Tvrtko Ursulin [Tue, 2 Mar 2021 11:42:13 +0000 (11:42 +0000)]
drm/i915: Wedge the GPU if command parser setup fails
Commit 311a50e76a33 ("drm/i915: Add support for mandatory cmdparsing")
introduced mandatory command parsing but setup failures were not
translated into wedging the GPU which was probably the intent.
Possible errors come in two categories. Either the sanity check on
internal tables has failed, which should be caught in CI unless an
affected platform would be missed in testing; or memory allocation failure
happened during driver load, which should be extremely unlikely but for
correctness should still be handled.
Dave Airlie [Fri, 12 Mar 2021 00:19:55 +0000 (10:19 +1000)]
Merge tag 'drm-misc-fixes-2021-03-11' of git://anongit.freedesktop.org/drm/drm-misc into drm-fixes
drm-misc-fixes for rc3, rebased on rc2:
- Fix oops in drm_fbdev_cleanup()
- unpin qxl bos created as pinned when freeing them,
and make ttm only warn once on this behavior.
- Use LCD management for atyfb on PPC_MAC.
- Use gitlab for drm bugzilla now.
- Fix ttm page pool accounting.
- Zero head.surface_id correctly in qxl.
- Assorted fixes for shmem helpers.
- Shutdown kms poll helper in meson correctly.
- Clear holes when converting compat ioctl's between 32-bits and 64-bits.
block: Discard page cache of zone reset target range
When zone reset ioctl and data read race for a same zone on zoned block
devices, the data read leaves stale page cache even though the zone
reset ioctl zero clears all the zone data on the device. To avoid
non-zero data read from the stale page cache after zone reset, discard
page cache of reset target zones in blkdev_zone_mgmt_ioctl(). Introduce
the helper function blkdev_truncate_zone_range() to discard the page
cache. Ensure the page cache discarded by calling the helper function
before and after zone reset in same manner as fallocate does.
This patch can be applied back to the stable kernel version v5.10.y.
Rework is needed for older stable kernels.
Daniel Wagner [Thu, 11 Mar 2021 15:19:17 +0000 (16:19 +0100)]
block: Suppress uevent for hidden device when removed
register_disk() suppress uevents for devices with the GENHD_FL_HIDDEN
but enables uevents at the end again in order to announce disk after
possible partitions are created.
When the device is removed the uevents are still on and user land sees
'remove' messages for devices which were never 'add'ed to the system.
J. Bruce Fields [Thu, 28 Jan 2021 22:36:38 +0000 (17:36 -0500)]
nfs: we don't support removing system.nfs4_acl
The NFSv4 protocol doesn't have any notion of reomoving an attribute, so
removexattr(path,"system.nfs4_acl") doesn't make sense.
There's no documented return value. Arguably it could be EOPNOTSUPP but
I'm a little worried an application might take that to mean that we
don't support ACLs or xattrs. How about EINVAL?
Jens Axboe [Thu, 11 Mar 2021 17:49:20 +0000 (10:49 -0700)]
io_uring: perform IOPOLL reaping if canceler is thread itself
We bypass IOPOLL completion polling (and reaping) for the SQPOLL thread,
but if it's the thread itself invoking cancelations, then we still need
to perform it or no one will.
Jens Axboe [Thu, 11 Mar 2021 17:17:56 +0000 (10:17 -0700)]
io_uring: force creation of separate context for ATTACH_WQ and non-threads
Earlier kernels had SQPOLL threads that could share across anything, as
we grabbed the context we needed on a per-ring basis. This is no longer
the case, so only allow attaching directly if we're in the same thread
group. That is the common use case. For non-group tasks, just setup a
new context and thread as we would've done if sharing wasn't set. This
isn't 100% ideal in terms of CPU utilization for the forked and share
case, but hopefully that isn't much of a concern. If it is, there are
plans in motion for how to improve that. Most importantly, we want to
avoid app side regressions where sharing worked before and now doesn't.
With this patch, functionality is equivalent to previous kernels that
supported IORING_SETUP_ATTACH_WQ with SQPOLL.
Ever since the addition of multipage bio_vecs BIO_MAX_PAGES has been
horribly confusingly misnamed. Rename it to BIO_MAX_VECS to stop
confusing users of the bio API.
These routines lost all existing users during the latest merge window so
we can remove them. This avoids the need to fix them in the context of
fixing a regression related to the ID map on 52-bit VA kernels.
Ard Biesheuvel [Wed, 10 Mar 2021 17:15:11 +0000 (18:15 +0100)]
arm64: mm: use a 48-bit ID map when possible on 52-bit VA builds
52-bit VA kernels can run on hardware that is only 48-bit capable, but
configure the ID map as 52-bit by default. This was not a problem until
recently, because the special T0SZ value for a 52-bit VA space was never
programmed into the TCR register anwyay, and because a 52-bit ID map
happens to use the same number of translation levels as a 48-bit one.
This behavior was changed by commit 1401bef703a4 ("arm64: mm: Always update
TCR_EL1 from __cpu_set_tcr_t0sz()"), which causes the unsupported T0SZ
value for a 52-bit VA to be programmed into TCR_EL1. While some hardware
simply ignores this, Mark reports that Amberwing systems choke on this,
resulting in a broken boot. But even before that commit, the unsupported
idmap_t0sz value was exposed to KVM and used to program TCR_EL2 incorrectly
as well.
Given that we already have to deal with address spaces being either 48-bit
or 52-bit in size, the cleanest approach seems to be to simply default to
a 48-bit VA ID map, and only switch to a 52-bit one if the placement of the
kernel in DRAM requires it. This is guaranteed not to happen unless the
system is actually 52-bit VA capable.
Mathias Nyman [Thu, 11 Mar 2021 11:53:53 +0000 (13:53 +0200)]
xhci: Fix repeated xhci wake after suspend due to uncleared internal wake state
If port terminations are detected in suspend, but link never reaches U0
then xHCI may have an internal uncleared wake state that will cause an
immediate wake after suspend.
This wake state is normally cleared when driver clears the PORT_CSC bit,
which is set after a device is enabled and in U0.
Write 1 to clear PORT_CSC for ports that don't have anything connected
when suspending. This makes sure any pending internal wake states in
xHCI are cleared.
Forest Crossman [Thu, 11 Mar 2021 11:53:52 +0000 (13:53 +0200)]
usb: xhci: Fix ASMedia ASM1042A and ASM3242 DMA addressing
I've confirmed that both the ASMedia ASM1042A and ASM3242 have the same
problem as the ASM1142 and ASM2142/ASM3142, where they lose some of the
upper bits of 64-bit DMA addresses. As with the other chips, this can
cause problems on systems where the upper bits matter, and adding the
XHCI_NO_64BIT_SUPPORT quirk completely fixes the issue.
Mathias Nyman [Thu, 11 Mar 2021 11:53:51 +0000 (13:53 +0200)]
xhci: Improve detection of device initiated wake signal.
A xHC USB 3 port might miss the first wake signal from a USB 3 device
if the port LFPS reveiver isn't enabled fast enough after xHC resume.
xHC host will anyway be resumed by a PME# signal, but will go back to
suspend if no port activity is seen.
The device resends the U3 LFPS wake signal after a 100ms delay, but
by then host is already suspended, starting all over from the
beginning of this issue.
USB 3 specs say U3 wake LFPS signal is sent for max 10ms, then device
needs to delay 100ms before resending the wake.
Don't suspend immediately if port activity isn't detected in resume.
Instead add a retry. If there is no port activity then delay for 120ms,
and re-check for port activity.
usb: xhci: do not perform Soft Retry for some xHCI hosts
On some systems rt2800usb and mt7601u devices are unable to operate since
commit f8f80be501aa ("xhci: Use soft retry to recover faster from
transaction errors")
Seems that some xHCI controllers can not perform Soft Retry correctly,
affecting those devices.
To avoid the problem add xhci->quirks flag that restore pre soft retry
xhci behaviour for affected xHCI controllers. Currently those are
AMD_PROMONTORYA_4 and AMD_PROMONTORYA_2, since it was confirmed
by the users: on those xHCI hosts issue happen and is gone after
disabling Soft Retry.
[minor commit message rewording for checkpatch -Mathias]
Daiyue Zhang [Mon, 1 Mar 2021 06:10:53 +0000 (14:10 +0800)]
configfs: fix a use-after-free in __configfs_open_file
Commit b0841eefd969 ("configfs: provide exclusion between IO and removals")
uses ->frag_dead to mark the fragment state, thus no bothering with extra
refcount on config_item when opening a file. The configfs_get_config_item
was removed in __configfs_open_file, but not with config_item_put. So the
refcount on config_item will lost its balance, causing use-after-free
issues in some occasions like this:
2. Then run:
for file in /config
do
echo $file
grep -R 'key' $file
done
3. __configfs_open_file will be called in parallel, the first one
got called will do:
if (file->f_mode & FMODE_READ) {
if (!(inode->i_mode & S_IRUGO))
goto out_put_module;
config_item_put(buffer->item);
kref_put()
package_details_release()
kfree()
the other one will run into use-after-free issues like this:
BUG: KASAN: use-after-free in __configfs_open_file+0x1bc/0x3b0
Read of size 8 at addr fffffff155f02480 by task grep/13096
CPU: 0 PID: 13096 Comm: grep VIP: 00 Tainted: G W 4.14.116-kasan #1
TGID: 13096 Comm: grep
Call trace:
dump_stack+0x118/0x160
kasan_report+0x22c/0x294
__asan_load8+0x80/0x88
__configfs_open_file+0x1bc/0x3b0
configfs_open_file+0x28/0x34
do_dentry_open+0x2cc/0x5c0
vfs_open+0x80/0xe0
path_openat+0xd8c/0x2988
do_filp_open+0x1c4/0x2fc
do_sys_open+0x23c/0x404
SyS_openat+0x38/0x48
Dmitry Monakhov [Wed, 10 Mar 2021 12:06:41 +0000 (12:06 +0000)]
nvme-pci: add the DISABLE_WRITE_ZEROES quirk for a Samsung PM1725a
This adds a quirk for Samsung PM1725a drive which fixes timeouts and
I/O errors due to the fact that the controller does not properly
handle the Write Zeroes command, dmesg log:
nvme nvme0: I/O 528 QID 10 timeout, aborting
nvme nvme0: I/O 529 QID 10 timeout, aborting
nvme nvme0: I/O 530 QID 10 timeout, aborting
nvme nvme0: I/O 531 QID 10 timeout, aborting
nvme nvme0: I/O 532 QID 10 timeout, aborting
nvme nvme0: I/O 533 QID 10 timeout, aborting
nvme nvme0: I/O 534 QID 10 timeout, aborting
nvme nvme0: I/O 535 QID 10 timeout, aborting
nvme nvme0: Abort status: 0x0
nvme nvme0: Abort status: 0x0
nvme nvme0: Abort status: 0x0
nvme nvme0: Abort status: 0x0
nvme nvme0: Abort status: 0x0
nvme nvme0: Abort status: 0x0
nvme nvme0: Abort status: 0x0
nvme nvme0: Abort status: 0x0
nvme nvme0: I/O 528 QID 10 timeout, reset controller
nvme nvme0: controller is down; will reset: CSTS=0x3, PCI_STATUS=0x10
nvme nvme0: Device not ready; aborting reset, CSTS=0x3
nvme nvme0: Device not ready; aborting reset, CSTS=0x3
nvme nvme0: Removing after probe failure status: -19
nvme0n1: detected capacity change from 6251233968 to 0
blk_update_request: I/O error, dev nvme0n1, sector 32776 op 0x1:(WRITE) flags 0x3000 phys_seg 6 prio class 0
blk_update_request: I/O error, dev nvme0n1, sector 113319936 op 0x9:(WRITE_ZEROES) flags 0x800 phys_seg 0 prio class 0
Buffer I/O error on dev nvme0n1p2, logical block 1, lost async page write
blk_update_request: I/O error, dev nvme0n1, sector 113319680 op 0x9:(WRITE_ZEROES) flags 0x0 phys_seg 0 prio class 0
Buffer I/O error on dev nvme0n1p2, logical block 2, lost async page write
blk_update_request: I/O error, dev nvme0n1, sector 113319424 op 0x9:(WRITE_ZEROES) flags 0x0 phys_seg 0 prio class 0
Buffer I/O error on dev nvme0n1p2, logical block 3, lost async page write
blk_update_request: I/O error, dev nvme0n1, sector 113319168 op 0x9:(WRITE_ZEROES) flags 0x0 phys_seg 0 prio class 0
Buffer I/O error on dev nvme0n1p2, logical block 4, lost async page write
blk_update_request: I/O error, dev nvme0n1, sector 113318912 op 0x9:(WRITE_ZEROES) flags 0x0 phys_seg 0 prio class 0
Buffer I/O error on dev nvme0n1p2, logical block 5, lost async page write
blk_update_request: I/O error, dev nvme0n1, sector 113318656 op 0x9:(WRITE_ZEROES) flags 0x0 phys_seg 0 prio class 0
Buffer I/O error on dev nvme0n1p2, logical block 6, lost async page write
blk_update_request: I/O error, dev nvme0n1, sector 113318400 op 0x9:(WRITE_ZEROES) flags 0x0 phys_seg 0 prio class 0
blk_update_request: I/O error, dev nvme0n1, sector 113318144 op 0x9:(WRITE_ZEROES) flags 0x0 phys_seg 0 prio class 0
blk_update_request: I/O error, dev nvme0n1, sector 113317888 op 0x9:(WRITE_ZEROES) flags 0x0 phys_seg 0 prio class 0
Lv Yunlong [Thu, 11 Mar 2021 05:44:13 +0000 (21:44 -0800)]
nvme-rdma: Fix a use after free in nvmet_rdma_write_data_done
In nvmet_rdma_write_data_done, rsp is recoverd by wc->wr_cqe and freed by
nvmet_rdma_release_rsp(). But after that, pr_info() used the freed
chunk's member object and could leak the freed chunk address with
wc->wr_cqe by computing the offset.
James Smart [Tue, 9 Mar 2021 00:51:26 +0000 (16:51 -0800)]
nvme-fc: fix racing controller reset and create association
Recent patch to prevent calling __nvme_fc_abort_outstanding_ios in
interrupt context results in a possible race condition. A controller
reset results in errored io completions, which schedules error
work. The change of error work to a work element allows it to fire
after the ctrl state transition to NVME_CTRL_CONNECTING, causing
any outstanding io (used to initialize the controller) to fail and
cause problems for connect_work.
Add a state check to only schedule error work if not in the RESETTING
state.
Fixes: 19fce0470f05 ("nvme-fc: avoid calling _nvme_fc_abort_outstanding_ios from interrupt context") Signed-off-by: Nigel Kirkland <[email protected]> Signed-off-by: James Smart <[email protected]> Signed-off-by: Christoph Hellwig <[email protected]>
Hannes Reinecke [Fri, 26 Feb 2021 07:17:27 +0000 (08:17 +0100)]
nvme-fc: set NVME_REQ_CANCELLED in nvme_fc_terminate_exchange()
nvme_fc_terminate_exchange() is being called when exchanges are
being deleted, and as such we should be setting the NVME_REQ_CANCELLED
flag to have identical behaviour on all transports.
Hannes Reinecke [Fri, 26 Feb 2021 07:17:26 +0000 (08:17 +0100)]
nvme: add NVME_REQ_CANCELLED flag in nvme_cancel_request()
NVME_REQ_CANCELLED is translated into -EINTR in nvme_submit_sync_cmd(),
so we should be setting this flags during nvme_cancel_request() to
ensure that the callers to nvme_submit_sync_cmd() will get the correct
error code when the controller is reset.
Hannes Reinecke [Fri, 26 Feb 2021 07:17:25 +0000 (08:17 +0100)]
nvme: simplify error logic in nvme_validate_ns()
We only should remove namespaces when we get fatal error back from
the device or when the namespace IDs have changed.
So instead of painfully masking out error numbers which might indicate
that the error should be ignored we could use an NVME status code
to indicated when the namespace should be removed.
That simplifies the final logic and makes it less error-prone.
The histogram mode is set using 'rkisp1_params_set_bits'.
Only the bits of the mode should be the value argument for
that function. Otherwise bits outside the mode mask are
turned on which is not what was intended.
usbtv doesn't support power management, so on system suspend the
.disconnect callback of the driver is called. The teardown sequence
includes a call to snd_card_free. Its implementation waits until the
refcount of the sound card device drops to zero, however, if its file is
open, snd_card_file_add takes a reference, which can't be dropped during
the suspend, because the userspace processes are already frozen at this
point. snd_card_free waits for completion forever, leading to a hang on
suspend.
This commit fixes this deadlock condition by replacing snd_card_free
with snd_card_free_when_closed, that doesn't wait until all references
are released, allowing suspend to progress.
Hans Verkuil [Fri, 26 Feb 2021 10:37:47 +0000 (11:37 +0100)]
media: rc: compile rc-cec.c into rc-core
The rc-cec keymap is unusual in that it can't be built as a module,
instead it is registered directly in rc-main.c if CONFIG_MEDIA_CEC_RC
is set. This is because it can be called from drm_dp_cec_set_edid() via
cec_register_adapter() in an asynchronous context, and it is not
allowed to use request_module() to load rc-cec.ko in that case. Trying to
do so results in a 'WARN_ON_ONCE(wait && current_is_async())'.
Since this keymap is only used if CONFIG_MEDIA_CEC_RC is set, we
just compile this keymap into the rc-core module and never as a
separate module.
Signed-off-by: Hans Verkuil <[email protected]> Fixes: 2c6d1fffa1d9 (drm: add support for DisplayPort CEC-Tunneling-over-AUX) Reported-by: Hans de Goede <[email protected]> Signed-off-by: Sean Young <[email protected]> Signed-off-by: Mauro Carvalho Chehab <[email protected]>
Daniel Vetter [Mon, 22 Feb 2021 10:06:43 +0000 (11:06 +0100)]
drm/compat: Clear bounce structures
Some of them have gaps, or fields we don't clear. Native ioctl code
does full copies plus zero-extends on size mismatch, so nothing can
leak. But compat is more hand-rolled so need to be careful.
None of these matter for performance, so just memset.
Also I didn't fix up the CONFIG_DRM_LEGACY or CONFIG_DRM_AGP ioctl, those
are security holes anyway.
Noralf Trønnes [Fri, 19 Feb 2021 12:22:03 +0000 (13:22 +0100)]
drm/shmem-helpers: vunmap: Don't put pages for dma-buf
dma-buf importing was reworked in commit 7d2cd72a9aa3
("drm/shmem-helpers: Simplify dma-buf importing"). Before that commit
drm_gem_shmem_prime_import_sg_table() did set ->pages_use_count=1 and
drm_gem_shmem_vunmap_locked() could call drm_gem_shmem_put_pages()
unconditionally. Now without the use count set, put pages is called also
on dma-bufs. Fix this by only putting pages if it's not imported.
Neil Roberts [Tue, 23 Feb 2021 15:51:25 +0000 (16:51 +0100)]
drm/shmem-helper: Don't remove the offset in vm_area_struct pgoff
When mmapping the shmem, it would previously adjust the pgoff in the
vm_area_struct to remove the fake offset that is added to be able to
identify the buffer. This patch removes the adjustment and makes the
fault handler use the vm_fault address to calculate the page offset
instead. Although using this address is apparently discouraged, several
DRM drivers seem to be doing it anyway.
The problem with removing the pgoff is that it prevents
drm_vma_node_unmap from working because that searches the mapping tree
by address. That doesn't work because all of the mappings are at offset
0. drm_vma_node_unmap is being used by the shmem helpers when purging
the buffer.
This fixes a bug in Panfrost which is using drm_gem_shmem_purge. Without
this the mapping for the purged buffer can still be accessed which might
mean it would access random pages from other buffers
v2: Don't check whether the unsigned page_offset is less than 0.
Neil Roberts [Tue, 23 Feb 2021 15:51:24 +0000 (16:51 +0100)]
drm/shmem-helper: Check for purged buffers in fault handler
When a buffer is madvised as not needed and then purged, any attempts to
access the buffer from user-space should cause a bus fault. This patch
adds a check for that.
Colin Ian King [Thu, 4 Mar 2021 09:49:28 +0000 (09:49 +0000)]
qxl: Fix uninitialised struct field head.surface_id
The surface_id struct field in head is not being initialized and
static analysis warns that this is being passed through to
dev->monitors_config->heads[i] on an assignment. Clear up this
warning by initializing it to zero.
Anthony DeRossi [Wed, 3 Mar 2021 01:17:25 +0000 (17:17 -0800)]
drm/ttm: Fix TTM page pool accounting
Freed pages are not subtracted from the allocated_pages counter in
ttm_pool_type_fini(), causing a leak in the count on device removal.
The next shrinker invocation loops forever trying to free pages that are
no longer in the pool:
drm: Use USB controller's DMA mask when importing dmabufs
USB devices cannot perform DMA and hence have no dma_mask set in their
device structure. Therefore importing dmabuf into a USB-based driver
fails, which breaks joining and mirroring of display in X11.
For USB devices, pick the associated USB controller as attachment device.
This allows the DRM import helpers to perform the DMA setup. If the DMA
controller does not support DMA transfers, we're out of luck and cannot
import. Our current USB-based DRM drivers don't use DMA, so the actual
DMA device is not important.
Tested by joining/mirroring displays of udl and radeon under Gnome/X11.
v8:
* release dmadev if device initialization fails (Noralf)
* fix commit description (Noralf)
v7:
* fix use-before-init bug in gm12u320 (Dan)
v6:
* implement workaround in DRM drivers and hold reference to
DMA device while USB device is in use
* remove dev_is_usb() (Greg)
* collapse USB helper into usb_intf_get_dma_device() (Alan)
* integrate Daniel's TODO statement (Daniel)
* fix typos (Greg)
v5:
* provide a helper for USB interfaces (Alan)
* add FIXME item to documentation and TODO list (Daniel)
v4:
* implement workaround with USB helper functions (Greg)
* use struct usb_device->bus->sysdev as DMA device (Takashi)
v3:
* drop gem_create_object
* use DMA mask of USB controller, if any (Daniel, Christian, Noralf)
v2:
* move fix to importer side (Christian, Daniel)
* update SHMEM and CMA helpers for new PRIME callbacks
Randy Dunlap [Wed, 24 Feb 2021 21:55:28 +0000 (13:55 -0800)]
fbdev: atyfb: always declare aty_{ld,st}_lcd()
The previously added stubs for aty_{ld,}st_lcd() make it
so that these functions are used regardless of the config
options that were guarding them, so remove the #ifdef/#endif
lines and make their declarations always visible.
This fixes build warnings that were reported by clang:
drivers/video/fbdev/aty/atyfb_base.c:180:6: warning: no previous prototype for function 'aty_st_lcd' [-Wmissing-prototypes]
void aty_st_lcd(int index, u32 val, const struct atyfb_par *par)
^
drivers/video/fbdev/aty/atyfb_base.c:180:1: note: declare 'static' if the function is not intended to be used outside of this translation unit
void aty_st_lcd(int index, u32 val, const struct atyfb_par *par)
drivers/video/fbdev/aty/atyfb_base.c:183:5: warning: no previous prototype for function 'aty_ld_lcd' [-Wmissing-prototypes]
u32 aty_ld_lcd(int index, const struct atyfb_par *par)
^
drivers/video/fbdev/aty/atyfb_base.c:183:1: note: declare 'static' if the function is not intended to be used outside of this translation unit
u32 aty_ld_lcd(int index, const struct atyfb_par *par)
They should not be marked as static since they are used in
mach64_ct.c.
Tong Zhang [Sun, 28 Feb 2021 04:46:25 +0000 (23:46 -0500)]
drm/fb-helper: only unmap if buffer not null
drm_fbdev_cleanup() can be called when fb_helper->buffer is null, hence
fb_helper->buffer should be checked before calling
drm_client_buffer_vunmap(). This buffer is also checked in
drm_client_framebuffer_delete(), so we should also do the same thing for
drm_client_buffer_vunmap().
Merge tag 'usb-serial-5.12-rc3' of https://git.kernel.org/pub/scm/linux/kernel/git/johan/usb-serial into usb-linus
Johan writes:
USB-serial fixes for 5.12-rc3
Here's a fix for a long-standing memory leak after probe failure in
io_edgeport and a fix for a NULL-deref on disconnect in the new xr
driver.
Included are also some new device ids.
All have been in linux-next with no reported issues.
* tag 'usb-serial-5.12-rc3' of https://git.kernel.org/pub/scm/linux/kernel/git/johan/usb-serial:
USB: serial: io_edgeport: fix memory leak in edge_startup
USB: serial: ch341: add new Product ID
USB: serial: xr: fix NULL-deref on disconnect
USB: serial: cp210x: add some more GE USB IDs
USB: serial: cp210x: add ID for Acuity Brands nLight Air Adapter
Masahiro Yamada [Wed, 10 Mar 2021 13:54:22 +0000 (22:54 +0900)]
kbuild: remove LLVM=1 test from HAS_LTO_CLANG
As Documentation/kbuild/llvm.rst notes, LLVM=1 switches the default of
tools, but you can still override CC, LD, etc. individually. This LLVM=1
check is unneeded because each tool is already checked separately.
"make CC=clang LD=ld.lld NM=llvm-nm AR=llvm-ar LLVM_IAS=1 menuconfig"
should be able to enable Clang LTO.
Masahiro Yamada [Wed, 10 Mar 2021 11:08:24 +0000 (20:08 +0900)]
kbuild: remove unneeded -O option to dtc
This piece of code converts the target suffix to the dtc -O option:
*.dtb -> -O dtb
*.dt.yaml -> -O yaml
Commit ce88c9c79455 ("kbuild: Add support to build overlays (%.dtbo)")
added the third case:
*.dtbo -> -O dtbo
This works thanks to commit 163f0469bf2e ("dtc: Allow overlays to have
.dtbo extension") in the upstream DTC, which has already been pulled in
the kernel.
However, I think it is a bit odd because "dtbo" is not a format name.
At least, it does not show up in the help message of dtc.
$ scripts/dtc/dtc --help
[ snip ]
-O, --out-format <arg>
Output formats are:
dts - device tree source text
dtb - device tree blob
yaml - device tree encoded as YAML
asm - assembler source
So, I am not a big fan of the second hunk of that change:
} else if (streq(outform, "dtbo")) {
dt_to_blob(outf, dti, outversion);
Anyway, we did not need to do this in Makefile in the first place.
guess_type_by_name() had already understood ".yaml" before commit 4f0e3a57d6eb ("kbuild: Add support for DT binding schema checks"),
and now does ".dtbo" as well.
Makefile does not need to duplicate the same logic. Let's leave it
to dtc.
Sami Tolvanen [Mon, 8 Mar 2021 18:46:56 +0000 (10:46 -0800)]
kbuild: Allow LTO to be selected with KASAN_HW_TAGS
While LTO with KASAN is normally not useful, hardware tag-based KASAN
can be used also in production kernels with ARM64_MTE. Therefore, allow
KASAN_HW_TAGS to be selected together with HAS_LTO_CLANG.
Jiri Slaby [Mon, 8 Mar 2021 06:28:20 +0000 (07:28 +0100)]
kbuild: dummy-tools: support MPROFILE_KERNEL checks for ppc
ppc64le checks for -mprofile-kernel to define MPROFILE_KERNEL Kconfig.
Kconfig calls arch/powerpc/tools/gcc-check-mprofile-kernel.sh for that
purpose. This script performs two checks:
1) build with -mprofile-kernel should contain "_mcount"
2) build with -mprofile-kernel with a function marked as "notrace"
should not produce "_mcount"
So support this in dummy-tools' gcc, so that we have MPROFILE_KERNEL
always true.
Masahiro Yamada [Thu, 4 Mar 2021 11:37:08 +0000 (20:37 +0900)]
kbuild: rebuild GCC plugins when the compiler is upgraded
Linus reported a build error due to the GCC plugin incompatibility
when the compiler is upgraded. [1]
GCC plugins are tied to a particular GCC version. So, they must be
rebuilt when the compiler is upgraded.
This seems to be a long-standing flaw since the initial support of
GCC plugins.
Extend commit 8b59cd81dc5e ("kbuild: ensure full rebuild when the
compiler is updated"), so that GCC plugins are covered by the
compiler upgrade detection.
Jan Beulich [Wed, 10 Mar 2021 10:46:13 +0000 (11:46 +0100)]
Xen/gntdev: don't needlessly use kvcalloc()
Requesting zeroed memory when all of it will be overwritten subsequently
by all ones is a waste of processing bandwidth. In fact, rather than
recording zeroed ->grants[], fill that array too with more appropriate
"invalid" indicators.
Jan Beulich [Wed, 10 Mar 2021 10:45:26 +0000 (11:45 +0100)]
Xen/gnttab: introduce common INVALID_GRANT_{HANDLE,REF}
It's not helpful if every driver has to cook its own. Generalize
xenbus'es INVALID_GRANT_HANDLE and pcifront's INVALID_GRANT_REF (which
shouldn't have expanded to zero to begin with). Use the constants in
p2m.c and gntdev.c right away, and update field types where necessary so
they would match with the constants' types (albeit without touching
struct ioctl_gntdev_grant_ref's ref field, as that's part of the public
interface of the kernel and would require introducing a dependency on
Xen's grant_table.h public header).
Jan Beulich [Tue, 9 Mar 2021 17:00:44 +0000 (18:00 +0100)]
Xen: drop exports of {set,clear}_foreign_p2m_mapping()
They're only used internally, and the layering violation they contain
(x86) or imply (Arm) of calling HYPERVISOR_grant_table_op() strongly
advise against any (uncontrolled) use from a module. The functions also
never had users except the ones from drivers/xen/grant-table.c forever
since their introduction in 3.15.
Juergen Gross [Sat, 6 Mar 2021 16:18:33 +0000 (17:18 +0100)]
xen/events: avoid handling the same event on two cpus at the same time
When changing the cpu affinity of an event it can happen today that
(with some unlucky timing) the same event will be handled on the old
and the new cpu at the same time.
Avoid that by adding an "event active" flag to the per-event data and
call the handler only if this flag isn't set.
Juergen Gross [Sat, 6 Mar 2021 16:18:32 +0000 (17:18 +0100)]
xen/events: don't unmask an event channel when an eoi is pending
An event channel should be kept masked when an eoi is pending for it.
When being migrated to another cpu it might be unmasked, though.
In order to avoid this keep three different flags for each event channel
to be able to distinguish "normal" masking/unmasking from eoi related
masking/unmasking and temporary masking. The event channel should only
be able to generate an interrupt if all flags are cleared.
Kenneth Feng [Tue, 9 Mar 2021 13:10:16 +0000 (21:10 +0800)]
drm/amd/pm: bug fix for pcie dpm
Currently the pcie dpm has two problems.
1. Only the high dpm level speed/width can be overrided
if the requested values are out of the pcie capability.
2. The high dpm level is always overrided though sometimes
it's not necesarry.
Dillon Varone [Fri, 19 Feb 2021 23:15:30 +0000 (18:15 -0500)]
drm/amd/display: Enabled pipe harvesting in dcn30
[Why & How]
Ported logic from dcn21 for reading in pipe fusing to dcn30.
Supported configurations are 1 and 6 pipes. Invalid fusing
will revert to 1 pipe being enabled.
Linus Torvalds [Wed, 10 Mar 2021 21:15:16 +0000 (13:15 -0800)]
Merge tag 's390-5.12-3' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Pull s390 fixes from Heiko Carstens:
- fix various user space visible copy_to_user() instances which return
the number of bytes left to copy instead of -EFAULT
- make TMPFS_INODE64 available again for s390 and alpha, now that both
architectures have been switched to 64-bit ino_t (see commit 96c0a6a72d18: "s390,alpha: switch to 64-bit ino_t")
- make sure to release a shared hypervisor resource within the zcore
device driver also on restart and power down; also remove unneeded
surrounding debugfs_create return value checks
- for the new hardware counter set device driver rename the uapi header
file to be a bit more generic; also remove 60 second read limit which
is not really necessary and without the limit the interface can be
easier tested
- some small cleanups, the largest being to convert all long long in
our time and idle code to longs
- update defconfigs
* tag 's390-5.12-3' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
s390: remove IBM_PARTITION and CONFIGFS_FS from zfcpdump defconfig
s390: update defconfigs
s390,alpha: make TMPFS_INODE64 available again
s390/cio: return -EFAULT if copy_to_user() fails
s390/tty3270: avoid comma separated statements
s390/cpumf: remove unneeded semicolon
s390/crypto: return -EFAULT if copy_to_user() fails
s390/cio: return -EFAULT if copy_to_user() fails
s390/cpumf: rename header file to hwctrset.h
s390/zcore: release dump save area on restart or power down
s390/zcore: no need to check return value of debugfs_create functions
s390/cpumf: remove 60 seconds read limit
s390/topology: remove always false if check
s390/time,idle: get rid of unsigned long long
drm/amdgpu/display: use GFP_ATOMIC in dcn21_validate_bandwidth_fp()
After fixing nested FPU contexts caused by 41401ac67791 we're still seeing
complaints about spurious kernel_fpu_end(). As it turns out this was
already fixed for dcn20 in commit f41ed88cbd ("drm/amdgpu/display:
use GFP_ATOMIC in dcn20_validate_bandwidth_internal") but never moved
forward to dcn21.
drm/amd/display: Fix nested FPU context in dcn21_validate_bandwidth()
Commit 41401ac67791 added FPU wrappers to dcn21_validate_bandwidth(),
which was correct. Unfortunately a nested function alredy contained
DC_FP_START()/DC_FP_END() calls, which results in nested FPU context
enter/exit and complaints by kernel_fpu_begin_mask().
This can be observed e.g. with 5.10.20, which backported 41401ac67791
and now emits the following warning on boot:
The warning is emitted due to the additional DC_FP_START/END calls in
patch_bounding_box(), which is inlined into dcn21_calculate_wm(),
its only caller. Removing the calls brings the code in line with
dcn20 and makes the warning disappear.
Takashi Iwai [Wed, 3 Feb 2021 12:42:41 +0000 (13:42 +0100)]
drm/amd/display: Add a backlight module option
There seem devices that don't work with the aux channel backlight
control. For allowing such users to test with the other backlight
control method, provide a new module option, aux_backlight, to specify
enabling or disabling the aux backport support explicitly. As
default, the aux support is detected by the hardware capability.
v2: make the backlight option generic in case we add future
backlight types (Alex)
Shuah Khan [Mon, 8 Mar 2021 03:53:31 +0000 (20:53 -0700)]
usbip: fix vudc usbip_sockfd_store races leading to gpf
usbip_sockfd_store() is invoked when user requests attach (import)
detach (unimport) usb gadget device from usbip host. vhci_hcd sends
import request and usbip_sockfd_store() exports the device if it is
free for export.
Export and unexport are governed by local state and shared state
- Shared state (usbip device status, sockfd) - sockfd and Device
status are used to determine if stub should be brought up or shut
down. Device status is shared between host and client.
- Local state (tcp_socket, rx and tx thread task_struct ptrs)
A valid tcp_socket controls rx and tx thread operations while the
device is in exported state.
- While the device is exported, device status is marked used and socket,
sockfd, and thread pointers are valid.
Export sequence (stub-up) includes validating the socket and creating
receive (rx) and transmit (tx) threads to talk to the client to provide
access to the exported device. rx and tx threads depends on local and
shared state to be correct and in sync.
Unexport (stub-down) sequence shuts the socket down and stops the rx and
tx threads. Stub-down sequence relies on local and shared states to be
in sync.
There are races in updating the local and shared status in the current
stub-up sequence resulting in crashes. These stem from starting rx and
tx threads before local and global state is updated correctly to be in
sync.
1. Doesn't handle kthread_create() error and saves invalid ptr in local
state that drives rx and tx threads.
2. Updates tcp_socket and sockfd, starts stub_rx and stub_tx threads
before updating usbip_device status to SDEV_ST_USED. This opens up a
race condition between the threads and usbip_sockfd_store() stub up
and down handling.
Fix the above problems:
- Stop using kthread_get_run() macro to create/start threads.
- Create threads and get task struct reference.
- Add kthread_create() failure handling and bail out.
- Hold usbip_device lock to update local and shared states after
creating rx and tx threads.
- Update usbip_device status to SDEV_ST_USED.
- Update usbip_device tcp_socket, sockfd, tcp_rx, and tcp_tx
- Start threads after usbip_device (tcp_socket, sockfd, tcp_rx, tcp_tx,
and status) is complete.
Credit goes to syzbot and Tetsuo Handa for finding and root-causing the
kthread_get_run() improper error handling problem and others. This is a
hard problem to find and debug since the races aren't seen in a normal
case. Fuzzing forces the race window to be small enough for the
kthread_get_run() error path bug and starting threads before updating the
local and shared state bug in the stub-up sequence.
Shuah Khan [Mon, 8 Mar 2021 03:53:30 +0000 (20:53 -0700)]
usbip: fix vhci_hcd attach_store() races leading to gpf
attach_store() is invoked when user requests import (attach) a device
from usbip host.
Attach and detach are governed by local state and shared state
- Shared state (usbip device status) - Device status is used to manage
the attach and detach operations on import-able devices.
- Local state (tcp_socket, rx and tx thread task_struct ptrs)
A valid tcp_socket controls rx and tx thread operations while the
device is in exported state.
- Device has to be in the right state to be attached and detached.
Attach sequence includes validating the socket and creating receive (rx)
and transmit (tx) threads to talk to the host to get access to the
imported device. rx and tx threads depends on local and shared state to
be correct and in sync.
Detach sequence shuts the socket down and stops the rx and tx threads.
Detach sequence relies on local and shared states to be in sync.
There are races in updating the local and shared status in the current
attach sequence resulting in crashes. These stem from starting rx and
tx threads before local and global state is updated correctly to be in
sync.
1. Doesn't handle kthread_create() error and saves invalid ptr in local
state that drives rx and tx threads.
2. Updates tcp_socket and sockfd, starts stub_rx and stub_tx threads
before updating usbip_device status to VDEV_ST_NOTASSIGNED. This opens
up a race condition between the threads, port connect, and detach
handling.
Fix the above problems:
- Stop using kthread_get_run() macro to create/start threads.
- Create threads and get task struct reference.
- Add kthread_create() failure handling and bail out.
- Hold vhci and usbip_device locks to update local and shared states after
creating rx and tx threads.
- Update usbip_device status to VDEV_ST_NOTASSIGNED.
- Update usbip_device tcp_socket, sockfd, tcp_rx, and tcp_tx
- Start threads after usbip_device (tcp_socket, sockfd, tcp_rx, tcp_tx,
and status) is complete.
Credit goes to syzbot and Tetsuo Handa for finding and root-causing the
kthread_get_run() improper error handling problem and others. This is
hard problem to find and debug since the races aren't seen in a normal
case. Fuzzing forces the race window to be small enough for the
kthread_get_run() error path bug and starting threads before updating the
local and shared state bug in the attach sequence.
- Update usbip_device tcp_rx and tcp_tx pointers holding vhci and
usbip_device locks.
Tested with syzbot reproducer:
- https://syzkaller.appspot.com/text?tag=ReproC&x=14801034d00000
Shuah Khan [Mon, 8 Mar 2021 03:53:29 +0000 (20:53 -0700)]
usbip: fix stub_dev usbip_sockfd_store() races leading to gpf
usbip_sockfd_store() is invoked when user requests attach (import)
detach (unimport) usb device from usbip host. vhci_hcd sends import
request and usbip_sockfd_store() exports the device if it is free
for export.
Export and unexport are governed by local state and shared state
- Shared state (usbip device status, sockfd) - sockfd and Device
status are used to determine if stub should be brought up or shut
down.
- Local state (tcp_socket, rx and tx thread task_struct ptrs)
A valid tcp_socket controls rx and tx thread operations while the
device is in exported state.
- While the device is exported, device status is marked used and socket,
sockfd, and thread pointers are valid.
Export sequence (stub-up) includes validating the socket and creating
receive (rx) and transmit (tx) threads to talk to the client to provide
access to the exported device. rx and tx threads depends on local and
shared state to be correct and in sync.
Unexport (stub-down) sequence shuts the socket down and stops the rx and
tx threads. Stub-down sequence relies on local and shared states to be
in sync.
There are races in updating the local and shared status in the current
stub-up sequence resulting in crashes. These stem from starting rx and
tx threads before local and global state is updated correctly to be in
sync.
1. Doesn't handle kthread_create() error and saves invalid ptr in local
state that drives rx and tx threads.
2. Updates tcp_socket and sockfd, starts stub_rx and stub_tx threads
before updating usbip_device status to SDEV_ST_USED. This opens up a
race condition between the threads and usbip_sockfd_store() stub up
and down handling.
Fix the above problems:
- Stop using kthread_get_run() macro to create/start threads.
- Create threads and get task struct reference.
- Add kthread_create() failure handling and bail out.
- Hold usbip_device lock to update local and shared states after
creating rx and tx threads.
- Update usbip_device status to SDEV_ST_USED.
- Update usbip_device tcp_socket, sockfd, tcp_rx, and tcp_tx
- Start threads after usbip_device (tcp_socket, sockfd, tcp_rx, tcp_tx,
and status) is complete.
Credit goes to syzbot and Tetsuo Handa for finding and root-causing the
kthread_get_run() improper error handling problem and others. This is a
hard problem to find and debug since the races aren't seen in a normal
case. Fuzzing forces the race window to be small enough for the
kthread_get_run() error path bug and starting threads before updating the
local and shared state bug in the stub-up sequence.
Tested with syzbot reproducer:
- https://syzkaller.appspot.com/text?tag=ReproC&x=14801034d00000
Shuah Khan [Mon, 8 Mar 2021 03:53:28 +0000 (20:53 -0700)]
usbip: fix vudc to check for stream socket
Fix usbip_sockfd_store() to validate the passed in file descriptor is
a stream socket. If the file descriptor passed was a SOCK_DGRAM socket,
sock_recvmsg() can't detect end of stream.
Shuah Khan [Mon, 8 Mar 2021 03:53:27 +0000 (20:53 -0700)]
usbip: fix vhci_hcd to check for stream socket
Fix attach_store() to validate the passed in file descriptor is a
stream socket. If the file descriptor passed was a SOCK_DGRAM socket,
sock_recvmsg() can't detect end of stream.
Shuah Khan [Mon, 8 Mar 2021 03:53:26 +0000 (20:53 -0700)]
usbip: fix stub_dev to check for stream socket
Fix usbip_sockfd_store() to validate the passed in file descriptor is
a stream socket. If the file descriptor passed was a SOCK_DGRAM socket,
sock_recvmsg() can't detect end of stream.
The kernel test robot reports a huge performance regression due to the
commit, and the reason seems fairly straightforward: when there is
contention on the page list (which is what causes acquire_slab() to
fail), we do _not_ want to just loop and try again, because that will
transfer the contention to the 'n->list_lock' spinlock we hold, and
just make things even worse.
This is admittedly likely a problem only on big machines - the kernel
test robot report comes from a 96-thread dual socket Intel Xeon Gold
6252 setup, but the regression there really is quite noticeable:
-47.9% regression of stress-ng.rawpkt.ops_per_sec
and the commit that was marked as being fixed (7ced37197196: "slub:
Acquire_slab() avoid loop") actually did the loop exit early very
intentionally (the hint being that "avoid loop" part of that commit
message), exactly to avoid this issue.
The correct thing to do may be to pick some kind of reasonable middle
ground: instead of breaking out of the loop on the very first sign of
contention, or trying over and over and over again, the right thing may
be to re-try _once_, and then give up on the second failure (or pick
your favorite value for "once"..).
Linus Torvalds [Wed, 10 Mar 2021 18:01:35 +0000 (10:01 -0800)]
Merge tag 'for-linus-2021-03-10' of git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux
Pull detached mounts fix from Christian Brauner:
"Creating a series of detached mounts, attaching them to the
filesystem, and unmounting them can be used to trigger an integer
overflow in ns->mounts causing the kernel to block any new mounts in
count_mounts() and returning ENOSPC because it falsely assumes that
the maximum number of mounts in the mount namespace has been reached,
i.e. it thinks it can't fit the new mounts into the mount namespace
anymore.
Without this fix heavy use of the new mount API with move_mount() will
cause the host to become unuseable and thus blocks some xfstest
patches I want to resend.
Depending on the number of mounts in your system, this can be
reproduced on any kernel that supportes open_tree() and move_mount().
A reproducer has been sent for inclusion with xfstests. It takes care
to do this in another mount namespace, not in the host's mount
namespace so there shouldn't be any risk in running it but if one did
run it on the host it would require a reboot in order to be able to
mount again. See
The root cause of this is that detached mounts aren't handled
correctly when source and target mount are identical and reside on a
shared mount causing a broken mount tree where the detached source
itself is propagated which propagation prevents for regular
bind-mounts and new mounts.
This ultimately leads to a miscalculation of the number of mounts in
the mount namespace.
Detached mounts created via 'open_tree(fd, path, OPEN_TREE_CLONE)' are
essentially like an unattached bind-mount. They can then later on be
attached to the filesystem via move_mount() which calls into
attach_recursive_mount().
Part of attaching it to the filesystem is making sure that mounts get
correctly propagated in case the destination mountpoint is MS_SHARED,
i.e. is a shared mountpoint. This is done by calling into
propagate_mnt() which walks the list of peers calling propagate_one()
on each mount in this list making sure it receives the propagation
event. The propagate_one() function thereby skips both new mounts and
bind mounts to not propagate them "into themselves". Both are
identified by checking whether the mount is already attached to any
mount namespace in mnt->mnt_ns. The is what the IS_MNT_NEW() helper is
responsible for.
However, detached mounts have an anonymous mount namespace attached to
them stashed in mnt->mnt_ns which means that IS_MNT_NEW() doesn't
realize they need to be skipped causing the mount to propagate "into
itself" breaking the mount table and causing a disconnect between the
number of mounts recorded as being beneath or reachable from the
target mountpoint and the number of mounts actually recorded/counted
in ns->mounts ultimately causing an overflow which in turn prevents
any new mounts via the ENOSPC issue.
So teach propagation to handle detached mounts by making it aware of
them. I've been tracking this issue down for the last couple of days
and then verifying that the fix is correct by unmounting everything in
my current mount table leaving only /proc and /sys mounted and running
the reproducer above overnight verifying the number of mounts counted
in ns->mounts. With this fix the counts are correct and the ENOSPC
issue can't be reproduced.
This change will only have an effect on mounts created with the new
mount API since detached mounts cannot be created with the old mount
API so regressions are extremely unlikely.
* tag 'for-linus-2021-03-10' of git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux:
mount: fix mounting of detached mounts onto targets that reside on shared mounts
Linus Torvalds [Wed, 10 Mar 2021 17:55:06 +0000 (09:55 -0800)]
Merge tag '5.12-rc2-smb3' of git://git.samba.org/sfrench/cifs-2.6
Pull cifs fixes from Steve French:
"Six cifs/smb3 fixes, three of them for stable, including some
important mulitchannel crediting fixes, and a fix for statfs error
handling"
* tag '5.12-rc2-smb3' of git://git.samba.org/sfrench/cifs-2.6:
cifs: do not send close in compound create+close requests
cifs: return proper error code in statfs(2)
cifs: change noisy error message to FYI
cifs: print MIDs in decimal notation
cifs: ask for more credit on async read/write code paths
cifs: fix credit accounting for extra channel
Shile Zhang [Thu, 18 Feb 2021 12:31:16 +0000 (20:31 +0800)]
misc/pvpanic: Export module FDT device table
Export the module FDT device table to ensure the FDT compatible strings
are listed in the module alias. This help the pvpanic driver can be
loaded on boot automatically not only the ACPI device, but also the FDT
device.
Dmitry Baryshkov [Fri, 12 Feb 2021 19:26:58 +0000 (22:26 +0300)]
misc: fastrpc: restrict user apps from sending kernel RPC messages
Verify that user applications are not using the kernel RPC message
handle to restrict them from directly attaching to guest OS on the
remote subsystem. This is a port of CVE-2019-2308 fix.
x86/perf: Use RET0 as default for guest_get_msrs to handle "no PMU" case
Initialize x86_pmu.guest_get_msrs to return 0/NULL to handle the "nop"
case. Patching in perf_guest_get_msrs_nop() during setup does not work
if there is no PMU, as setup bails before updating the static calls,
leaving x86_pmu.guest_get_msrs NULL and thus a complete nop. Ultimately,
this causes VMX abort on VM-Exit due to KVM putting random garbage from
the stack into the MSR load list.
Add a comment in KVM to note that nr_msrs is valid if and only if the
return value is non-NULL.
Pavel Begunkov [Wed, 10 Mar 2021 13:13:54 +0000 (13:13 +0000)]
io_uring: remove indirect ctx into sqo injection
We use ->ctx_new_list to notify sqo about new ctx pending, then sqo
should stop and splice it to its sqd->ctx_list, paired with
->sq_thread_comp.
The last one is broken because nobody reinitialises it, and trying to
fix it would only add more complexity and bugs. And the first isn't
really needed as is done under park(), that protects from races well.
Add ctx into sqd->ctx_list directly (under park()), it's much simpler
and allows to kill both, ctx_new_list and sq_thread_comp.
note: apparently there is no real problem at the moment, because
sq_thread_comp is used only by io_sq_thread_finish() followed by
parking, where list_del(&ctx->sqd_list) removes it well regardless
whether it's in the new or the active list.
Jens Axboe [Wed, 10 Mar 2021 02:49:02 +0000 (19:49 -0700)]
kernel: make IO threads unfreezable by default
The io-wq threads were already marked as no-freeze, but the manager was
not. On resume, we perpetually have signal_pending() being true, and
hence the manager will loop and spin 100% of the time.
Just mark the tasks created by create_io_thread() as PF_NOFREEZE by
default, and remove any knowledge of it in io-wq and io_uring.
Jens Axboe [Tue, 9 Mar 2021 23:32:13 +0000 (16:32 -0700)]
io_uring: always wait for sqd exited when stopping SQPOLL thread
We have a tiny race where io_put_sq_data() calls io_sq_thead_stop()
and finds the thread gone, but the thread has indeed not fully
exited or called complete() yet. Close it up by always having
io_sq_thread_stop() wait on completion of the exit event.
projects / linux.git / log