Fix netfs_read_gaps() to release the sink page it uses after waiting for
the request to complete. The way the sink page is used is that an
ITER_BVEC-class iterator is created that has the gaps from the target folio
at either end, but has the sink page tiled over the middle so that a single
read op can fill in both gaps.
The bug was found by KASAN detecting a UAF on the generic/075 xfstest in
the cifsd kernel thread that handles reception of data from the TCP socket:
BUG: KASAN: use-after-free in _copy_to_iter+0x48a/0xa20
Write of size 885 at addr ffff888107f92000 by task cifsd/1285
CPU: 2 UID: 0 PID: 1285 Comm: cifsd Not tainted 7.0.0 #6 PREEMPT(lazy)
Call Trace:
dump_stack_lvl+0x5d/0x80
print_report+0x17f/0x4f1
kasan_report+0x100/0x1e0
kasan_check_range+0x10f/0x1e0
__asan_memcpy+0x3c/0x60
_copy_to_iter+0x48a/0xa20
__skb_datagram_iter+0x2c9/0x430
skb_copy_datagram_iter+0x6e/0x160
tcp_recvmsg_locked+0xce0/0x1130
tcp_recvmsg+0xeb/0x300
inet_recvmsg+0xcf/0x3a0
sock_recvmsg+0xea/0x100
cifs_readv_from_socket+0x3a6/0x4d0 [cifs]
cifs_read_iter_from_socket+0xdd/0x130 [cifs]
cifs_readv_receive+0xaad/0xb10 [cifs]
cifs_demultiplex_thread+0x1148/0x1740 [cifs]
kthread+0x1cf/0x210
Fixes: ee4cdf7ba8 ("netfs: Speed up buffered reading")
Reported-by: Steve French <sfrench@samba.org>
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://patch.msgid.link/20260512123404.719402-18-dhowells@redhat.com
Reviewed-by: Paulo Alcantara (Red Hat) <pc@manguebit.org>
cc: Paulo Alcantara <pc@manguebit.org>
cc: Matthew Wilcox <willy@infradead.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
In netfs_perform_write(), "write streaming" (the caching of dirty data in
dirty but !uptodate folios) is performed to avoid the need to read data
that is just going to get immediately overwritten. However, this is/will
be disabled in three circumstances: if the fd is open O_RDWR, if fscache is
in use (as we need to round out the blocks for DIO) or if content
encryption is enabled (again for rounding out purposes).
The idea behind disabling it if the fd is open O_RDWR is that we'd need to
flush the write-streaming page before we could read the data, particularly
through mmap. But netfs now fills in the gaps if ->read_folio() is called
on the page, so that is unnecessary. Further, this doesn't actually work
if a separate fd is open for reading.
Fix this by removing the check for O_RDWR, thereby allowing streaming
writes even when we might read.
This caused a number of problems with the generic/522 xfstest, but those
are now fixed.
Fixes: c38f4e96e6 ("netfs: Provide func to copy data to pagecache for buffered write")
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://patch.msgid.link/20260512123404.719402-17-dhowells@redhat.com
cc: Paulo Alcantara <pc@manguebit.org>
cc: Matthew Wilcox <willy@infradead.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
Fix netfs_read_gaps() to remove the netfs_folio record from the folio
record before marking the folio uptodate if it successfully fills the gaps
around the dirty data in a streaming write folio (dirty, but not uptodate).
Found with:
fsx -q -N 1000000 -p 10000 -o 128000 -l 600000 \
/xfstest.test/junk --replay-ops=junk.fsxops
using the following as junk.fsxops:
truncate 0x0 0x138b1 0x8b15d *
write 0x507ee 0x10df7 0x927c0
write 0x19993 0x10e04 0x927c0 *
mapwrite 0x66214 0x1a253 0x927c0
copy_range 0xb704 0x89b9 0x24429 0x79380
write 0x2402b 0x144a2 0x90660 *
mapwrite 0x204d5 0x140a0 0x927c0 *
copy_range 0x1f72c 0x137d0 0x7a906 0x927c0 *
read 0 0x9157c 0x9157c
on cifs with the default cache option.
It shows folio 0x24 misbehaving if the FMODE_READ check is commented out in
netfs_perform_write():
if (//(file->f_mode & FMODE_READ) ||
netfs_is_cache_enabled(ctx)) {
and no fscache. This was initially found with the generic/522 xfstest.
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://patch.msgid.link/20260512123404.719402-16-dhowells@redhat.com
Fixes: ee4cdf7ba8 ("netfs: Speed up buffered reading")
cc: Paulo Alcantara <pc@manguebit.org>
cc: Matthew Wilcox <willy@infradead.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
Fix netfs_advance_writethrough() to always unlock the supplied folio and to
mark it dirty if it isn't yet written to the end. Unfortunately, it can't
be marked for writeback until the folio is done with as that may cause a
deadlock against mmapped reads and writes.
Even though it has been marked dirty, premature writeback can't occur as
the caller is holding both inode->i_rwsem (which will prevent concurrent
truncation, fallocation, DIO and other writes) and ictx->wb_lock (which
will cause flushing to wait and writeback to skip or wait).
Note that this may be easier to deal with once the queuing of folios is
split from the generation of subrequests.
Fixes: 288ace2f57 ("netfs: New writeback implementation")
Closes: https://sashiko.dev/#/patchset/20260427154639.180684-1-dhowells%40redhat.com
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://patch.msgid.link/20260512123404.719402-15-dhowells@redhat.com
cc: Paulo Alcantara <pc@manguebit.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
In order to avoid reading whilst writing, netfslib will allow "streaming
writes" in which dirty data is stored directly into folios without reading
them first. Such folios are marked dirty but may not be marked uptodate.
If a folio is entirely written by a streaming write, uptodate will be set,
otherwise it will have a netfs_folio struct attached to ->private recording
the dirty region.
In the event that a partially written streaming write page is to be
overwritten entirely by a single write(), netfs_perform_write() will try to
copy over it, but doesn't discard the netfs_folio if it succeeds; further,
it doesn't correctly handle a partial copy that overwrites some of the
dirty data.
Fix this by the following:
(1) If the folio is successfully overwritten, free the netfs_folio struct
before marking the page uptodate.
(2) If the copy to the folio partially fails, but short of the dirty data,
just ignore the copy.
(3) If the copy partially fails and overwrites some of the dirty data,
accept the copy, update the netfs_folio struct to record the new data.
If the folio is now filled, free the netfs_folio and set uptodate,
otherwise return a partial write.
Found with:
fsx -q -N 1000000 -p 10000 -o 128000 -l 600000 \
/xfstest.test/junk --replay-ops=junk.fsxops
using the following as junk.fsxops:
truncate 0x0 0 0x927c0
write 0x63fb8 0x53c8 0
copy_range 0xb704 0x19b9 0x24429 0x79380
write 0x2402b 0x144a2 0x90660 *
write 0x204d5 0x140a0 0x927c0 *
copy_range 0x1f72c 0x137d0 0x7a906 0x927c0 *
read 0x00000 0x20000 0x9157c
read 0x20000 0x20000 0x9157c
read 0x40000 0x20000 0x9157c
read 0x60000 0x20000 0x9157c
read 0x7e1a0 0xcfb9 0x9157c
on cifs with the default cache option.
It shows folio 0x24 misbehaving if the FMODE_READ check is commented out in
netfs_perform_write():
if (//(file->f_mode & FMODE_READ) ||
netfs_is_cache_enabled(ctx)) {
and no fscache. This was initially found with the generic/522 xfstest.
Fixes: 8f52de0077 ("netfs: Reduce number of conditional branches in netfs_perform_write()")
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://patch.msgid.link/20260512123404.719402-14-dhowells@redhat.com
cc: Paulo Alcantara <pc@manguebit.org>
cc: Matthew Wilcox <willy@infradead.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
If a streaming write is made, this will leave the relevant modified folio
in a not-uptodate, but dirty state with a netfs_folio struct hung off of
folio->private indicating the dirty range. Subsequently truncating the
file such that the dirty data in the folio is removed, but the first part
of the folio theoretically remains will cause the netfs_folio struct to be
discarded... but will leave the dirty flag set.
If the folio is then read via mmap(), netfs_read_folio() will see that the
page is dirty and jump to netfs_read_gaps() to fill in the missing bits.
netfs_read_gaps(), however, expects there to be a netfs_folio struct
present and can oops because truncate removed it.
Fix this by calling folio_cancel_dirty() in netfs_invalidate_folio() in the
event that all the dirty data in the folio is erased (as nfs does).
Also add some tracepoints to log modifications to a dirty page.
This can be reproduced with something like:
dd if=/dev/zero of=/xfstest.test/foo bs=1M count=1
umount /xfstest.test
mount /xfstest.test
xfs_io -c "w 0xbbbf 0xf96c" \
-c "truncate 0xbbbf" \
-c "mmap -r 0xb000 0x11000" \
-c "mr 0xb000 0x11000" \
/xfstest.test/foo
with fscaching disabled (otherwise streaming writes are suppressed) and a
change to netfs_perform_write() to disallow streaming writes if the fd is
open O_RDWR:
if (//(file->f_mode & FMODE_READ) || <--- comment this out
netfs_is_cache_enabled(ctx)) {
It should be reproducible even without this change, but if prevents the
above trivial xfs_io command from reproducing it.
Note that the initial dd is important: the file must start out sufficiently
large that the zero-point logic doesn't just clear the gaps because it
knows there's nothing in the file to read yet. Unmounting and mounting is
needed to clear the pagecache (there are other ways to do that that may
also work).
This was initially reproduced with the generic/522 xfstest on some patches
that remove the FMODE_READ restriction.
Fixes: 9ebff83e64 ("netfs: Prep to use folio->private for write grouping and streaming write")
Reported-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://patch.msgid.link/20260512123404.719402-12-dhowells@redhat.com
cc: Paulo Alcantara <pc@manguebit.org>
cc: Matthew Wilcox <willy@infradead.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
In netfs_extract_user_iter(), if iov_iter_extract_pages() failed to
extract user pages, bail out on -ENOMEM, otherwise return the error
code only if @npages == 0, allowing short DIO reads and writes to be
issued.
This fixes mmapstress02 from LTP tests against CIFS.
Fixes: 85dd2c8ff3 ("netfs: Add a function to extract a UBUF or IOVEC into a BVEC iterator")
Reported-by: Xiaoli Feng <xifeng@redhat.com>
Signed-off-by: Paulo Alcantara (Red Hat) <pc@manguebit.org>
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://patch.msgid.link/20260512123404.719402-10-dhowells@redhat.com
Cc: netfs@lists.linux.dev
Cc: stable@vger.kernel.org
Cc: linux-cifs@vger.kernel.org
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
Fix the update of the zero point[*] by netfs_release_folio() when there is
uncommitted data in the pagecache beyond the folio being released but the
on-server EOF is in this folio (ie. i_size > remote_i_size). The update
needs to limit zero_point to remote_i_size, not i_size as i_size is a local
phenomenon reflecting updates made locally to the pagecache, not stuff
written to the server. remote_i_size tracks the server's i_size.
[*] The zero point is the file position from which we can assume that the
server will just return zeros, so we can avoid generating reads.
Note that netfs_invalidate_folio() probably doesn't need fixing as
zero_point should be updated by setattr after truncation or fallocate.
Found with:
fsx -q -N 1000000 -p 10000 -o 128000 -l 600000 \
/xfstest.test/junk --replay-ops=junk.fsxops
using the following as junk.fsxops:
truncate 0x0 0x1bbae 0x82864
write 0x3ef2e 0xf9c8 0x1bbae
write 0x67e05 0xcb5a 0x4e8f6
mapread 0x57781 0x85b6 0x7495f
copy_range 0x5d3d 0x10329 0x54fac 0x7495f
write 0x64710 0x1c2b 0x7495f
mapread 0x64000 0x1000 0x7495f
on cifs with the default cache option.
It shows read-gaps on folio 0x64 failing with a short read (ie. it hits
EOF) if the FMODE_READ check is commented out in netfs_perform_write():
if (//(file->f_mode & FMODE_READ) ||
netfs_is_cache_enabled(ctx)) {
and no fscache. This was initially found with the generic/522 xfstest.
Fixes: cce6bfa6ca ("netfs: Fix trimming of streaming-write folios in netfs_inval_folio()")
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://patch.msgid.link/20260512123404.719402-7-dhowells@redhat.com
cc: Paulo Alcantara <pc@manguebit.org>
cc: Matthew Wilcox <willy@infradead.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
Fix potential tearing in using ->remote_i_size and ->zero_point by copying
i_size_read() and i_size_write() and using the same seqcount as for i_size.
We need to make sure that netfslib and the filesystems that use it always
hold i_lock whilst updating any of the sizes to prevent i_size_seqcount
from getting corrupted.
Fixes: 4058f74210 ("netfs: Keep track of the actual remote file size")
Fixes: 100ccd18bb ("netfs: Optimise away reads above the point at which there can be no data")
Closes: https://sashiko.dev/#/patchset/20260414082004.3756080-1-dhowells%40redhat.com
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://patch.msgid.link/20260512123404.719402-6-dhowells@redhat.com
cc: Paulo Alcantara <pc@manguebit.org>
cc: Matthew Wilcox <willy@infradead.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
The list of subrequests attached to stream->subrequests is accessed without
locks by netfs_collect_read_results() and netfs_collect_write_results(),
and then they access subreq->flags without taking a barrier after getting
the subreq pointer from the list. Relatedly, the functions that build the
list don't use any sort of write barrier when constructing the list to make
sure that the NETFS_SREQ_IN_PROGRESS flag is perceived to be set first if
no lock is taken.
Fix this by:
(1) Add a new list_add_tail_release() function that uses a release barrier
to set the pointer to the new member of the list.
(2) Add a new list_first_entry_or_null_acquire() function that uses an
acquire barrier to read the pointer to the first member in a list (or
return NULL).
(3) Use list_add_tail_release() when adding a subreq to ->subrequests.
(4) Use list_first_entry_or_null_acquire() when initially accessing the
front of the list (when an item is removed, the pointer to the new
front iterm is obtained under the same lock).
Fixes: e2d46f2ec3 ("netfs: Change the read result collector to only use one work item")
Fixes: 288ace2f57 ("netfs: New writeback implementation")
Link: https://sashiko.dev/#/patchset/20260326104544.509518-1-dhowells%40redhat.com
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://patch.msgid.link/20260512123404.719402-4-dhowells@redhat.com
cc: Paulo Alcantara <pc@manguebit.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
When the preparation of a new subrequest for a read fails, if the
subrequest has already been added to the stream->subrequests list, it can't
simply be put and abandoned as the collector may see it. Also, if it
hasn't been queued yet, it has two outstanding refs that both need to be
put. Both DIO read and single-read dispatch fail at this; further, both
differ in the order they do things to the way buffered read works.
Fix cancellation of both DIO-read and single-read subrequests that failed
preparation by the following steps:
(1) Harmonise all three reads (buffered, dio, single) to queue the subreq
before prepping it.
(2) Make all three call netfs_queue_read() to do the queuing.
(3) Set NETFS_RREQ_ALL_QUEUED independently of the queuing as we don't
know the length of the subreq at this point.
(4) In all cases, set the error and NETFS_SREQ_FAILED flag on the subreq
and then call netfs_read_subreq_terminated() to deal with it. This
will pass responsibility off to the collector for dealing with it.
Fixes: e2d46f2ec3 ("netfs: Change the read result collector to only use one work item")
Closes: https://sashiko.dev/#/patchset/20260425125426.3855807-1-dhowells%40redhat.com
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://patch.msgid.link/20260512123404.719402-2-dhowells@redhat.com
cc: Paulo Alcantara <pc@manguebit.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
kern_select() normalises the user-supplied struct __kernel_old_timeval
with
tv.tv_sec + (tv.tv_usec / USEC_PER_SEC)
(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC
before calling poll_select_set_timeout() -> timespec64_valid(). Both
operands of the seconds sum are unbounded user-controlled signed long.
A crafted pair where tv_usec is a negative multiple of USEC_PER_SEC
drives the sum across the wrap boundary - e.g.
{ .tv_sec = LONG_MIN, .tv_usec = -1000000 }
yields sec = LONG_MAX, nsec = 0, which passes timespec64_valid() and
then flows through timespec64_add_safe(), which saturates the absolute
deadline to TIME64_MAX (clamped further to KTIME_MAX downstream).
select(2) therefore blocks effectively forever instead of returning
-EINVAL as POSIX requires for a negative timeout.
Only the legacy __NR_select syscall takes this path. pselect6, ppoll,
poll and epoll_pwait2 all hand the user's two fields directly to
poll_select_set_timeout(), which validates *before* doing any
arithmetic:
/* fs/select.c:271 -- the validator */
int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
{
struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
if (!timespec64_valid(&ts))
return -EINVAL;
...
}
/* include/linux/time64.h:97 -- timespec64_valid */
if (ts->tv_sec < 0) return false;
if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) return false;
/* fs/select.c:744 do_pselect() (pselect6, pselect6_time32) */
if (get_timespec64(&ts, tsp)) return -EFAULT;
if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) return -EINVAL;
/* fs/select.c:1097 ppoll */
if (get_timespec64(&ts, tsp)) return -EFAULT;
if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) return -EINVAL;
/* fs/select.c:1065 poll -- timeout_msecs is int; >= 0 gates the math */
if (timeout_msecs >= 0)
poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
/* fs/eventpoll.c:2512 epoll_pwait2 */
if (get_timespec64(&ts, timeout)) return -EFAULT;
if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) return -EINVAL;
In every one of these the wrap-prone arithmetic from kern_select()
simply does not exist; the user fields reach timespec64_valid()
unmodified. glibc routes the C-library select() through pselect6,
so the bug is reachable only via a direct syscall(__NR_select, ...).
The pre-validation negative check that used to live here was lost
when the syscall was switched to the poll_select_set_timeout() helper.
Restore it: reject tv_sec < 0 || tv_usec < 0 up front, mirroring what
glibc does in userspace. do_compat_select() has the same arithmetic
pattern but is only reachable on 32-bit compat and from a different
syscall entry; left for a follow-up so this change stays minimal.
Reproducer (returns -1/EINVAL on a fixed kernel; blocks indefinitely
on an unfixed one):
struct timeval tv = { .tv_sec = LONG_MIN, .tv_usec = -1000000 };
fd_set r;
int pfd[2];
pipe(pfd);
FD_ZERO(&r);
FD_SET(pfd[0], &r);
syscall(__NR_select, pfd[0] + 1, &r, NULL, NULL, &tv);
Fixes: 4d36a9e65d ("select: deal with math overflow from borderline valid userland data")
Signed-off-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20260429-timeval-v1-1-4448e2588bbf@debian.org
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Stefano Garzarella says:
====================
vsock/virtio: fix vsockmon tap skb construction
While reviewing the patch posted by Yiqi Sun [1] to fix an issue in
virtio_transport_build_skb(), I discovered another issue related to
the offset and length of the payload to be copied in the new skb.
This was introduced when we did the skb conversion, and fixed by
patch 1.
Patch 2 fixes the issue found by Yiqi Sun in a different way: using
iov_iter_kvec() to properly initialize all the iov_iter fields and
removing the linear vs non-linear split like we alredy do in
vhost-vsock.
It could have been a single patch, but since there were two affected
commits, I decided to keep the fixes separate.
[1] https://lore.kernel.org/netdev/20260430071110.380509-1-sunyiqixm@gmail.com/
====================
Link: https://patch.msgid.link/20260508164411.261440-1-sgarzare@redhat.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
For non-linear skbs, virtio_transport_build_skb() goes through
virtio_transport_copy_nonlinear_skb() to copy the original payload
in the new skb to be delivered to the vsockmon tap device.
This manually initializes an iov_iter but does not set iov_iter.count.
Since the iov_iter is zero-initialized, the copy length is zero and no
payload is actually copied to the monitor interface, leaving data
un-initialized.
Fix this by removing the linear vs non-linear split and using
skb_copy_datagram_iter() with iov_iter_kvec() for all cases, as
vhost-vsock already does. This handles both linear and non-linear skbs,
properly initializes the iov_iter, and removes the now unused
virtio_transport_copy_nonlinear_skb().
While touching this code, let's also check the return value of
skb_copy_datagram_iter(), even though it's unlikely to fail.
Fixes: 4b0bf10eb0 ("vsock/virtio: non-linear skb handling for tap")
Reported-by: Yiqi Sun <sunyiqixm@gmail.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Bobby Eshleman <bobbyeshleman@meta.com>
Reviewed-by: Arseniy Krasnov <avkrasnov@rulkc.org>
Link: https://patch.msgid.link/20260508164411.261440-3-sgarzare@redhat.com
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
virtio_transport_build_skb() builds a new skb to be delivered to the
vsockmon tap device. To build the new skb, it uses the original skb
data length as payload length, but as the comment notes, the original
packet stored in the skb may have been split in multiple packets, so we
need to use the length in the header, which is correctly updated before
the packet is delivered to the tap, and the offset for the data.
This was also similar to what we did before commit 71dc9ec9ac
("virtio/vsock: replace virtio_vsock_pkt with sk_buff") where we probably
missed something during the skb conversion.
Also update the comment above, which was left stale by the skb
conversion and still mentioned a buffer pointer that no longer exists.
Fixes: 71dc9ec9ac ("virtio/vsock: replace virtio_vsock_pkt with sk_buff")
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Bobby Eshleman <bobbyeshleman@meta.com>
Reviewed-by: Arseniy Krasnov <avkrasnov@rulkc.org>
Link: https://patch.msgid.link/20260508164411.261440-2-sgarzare@redhat.com
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
In the HSR (High-availability Seamless Redundancy) protocol, node
information is maintained in the node_db. When a supervision frame is
received, node->addr_B_port is updated to track the receiving port type
(e.g., HSR_PT_SLAVE_B).
If the underlying physical interface associated with this slave port is
removed (e.g., via `ip link del`), hsr_del_port() frees the hsr_port
object. However, the stale node->addr_B_port reference is kept in the
node_db until the node ages out.
Subsequently, if userspace queries the node status via the Netlink
command HSR_C_GET_NODE_STATUS, the kernel calls hsr_get_node_data().
This function unconditionally dereferences the pointer returned by
hsr_port_get_hsr():
if (node->addr_B_port != HSR_PT_NONE) {
port = hsr_port_get_hsr(hsr, node->addr_B_port);
*addr_b_ifindex = port->dev->ifindex; // <-- NULL deref
}
If the slave port has been deleted, hsr_port_get_hsr() returns NULL,
resulting in a kernel panic.
Oops: general protection fault, probably for non-canonical address
KASAN: null-ptr-deref in range [0x0000000000000010-0x0000000000000017]
RIP: 0010:hsr_get_node_data+0x7b6/0x9e0
Call Trace:
<TASK>
hsr_get_node_status+0x445/0xa40
Fix this by adding a proper NULL pointer check. If the port lookup fails
due to a stale port type, gracefully treat it as if no valid port exists
and assign -1 to the interface index.
Steps to reproduce:
1. Create an HSR interface with two slave devices.
2. Receive a supervision frame to populate node_db with
addr_B_port assigned to SLAVE_B.
3. Delete the underlying slave device B.
4. Send an HSR_C_GET_NODE_STATUS Netlink message.
Fixes: c5a7591172 ("net/hsr: Use list_head (and rcu) instead of array for slave devices.")
Signed-off-by: Quan Sun <2022090917019@std.uestc.edu.cn>
Link: https://patch.msgid.link/20260508124636.1462346-1-2022090917019@std.uestc.edu.cn
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
After a GPU reset the HWSP is zeroed, so previously completed
requests appear incomplete. If such a request is picked up during
reset_rewind() and marked guilty, i915_request_set_error_once()
returns early (fence already signaled), leaving fence.error without
a fatal error code. The subsequent __i915_request_skip() then hits:
```
GEM_BUG_ON(!fatal_error(rq->fence.error))
```
Fixes a kernel BUG observed on Sandy Bridge (Gen6) during
heartbeat-triggered engine resets.
```
kernel BUG at drivers/gpu/drm/i915/i915_request.c:556!
RIP: __i915_request_skip+0x15e/0x1d0 [i915]
...
__i915_request_reset+0x212/0xa70 [i915]
reset_rewind+0xe4/0x280 [i915]
intel_gt_reset+0x30d/0x5b0 [i915]
heartbeat+0x516/0x530 [i915]
```
Guard __i915_request_skip() with i915_request_signaled(), if the
fence is already signaled, the ring content is committed and there
is nothing left to skip.
Fixes: 36e191f064 ("drm/i915: Apply i915_request_skip() on submission")
Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/work_items/13729
Signed-off-by: Sebastian Brzezinka <sebastian.brzezinka@intel.com>
Cc: stable@vger.kernel.org # v5.7+
Reviewed-by: Krzysztof Karas <krzysztof.karas@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://lore.kernel.org/r/fe76921d35b6ae85aa651822726d0d9815aa5362.1776339012.git.sebastian.brzezinka@intel.com
(cherry picked from commit 5ba54393dc)
Signed-off-by: Tvrtko Ursulin <tursulin@ursulin.net>
Four sysfs show() callbacks in hv-gpci take get_cpu_var(hv_gpci_reqb)
(which calls preempt_disable()) but only call the matching put_cpu_var()
on the error path under the 'out:' label. Every successful read leaks
one preempt_disable():
processor_bus_topology_show()
processor_config_show()
affinity_domain_via_virtual_processor_show()
affinity_domain_via_domain_show()
(affinity_domain_via_partition_show() was already correct.)
On a CONFIG_PREEMPT=y kernel, repeated reads raise preempt_count and
eventually return to userspace with preemption still disabled. The
next user-mode page fault then hits faulthandler_disabled() == 1,
gets forced to SIGSEGV, and the resulting coredump trips
'BUG: scheduling while atomic' in call_usermodehelper_exec ->
wait_for_completion_state -> schedule:
BUG: scheduling while atomic: <task>/<pid>/0x00000004
...
__schedule_bug+0x6c/0x90
__schedule+0x58c/0x13a0
schedule+0x48/0x1a0
schedule_timeout+0x104/0x170
wait_for_completion_state+0x16c/0x330
call_usermodehelper_exec+0x254/0x2d0
vfs_coredump+0x1050/0x2590
get_signal+0xb9c/0xc80
do_notify_resume+0xf8/0x470
Add an out_success label that calls put_cpu_var() before returning
the byte count, mirroring affinity_domain_via_partition_show().
Fixes: 71f1c39647 ("powerpc/hv_gpci: Add sysfs file inside hv_gpci device to show processor bus topology information")
Fixes: 1a160c2a13 ("powerpc/hv_gpci: Add sysfs file inside hv_gpci device to show processor config information")
Fixes: 71a7ccb478 ("powerpc/hv_gpci: Add sysfs file inside hv_gpci device to show affinity domain via virtual processor information")
Fixes: a69a57cac1 ("powerpc/hv_gpci: Add sysfs file inside hv_gpci device to show affinity domain via domain information")
Signed-off-by: Aboorva Devarajan <aboorvad@linux.ibm.com>
Reviewed-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Link: https://patch.msgid.link/20260508041256.3447113-1-aboorvad@linux.ibm.com
The GUEST_STATE_BUFFER_TEST config option should default
to KUNIT_ALL_TESTS so that if all tests are enabled then
it is included, but currently the 'default KUNIT_ALL_TESTS'
statement is shadowed by 'def_tristate n',
meaning that this second default statement is currently dead code.
It looks to me like the commit
6ccbbc33f0 ("KVM: PPC: Add helper library for Guest State Buffers")
intended to set the default to KUNIT_ALL_TESTS, but mistakenly
missed the def_tristate.
This dead code was found by kconfirm, a static analysis tool for Kconfig.
Fixes: 6ccbbc33f0 ("KVM: PPC: Add helper library for Guest State Buffers")
Signed-off-by: Julian Braha <julianbraha@gmail.com>
Tested-by: Gautam Menghani <gautam@linux.ibm.com>
Reviewed-by: Amit Machhiwal <amachhiw@linux.ibm.com>
Reviewed-by: Harsh Prateek Bora <harshpb@linux.ibm.com>
Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Link: https://patch.msgid.link/20260405161545.161006-1-julianbraha@gmail.com
Commit a28d3af2a2 ("[PATCH] 2/5 powerpc: Rework PowerMac i2c part 2")
removed the last calls to the pmac_low_i2c_{lock,unlock}() functions.
Hence, remove these two functions.
Reviewed-by: Christophe Leroy (CS GROUP) <chleroy@kernel.org>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Link: https://patch.msgid.link/20260316174747.3871924-1-bvanassche@acm.org
pika_dtm_thread() acquires client through of_find_i2c_device_by_node()
but fails to release it in error handling path. This could result in a
reference count leak, preventing proper cleanup and potentially
leading to resource exhaustion. Add put_device() to release the
reference in the error handling path.
Found by code review.
Cc: stable@vger.kernel.org
Fixes: 3984114f05 ("powerpc/warp: Platform fix for i2c change")
Signed-off-by: Ma Ke <make24@iscas.ac.cn>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Link: https://patch.msgid.link/20251116024411.21968-1-make24@iscas.ac.cn
Uninitialized pointers with `__free` attribute can cause undefined
behavior as the memory allocated to the pointer is freed automatically
when the pointer goes out of scope.
powerpc/km82xx doesn't have any bugs related to this as of now, but,
it is better to initialize and assign pointers with `__free` attribute
in one statement to ensure proper scope-based cleanup
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Closes: https://lore.kernel.org/all/aPiG_F5EBQUjZqsl@stanley.mountain/
Signed-off-by: Ally Heev <allyheev@gmail.com>
Fixes: 4aa5cc1e00 ("powerpc-km82xx.c: replace of_node_put() with __free")
Reviewed-by: Christophe Leroy (CS GROUP) <chleroy@kernel.org>
Reviewed-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Link: https://patch.msgid.link/20251116-aheev-uninitialized-free-attr-km82xx-v2-1-4307e2b5300d@gmail.com
The G5 defconfig is clearly intended for the G5 Powermac
series, and that should enable all the available
windfarm drivers, or the machine will overheat a short
while after booting and shut itself down, which is
annoying.
Signed-off-by: Linus Walleij <linusw@kernel.org>
Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Link: https://patch.msgid.link/20260505-powermac-g5-config-v3-1-7747bf72f874@kernel.org
Make sure resources are not improperly shared in the op cache and
cause instruction corruption this way.
Signed-off-by: Prathyushi Nangia <prathyushi.nangia@amd.com>
Co-developed-by: Borislav Petkov (AMD) <bp@alien8.de>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
imx8qxp_pxl2dpi_get_available_ep_from_port() returns ERR_PTR()
on errors. imx8qxp_pxl2dpi_find_next_bridge() stores its return
value in a __free(device_node) variable before checking IS_ERR().
When the function returns on the error path, the cleanup action calls
of_node_put() on the ERR_PTR() value.
Do not let a device_node cleanup variable hold error pointers. Change
imx8qxp_pxl2dpi_get_available_ep_from_port() to return an int and pass
the endpoint node through an output argument. Initialize the output
argument to NULL so callers hold either NULL on error paths or a valid
device_node pointer on successful path.
Fixes: ceea3f7806 ("drm/bridge: imx8qxp-pxl2dpi: simplify put of device_node pointers")
Cc: stable@vger.kernel.org
Reviewed-by: Liu Ying <victor.liu@nxp.com>
Signed-off-by: Guangshuo Li <lgs201920130244@gmail.com>
Link: https://patch.msgid.link/20260507100604.667731-1-lgs201920130244@gmail.com
Signed-off-by: Liu Ying <victor.liu@nxp.com>
In qed_init_wfq_param(), variable non_requested_count can become zero
when the number of vports with the configured flag set (including the
current vport being configured) equals total num_vports. This happens
when configuring the last unconfigured vport or when re-configuring
an already configured vport.
The function then calculates left_rate_per_vp = total_left_rate /
non_requested_count, which causes division by zero.
Fix this by skipping the division when non_requested_count is zero.
In that case, there is no remaining bandwidth to distribute, so just
record the configuration for the current vport and return success.
Fixes: bcd197c81f ("qed: Add vport WFQ configuration APIs")
Signed-off-by: Evgenii Burenchev <evg28bur@yandex.ru>
Link: https://patch.msgid.link/20260507145520.23106-1-evg28bur@yandex.ru
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
lockdep_sock_is_held() was added in tcp_ao_established_key()
by the cited commit.
It can be called from tcp_v[46]_timewait_ack() with twsk.
Since it does not have sk->sk_lock, the lockdep annotation
results in out-of-bound access.
$ pahole -C tcp_timewait_sock vmlinux | grep size
/* size: 288, cachelines: 5, members: 8 */
$ pahole -C sock vmlinux | grep sk_lock
socket_lock_t sk_lock; /* 440 192 */
Let's not use lockdep_sock_is_held() for TCP_TIME_WAIT.
Fixes: 6b2d11e2d8 ("net/tcp: Add missing lockdep annotations for TCP-AO hlist traversals")
Reported-by: Damiano Melotti <melotti@google.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20260508120853.4098365-1-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
ena_phc_gettimex64() is setting the output parameter regardless
of whether ena_com_phc_get_timestamp() succeeded or failed.
When ena_com_phc_get_timestamp() returns an error, the timestamp
parameter may contain uninitialized stack memory (e.g., when PHC is
disabled or in blocked state) or invalid hardware values. Passing
these to userspace via the PTP ioctl is both a security issue
(information leak) and a correctness bug.
Fix by checking the return code after releasing the lock and only
setting the output timestamp on success.
Fixes: e0ea34158e ("net: ena: Add PHC support in the ENA driver")
Cc: stable@vger.kernel.org
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://patch.msgid.link/20260507003518.22554-1-akiyano@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When iov_iter_get_pages2() fails in rds_message_zcopy_from_user(),
the pinned pages are released with put_page(), and
rm->data.op_mmp_znotifier is cleared. But we fail to properly
clear rm->data.op_nents.
Later when rds_message_purge() is called from rds_sendmsg() the
cleanup loop iterates over the incorrectly non zero number of
op_nents and frees them again.
Fix this by properly resetting op_nents when it should be in
rds_message_zcopy_from_user().
Fixes: 0cebaccef3 ("rds: zerocopy Tx support.")
Signed-off-by: Allison Henderson <achender@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260505234336.2132721-1-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Fix to decouple KUNIT_DEBUGFS and KUNIT_ALL_TESTS options and fix
KUNIT_DEBUGFS dependencies so it depends on DEBUG_FS without which
it will not be useful.
-----BEGIN PGP SIGNATURE-----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=QdSk
-----END PGP SIGNATURE-----
Merge tag 'linux_kselftest-kunit-fixes-7.1-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest
Pull kunit fixes from Shuah Khan:
"Fix to decouple KUNIT_DEBUGFS and KUNIT_ALL_TESTS options and fix
KUNIT_DEBUGFS dependencies so it depends on DEBUG_FS without which it
will not be useful"
* tag 'linux_kselftest-kunit-fixes-7.1-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest:
kunit: config: KUNIT_DEBUGFS should depend on DEBUG_FS
kunit: config: Enable KUNIT_DEBUGFS by default
In scx_root_enable_workfn(), put_task_struct(p) is called before scx_error()
dereferences p->comm and p->pid. If the iterator's reference is the last
drop, the task is freed synchronously and the deref becomes a UAF.
Move put_task_struct() past scx_error().
Reported-by: Sashiko <sashiko-bot@kernel.org>
Closes: https://lore.kernel.org/all/20260511214031.AF5E9C2BCB0@smtp.kernel.org/
Fixes: f0e1a0643a ("sched_ext: Implement BPF extensible scheduler class")
Cc: stable@vger.kernel.org # v6.12+
Signed-off-by: Tejun Heo <tj@kernel.org>
gfx_v12_0_init_microcode() always loads RS64 CP ucode but never set
adev->gfx.rs64_enable, so it stayed false and code that branches on it
(e.g. MEC pipe reset) used the legacy CP_MEC_CNTL path incorrectly.
Match GFX11: derive RS64 mode from the PFP firmware header (v2.0) via
amdgpu_ucode_hdr_version(). Log at debug when RS64 is enabled.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Jesse Zhang <jesse.zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit b03d53598b)
The legacy CPER debugfs reader can reach the payload path without a
valid pointer snapshot. The remaining user byte count is also treated as
the ring occupancy in dwords, so reads past the header can copy more than
requested.
Take the CPER lock before sampling pointers. Resample rptr/wptr for
payload reads, bound the payload copy by available dwords and the
remaining user size, and advance the file position for each dword copied.
Signed-off-by: Xiang Liu <xiang.liu@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 1e40ef87ff)
[Why]
dcn32_validate_bandwidth() wraps dcn32_internal_validate_bw() with
DC_FP_START()/DC_FP_END(). In x86 non-RT, DC_FP_START takes fpregs_lock(),
which disables local softirqs.
The DML1 path through dcn32_enable_phantom_plane() calls kvzalloc() to
allocate ~335 KiB for dc_plane_state. This triggers the vmalloc path,
which calls BUG_ON(in_interrupt()) because it's invoked within the
FPU-enabled (softirq disabled) region, leading to a kernel crash.
[How]
Wrap the dc_state_create_phantom_plane() call with the
DC_RUN_WITH_PREEMPTION_ENABLED() macro to allow preemption during
this memory allocation.
Fixes: 235c676342 ("drm/amd/display: add DCN32/321 specific files for Display Core")
Closes: https://gitlab.freedesktop.org/drm/amd/-/work_items/4470
Reviewed-by: Aurabindo Pillai <aurabindo.pillai@amd.com>
Signed-off-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
Signed-off-by: James Lin <pinglei.lin@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 885ccbef7b)
Cc: stable@vger.kernel.org
Fix lock inversions pointed out by Prike and Sunil. The hang detection
timeout *CAN'T* grab locks under which we wait for fences, especially
not the userq_mutex lock.
Then instead of this completely broken handling with the
hang_detect_fence just cancel the work when fences are processed and
re-start if necessary.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Sunil Khatri <sunil.khatri@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 1b62077f04)
Well the reset handling seems broken on multiple levels.
As first step of fixing this remove most calls to the hang detection.
That function should only be called after we run into a timeout! And *NOT*
as random check spread over the code in multiple places.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Sunil Khatri <sunil.khatri@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 71bea36b54)
This one was fortunately not looking so bad as the wait ioctl path, but
there were still a few things which could be fixed/improved:
1. Allocating with GFP_ATOMIC was quite unnecessary, we can do that
before taking the userq_lock.
2. Use a new mutex as protection for the fence_drv_xa so that we can do
memory allocations while holding it.
3. Starting the reset timer is unnecessary when the fence is already
signaled when we create it.
4. Cleanup error handling, avoid trying to free the queue when we don't
even got one.
v2: fix incorrect usage of xa_find, destroy the new mutex on error
v3: cleanup ref ordering
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Sunil Khatri <sunil.khatri@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 1609eb0f81)
The purpose of a GPU reset is to make sure that fence can be signaled
again and the signal and resume workers can make progress again.
So waiting for the resume worker or any fence in the GPU reset path is
just utterly nonsense.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Prike Liang <Prike.Liang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit fcd5f065ea)
Retry doesn't work here, since bo will be freed on error, leading to
UAF. However, now that we do the alloc & init before the attach, we can
now combine this as one unit and have the init do the alloc for us. This
should make the retry safe.
Reported by Sashiko.
v2: Fix up the error unwind (CI)
Closes: https://sashiko.dev/#/patchset/20260506184332.86743-2-matthew.auld%40intel.com
Fixes: eb289a5f6c ("drm/xe: Convert xe_dma_buf.c for exhaustive eviction")
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: <stable@vger.kernel.org> # v6.18+
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Link: https://patch.msgid.link/20260508102635.149172-4-matthew.auld@intel.com
(cherry picked from commit 4796694182)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
There look to be some nasty races here when triggering the
invalidate_mappings hook:
1) We do xe_bo_alloc() followed by the attach, before the actual full bo
init step in xe_dma_buf_init_obj(). However the bo is visible on the
attachments list after the attach. This is bad since exporter driver,
say amdgpu, can at any time call back into our invalidate_mappings hook,
with an empty/bogus bo, leading to potential bugs/crashes.
2) Similar to 1) but here we get a UAF, when the invalidate_mappings
hook is triggered. For example, we get as far as xe_bo_init_locked()
but this fails in some way. But here the bo will be freed on error, but
we still have it attached from dma-buf pov, so if the
invalidate_mappings is now triggered then the bo we access is gone and
we trigger UAF and more bugs/crashes.
To fix this, move the attach step until after we actually have a fully
set up buffer object. Note that the bo is not published to userspace
until later, so not sure what the comment "Don't publish the bo
until we have a valid attachment", is referring to.
We have at least two different customers reporting hitting a NULL ptr
deref in evict_flags when importing something from amdgpu, followed by
triggering the evict flow. Hit rate is also pretty low, which would
hint at some kind of race, so something like 1) or 2) might explain
this.
v2:
- Shuffle the order of the ops slightly (no functional change)
- Improve the comment to better explain the ordering (Matt B)
Assisted-by: Gemini:gemini-3 #debug
Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/7903
Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/4055
Fixes: dd08ebf6c3 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: <stable@vger.kernel.org> # v6.8+
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Acked-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Link: https://patch.msgid.link/20260508102635.149172-3-matthew.auld@intel.com
(cherry picked from commit af1f2ad0c5)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
The blitter engines' MEM_COPY and MEM_SET instructions were added as
part of the same hardware change that introduced service copy engines
(i.e., BCS1-BCS8) which is why the driver checks for service copy engine
presence when deciding whether to use these instructions or the older
XY_* instructions. However when making this decision the driver should
consider which engines are part of the hardware architecture, not which
engines are present/usable on the current device. For graphics IP
versions that architecturally include service copy engines (i.e.,
everything Xe2 and later, plus PVC's Xe_HPC) we should use MEM_SET and
MEM_COPY even in if all of the service copy engines wind up getting
fused off. I.e., we need to decide based on whether the platform's
graphics descriptor contains these engines, rather than whether the
usable engine mask contains them. This logic got broken when
gt->info.__engine_mask was removed, although in practice that mistake
has been harmless so far because there haven't been any hardware
SKUs that fuse off all of the service copy engines yet.
Replace the incorrect has_service_copy_support() function with a GT
feature flag that tracks more accurately whether the new blitter
instructions are usable. In addition to fixing incorrect logic if all
service copies are fused off, the flag also makes it more obvious what
the calling code is trying to do; previously it wasn't terribly obvious
why "has service copy engines" was being used as the condition for using
different instructions on all copy engine types.
The new feature flag is named 'has_xe2_blt_instructions' because we
expect this flag to be set for all Xe2 and later platforms (i.e.,
everything officially supported by the Xe driver). Technically there's
also one Xe1-era platform (PVC) that supports these engines/instructions
and will set this flag, but this still seems to be the most clear and
understandable name for the flag.
Fixes: 61549a2ee5 ("drm/xe: Drop __engine_mask")
Cc: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Link: https://patch.msgid.link/20260507-xe2_copy-v1-1-26506381b821@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
(cherry picked from commit 09b3998429)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
xe_bo_recompute_purgeable_state() walks all VMAs of a BO to determine
whether the BO can be made purgeable. This makes VMA create/destroy and
madvise updates O(n) in the number of mappings.
Replace the walk with BO-local counters protected by the BO dma-resv
lock:
- vma_count tracks the number of VMAs mapping the BO.
- willneed_count tracks active WILLNEED holders, including WILLNEED
VMAs and active dma-buf exports for non-imported BOs.
A DONTNEED BO is promoted back to WILLNEED on a 0->1 transition of
willneed_count. A BO is demoted to DONTNEED on a 1->0 transition only
when it still has VMAs, preserving the previous behaviour where a BO
with no mappings keeps its current madvise state.
PURGED remains terminal, preserving the existing "once purged, always
purged" rule.
Fixes: 4f44961eab ("drm/xe/vm: Prevent binding of purged buffer objects")
v2:
- Use early return for imported BOs in all four helpers to avoid
nesting (Matt B).
- Group purgeability state into a purgeable sub-struct on struct
xe_bo (Matt B).
- Reword xe_bo_willneed_put_locked() kernel-doc to explain that a 1->0
transition means all remaining active VMAs are DONTNEED (Matt B).
v3:
- Move DONTNEED/PURGED reject from vma_lock_and_validate() into
xe_vma_create(), gated on attr->purgeable_state == WILLNEED.
Fixes vm_bind bypass and partial-unbind rejection on DONTNEED
BOs (Matt B).
- Drop .check_purged from MAP and REMAP; keep it for PREFETCH and
add a comment why (Matt B).
- Skip BO validation in vma_lock_and_validate() for non-WILLNEED
VMA remnants so cleanup/remap paths do not repopulate
DONTNEED/PURGED BOs.
Suggested-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patch.msgid.link/20260506132027.2556046-1-arvind.yadav@intel.com
Signed-off-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
(cherry picked from commit 23fb2ea56c)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
