MEI GSC interrupt comes from i915. It has top half and bottom half.
Top half is called from i915 interrupt handler. It should be in
irq disabled context.
With RT kernel, by default i915 IRQ handler is in threaded IRQ. MEI GSC
top half might be in threaded IRQ context. generic_handle_irq_safe API
could be called from either IRQ or process context, it disables local
IRQ then calls MEI GSC interrupt top half.
This change fixes A380/A770 GPU boot hang issue with RT kernel.
Fixes: 1e3dc1d862 ("drm/i915/gsc: add gsc as a mei auxiliary device")
Tested-by: Furong Zhou <furong.zhou@intel.com>
Suggested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Junxiao Chang <junxiao.chang@intel.com>
Link: https://lore.kernel.org/r/20250425151108.643649-1-junxiao.chang@intel.com
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(cherry picked from commit dccf655f69)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
The following error has been reported sporadically by CI when a test
unbinds the i915 driver on a ring submission platform:
<4> [239.330153] ------------[ cut here ]------------
<4> [239.330166] i915 0000:00:02.0: [drm] drm_WARN_ON(dev_priv->mm.shrink_count)
<4> [239.330196] WARNING: CPU: 1 PID: 18570 at drivers/gpu/drm/i915/i915_gem.c:1309 i915_gem_cleanup_early+0x13e/0x150 [i915]
...
<4> [239.330640] RIP: 0010:i915_gem_cleanup_early+0x13e/0x150 [i915]
...
<4> [239.330942] Call Trace:
<4> [239.330944] <TASK>
<4> [239.330949] i915_driver_late_release+0x2b/0xa0 [i915]
<4> [239.331202] i915_driver_release+0x86/0xa0 [i915]
<4> [239.331482] devm_drm_dev_init_release+0x61/0x90
<4> [239.331494] devm_action_release+0x15/0x30
<4> [239.331504] release_nodes+0x3d/0x120
<4> [239.331517] devres_release_all+0x96/0xd0
<4> [239.331533] device_unbind_cleanup+0x12/0x80
<4> [239.331543] device_release_driver_internal+0x23a/0x280
<4> [239.331550] ? bus_find_device+0xa5/0xe0
<4> [239.331563] device_driver_detach+0x14/0x20
...
<4> [357.719679] ---[ end trace 0000000000000000 ]---
If the test also unloads the i915 module then that's followed with:
<3> [357.787478] =============================================================================
<3> [357.788006] BUG i915_vma (Tainted: G U W N ): Objects remaining on __kmem_cache_shutdown()
<3> [357.788031] -----------------------------------------------------------------------------
<3> [357.788204] Object 0xffff888109e7f480 @offset=29824
<3> [357.788670] Allocated in i915_vma_instance+0xee/0xc10 [i915] age=292729 cpu=4 pid=2244
<4> [357.788994] i915_vma_instance+0xee/0xc10 [i915]
<4> [357.789290] init_status_page+0x7b/0x420 [i915]
<4> [357.789532] intel_engines_init+0x1d8/0x980 [i915]
<4> [357.789772] intel_gt_init+0x175/0x450 [i915]
<4> [357.790014] i915_gem_init+0x113/0x340 [i915]
<4> [357.790281] i915_driver_probe+0x847/0xed0 [i915]
<4> [357.790504] i915_pci_probe+0xe6/0x220 [i915]
...
Closer analysis of CI results history has revealed a dependency of the
error on a few IGT tests, namely:
- igt@api_intel_allocator@fork-simple-stress-signal,
- igt@api_intel_allocator@two-level-inception-interruptible,
- igt@gem_linear_blits@interruptible,
- igt@prime_mmap_coherency@ioctl-errors,
which invisibly trigger the issue, then exhibited with first driver unbind
attempt.
All of the above tests perform actions which are actively interrupted with
signals. Further debugging has allowed to narrow that scope down to
DRM_IOCTL_I915_GEM_EXECBUFFER2, and ring_context_alloc(), specific to ring
submission, in particular.
If successful then that function, or its execlists or GuC submission
equivalent, is supposed to be called only once per GEM context engine,
followed by raise of a flag that prevents the function from being called
again. The function is expected to unwind its internal errors itself, so
it may be safely called once more after it returns an error.
In case of ring submission, the function first gets a reference to the
engine's legacy timeline and then allocates a VMA. If the VMA allocation
fails, e.g. when i915_vma_instance() called from inside is interrupted
with a signal, then ring_context_alloc() fails, leaving the timeline held
referenced. On next I915_GEM_EXECBUFFER2 IOCTL, another reference to the
timeline is got, and only that last one is put on successful completion.
As a consequence, the legacy timeline, with its underlying engine status
page's VMA object, is still held and not released on driver unbind.
Get the legacy timeline only after successful allocation of the context
engine's VMA.
v2: Add a note on other submission methods (Krzysztof Karas):
Both execlists and GuC submission use lrc_alloc() which seems free
from a similar issue.
Fixes: 75d0a7f31e ("drm/i915: Lift timeline into intel_context")
Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/12061
Cc: Chris Wilson <chris.p.wilson@linux.intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Krzysztof Karas <krzysztof.karas@intel.com>
Reviewed-by: Sebastian Brzezinka <sebastian.brzezinka@intel.com>
Reviewed-by: Krzysztof Niemiec <krzysztof.niemiec@intel.com>
Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com>
Reviewed-by: Nitin Gote <nitin.r.gote@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://lore.kernel.org/r/20250611104352.1014011-2-janusz.krzysztofik@linux.intel.com
(cherry picked from commit cc43422b3c)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Commit 81256a50aa ("x86/mm: Make memremap(MEMREMAP_WB) map memory as
encrypted by default") changed the default behavior of
memremap(MEMREMAP_WB) and started mapping memory as encrypted.
The driver requires the fifo memory to be decrypted to communicate with
the host but was relaying on the old default behavior of
memremap(MEMREMAP_WB) and thus broke.
Fix it by explicitly specifying the desired behavior and passing
MEMREMAP_DEC to memremap.
Fixes: 81256a50aa ("x86/mm: Make memremap(MEMREMAP_WB) map memory as encrypted by default")
Signed-off-by: Marko Kiiskila <marko.kiiskila@broadcom.com>
Signed-off-by: Zack Rusin <zack.rusin@broadcom.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: linux-mm@kvack.org
Cc: linux-kernel@vger.kernel.org
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Link: https://lore.kernel.org/r/20250618192926.1092450-1-zack.rusin@broadcom.com
struct ublk_io is already 56 bytes on 64-bit architectures, so round it
up to a full cache line (typically 64 bytes). This ensures a single
ublk_io doesn't span multiple cache lines and prevents false sharing if
consecutive ublk_io's are accessed by different daemon tasks.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-15-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
ublk_get_req_ref() and ublk_put_req_ref() currently call
ublk_need_req_ref(ubq) to check whether the ublk device features require
reference counting of its requests. However, all callers already know
that reference counting is required:
- __ublk_check_and_get_req() is only called from
ublk_check_and_get_req() if user copy is enabled, and from
ublk_register_io_buf() if zero copy is enabled
- ublk_io_release() is only called for requests registered by
ublk_register_io_buf(), which requires zero copy
- ublk_ch_read_iter() and ublk_ch_write_iter() only call
ublk_put_req_ref() if ublk_check_and_get_req() succeeded, which
requires user copy to be enabled
So drop the ublk_need_req_ref() check and the ubq argument in
ublk_get_req_ref() and ublk_put_req_ref().
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-14-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
ublk_io_release() performs an expensive atomic refcount decrement. This
atomic operation is unnecessary in the common case where the request's
buffer is registered and unregistered on the daemon task before handling
UBLK_IO_COMMIT_AND_FETCH_REQ for the I/O. So if ublk_io_release() is
called on the daemon task and task_registered_buffers is positive, just
decrement task_registered_buffers (nonatomically). ublk_sub_req_ref()
will apply this decrement when it atomically subtracts from io->ref.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-13-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
ublk_register_io_buf() performs an expensive atomic refcount increment,
as well as a lot of pointer chasing to look up the struct request.
Create a separate ublk_daemon_register_io_buf() for the daemon task to
call. Initialize ublk_io's reference count to a large number, introduce
a field task_registered_buffers to count the buffers registered on the
daemon task, and atomically subtract the large number minus
task_registered_buffers in ublk_commit_and_fetch().
Also obtain the struct request directly from ublk_io's req field instead
of looking it up on the tagset.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-12-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Make the unlikely case blk_should_fake_timeout() return early to reduce
the indentation of the successful path.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-11-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently, UBLK_IO_REGISTER_IO_BUF and UBLK_IO_UNREGISTER_IO_BUF are
only permitted on the ublk_io's daemon task. But this restriction is
unnecessary. ublk_register_io_buf() calls __ublk_check_and_get_req() to
look up the request from the tagset and atomically take a reference on
the request without accessing the ublk_io. ublk_unregister_io_buf()
doesn't use the q_id or tag at all.
So allow these opcodes even on tasks other than io->task.
Handle UBLK_IO_UNREGISTER_IO_BUF before obtaining the ubq and io since
the buffer index being unregistered is not necessarily related to the
specified q_id and tag.
Add a feature flag UBLK_F_BUF_REG_OFF_DAEMON that userspace can use to
determine whether the kernel supports off-daemon buffer registration.
Suggested-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-10-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
UBLK_IO_UNREGISTER_IO_BUF currently requires a valid q_id and tag to be
passed in the ublksrv_io_cmd. However, only the addr (registered buffer
index) is actually used to unregister the buffer. There is no check that
the q_id and tag are for the ublk request whose buffer is registered at
the given index. To prepare to allow userspace to omit the q_id and tag,
check the UBLK_F_SUPPORT_ZERO_COPY flag on the ublk_device instead of
the ublk_queue.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-9-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
- MGMT: set_mesh: update LE scan interval and window
- MGMT: mesh_send: check instances prior disabling advertising
- hci_sync: revert some mesh modifications
- hci_sync: Set extended advertising data synchronously
- hci_sync: Prevent unintended pause by checking if advertising is active
-----BEGIN PGP SIGNATURE-----
iQJNBAABCgA3FiEE7E6oRXp8w05ovYr/9JCA4xAyCykFAmhe33EZHGx1aXoudm9u
LmRlbnR6QGludGVsLmNvbQAKCRD0kIDjEDILKWsGD/9BmzTKlTpTzwRPa8xDUuNG
Rl4qEBJVcwdah6wFMahl+SK5LaFIyU9Jyjdhnebor2uHqff9NDdFEOgJ6yu2h42V
+Delf9lzpqFetEd7E8EgA0ynipUkfvvgLJ5DhulMptK/icESfh3vY45jsuW93hks
iIqsdLZr9840+CwGQ6BWSg8XkvweU/LqtKF7xx7CDx9ZBONClf92mZD9S8fPDbFL
s9GBnLyO+0nuFU7cZwmVlU3w/m9Sb5AALexbaPYQhDR8yk10Wui9viVD2LZyyc8c
+QtQHgYLETfGDAlH8NcabZVEuIpSmeMNQIn3VBPj9hm8FYc1fqJ4kXcLtw4xvGzJ
Xj1Gao8zt0xVEOQCyHiW1J87+AbxHf8XzIMGMEw2rk9qbgNXEIs44yO1pDvhpNEa
Fs2bZ4v81BnueOWI4ILH/y548IPrI8mZX7bCro66OW53mlZXHprlj20z2/jrF5WI
UCsBq4a0h1T1yfYLu54GykORvlXHqRB16U3av9K0T6sHNIqees1Q6NjZLj5ztX/y
MwMph//lSar1BbfLZapjo2itNce8c8fkBDTW25puttyS2tX0kFoWwN9VRLQg5U/P
j4meJ9FOzNAgXDTLKrXj0lLDmcP9Z4uKO0xsOvee7PfJbsI1iD+NCeBrpHoq5Hot
w1oNI/L/VufL2dhhY0L94Q==
=QBPI
-----END PGP SIGNATURE-----
Merge tag 'for-net-2025-06-27' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth
Luiz Augusto von Dentz says:
====================
bluetooth pull request for net:
- MGMT: set_mesh: update LE scan interval and window
- MGMT: mesh_send: check instances prior disabling advertising
- hci_sync: revert some mesh modifications
- hci_sync: Set extended advertising data synchronously
- hci_sync: Prevent unintended pause by checking if advertising is active
* tag 'for-net-2025-06-27' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth:
Bluetooth: HCI: Set extended advertising data synchronously
Bluetooth: MGMT: mesh_send: check instances prior disabling advertising
Bluetooth: MGMT: set_mesh: update LE scan interval and window
Bluetooth: hci_sync: revert some mesh modifications
Bluetooth: Prevent unintended pause by checking if advertising is active
====================
Link: https://patch.msgid.link/20250627181601.520435-1-luiz.dentz@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
At the time of commit bc51dddf98 ("netns: avoid disabling irq
for netns id") peernet2id() was not yet using RCU.
Commit 2dce224f46 ("netns: protect netns
ID lookups with RCU") changed peernet2id() to no longer
acquire net->nsid_lock (potentially from BH context).
We do not need to block soft interrupts when acquiring
net->nsid_lock anymore.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Guillaume Nault <gnault@redhat.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20250627163242.230866-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit d53320aeef ("dt-bindings: net: Rename
renesas,r9a09g057-gbeth.yaml") renames the net devicetree binding
renesas,r9a09g057-gbeth.yaml to renesas,rzv2h-gbeth.yaml, but misses to
adjust the file entry in the RENESAS RZ/V2H(P) DWMAC GBETH GLUE LAYER
DRIVER section in MAINTAINERS.
Adjust the file entry after this file renaming.
Signed-off-by: Lukas Bulwahn <lukas.bulwahn@redhat.com>
Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
Link: https://patch.msgid.link/20250627134453.51780-1-lukas.bulwahn@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The counters of port MAC are all 64-bit registers, and the statistics of
ethtool are u64 type, so replace enetc_port_rd() with enetc_port_rd64()
to read 64-bit statistics.
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Reviewed-by: Frank Li <Frank.Li@nxp.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250627021108.3359642-4-wei.fang@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Some counters in enetc_port_counters are 32-bit registers, and some are
64-bit registers. But in the current driver, they are all read through
enetc_port_rd(), which can only read a 32-bit value. Therefore, separate
64-bit counters (enetc_pm_counters) from enetc_port_counters and use
enetc_port_rd64() to read the 64-bit statistics.
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Reviewed-by: Frank Li <Frank.Li@nxp.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250627021108.3359642-3-wei.fang@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The statistics of the ring are all unsigned int type, so the statistics
will overflow quickly under heavy traffic. In addition, the statistics
of struct net_device_stats are obtained from struct enetc_ring_stats,
but the statistics of net_device_stats are unsigned long type. So it is
better to keep the statistics types consistent in these two structures.
Considering these two factors, and the fact that both LS1028A and i.MX95
are arm64 architecture, the statistics of enetc_ring_stats are changed
to unsigned long type. Note that unsigned int and unsigned long are the
same thing on some systems, and on such systems there is no overflow
advantage of one over the other.
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Reviewed-by: Frank Li <Frank.Li@nxp.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250627021108.3359642-2-wei.fang@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In the current implementation, IP coalescing is always enabled and
cannot be disabled.
As setting maximum frames to 0 or 1, or setting delay to zero implies
immediate delivery of single packets/IRQs, disable coalescing in
hardware in these cases.
This also guarantees that coalescing is never enabled with ICFT or ICTT
set to zero, a configuration that could lead to unpredictable behaviour
according to i.MX8MP reference manual.
Signed-off-by: Jonas Rebmann <jre@pengutronix.de>
Reviewed-by: Wei Fang <wei.fang@nxp.com>
Link: https://patch.msgid.link/20250626-fec_deactivate_coalescing-v2-1-0b217f2e80da@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
There is a occasional problem that ping is failed between AML devices.
That is because the manual enablement of the security Tx path on the
hardware is missing, no matter what its previous state was.
Fixes: 6f8b4c01a8 ("net: txgbe: Implement PHYLINK for AML 25G/10G devices")
Signed-off-by: Jiawen Wu <jiawenwu@trustnetic.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/5BDFB14C57D1C42A+20250626085153.86122-1-jiawenwu@trustnetic.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Ido Schimmel says:
====================
Add support for externally validated neighbor entries
Patch #1 adds a new neighbor flag ("extern_valid") that prevents the
kernel from invalidating or removing a neighbor entry, while allowing
the kernel to notify user space when the entry becomes reachable. See
motivation and implementation details in the commit message.
Patch #2 adds a selftest.
v1: https://lore.kernel.org/20250611141551.462569-1-idosch@nvidia.com
====================
Link: https://patch.msgid.link/20250626073111.244534-1-idosch@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add test cases for externally validated neighbor entries, testing both
IPv4 and IPv6. Name the file "test_neigh.sh" so that it could be
possibly extended in the future with more neighbor test cases.
Example output:
# ./test_neigh.sh
TEST: IPv4 "extern_valid" flag: Add entry [ OK ]
TEST: IPv4 "extern_valid" flag: Add with an invalid state [ OK ]
TEST: IPv4 "extern_valid" flag: Add with "use" flag [ OK ]
TEST: IPv4 "extern_valid" flag: Replace entry [ OK ]
TEST: IPv4 "extern_valid" flag: Replace entry with "managed" flag [ OK ]
TEST: IPv4 "extern_valid" flag: Replace with an invalid state [ OK ]
TEST: IPv4 "extern_valid" flag: Interface down [ OK ]
TEST: IPv4 "extern_valid" flag: Carrier down [ OK ]
TEST: IPv4 "extern_valid" flag: Transition to "reachable" state [ OK ]
TEST: IPv4 "extern_valid" flag: Transition back to "stale" state [ OK ]
TEST: IPv4 "extern_valid" flag: Forced garbage collection [ OK ]
TEST: IPv4 "extern_valid" flag: Periodic garbage collection [ OK ]
TEST: IPv6 "extern_valid" flag: Add entry [ OK ]
TEST: IPv6 "extern_valid" flag: Add with an invalid state [ OK ]
TEST: IPv6 "extern_valid" flag: Add with "use" flag [ OK ]
TEST: IPv6 "extern_valid" flag: Replace entry [ OK ]
TEST: IPv6 "extern_valid" flag: Replace entry with "managed" flag [ OK ]
TEST: IPv6 "extern_valid" flag: Replace with an invalid state [ OK ]
TEST: IPv6 "extern_valid" flag: Interface down [ OK ]
TEST: IPv6 "extern_valid" flag: Carrier down [ OK ]
TEST: IPv6 "extern_valid" flag: Transition to "reachable" state [ OK ]
TEST: IPv6 "extern_valid" flag: Transition back to "stale" state [ OK ]
TEST: IPv6 "extern_valid" flag: Forced garbage collection [ OK ]
TEST: IPv6 "extern_valid" flag: Periodic garbage collection [ OK ]
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/20250626073111.244534-3-idosch@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tl;dr
=====
Add a new neighbor flag ("extern_valid") that can be used to indicate to
the kernel that a neighbor entry was learned and determined to be valid
externally. The kernel will not try to remove or invalidate such an
entry, leaving these decisions to the user space control plane. This is
needed for EVPN multi-homing where a neighbor entry for a multi-homed
host needs to be synced across all the VTEPs among which the host is
multi-homed.
Background
==========
In a typical EVPN multi-homing setup each host is multi-homed using a
set of links called ES (Ethernet Segment, i.e., LAG) to multiple leaf
switches (VTEPs). VTEPs that are connected to the same ES are called ES
peers.
When a neighbor entry is learned on a VTEP, it is distributed to both ES
peers and remote VTEPs using EVPN MAC/IP advertisement routes. ES peers
use the neighbor entry when routing traffic towards the multi-homed host
and remote VTEPs use it for ARP/NS suppression.
Motivation
==========
If the ES link between a host and the VTEP on which the neighbor entry
was locally learned goes down, the EVPN MAC/IP advertisement route will
be withdrawn and the neighbor entries will be removed from both ES peers
and remote VTEPs. Routing towards the multi-homed host and ARP/NS
suppression can fail until another ES peer locally learns the neighbor
entry and distributes it via an EVPN MAC/IP advertisement route.
"draft-rbickhart-evpn-ip-mac-proxy-adv-03" [1] suggests avoiding these
intermittent failures by having the ES peers install the neighbor
entries as before, but also injecting EVPN MAC/IP advertisement routes
with a proxy indication. When the previously mentioned ES link goes down
and the original EVPN MAC/IP advertisement route is withdrawn, the ES
peers will not withdraw their neighbor entries, but instead start aging
timers for the proxy indication.
If an ES peer locally learns the neighbor entry (i.e., it becomes
"reachable"), it will restart its aging timer for the entry and emit an
EVPN MAC/IP advertisement route without a proxy indication. An ES peer
will stop its aging timer for the proxy indication if it observes the
removal of the proxy indication from at least one of the ES peers
advertising the entry.
In the event that the aging timer for the proxy indication expired, an
ES peer will withdraw its EVPN MAC/IP advertisement route. If the timer
expired on all ES peers and they all withdrew their proxy
advertisements, the neighbor entry will be completely removed from the
EVPN fabric.
Implementation
==============
In the above scheme, when the control plane (e.g., FRR) advertises a
neighbor entry with a proxy indication, it expects the corresponding
entry in the data plane (i.e., the kernel) to remain valid and not be
removed due to garbage collection or loss of carrier. The control plane
also expects the kernel to notify it if the entry was learned locally
(i.e., became "reachable") so that it will remove the proxy indication
from the EVPN MAC/IP advertisement route. That is why these entries
cannot be programmed with dummy states such as "permanent" or "noarp".
Instead, add a new neighbor flag ("extern_valid") which indicates that
the entry was learned and determined to be valid externally and should
not be removed or invalidated by the kernel. The kernel can probe the
entry and notify user space when it becomes "reachable" (it is initially
installed as "stale"). However, if the kernel does not receive a
confirmation, have it return the entry to the "stale" state instead of
the "failed" state.
In other words, an entry marked with the "extern_valid" flag behaves
like any other dynamically learned entry other than the fact that the
kernel cannot remove or invalidate it.
One can argue that the "extern_valid" flag should not prevent garbage
collection and that instead a neighbor entry should be programmed with
both the "extern_valid" and "extern_learn" flags. There are two reasons
for not doing that:
1. Unclear why a control plane would like to program an entry that the
kernel cannot invalidate but can completely remove.
2. The "extern_learn" flag is used by FRR for neighbor entries learned
on remote VTEPs (for ARP/NS suppression) whereas here we are
concerned with local entries. This distinction is currently irrelevant
for the kernel, but might be relevant in the future.
Given that the flag only makes sense when the neighbor has a valid
state, reject attempts to add a neighbor with an invalid state and with
this flag set. For example:
# ip neigh add 192.0.2.1 nud none dev br0.10 extern_valid
Error: Cannot create externally validated neighbor with an invalid state.
# ip neigh add 192.0.2.1 lladdr 00:11:22:33:44:55 nud stale dev br0.10 extern_valid
# ip neigh replace 192.0.2.1 nud failed dev br0.10 extern_valid
Error: Cannot mark neighbor as externally validated with an invalid state.
The above means that a neighbor cannot be created with the
"extern_valid" flag and flags such as "use" or "managed" as they result
in a neighbor being created with an invalid state ("none") and
immediately getting probed:
# ip neigh add 192.0.2.1 lladdr 00:11:22:33:44:55 nud stale dev br0.10 extern_valid use
Error: Cannot create externally validated neighbor with an invalid state.
However, these flags can be used together with "extern_valid" after the
neighbor was created with a valid state:
# ip neigh add 192.0.2.1 lladdr 00:11:22:33:44:55 nud stale dev br0.10 extern_valid
# ip neigh replace 192.0.2.1 lladdr 00:11:22:33:44:55 nud stale dev br0.10 extern_valid use
One consequence of preventing the kernel from invalidating a neighbor
entry is that by default it will only try to determine reachability
using unicast probes. This can be changed using the "mcast_resolicit"
sysctl:
# sysctl net.ipv4.neigh.br0/10.mcast_resolicit
0
# tcpdump -nn -e -i br0.10 -Q out arp &
# ip neigh replace 192.0.2.1 lladdr 00:11:22:33:44:55 nud stale dev br0.10 extern_valid use
62:50:1d:11:93:6f > 00:11:22:33:44:55, ethertype ARP (0x0806), length 42: Request who-has 192.0.2.1 tell 192.0.2.2, length 28
62:50:1d:11:93:6f > 00:11:22:33:44:55, ethertype ARP (0x0806), length 42: Request who-has 192.0.2.1 tell 192.0.2.2, length 28
62:50:1d:11:93:6f > 00:11:22:33:44:55, ethertype ARP (0x0806), length 42: Request who-has 192.0.2.1 tell 192.0.2.2, length 28
# sysctl -wq net.ipv4.neigh.br0/10.mcast_resolicit=3
# ip neigh replace 192.0.2.1 lladdr 00:11:22:33:44:55 nud stale dev br0.10 extern_valid use
62:50:1d:11:93:6f > 00:11:22:33:44:55, ethertype ARP (0x0806), length 42: Request who-has 192.0.2.1 tell 192.0.2.2, length 28
62:50:1d:11:93:6f > 00:11:22:33:44:55, ethertype ARP (0x0806), length 42: Request who-has 192.0.2.1 tell 192.0.2.2, length 28
62:50:1d:11:93:6f > 00:11:22:33:44:55, ethertype ARP (0x0806), length 42: Request who-has 192.0.2.1 tell 192.0.2.2, length 28
62:50:1d:11:93:6f > ff:ff:ff:ff:ff:ff, ethertype ARP (0x0806), length 42: Request who-has 192.0.2.1 tell 192.0.2.2, length 28
62:50:1d:11:93:6f > ff:ff:ff:ff:ff:ff, ethertype ARP (0x0806), length 42: Request who-has 192.0.2.1 tell 192.0.2.2, length 28
62:50:1d:11:93:6f > ff:ff:ff:ff:ff:ff, ethertype ARP (0x0806), length 42: Request who-has 192.0.2.1 tell 192.0.2.2, length 28
iproute2 patches can be found here [2].
[1] https://datatracker.ietf.org/doc/html/draft-rbickhart-evpn-ip-mac-proxy-adv-03
[2] https://github.com/idosch/iproute2/tree/submit/extern_valid_v1
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://patch.msgid.link/20250626073111.244534-2-idosch@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This is just cosmetics and so no functional changes intended.
While at it, sort header in alphabetical order.
Signed-off-by: Nuno Sá <nuno.sa@analog.com>
Link: https://lore.kernel.org/r/20250519-dev-axi-clkgen-limits-v6-7-bc4b3b61d1d4@analog.com
Reviewed-by: David Lechner <dlechner@baylibre.com>
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
Instead of using the type versions of min/max(), use the plain ones as
now they are perfectly capable of handling different types like
unsigned and non negative integers that are compiletime constant.
Signed-off-by: Nuno Sá <nuno.sa@analog.com>
Link: https://lore.kernel.org/r/20250519-dev-axi-clkgen-limits-v6-6-bc4b3b61d1d4@analog.com
Reviewed-by: David Lechner <dlechner@baylibre.com>
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
This patch adds support for setting the limits in struct
axi_clkgen_limits in accordance with fpga speed grade, voltage,
technology and family. This new information is extracted from
two new registers implemented in the ip core that are only available
for core versions higher or equal to 4.
Signed-off-by: Nuno Sá <nuno.sa@analog.com>
Link: https://lore.kernel.org/r/20250519-dev-axi-clkgen-limits-v6-5-bc4b3b61d1d4@analog.com
Reviewed-by: David Lechner <dlechner@baylibre.com>
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
Add new helper macros and enums to help identifying the platform and some
characteristics of it at runtime.
Signed-off-by: Nuno Sá <nuno.sa@analog.com>
Link: https://lore.kernel.org/r/20250519-dev-axi-clkgen-limits-v6-4-bc4b3b61d1d4@analog.com
Reviewed-by: David Lechner <dlechner@baylibre.com>
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
The adi-axi-common.h header has some common defines used in various ADI
IPs. However they are not specific for any fpga manager so it's
questionable for the header to live under include/linux/fpga. Hence
let's just move one directory up and update all users.
Suggested-by: Xu Yilun <yilun.xu@linux.intel.com>
Acked-by: Xu Yilun <yilun.xu@intel.com>
Acked-by: Jonathan Cameron <Jonathan.Cameron@huawei.com> # for IIO
Signed-off-by: Nuno Sá <nuno.sa@analog.com>
Link: https://lore.kernel.org/r/20250519-dev-axi-clkgen-limits-v6-3-bc4b3b61d1d4@analog.com
Acked-by: Mark Brown <broonie@kernel.org>
Acked-by: Uwe Kleine-König <ukleinek@kernel.org>
Reviewed-by: David Lechner <dlechner@baylibre.com>
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
The fpfd_max frequency should be set to 450 MHz instead of 300 MHz.
Well, it actually depends on the platform speed grade but we are being
conservative for ultrascale so let's be consistent. In a following
change we will set these limits at runtime.
Fixes: 0e646c52cf ("clk: Add axi-clkgen driver")
Signed-off-by: Nuno Sá <nuno.sa@analog.com>
Link: https://lore.kernel.org/r/20250519-dev-axi-clkgen-limits-v6-1-bc4b3b61d1d4@analog.com
Reviewed-by: David Lechner <dlechner@baylibre.com>
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
The dev_dbg() message in cxl_set_ecs_threshold() used %d for
an unsigned value, which could lead to incorrect logging.
Update the format specifier to %u to match variable type.
Signed-off-by: Alok Tiwari <alok.a.tiwari@oracle.com>
Reviewed-by: Shiju Jose <shiju.jose@huawei.com>
Reviewed-by: Alison Schofield <alison.schofield@intel.com?>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Link: https://patch.msgid.link/20250622183919.4156343-1-alok.a.tiwari@oracle.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Fix several typos and improve comment clarity in the CXL device types
docs:
"w/" replaced with "with"
"sill" -> "still"
"The allows" -> "This allows"
"capacity" corrected to "capable"
"more devices" corrected to "more upstream devices" in MLD description
These changes improve readability and enhance the documentation quality.
[ dj: Fix up "one or more hosts" to "one or more upstream devices" from
Gregory ]
Signed-off-by: Alok Tiwari <alok.a.tiwari@oracle.com>
Reviewed-by: Gregory Price <gourry@gourry.net>
Link: https://patch.msgid.link/20250616060737.1645393-1-alok.a.tiwari@oracle.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Fix typo 'enumates' to 'enumerate' in CXL driver operation
documentation to improve readability.
Signed-off-by: Nai-Chen Cheng <bleach1827@gmail.com>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Reviewed-by: Li Ming <ming.li@zohomail.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Link: https://patch.msgid.link/20250610173152.33566-1-bleach1827@gmail.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
This patch corrects several typographical issues and improves phrasing
in memory-devices.rst:
- Fixes duplicate word ("1 one") and adjusts phrasing for clarity.
- Adds missing hyphen in "on-device".
- Corrects "a give memory device" to "a given memory device".
- fix singular/plural "decoder resource" -> "decoder resources".
- Clarifies "spans to Host Bridges" -> "spans two Host Bridges".
- change "at a" -> "a"
These changes improve readability and accuracy of the documentation.
Signed-off-by: Alok Tiwari <alok.a.tiwari@oracle.com>
Reviewed-by: Randy Dunlap <rdunlap@infradead.org>
Reviewed-by: Gregory Price <gourry@gourry.net>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Link: https://patch.msgid.link/20250609171130.2375901-1-alok.a.tiwari@oracle.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Some architectures (e.g. arm64) only support memory hotplug operations on
a restricted set of physical addresses. This applies even when we are
faking some CXL fixed memory windows for the purposes of cxl_test.
That range can be queried with mhp_get_pluggable_range(true). Use the
minimum of that the top of that range and iomem_resource.end to establish
the 64GiB region used by cxl_test.
From thread #2 which was related to the issue in #1.
[ dj: Add CONFIG_MEMORY_HOTPLUG config check, from Alison ]
Link: https://lore.kernel.org/linux-cxl/20250522145622.00002633@huawei.com/#2
Reported-by: Itaru Kitayama <itaru.kitayama@linux.dev>
Closes: https://github.com/pmem/ndctl/issues/278#1
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Tested-by: Itaru Kitayama <itaru.kitayama@fujitsu.com <mailto:itaru.kitayama@fujitsu.com>
Tested-by: Marc Herbert <marc.herbert@linux.intel.com>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Link: https://patch.msgid.link/20250527153451.82858-1-Jonathan.Cameron@huawei.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
mutex_lock_io() differs from mutex_lock() in that it may call
io_schedule() when a task must sleep waiting for the lock. This
distinction only makes sense in block I/O or memory reclaim paths,
where giving I/O a chance to make progress is useful.
At this call site, cxl_pci_mbox_send(), the mutex protects an MMIO
mailbox. The task holding the lock is not blocking I/O progress, so
calling io_schedule(), as mutex_lock_io() may do, has no practical
effect.
Although there is no functional change, using the correct locking
primitive, that more accurately reflects the semantics and intended
use of the lock, improves code clarity and avoids misleading readers
and tools.
[ dj: Dropped fixes tag, no need to backport ]
Reported-by: Alok Tiwari <alok.a.tiwari@oracle.com>
Closes: https://lore.kernel.org/linux-cxl/0d2af1e8-7f1b-438c-a090-fd366c8c63e0@oracle.com/
Suggested-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Alison Schofield <alison.schofield@intel.com>
Reviewed-by: Davidlohr Bueso <dave@stgolabs.net>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Link: https://patch.msgid.link/20250529205117.1990465-1-alison.schofield@intel.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
There exists shipping platforms that bend, break, or otherwise lean on
ambiguities in the CXL specification. Without driver changes to accommodate
these deviations, end users are left without CXL subsystem RAS features.
Specifically, provisioning, error translation, and other flows require the
CXL subsystem to understand the platforms CXL topology beyond undecorated
memory address ranges.
Those isolated compatibility problems risk growing into deeper upstream
maintenance burden if different platform vendors arrive at diverging
solutions. For example, there are multiple options for resolving
low-memory-mmio intersecting large-interleave-ways CXL windows. Linux
should only entertain one solution to that problem.
Now, with the ACPI Specification Working Group, situations like this would
be resolved with the "Code First ECN" process to codify Linux expectations
in a specification. In the absence of such a process for the CXL
specification, create a file in Linux documentation to detail the
motivations, assumptions, tradeoffs, and proposals for amending
specification language.
The goal is to capture the issues such that platform vendors arrive at
compatible solutions for these problems and serve as a repository for
potential specification updates. The expectation is to update
conventions.rst along with CXL subsystem code changes to accommodate the
platform topology.
[ dj: Rebased against v6.16-rc1 ]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Robert Richter <rrichter@amd.com>
Link: https://patch.msgid.link/20250603185254.3730099-1-dan.j.williams@intel.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
-----BEGIN PGP SIGNATURE-----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=F1L2
-----END PGP SIGNATURE-----
Merge tag 'io_uring-6.16-20250630' of git://git.kernel.dk/linux
Pull io_uring fix from Jens Axboe:
"Now that anonymous inodes set S_IFREG, this breaks the io_uring
read/write retries for short reads/writes. As things like timerfd and
eventfd are anon inodes, applications that previously did:
unsigned long event_data[2];
io_uring_prep_read(sqe, evfd, event_data, sizeof(event_data), 0);
and just got a short read when 1 event was posted, will now wait for
the full amount before posting a completion.
This caused issues for the ghostty application, making it basically
unusable due to excessive buffering"
* tag 'io_uring-6.16-20250630' of git://git.kernel.dk/linux:
io_uring: gate REQ_F_ISREG on !S_ANON_INODE as well
When using Secure TSC, the GUEST_TSC_FREQ MSR reports a frequency based on
the nominal P0 frequency, which deviates slightly (typically ~0.2%) from
the actual mean TSC frequency due to clocking parameters.
Over extended VM uptime, this discrepancy accumulates, causing clock skew
between the hypervisor and a SEV-SNP VM, leading to early timer interrupts as
perceived by the guest.
The guest kernel relies on the reported nominal frequency for TSC-based
timekeeping, while the actual frequency set during SNP_LAUNCH_START may
differ. This mismatch results in inaccurate time calculations, causing the
guest to perceive hrtimers as firing earlier than expected.
Utilize the TSC_FACTOR from the SEV firmware's secrets page (see "Secrets
Page Format" in the SNP Firmware ABI Specification) to calculate the mean
TSC frequency, ensuring accurate timekeeping and mitigating clock skew in
SEV-SNP VMs.
Use early_ioremap_encrypted() to map the secrets page as
ioremap_encrypted() uses kmalloc() which is not available during early TSC
initialization and causes a panic.
[ bp: Drop the silly dummy var:
https://lore.kernel.org/r/20250630192726.GBaGLlHl84xIopx4Pt@fat_crate.local ]
Fixes: 73bbf3b0fb ("x86/tsc: Init the TSC for Secure TSC guests")
Signed-off-by: Nikunj A Dadhania <nikunj@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/20250630081858.485187-1-nikunj@amd.com
UBLK_IO_FLAG_ACTIVE and UBLK_IO_FLAG_OWNED_BY_SRV are mutually
exclusive. So just check that UBLK_IO_FLAG_OWNED_BY_SRV is set in
__ublk_ch_uring_cmd(); that implies UBLK_IO_FLAG_ACTIVE is unset.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-7-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The variable is computed from a simple expression and used once, so just
replace it with the expression.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-6-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Check for UBLK_IO_FETCH_REQ early in __ublk_ch_uring_cmd() and skip the
rest of the checks in this case. This allows removing the checks for
NULL io->task and UBLK_IO_FLAG_OWNED_BY_SRV unset in io->flags, which
are only allowed for FETCH.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-5-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In preparation for skipping some of the other checks for certain IO
opcodes, move the cmd_op check earlier.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-4-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
__ublk_check_and_get_req() attempts to atomically look up the struct
request for a ublk I/O and take a reference on it. However, the request
can be freed between the lookup on the tagset in blk_mq_tag_to_rq() and
the increment of its reference count in ublk_get_req_ref(), for example
if an elevator switch happens concurrently.
Fix the potential use after free by moving the reference count from
ublk_rq_data to ublk_io. Move the fields buf_index and buf_ctx_handle
too to reduce the number of cache lines touched when dispatching and
completing a ublk I/O, allowing ublk_rq_data to be removed entirely.
Suggested-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Fixes: 62fe99cef9 ("ublk: add read()/write() support for ublk char device")
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-3-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
struct ublk_device's __queues points to an allocation with up to
UBLK_MAX_NR_QUEUES (4096) queues, each of which have:
- struct ublk_queue (48 bytes)
- Tail array of up to UBLK_MAX_QUEUE_DEPTH (4096) struct ublk_io's,
32 bytes each
This means the full allocation can exceed 512 MB, which may well be
impossible to service with contiguous physical pages. Switch to
kvcalloc() and kvfree(), since there is no need for physically
contiguous memory.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Fixes: 71f28f3136 ("ublk_drv: add io_uring based userspace block driver")
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250620151008.3976463-2-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The current use of an always_inline helper is a bit convoluted.
Instead use macros that represent the arithmetics used for building
up the PRP chain.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Daniel Gomez <da.gomez@samsung.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20250625113531.522027-9-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Having a define in kiB units is a bit weird. Also update the
comment now that there is not scatterlist limit.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Daniel Gomez <da.gomez@samsung.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20250625113531.522027-8-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Use the blk_rq_dma_map API to DMA map requests instead of scatterlists.
This removes the need to allocate a scatterlist covering every segment,
and thus the overall transfer length limit based on the scatterlist
allocation.
Instead the DMA mapping is done by iterating the bio_vec chain in the
request directly. The unmap is handled differently depending on how
we mapped:
- when using an IOMMU only a single IOVA is used, and it is stored in
iova_state
- for direct mappings that don't use swiotlb and are cache coherent,
unmap is not needed at all
- for direct mappings that are not cache coherent or use swiotlb, the
physical addresses are rebuild from the PRPs or SGL segments
The latter unfortunately adds a fair amount of code to the driver, but
it is code not used in the fast path.
The conversion only covers the data mapping path, and still uses a
scatterlist for the multi-segment metadata case. I plan to convert that
as soon as we have good test coverage for the multi-segment metadata
path.
Thanks to Chaitanya Kulkarni for an initial attempt at a new DMA API
conversion for nvme-pci, Kanchan Joshi for bringing back the single
segment optimization, Leon Romanovsky for shepherding this through a
gazillion rebases and Nitesh Shetty for various improvements.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20250625113531.522027-7-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
