found that TCP would
sometimes send big skbs followed by a single MSS skb,
in a 'locked' pattern.
For instance, BIG TCP is enabled, MSS is set to have 4096 bytes
of payload per segment. gso_max_size is set to 181000.
This means that an optimal TCP packet size should contain
44 * 4096 = 180224 bytes of payload,
However, I was seeing packets sizes interleaved in this pattern:
172032, 8192, 172032, 8192, 172032, 8192, <repeat>
tcp_tso_should_defer() heuristic is defeated, because after a split of
a packet in write queue for whatever reason (this might be a too small
CWND or a small enough pacing_rate),
the leftover packet in the queue is smaller than the optimal size.
It is time to try to make 'leftover packets' bigger so that
tcp_tso_should_defer() can give its full potential.
After this patch, we can see the following output:
14:13:34.009273 IP6 sender > receiver: Flags [P.], seq 4048380:4098360, ack 1, win 256, options [nop,nop,TS val 3425678144 ecr 1561784500], length 49980
14:13:34.010272 IP6 sender > receiver: Flags [P.], seq 4098360:4148340, ack 1, win 256, options [nop,nop,TS val 3425678145 ecr 1561784501], length 49980
14:13:34.011271 IP6 sender > receiver: Flags [P.], seq 4148340:4198320, ack 1, win 256, options [nop,nop,TS val 3425678146 ecr 1561784502], length 49980
14:13:34.012271 IP6 sender > receiver: Flags [P.], seq 4198320:4248300, ack 1, win 256, options [nop,nop,TS val 3425678147 ecr 1561784503], length 49980
14:13:34.013272 IP6 sender > receiver: Flags [P.], seq 4248300:4298280, ack 1, win 256, options [nop,nop,TS val 3425678148 ecr 1561784504], length 49980
14:13:34.014271 IP6 sender > receiver: Flags [P.], seq 4298280:4348260, ack 1, win 256, options [nop,nop,TS val 3425678149 ecr 1561784505], length 49980
14:13:34.015272 IP6 sender > receiver: Flags [P.], seq 4348260:4398240, ack 1, win 256, options [nop,nop,TS val 3425678150 ecr 1561784506], length 49980
14:13:34.016270 IP6 sender > receiver: Flags [P.], seq 4398240:4448220, ack 1, win 256, options [nop,nop,TS val 3425678151 ecr 1561784507], length 49980
14:13:34.017269 IP6 sender > receiver: Flags [P.], seq 4448220:4498200, ack 1, win 256, options [nop,nop,TS val 3425678152 ecr 1561784508], length 49980
14:13:34.018276 IP6 sender > receiver: Flags [P.], seq 4498200:4548180, ack 1, win 256, options [nop,nop,TS val 3425678153 ecr 1561784509], length 49980
14:13:34.019259 IP6 sender > receiver: Flags [P.], seq 4548180:4598160, ack 1, win 256, options [nop,nop,TS val 3425678154 ecr 1561784510], length 49980
With 200 concurrent flows on a 100Gbit NIC, we can see a reduction
of TSO packets (and ACK packets) of about 30 %.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240418214600.1291486-4-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_write_xmit() calls tcp_init_tso_segs()
to set gso_size and gso_segs on the packet.
tcp_init_tso_segs() requires the stack to maintain
an up to date tcp_skb_pcount(), and this makes sense
for packets in rtx queue. Not so much for packets
still in the write queue.
In the following patch, we don't want to deal with
tcp_skb_pcount() when moving payload from 2nd
skb to 1st skb in the write queue.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240418214600.1291486-3-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_cwnd_test() has a special handing for the last packet in
the write queue if it is smaller than one MSS and has the FIN flag.
This is in violation of TCP RFC, and seems quite dubious.
This packet can be sent only if the current CWND is bigger
than the number of packets in flight.
Making tcp_cwnd_test() result independent of the first skb
in the write queue is needed for the last patch of the series.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240418214600.1291486-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tariq Toukan says:
====================
mlx5e per-queue coalescing
This patchset adds ethtool per-queue coalescing support for the mlx5e
driver.
The series introduce some changes needed as preparations for the final
patch which adds the support and implements the callbacks. Main
changes:
- DIM code movements into its own header file.
- Switch to dynamic allocation of the DIM struct in the RQs/SQs.
- Allow coalescing config change without channels reset when possible.
====================
Link: https://lore.kernel.org/r/20240419080445.417574-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When CQE mode or DIM state is changed, gracefully reconfigure channels to
handle new configuration. Previously, would create new channels that would
reflect the changes rather than update the original channels.
Co-developed-by: Nabil S. Alramli <dev@nalramli.com>
Signed-off-by: Nabil S. Alramli <dev@nalramli.com>
Co-developed-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://lore.kernel.org/r/20240419080445.417574-5-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Make it possible for the DIM structure to be torn down while an SQ or RQ is
still active. Changing the CQ period mode is an example where the previous
sampling done with the DIM structure would need to be invalidated.
Co-developed-by: Nabil S. Alramli <dev@nalramli.com>
Signed-off-by: Nabil S. Alramli <dev@nalramli.com>
Co-developed-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://lore.kernel.org/r/20240419080445.417574-4-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Use core DIM CQ period mode enum values for the CQ parameter for the period
mode. Translate the value to the specific mlx5 device constant for the
selected period mode when creating a CQ. Avoid needing to translate mlx5
device constants to DIM constants for core DIM functionality.
Co-developed-by: Nabil S. Alramli <dev@nalramli.com>
Signed-off-by: Nabil S. Alramli <dev@nalramli.com>
Co-developed-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://lore.kernel.org/r/20240419080445.417574-3-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Create a header specifically for DIM-related declarations. Move existing
DIM-specific functionality from en.h. Future DIM-related functionality will
be declared in en/dim.h in subsequent patches.
Co-developed-by: Nabil S. Alramli <dev@nalramli.com>
Signed-off-by: Nabil S. Alramli <dev@nalramli.com>
Co-developed-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://lore.kernel.org/r/20240419080445.417574-2-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pawel Dembicki says:
====================
net: dsa: vsc73xx: convert to PHYLINK and do some cleanup
This patch series is a result of splitting a larger patch series [0],
where some parts needed to be refactored.
The first patch switches from a poll loop to read_poll_timeout.
The second patch is a simple conversion to phylink because adjust_link
won't work anymore.
The third patch is preparation for future use. Using the
"phy_interface_mode_is_rgmii" macro allows for the proper recognition
of all RGMII modes.
Patches 4-5 involve some cleanup: The fourth patch introduces
a definition with the maximum number of ports to avoid using
magic numbers. The next one fills in documentation.
[0] https://patchwork.kernel.org/project/netdevbpf/list/?series=841034&state=%2A&archive=both
====================
Link: https://lore.kernel.org/r/20240417205048.3542839-1-paweldembicki@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This commit adds updates to the documentation describing the structures
used in vsc73xx. This will help prevent kdoc-related issues in the future.
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Pawel Dembicki <paweldembicki@gmail.com>
Link: https://lore.kernel.org/r/20240417205048.3542839-6-paweldembicki@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch introduces a new define: VSC73XX_MAX_NUM_PORTS, which can be
used in the future instead of a hardcoded value.
Currently, the only hardcoded value is vsc->ds->num_ports. It is being
replaced with the new define.
Suggested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Pawel Dembicki <paweldembicki@gmail.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Link: https://lore.kernel.org/r/20240417205048.3542839-5-paweldembicki@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
It's preparation for future use. At this moment, the RGMII port is used
only for a connection to the MAC interface, but in the future, someone
could connect a PHY to it. Using the "phy_interface_mode_is_rgmii" macro
allows for the proper recognition of all RGMII modes.
Suggested-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: Pawel Dembicki <paweldembicki@gmail.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/20240417205048.3542839-4-paweldembicki@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch replaces the adjust_link api with the phylink apis that provide
equivalent functionality.
The remaining functionality from the adjust_link is now covered in the
mac_link_* and mac_config from phylink_mac_ops structure.
Removes:
.adjust_link
Adds phylink_mac_ops structure:
.mac_config
.mac_link_up
.mac_link_down
Signed-off-by: Pawel Dembicki <paweldembicki@gmail.com>
Link: https://lore.kernel.org/r/20240417205048.3542839-3-paweldembicki@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Switch the delay loop during the Arbiter empty check from
vsc73xx_adjust_link() to use read_poll_timeout(). Functionally,
one msleep() call is eliminated at the end of the loop in the timeout
case.
As Russell King suggested:
"This [change] avoids the issue that on the last iteration, the code reads
the register, tests it, finds the condition that's being waiting for is
false, _then_ waits and end up printing the error message - that last
wait is rather useless, and as the arbiter state isn't checked after
waiting, it could be that we had success during the last wait."
Suggested-by: Russell King <linux@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Signed-off-by: Pawel Dembicki <paweldembicki@gmail.com>
Link: https://lore.kernel.org/r/20240417205048.3542839-2-paweldembicki@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
During module probe, regulator 'vin' and 'vdd-io' are used and enabled,
but the vdd-io regulator overwrites the 'vin' regulator pointer. During
remove, only the vdd-io is disabled, as the vin regulator pointer is not
available anymore. When regulator_put() is called during resource
cleanup a kernel warning is given, as the regulator is still enabled.
Store the two regulators in separate pointers and disable both the
regulators on module remove.
Fixes: 49d22c70aa ("NFC: trf7970a: Add device tree option of 1.8 Volt IO voltage")
Signed-off-by: Paul Geurts <paul_geurts@live.nl>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Link: https://lore.kernel.org/r/DB7PR09MB26847A4EBF88D9EDFEB1DA0F950E2@DB7PR09MB2684.eurprd09.prod.outlook.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Current maintainer Douglas Miller has left IBM and no replacement has
been assigned for the driver. The eHEA hardware was last used on
IBM POWER7 systems, the last of which reached end-of-support at the
end of 2020.
Signed-off-by: David Christensen <drc@linux.ibm.com>
Reviewed-by: Pradeep Satyanarayana <pradeeps@linux.ibm.com>
Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc)
Link: https://lore.kernel.org/r/20240418195517.528577-1-drc@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This commit cleans up the uapi for vhost_vdpa by
better naming some of the enums which report blk
information to user space, and they are not
in any official releases yet.
Fixes: 1ac61ddfee ("vDPA: report virtio-blk flush info to user space")
Fixes: ae1374b7f7 ("vDPA: report virtio-block read-only info to user space")
Fixes: 330b8aea69 ("vDPA: report virtio-block max segment size to user space")
Signed-off-by: Zhu Lingshan <lingshan.zhu@intel.com>
Message-Id: <20240415111047.1047774-1-lingshan.zhu@intel.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
- fix a few more deadlocks in recovery
- fix u32/u64 issues in mi_btree_bitmap
- btree key cache shrinker now actually frees, with more instrumentation
coming so we can verify that it's working correctly more easily in the
future
-----BEGIN PGP SIGNATURE-----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=lCNd
-----END PGP SIGNATURE-----
Merge tag 'bcachefs-2024-04-22' of https://evilpiepirate.org/git/bcachefs
Pull bcachefs fixes from Kent Overstreet:
"Nothing too crazy in this one, and it looks like (fingers crossed) the
recovery and repair issues are settling down - although there's going
to be a long tail there, as we've still yet to really ramp up on error
injection or syzbot.
- fix a few more deadlocks in recovery
- fix u32/u64 issues in mi_btree_bitmap
- btree key cache shrinker now actually frees, with more
instrumentation coming so we can verify that it's working
correctly more easily in the future"
* tag 'bcachefs-2024-04-22' of https://evilpiepirate.org/git/bcachefs:
bcachefs: If we run merges at a lower watermark, they must be nonblocking
bcachefs: Fix inode early destruction path
bcachefs: Fix deadlock in journal write path
bcachefs: Tweak btree key cache shrinker so it actually frees
bcachefs: bkey_cached.btree_trans_barrier_seq needs to be a ulong
bcachefs: Fix missing call to bch2_fs_allocator_background_exit()
bcachefs: Check for journal entries overruning end of sb clean section
bcachefs: Fix bio alloc in check_extent_checksum()
bcachefs: fix leak in bch2_gc_write_reflink_key
bcachefs: KEY_TYPE_error is allowed for reflink
bcachefs: Fix bch2_dev_btree_bitmap_marked_sectors() shift
bcachefs: make sure to release last journal pin in replay
bcachefs: node scan: ignore multiple nodes with same seq if interior
bcachefs: Fix format specifier in validate_bset_keys()
bcachefs: Fix null ptr deref in twf from BCH_IOCTL_FSCK_OFFLINE
To get the changes in:
2855c2a782 ("vhost-vdpa: change ioctl # for VDPA_GET_VRING_SIZE")
1496c47065 ("vhost-vdpa: uapi to support reporting per vq size")
To pick up these changes and support them:
$ tools/perf/trace/beauty/vhost_virtio_ioctl.sh > before
$ cp include/uapi/linux/vhost.h tools/perf/trace/beauty/include/uapi/linux/vhost.h
$ tools/perf/trace/beauty/vhost_virtio_ioctl.sh > after
$ diff -u before after
--- before 2024-04-22 13:39:37.185674799 -0300
+++ after 2024-04-22 13:39:52.043344784 -0300
@@ -50,5 +50,6 @@
[0x7F] = "VDPA_GET_VRING_DESC_GROUP",
[0x80] = "VDPA_GET_VQS_COUNT",
[0x81] = "VDPA_GET_GROUP_NUM",
+ [0x82] = "VDPA_GET_VRING_SIZE",
[0x8] = "NEW_WORKER",
};
$
For instance, see how those 'cmd' ioctl arguments get translated, now
VDPA_GET_VRING_SIZE will be as well:
# perf trace -a -e ioctl --max-events=10
0.000 ( 0.011 ms): pipewire/2261 ioctl(fd: 60, cmd: SNDRV_PCM_HWSYNC, arg: 0x1) = 0
21.353 ( 0.014 ms): pipewire/2261 ioctl(fd: 60, cmd: SNDRV_PCM_HWSYNC, arg: 0x1) = 0
25.766 ( 0.014 ms): gnome-shell/2196 ioctl(fd: 14, cmd: DRM_I915_IRQ_WAIT, arg: 0x7ffe4a22c740) = 0
25.845 ( 0.034 ms): gnome-shel:cs0/2212 ioctl(fd: 14, cmd: DRM_I915_IRQ_EMIT, arg: 0x7fd43915dc70) = 0
25.916 ( 0.011 ms): gnome-shell/2196 ioctl(fd: 9, cmd: DRM_MODE_ADDFB2, arg: 0x7ffe4a22c8a0) = 0
25.941 ( 0.025 ms): gnome-shell/2196 ioctl(fd: 9, cmd: DRM_MODE_ATOMIC, arg: 0x7ffe4a22c840) = 0
32.915 ( 0.009 ms): gnome-shell/2196 ioctl(fd: 9, cmd: DRM_MODE_RMFB, arg: 0x7ffe4a22cf9c) = 0
42.522 ( 0.013 ms): gnome-shell/2196 ioctl(fd: 14, cmd: DRM_I915_IRQ_WAIT, arg: 0x7ffe4a22c740) = 0
42.579 ( 0.031 ms): gnome-shel:cs0/2212 ioctl(fd: 14, cmd: DRM_I915_IRQ_EMIT, arg: 0x7fd43915dc70) = 0
42.644 ( 0.010 ms): gnome-shell/2196 ioctl(fd: 9, cmd: DRM_MODE_ADDFB2, arg: 0x7ffe4a22c8a0) = 0
#
This addresses this perf tools build warning:
diff -u tools/perf/trace/beauty/include/uapi/linux/vhost.h include/uapi/linux/vhost.h
But this specific process, usually boring, this time around catch a
problem, namely the addition of VDPA_GET_VRING_SIZE used an ioctl number
already taken, which went on unnoticed and only got caught when the
tools/perf/trace/beauty/vhost_virtio_ioctl.sh script was run as part of
the perf tools process of updating the tools copies of system headers it
uses for creating id->string tables that, well, broke the perf tools
build because there were multiple initializations in the strings table
for the 0x80 entry...
I'm adding here a link to the discussion, that is lacking in the fix for
the reported problem, and a quote from one of the developers involved:
"Thanks a lot for taking care of this! So given the header is actually
buggy pls hang on to this change until I merge the fix for the header
(you were CC'd on the patch). It's great we have this redundancy which
allowed us to catch the bug in time, and many thanks to Namhyung Kim for
reporting the issue!"
This is here as a hint for anyone thinking about ways to automate
checking these issues in a more automated way... ;-)
Link: https://lore.kernel.org/lkml/20240402172151-mutt-send-email-mst@kernel.org
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zhu Lingshan <lingshan.zhu@intel.com>
Link: https://lore.kernel.org/lkml/ZiaW-csEZLKK48BE@x1
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
With the recent PHYLINK changes requiring supported_interfaces to be set,
MV88E6250 family switches like the 88E6020 fail to probe - cmode is
never initialized on these devices, so mv88e6250_phylink_get_caps() does
not set any supported_interfaces flags.
Instead of a cmode, on 88E6250 we have a read-only port mode value that
encodes similar information. There is no reason to bother mapping port
mode to the cmodes of other switch models; instead we introduce a
mv88e6250_setup_supported_interfaces() that is called directly from
mv88e6250_phylink_get_caps().
Fixes: de5c9bf40c ("net: phylink: require supported_interfaces to be filled")
Signed-off-by: Matthias Schiffer <matthias.schiffer@ew.tq-group.com>
Link: https://lore.kernel.org/r/20240417103737.166651-1-matthias.schiffer@ew.tq-group.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The return value of fn() wasn't used for a long time,
so no need to assign it to a variable, addressing a
W=1 warning.
This seems to be - with patches from others posted to
the list before - the last W=1 warning in arch/um/.
Fixes: 22e2430d60 ("x86, um: convert to saner kernel_execve() semantics")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Reviewed-by: Tiwei Bie <tiwei.btw@antgroup.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
The len variable is unused, so not needed, remove it.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Reviewed-by: Tiwei Bie <tiwei.btw@antgroup.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
With W=1 the build complains about a pointer compared to
zero, clearly the result should've been compared.
Fixes: 9807019a62 ("um: Loadable BPF "Firmware" for vector drivers")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Reviewed-by: Tiwei Bie <tiwei.btw@antgroup.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
The code doesn't use 'pid' here, just remove it.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Reviewed-by: Tiwei Bie <tiwei.btw@antgroup.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
We have W=1 warnings on 64-bit because the pid is only
used in branches on 32-bit; move it inside to get rid of
the warnings.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Reviewed-by: Tiwei Bie <tiwei.btw@antgroup.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
Set Makefile SHELL to bash instead of /bin/bash for better portability.
Some systems do not install binaries to /bin, and therefore do not
provide /bin/bash. This includes Linux distros which intentionally avoid
implementing the Filesystem Hierarchy Standard (FHS), such as NixOS and
Guix System.
The recipies inside arch/um/Makefile don't require top-level Bash to
build, and setting "SHELL" to "bash" makes Make pick the Bash executable
from the environment, hence this patch.
Changes since last roll:
- Rebase onto a more recent commit on the master branch.
- Remove a dangling in-text citation from the change log.
- Reword the change log.
Signed-off-by: Yueh-Shun Li <shamrocklee@posteo.net>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net
Signed-off-by: Richard Weinberger <richard@nod.at>
Registering a winch IRQ is racy, an interrupt may occur before the winch is
added to the winch_handlers list.
If that happens, register_winch_irq() adds to that list a winch that is
scheduled to be (or has already been) freed, causing a panic later in
winch_cleanup().
Avoid the race by adding the winch to the winch_handlers list before
registering the IRQ, and rolling back if um_request_irq() fails.
Fixes: 42a359e31a ("uml: SIGIO support cleanup")
Signed-off-by: Roberto Sassu <roberto.sassu@huawei.com>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Richard Weinberger <richard@nod.at>
mlx5 has a built in self-test at driver startup to evaluate if the
platform supports write combining to generate a 64 byte PCIe TLP or
not. This has proven necessary because a lot of common scenarios end up
with broken write combining (especially inside virtual machines) and there
is other way to learn this information.
This self test has been consistently failing on new ARM64 CPU
designs (specifically with NVIDIA Grace's implementation of Neoverse
V2). The C loop around writeq() generates some pretty terrible ARM64
assembly, but historically this has worked on a lot of existing ARM64 CPUs
till now.
We see it succeed about 1 time in 10,000 on the worst effected
systems. The CPU architects speculate that the load instructions
interspersed with the stores makes the WC buffers statistically flush too
often and thus the generation of large TLPs becomes infrequent. This makes
the boot up test unreliable in that it indicates no write-combining,
however userspace would be fine since it uses a ST4 instruction.
Further, S390 has similar issues where only the special zpci_memcpy_toio()
will actually generate large TLPs, and the open coded loop does not
trigger it at all.
Fix both ARM64 and S390 by switching to __iowrite64_copy() which now
provides architecture specific variants that have a high change of
generating a large TLP with write combining. x86 continues to use a
similar writeq loop in the generate __iowrite64_copy().
Fixes: 11f552e217 ("IB/mlx5: Test write combining support")
Link: https://lore.kernel.org/r/6-v3-1893cd8b9369+1925-mlx5_arm_wc_jgg@nvidia.com
Tested-by: Niklas Schnelle <schnelle@linux.ibm.com>
Acked-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Now that the ARM64 arch implementation does the DGH as part of
__iowrite64_copy() there is no reason to open code this in drivers.
Link: https://lore.kernel.org/r/5-v3-1893cd8b9369+1925-mlx5_arm_wc_jgg@nvidia.com
Reviewed-by: Jijie Shao<shaojijie@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The kernel provides driver support for using write combining IO memory
through the __iowriteXX_copy() API which is commonly used as an optional
optimization to generate 16/32/64 byte MemWr TLPs in a PCIe environment.
iomap_copy.c provides a generic implementation as a simple 4/8 byte at a
time copy loop that has worked well with past ARM64 CPUs, giving a high
frequency of large TLPs being successfully formed.
However modern ARM64 CPUs are quite sensitive to how the write combining
CPU HW is operated and a compiler generated loop with intermixed
load/store is not sufficient to frequently generate a large TLP. The CPUs
would like to see the entire TLP generated by consecutive store
instructions from registers. Compilers like gcc tend to intermix loads and
stores and have poor code generation, in part, due to the ARM64 situation
that writeq() does not codegen anything other than "[xN]". However even
with that resolved compilers like clang still do not have good code
generation.
This means on modern ARM64 CPUs the rate at which __iowriteXX_copy()
successfully generates large TLPs is very small (less than 1 in 10,000)
tries), to the point that the use of WC is pointless.
Implement __iowrite32/64_copy() specifically for ARM64 and use inline
assembly to build consecutive blocks of STR instructions. Provide direct
support for 64/32/16 large TLP generation in this manner. Optimize for
common constant lengths so that the compiler can directly inline the store
blocks.
This brings the frequency of large TLP generation up to a high level that
is comparable with older CPU generations.
As the __iowriteXX_copy() family of APIs is intended for use with WC
incorporate the DGH hint directly into the function.
Link: https://lore.kernel.org/r/4-v3-1893cd8b9369+1925-mlx5_arm_wc_jgg@nvidia.com
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: linux-arch@vger.kernel.org
Cc: linux-arm-kernel@lists.infradead.org
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Complete switching the __iowriteXX_copy() routines over to use #define and
arch provided inline/macro functions instead of weak symbols.
S390 has an implementation that simply calls another memcpy
function. Inline this so the callers don't have to do two jumps.
Link: https://lore.kernel.org/r/3-v3-1893cd8b9369+1925-mlx5_arm_wc_jgg@nvidia.com
Acked-by: Niklas Schnelle <schnelle@linux.ibm.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
It is trivial to implement an inline to do this, so provide it in the s390
headers. Like the 64 bit version it should just invoke zpci_memcpy_toio()
with the correct size.
Link: https://lore.kernel.org/r/2-v3-1893cd8b9369+1925-mlx5_arm_wc_jgg@nvidia.com
Acked-by: Niklas Schnelle <schnelle@linux.ibm.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Start switching iomap_copy routines over to use #define and arch provided
inline/macro functions instead of weak symbols.
Inline functions allow more compiler optimization and this is often a
driver hot path.
x86 has the only weak implementation for __iowrite32_copy(), so replace it
with a static inline containing the same single instruction inline
assembly. The compiler will generate the "mov edx,ecx" in a more optimal
way.
Remove iomap_copy_64.S
Link: https://lore.kernel.org/r/1-v3-1893cd8b9369+1925-mlx5_arm_wc_jgg@nvidia.com
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Previously, it was using "remaining args" without leading "@" which isn't
valid. Let's follow snprintf()'s example and use "@...".
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
New driver specific parameter 'tx_scheduling_layers' was introduced.
Describe parameter in the documentation.
Signed-off-by: Michal Wilczynski <michal.wilczynski@intel.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Co-developed-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
It was observed that Tx performance was inconsistent across all queues
and/or VSIs and that it was directly connected to existing 9-layer
topology of the Tx scheduler.
Introduce new private devlink param - tx_scheduling_layers. This parameter
gives user flexibility to choose the 5-layer transmit scheduler topology
which helps to smooth out the transmit performance.
Allowed parameter values are 5 and 9.
Example usage:
Show:
devlink dev param show pci/0000:4b:00.0 name tx_scheduling_layers
pci/0000:4b:00.0:
name tx_scheduling_layers type driver-specific
values:
cmode permanent value 9
Set:
devlink dev param set pci/0000:4b:00.0 name tx_scheduling_layers value 5
cmode permanent
devlink dev param set pci/0000:4b:00.0 name tx_scheduling_layers value 9
cmode permanent
Signed-off-by: Lukasz Czapnik <lukasz.czapnik@intel.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Co-developed-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
Tested-by: Pucha Himasekhar Reddy <himasekharx.reddy.pucha@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Introduce support for Tx scheduler topology change, based on user
selection, from default 9-layer to 5-layer.
Change requires NVM (version 3.20 or newer) and DDP package (OS Package
1.3.30 or newer - available for over a year in linux-firmware, since
commit aed71f296637 in linux-firmware ("ice: Update package to 1.3.30.0"))
https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/commit/?id=aed71f296637
Enable 5-layer topology switch in init path of the driver. To accomplish
that upload of the DDP package needs to be delayed, until change in Tx
topology is finished. To trigger the Tx change user selection should be
changed in NVM using devlink. Then the platform should be rebooted.
Signed-off-by: Michal Wilczynski <michal.wilczynski@intel.com>
Co-developed-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
Tested-by: Pucha Himasekhar Reddy <himasekharx.reddy.pucha@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Adjust the VSI/Aggregator layers based on the number of logical layers
supported by the FW. Currently the VSI and Aggregator layers are
fixed based on the 9 layer scheduler tree layout. Due to performance
reasons the number of layers of the scheduler tree is changing from
9 to 5. It requires a readjustment of these VSI/Aggregator layer values.
Signed-off-by: Raj Victor <victor.raj@intel.com>
Co-developed-by: Michal Wilczynski <michal.wilczynski@intel.com>
Signed-off-by: Michal Wilczynski <michal.wilczynski@intel.com>
Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
Tested-by: Pucha Himasekhar Reddy <himasekharx.reddy.pucha@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
There is a performance issue when the number of VSIs are not multiple
of 8. This is caused due to the max children limitation per node(8) in
9 layer topology. The BW credits are shared evenly among the children
by default. Assume one node has 8 children and the other has 1.
The parent of these nodes share the BW credit equally among them.
Apparently this causes a problem for the first node which has 8 children.
The 9th VM get more BW credits than the first 8 VMs.
Example:
1) With 8 VM's:
for x in 0 1 2 3 4 5 6 7;
do taskset -c ${x} netperf -P0 -H 172.68.169.125 & sleep .1 ; done
tx_queue_0_packets: 23283027
tx_queue_1_packets: 23292289
tx_queue_2_packets: 23276136
tx_queue_3_packets: 23279828
tx_queue_4_packets: 23279828
tx_queue_5_packets: 23279333
tx_queue_6_packets: 23277745
tx_queue_7_packets: 23279950
tx_queue_8_packets: 0
2) With 9 VM's:
for x in 0 1 2 3 4 5 6 7 8;
do taskset -c ${x} netperf -P0 -H 172.68.169.125 & sleep .1 ; done
tx_queue_0_packets: 24163396
tx_queue_1_packets: 24164623
tx_queue_2_packets: 24163188
tx_queue_3_packets: 24163701
tx_queue_4_packets: 24163683
tx_queue_5_packets: 24164668
tx_queue_6_packets: 23327200
tx_queue_7_packets: 24163853
tx_queue_8_packets: 91101417
So on average queue 8 statistics show that 3.7 times more packets were
send there than to the other queues.
The FW starting with version 3.20, has increased the max number of
children per node by reducing the number of layers from 9 to 5. Reflect
this on driver side.
Signed-off-by: Raj Victor <victor.raj@intel.com>
Co-developed-by: Michal Wilczynski <michal.wilczynski@intel.com>
Signed-off-by: Michal Wilczynski <michal.wilczynski@intel.com>
Co-developed-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
Tested-by: Pucha Himasekhar Reddy <himasekharx.reddy.pucha@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Extend devlink_param *set function pointer to take extack as a param.
Sometimes it is needed to pass information to the end user from set
function. It is more proper to use for that netlink instead of passing
message to dmesg.
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
These functions are not called explicitly. Let's just workaround
the -Wmissing-prototypes warnings by declaring them locally similar
to what was done in arch/x86/kernel/asm-offsets_32.c.
This will address below -Wmissing-prototypes warnings:
./arch/x86/um/shared/sysdep/kernel-offsets.h:9:6: warning: no previous prototype for ‘foo’ [-Wmissing-prototypes]
arch/um/os-Linux/main.c:187:7: warning: no previous prototype for ‘__wrap_malloc’ [-Wmissing-prototypes]
arch/um/os-Linux/main.c:208:7: warning: no previous prototype for ‘__wrap_calloc’ [-Wmissing-prototypes]
arch/um/os-Linux/main.c:222:6: warning: no previous prototype for ‘__wrap_free’ [-Wmissing-prototypes]
arch/x86/um/user-offsets.c:17:6: warning: no previous prototype for ‘foo’ [-Wmissing-prototypes]
Signed-off-by: Tiwei Bie <tiwei.btw@antgroup.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
The prototypes for text_poke* are declared in asm/text-patching.h
under arch/x86/include/. It's safe to include this header, as it's
UML-aware (by checking CONFIG_UML_X86).
This will address below -Wmissing-prototypes warnings:
arch/um/kernel/um_arch.c:461:7: warning: no previous prototype for ‘text_poke’ [-Wmissing-prototypes]
arch/um/kernel/um_arch.c:473:6: warning: no previous prototype for ‘text_poke_sync’ [-Wmissing-prototypes]
Signed-off-by: Tiwei Bie <tiwei.btw@antgroup.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
This will address below -Wmissing-prototypes warnings:
arch/um/kernel/initrd.c:18:12: warning: no previous prototype for ‘read_initrd’ [-Wmissing-prototypes]
arch/um/kernel/um_arch.c:408:19: warning: no previous prototype for ‘read_initrd’ [-Wmissing-prototypes]
arch/um/os-Linux/start_up.c:301:12: warning: no previous prototype for ‘parse_iomem’ [-Wmissing-prototypes]
arch/x86/um/ptrace_32.c:15:6: warning: no previous prototype for ‘arch_switch_to’ [-Wmissing-prototypes]
arch/x86/um/ptrace_32.c:101:5: warning: no previous prototype for ‘poke_user’ [-Wmissing-prototypes]
arch/x86/um/ptrace_32.c:153:5: warning: no previous prototype for ‘peek_user’ [-Wmissing-prototypes]
arch/x86/um/ptrace_64.c:111:5: warning: no previous prototype for ‘poke_user’ [-Wmissing-prototypes]
arch/x86/um/ptrace_64.c:171:5: warning: no previous prototype for ‘peek_user’ [-Wmissing-prototypes]
arch/x86/um/syscalls_64.c:48:6: warning: no previous prototype for ‘arch_switch_to’ [-Wmissing-prototypes]
arch/x86/um/tls_32.c:184:5: warning: no previous prototype for ‘arch_switch_tls’ [-Wmissing-prototypes]
Signed-off-by: Tiwei Bie <tiwei.btw@antgroup.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
In some situations a sent packet may get queued in the NIC longer than
than timeout of a ULP. Currently if this happens the ULP may try to reset
the link by destroying the qp and setting up an alternate connection but
will fail because the rxe driver is waiting for the packet to finish
getting sent and be returned to the skb destructor function where the qp
reference holding things up will be dropped. This patch modifies the way
that the qp is passed to the destructor to pass the qp index and not a qp
pointer. Then the destructor will attempt to lookup the qp from its index
and if it fails exit early. This requires taking a reference on the struct
sock rather than the qp allowing the qp to be destroyed while the sk is
still around waiting for the packet to finish.
Link: https://lore.kernel.org/r/20240329145513.35381-15-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Currently the rxe_driver detects packet drops by ip_local_out() which
occur before the packet is sent on the wire and attempts to resend
them. This is redundant with the usual retry mechanism which covers
packets that get dropped in transit to or from the remote node.
The way this is implemented is not robust since it sets need_req_skb and
waits for the number of local skbs outstanding for this qp to drop below a
low water mark. This is racy since the skb may be sent to the destructor
before the requester can set the need_req_skb flag. This will cause a
deadlock in the send path for that qp.
This patch removes this mechanism since the normal retry path will correct
the error and resend the packet and it makes no difference if the packet
is dropped locally or later.
Link: https://lore.kernel.org/r/20240329145513.35381-14-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The rxe send path currently counts the number of skbs outstanding between
the rxe driver and the ethernet driver to prevent too many packets to
accumulate waiting to send. This patch makes the local loopback path
behave the same way. The loopback path forwards the packets to the receive
path which will eventually call kfree_skb on all packets and drop the qp
references. This makes the loopback path more useful for software testing.
Link: https://lore.kernel.org/r/20240329145513.35381-13-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In rxe_send() a ref is taken on the qp to keep it alive until the
kfree_skb() has a chance to call the skb destructor rxe_skb_tx_dtor()
which drops the reference. If the packet has an incorrect protocol the
error path just calls kfree_skb() which will call the destructor which
will drop the ref. Currently the driver also calls rxe_put() which is
incorrect. Additionally since the packets sent to rxe_send() are under the
control of the driver and it only ever produces IPV4 or IPV6 packets the
simplest fix is to remove all the code in this block.
Link: https://lore.kernel.org/r/20240329145513.35381-12-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Fixes: 9eb7f8e44d ("IB/rxe: Move refcounting earlier in rxe_send()")
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Replace calls to rxe_run_task() with rxe_sched_task(). This prevents the
tasks from all running on the same cpu.
This change slightly reduces performance for single qp send and write
benchmarks in loopback mode but greatly improves the performance with
multiple qps because if run task is used all the work tends to be
performed on one cpu. For actual on the wire benchmarks there is no
noticeable performance change.
Link: https://lore.kernel.org/r/20240329145513.35381-11-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
