Turns out that just writing CURPOS isn't sufficient to move the cursor
on some platforms. My 830 works just fine, but eg. 945 and PNV don't.
On those platforms we need to arm even the CURPOS update with a
CURBASE write.
Even worse, a write to any of the cursor register apart from CURBASE
will cancel an already pending cursor update. So if we have armed a
CURCNTR/CURBASE update, a subsequent CURPOS write prior to vblank
would cancel that armed update. Thus we're left with a cursor that
doesn't appear to move, or even change shape.
Fix the problem by always performing the CURBASE write after a
CURPOS write. Bspec is somewhat unclear which platforms actually
require this CURBASE write and which don't. So to keep it simple
and to make sure we really fix the problem across all supported
devices, let's just perform the CURBASE write unconditionally.
Cc: Paul Menzel <pmenzel@molgen.mpg.de>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101790
Fixes: 75343a44c9 ("drm/i915: Drop useless posting reads from cursor commit")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Tested-by: Paul Menzel <paulepanter@users.sourceforge.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20170714155227.6089-1-ville.syrjala@linux.intel.com
(cherry picked from commit 8753d2bc5e)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Commit b1f5bfc27a ("sctp: don't dereference ptr before leaving
_sctp_walk_{params, errors}()") tried to fix the issue that it
may overstep the chunk end for _sctp_walk_{params, errors} with
'chunk_end > offset(length) + sizeof(length)'.
But it introduced a side effect: When processing INIT, it verifies
the chunks with 'param.v == chunk_end' after iterating all params
by sctp_walk_params(). With the check 'chunk_end > offset(length)
+ sizeof(length)', it would return when the last param is not yet
accessed. Because the last param usually is fwdtsn supported param
whose size is 4 and 'chunk_end == offset(length) + sizeof(length)'
This is a badly issue even causing sctp couldn't process 4-shakes.
Client would always get abort when connecting to server, due to
the failure of INIT chunk verification on server.
The patch is to use 'chunk_end <= offset(length) + sizeof(length)'
instead of 'chunk_end < offset(length) + sizeof(length)' for both
_sctp_walk_params and _sctp_walk_errors.
Fixes: b1f5bfc27a ("sctp: don't dereference ptr before leaving _sctp_walk_{params, errors}()")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In dccp_feat_init, when ccid_get_builtin_ccids failsto alloc
memory for rx.val, it should free tx.val before returning an
error.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The patch "dccp: fix a memleak that dccp_ipv6 doesn't put reqsk
properly" fixed reqsk refcnt leak for dccp_ipv6. The same issue
exists on dccp_ipv4.
This patch is to fix it for dccp_ipv4.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In dccp_v6_conn_request, after reqsk gets alloced and hashed into
ehash table, reqsk's refcnt is set 3. one is for req->rsk_timer,
one is for hlist, and the other one is for current using.
The problem is when dccp_v6_conn_request returns and finishes using
reqsk, it doesn't put reqsk. This will cause reqsk refcnt leaks and
reqsk obj never gets freed.
Jianlin found this issue when running dccp_memleak.c in a loop, the
system memory would run out.
dccp_memleak.c:
int s1 = socket(PF_INET6, 6, IPPROTO_IP);
bind(s1, &sa1, 0x20);
listen(s1, 0x9);
int s2 = socket(PF_INET6, 6, IPPROTO_IP);
connect(s2, &sa1, 0x20);
close(s1);
close(s2);
This patch is to put the reqsk before dccp_v6_conn_request returns,
just as what tcp_conn_request does.
Reported-by: Jianlin Shi <jishi@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
"qd.id" comes directly from the copy_from_user() on the line before so
we should verify that it's within bounds.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
FCOE offloading failed with:
[qed_sp_fcoe_func_start:150(sp-0-3b:00.02)]Cannot satisfy CQ amount. CQs
requested 8, CQs available 6. Aborting function start
[qed_fcoe_start:821()]Failed to start fcoe
[__qedf_probe:3041]:6: Cannot start FCoE function.
The reason is a newly introduced check in the qed main part. This change
also provides the information about how many CQs are available, so we
simply limit the number of requested CQs..
Fixes: 3c5da94278 ("qed: Share additional information with qedf")
Signed-off-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Acked-by: Chad Dupuis <chad.dupuis@cavium.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The CPU hotplug related code of this driver can be simplified by:
1) Consolidating the callbacks into a single state. The CPU thread can be
torn down on the CPU which goes offline. There is no point in delaying
that to the CPU dead state
2) Let the core code invoke the online/offline callbacks and remove the
extra for_each_online_cpu() loops.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Chad Dupuis <chad.dupuis@cavium.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The CPU hotplug related code of this driver can be simplified by:
1) Consolidating the callbacks into a single state. The CPU thread can be
torn down on the CPU which goes offline. There is no point in delaying
that to the CPU dead state
2) Let the core code invoke the online/offline callbacks and remove the
extra for_each_online_cpu() loops.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The BNX2I module init/exit code installs/removes the hotplug callbacks with
the cpu hotplug lock held. This worked with the old CPU locking
implementation which allowed recursive locking, but with the new percpu
rwsem based mechanism this is not longer allowed.
Use the _cpuslocked() variants to fix this.
Reported-by: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Chad Dupuis <chad.dupuis@cavium.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The BNX2FC module init/exit code installs/removes the hotplug callbacks with
the cpu hotplug lock held. This worked with the old CPU locking
implementation which allowed recursive locking, but with the new percpu
rwsem based mechanism this is not longer allowed.
Use the _cpuslocked() variants to fix this.
Reported-by: kernel test robot <fengguang.wu@intel.com>
Acked-by: Chad Dupuis <chad.dupuis@cavium.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
bnx2fc_process_new_cqes() has protection against CPU hotplug, which relies
on the per cpu thread pointer. This protection is racy because it happens
only partially with the per cpu fp_work_lock held.
If the CPU is unplugged after the lock is dropped, the wakeup code can
dereference a NULL pointer or access freed and potentially reused memory.
Restructure the code so the thread check and wakeup happens with the
fp_work_lock held.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Chad Dupuis <chad.dupuis@cavium.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The rework of the exynos DRM clock handling introduced
warnings for configurations that have CONFIG_PM disabled:
drivers/gpu/drm/exynos/exynos_hdmi.c:736:13: error: 'hdmi_clk_disable_gates' defined but not used [-Werror=unused-function]
static void hdmi_clk_disable_gates(struct hdmi_context *hdata)
^~~~~~~~~~~~~~~~~~~~~~
drivers/gpu/drm/exynos/exynos_hdmi.c:717:12: error: 'hdmi_clk_enable_gates' defined but not used [-Werror=unused-function]
static int hdmi_clk_enable_gates(struct hdmi_context *hdata)
The problem is that the PM functions themselves are inside of
an #ifdef, but some functions they call are not.
This patch removes the #ifdef and instead marks the PM functions
as __maybe_unused, which is a more reliable way to get it right.
Link: https://patchwork.kernel.org/patch/8436281/
Fixes: 9be7e98984 ("drm/exynos/hdmi: clock code re-factoring")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
If the s5p-cec driver is a module and the drm exynos driver is built-in, then
the CEC core will be a module also, causing the CEC notifier to fail (will be
compiled as empty functions).
To prevent this select CEC_CORE if CEC_NOTIFIER is set to ensure the CEC core
is also built into the kernel.
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
The "Fixes" patch was incorrectly merged, as a result PHY is prematurely
powered off and for example Odroid-U3 cannot disable TV power domain
when HDMI cable is unplugged.
Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reported-by: Marek Szyprowski <m.szyprowski@samsung.com>
Fixes: 625e63e2 ("drm/exynos/hdmi: fix pipeline disable order")
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
This patch moves drm_bridge_add call into probe.
It doesn't need to call drm_bridge_add call every time
bind callback is called.
Changelog v2
- moved drm_bridge_remove call into remove callback.
- corrected description.
Suggested-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Hoegeun Kwon <hoegeun.kwon@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
Remove the error handling of bridge_node because the bridge_node is
optional.
For example, In case of Exynos SoC, a bridge device such as mDNIe and
MIC could be placed between Display Controller and MIPI DSI device but
the bridge device is optional.
Signed-off-by: Hoegeun Kwon <hoegeun.kwon@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
of_device_ids are not supposed to change at runtime. All functions
working with of_device_ids provided by <linux/of.h> work with const
of_device_ids. So mark the non-const structs as const.
File size before:
text data bss dec hex filename
12294 1192 0 13486 34ae drivers/gpu/drm/exynos/exynos_hdmi.o
File size after constify hdmi_match_types.
text data bss dec hex filename
13318 176 0 13494 34b6 drivers/gpu/drm/exynos/exynos_hdmi.o
Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
num_ioctls is already assigned when declaring the exynos_drm_driver
structure. No need to duplicate it here.
Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>
Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
The buffer passed to bpf_obj_get_info_by_fd() should be initialized
to zeros. Kernel will enforce that to guarantee we can safely extend
info structures in the future.
Making the bpf_obj_get_info_by_fd() call in libbpf perform the zeroing
is problematic, however, since some members of the info structures
may need to be initialized by the callers (for instance pointers
to buffers to which kernel is to dump translated and jited images).
Remove the zeroing and fix up the in-tree callers before any kernel
has been released with this code.
As Daniel points out this seems to be the intended operation anyway,
since commit 95b9afd398 ("bpf: Test for bpf ID") is itself setting
the buffer pointers before calling bpf_obj_get_info_by_fd().
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Apparently netpoll_setup() assumes that netpoll.dev_name is a pointer
when checking if the device name is set:
if (np->dev_name) {
...
However the field is a character array, therefore the condition always
yields true. Check instead whether the first byte of the array has a
non-zero value.
Signed-off-by: Matthias Kaehlcke <mka@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Three misc amd fixes.
* 'drm-fixes-4.13' of git://people.freedesktop.org/~agd5f/linux:
drm/amd/powerplay: fix AVFS voltage offset for Vega10
drm/amdgpu/gfx9: simplify and fix GRBM index selection
drm/amdgpu: Fix blocking in RCU critical section(v2)
for_each_compatible_node performs an of_node_get on each iteration, so a
return from the loop requires an of_node_put.
The semantic patch that fixes this problem is as follows
(http://coccinelle.lip6.fr):
// <smpl>
@@
local idexpression n;
expression e,e1,e2;
statement S;
iterator i1;
iterator name for_each_compatible_node;
@@
for_each_compatible_node(n,e1,e2) {
...
(
of_node_put(n);
|
e = n
|
return n;
|
i1(...,n,...) S
|
+ of_node_put(n);
? return ...;
)
...
}
// </smpl>
Additionally, call of_node_put on the previous value of np, obtained from
of_find_compatible_node, that is no longer accessible at the point of the
for_each_compatible_node.
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Commit bd8b244174 ("NFS: Store the raw NFS access mask in the inode's
access cache") changed how the access results are stored after an
access() call. An NFS v4 OPEN might have access bits returned with the
opendata, so we should use the NFS4_ACCESS values when determining the
return value in nfs4_opendata_access().
Fixes: bd8b244174 ("NFS: Store the raw NFS access mask in the inode's
access cache")
Reported-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Make iowait_boost and iowait_boost_max as unsigned int since its unit
is kHz and this is consistent with struct cpufreq_policy. Also change
the local variables in sugov_iowait_boost() to match this.
Signed-off-by: Joel Fernandes <joelaf@google.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Currently the iowait_boost feature in schedutil makes the frequency
go to max on iowait wakeups. This feature was added to handle a case
that Peter described where the throughput of operations involving
continuous I/O requests [1] is reduced due to running at a lower
frequency, however the lower throughput itself causes utilization to
be low and hence causing frequency to be low hence its "stuck".
Instead of going to max, its also possible to achieve the same effect
by ramping up to max if there are repeated in_iowait wakeups
happening. This patch is an attempt to do that. We start from a lower
frequency (policy->min) and double the boost for every consecutive
iowait update until we reach the maximum iowait boost frequency
(iowait_boost_max).
I ran a synthetic test (continuous O_DIRECT writes in a loop) on an
x86 machine with intel_pstate in passive mode using schedutil. In
this test the iowait_boost value ramped from 800MHz to 4GHz in 60ms.
The patch achieves the desired improved throughput as the existing
behavior.
[1] https://patchwork.kernel.org/patch/9735885/
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Suggested-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Joel Fernandes <joelaf@google.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Commit de77ecd4ef ("bonding: improve link-status update in mii-monitoring")
moves link status commitment into bond_mii_monitor(), but it still relies
on the return value of bond_miimon_inspect() as the hint. We need to return
non-zero as long as we propose a link status change.
Fixes: de77ecd4ef ("bonding: improve link-status update in mii-monitoring")
Reported-by: Benjamin Gilbert <benjamin.gilbert@coreos.com>
Tested-by: Benjamin Gilbert <benjamin.gilbert@coreos.com>
Cc: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The converter function for translating ns timings in register values was
initialized with a wrong function pointer. This resulted in wrong
register values also for the setup and pulse registers when configuring
the EBI interface trough dts.
Includes a small fix in a comment of the smc driver, which was probably
just a copy'n'paste mistake.
Signed-off-by: Alexander Dahl <ada@thorsis.com>
Acked-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Acked-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Setting optional EBI/SMC properties through device tree always fails due
to wrong evaluation of the return value of
atmel_ebi_xslate_smc_timings().
If you put some of those properties in your dts file, but not
'atmel,smc-tdf-ns' the local variable 'required' in
atmel_ebi_xslate_smc_timings() stays on 'false' after the first 'if'
block. This leads to setting 'ret' to -EINVAL in the first run of the
following 'for' loop which is then the return value of this function.
However if you set 'atmel,smc-tdf-ns' in the dts file and everything in
atmel_ebi_xslate_smc_timings() works well, it returns the content of
'required' which is 'true' then.
So the function atmel_ebi_xslate_smc_timings() always returns non-zero
which lets its call in atmel_ebi_xslate_smc_config() always fail and
thus returning -EINVAL, so the EBI configuration for this node fails.
Judging from the following code evaluating the local 'required' variable
in atmel_ebi_xslate_smc_config() and the call of caps->xlate_config in
atmel_ebi_dev_setup() it's probably right to only let the call fail if a
negative error code is returned.
Signed-off-by: Alexander Dahl <ada@thorsis.com>
Acked-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Signed-off-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Device lock bites again; if a device .remove() callback races a user
calling ioctl(VFIO_GROUP_GET_DEVICE_FD), the unbind request will hold
the device lock, but the user ioctl may have already taken a vfio_device
reference. In the case of a PCI device, the initial open will attempt
to reset the device, which again attempts to get the device lock,
resulting in deadlock. Use the trylock PCI reset interface and return
error on the open path if reset fails due to lock contention.
Link: https://lkml.org/lkml/2017/7/25/381
Reported-by: Wen Congyang <wencongyang2@huawei.com>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Currently dm_dax_flush() is not being called, even if underlying dax
device supports write cache, because DAXDEV_WRITE_CACHE is not being
propagated up to the DM dax device.
If the underlying dax device supports write cache, set
DAXDEV_WRITE_CACHE on the DM dax device. This will cause dm_dax_flush()
to be called.
Fixes: abebfbe2f7 ("dm: add ->flush() dax operation support")
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Acked-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
mempool_alloc() cannot fail for GFP_NOIO allocation, so there is no
point testing for failure.
One place the code tested for failure was passing "0" as the GFP
flags. This is most unusual and is probably meant to be GFP_NOIO,
so that is changed.
Also, allocation from ->extra_pool and ->prealloc_pool are repeated
before releasing the previous allocation. This can deadlock if the code
is servicing a write under high memory pressure. To avoid deadlocks,
change these to use GFP_NOWAIT and leave the error handling in place.
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Use GFP_NOIO for memory allocations in the I/O path. Other memory
allocations in the initialization path can use GFP_KERNEL.
Reported-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Reviewed-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
When CONFIG_EEH=y and CONFIG_VFIO_SPAPR_EEH=n, build fails with the
following:
drivers/vfio/pci/vfio_pci.o: In function `.vfio_pci_release':
vfio_pci.c:(.text+0xa98): undefined reference to `.vfio_spapr_pci_eeh_release'
drivers/vfio/pci/vfio_pci.o: In function `.vfio_pci_open':
vfio_pci.c:(.text+0x1420): undefined reference to `.vfio_spapr_pci_eeh_open'
In this case, vfio_pci.c should use the empty definitions of
vfio_spapr_pci_eeh_open and vfio_spapr_pci_eeh_release functions.
This patch fixes it by guarding these function definitions with
CONFIG_VFIO_SPAPR_EEH, the symbol that controls whether vfio_spapr_eeh.c is
built, which is where the non-empty versions of these functions are. We need to
make use of IS_ENABLED() macro because CONFIG_VFIO_SPAPR_EEH is a tristate
option.
This issue was found during a randconfig build. Logs are here:
http://kisskb.ellerman.id.au/kisskb/buildresult/12982362/
Signed-off-by: Murilo Opsfelder Araujo <mopsfelder@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
All systems use the same P-state selection "powersave" algorithm
in the active mode if HWP is not used, so there's no need to provide
a pointer for it in struct pstate_funcs any more.
Drop ->update_util from struct pstate_funcs and make
intel_pstate_set_update_util_hook() use intel_pstate_update_util()
directly.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
All systems with a defined ACPI preferred profile that are not
"servers" have been using the load-based P-state selection algorithm
in intel_pstate since 4.12-rc1 (mobile systems and laptops have been
using it since 4.10-rc1) and no problems with it have been reported
to date. In particular, no regressions with respect to the PID-based
P-state selection have been reported. Also testing indicates that
the P-state selection algorithm based on CPU load is generally on par
with the PID-based algorithm performance-wise, and for some workloads
it turns out to be better than the other one, while being more
straightforward and easier to understand at the same time.
Moreover, the PID-based P-state selection algorithm in intel_pstate
is known to be unstable in some situation and generally problematic,
the issues with it are hard to address and it has become a
significant maintenance burden.
For these reasons, make intel_pstate use the "powersave" P-state
selection algorithm based on CPU load in the active mode on all
systems and drop the PID-based P-state selection code along with
all things related to it from the driver. Also update the
documentation accordingly.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Pull virtio fixes and cleanups from Michael Tsirkin:
"Fixes some minor issues all over the codebase"
* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost:
virtio-net: fix module unloading
virtio-balloon: coding format cleanup
virtio-balloon: deflate via a page list
virtio_blk: Use sysfs_match_string() helper
Using variables instead of hard paths makes the requirements information
more accurate.
Signed-off-by: Lin Ma <lma@suse.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
KVM: s390: fixup missing srcu lock
We need to hold the srcu lock when accessing memory slots
during migration
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Run kvm-unit-tests/eventinj.flat in L1 w/ ept=0 on both L0 and L1:
Before NMI IRET test
Sending NMI to self
NMI isr running stack 0x461000
Sending nested NMI to self
After nested NMI to self
Nested NMI isr running rip=40038e
After iret
After NMI to self
FAIL: NMI
Commit 4c4a6f790e (KVM: nVMX: track NMI blocking state separately
for each VMCS) tracks NMI blocking state separately for vmcs01 and
vmcs02. However it is not enough:
- The L2 (kvm-unit-tests/eventinj.flat) generates NMI that will fault
on IRET, so the L2 can generate #PF which can be intercepted by L0.
- L0 walks L1's guest page table and sees the mapping is invalid, it
resumes the L1 guest and injects the #PF into L1. At this point the
vmcs02 has nmi_known_unmasked=true.
- L1 sets set bit 3 (blocking by NMI) in the interruptibility-state field
of vmcs12 (and fixes the shadow page table) before resuming L2 guest.
- L1 executes VMRESUME to resume L2, causing a vmexit to L0
- during VMRESUME emulation, prepare_vmcs02 sets bit 3 in the
interruptibility-state field of vmcs02, but nmi_known_unmasked is
still true.
- L2 immediately exits to L0 with another page fault, because L0 still has
not updated the NGVA->HPA page tables. However, nmi_known_unmasked is
true so vmx_recover_nmi_blocking does not do anything.
The fix is to update nmi_known_unmasked when preparing vmcs02 from vmcs12.
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The PI vector for L0 and L1 must be different. If dest vcpu0
is in guest mode while vcpu1 is delivering a non-nested PI to
vcpu0, there wont't be any vmexit so that the non-nested interrupt
will be delayed.
Signed-off-by: Wincy Van <fanwenyi0529@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We are using the same vector for nested/non-nested posted
interrupts delivery, this may cause interrupts latency in
L1 since we can't kick the L2 vcpu out of vmx-nonroot mode.
This patch introduces a new vector which is only for nested
posted interrupts to solve the problems above.
Signed-off-by: Wincy Van <fanwenyi0529@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
