When destroying a queue, when calling sock_release, the network stack
might need to allocate an skb to send a FIN/RST. When that happens
during memory pressure, there is a need to reclaim memory, which
in turn may ask the nvme-tcp device to write out dirty pages, however
this is not possible due to a ctrl teardown that is going on.
Set PF_MEMALLOC to the task that releases the socket to grant access
to PF_MEMALLOC reserves. In addition, do the same for the nvme-tcp
thread as this may also originate from the swap itself and should
be more resilient to memory pressure situations.
This fixes the following lockdep complaint:
--
======================================================
WARNING: possible circular locking dependency detected
6.0.0-rc2+ #25 Tainted: G W
------------------------------------------------------
kswapd0/92 is trying to acquire lock:
ffff888114003240 (sk_lock-AF_INET-NVME){+.+.}-{0:0}, at: tcp_sendpage+0x23/0xa0
but task is already holding lock:
ffffffff97e95ca0 (fs_reclaim){+.+.}-{0:0}, at: balance_pgdat+0x987/0x10d0
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #1 (fs_reclaim){+.+.}-{0:0}:
fs_reclaim_acquire+0x11e/0x160
kmem_cache_alloc_node+0x44/0x530
__alloc_skb+0x158/0x230
tcp_send_active_reset+0x7e/0x730
tcp_disconnect+0x1272/0x1ae0
__tcp_close+0x707/0xd90
tcp_close+0x26/0x80
inet_release+0xfa/0x220
sock_release+0x85/0x1a0
nvme_tcp_free_queue+0x1fd/0x470 [nvme_tcp]
nvme_do_delete_ctrl+0x130/0x13d [nvme_core]
nvme_sysfs_delete.cold+0x8/0xd [nvme_core]
kernfs_fop_write_iter+0x356/0x530
vfs_write+0x4e8/0xce0
ksys_write+0xfd/0x1d0
do_syscall_64+0x58/0x80
entry_SYSCALL_64_after_hwframe+0x63/0xcd
-> #0 (sk_lock-AF_INET-NVME){+.+.}-{0:0}:
__lock_acquire+0x2a0c/0x5690
lock_acquire+0x18e/0x4f0
lock_sock_nested+0x37/0xc0
tcp_sendpage+0x23/0xa0
inet_sendpage+0xad/0x120
kernel_sendpage+0x156/0x440
nvme_tcp_try_send+0x48a/0x2630 [nvme_tcp]
nvme_tcp_queue_rq+0xefb/0x17e0 [nvme_tcp]
__blk_mq_try_issue_directly+0x452/0x660
blk_mq_plug_issue_direct.constprop.0+0x207/0x700
blk_mq_flush_plug_list+0x6f5/0xc70
__blk_flush_plug+0x264/0x410
blk_finish_plug+0x4b/0xa0
shrink_lruvec+0x1263/0x1ea0
shrink_node+0x736/0x1a80
balance_pgdat+0x740/0x10d0
kswapd+0x5f2/0xaf0
kthread+0x256/0x2f0
ret_from_fork+0x1f/0x30
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0 CPU1
---- ----
lock(fs_reclaim);
lock(sk_lock-AF_INET-NVME);
lock(fs_reclaim);
lock(sk_lock-AF_INET-NVME);
*** DEADLOCK ***
3 locks held by kswapd0/92:
#0: ffffffff97e95ca0 (fs_reclaim){+.+.}-{0:0}, at: balance_pgdat+0x987/0x10d0
#1: ffff88811f21b0b0 (q->srcu){....}-{0:0}, at: blk_mq_flush_plug_list+0x6b3/0xc70
#2: ffff888170b11470 (&queue->send_mutex){+.+.}-{3:3}, at: nvme_tcp_queue_rq+0xeb9/0x17e0 [nvme_tcp]
Fixes: 3f2304f8c6 ("nvme-tcp: add NVMe over TCP host driver")
Reported-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Tested-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
In nvme_tcp_ddgst_update(), sg_init_marker() is called with an
uninitialized scatterlist. This is probably fine, but gcc complains:
CC [M] drivers/nvme/host/tcp.o
In file included from ./include/linux/dma-mapping.h:10,
from ./include/linux/skbuff.h:31,
from ./include/net/net_namespace.h:43,
from ./include/linux/netdevice.h:38,
from ./include/net/sock.h:46,
from drivers/nvme/host/tcp.c:12:
In function ‘sg_mark_end’,
inlined from ‘sg_init_marker’ at ./include/linux/scatterlist.h:356:2,
inlined from ‘nvme_tcp_ddgst_update’ at drivers/nvme/host/tcp.c:390:2:
./include/linux/scatterlist.h:234:11: error: ‘sg.page_link’ is used uninitialized [-Werror=uninitialized]
234 | sg->page_link |= SG_END;
| ~~^~~~~~~~~~~
drivers/nvme/host/tcp.c: In function ‘nvme_tcp_ddgst_update’:
drivers/nvme/host/tcp.c:388:28: note: ‘sg’ declared here
388 | struct scatterlist sg;
| ^~
cc1: all warnings being treated as errors
Use sg_init_table() instead, which basically memset the scatterlist to
zero first before calling sg_init_marker().
Signed-off-by: Nam Cao <namcaov@gmail.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Odroid N2/N2+ support active cooling via gpio-fan controller.
Add fan controls and tip point for cpu and ddr thermal sensor
on this boards.
Signed-off-by: Anand Moon <linux.amoon@gmail.com>
Link: https://lore.kernel.org/r/20221021050906.1158-1-linux.amoon@gmail.com
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
Add SPICC Controller pin nodes for CLK line when idle for Amlogic GXBB
SoCs.
Signed-off-by: Amjad Ouled-Ameur <aouledameur@baylibre.com>
Reviewed-by: Neil Armstrong <neil.armstrong@linaro.org>
Reviewed-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Link: https://lore.kernel.org/r/20221004-up-aml-fix-spi-v4-4-0342d8e10c49@baylibre.com
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
Add SPICC Controller pin nodes for CLK line when idle for Amlogic GXL
SoCs.
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Signed-off-by: Amjad Ouled-Ameur <aouledameur@baylibre.com>
Reviewed-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Link: https://lore.kernel.org/r/20221004-up-aml-fix-spi-v4-3-0342d8e10c49@baylibre.com
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
Now that blk_mq_destroy_queue does not release the queue reference, there
is no need for a second admin queue reference to be held by the
apple_nvme structure.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Sven Peter <sven@svenpeter.dev>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Link: https://lore.kernel.org/r/20221018135720.670094-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Now that blk_mq_destroy_queue does not release the queue reference, there
is no need for a second admin queue reference to be held by the nvme_dev.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Link: https://lore.kernel.org/r/20221018135720.670094-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Now that blk_mq_destroy_queue does not release the queue reference, there
is no need for a second queue reference to be held by the scsi_device.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Link: https://lore.kernel.org/r/20221018135720.670094-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The fact that blk_mq_destroy_queue also drops a queue reference leads
to various places having to grab an extra reference. Move the call to
blk_put_queue into the callers to allow removing the extra references.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Link: https://lore.kernel.org/r/20221018135720.670094-2-hch@lst.de
[axboe: fix fabrics_q vs admin_q conflict in nvme core.c]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
With char becoming unsigned by default, and with `char` alone being
ambiguous and based on architecture, signed chars need to be marked
explicitly as such. This fixes warnings like:
sound/pci/rme9652/hdsp.c:3953 hdsp_channel_buffer_location() warn: 'hdsp->channel_map[channel]' is unsigned
sound/pci/rme9652/hdsp.c:4153 snd_hdsp_channel_info() warn: impossible condition '(hdsp->channel_map[channel] < 0) => (0-255 < 0)'
sound/pci/rme9652/rme9652.c:1833 rme9652_channel_buffer_location() warn: 'rme9652->channel_map[channel]' is unsigned
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20221025000313.546261-1-Jason@zx2c4.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
With char becoming unsigned by default, and with `char` alone being
ambiguous and based on architecture, signed chars need to be marked
explicitly as such. This fixes warnings like:
sound/pci/au88x0/au88x0_core.c:2029 vortex_adb_checkinout() warn: signedness bug returning '(-22)'
sound/pci/au88x0/au88x0_core.c:2046 vortex_adb_checkinout() warn: signedness bug returning '(-12)'
sound/pci/au88x0/au88x0_core.c:2125 vortex_adb_allocroute() warn: 'vortex_adb_checkinout(vortex, (0), en, 0)' is unsigned
sound/pci/au88x0/au88x0_core.c:2170 vortex_adb_allocroute() warn: 'vortex_adb_checkinout(vortex, stream->resources, en, 4)' is unsigned
As well, since one function returns errnos, return an `int` rather than
a `signed char`.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20221024162929.536004-1-Jason@zx2c4.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Commit 46573fd636 ("cpufreq: intel_pstate: hybrid: Rework HWP
calibration") attempted to use the information from CPPC (the nominal
performance in particular) to obtain the scaling factor allowing the
frequency to be computed if the HWP performance level of the given CPU
is known or vice versa.
However, it turns out that on some platforms this doesn't work, because
the CPPC information on them does not align with the contents of the
MSR_HWP_CAPABILITIES registers.
This basically means that the only way to make intel_pstate work on all
of the hybrid platforms to date is to use the observation that on all
of them the scaling factor between the HWP performance levels and
frequency for P-cores is 78741 (approximately 100000/1.27). For
E-cores it is 100000, which is the same as for all of the non-hybrid
"core" platforms and does not require any changes.
Accordingly, make intel_pstate use 78741 as the scaling factor between
HWP performance levels and frequency for P-cores on all hybrid platforms
and drop the dependency of the HWP calibration code on CPPC.
Fixes: 46573fd636 ("cpufreq: intel_pstate: hybrid: Rework HWP calibration")
Reported-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Acked-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Tested-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: 5.15+ <stable@vger.kernel.org> # 5.15+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Some of the MSR accesses in intel_pstate are carried out on the CPU
that is running the code, but the values coming from them are used
for the performance scaling of the other CPUs.
This is problematic, for example, on hybrid platforms where
MSR_TURBO_RATIO_LIMIT for P-cores and E-cores is different, so the
values read from it on a P-core are generally not applicable to E-cores
and the other way around.
For this reason, make the driver access all MSRs on the target CPU on
platforms using the "core" pstate_funcs callbacks which is the case for
all of the hybrid platforms released to date. For this purpose, pass
a CPU argument to the ->get_max(), ->get_max_physical(), ->get_min()
and ->get_turbo() pstate_funcs callbacks and from there pass it to
rdmsrl_on_cpu() or rdmsrl_safe_on_cpu() to access the MSR on the target
CPU.
Fixes: 46573fd636 ("cpufreq: intel_pstate: hybrid: Rework HWP calibration")
Acked-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Tested-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: 5.15+ <stable@vger.kernel.org> # 5.15+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Hybrid sleep is currently hardcoded to only operate with S3 even
on systems that might not support it.
Instead of assuming this mode is what the user wants to use, for
hybrid sleep follow the setting of `mem_sleep_current` which
will respect mem_sleep_default kernel command line and policy
decisions made by the presence of the FADT low power idle bit.
Fixes: 81d45bdf89 ("PM / hibernate: Untangle power_down()")
Reported-and-tested-by: kolAflash <kolAflash@kolahilft.de>
Link: https://bugzilla.kernel.org/show_bug.cgi?id=216574
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Set inner tunnel offset to the GRE header, this is redundant to existing
transport header offset, but this normalizes the handling of the tunnel
header regardless its location in the layering. GRE version 0 is overloaded
with RFCs, the type decorator in the inner expression might also be useful
to interpret matching fields from the netlink delinearize path in userspace.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Geneve tunnel header may contain options, parse geneve header and update
offset to point to the link layer header according to the opt_len field.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Add support for inner meta matching on:
- NFT_META_PROTOCOL: to match on the ethertype, this can be used
regardless tunnel protocol provides no link layer header, in that case
nft_inner sets on the ethertype based on the IP header version field.
- NFT_META_L4PROTO: to match on the layer 4 protocol.
These meta expression are usually autogenerated as dependencies by
userspace nftables.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Add NFT_PKTINFO_INNER_FULL flag to annotate that inner offsets are
available. Store nft_inner_tun_ctx object in percpu area to cache
existing inner offsets for this skbuff.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This new expression allows you to match on the inner headers that are
encapsulated by any of the existing tunneling protocols.
This expression parses the inner packet to set the link, network and
transport offsets, so the existing expressions (with a few updates) can
be reused to match on the inner headers.
The inner expression supports for different tunnel combinations such as:
- ethernet frame over IPv4/IPv6 packet, eg. VxLAN.
- IPv4/IPv6 packet over IPv4/IPv6 packet, eg. IPIP.
- IPv4/IPv6 packet over IPv4/IPv6 + transport header, eg. GRE.
- transport header (ESP or SCTP) over transport header (usually UDP)
The following fields are used to describe the tunnel protocol:
- flags, which describe how to parse the inner headers:
NFT_PAYLOAD_CTX_INNER_TUN, the tunnel provides its own header.
NFT_PAYLOAD_CTX_INNER_ETHER, the ethernet frame is available as inner header.
NFT_PAYLOAD_CTX_INNER_NH, the network header is available as inner header.
NFT_PAYLOAD_CTX_INNER_TH, the transport header is available as inner header.
For example, VxLAN sets on all of these flags. While GRE only sets on
NFT_PAYLOAD_CTX_INNER_NH and NFT_PAYLOAD_CTX_INNER_TH. Then, ESP over
UDP only sets on NFT_PAYLOAD_CTX_INNER_TH.
The tunnel description is composed of the following attributes:
- header size: in case the tunnel comes with its own header, eg. VxLAN.
- type: this provides a hint to userspace on how to delinearize the rule.
This is useful for VxLAN and Geneve since they run over UDP, since
transport does not provide a hint. This is also useful in case hardware
offload is ever supported. The type is not currently interpreted by the
kernel.
- expression: currently only payload supported. Follow up patch adds
also inner meta support which is required by autogenerated
dependencies. The exthdr expression should be supported too
at some point. There is a new inner_ops operation that needs to be
set on to allow to use an existing expression from the inner expression.
This patch adds a new NFT_PAYLOAD_TUN_HEADER base which allows to match
on the tunnel header fields, eg. vxlan vni.
The payload expression is embedded into nft_inner private area and this
private data area is passed to the payload inner eval function via
direct call.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
ipip is an special case, transport and inner header offset are set to
the same offset to use the upcoming inner expression for matching on
inner tunnel headers.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
nft_objref is needed to reference named objects, it makes
no sense to disable it.
Before:
text data bss dec filename
4014 424 0 4438 nft_objref.o
4174 1128 0 5302 nft_objref.ko
359351 15276 864 375491 nf_tables.ko
After:
text data bss dec filename
3815 408 0 4223 nft_objref.o
363161 15692 864 379717 nf_tables.ko
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The current code provokes some kernel-doc warnings:
drivers/char/ipmi/ipmi_msghandler.c:618: warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst
Signed-off-by: Bo Liu <liubo03@inspur.com>
Message-Id: <20221025060436.4372-1-liubo03@inspur.com>
Signed-off-by: Corey Minyard <cminyard@mvista.com>
structure is reduced from 32 to 24 bytes. While at it, also check
that iphdrlen is sane, this is guaranteed for NFPROTO_IPV4 but not
for ingress or bridge, so add checks for this.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Not required to expose this header in nf_tables_core.h, move it to where
it is used, ie. nft_payload.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Document V4L2_CID_HDR_MODE possible values for this sensor.
Signed-off-by: Benjamin Mugnier <benjamin.mugnier@foss.st.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Add device tree binding for the ST VGXY61 camera sensor, and update
MAINTAINERS file.
Signed-off-by: Benjamin Mugnier <benjamin.mugnier@foss.st.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Add V4L2_CID_HDR_MODE as a menu item control to set the HDR mode of the
sensor, and its documentation.
Menu items are not standardized as they differ for each sensors.
Signed-off-by: Benjamin Mugnier <benjamin.mugnier@foss.st.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
This extends the greyscale media bus family originally from
MEDIA_BUS_FMT_Y8_1X8 up to MEDIA_BUS_FMT_Y14_1X14 by adding
MEDIA_BUS_FMT_Y16_1X16, and behaves the same way with 16 bits.
Add its documentation in subdev-formats.rst
Signed-off-by: Benjamin Mugnier <benjamin.mugnier@foss.st.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Add a V4L2 sub-device driver for OmniVision OV4689 image sensor. This
is a 4 Mpx image sensor using the I2C bus for control and the CSI-2
bus for data.
This driver supports following features:
- manual exposure and analog gain control support
- test pattern support
- media controller support
- runtime PM support
- support following resolutions:
+ 2688x1520 at 30 fps
The driver provides all mandatory V4L2 controls for compatibility with
libcamera. The sensor supports 1/2/4-lane CSI-2 modes, but the driver
implements 4 lane mode only at this moment.
Signed-off-by: Mikhail Rudenko <mike.rudenko@gmail.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Add a V4L2 sub-device driver for Omnivision ov08X40 image sensor.
This is a camera sensor using the I2C bus for control and the
CSI-2 bus for data.
This driver supports following features:
- manual exposure and analog/digital gain control support
- vblank/hblank control support
- test pattern support
- media controller support
- runtime PM support
- support following resolutions:
+ 3856x2464 at 30FPS
+ 1928x1208 at 30FPS
Signed-off-by: Jason Chen <jason.z.chen@intel.com>
Signed-off-by: Shawn Tu <shawnx.tu@intel.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Commit 1f391df446 ("media: v4l2-async: Use endpoints in
__v4l2_async_nf_add_fwnode_remote()") changed the data that is stored in
the v4l2_async_subdev internals from the fwnode pointer to the parent
device to the fwnode pointer to the matched endpoint. This broke the
sensor matching code, which relied on the particular fwnode data in the
v4l2_async_subdev internals. Fix this by simply matching the
v4l2_async_subdev pointer, which is already available there.
Reported-by: Daniel Scally <djrscally@gmail.com>
Fixes: fa91f1056f ("[media] exynos4-is: Add support for asynchronous subdevices registration")
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Reviewed-by: Daniel Scally <djrscally@gmail.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
The V4L2 fwnode data structure doesn't get freed on unbind, which leads to
a memleak.
Fixes: e43ccb0a04 ("media: i2c: Add support for the OV5648 image sensor")
Signed-off-by: Rafael Mendonca <rafaelmendsr@gmail.com>
Reviewed-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
Reviewed-by: Paul Kocialkowski <paul.kocialkowski@bootlin.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
There were no runtime PM callbacks registered, leaving regulators being
enabled while the device is suspended on DT systems. Adjust and register
existing power controlling functions to turn them off/on.
Signed-off-by: Hidenori Kobayashi <hidenorik@chromium.org>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Round up allocations with kmalloc_size_roundup() so that build_skb()'s
use of ksize() is always accurate and no special handling of the memory
is needed by KASAN, UBSAN_BOUNDS, nor FORTIFY_SOURCE.
Cc: Rasesh Mody <rmody@marvell.com>
Cc: GR-Linux-NIC-Dev@marvell.com
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: netdev@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20221022021004.gonna.489-kees@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The currently default Round-Robin GPU scheduling can result in starvation
of entities which have a large number of jobs, over entities which have
a very small number of jobs (single digit).
This can be illustrated in the following diagram, where jobs are
alphabetized to show their chronological order of arrival, where job A is
the oldest, B is the second oldest, and so on, to J, the most recent job to
arrive.
---> entities
j | H-F-----A--E--I--
o | --G-----B-----J--
b | --------C--------
s\/ --------D--------
WLOG, assuming all jobs are "ready", then a R-R scheduling will execute them
in the following order (a slice off of the top of the entities' list),
H, F, A, E, I, G, B, J, C, D.
However, to mitigate job starvation, we'd rather execute C and D before E,
and so on, given, of course, that they're all ready to be executed.
So, if all jobs are ready at this instant, the order of execution for this
and the next 9 instances of picking the next job to execute, should really
be,
A, B, C, D, E, F, G, H, I, J,
which is their chronological order. The only reason for this order to be
broken, is if an older job is not yet ready, but a younger job is ready, at
an instant of picking a new job to execute. For instance if job C wasn't
ready at time 2, but job D was ready, then we'd pick job D, like this:
0 +1 +2 ...
A, B, D, ...
And from then on, C would be preferred before all other jobs, if it is ready
at the time when a new job for execution is picked. So, if C became ready
two steps later, the execution order would look like this:
......0 +1 +2 ...
A, B, D, E, C, F, G, H, I, J
This is what the FIFO GPU scheduling algorithm achieves. It uses a
Red-Black tree to keep jobs sorted in chronological order, where picking
the oldest job is O(1) (we use the "cached" structure), and balancing the
tree is O(log n). IOW, it picks the *oldest ready* job to execute now.
The implementation is already in the kernel, and this commit only changes
the default GPU scheduling algorithm to use.
This was tested and achieves about 1% faster performance over the Round
Robin algorithm.
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <Alexander.Deucher@amd.com>
Cc: Direct Rendering Infrastructure - Development <dri-devel@lists.freedesktop.org>
Signed-off-by: Luben Tuikov <luben.tuikov@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221024212634.27230-1-luben.tuikov@amd.com
Signed-off-by: Christian König <christian.koenig@amd.com>
Mat Martineau says:
====================
mptcp: Socket option updates
Patches 1 and 3 refactor a recent socket option helper function for more
generic use, and make use of it in a couple of places.
Patch 2 adds TCP_FASTOPEN_NO_COOKIE functionality to MPTCP sockets,
similar to TCP_FASTOPEN_CONNECT support recently added in v6.1
====================
Link: https://lore.kernel.org/r/20221022004505.160988-1-mathew.j.martineau@linux.intel.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
mptcp_setsockopt_sol_tcp_defer() was doing the same thing as
mptcp_setsockopt_first_sf_only() except for the returned code in case of
error.
Ignoring the error is needed to mimic how TCP_DEFER_ACCEPT is handled
when used with "plain" TCP sockets.
The specific function for TCP_DEFER_ACCEPT can be replaced by the new
mptcp_setsockopt_first_sf_only() helper and errors can be ignored to
stay compatible with TCP. A bit of cleanup.
Suggested-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The goal of this socket option is to configure MPTCP + TFO without
cookie per socket.
It was already possible to enable TFO without a cookie per netns by
setting net.ipv4.tcp_fastopen sysctl knob to the right value. Per route
was also supported by setting 'fastopen_no_cookie' option. This patch
adds a per socket support like it is possible to do with TCP thanks to
TCP_FASTOPEN_NO_COOKIE socket option.
The only thing to do here is to relay the request to the first subflow
like it is already done for TCP_FASTOPEN_CONNECT.
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
There are other socket options that need to act only on the first
subflow, e.g. all TCP_FASTOPEN* socket options.
This is similar to the getsockopt version.
In the next commit, this new mptcp_setsockopt_first_sf_only() helper is
used by other another option.
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Kuniyuki Iwashima says:
====================
soreuseport: Fix broken SO_INCOMING_CPU.
setsockopt(SO_INCOMING_CPU) for UDP/TCP is broken since 4.5/4.6 due to
these commits:
* e32ea7e747 ("soreuseport: fast reuseport UDP socket selection")
* c125e80b88 ("soreuseport: fast reuseport TCP socket selection")
These commits introduced the O(1) socket selection algorithm and removed
O(n) iteration over the list, but it ignores the score calculated by
compute_score(). As a result, it caused two misbehaviours:
* Unconnected sockets receive packets sent to connected sockets
* SO_INCOMING_CPU does not work
The former is fixed by commit acdcecc612 ("udp: correct reuseport
selection with connected sockets"). This series fixes the latter and
adds some tests for SO_INCOMING_CPU.
====================
Link: https://lore.kernel.org/r/20221021204435.4259-1-kuniyu@amazon.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Some highly optimised applications use SO_INCOMING_CPU to make them
efficient, but they didn't test if it's working correctly by getsockopt()
to avoid slowing down. As a result, no one noticed it had been broken
for years, so it's a good time to add a test to catch future regression.
The test does
1) Create $(nproc) TCP listeners associated with each CPU.
2) Create 32 child sockets for each listener by calling
sched_setaffinity() for each CPU.
3) Check if accept()ed sockets' sk_incoming_cpu matches
listener's one.
If we see -EAGAIN, SO_INCOMING_CPU is broken. However, we might not see
any error even if broken; the kernel could miraculously distribute all SYN
to correct listeners. Not to let that happen, we must increase the number
of clients and CPUs to some extent, so the test requires $(nproc) >= 2 and
creates 64 sockets at least.
Test:
$ nproc
96
$ ./so_incoming_cpu
Before the previous patch:
# Starting 12 tests from 5 test cases.
# RUN so_incoming_cpu.before_reuseport.test1 ...
# so_incoming_cpu.c:191:test1:Expected cpu (5) == i (0)
# test1: Test terminated by assertion
# FAIL so_incoming_cpu.before_reuseport.test1
not ok 1 so_incoming_cpu.before_reuseport.test1
...
# FAILED: 0 / 12 tests passed.
# Totals: pass:0 fail:12 xfail:0 xpass:0 skip:0 error:0
After:
# Starting 12 tests from 5 test cases.
# RUN so_incoming_cpu.before_reuseport.test1 ...
# so_incoming_cpu.c:199:test1:SO_INCOMING_CPU is very likely to be working correctly with 3072 sockets.
# OK so_incoming_cpu.before_reuseport.test1
ok 1 so_incoming_cpu.before_reuseport.test1
...
# PASSED: 12 / 12 tests passed.
# Totals: pass:12 fail:0 xfail:0 xpass:0 skip:0 error:0
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Kazuho Oku reported that setsockopt(SO_INCOMING_CPU) does not work
with setsockopt(SO_REUSEPORT) since v4.6.
With the combination of SO_REUSEPORT and SO_INCOMING_CPU, we could
build a highly efficient server application.
setsockopt(SO_INCOMING_CPU) associates a CPU with a TCP listener
or UDP socket, and then incoming packets processed on the CPU will
likely be distributed to the socket. Technically, a socket could
even receive packets handled on another CPU if no sockets in the
reuseport group have the same CPU receiving the flow.
The logic exists in compute_score() so that a socket will get a higher
score if it has the same CPU with the flow. However, the score gets
ignored after the blamed two commits, which introduced a faster socket
selection algorithm for SO_REUSEPORT.
This patch introduces a counter of sockets with SO_INCOMING_CPU in
a reuseport group to check if we should iterate all sockets to find
a proper one. We increment the counter when
* calling listen() if the socket has SO_INCOMING_CPU and SO_REUSEPORT
* enabling SO_INCOMING_CPU if the socket is in a reuseport group
Also, we decrement it when
* detaching a socket out of the group to apply SO_INCOMING_CPU to
migrated TCP requests
* disabling SO_INCOMING_CPU if the socket is in a reuseport group
When the counter reaches 0, we can get back to the O(1) selection
algorithm.
The overall changes are negligible for the non-SO_INCOMING_CPU case,
and the only notable thing is that we have to update sk_incomnig_cpu
under reuseport_lock. Otherwise, the race prevents transitioning to
the O(n) algorithm and results in the wrong socket selection.
cpu1 (setsockopt) cpu2 (listen)
+-----------------+ +-------------+
lock_sock(sk1) lock_sock(sk2)
reuseport_update_incoming_cpu(sk1, val)
.
| /* set CPU as 0 */
|- WRITE_ONCE(sk1->incoming_cpu, val)
|
| spin_lock_bh(&reuseport_lock)
| reuseport_grow(sk2, reuse)
| .
| |- more_socks_size = reuse->max_socks * 2U;
| |- if (more_socks_size > U16_MAX &&
| | reuse->num_closed_socks)
| | .
| | |- RCU_INIT_POINTER(sk1->sk_reuseport_cb, NULL);
| | `- __reuseport_detach_closed_sock(sk1, reuse)
| | .
| | `- reuseport_put_incoming_cpu(sk1, reuse)
| | .
| | | /* Read shutdown()ed sk1's sk_incoming_cpu
| | | * without lock_sock().
| | | */
| | `- if (sk1->sk_incoming_cpu >= 0)
| | .
| | | /* decrement not-yet-incremented
| | | * count, which is never incremented.
| | | */
| | `- __reuseport_put_incoming_cpu(reuse);
| |
| `- spin_lock_bh(&reuseport_lock)
|
|- spin_lock_bh(&reuseport_lock)
|
|- reuse = rcu_dereference_protected(sk1->sk_reuseport_cb, ...)
|- if (!reuse)
| .
| | /* Cannot increment reuse->incoming_cpu. */
| `- goto out;
|
`- spin_unlock_bh(&reuseport_lock)
Fixes: e32ea7e747 ("soreuseport: fast reuseport UDP socket selection")
Fixes: c125e80b88 ("soreuseport: fast reuseport TCP socket selection")
Reported-by: Kazuho Oku <kazuhooku@gmail.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
We miss ISOFS entry in MAINTAINERS file. Add it and write me as the
maintainer there since ISOFS is pretty low effort these days. Less
random patches for Andrew to merge ;-).
Signed-off-by: Jan Kara <jack@suse.cz>
Alex Elder says:
====================
net: ipa: validation cleanup
This series gathers a set of IPA driver cleanups, mostly involving
code that ensures certain things are known to be correct *early*
(either at build or initializatin time), so they can be assumed good
during normal operation.
The first removes three constant symbols, by making a (reasonable)
assumption that a routing table consists of entries for the modem
followed by entries for the AP, with no unused entries between them.
The second removes two checks that are redundant (they verify the
sizes of two memory regions are in range, which will have been done
earlier for all regions).
The third adds some new checks to routing and filter tables that
can be done at "init time" (without requiring any access to IPA
hardware).
The fourth moves a check that routing and filter table addresses can
be encoded within certain IPA immediate commands, so it's performed
earlier; the checks can be done without touching IPA hardware. The
fifth moves some other command-related checks earlier, for the same
reason.
The sixth removes the definition ipa_table_valid(), because what it
does has become redundant. Finally, the last patch moves two more
validation calls so they're done very early in the probe process.
This will be required by some upcoming patches, which will record
the size of the routing and filter tables at this time so they're
available for subsequent initialization.
====================
Link: https://lore.kernel.org/r/20221021191340.4187935-1-elder@linaro.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Verify that the sizes of the routing and filter table memory regions
are valid as part of memory initialization, rather than waiting for
table initialization. The main reason to do this is that upcoming
patches use these memory region sizes to determine the number of
entries in these tables, and we'll want to know these sizes are good
sooner.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
What ipa_table_valid() (and ipa_table_valid_one(), which it calls)
does is ensure that the memory regions that hold routing and filter
tables have reasonable size. Specifically, it checks that the size
of a region is sufficient (or rather, exactly the right size) to
hold the maximum number of entries supported by the driver. (There
is an additional check that's erroneous, but in practice it is never
reached.)
Recently ipa_table_mem_valid() was added, which is called by
ipa_table_init(). That function verifies that all table memory
regions are of sufficient size, and requires hashed tables to have
zero size if hashing is not supported. It only ensures the filter
table is large enough to hold the number of endpoints that support
filtering, but that is adequate.
Therefore everything that ipa_table_valid() does is redundant, so
get rid of it.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Currently, ipa_cmd_data_valid() is called by ipa_mem_config().
Nothing it does requires access to hardware though, so it can be
done during the init phase of IPA driver startup.
Create a new function ipa_cmd_init(), whose purpose is to do early
initialization related to IPA immediate commands. It will call the
build-time validation function, then will make the two calls made
previously by ipa_cmd_data_valid(). This make ipa_cmd_data_valid()
unnecessary, so get rid of it.
Rename ipa_cmd_header_valid() to be ipa_cmd_header_init_local_valid(),
so its name is clearer about which IPA immediate command it is
associated with.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
