For SOC21 ASICs, there is an issue in re-enabling PM features if a
suspend got aborted. In such cases, reset the device during resume
phase. This is a workaround till a proper solution is finalized.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Add FW information of all the IP's in the devcoredump.
Signed-off-by: Sunil Khatri <sunil.khatri@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Otherwise the old one will be used during GPU reset.
That's not expected.
Signed-off-by: Lang Yu <Lang.Yu@amd.com>
Reviewed-by: Feifei Xu <Feifei.Xu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Downgrade to debug information when IBs are skipped. Also, use dev_* to
identify the device.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
v1:
add a new enumeration type to identify device attribute node,
this method is relatively more efficient compared with 'strcmp' in
update_attr() function.
v2:
rename device_attr_type to device_attr_id.
Signed-off-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Ma Jun <majun@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
With the changes in the last patches, napi_frag_unref() is now
reduandant. Remove it and use skb_page_unref directly.
Signed-off-by: Mina Almasry <almasrymina@google.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20240408153000.2152844-4-almasrymina@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The implementations of these 2 functions are almost identical. Remove
the implementation of napi_frag_unref, and make it a call into
skb_page_unref so we don't duplicate the implementation.
Signed-off-by: Mina Almasry <almasrymina@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20240408153000.2152844-2-almasrymina@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tony Nguyen says:
====================
net/e1000e, igb, igc: Remove redundant runtime resume
Bjorn Helgaas says:
e1000e, igb, and igc all have code to runtime resume the device during
ethtool operations.
Since f32a213765 ("ethtool: runtime-resume netdev parent before ethtool
ioctl ops"), dev_ethtool() does this for us, so remove it from the
individual drivers.
* '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue:
igc: Remove redundant runtime resume for ethtool ops
igb: Remove redundant runtime resume for ethtool_ops
e1000e: Remove redundant runtime resume for ethtool_ops
====================
Link: https://lore.kernel.org/r/20240408210849.3641172-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Eric Dumazet says:
====================
bonding: remove RTNL from three sysfs files
First patch might fix a potential deadlock.
sysfs handlers should use rtnl_trylock() instead of rtnl_lock().
Following files can be read without acquiring RTNL :
- /sys/class/net/bonding_masters
- /sys/class/net/<name>/bonding/slaves
- /sys/class/net/<name>/bonding/queue_id
====================
Link: https://lore.kernel.org/r/20240408190437.2214473-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Annotate lockless reads of slave->queue_id.
Annotate writes of slave->queue_id.
Switch bonding_show_queue_id() to rcu_read_lock()
and bond_for_each_slave_rcu().
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Link: https://lore.kernel.org/r/20240408190437.2214473-4-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Slave devices are already RCU protected, simply
switch to bond_for_each_slave_rcu(),
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Link: https://lore.kernel.org/r/20240408190437.2214473-3-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
netdev structures are already RCU protected.
Change bond_init() and bond_uninit() to use RCU
enabled list_add_tail_rcu() and list_del_rcu().
Then bonding_show_bonds() can use rcu_read_lock()
while iterating through bn->dev_list.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Link: https://lore.kernel.org/r/20240408190437.2214473-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When constructing a heap, heapify operations are required on all
non-leaf nodes. Thus, determining the index of the first non-leaf node
is crucial. In a heap, the left child's index of node i is 2 * i + 1
and the right child's index is 2 * i + 2. Node CAKE_MAX_TINS *
CAKE_QUEUES / 2 has its left and right children at indexes
CAKE_MAX_TINS * CAKE_QUEUES + 1 and CAKE_MAX_TINS * CAKE_QUEUES + 2,
respectively, which are beyond the heap's range, indicating it as a
leaf node. Conversely, node CAKE_MAX_TINS * CAKE_QUEUES / 2 - 1 has a
left child at index CAKE_MAX_TINS * CAKE_QUEUES - 1, confirming its
non-leaf status. The loop should start from it since it's not a leaf
node.
By starting the loop from CAKE_MAX_TINS * CAKE_QUEUES / 2 - 1, we
minimize function calls and branch condition evaluations. This
adjustment theoretically reduces two function calls (one for
cake_heapify() and another for cake_heap_get_backlog()) and five branch
evaluations (one for iterating all non-leaf nodes, one within
cake_heapify()'s while loop, and three more within the while loop
with if conditions).
Signed-off-by: Kuan-Wei Chiu <visitorckw@gmail.com>
Acked-by: Toke Høiland-Jørgensen <toke@toke.dk>
Link: https://lore.kernel.org/r/20240408174716.751069-1-visitorckw@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The J722S EVM has an on-board eMMC. Enable the SDHC interface for it.
There is no pinmuxing required because the interface has dedicated pins.
Signed-off-by: Michael Walle <mwalle@kernel.org>
Link: https://lore.kernel.org/r/20240403102302.3934932-1-mwalle@kernel.org
Signed-off-by: Nishanth Menon <nm@ti.com>
Device tree best practice is to disable any external interface in the
dtsi and just enable them if needed in the device tree. Thus, disable
the ethernet switch and its ports by default and just enable the ones
used by the EVMs in their device trees.
There is no functional change.
Signed-off-by: Michael Walle <mwalle@kernel.org>
Acked-by: Andrew Davis <afd@ti.com>
Link: https://lore.kernel.org/r/20240403101545.3932437-1-mwalle@kernel.org
Signed-off-by: Nishanth Menon <nm@ti.com>
The phyBOARD-Electra has two TCAN1044VDD CAN transceivers which
support CAN FD at 8 Mbps.
Increase the maximum bitrate to 8 Mbps.
Signed-off-by: Nathan Morrisson <nmorrisson@phytec.com>
Reviewed-by: Wadim Egorov <w.egorov@phytec.de>
Link: https://lore.kernel.org/r/20240402160825.1516036-3-nmorrisson@phytec.com
Signed-off-by: Nishanth Menon <nm@ti.com>
The phyBOARD-Lyra has one TCAN1044VDD CAN transceiver which supports
CAN FD at 8 Mbps.
Increase the maximum bitrate to 8 Mbps.
Signed-off-by: Nathan Morrisson <nmorrisson@phytec.com>
Reviewed-by: Wadim Egorov <w.egorov@phytec.de>
Link: https://lore.kernel.org/r/20240402160825.1516036-2-nmorrisson@phytec.com
Signed-off-by: Nishanth Menon <nm@ti.com>
Add a GPIO hog to release PCIe reset on the carrier board, this is
required to use M.2 or mPCIe cards.
Verdin AM62 does not have any PCIe interface, however the Verdin family
has PCIe and normally an M.2 or mPCIe slot is available in the carrier
board that can be used with cards that use only the USB interface toward
the host CPU, for example cellular network modem.
Signed-off-by: Francesco Dolcini <francesco.dolcini@toradex.com>
Link: https://lore.kernel.org/r/20240327182801.5997-3-francesco@dolcini.it
Signed-off-by: Nishanth Menon <nm@ti.com>
Generic GPIOs pinctrl nodes are not correct, gpio[1-4] are into the MCU
domain and should be into &mcu_gpio0, gpio[5-8] were missing and are added
in this commit.
Fixes: 7698622fbc ("arm64: dts: ti: Add verdin am62 mallow board")
Signed-off-by: Francesco Dolcini <francesco.dolcini@toradex.com>
Link: https://lore.kernel.org/r/20240327182801.5997-2-francesco@dolcini.it
Signed-off-by: Nishanth Menon <nm@ti.com>
As described in the binding document for the "current-speed" property:
"This should only be present in case a driver has no chance to know the
baud rate of the slave device."
This is not the case for the UART used in K3 devices, the current
baud-rate can be calculated from the registers. Having this property
has the effect of actually skipping the baud-rate setup in some drivers
as it assumes it will already be set to this rate, which may not always
be the case.
It seems this property's purpose was mistaken as selecting the desired
baud-rate, which it does not. It would have been wrong to select that
here anyway as DT is not the place for configuration, especially when
there are already more standard ways to set serial baud-rates.
Signed-off-by: Andrew Davis <afd@ti.com>
Link: https://lore.kernel.org/r/20240326185441.29656-6-afd@ti.com
Signed-off-by: Nishanth Menon <nm@ti.com>
As described in the binding document for the "current-speed" property:
"This should only be present in case a driver has no chance to know the
baud rate of the slave device."
This is not the case for the UART used in K3 devices, the current
baud-rate can be calculated from the registers. Having this property
has the effect of actually skipping the baud-rate setup in some drivers
as it assumes it will already be set to this rate, which may not always
be the case.
It seems this property's purpose was mistaken as selecting the desired
baud-rate, which it does not. It would have been wrong to select that
here anyway as DT is not the place for configuration, especially when
there are already more standard ways to set serial baud-rates.
Signed-off-by: Andrew Davis <afd@ti.com>
Link: https://lore.kernel.org/r/20240326185441.29656-5-afd@ti.com
Signed-off-by: Nishanth Menon <nm@ti.com>
As described in the binding document for the "current-speed" property:
"This should only be present in case a driver has no chance to know the
baud rate of the slave device."
This is not the case for the UART used in K3 devices, the current
baud-rate can be calculated from the registers. Having this property
has the effect of actually skipping the baud-rate setup in some drivers
as it assumes it will already be set to this rate, which may not always
be the case.
It seems this property's purpose was mistaken as selecting the desired
baud-rate, which it does not. It would have been wrong to select that
here anyway as DT is not the place for configuration, especially when
there are already more standard ways to set serial baud-rates.
Signed-off-by: Andrew Davis <afd@ti.com>
Link: https://lore.kernel.org/r/20240326185441.29656-4-afd@ti.com
Signed-off-by: Nishanth Menon <nm@ti.com>
As described in the binding document for the "current-speed" property:
"This should only be present in case a driver has no chance to know the
baud rate of the slave device."
This is not the case for the UART used in K3 devices, the current
baud-rate can be calculated from the registers. Having this property
has the effect of actually skipping the baud-rate setup in some drivers
as it assumes it will already be set to this rate, which may not always
be the case.
It seems this property's purpose was mistaken as selecting the desired
baud-rate, which it does not. It would have been wrong to select that
here anyway as DT is not the place for configuration, especially when
there are already more standard ways to set serial baud-rates.
Signed-off-by: Andrew Davis <afd@ti.com>
Link: https://lore.kernel.org/r/20240326185441.29656-3-afd@ti.com
Signed-off-by: Nishanth Menon <nm@ti.com>
As described in the binding document for the "current-speed" property:
"This should only be present in case a driver has no chance to know the
baud rate of the slave device."
This is not the case for the UART used in K3 devices, the current
baud-rate can be calculated from the registers. Having this property
has the effect of actually skipping the baud-rate setup in some drivers
as it assumes it will already be set to this rate, which may not always
be the case.
It seems this property's purpose was mistaken as selecting the desired
baud-rate, which it does not. It would have been wrong to select that
here anyway as DT is not the place for configuration, especially when
there are already more standard ways to set serial baud-rates.
Signed-off-by: Andrew Davis <afd@ti.com>
Link: https://lore.kernel.org/r/20240326185441.29656-2-afd@ti.com
Signed-off-by: Nishanth Menon <nm@ti.com>
As described in the binding document for the "current-speed" property:
"This should only be present in case a driver has no chance to know the
baud rate of the slave device."
This is not the case for the UART used in K3 devices, the current
baud-rate can be calculated from the registers. Having this property
has the effect of actually skipping the baud-rate setup in some drivers
as it assumes it will already be set to this rate, which may not always
be the case.
It seems this property's purpose was mistaken as selecting the desired
baud-rate, which it does not. It would have been wrong to select that
here anyway as DT is not the place for configuration, especially when
there are already more standard ways to set serial baud-rates.
Signed-off-by: Andrew Davis <afd@ti.com>
Link: https://lore.kernel.org/r/20240326185441.29656-1-afd@ti.com
Signed-off-by: Nishanth Menon <nm@ti.com>
On am62-lp-sk the PMIC is not wired up to a power button. Remove this
property. This fixes issues observed when entering a very deep sleep
state that is not yet available upstream.
Fixes: e6a51ffabf ("arm64: ti: dts: Add support for AM62x LP SK")
Signed-off-by: Markus Schneider-Pargmann <msp@baylibre.com>
Link: https://lore.kernel.org/r/20240325152029.2933445-1-msp@baylibre.com
Signed-off-by: Nishanth Menon <nm@ti.com>
BeaglePlay SBC[1] has Texas Instrument's WL18xx WiFi chipset[2].
Currently, WLAN_EN is configured as regulator and regulator-always-on.
However, the timing and wlan_en sequencing is not correctly modelled.
This causes the sdio access to fail during runtime-pm power operations
saving or during system suspend/resume/hibernation/freeze operations.
This is because the WLAN_EN line is not deasserted to low '0' to power
down the WiFi. So during restore, the WiFi driver tries to load the FW
without following correct power sequence. WLAN_EN => '1'/assert (high)
to power-up the chipset.
Use mmc-pwrseq-simple to drive TI's WiFi (WL18xx) chipset enable
'WLAN_EN'. mmc-pwrseq-simple provides power sequence flexibility with
support for post power-on and power-off delays.
Typical log signature that indicates this bug is:
wl1271_sdio mmc2:0001:2: sdio write failed (-110)
Followed by possibly a kernel warning (depending on firmware present):
WARNING: CPU: 1 PID: 45 at drivers/net/wireless/ti/wlcore/sdio.c:123 wl12xx_sdio_raw_write+0xe4/0x168 [wlcore_sdio]
[1] https://www.beagleboard.org/boards/beagleplay
[2] https://www.ti.com/lit/ds/symlink/wl1807mod.pdf
Fixes: f5a731f078 ("arm64: dts: ti: Add k3-am625-beagleplay")
Suggested-by: Shengyu Qu <wiagn233@outlook.com>
Signed-off-by: Sukrut Bellary <sukrut.bellary@linux.com>
Tested-by: Robert Nelson <robertcnelson@gmail.com>
Link: https://lore.kernel.org/r/20240325143511.2144768-1-nm@ti.com
Signed-off-by: Nishanth Menon <nm@ti.com>
TI SDHCI instance has a hardware debounce timer of 1 second as described
in commit 7ca0f166f5 ("mmc: sdhci_am654: Add workaround for card detect
debounce timer"), because of this the boot time increases of up to 1
second.
Workaround the issue the same way that is done on
arch/arm64/boot/dts/ti/k3-am625-beagleplay.dts, using the SD1 CD as
GPIO.
Suggested-by: Nishanth Menon <nm@ti.com>
Reported-by: João Paulo Silva Gonçalves <joao.goncalves@toradex.com>
Closes: https://lore.kernel.org/all/0e81af80de3d55e72f79af83fa5db87f5c9938f8.camel@toradex.com/
Signed-off-by: Francesco Dolcini <francesco.dolcini@toradex.com>
Link: https://lore.kernel.org/r/20240325083340.89568-1-francesco@dolcini.it
Signed-off-by: Nishanth Menon <nm@ti.com>
Replace netdev_{warn,err} with NL_SET_ERR_MSG_{FMT_,}MOD
to better inform the user about the problem.
Only compile-tested, no access to HW.
Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Link: https://lore.kernel.org/r/20240408165506.94483-1-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Eric Dumazet says:
====================
net: start to replace copy_from_sockptr()
We got several syzbot reports about unsafe copy_from_sockptr()
calls. After fixing some of them, it appears that we could
use a new helper to factorize all the checks in one place.
This series targets net tree, we can later start converting
many call sites in net-next.
====================
Link: https://lore.kernel.org/r/20240408082845.3957374-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
copy_from_sockptr() helper is unsafe, unless callers
did the prior check against user provided optlen.
Too many callers get this wrong, lets add a helper to
fix them and avoid future copy/paste bugs.
Instead of :
if (optlen < sizeof(opt)) {
err = -EINVAL;
break;
}
if (copy_from_sockptr(&opt, optval, sizeof(opt)) {
err = -EFAULT;
break;
}
Use :
err = copy_safe_from_sockptr(&opt, sizeof(opt),
optval, optlen);
if (err)
break;
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240408082845.3957374-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Unlike other ethernet PHYs from TI, PHY dp8382x has WOL enabled
at reset. The driver explicitly disables WOL in config_init callback
which is called during init and during resume from suspend. Hence,
WOL is unconditionally disabled during resume, even if it was enabled
before the suspend. We make sure that WOL configuration is persistent
across suspends.
Signed-off-by: Catalin Popescu <catalin.popescu@leica-geosystems.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240408082602.3654090-1-catalin.popescu@leica-geosystems.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In 9p2000 legacy mode, stat2inode initializes nlink to 1,
which is redundant with what alloc_inode should have already set.
9p2000.u overrides this with extensions if present in the stat
structure, and 9p2000.L incorporates nlink into its stat structure.
At the very least this probably messes with directory nlink
accounting in legacy mode.
Signed-off-by: Eric Van Hensbergen <ericvh@kernel.org>
Merge series from Richard Fitzgerald <rf@opensource.cirrus.com>:
This series adds some more test cases, mainly for testing:
commit eaa03486d9 ("regmap: maple: Fix uninitialized symbol 'ret' warnings")
commit 00bb549d7d ("regmap: maple: Fix cache corruption in regcache_maple_drop()")
And the pending patch ("regmap: Add regmap_read_bypassed()")
There are also a few small improvements to the KUnit implementation.
Merge series from Bastien Curutchet <bastien.curutchet@bootlin.com>:
This series aims to add some features to McBSP driver.
Convert bindings from .txt to .yaml.
Add possibility to use an external clock as sample rate generator's
input.
Add handling of new formats (TDM, S24_LE, BP_FC).
Enable the detection of unexpected frame pulses.
Set the clock free-running mode according to SND_SOC_DAIFMT_[GATED/CONT]
configuration in DAI format.
Add ti,T1-framing[tx/rx] properties in DT. They allow to set the data
delay to two bit-clock periods.
This has been tested on a platform designed off of the DAVINCI/OMAP-L138
connected to 3 daisy-chained AD7767. An external clock drives the
sample rate generator through the CLKS pin.
The hardware I have only allowed me to test acquisition side of McBSP.
It is connected to a 6 channels TDM and acts as Bit clock provider and
Frame clock consumer.
Merge series from Peter Ujfalusi <peter.ujfalusi@linux.intel.com>:
The beginning of the first SRAM window contains various fw registers and
additional information which can be very beneficial to read to gather
information on the current states to debug issues.
Merge series from Richard Fitzgerald <rf@opensource.cirrus.com>:
This chain fixes some problems with some previous patches for handling
the ASP1 config registers. The root of the problem is that the ownership
of these registers can be either with the firmware or the driver, and that
the chip has to be soft-reset after downloading the firmware.
This chain adds and uses a regmap_read_bypassed() function so that the
driver can leave the regmap in cache-only until the chip has rebooted,
but still poll a register to detect when the chip has rebooted.
Richard Fitzgerald (4):
regmap: Add regmap_read_bypassed()
ALSA: hda: cs35l56: Exit cache-only after
cs35l56_wait_for_firmware_boot()
ASoC: cs35l56: Fix unintended bus access while resetting amp
ASoC: cs35l56: Prevent overwriting firmware ASP config
drivers/base/regmap/regmap.c | 37 ++++++++++++++
include/linux/regmap.h | 8 +++
include/sound/cs35l56.h | 2 +
sound/pci/hda/cs35l56_hda.c | 4 ++
sound/soc/codecs/cs35l56-sdw.c | 2 -
sound/soc/codecs/cs35l56-shared.c | 83 ++++++++++++++++++++-----------
sound/soc/codecs/cs35l56.c | 26 +++++++++-
7 files changed, 130 insertions(+), 32 deletions(-)
--
2.39.2
Merge series from Peter Ujfalusi <peter.ujfalusi@linux.intel.com>:
The current code will reset the ChainDMA on release unconditionally which
can result the following error when the CHainDMA is not allocated:
ipc tx : 0xe040000|0x0: GLB_CHAIN_DMA
ipc tx reply: 0x2e000007|0x0: GLB_CHAIN_DMA
FW reported error: 7 - Unsupported operation requested
ipc error for msg 0xe040000|0x0
sof_pcm_stream_free: pcm_ops hw_free failed -22
Background:
Pulseaudio and Pipewire on startup opens all available streams and
closes them without triggering a start (after probing it's capabilities).
Merge series from Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>:
We somehow missed the default path for DSP libraries for LNL, and need
to restrict support for D0i3 w/ IPC4. Also add debugfs support for
firmware profile information so that sof-test scripts can show what is
being tested.
Fixes the following Coccinelle/coccicheck warning reported by
string_choices.cocci:
opportunity for str_plural(zgroup->g_nr_zones)
Signed-off-by: Thorsten Blum <thorsten.blum@toblux.com>
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
For easier use of the tests in automation and for having some
status information for the user while the test is running, let's
provide some TAP output in this test.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Muhammad Usama Anjum <usama.anjum@collabora.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Zhao Liu <zhao1.liu@intel.com>
Link: https://lore.kernel.org/r/20231019095900.450467-1-thuth@redhat.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
With multiple reader threads POLLing a single UFFD, the demand paging test
suffers from the thundering herd problem: performance degrades as the
number of reader threads is increased. Solve this issue [1] by switching
the the polling mechanism to EPOLL + EPOLLEXCLUSIVE.
Also, change the error-handling convention of uffd_handler_thread_fn.
Instead of just printing errors and returning early from the polling
loop, check for them via TEST_ASSERT(). "return NULL" is reserved for a
successful exit from uffd_handler_thread_fn, i.e. one triggered by a
write to the exit pipe.
Performance samples generated by the command in [2] are given below.
Num Reader Threads, Paging Rate (POLL), Paging Rate (EPOLL)
1 249k 185k
2 201k 235k
4 186k 155k
16 150k 217k
32 89k 198k
[1] Single-vCPU performance does suffer somewhat.
[2] ./demand_paging_test -u MINOR -s shmem -v 4 -o -r <num readers>
Signed-off-by: Anish Moorthy <amoorthy@google.com>
Acked-by: James Houghton <jthoughton@google.com>
Link: https://lore.kernel.org/r/20240215235405.368539-13-amoorthy@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
At the moment, demand_paging_test does not support profiling/testing
multiple vCPU threads concurrently faulting on a single uffd because
(a) "-u" (run test in userfaultfd mode) creates a uffd for each vCPU's
region, so that each uffd services a single vCPU thread.
(b) "-u -o" (userfaultfd mode + overlapped vCPU memory accesses)
simply doesn't work: the test tries to register the same memory
to multiple uffds, causing an error.
Add support for many vcpus per uffd by
(1) Keeping "-u" behavior unchanged.
(2) Making "-u -a" create a single uffd for all of guest memory.
(3) Making "-u -o" implicitly pass "-a", solving the problem in (b).
In cases (2) and (3) all vCPU threads fault on a single uffd.
With potentially multiple vCPUs per UFFD, it makes sense to allow
configuring the number of reader threads per UFFD as well: add the "-r"
flag to do so.
Signed-off-by: Anish Moorthy <amoorthy@google.com>
Acked-by: James Houghton <jthoughton@google.com>
Link: https://lore.kernel.org/r/20240215235405.368539-12-amoorthy@google.com
[sean: fix kernel style violations, use calloc() for arrays]
Signed-off-by: Sean Christopherson <seanjc@google.com>