This modifies the ADI AXI SPI Engine driver to restore the clkdiv
configuration register at the end of a SPI message. Having the clkdiv in
a known state is needed to be able to add a new command in the future
that only performs a delay without any SPI transfers. Furthermore having
that state be the smallest possible divider will allow these delays to
have the highest possible precision.
Changing the initial value of clk_div from -1 to 1 is now possible
because we know the function will always be called with a known clkdiv
config register state. Making this change will also have the effect of
not emitting a clkdiv configuration register instruction in cases where
the maximum sclk rate is used. Having one less instruction to process
reduces delays on the bus which will be beneficial when we implement
offload support to enable reading data from devices at very high rates.
Signed-off-by: David Lechner <dlechner@baylibre.com>
Acked-by: Michael Hennerich <michael.hennerich@analog.com>
Acked-by: Nuno Sa <nuno.sa@analog.com>
Link: https://lore.kernel.org/r/20231204-axi-spi-engine-series-2-v1-7-063672323fce@baylibre.com
Signed-off-by: Mark Brown <broonie@kernel.org>
This fixes the sleep ticks calculation when generating sleep
instructions in the AXI SPI Engine driver. The previous calculation
was ignoring delays less than one microsecond and missed a microsecond
to second conversion factor.
This fixes the first issue by not rounding to microseconds. Now that
xfer->effective_speed_hz is guaranteed to be set correctly, we can use
that to simplify the calculation. This new calculation replaces the old
incorrect math.
Also add unit suffix to the delay variable for clarity while we are
touching this.
Signed-off-by: David Lechner <dlechner@baylibre.com>
Acked-by: Michael Hennerich <michael.hennerich@analog.com>
Acked-by: Nuno Sa <nuno.sa@analog.com>
Link: https://lore.kernel.org/r/20231204-axi-spi-engine-series-2-v1-4-063672323fce@baylibre.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Now that host->max_speed_hz and xfer->effective_speed_hz are properly
set, we can use them instead of having to do more complex calculations
to get the clock divider for each transfer.
This removes the spi_engine_get_clk_div() function and replaces it with
just dividing the two clock rates. Since the hardware register value is
the divider minus one, we need to subtract one. Subtracting one was
previously done in the spi_engine_get_clk_div() function.
Signed-off-by: David Lechner <dlechner@baylibre.com>
Acked-by: Michael Hennerich <michael.hennerich@analog.com>
Acked-by: Nuno Sa <nuno.sa@analog.com>
Link: https://lore.kernel.org/r/20231204-axi-spi-engine-series-2-v1-3-063672323fce@baylibre.com
Signed-off-by: Mark Brown <broonie@kernel.org>
This adds a new spi_engine_precompile_message() function to the ADI AXI
SPI Engine driver to populate the xfer->effective_speed_hz field since
the SPI core doesn't/can't do this for us.
This driver is already using spi_delay_to_ns() which depends on
effective_speed_hz to get an accurate value in some cases.
Having an effective_speed_hz value can also be used in future changes
to simplify other code.
Signed-off-by: David Lechner <dlechner@baylibre.com>
Acked-by: Michael Hennerich <michael.hennerich@analog.com>
Acked-by: Nuno Sa <nuno.sa@analog.com>
Link: https://lore.kernel.org/r/20231204-axi-spi-engine-series-2-v1-2-063672323fce@baylibre.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Merge series from David Lechner <dlechner@baylibre.com>:
We are working towards adding support for the offload feature[1] of the
AXI SPI Engine IP core. Before we can do that, we want to make some
general fixes and improvements to the driver. In order to avoid a giant
series with 35+ patches, we are splitting this up into a few smaller
series.
This first series mostly doing some housekeeping:
* Convert device tree bindings to yaml.
* Add a MAINTAINERS entry.
* Clean up probe and remove using devm.
* Separate message state from driver state.
* Add support for cs_off and variable word size.
Once this series is applied, we will follow up with a second series of
general improvements, and then finally a 3rd series that implements the
offload support. The offload support will also involve the IIO
subsystem (a new IIO driver will depend on the new SPI offload feature),
so I'm mentioning this now in case we want to do anything ahead of time
to prepare for that (e.g. putting all of these changes on a separate
branch).
[1]: https://wiki.analog.com/resources/fpga/peripherals/spi_engine/offload
The AXI SPI Engine IP supports any word size from 1 to 32 bits. This
adds support for this by setting the bits_per_word_mask and emitting
the appropriate instruction to the SPI Engine each time a transfer
requires a new word size.
The functions that transfer tx/rx buffers from/to the SPI Engine
registers (spi_engine_write_{tx,rx}_fifo()) as well as the function that
creates the transfer instruction (spi_engine_gen_xfer()) also have to be
modified to take into account the word size since xfer->len is the
size of the buffers in bytes rather than words.
Signed-off-by: David Lechner <dlechner@baylibre.com>
Link: https://lore.kernel.org/r/20231117-axi-spi-engine-series-1-v1-14-cc59db999b87@baylibre.com
Signed-off-by: Mark Brown <broonie@kernel.org>
In the AXI SPI Engine driver, the struct spi_engine::msg member was used
to keep track of the current message being processed. The SPI core is
already keeping track of this, so we don't need to duplicate the effort.
In most cases, we already have a pointer to the current message, so we
can pass it directly to the functions that need it. In the one case
where we don't have a pointer to the current message, we can get it
from struct spi_controller::cur_msg.
Signed-off-by: David Lechner <dlechner@baylibre.com>
Link: https://lore.kernel.org/r/20231117-axi-spi-engine-series-1-v1-12-cc59db999b87@baylibre.com
Signed-off-by: Mark Brown <broonie@kernel.org>
This modifies the AXI SPI Engine driver to make use of the
message_prepare and message_unprepare callbacks. This separates
the concerns of allocating and freeing the message state from the
transfer_one_message callback.
The main benfit of this is so that future callers of
spi_finalize_current_message() will not have to do manual cleanup
of the state.
Signed-off-by: David Lechner <dlechner@baylibre.com>
Link: https://lore.kernel.org/r/20231117-axi-spi-engine-series-1-v1-10-cc59db999b87@baylibre.com
Signed-off-by: Mark Brown <broonie@kernel.org>
This moves the message state in the AXI SPI Engine driver to a new
struct spi_engine_msg_state.
Previously, the driver state contained various pointers that pointed
to memory owned by a struct spi_message. However, it did not set any of
these pointers to NULL when a message was completed. This could lead to
use after free bugs.
Example of how this could happen:
1. SPI core calls into spi_engine_transfer_one_message() with msg1.
2. Assume something was misconfigured and spi_engine_tx_next() is not
called enough times in interrupt callbacks for msg1 such that
spi_engine->tx_xfer is never set to NULL before the msg1 completes.
3. SYNC interrupt is received and spi_finalize_current_message() is
called for msg1. spi_engine->msg is set to NULL but no other
message-specific state is reset.
4. Caller that sent msg1 is notified of the completion and frees msg1
and the associated xfers and tx/rx buffers.
4. SPI core calls into spi_engine_transfer_one_message() with msg2.
5. When spi_engine_tx_next() is called for msg2, spi_engine->tx_xfer is
still be pointing to an xfer from msg1, which was already freed.
spi_engine_xfer_next() tries to access xfer->transfer_list of one
of the freed xfers and we get a segfault or undefined behavior.
To avoid issues like this, instead of putting per-message state in the
driver state struct, we can make use of the struct spi_message::state
field to store a pointer to a new struct spi_engine_msg_state. This way,
all of the state that belongs to specific message stays with that
message and we don't have to remember to manually reset all aspects of
the message state when a message is completed. Rather, a new state is
allocated for each message.
Most of the changes are just renames where the state is accessed. One
place where this wasn't straightforward was the sync_id member. This
has been changed to use ida_alloc_range() since we needed to separate
the per-message sync_id from the per-controller next available sync_id.
Signed-off-by: David Lechner <dlechner@baylibre.com>
Link: https://lore.kernel.org/r/20231117-axi-spi-engine-series-1-v1-9-cc59db999b87@baylibre.com
Signed-off-by: Mark Brown <broonie@kernel.org>
This modifies the AXI SPI Engine driver to use devm_spi_alloc_host()
instead of spi_alloc_host() to simplify the code a bit.
In addition to simplifying the error paths in the probe function, we
can also remove spi_controller_get/put() calls in the remove function
since devm_spi_alloc_host() sets a flag to no longer decrement the
controller reference count in the spi_unregister_controller() function.
Signed-off-by: David Lechner <dlechner@baylibre.com>
Link: https://lore.kernel.org/r/20231117-axi-spi-engine-series-1-v1-4-cc59db999b87@baylibre.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Pull Kbuild fixes from Masahiro Yamada:
- Fix section mismatch warning messages for riscv and loongarch
- Remove CONFIG_IA64 left-over from linux/export-internal.h
- Fix the location of the quotes for UIMAGE_NAME
- Fix a memory leak bug in Kconfig
* tag 'kbuild-fixes-v6.7' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
kconfig: fix memory leak from range properties
kbuild: Move the single quotes for image name
linux/export: clean up the IA-64 KSYM_FUNC macro
modpost: fix section mismatch message for RELA
Pull irq fix from Borislav Petkov:
- Flush the translation service tables to prevent unpredictable
behavior on non-coherent GIC devices
* tag 'irq_urgent_for_v6.7_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
irqchip/gic-v3-its: Flush ITS tables correctly in non-coherent GIC designs
Pull x86 fixes from Borislav Petkov:
- Ignore invalid x2APIC entries in order to not waste per-CPU data
- Fix a back-to-back signals handling scenario when shadow stack is in
use
- A documentation fix
- Add Kirill as TDX maintainer
* tag 'x86_urgent_for_v6.7_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/acpi: Ignore invalid x2APIC entries
x86/shstk: Delay signal entry SSP write until after user accesses
x86/Documentation: Indent 'note::' directive for protocol version number note
MAINTAINERS: Add Intel TDX entry
Pull timer fix from Borislav Petkov:
- Do the push of pending hrtimers away from a CPU which is being
offlined earlier in the offlining process in order to prevent a
deadlock
* tag 'timers_urgent_for_v6.7_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
hrtimers: Push pending hrtimers away from outgoing CPU earlier
Pull scheduler fixes from Borislav Petkov:
- Fix virtual runtime calculation when recomputing a sched entity's
weights
- Fix wrongly rejected unprivileged poll requests to the cgroup psi
pressure files
- Make sure the load balancing is done by only one CPU
* tag 'sched_urgent_for_v6.7_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/fair: Fix the decision for load balance
sched: psi: fix unprivileged polling against cgroups
sched/eevdf: Fix vruntime adjustment on reweight
Pull locking fix from Borislav Petkov:
- Fix a hardcoded futex flags case which lead to one robust futex test
failure
* tag 'locking_urgent_for_v6.7_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
futex: Fix hardcoded flags
Pull perf fix from Borislav Petkov:
- Make sure the context refcount is transferred too when migrating perf
events
* tag 'perf_urgent_for_v6.7_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf/core: Fix cpuctx refcounting
Pull SCSI fixes from James Bottomley:
"Seven small fixes, six in drivers and one in sd.
The sd fix is so large because it changes a struct pointer to a struct
but otherwise is fairly simple"
* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
scsi: ufs: qcom-ufs: dt-bindings: Document the SM8650 UFS Controller
scsi: sd: Fix sshdr use in sd_suspend_common()
scsi: scsi_debug: Delete some bogus error checking
scsi: scsi_debug: Fix some bugs in sdebug_error_write()
scsi: ufs: core: Fix racing issue between ufshcd_mcq_abort() and ISR
scsi: ufs: core: Expand MCQ queue slot to DeviceQueueDepth + 1
scsi: qla2xxx: Fix system crash due to bad pointer access
Pull parisc fixes from Helge Deller:
"On parisc we still sometimes need writeable stacks, e.g. if programs
aren't compiled with gcc-14. To avoid issues with the upcoming
systemd-254 we therefore have to disable prctl(PR_SET_MDWE) for now
(for parisc only).
The other two patches are minor: a bugfix for the soft power-off on
qemu with 64-bit kernel and prefer strscpy() over strlcpy():
- Fix power soft-off on qemu
- Disable prctl(PR_SET_MDWE) since parisc sometimes still needs
writeable stacks
- Use strscpy instead of strlcpy in show_cpuinfo()"
* tag 'parisc-for-6.7-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
prctl: Disable prctl(PR_SET_MDWE) on parisc
parisc/power: Fix power soft-off when running on qemu
parisc: Replace strlcpy() with strscpy()
Pull xfs fixes from Chandan Babu:
- Fix deadlock arising due to intent items in AIL not being cleared
when log recovery fails
- Fix stale data exposure bug when remapping COW fork extents to data
fork
- Fix deadlock when data device flush fails
- Fix AGFL minimum size calculation
- Select DEBUG_FS instead of XFS_DEBUG when XFS_ONLINE_SCRUB_STATS is
selected
- Fix corruption of log inode's extent count field when NREXT64 feature
is enabled
* tag 'xfs-6.7-fixes-1' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux:
xfs: recovery should not clear di_flushiter unconditionally
xfs: inode recovery does not validate the recovered inode
xfs: fix again select in kconfig XFS_ONLINE_SCRUB_STATS
xfs: fix internal error from AGFL exhaustion
xfs: up(ic_sema) if flushing data device fails
xfs: only remap the written blocks in xfs_reflink_end_cow_extent
XFS: Update MAINTAINERS to catch all XFS documentation
xfs: abort intent items when recovery intents fail
xfs: factor out xfs_defer_pending_abort
Pull nfsd fixes from Chuck Lever:
- Fix several long-standing bugs in the duplicate reply cache
- Fix a memory leak
* tag 'nfsd-6.7-1' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux:
NFSD: Fix checksum mismatches in the duplicate reply cache
NFSD: Fix "start of NFS reply" pointer passed to nfsd_cache_update()
NFSD: Update nfsd_cache_append() to use xdr_stream
nfsd: fix file memleak on client_opens_release
Pull smb client fixes from Steve French:
- multichannel fixes (including a lock ordering fix and an important
refcounting fix)
- spnego fix
* tag '6.7-rc1-smb3-client-fixes' of git://git.samba.org/sfrench/cifs-2.6:
cifs: fix lock ordering while disabling multichannel
cifs: fix leak of iface for primary channel
cifs: fix check of rc in function generate_smb3signingkey
cifs: spnego: add ';' in HOST_KEY_LEN
Pull device mapper fixes from Mike Snitzer:
- Various fixes for the DM delay target to address regressions
introduced during the 6.7 merge window
- Fixes to both DM bufio and the verity target for no-sleep mode,
to address sleeping while atomic issues
- Update DM crypt target in response to the treewide change that
made MAX_ORDER inclusive
* tag 'for-6.7/dm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
dm-crypt: start allocating with MAX_ORDER
dm-verity: don't use blocking calls from tasklets
dm-bufio: fix no-sleep mode
dm-delay: avoid duplicate logic
dm-delay: fix bugs introduced by kthread mode
dm-delay: fix a race between delay_presuspend and delay_bio
Firmware returns the physical address of the power switch,
so need to use gsc_writel() instead of direct memory access.
Fixes: d0c2194729 ("parisc/power: Add power soft-off when running on qemu")
Signed-off-by: Helge Deller <deller@gmx.de>
Cc: stable@vger.kernel.org # v6.0+
Pull i2c fixes from Wolfram Sang:
"Revert a not-working conversion to generic recovery for PXA,
use proper IO accessors for designware, and use proper PM level
for ocores to allow accessing interrupt providers late"
* tag 'i2c-for-6.7-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
i2c: ocores: Move system PM hooks to the NOIRQ phase
i2c: designware: Fix corrupted memory seen in the ISR
Revert "i2c: pxa: move to generic GPIO recovery"
Pull turbostat updates from Len Brown:
- Turbostat features are now table-driven (Rui Zhang)
- Add support for some new platforms (Sumeet Pawnikar, Rui Zhang)
- Gracefully run in configs when CPUs are limited (Rui Zhang, Srinivas
Pandruvada)
- misc minor fixes
[ This came in during the merge window, but sorting out the signed tag
took a while, so thus the late merge - Linus ]
* tag 'turbostat-2023.11.07' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux: (86 commits)
tools/power turbostat: version 2023.11.07
tools/power/turbostat: bugfix "--show IPC"
tools/power/turbostat: Add initial support for LunarLake
tools/power/turbostat: Add initial support for ArrowLake
tools/power/turbostat: Add initial support for GrandRidge
tools/power/turbostat: Add initial support for SierraForest
tools/power/turbostat: Add initial support for GraniteRapids
tools/power/turbostat: Add MSR_CORE_C1_RES support for spr_features
tools/power/turbostat: Move process to root cgroup
tools/power/turbostat: Handle cgroup v2 cpu limitation
tools/power/turbostat: Abstrct function for parsing cpu string
tools/power/turbostat: Handle offlined CPUs in cpu_subset
tools/power/turbostat: Obey allowed CPUs for system summary
tools/power/turbostat: Obey allowed CPUs for primary thread/core detection
tools/power/turbostat: Abstract several functions
tools/power/turbostat: Obey allowed CPUs during startup
tools/power/turbostat: Obey allowed CPUs when accessing CPU counters
tools/power/turbostat: Introduce cpu_allowed_set
tools/power/turbostat: Remove PC7/PC9 support on ADL/RPL
tools/power/turbostat: Enable MSR_CORE_C1_RES on recent Intel client platforms
...
Pull bcachefs fixes from Kent Overstreet:
"Lots of small fixes for minor nits and compiler warnings.
Bigger items:
- The six locks lost wakeup is finally fixed: six_read_trylock() was
checking for the waiting bit before decrementing the number of
readers - validated the fix with a torture test.
- Fix for a memory reclaim issue: when needing to reallocate a key
cache key, we now do our usual GFP_NOWAIT; unlock(); GFP_KERNEL
dance.
- Multiple deleted inodes btree fixes
- Fix an issue in fsck, where i_nlink would be recalculated
incorrectly for hardlinked files if a snapshot had ever been taken.
- Kill journal pre-reservations: This is a bigger patch than I would
normally send at this point, but it deletes code and it fixes some
of our tests that would sporadically die with the journal getting
stuck, and it's a performance improvement, too"
* tag 'bcachefs-2023-11-17' of https://evilpiepirate.org/git/bcachefs: (22 commits)
bcachefs: Fix missing locking for dentry->d_parent access
bcachefs: six locks: Fix lost wakeup
bcachefs: Fix no_data_io mode checksum check
bcachefs: Fix bch2_check_nlinks() for snapshots
bcachefs: Don't decrease BTREE_ITER_MAX when LOCKDEP=y
bcachefs: Disable debug log statements
bcachefs: Fix missing transaction commit
bcachefs: Fix error path in bch2_mount()
bcachefs: Fix potential sleeping during mount
bcachefs: Fix iterator leak in may_delete_deleted_inode()
bcachefs: Kill journal pre-reservations
bcachefs: Check for nonce offset inconsistency in data_update path
bcachefs: Make sure to drop/retake btree locks before reclaim
bcachefs: btree_trans->write_locked
bcachefs: Run btree key cache shrinker less aggressively
bcachefs: Split out btree_key_cache_types.h
bcachefs: Guard against insufficient devices to create stripes
bcachefs: Fix null ptr deref in bch2_backpointer_get_node()
bcachefs: Fix multiple -Warray-bounds warnings
bcachefs: Use DECLARE_FLEX_ARRAY() helper and fix multiple -Warray-bounds warnings
...
Pull misc fixes from Andrew Morton:
"Thirteen hotfixes. Seven are cc:stable and the remainder pertain to
post-6.6 issues or aren't considered suitable for backporting"
* tag 'mm-hotfixes-stable-2023-11-17-14-04' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
mm: more ptep_get() conversion
parisc: fix mmap_base calculation when stack grows upwards
mm/damon/core.c: avoid unintentional filtering out of schemes
mm: kmem: drop __GFP_NOFAIL when allocating objcg vectors
mm/damon/sysfs-schemes: handle tried region directory allocation failure
mm/damon/sysfs-schemes: handle tried regions sysfs directory allocation failure
mm/damon/sysfs: check error from damon_sysfs_update_target()
mm: fix for negative counter: nr_file_hugepages
selftests/mm: add hugetlb_fault_after_madv to .gitignore
selftests/mm: restore number of hugepages
selftests: mm: fix some build warnings
selftests: mm: skip whole test instead of failure
mm/damon/sysfs: eliminate potential uninitialized variable warning
Pull block fix from Jens Axboe:
"Just a single fix from Christoph/Ming, fixing a case where integrity
IO could be called without having an appropriate queue reference"
* tag 'block-6.7-2023-11-17' of git://git.kernel.dk/linux:
blk-mq: make sure active queue usage is held for bio_integrity_prep()
Pull io_uring fix from Jens Axboe:
"Just a single fixup for a change we made in this release, which caused
a regression in sometimes missing fdinfo output if the SQPOLL thread
had the lock held when fdinfo output was retrieved.
This brings us back on par with what we had before, where just the
main uring_lock will prevent that output. We'd love to get rid of that
too, but that is beyond the scope of this release and will have to
wait for 6.8"
* tag 'io_uring-6.7-2023-11-17' of git://git.kernel.dk/linux:
io_uring/fdinfo: remove need for sqpoll lock for thread/pid retrieval
Pull drm fixes from Daniel Vetter:
"This is a 'blast from the bast' fixes pull, because it contains a
bunch of AGP fixes for amdgpu. Otherwise nothing out of the ordinary.
Next week is back to Dave unless he's knocked out by some conference
bug.
- amdgpu: fixes all over, including a set of AGP fixes
- nouvea: GSP + other bugfixes
- ivpu build fix
- lenovo legion go panel orientation quirk"
* tag 'drm-fixes-2023-11-17' of git://anongit.freedesktop.org/drm/drm: (30 commits)
drm/amdgpu/gmc9: disable AGP aperture
drm/amdgpu/gmc10: disable AGP aperture
drm/amdgpu/gmc11: disable AGP aperture
drm/amdgpu: add a module parameter to control the AGP aperture
drm/amdgpu/gmc11: fix logic typo in AGP check
drm/amd/display: Fix encoder disable logic
drm/amd/display: Change the DMCUB mailbox memory location from FB to inbox
drm/amdgpu: add and populate the port num into xgmi topology info
drm/amd/display: Negate IPS allow and commit bits
drm/amd/pm: Don't send unload message for reset
drm/amdgpu: fix ras err_data null pointer issue in amdgpu_ras.c
drm/amd/display: Clear dpcd_sink_ext_caps if not set
drm/amd/display: Enable fast plane updates on DCN3.2 and above
drm/amd/display: fix NULL dereference
drm/amd/display: fix a NULL pointer dereference in amdgpu_dm_i2c_xfer()
drm/amd/display: Add null checks for 8K60 lightup
drm/amd/pm: Fill pcie error counters for gpu v1_4
drm/amd/pm: Update metric table for smu v13_0_6
drm/amdgpu: correct chunk_ptr to a pointer to chunk.
drm/amd/display: Fix DSC not Enabled on Direct MST Sink
...
nfsd_cache_csum() currently assumes that the server's RPC layer has
been advancing rq_arg.head[0].iov_base as it decodes an incoming
request, because that's the way it used to work. On entry, it
expects that buf->head[0].iov_base points to the start of the NFS
header, and excludes the already-decoded RPC header.
These days however, head[0].iov_base now points to the start of the
RPC header during all processing. It no longer points at the NFS
Call header when execution arrives at nfsd_cache_csum().
In a retransmitted RPC the XID and the NFS header are supposed to
be the same as the original message, but the contents of the
retransmitted RPC header can be different. For example, for krb5,
the GSS sequence number will be different between the two. Thus if
the RPC header is always included in the DRC checksum computation,
the checksum of the retransmitted message might not match the
checksum of the original message, even though the NFS part of these
messages is identical.
The result is that, even if a matching XID is found in the DRC,
the checksum mismatch causes the server to execute the
retransmitted RPC transaction again.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Tested-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
