As a step towards making NFSD's maximum rsize and wsize variable at
run-time, make sk_pages a flexible array.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: NeilBrown <neil@brown.name>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Clean up: This array is no longer used.
On a system with 8-byte pointers and 4KB pages, pahole reports that
the rq_vec[] array accounts for 4144 bytes.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
If we can get rid of all uses of rq_vec, then it can be removed.
Replace one use of rqstp::rq_vec with rqstp::rq_bvec.
The feeling of layering violation grows stronger now that
<linux/sunrpc/xdr.h> is included in fs/nfsd/vfs.c.
Suggested-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
All three call sites do the same thing.
I'm struggling with this a bit, however. struct xdr_buf is an XDR
layer object and unmarshaling a WRITE payload is clearly a task
intended to be done by the proc and xdr functions, not by VFS. This
feels vaguely like a layering violation.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
If we can get rid of all uses of rq_vec, then it can be removed.
Replace one use of rqstp::rq_vec with rqstp::rq_bvec.
Suggested-by: Christoph Hellwig <hch@infradead.org>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
As a step towards making NFSD's maximum rsize and wsize variable at
run-time, replace the fixed-size rq_bvec[] array in struct svc_rqst
with a chunk of dynamically-allocated memory.
The rq_bvec[] array contains enough bio_vecs to handle each page in
a maximum size RPC message.
On a system with 8-byte pointers and 4KB pages, pahole reports that
the rq_bvec[] array is 4144 bytes. This patch replaces that array
with a single 8-byte pointer field.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: NeilBrown <neil@brown.name>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
As a step towards making NFSD's maximum rsize and wsize variable at
run-time, replace the fixed-size rq_vec[] array in struct svc_rqst
with a chunk of dynamically-allocated memory.
On a system with 8-byte pointers and 4KB pages, pahole reports that
the rq_pages[] array is 2080 bytes. This patch replaces that with
a single 8-byte pointer field.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: NeilBrown <neil@brown.name>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The server's backchannel uses struct svc_rqst, but does not use the
pages in svc_rqst::rq_pages. It's rq_arg::pages and rq_res::pages
comes from the RPC client's page allocator. Currently,
svc_init_buffer() skips allocating pages in rq_pages for that
reason.
Except that, svc_rqst::rq_pages is filled anyway when a backchannel
svc_rqst is passed to svc_recv() -> and then to svc_alloc_arg().
This isn't really a problem at the moment, except that these pages
are allocated but then never used, as far as I can tell.
The problem is that later in this series, in addition to populating
the entries of rq_pages[], svc_init_buffer() will also allocate the
memory underlying the rq_pages[] array itself. If that allocation is
skipped, then svc_alloc_args() chases a NULL pointer for ingress
backchannel requests.
This approach avoids introducing extra conditional logic in
svc_alloc_args(), which is a hot path.
Acked-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: NeilBrown <neil@brown.name>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This page count is to be used to allocate various arrays of pages
and bio_vecs, replacing the fixed RPCSVC_MAXPAGES value.
The documenting comment is somewhat stale -- of course NFSv4
COMPOUND procedures may have multiple payloads.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: NeilBrown <neil@brown.name>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
There is an upper bound on the number of rdma_rw contexts that can
be created per QP.
This invisible upper bound is because rdma_create_qp() adds one or
more additional SQEs for each ctxt that the ULP requests via
qp_attr.cap.max_rdma_ctxs. The QP's actual Send Queue length is on
the order of the sum of qp_attr.cap.max_send_wr and a factor times
qp_attr.cap.max_rdma_ctxs. The factor can be up to three, depending
on whether MR operations are required before RDMA Reads.
This limit is not visible to RDMA consumers via dev->attrs. When the
limit is surpassed, QP creation fails with -ENOMEM. For example:
svcrdma's estimate of the number of rdma_rw contexts it needs is
three times the number of pages in RPCSVC_MAXPAGES. When MAXPAGES
is about 260, the internally-computed SQ length should be:
64 credits + 10 backlog + 3 * (3 * 260) = 2414
Which is well below the advertised qp_max_wr of 32768.
If RPCSVC_MAXPAGES is increased to 4MB, that's 1040 pages:
64 credits + 10 backlog + 3 * (3 * 1040) = 9434
However, QP creation fails. Dynamic printk for mlx5 shows:
calc_sq_size:618:(pid 1514): send queue size (9326 * 256 / 64 -> 65536) exceeds limits(32768)
Although 9326 is still far below qp_max_wr, QP creation still
fails.
Because the total SQ length calculation is opaque to RDMA consumers,
there doesn't seem to be much that can be done about this except for
consumers to try to keep the requested rdma_rw ctxt count low.
Fixes: 2da0f610e7 ("svcrdma: Increase the per-transport rw_ctx count")
Reviewed-by: NeilBrown <neil@brown.name>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Add bootph-pre-ram hinting to jh7110-common.dtsi:
- i2c5_pins and i2c-pins subnode for connection to eeprom
- eeprom node
- qspi flash configuration subnode
- memory node
- mmc0 for eMMC
- mmc1 for SD Card
- uart0 for serial console
With this the U-Boot SPL secondary program loader may drop such overrides.
Signed-off-by: E Shattow <e@freeshell.de>
Acked-by: Emil Renner Berthing <emil.renner.berthing@canonical.com>
Signed-off-by: Conor Dooley <conor.dooley@microchip.com>
StarFive VisionFive2 and similar JH7110 boards have an eeprom compatible
with Atmel 24c04. Add the node so this may be used with the at24 driver.
Signed-off-by: E Shattow <e@freeshell.de>
Reviewed-by: Hal Feng <hal.feng@starfivetech.com>
Reviewed-by: Emil Renner Berthing <emil.renner.berthing@canonical.com>
Signed-off-by: Conor Dooley <conor.dooley@microchip.com>
Use qspi flash read-delay and spi-max-frequency settings compatible with
U-Boot bootloader.
Observations from testing on Pine64 Star64 hardware within U-Boot bootloader
and read-delay=2 are spi-max-frequency less than 49.8MHz fails to write,
corrupt data writes at 25MHz to 49.799999MHz, and valid data writes at
49.8MHz to 100MHz (not tested above 100MHz). No valid spi-max-frequency
was found for 1<read-delay<=3 and corrupt data with read-delay=3.
Looking around the Linux codebase it is common to see read-delay 2 cycles
with spi-max-frequency 100MHz and testing confirms this to work in both
U-Boot and Linux.
Signed-off-by: E Shattow <e@freeshell.de>
Reviewed-by: Hal Feng <hal.feng@starfivetech.com>
Acked-by: Emil Renner Berthing <emil.renner.berthing@canonical.com>
Signed-off-by: Conor Dooley <conor.dooley@microchip.com>
Add syscrg clock assignments for CPU, BUS, PERH, and QSPI as required by
boot loader before kernel.
Signed-off-by: E Shattow <e@freeshell.de>
Reviewed-by: Emil Renner Berthing <emil.renner.berthing@canonical.com>
Signed-off-by: Conor Dooley <conor.dooley@microchip.com>
The pin names of MMC0 pinmux is defined in the pinctrl dt binding header
associated with starfive,jh7110-pinctrl .
Include the header file and use these names instead of raw numbers for
defining MMC0 pinmux.
Signed-off-by: Icenowy Zheng <uwu@icenowy.me>
Reviewed-by: Emil Renner Berthing <emil.renner.berthing@canonical.com>
Signed-off-by: Conor Dooley <conor.dooley@microchip.com>
Add usb_cdns3 and usb0_pins configuration to support super speed USB
device on the FML13V01 board.
Signed-off-by: Sandie Cao <sandie.cao@deepcomputing.io>
Tested-by: Maud Spierings <maud_spierings@hotmail.com>
Reviewed-by: Emil Renner Berthing <emil.renner.berthing@canonical.com>
Signed-off-by: Conor Dooley <conor.dooley@microchip.com>
Most H2G messages are FAST_REQ which means no synchronous response is
expected. The messages are sent as fire-and-forget with no tracking.
However, errors can still be returned when something goes unexpectedly
wrong. That leads to confusion due to not being able to match up the
error response to the originating H2G.
So add support for tracking the FAST_REQ H2Gs and matching up an error
response to its originator. This is only enabled in XE_DEBUG builds
given that such errors should never happen in a working system and
there is an overhead for the tracking.
Further, if XE_DEBUG_GUC is enabled then even more memory and time is
used to record the call stack of each H2G and report that with the
error. That makes it much easier to work out where a specific H2G came
from if there are multiple code paths that can send it.
v2: Some re-wording of comments and prints, more consistent use of #if
vs stub functions - review feedback from Daniele & Michal).
v3: Split config change to separate patch, improve a debug print
(review feedback from Michal).
v4: Bunch of minor tweaks (review feedback from Michal).
Original-i915-code: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://lore.kernel.org/r/20250512215324.1457009-5-John.C.Harrison@Intel.com
Rename XE_LARGE_GUC_BUFFER to XE_DEBUG_GUC to allow for more debug
only code (in subsequent patch) without adding more config defines
that each control only a single thing.
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://lore.kernel.org/r/20250512215324.1457009-4-John.C.Harrison@Intel.com
These error codes are not actually used in the driver but it is
extremely useful to have them available to understand error messages.
v2: Add a bunch more error codes and drop 'status' from names (review
feedback by Michal W).
v3: Drop 'SUCCESS' response as meaningless in current API (review
feedback by Michal W).
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://lore.kernel.org/r/20250512215324.1457009-3-John.C.Harrison@Intel.com
An earlier patch moved a drm_print a few lines lower but accidentally
left a double blank line behind. So fix that.
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://lore.kernel.org/r/20250512215324.1457009-2-John.C.Harrison@Intel.com
Four small fixes for crashes:
- Double free in rxe
- UAF in irdma from early freeing the rf
- Off by one undoing the IRQ allocations during error unwind in irdma
- Another race with device rename and uevent generation. uevents access
the struct device name and UAF when it is changed.
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQRRRCHOFoQz/8F5bUaFwuHvBreFYQUCaCY7/AAKCRCFwuHvBreF
YcdvAQCr6MYgSUEczhCCtG7nS3lWHcQV/CgBeph18qNJQnChQgEApKwpTuueODEy
ml/6Ho0Ge2NmBBBSPxEfbCDsBFGaNgM=
=Nr/y
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Pull rdma fixes from Jason Gunthorpe:
"Four small fixes for crashes:
- Double free in rxe
- UAF in irdma from early freeing the rf
- Off by one undoing the IRQ allocations during error unwind in irdma
- Another race with device rename and uevent generation. uevents
accesses the struct device name and UAF when it is changed"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
RDMA/core: Fix "KASAN: slab-use-after-free Read in ib_register_device" problem
ice, irdma: fix an off by one in error handling code
irdma: free iwdev->rf after removing MSI-X
RDMA/rxe: Fix slab-use-after-free Read in rxe_queue_cleanup bug
Clearing the table_sz on cleanup seemed reasonable, but further
discussions concluded that this merely working around the issue
and that the fix is incomplete.
As such, revert commit efdde3d73a ("remoteproc: core: Clear table_sz
when rproc_shutdown") to avoid carrying a partial fix.
Signed-off-by: Bjorn Andersson <andersson@kernel.org>
Link: https://lore.kernel.org/r/20250513-revert-rproc-table-sz-v1-1-a8c6b5d6f8a7@kernel.org
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Because ARM's MPAM controls are probed using MMIO, resctrl can't be
initialised until enough CPUs are online to have determined the system-wide
supported num_closid. Arm64 also supports 'late onlined secondaries', where
only a subset of CPUs are online during boot.
These two combine to mean the MPAM driver may not be able to initialise
resctrl until user-space has brought 'enough' CPUs online.
To allow MPAM to initialise resctrl after __init text has been free'd, remove
all the __init markings from resctrl.
The existing __exit markings cause these functions to be removed by the linker
as it has never been possible to build resctrl as a module. MPAM has an error
interrupt which causes the driver to reset and disable itself. Remove the
__exit markings to allow the MPAM driver to tear down resctrl when an error
occurs.
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Shaopeng Tan <tan.shaopeng@jp.fujitsu.com>
Reviewed-by: Tony Luck <tony.luck@intel.com>
Reviewed-by: Reinette Chatre <reinette.chatre@intel.com>
Reviewed-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: Carl Worth <carl@os.amperecomputing.com> # arm64
Tested-by: Shaopeng Tan <tan.shaopeng@jp.fujitsu.com>
Tested-by: Peter Newman <peternewman@google.com>
Tested-by: Amit Singh Tomar <amitsinght@marvell.com> # arm64
Tested-by: Shanker Donthineni <sdonthineni@nvidia.com> # arm64
Tested-by: Babu Moger <babu.moger@amd.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Link: https://lore.kernel.org/20250515165855.31452-10-james.morse@arm.com
resctrl_exit() was intended for use when the 'resctrl' module was unloaded.
resctrl can't be built as a module, and the kernfs helpers are not exported so
this is unlikely to change. MPAM has an error interrupt which indicates the
MPAM driver has gone haywire. Should this occur tasks could run with the wrong
control values, leading to bad performance for important tasks. In this
scenario the MPAM driver will reset the hardware, but it needs a way to tell
resctrl that no further configuration should be attempted.
In particular, moving tasks between control or monitor groups does not
interact with the architecture code, so there is no opportunity for the arch
code to indicate that the hardware is no-longer functioning.
Using resctrl_exit() for this leaves the system in a funny state as resctrl is
still mounted, but cannot be un-mounted because the sysfs directory that is
typically used has been removed. Dave Martin suggests this may cause systemd
trouble in the future as not all filesystems can be unmounted.
Add calls to remove all the files and directories in resctrl, and remove the
sysfs_remove_mount_point() call that leaves the system in a funny state. When
triggered, this causes all the resctrl files to disappear. resctrl can be
unmounted, but not mounted again.
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Shaopeng Tan <tan.shaopeng@jp.fujitsu.com>
Reviewed-by: Tony Luck <tony.luck@intel.com>
Reviewed-by: Reinette Chatre <reinette.chatre@intel.com>
Reviewed-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: Carl Worth <carl@os.amperecomputing.com> # arm64
Tested-by: Shaopeng Tan <tan.shaopeng@jp.fujitsu.com>
Tested-by: Peter Newman <peternewman@google.com>
Tested-by: Amit Singh Tomar <amitsinght@marvell.com> # arm64
Tested-by: Shanker Donthineni <sdonthineni@nvidia.com> # arm64
Tested-by: Babu Moger <babu.moger@amd.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Link: https://lore.kernel.org/20250515165855.31452-9-james.morse@arm.com
resctrl_exit() removes things like the resctrl mount point directory
and unregisters the filesystem prior to freeing data structures that
were allocated during resctrl_init().
This assumes that there are no online domains when resctrl_exit() is
called. If any domain were online, the limbo or overflow handler could
be scheduled to run.
Add a check for any online control or monitor domains, and document that
the architecture code is required to offline all monitor and control
domains before calling resctrl_exit().
Suggested-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Reinette Chatre <reinette.chatre@intel.com>
Reviewed-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: Babu Moger <babu.moger@amd.com>
Tested-by: Shaopeng Tan <tan.shaopeng@jp.fujitsu.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Link: https://lore.kernel.org/20250515165855.31452-8-james.morse@arm.com
resctrl_sched_in() loads the architecture specific CPU MSRs with the
CLOSID and RMID values. This function was named before resctrl was
split to have architecture specific code, and generic filesystem code.
This function is obviously architecture specific, but does not begin
with 'resctrl_arch_', making it the odd one out in the functions an
architecture needs to support to enable resctrl.
Rename it for consistency. This is purely cosmetic.
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Shaopeng Tan <tan.shaopeng@jp.fujitsu.com>
Reviewed-by: Tony Luck <tony.luck@intel.com>
Reviewed-by: Reinette Chatre <reinette.chatre@intel.com>
Reviewed-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: Carl Worth <carl@os.amperecomputing.com> # arm64
Tested-by: Shaopeng Tan <tan.shaopeng@jp.fujitsu.com>
Tested-by: Peter Newman <peternewman@google.com>
Tested-by: Amit Singh Tomar <amitsinght@marvell.com> # arm64
Tested-by: Shanker Donthineni <sdonthineni@nvidia.com> # arm64
Tested-by: Babu Moger <babu.moger@amd.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Link: https://lore.kernel.org/20250515165855.31452-7-james.morse@arm.com
CONFIG_AUXILIARY_BUS cannot be enabled explicitly, and unless we select
it we have no way to include it (and thus to enable NOVA_DRM) unless
another driver happens to do it for us.
Fixes: cdeaeb9dd7 ("drm: nova-drm: add initial driver skeleton")
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Link: https://lore.kernel.org/r/20250515-aux_bus-v2-3-47c70f96ae9b@nvidia.com
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
CONFIG_AUXILIARY_BUS cannot be enabled explicitly, and unless we select
it we have no way to include it (and thus to enable NOVA_CORE) unless
another driver happens to do it for us.
Fixes: e041d81a03 ("gpu: nova-core: register auxiliary device for nova-drm")
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Link: https://lore.kernel.org/r/20250515-aux_bus-v2-2-47c70f96ae9b@nvidia.com
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
CONFIG_AUXILIARY_BUS cannot be enabled explicitly, and unless we select
it we have no way to include it (and thus to enable the auxiliary driver
sample) unless a driver happens to do it for us.
Fixes: 96609a1969 ("samples: rust: add Rust auxiliary driver sample")
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Link: https://lore.kernel.org/r/20250515-aux_bus-v2-1-47c70f96ae9b@nvidia.com
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
Resctrl allocates and finds free CLOSID values using the bits of a u32.
This restricts the number of control groups that can be created by
user-space.
MPAM has an architectural limit of 2^16 CLOSID values, Intel x86 could
be extended beyond 32 values. There is at least one MPAM platform which
supports more than 32 CLOSID values.
Replace the fixed size bitmap with calls to the bitmap API to allocate
an array of a sufficient size.
ffs() returns '1' for bit 0, hence the existing code subtracts 1 from
the index to get the CLOSID value. find_first_bit() returns the bit
number which does not need adjusting.
[ morse: fixed the off-by-one in the allocator and the wrong not-found
value. Removed the limit. Rephrase the commit message. ]
Signed-off-by: Amit Singh Tomar <amitsinght@marvell.com>
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Fenghua Yu <fenghuay@nvidia.com>
Reviewed-by: Reinette Chatre <reinette.chatre@intel.com>
Tested-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: Peter Newman <peternewman@google.com>
Tested-by: Shaopeng Tan <tan.shaopeng@jp.fujitsu.com>
Tested-by: Amit Singh Tomar <amitsinght@marvell.com> # arm64
Tested-by: Shanker Donthineni <sdonthineni@nvidia.com> # arm64
Tested-by: Babu Moger <babu.moger@amd.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Link: https://lore.kernel.org/20250515165855.31452-6-james.morse@arm.com
With the lack of cpumask_any_andnot_but(), cpumask_any_housekeeping()
has to abuse cpumask_nth() functions.
Update cpumask_any_housekeeping() to use the new cpumask_any_but()
and cpumask_any_andnot_but(). These two functions understand
RESCTRL_PICK_ANY_CPU, which simplifies cpumask_any_housekeeping()
significantly.
Signed-off-by: Yury Norov [NVIDIA] <yury.norov@gmail.com>
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: James Morse <james.morse@arm.com>
Reviewed-by: Reinette Chatre <reinette.chatre@intel.com>
Reviewed-by: Shaopeng Tan <tan.shaopeng@jp.fujitsu.com>
Reviewed-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: James Morse <james.morse@arm.com>
Tested-by: Shaopeng Tan <tan.shaopeng@jp.fujitsu.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Link: https://lore.kernel.org/20250515165855.31452-5-james.morse@arm.com
The TCA_FLOWER_KEY_CFM enum has a UNSPEC and MAX with _OPT
in the name, but the real attributes don't. Add a MAX that
more reasonably matches the attrs.
The PAD in TCA_TAPRIO is the only attr which doesn't have
_ATTR in it, perhaps signifying that it's not a real attr?
If so interesting idea in abstract but it makes codegen painful.
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250513221752.843102-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Eric Dumazet says:
====================
tcp: receive side improvements
We have set tcp_rmem[2] to 15 MB for about 8 years at Google,
but had some issues for high speed flows on very small RTT.
TCP rx autotuning has a tendency to overestimate the RTT,
thus tp->rcvq_space.space and sk->sk_rcvbuf.
This makes TCP receive queues much bigger than necessary,
to a point cpu caches are evicted before application can
copy the data, on cpus using DDIO.
This series aims to fix this.
- First patch adds tcp_rcvbuf_grow() tracepoint, which was very
convenient to study the various issues fixed in this series.
- Seven patches fix receiver autotune issues.
- Two patches fix sender side issues.
- Final patch increases tcp_rmem[2] so that TCP speed over WAN
can meet modern needs.
Tested on a 200Gbit NIC, average max throughput of a single flow:
Before:
73593 Mbit.
After:
122514 Mbit.
====================
Link: https://patch.msgid.link/20250513193919.1089692-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Last change to tcp_rmem[2] happened in 2012, in commit b49960a05e
("tcp: change tcp_adv_win_scale and tcp_rmem[2]")
TCP performance on WAN is mostly limited by tcp_rmem[2] for receivers.
After this series improvements, it is time to increase the default.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20250513193919.1089692-12-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This partially reverts commit c73e5807e4 ("tcp: tsq: no longer use
limit_output_bytes for paced flows")
Overriding the tcp_limit_output_bytes sysctl value
for FQ enabled flows has the following problem:
It allows TCP to queue around 2 ms worth of data per flow,
defeating tcp_rcv_rtt_update() accuracy on the receiver,
forcing it to increase sk->sk_rcvbuf even if the real
RTT is around 100 us.
After this change, we keep enough packets in flight to fill
the pipe, and let receive queues small enough to get
good cache behavior (cpu caches and/or NIC driver page pools).
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20250513193919.1089692-11-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Last change happened in 2018 with commit c73e5807e4
("tcp: tsq: no longer use limit_output_bytes for paced flows")
Modern NIC speeds got a 4x increase since then.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20250513193919.1089692-10-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_rcv_rtt_update() role is to keep an estimation
of RTT (tp->rcv_rtt_est.rtt_us) for receivers.
If an application is too slow to drain the TCP receive
queue, it is better to leave the RTT estimation small,
so that tcp_rcv_space_adjust() does not inflate
tp->rcvq_space.space and sk->sk_rcvbuf.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20250513193919.1089692-9-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_rcv_rtt_update() goal is to maintain an estimation of the RTT
in tp->rcv_rtt_est.rtt_us, used by tcp_rcv_space_adjust()
When TCP TS are enabled, tcp_rcv_rtt_update() is using
EWMA to smooth the samples.
Change this to immediately latch the incoming value if it
is lower than tp->rcv_rtt_est.rtt_us, so that tcp_rcv_space_adjust()
does not overshoot tp->rcvq_space.space and sk->sk_rcvbuf.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20250513193919.1089692-8-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_rcv_state_process() must tweak tp->advmss for TS enabled flows
before the call to tcp_init_transfer() / tcp_init_buffer_space().
Otherwise tp->rcvq_space.space is off by 120 bytes
(TCP_INIT_CWND * TCPOLEN_TSTAMP_ALIGNED).
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Wei Wang <weiwan@google.com>
Link: https://patch.msgid.link/20250513193919.1089692-7-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If the application can not drain fast enough a TCP socket queue,
tcp_rcv_space_adjust() can overestimate tp->rcvq_space.space.
Then sk->sk_rcvbuf can grow and hit tcp_rmem[2] for no good reason.
Fix this by taking into acount the number of available bytes.
Keeping sk->sk_rcvbuf at the right size allows better cache efficiency.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Wei Wang <weiwan@google.com>
Link: https://patch.msgid.link/20250513193919.1089692-5-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch takes care of the needed provisioning
when incoming packets are stored in the out of order queue.
This part was not implemented in the correct way, we need
to decouple it from tcp_rcv_space_adjust() logic.
Without it, stalls in the pipe could happen.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20250513193919.1089692-4-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Current autosizing in tcp_rcv_space_adjust() is too aggressive.
Instead of betting on possible losses and over estimate BDP,
it is better to only account for slow start.
The following patch is then adding a more precise tuning
in the events of packet losses.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20250513193919.1089692-3-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
With the lack of the functions, client code has to abuse less efficient
cpumask_nth().
Signed-off-by: Yury Norov [NVIDIA] <yury.norov@gmail.com>
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: James Morse <james.morse@arm.com>
Reviewed-by: Reinette Chatre <reinette.chatre@intel.com>
Reviewed-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: Fenghua Yu <fenghuay@nvidia.com>
Tested-by: James Morse <james.morse@arm.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Link: https://lore.kernel.org/20250515165855.31452-4-james.morse@arm.com
