platform/kernel/linux-rpi.git
4 years agobtrfs: allow to use up to 90% of the global block rsv for unlink
Josef Bacik [Fri, 13 Mar 2020 19:58:06 +0000 (15:58 -0400)]
btrfs: allow to use up to 90% of the global block rsv for unlink

We previously had a limit of stealing 50% of the global reserve for
unlink.  This was from a time when the global reserve was used for the
delayed refs as well.  However now those reservations are kept separate,
so the global reserve can be depleted much more to allow us to make
progress for space restoring operations like unlink.  Change the minimum
amount of space required to be left in the global reserve to 10%.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Tested-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: improve global reserve stealing logic
Josef Bacik [Fri, 13 Mar 2020 19:58:05 +0000 (15:58 -0400)]
btrfs: improve global reserve stealing logic

For unlink transactions and block group removal
btrfs_start_transaction_fallback_global_rsv will first try to start an
ordinary transaction and if it fails it will fall back to reserving the
required amount by stealing from the global reserve. This is problematic
because of all the same reasons we had with previous iterations of the
ENOSPC handling, thundering herd.  We get a bunch of failures all at
once, everybody tries to allocate from the global reserve, some win and
some lose, we get an ENSOPC.

Fix this behavior by introducing BTRFS_RESERVE_FLUSH_ALL_STEAL. It's
used to mark unlink reservation. To fix this we need to integrate this
logic into the normal ENOSPC infrastructure.  We still go through all of
the normal flushing work, and at the moment we begin to fail all the
tickets we try to satisfy any tickets that are allowed to steal by
stealing from the global reserve.  If this works we start the flushing
system over again just like we would with a normal ticket satisfaction.
This serializes our global reserve stealing, so we don't have the
thundering herd problem.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Tested-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: distinguish reloc and non-reloc use of indirect resolution
Qu Wenruo [Mon, 16 Mar 2020 06:10:01 +0000 (14:10 +0800)]
btrfs: backref: distinguish reloc and non-reloc use of indirect resolution

For relocation tree detection, relocation backref cache uses
btrfs_should_ignore_reloc_root() which uses relocation-specific checks
like checking the DEAD_RELOC_ROOT bit.

However for general purpose backref cache, we can rely on that check, as
it's possible that relocation is also running.

For generic purposed backref cache, we detect reloc root by
SHARED_BLOCK_REF item.  Only reloc root node has its parent bytenr
pointing back to itself.

And in that case, backref cache will mark the reloc root node useless,
dropping any child orphan nodes.

So only call btrfs_should_ignore_reloc_root() if the backref cache is
for relocation.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: move error handling of build_backref_tree() to backref.c
Qu Wenruo [Mon, 23 Mar 2020 08:57:15 +0000 (16:57 +0800)]
btrfs: reloc: move error handling of build_backref_tree() to backref.c

The error cleanup will be extracted as a new function,
btrfs_backref_error_cleanup(), and moved to backref.c and exported for
later usage.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: rename and move finish_upper_links()
Qu Wenruo [Mon, 23 Mar 2020 08:14:08 +0000 (16:14 +0800)]
btrfs: backref: rename and move finish_upper_links()

This the the 2nd major part of generic backref cache. Move it to
backref.c so we can reuse it.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: rename and move handle_one_tree_block()
Qu Wenruo [Mon, 23 Mar 2020 08:08:34 +0000 (16:08 +0800)]
btrfs: backref: rename and move handle_one_tree_block()

This function is the major part of backref cache build process, move it
to backref.c so we can reuse it later.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: open code read_fs_root() for handle_indirect_tree_backref()
Qu Wenruo [Thu, 5 Mar 2020 07:16:09 +0000 (15:16 +0800)]
btrfs: reloc: open code read_fs_root() for handle_indirect_tree_backref()

The backref code is going to be moved to backref.c, and read_fs_root()
is just a simple wrapper, open-code it to prepare to the incoming code
move.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: rename and move should_ignore_root()
Qu Wenruo [Tue, 3 Mar 2020 06:26:02 +0000 (14:26 +0800)]
btrfs: backref: rename and move should_ignore_root()

This function is mostly single purpose to relocation backref cache, but
since we're moving the main part of backref cache to backref.c, we need
to export such function.

And to avoid confusion, rename the function to
btrfs_should_ignore_reloc_root() make the name a little more clear.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: rename and move backref_tree_panic()
Qu Wenruo [Thu, 26 Mar 2020 06:21:36 +0000 (14:21 +0800)]
btrfs: backref: rename and move backref_tree_panic()

Also change the parameter, since all callers can easily grab an fs_info,
there is no need for all the pointer chasing.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: rename and move backref_cache_cleanup()
Qu Wenruo [Tue, 3 Mar 2020 05:55:12 +0000 (13:55 +0800)]
btrfs: backref: rename and move backref_cache_cleanup()

Since we're releasing all existing nodes/edges, other than cleanup the
mess after error, "release" is a more proper naming here.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: rename and move remove_backref_node()
Qu Wenruo [Mon, 23 Mar 2020 07:42:25 +0000 (15:42 +0800)]
btrfs: backref: rename and move remove_backref_node()

Also add comment explaining the cleanup progress, to differ it from
btrfs_backref_drop_node().

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: rename and move drop_backref_node()
Qu Wenruo [Tue, 3 Mar 2020 05:35:27 +0000 (13:35 +0800)]
btrfs: backref: rename and move drop_backref_node()

With extra comment for drop_backref_node() as it has some similarity
with remove_backref_node(), thus we need extra comment explaining the
difference.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: rename and move free_backref_(node|edge)
Qu Wenruo [Tue, 3 Mar 2020 05:26:12 +0000 (13:26 +0800)]
btrfs: backref: rename and move free_backref_(node|edge)

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: rename and move link_backref_edge()
Qu Wenruo [Tue, 3 Mar 2020 05:24:06 +0000 (13:24 +0800)]
btrfs: backref: rename and move link_backref_edge()

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: rename and move alloc_backref_edge()
Qu Wenruo [Tue, 3 Mar 2020 05:22:57 +0000 (13:22 +0800)]
btrfs: backref: rename and move alloc_backref_edge()

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: rename and move alloc_backref_node()
Qu Wenruo [Tue, 3 Mar 2020 05:21:30 +0000 (13:21 +0800)]
btrfs: backref: rename and move alloc_backref_node()

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: rename and move backref_cache_init()
Qu Wenruo [Tue, 3 Mar 2020 05:14:41 +0000 (13:14 +0800)]
btrfs: backref: rename and move backref_cache_init()

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: rename tree_entry to rb_simple_node and export it
Qu Wenruo [Thu, 26 Mar 2020 06:11:09 +0000 (14:11 +0800)]
btrfs: rename tree_entry to rb_simple_node and export it

Structure tree_entry provides a very simple rb_tree which only uses
bytenr as search index.

That tree_entry is used in 3 structures: backref_node, mapping_node and
tree_block.

Since we're going to make backref_node independnt from relocation, it's
a good time to extract the tree_entry into rb_simple_node, and export it
into misc.h.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: move btrfs_backref_(node|edge|cache) structures to backref.h
Qu Wenruo [Mon, 23 Mar 2020 07:03:56 +0000 (15:03 +0800)]
btrfs: backref: move btrfs_backref_(node|edge|cache) structures to backref.h

These 3 structures are the main part of btrfs backref cache, move them
to backref.h to build the basis for later reuse.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: add btrfs_ prefix for backref_node/edge/cache
Qu Wenruo [Mon, 23 Mar 2020 06:59:06 +0000 (14:59 +0800)]
btrfs: reloc: add btrfs_ prefix for backref_node/edge/cache

Those three structures are the main elements of backref cache. Add the
"btrfs_" prefix for later export.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: refactor useless nodes handling into its own function
Qu Wenruo [Wed, 26 Feb 2020 05:08:36 +0000 (13:08 +0800)]
btrfs: reloc: refactor useless nodes handling into its own function

This patch will also add some comment for the cleanup.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: refactor finishing part of upper linkage into finish_upper_links()
Qu Wenruo [Tue, 25 Feb 2020 06:20:13 +0000 (14:20 +0800)]
btrfs: reloc: refactor finishing part of upper linkage into finish_upper_links()

After handle_one_tree_backref(), all newly added (not cached) edges and
nodes have the following features:

- Only backref_edge::list[LOWER] is linked.
  This means, we can only iterate from botton to top, not the other
  direction.

- Newly added nodes are not added to cache rb_tree yet

So to finish the backref cache, we still need to finish the links and
add all nodes into backref cache rb_tree.

This patch will refactor the existing code into finish_upper_links(),
add more comments of each branch, and why we need to do all the work.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: remove the open-coded goto loop for breadth-first search
Qu Wenruo [Thu, 5 Mar 2020 06:49:29 +0000 (14:49 +0800)]
btrfs: reloc: remove the open-coded goto loop for breadth-first search

build_backref_tree() uses "goto again;" to implement a breadth-first
search to build backref cache.

This patch will extract most of its work into a wrapper,
handle_one_tree_block(), and use a do {} while() loop to implement the
same thing.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: pass essential members for alloc_backref_node()
Qu Wenruo [Mon, 24 Feb 2020 01:34:55 +0000 (09:34 +0800)]
btrfs: reloc: pass essential members for alloc_backref_node()

Bytenr and level are essential parameters for backref_node, thus it
makes sense to initialize them at allocation time.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: use wrapper to replace open-coded edge linking
Qu Wenruo [Mon, 24 Feb 2020 01:19:02 +0000 (09:19 +0800)]
btrfs: reloc: use wrapper to replace open-coded edge linking

Since backref_edge is used to connect upper and lower backref nodes, and
needs to access both nodes, some code can look pretty nasty:

list_add_tail(&edge->list[LOWER], &cur->upper);

The above code will link @cur to the LOWER side of the edge, while both
"LOWER" and "upper" words show up.  This can sometimes be very confusing
for reader to grasp.

This patch introduces a new wrapper, link_backref_edge(), to handle the
linking behavior.  Which also has extra ASSERT() to ensure caller won't
pass wrong nodes.

Also, this updates the comment of related lists of backref_node and
backref_edge, to make it more clear that each list points to what.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: refactor indirect tree backref processing into its own function
Qu Wenruo [Thu, 5 Mar 2020 06:22:43 +0000 (14:22 +0800)]
btrfs: reloc: refactor indirect tree backref processing into its own function

The processing of indirect tree backref (TREE_BLOCK_REF) is the most
complex work.

We need to grab the fs root, do a tree search to locate all its parent
nodes, link all needed edges, and put all uncached edges to pending edge
list.

This is definitely worth a helper function.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: refactor direct tree backref processing into its own function
Qu Wenruo [Thu, 5 Mar 2020 06:06:29 +0000 (14:06 +0800)]
btrfs: reloc: refactor direct tree backref processing into its own function

For BTRFS_SHARED_BLOCK_REF_KEY, its processing is straightforward, as we
now the parent node bytenr directly.

If the parent is already cached, or a root, call it a day.
If the parent is not cached, add it pending list.

This patch will just refactor this part into its own function,
handle_direct_tree_backref() and add some comment explaining the
@ref_key parameter.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: make reloc root search-specific for relocation backref cache
Qu Wenruo [Fri, 6 Mar 2020 06:04:12 +0000 (14:04 +0800)]
btrfs: reloc: make reloc root search-specific for relocation backref cache

find_reloc_root() searches reloc_control::reloc_root_tree to find the
reloc root.  This behavior is only useful for relocation backref cache.

For the incoming more generic purpose backref cache, we don't care
about who owns the reloc root, but only care if it's a reloc root.

So this patch makes the following modifications to make the reloc root
search more specific to relocation backref:

- Add backref_node::is_reloc_root
  This will be an extra indicator for generic purposed backref cache.
  User doesn't need to read root key from backref_node::root to
  determine if it's a reloc root.
  Also for reloc tree root, it's useless and will be queued to useless
  list.

- Add backref_cache::is_reloc
  This will allow backref cache code to do different behavior for
  generic purpose backref cache and relocation backref cache.

- Pass fs_info to find_reloc_root()

- Export find_reloc_root()
  So backref.c can utilize this function.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: add backref_cache::fs_info member
Qu Wenruo [Thu, 5 Mar 2020 05:54:11 +0000 (13:54 +0800)]
btrfs: reloc: add backref_cache::fs_info member

Add this member so that we can grab fs_info without the help from
reloc_control.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: add backref_cache::pending_edge and backref_cache::useless_node
Qu Wenruo [Thu, 5 Mar 2020 05:48:31 +0000 (13:48 +0800)]
btrfs: reloc: add backref_cache::pending_edge and backref_cache::useless_node

These two new members will act the same as the existing local lists,
@useless and @list in build_backref_tree().

Currently build_backref_tree() is only executed serially, thus moving
such local list into backref_cache is still safe.

Also since we're here, use list_first_entry() to replace a lot of
list_entry() calls after !list_empty().

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: rename mark_block_processed and __mark_block_processed
Qu Wenruo [Thu, 20 Feb 2020 07:16:16 +0000 (15:16 +0800)]
btrfs: reloc: rename mark_block_processed and __mark_block_processed

These two functions are weirdly named, mark_block_processed() in fact
just marks a range dirty unconditionally, while __mark_block_processed()
does extra check before doing the marking.

This patch will open code old mark_block_processed, and rename
__mark_block_processed() to remove the "__" prefix.

Since we're here, also kill the forward declaration, which could also
kill in_block_group() with in_range() macro.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: reloc: use btrfs_backref_iter infrastructure
Qu Wenruo [Fri, 14 Feb 2020 05:48:20 +0000 (13:48 +0800)]
btrfs: reloc: use btrfs_backref_iter infrastructure

In the core function of relocation, build_backref_tree, it needs to
iterate all backref items of one tree block.

Use btrfs_backref_iter infrastructure to do the loop and make the code
more readable.

The backref items look would be much more easier to read:

ret = btrfs_backref_iter_start(iter, cur->bytenr);
for (; ret == 0; ret = btrfs_backref_iter_next(iter)) {
/* The really important work */
}

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: implement btrfs_backref_iter_next()
Qu Wenruo [Thu, 13 Feb 2020 07:04:04 +0000 (15:04 +0800)]
btrfs: backref: implement btrfs_backref_iter_next()

This function will go to the next inline/keyed backref for
btrfs_backref_iter infrastructure.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: backref: introduce the skeleton of btrfs_backref_iter
Qu Wenruo [Thu, 13 Feb 2020 06:11:04 +0000 (14:11 +0800)]
btrfs: backref: introduce the skeleton of btrfs_backref_iter

Due to the complex nature of btrfs extent tree, when we want to iterate
all backrefs of one extent, this involves quite a lot of work, like
searching the EXTENT_ITEM/METADATA_ITEM, iteration through inline and keyed
backrefs.

Normally this would result in a complex code, something like:

  btrfs_search_slot()
  /* Ensure we are at EXTENT_ITEM/METADATA_ITEM */
  while (1) { /* Loop for extent tree items */
while (ptr < end) { /* Loop for inlined items */
/* Real work here */
}
  next:
   ret = btrfs_next_item()
/* Ensure we're still at keyed item for specified bytenr */
  }

The idea of btrfs_backref_iter is to avoid such complex and hard to
read code structure, but something like the following:

  iter = btrfs_backref_iter_alloc();
  ret = btrfs_backref_iter_start(iter, bytenr);
  if (ret < 0)
goto out;
  for (; ; ret = btrfs_backref_iter_next(iter)) {
/* Real work here */
  }
  out:
  btrfs_backref_iter_free(iter);

This patch is just the skeleton + btrfs_backref_iter_start() code.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: add missing annotation for btrfs_tree_lock()
Jules Irenge [Tue, 31 Mar 2020 20:46:42 +0000 (21:46 +0100)]
btrfs: add missing annotation for btrfs_tree_lock()

Sparse reports a warning at btrfs_tree_lock()

warning: context imbalance in btrfs_tree_lock() - wrong count at exit

The root cause is the missing annotation at btrfs_tree_lock()
Add the missing __acquires(&eb->lock) annotation

Signed-off-by: Jules Irenge <jbi.octave@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agobtrfs: add missing annotation for btrfs_lock_cluster()
Jules Irenge [Tue, 31 Mar 2020 20:46:41 +0000 (21:46 +0100)]
btrfs: add missing annotation for btrfs_lock_cluster()

Sparse reports a warning at btrfs_lock_cluster()

warning: context imbalance in btrfs_lock_cluster()
- wrong count

The root cause is the missing annotation at btrfs_lock_cluster()
Add the missing __acquires(&cluster->refill_lock) annotation.

Signed-off-by: Jules Irenge <jbi.octave@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agoLinux 5.7-rc7
Linus Torvalds [Sun, 24 May 2020 22:32:54 +0000 (15:32 -0700)]
Linux 5.7-rc7

4 years agoMerge tag 'efi-urgent-2020-05-24' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 24 May 2020 17:24:10 +0000 (10:24 -0700)]
Merge tag 'efi-urgent-2020-05-24' of git://git./linux/kernel/git/tip/tip

Pull EFI fixes from Thomas Gleixner:
 "A set of EFI fixes:

   - Don't return a garbage screen info when EFI framebuffer is not
     available

   - Make the early EFI console work properly with wider fonts instead
     of drawing garbage

   - Prevent a memory buffer leak in allocate_e820()

   - Print the firmware error record properly so it can be decoded by
     users

   - Fix a symbol clash in the host tool build which only happens with
     newer compilers.

   - Add a missing check for the event log version of TPM which caused
     boot failures on several Dell systems due to an attempt to decode
     SHA-1 format with the crypto agile algorithm"

* tag 'efi-urgent-2020-05-24' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  tpm: check event log version before reading final events
  efi: Pull up arch-specific prototype efi_systab_show_arch()
  x86/boot: Mark global variables as static
  efi: cper: Add support for printing Firmware Error Record Reference
  efi/libstub/x86: Avoid EFI map buffer alloc in allocate_e820()
  efi/earlycon: Fix early printk for wider fonts
  efi/libstub: Avoid returning uninitialized data from setup_graphics()

4 years agoMerge tag 'x86-urgent-2020-05-24' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 24 May 2020 17:21:02 +0000 (10:21 -0700)]
Merge tag 'x86-urgent-2020-05-24' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "Two fixes for x86:

   - Unbreak stack dumps for inactive tasks by interpreting the special
     first frame left by __switch_to_asm() correctly.

     The recent change not to skip the first frame so ORC and frame
     unwinder behave in the same way caused all entries to be
     unreliable, i.e. prepended with '?'.

   - Use cpumask_available() instead of an implicit NULL check of a
     cpumask_var_t in mmio trace to prevent a Clang build warning"

* tag 'x86-urgent-2020-05-24' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/unwind/orc: Fix unwind_get_return_address_ptr() for inactive tasks
  x86/mmiotrace: Use cpumask_available() for cpumask_var_t variables

4 years agoMerge tag 'sched-urgent-2020-05-24' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 24 May 2020 17:14:58 +0000 (10:14 -0700)]
Merge tag 'sched-urgent-2020-05-24' of git://git./linux/kernel/git/tip/tip

Pull scheduler fixes from Thomas Gleixner:
 "A set of fixes for the scheduler:

   - Fix handling of throttled parents in enqueue_task_fair() completely.

     The recent fix overlooked a corner case where the first iteration
     terminates due to an entity already being on the runqueue which
     makes the list management incomplete and later triggers the
     assertion which checks for completeness.

   - Fix a similar problem in unthrottle_cfs_rq().

   - Show the correct uclamp values in procfs which prints the effective
     value twice instead of requested and effective"

* tag 'sched-urgent-2020-05-24' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/fair: Fix unthrottle_cfs_rq() for leaf_cfs_rq list
  sched/debug: Fix requested task uclamp values shown in procfs
  sched/fair: Fix enqueue_task_fair() warning some more

4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Sun, 24 May 2020 00:16:18 +0000 (17:16 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Pull networking fixes from David Miller:

 1) Fix RCU warnings in ipv6 multicast router code, from Madhuparna
    Bhowmik.

 2) Nexthop attributes aren't being checked properly because of
    mis-initialized iterator, from David Ahern.

 3) Revert iop_idents_reserve() change as it caused performance
    regressions and was just working around what is really a UBSAN bug
    in the compiler. From Yuqi Jin.

 4) Read MAC address properly from ROM in bmac driver (double iteration
    proceeds past end of address array), from Jeremy Kerr.

 5) Add Microsoft Surface device IDs to r8152, from Marc Payne.

 6) Prevent reference to freed SKB in __netif_receive_skb_core(), from
    Boris Sukholitko.

 7) Fix ACK discard behavior in rxrpc, from David Howells.

 8) Preserve flow hash across packet scrubbing in wireguard, from Jason
    A. Donenfeld.

 9) Cap option length properly for SO_BINDTODEVICE in AX25, from Eric
    Dumazet.

10) Fix encryption error checking in kTLS code, from Vadim Fedorenko.

11) Missing BPF prog ref release in flow dissector, from Jakub Sitnicki.

12) dst_cache must be used with BH disabled in tipc, from Eric Dumazet.

13) Fix use after free in mlxsw driver, from Jiri Pirko.

14) Order kTLS key destruction properly in mlx5 driver, from Tariq
    Toukan.

15) Check devm_platform_ioremap_resource() return value properly in
    several drivers, from Tiezhu Yang.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (71 commits)
  net: smsc911x: Fix runtime PM imbalance on error
  net/mlx4_core: fix a memory leak bug.
  net: ethernet: ti: cpsw: fix ASSERT_RTNL() warning during suspend
  net: phy: mscc: fix initialization of the MACsec protocol mode
  net: stmmac: don't attach interface until resume finishes
  net: Fix return value about devm_platform_ioremap_resource()
  net/mlx5: Fix error flow in case of function_setup failure
  net/mlx5e: CT: Correctly get flow rule
  net/mlx5e: Update netdev txq on completions during closure
  net/mlx5: Annotate mutex destroy for root ns
  net/mlx5: Don't maintain a case of del_sw_func being null
  net/mlx5: Fix cleaning unmanaged flow tables
  net/mlx5: Fix memory leak in mlx5_events_init
  net/mlx5e: Fix inner tirs handling
  net/mlx5e: kTLS, Destroy key object after destroying the TIS
  net/mlx5e: Fix allowed tc redirect merged eswitch offload cases
  net/mlx5: Avoid processing commands before cmdif is ready
  net/mlx5: Fix a race when moving command interface to events mode
  net/mlx5: Add command entry handling completion
  rxrpc: Fix a memory leak in rxkad_verify_response()
  ...

4 years agonet: smsc911x: Fix runtime PM imbalance on error
Dinghao Liu [Sat, 23 May 2020 08:08:20 +0000 (16:08 +0800)]
net: smsc911x: Fix runtime PM imbalance on error

Remove runtime PM usage counter decrement when the
increment function has not been called to keep the
counter balanced.

Signed-off-by: Dinghao Liu <dinghao.liu@zju.edu.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'mlx5-fixes-2020-05-22' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Sat, 23 May 2020 23:39:45 +0000 (16:39 -0700)]
Merge tag 'mlx5-fixes-2020-05-22' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5 fixes 2020-05-22

This series introduces some fixes to mlx5 driver.

Please pull and let me know if there is any problem.

For -stable v4.13
   ('net/mlx5: Add command entry handling completion')

For -stable v5.2
   ('net/mlx5: Fix error flow in case of function_setup failure')
   ('net/mlx5: Fix memory leak in mlx5_events_init')

For -stable v5.3
   ('net/mlx5e: Update netdev txq on completions during closure')
   ('net/mlx5e: kTLS, Destroy key object after destroying the TIS')
   ('net/mlx5e: Fix inner tirs handling')

For -stable v5.6
   ('net/mlx5: Fix cleaning unmanaged flow tables')
   ('net/mlx5: Fix a race when moving command interface to events mode')
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlx4_core: fix a memory leak bug.
Qiushi Wu [Fri, 22 May 2020 19:07:15 +0000 (14:07 -0500)]
net/mlx4_core: fix a memory leak bug.

In function mlx4_opreq_action(), pointer "mailbox" is not released,
when mlx4_cmd_box() return and error, causing a memory leak bug.
Fix this issue by going to "out" label, mlx4_free_cmd_mailbox() can
free this pointer.

Fixes: fe6f700d6cbb ("net/mlx4_core: Respond to operation request by firmware")
Signed-off-by: Qiushi Wu <wu000273@umn.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ethernet: ti: cpsw: fix ASSERT_RTNL() warning during suspend
Grygorii Strashko [Fri, 22 May 2020 17:09:28 +0000 (20:09 +0300)]
net: ethernet: ti: cpsw: fix ASSERT_RTNL() warning during suspend

vlan_for_each() are required to be called with rtnl_lock taken, otherwise
ASSERT_RTNL() warning will be triggered - which happens now during System
resume from suspend:
  cpsw_suspend()
  |- cpsw_ndo_stop()
    |- __hw_addr_ref_unsync_dev()
      |- cpsw_purge_all_mc()
         |- vlan_for_each()
            |- ASSERT_RTNL();

Hence, fix it by surrounding cpsw_ndo_stop() by rtnl_lock/unlock() calls.

Fixes: 15180eca569b ("net: ethernet: ti: cpsw: fix vlan mcast")
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: mscc: fix initialization of the MACsec protocol mode
Antoine Tenart [Fri, 22 May 2020 15:55:45 +0000 (17:55 +0200)]
net: phy: mscc: fix initialization of the MACsec protocol mode

At the very end of the MACsec block initialization in the MSCC PHY
driver, the MACsec "protocol mode" is set. This setting should be set
based on the PHY id within the package, as the bank used to access the
register used depends on this. This was not done correctly, and only the
first bank was used leading to the two upper PHYs being unstable when
using the VSC8584. This patch fixes it.

Fixes: 1bbe0ecc2a1a ("net: phy: mscc: macsec initialization")
Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: stmmac: don't attach interface until resume finishes
Leon Yu [Fri, 22 May 2020 15:29:43 +0000 (23:29 +0800)]
net: stmmac: don't attach interface until resume finishes

Commit 14b41a2959fb ("net: stmmac: Delete txtimer in suspend") was the
first attempt to fix a race between mod_timer() and setup_timer()
during stmmac_resume(). However the issue still exists as the commit
only addressed half of the issue.

Same race can still happen as stmmac_resume() re-attaches interface
way too early - even before hardware is fully initialized.  Worse,
doing so allows network traffic to restart and stmmac_tx_timer_arm()
being called in the middle of stmmac_resume(), which re-init tx timers
in stmmac_init_coalesce().  timer_list will be corrupted and system
crashes as a result of race between mod_timer() and setup_timer().

  systemd--1995    2.... 552950018us : stmmac_suspend: 4994
  ksoftirq-9       0..s2 553123133us : stmmac_tx_timer_arm: 2276
  systemd--1995    0.... 553127896us : stmmac_resume: 5101
  systemd--320     7...2 553132752us : stmmac_tx_timer_arm: 2276
  (sd-exec-1999    5...2 553135204us : stmmac_tx_timer_arm: 2276
  ---------------------------------
  pc : run_timer_softirq+0x468/0x5e0
  lr : run_timer_softirq+0x570/0x5e0
  Call trace:
   run_timer_softirq+0x468/0x5e0
   __do_softirq+0x124/0x398
   irq_exit+0xd8/0xe0
   __handle_domain_irq+0x6c/0xc0
   gic_handle_irq+0x60/0xb0
   el1_irq+0xb8/0x180
   arch_cpu_idle+0x38/0x230
   default_idle_call+0x24/0x3c
   do_idle+0x1e0/0x2b8
   cpu_startup_entry+0x28/0x48
   secondary_start_kernel+0x1b4/0x208

Fix this by deferring netif_device_attach() to the end of
stmmac_resume().

Signed-off-by: Leon Yu <leoyu@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: Fix return value about devm_platform_ioremap_resource()
Tiezhu Yang [Fri, 22 May 2020 11:03:21 +0000 (19:03 +0800)]
net: Fix return value about devm_platform_ioremap_resource()

When call function devm_platform_ioremap_resource(), we should use IS_ERR()
to check the return value and return PTR_ERR() if failed.

Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosparc32: fix page table traversal in srmmu_nocache_init()
Mike Rapoport [Sat, 23 May 2020 19:57:18 +0000 (22:57 +0300)]
sparc32: fix page table traversal in srmmu_nocache_init()

The srmmu_nocache_init() uses __nocache_fix() macro to add an offset to
page table entry to access srmmu_nocache_pool.

But since sparc32 has only three actual page table levels, pgd, p4d and
pud are essentially the same thing and pgd_offset() and p4d_offset() are
no-ops, the __nocache_fix() should be done only at PUD level.

Remove __nocache_fix() for p4d_offset() and pud_offset() and keep it
only for PUD and lower levels.

Fixes: c2bc26f7ca1f ("sparc32: use PUD rather than PGD to get PMD in srmmu_nocache_init()")
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Anatoly Pugachev <matorola@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Sat, 23 May 2020 18:21:47 +0000 (11:21 -0700)]
Merge branch 'akpm' (patches from Andrew)

Merge misc fixes from Andrew Morton:
 "11 fixes"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  MAINTAINERS: add files related to kdump
  z3fold: fix use-after-free when freeing handles
  sparc32: use PUD rather than PGD to get PMD in srmmu_nocache_init()
  MAINTAINERS: update email address for Naoya Horiguchi
  sh: include linux/time_types.h for sockios
  kasan: disable branch tracing for core runtime
  selftests/vm/write_to_hugetlbfs.c: fix unused variable warning
  selftests/vm/.gitignore: add mremap_dontunmap
  rapidio: fix an error in get_user_pages_fast() error handling
  x86: bitops: fix build regression
  device-dax: don't leak kernel memory to user space after unloading kmem

4 years agoMerge tag 'driver-core-5.7-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 23 May 2020 18:06:13 +0000 (11:06 -0700)]
Merge tag 'driver-core-5.7-rc7' of git://git./linux/kernel/git/gregkh/driver-core

Pull driver core fixes from Greg KH:
 "So, turns out the kobject fix didn't quite work, so here are four
  patches that in the end, result in just two driver core fixes for
  reported issues that no one has had problems with.

  The kobject patch that was originally in here has now been reverted,
  as Guenter reported boot problems with it on some of his systems"

* tag 'driver-core-5.7-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core:
  Revert "kobject: Make sure the parent does not get released before its children"
  kobject: Make sure the parent does not get released before its children
  driver core: Fix handling of SYNC_STATE_ONLY + STATELESS device links
  driver core: Fix SYNC_STATE_ONLY device link implementation

4 years agoMerge tag 'char-misc-5.7-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sat, 23 May 2020 18:02:42 +0000 (11:02 -0700)]
Merge tag 'char-misc-5.7-rc7' of git://git./linux/kernel/git/gregkh/char-misc

Pull char/misc fixes from Greg KH:
 "Here are some small char/misc driver fixes for 5.7-rc7 that resolve
  some reported issues. Included in here are tiny fixes for the mei,
  coresight, rtsx, ipack, and mhi drivers.

  All of these have been in linux-next with no reported issues"

* tag 'char-misc-5.7-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc:
  misc: rtsx: Add short delay after exit from ASPM
  bus: mhi: core: Fix some error return code
  ipack: tpci200: fix error return code in tpci200_register()
  coresight: cti: remove incorrect NULL return check
  mei: release me_cl object reference

4 years agoMerge tag 'staging-5.7-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sat, 23 May 2020 17:57:55 +0000 (10:57 -0700)]
Merge tag 'staging-5.7-rc7' of git://git./linux/kernel/git/gregkh/staging

Pull staging/iio fixes from Greg KH:
 "Here are some small staging and IIO driver fixes for 5.7-rc7

  Nothing major, just a collection of IIO driver fixes for reported
  issues, and a few small staging driver fixes that people have found.
  Full details are in the shortlog.

  All of these have been in linux-next for a while with no reported
  issues"

* tag 'staging-5.7-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  staging: wfx: unlock on error path
  staging: greybus: Fix uninitialized scalar variable
  staging: kpc2000: fix error return code in kp2000_pcie_probe()
  iio: sca3000: Remove an erroneous 'get_device()'
  iio: adc: stm32-dfsdm: fix device used to request dma
  iio: adc: stm32-adc: fix device used to request dma
  iio: adc: ti-ads8344: Fix channel selection
  staging: iio: ad2s1210: Fix SPI reading
  iio: dac: vf610: Fix an error handling path in 'vf610_dac_probe()'
  iio: imu: st_lsm6dsx: unlock on error in st_lsm6dsx_shub_write_raw()
  iio: chemical: atlas-sensor: correct DO-SM channels

4 years agoMerge tag 'tty-5.7-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty
Linus Torvalds [Sat, 23 May 2020 17:50:38 +0000 (10:50 -0700)]
Merge tag 'tty-5.7-rc7' of git://git./linux/kernel/git/gregkh/tty

Pull tty/serial fix from Greg KH:
 "Here is a single serial driver fix for 5.7-rc7. It resolves an issue
  with the SiFive serial console init sequence that was reported a
  number of times.

  It has been in linux-next for a while now with no reported issues"

* tag 'tty-5.7-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty:
  tty: serial: add missing spin_lock_init for SiFive serial console

4 years agoMerge tag 's390-5.7-4' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Linus Torvalds [Sat, 23 May 2020 17:42:12 +0000 (10:42 -0700)]
Merge tag 's390-5.7-4' of git://git./linux/kernel/git/s390/linux

Pull s390 fixes from Vasily Gorbik:

 - Add missing R_390_JMP_SLOT relocation type in KASLR code.

 - Fix set_huge_pte_at for empty ptes issue which has been uncovered
   with arch page table helper tests.

 - Correct initrd location for kdump kernel.

 - Fix s390_mmio_read/write with MIO in PCI code.

* tag 's390-5.7-4' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390/kaslr: add support for R_390_JMP_SLOT relocation type
  s390/mm: fix set_huge_pte_at() for empty ptes
  s390/kexec_file: fix initrd location for kdump kernel
  s390/pci: Fix s390_mmio_read/write with MIO

4 years agoMAINTAINERS: add files related to kdump
Baoquan He [Sat, 23 May 2020 05:23:15 +0000 (22:23 -0700)]
MAINTAINERS: add files related to kdump

Kdump is implemented based on kexec, however some files are only related
to crash dumping and missing, add them to KDUMP entry.

Signed-off-by: Baoquan He <bhe@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Dave Young <dyoung@redhat.com>
Link: http://lkml.kernel.org/r/20200520103633.GW5029@MiWiFi-R3L-srv
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoz3fold: fix use-after-free when freeing handles
Uladzislau Rezki [Sat, 23 May 2020 05:23:12 +0000 (22:23 -0700)]
z3fold: fix use-after-free when freeing handles

free_handle() for a foreign handle may race with inter-page compaction,
what can lead to memory corruption.

To avoid that, take write lock not read lock in free_handle to be
synchronized with __release_z3fold_page().

For example KASAN can detect it:

  ==================================================================
  BUG: KASAN: use-after-free in LZ4_decompress_safe+0x2c4/0x3b8
  Read of size 1 at addr ffffffc976695ca3 by task GoogleApiHandle/4121

  CPU: 0 PID: 4121 Comm: GoogleApiHandle Tainted: P S         OE     4.19.81-perf+ #162
  Hardware name: Sony Mobile Communications. PDX-203(KONA) (DT)
  Call trace:
     LZ4_decompress_safe+0x2c4/0x3b8
     lz4_decompress_crypto+0x3c/0x70
     crypto_decompress+0x58/0x70
     zcomp_decompress+0xd4/0x120
     ...

Apart from that, initialize zhdr->mapped_count in init_z3fold_page() and
remove "newpage" variable because it is not used anywhere.

Signed-off-by: Uladzislau Rezki <uladzislau.rezki@sony.com>
Signed-off-by: Vitaly Wool <vitaly.wool@konsulko.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Qian Cai <cai@lca.pw>
Cc: Raymond Jennings <shentino@gmail.com>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20200520082100.28876-1-vitaly.wool@konsulko.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agosparc32: use PUD rather than PGD to get PMD in srmmu_nocache_init()
Mike Rapoport [Sat, 23 May 2020 05:23:09 +0000 (22:23 -0700)]
sparc32: use PUD rather than PGD to get PMD in srmmu_nocache_init()

The kbuild test robot reported the following warning:

  arch/sparc/mm/srmmu.c: In function 'srmmu_nocache_init': arch/sparc/mm/srmmu.c:300:9: error: variable 'pud' set but not used [-Werror=unused-but-set-variable]
  300 |  pud_t *pud;

This warning is caused by misprint in the page table traversal in
srmmu_nocache_init() function which accessed a PMD entry using PGD
rather than PUD.

Since sparc32 has only 3 page table levels, the PGD and PUD are
essentially the same and usage of __nocache_fix() removed the type
checking.

Use PUD for the consistency and to silence the compiler warning.

Fixes: 7235db268a2777bc38 ("sparc32: use pgtable-nopud instead of 4level-fixup")
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Anatoly Pugachev <matorola@gmail.com>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20200520132005.GM1059226@linux.ibm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoMAINTAINERS: update email address for Naoya Horiguchi
Naoya Horiguchi [Sat, 23 May 2020 05:23:06 +0000 (22:23 -0700)]
MAINTAINERS: update email address for Naoya Horiguchi

My email address has changed due to system upgrade, so please update it
in MAINTAINERS list.  My old address (n-horiguchi@ah.jp.nec.com) will be
still active for a few months.

Note that my email system has some encoding issue and can't send patches
in raw format via git-send-email.  So patches from me will be delivered
via my free address (nao.horiguchi@gmail.com) or GitHub.

Signed-off-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/1589874488-9247-1-git-send-email-naoya.horiguchi@nec.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agosh: include linux/time_types.h for sockios
Arnd Bergmann [Sat, 23 May 2020 05:23:02 +0000 (22:23 -0700)]
sh: include linux/time_types.h for sockios

Using the socket ioctls on arch/sh (and only there) causes build time
problems when __kernel_old_timeval/__kernel_old_timespec are not already
visible to the compiler.

Add an explict include line for the header that defines these
structures.

Fixes: 8c709f9a0693 ("y2038: sh: remove timeval/timespec usage from headers")
Fixes: 0768e17073dc ("net: socket: implement 64-bit timestamps")
Reported-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Rich Felker <dalias@libc.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20200519131327.1836482-1-arnd@arndb.de
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agokasan: disable branch tracing for core runtime
Marco Elver [Sat, 23 May 2020 05:22:59 +0000 (22:22 -0700)]
kasan: disable branch tracing for core runtime

During early boot, while KASAN is not yet initialized, it is possible to
enter reporting code-path and end up in kasan_report().

While uninitialized, the branch there prevents generating any reports,
however, under certain circumstances when branches are being traced
(TRACE_BRANCH_PROFILING), we may recurse deep enough to cause kernel
reboots without warning.

To prevent similar issues in future, we should disable branch tracing
for the core runtime.

[elver@google.com: remove duplicate DISABLE_BRANCH_PROFILING, per Qian Cai]
Link: https://lore.kernel.org/lkml/20200517011732.GE24705@shao2-debian/
Link: http://lkml.kernel.org/r/20200522075207.157349-1-elver@google.com
Reported-by: kernel test robot <rong.a.chen@intel.com>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Qian Cai <cai@lca.pw>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r//20200517011732.GE24705@shao2-debian/
Link: http://lkml.kernel.org/r/20200519182459.87166-1-elver@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoselftests/vm/write_to_hugetlbfs.c: fix unused variable warning
John Hubbard [Sat, 23 May 2020 05:22:56 +0000 (22:22 -0700)]
selftests/vm/write_to_hugetlbfs.c: fix unused variable warning

Remove unused variable "i", which was triggering a compiler warning.

Fixes: 29750f71a9b4 ("hugetlb_cgroup: add hugetlb_cgroup reservation tests")
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-By: Mina Almasry <almasrymina@google.com>
Cc: Brian Geffon <bgeffon@google.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Link: http://lkml.kernel.org/r/20200517001245.361762-2-jhubbard@nvidia.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoselftests/vm/.gitignore: add mremap_dontunmap
John Hubbard [Sat, 23 May 2020 05:22:53 +0000 (22:22 -0700)]
selftests/vm/.gitignore: add mremap_dontunmap

Add mremap_dontunmap to .gitignore.

Fixes: 0c28759ee3c9 ("selftests: add MREMAP_DONTUNMAP selftest")
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Brian Geffon <bgeffon@google.com>
Link: http://lkml.kernel.org/r/20200517002509.362401-2-jhubbard@nvidia.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agorapidio: fix an error in get_user_pages_fast() error handling
John Hubbard [Sat, 23 May 2020 05:22:48 +0000 (22:22 -0700)]
rapidio: fix an error in get_user_pages_fast() error handling

In the case of get_user_pages_fast() returning fewer pages than
requested, rio_dma_transfer() does not quite do the right thing.  It
attempts to release all the pages that were requested, rather than just
the pages that were pinned.

Fix the error handling so that only the pages that were successfully
pinned are released.

Fixes: e8de370188d0 ("rapidio: add mport char device driver")
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Alexandre Bounine <alex.bou9@gmail.com>
Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20200517235620.205225-2-jhubbard@nvidia.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agox86: bitops: fix build regression
Nick Desaulniers [Sat, 23 May 2020 05:22:45 +0000 (22:22 -0700)]
x86: bitops: fix build regression

This is easily reproducible via CC=clang + CONFIG_STAGING=y +
CONFIG_VT6656=m.

It turns out that if your config tickles __builtin_constant_p via
differences in choices to inline or not, these statements produce
invalid assembly:

    $ cat foo.c
    long a(long b, long c) {
      asm("orb %1, %0" : "+q"(c): "r"(b));
      return c;
    }
    $ gcc foo.c
    foo.c: Assembler messages:
    foo.c:2: Error: `%rax' not allowed with `orb'

Use the `%b` "x86 Operand Modifier" to instead force register allocation
to select a lower-8-bit GPR operand.

The "q" constraint only has meaning on -m32 otherwise is treated as
"r".  Not all GPRs have low-8-bit aliases for -m32.

Fixes: 1651e700664b4 ("x86: Fix bitops.h warning with a moved cast")
Reported-by: kernelci.org bot <bot@kernelci.org>
Suggested-by: Andy Shevchenko <andriy.shevchenko@intel.com>
Suggested-by: Brian Gerst <brgerst@gmail.com>
Suggested-by: H. Peter Anvin <hpa@zytor.com>
Suggested-by: Ilie Halip <ilie.halip@gmail.com>
Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Tested-by: Nathan Chancellor <natechancellor@gmail.com> [build, clang-11]
Reviewed-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-By: Brian Gerst <brgerst@gmail.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Marco Elver <elver@google.com>
Cc: "Paul E. McKenney" <paulmck@kernel.org>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Luc Van Oostenryck <luc.vanoostenryck@gmail.com>
Cc: Masahiro Yamada <yamada.masahiro@socionext.com>
Cc: Daniel Axtens <dja@axtens.net>
Cc: "Peter Zijlstra (Intel)" <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20200508183230.229464-1-ndesaulniers@google.com
Link: https://github.com/ClangBuiltLinux/linux/issues/961
Link: https://lore.kernel.org/lkml/20200504193524.GA221287@google.com/
Link: https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#x86Operandmodifiers
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agodevice-dax: don't leak kernel memory to user space after unloading kmem
David Hildenbrand [Sat, 23 May 2020 05:22:42 +0000 (22:22 -0700)]
device-dax: don't leak kernel memory to user space after unloading kmem

Assume we have kmem configured and loaded:

  [root@localhost ~]# cat /proc/iomem
  ...
  140000000-33fffffff : Persistent Memory$
    140000000-1481fffff : namespace0.0
    150000000-33fffffff : dax0.0
      150000000-33fffffff : System RAM

Assume we try to unload kmem. This force-unloading will work, even if
memory cannot get removed from the system.

  [root@localhost ~]# rmmod kmem
  [   86.380228] removing memory fails, because memory [0x0000000150000000-0x0000000157ffffff] is onlined
  ...
  [   86.431225] kmem dax0.0: DAX region [mem 0x150000000-0x33fffffff] cannot be hotremoved until the next reboot

Now, we can reconfigure the namespace:

  [root@localhost ~]# ndctl create-namespace --force --reconfig=namespace0.0 --mode=devdax
  [  131.409351] nd_pmem namespace0.0: could not reserve region [mem 0x140000000-0x33fffffff]dax
  [  131.410147] nd_pmem: probe of namespace0.0 failed with error -16namespace0.0 --mode=devdax
  ...

This fails as expected due to the busy memory resource, and the memory
cannot be used.  However, the dax0.0 device is removed, and along its
name.

The name of the memory resource now points at freed memory (name of the
device):

  [root@localhost ~]# cat /proc/iomem
  ...
  140000000-33fffffff : Persistent Memory
    140000000-1481fffff : namespace0.0
    150000000-33fffffff : �_�^7_��/_��wR��WQ���^��� ...
    150000000-33fffffff : System RAM

We have to make sure to duplicate the string.  While at it, remove the
superfluous setting of the name and fixup a stale comment.

Fixes: 9f960da72b25 ("device-dax: "Hotremove" persistent memory that is used like normal RAM")
Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Vishal Verma <vishal.l.verma@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: <stable@vger.kernel.org> [5.3]
Link: http://lkml.kernel.org/r/20200508084217.9160-2-david@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoRevert "kobject: Make sure the parent does not get released before its children"
Greg Kroah-Hartman [Sat, 23 May 2020 15:11:11 +0000 (17:11 +0200)]
Revert "kobject: Make sure the parent does not get released before its children"

This reverts commit 4ef12f7198023c09ad6d25b652bd8748c965c7fa.

Guenter reports:

All my arm64be (arm64 big endian) boot tests crash with this
patch applied. Reverting it fixes the problem. Crash log and
bisect results (from pending-fixes branch) below.

And also:
arm64 images don't crash but report lots of "poison overwritten"
backtraces like the one below. On arm, I see "refcount_t:
underflow", also attached.  I didn't bisect those, but given the
context I would suspect the same culprit.

Reported-by: Guenter Roeck <linux@roeck-us.net>
Link: https://lore.kernel.org/r/20200513151840.36400-1-heikki.krogerus@linux.intel.com
Cc: Naresh Kamboju <naresh.kamboju@linaro.org>
Cc: kernel test robot <rong.a.chen@intel.com>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Cc: Brendan Higgins <brendanhiggins@google.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Sat, 23 May 2020 04:43:02 +0000 (21:43 -0700)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Three minor fixes, two in drivers, one to fix a hang after reset with
  iSCSI, and one to avoid a spurious log message; and the final core one
  to correct a suspend/resume miscount with quiesced devices"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: target: Put lun_ref at end of tmr processing
  scsi: pm: Balance pm_only counter of request queue during system resume
  scsi: qla2xxx: Do not log message when reading port speed via sysfs

4 years agoMerge tag 'drm-fixes-2020-05-23' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Sat, 23 May 2020 04:37:09 +0000 (21:37 -0700)]
Merge tag 'drm-fixes-2020-05-23' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "Things seemed to have quieten down, though no i915 pull (I even gave
  them an extra 12 hours in case they were late).

  The amdgpu floating point fix is probably the largest, but it just
  moves some code around to it doesn't do fpu stuff outside the fpu
  boundaries. Otherwise it's just a couple of vmwgfx fixes (maintainer
  change) and two etnaviv fixes.

  vmwgfx:
   - change maintainers
   - fix redundant assignment
   - fix parameter name
   - fix return value

  etnaviv:
   - memory leak fix when userspace passes a invalid softpin address
   - off-by-one crashing the kernel in the perfmon domain iteration when
     the GPU core has both 2D and 3D capabilities

  amdgpu:
   - DP fix
   - Floating point fix
   - Fix cursor stutter issue"

* tag 'drm-fixes-2020-05-23' of git://anongit.freedesktop.org/drm/drm:
  drm/amd/display: Defer cursor lock until after VUPDATE
  drm/amd/display: Remove dml_common_def file
  drm/amd/display: DP training to set properly SCRAMBLING_DISABLE
  drm/edid: Add Oculus Rift S to non-desktop list
  drm/etnaviv: Fix a leak in submit_pin_objects()
  drm/etnaviv: fix perfmon domain interation
  drm/vmwgfx: Return true in function vmw_fence_obj_signaled()
  drm/vmwgfx: remove redundant assignment to variable ret
  drm/vmwgfx: Fix parameter name in vmw_bo_init
  drm/vmwgfx: update MAINTAINERS entry

4 years agonet/mlx5: Fix error flow in case of function_setup failure
Shay Drory [Wed, 6 May 2020 11:52:04 +0000 (14:52 +0300)]
net/mlx5: Fix error flow in case of function_setup failure

Currently, if an error occurred during mlx5_function_setup(), we
keep dev->state as DEVICE_STATE_UP.
Fixing it by adding a goto label.

Fixes: e161105e58da ("net/mlx5: Function setup/teardown procedures")
Signed-off-by: Shay Drory <shayd@mellanox.com>
Reviewed-by: Moshe Shemesh <moshe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: CT: Correctly get flow rule
Roi Dayan [Mon, 18 May 2020 17:21:11 +0000 (20:21 +0300)]
net/mlx5e: CT: Correctly get flow rule

The correct way is to us the flow_cls_offload_flow_rule() wrapper
instead of f->rule directly.

Fixes: 4c3844d9e97e ("net/mlx5e: CT: Introduce connection tracking")
Signed-off-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Update netdev txq on completions during closure
Moshe Shemesh [Tue, 7 Apr 2020 14:38:28 +0000 (17:38 +0300)]
net/mlx5e: Update netdev txq on completions during closure

On sq closure when we free its descriptors, we should also update netdev
txq on completions which would not arrive. Otherwise if we reopen sqs
and attach them back, for example on fw fatal recovery flow, we may get
tx timeout.

Fixes: 29429f3300a3 ("net/mlx5e: Timeout if SQ doesn't flush during close")
Signed-off-by: Moshe Shemesh <moshe@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Annotate mutex destroy for root ns
Roi Dayan [Thu, 14 May 2020 20:44:38 +0000 (23:44 +0300)]
net/mlx5: Annotate mutex destroy for root ns

Invoke mutex_destroy() to catch any errors.

Fixes: 2cc43b494a6c ("net/mlx5_core: Managing root flow table")
Signed-off-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Don't maintain a case of del_sw_func being null
Roi Dayan [Mon, 11 May 2020 13:37:11 +0000 (16:37 +0300)]
net/mlx5: Don't maintain a case of del_sw_func being null

Add del_sw_func cb for root ns. Now there is no need to
maintain a case of del_sw_func being null when freeing the node.

Fixes: 2cc43b494a6c ("net/mlx5_core: Managing root flow table")
Signed-off-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Fix cleaning unmanaged flow tables
Roi Dayan [Mon, 11 May 2020 13:32:09 +0000 (16:32 +0300)]
net/mlx5: Fix cleaning unmanaged flow tables

Unmanaged flow tables doesn't have a parent and tree_put_node()
assume there is always a parent if cleaning is needed. fix that.

Fixes: 5281a0c90919 ("net/mlx5: fs_core: Introduce unmanaged flow tables")
Signed-off-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Fix memory leak in mlx5_events_init
Moshe Shemesh [Wed, 29 Apr 2020 20:56:58 +0000 (23:56 +0300)]
net/mlx5: Fix memory leak in mlx5_events_init

Fix memory leak in mlx5_events_init(), in case
create_single_thread_workqueue() fails, events
struct should be freed.

Fixes: 5d3c537f9070 ("net/mlx5: Handle event of power detection in the PCIE slot")
Signed-off-by: Moshe Shemesh <moshe@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Fix inner tirs handling
Roi Dayan [Thu, 30 Apr 2020 06:16:01 +0000 (09:16 +0300)]
net/mlx5e: Fix inner tirs handling

In the cited commit inner_tirs argument was added to create and destroy
inner tirs, and no indication was added to mlx5e_modify_tirs_hash()
function. In order to have a consistent handling, use
inner_indir_tir[0].tirn in tirs destroy/modify function as an indication
to whether inner tirs are created.
Inner tirs are not created for representors and before this commit,
a call to mlx5e_modify_tirs_hash() was sending HW commands to
modify non-existent inner tirs.

Fixes: 46dc933cee82 ("net/mlx5e: Provide explicit directive if to create inner indirect tirs")
Signed-off-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: kTLS, Destroy key object after destroying the TIS
Tariq Toukan [Mon, 27 Apr 2020 13:56:59 +0000 (16:56 +0300)]
net/mlx5e: kTLS, Destroy key object after destroying the TIS

The TLS TIS object contains the dek/key ID.
By destroying the key first, the TIS would contain an invalid
non-existing key ID.
Reverse the destroy order, this also acheives the desired assymetry
between the destroy and the create flows.

Fixes: d2ead1f360e8 ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Fix allowed tc redirect merged eswitch offload cases
Maor Dickman [Thu, 23 Apr 2020 12:16:17 +0000 (15:16 +0300)]
net/mlx5e: Fix allowed tc redirect merged eswitch offload cases

After changing the parent_id to be the same for both NICs of same
The cited commit wrongly allow offload of tc redirect flows from
VF to uplink and vice versa when devcies are on different eswitch,
these cases aren't supported by HW.

Disallow the above offloads when devcies are on different eswitch
and VF LAG is not configured.

Fixes: f6dc1264f1c0 ("net/mlx5e: Disallow tc redirect offload cases we don't support")
Signed-off-by: Maor Dickman <maord@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Avoid processing commands before cmdif is ready
Eran Ben Elisha [Thu, 19 Mar 2020 19:43:13 +0000 (21:43 +0200)]
net/mlx5: Avoid processing commands before cmdif is ready

When driver is reloading during recovery flow, it can't get new commands
till command interface is up again. Otherwise we may get to null pointer
trying to access non initialized command structures.

Add cmdif state to avoid processing commands while cmdif is not ready.

Fixes: e126ba97dba9 ("mlx5: Add driver for Mellanox Connect-IB adapters")
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Moshe Shemesh <moshe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Fix a race when moving command interface to events mode
Eran Ben Elisha [Wed, 18 Mar 2020 19:44:32 +0000 (21:44 +0200)]
net/mlx5: Fix a race when moving command interface to events mode

After driver creates (via FW command) an EQ for commands, the driver will
be informed on new commands completion by EQE. However, due to a race in
driver's internal command mode metadata update, some new commands will
still be miss-handled by driver as if we are in polling mode. Such commands
can get two non forced completion, leading to already freed command entry
access.

CREATE_EQ command, that maps EQ to the command queue must be posted to the
command queue while it is empty and no other command should be posted.

Add SW mechanism that once the CREATE_EQ command is about to be executed,
all other commands will return error without being sent to the FW. Allow
sending other commands only after successfully changing the driver's
internal command mode metadata.
We can safely return error to all other commands while creating the command
EQ, as all other commands might be sent from the user/application during
driver load. Application can rerun them later after driver's load was
finished.

Fixes: e126ba97dba9 ("mlx5: Add driver for Mellanox Connect-IB adapters")
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Moshe Shemesh <moshe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Add command entry handling completion
Moshe Shemesh [Fri, 27 Dec 2019 05:01:53 +0000 (07:01 +0200)]
net/mlx5: Add command entry handling completion

When FW response to commands is very slow and all command entries in
use are waiting for completion we can have a race where commands can get
timeout before they get out of the queue and handled. Timeout
completion on uninitialized command will cause releasing command's
buffers before accessing it for initialization and then we will get NULL
pointer exception while trying access it. It may also cause releasing
buffers of another command since we may have timeout completion before
even allocating entry index for this command.
Add entry handling completion to avoid this race.

Fixes: e126ba97dba9 ("mlx5: Add driver for Mellanox Connect-IB adapters")
Signed-off-by: Moshe Shemesh <moshe@mellanox.com>
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agoMerge tag 'rxrpc-fixes-20200523-v2' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Fri, 22 May 2020 23:43:58 +0000 (16:43 -0700)]
Merge tag 'rxrpc-fixes-20200523-v2' of git://git./linux/kernel/git/dhowells/linux-fs

David Howells says:

====================
rxrpc: Fix a warning and a leak [ver #2]

Here are a couple of fixes for AF_RXRPC:

 (1) Fix an uninitialised variable warning.

 (2) Fix a leak of the ticket on error in rxkad.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agorxrpc: Fix a memory leak in rxkad_verify_response()
Qiushi Wu [Fri, 22 May 2020 18:45:18 +0000 (13:45 -0500)]
rxrpc: Fix a memory leak in rxkad_verify_response()

A ticket was not released after a call of the function
"rxkad_decrypt_ticket" failed. Thus replace the jump target
"temporary_error_free_resp" by "temporary_error_free_ticket".

Fixes: 8c2f826dc3631 ("rxrpc: Don't put crypto buffers on the stack")
Signed-off-by: Qiushi Wu <wu000273@umn.edu>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Markus Elfring <Markus.Elfring@web.de>

4 years agorxrpc: Fix a warning
David Howells [Fri, 22 May 2020 22:58:28 +0000 (23:58 +0100)]
rxrpc: Fix a warning

Fix a warning due to an uninitialised variable.

le included from ../fs/afs/fs_probe.c:11:
../fs/afs/fs_probe.c: In function 'afs_fileserver_probe_result':
../fs/afs/internal.h:1453:2: warning: 'rtt_us' may be used uninitialized in this function [-Wmaybe-uninitialized]
 1453 |  printk("[%-6.6s] "FMT"\n", current->comm ,##__VA_ARGS__)
      |  ^~~~~~
../fs/afs/fs_probe.c:35:15: note: 'rtt_us' was declared here

Signed-off-by: David Howells <dhowells@redhat.com>
4 years agonet: sun: fix missing release regions in cas_init_one().
Qiushi Wu [Fri, 22 May 2020 21:50:27 +0000 (16:50 -0500)]
net: sun: fix missing release regions in cas_init_one().

In cas_init_one(), "pdev" is requested by "pci_request_regions", but it
was not released after a call of the function “pci_write_config_byte”
failed. Thus replace the jump target “err_write_cacheline” by
"err_out_free_res".

Fixes: 1f26dac32057 ("[NET]: Add Sun Cassini driver.")
Signed-off-by: Qiushi Wu <wu000273@umn.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: fix address ageing time (again)
Vladimir Oltean [Thu, 21 May 2020 21:31:23 +0000 (00:31 +0300)]
net: mscc: ocelot: fix address ageing time (again)

ocelot_set_ageing_time has 2 callers:
 - felix_set_ageing_time: from drivers/net/dsa/ocelot/felix.c
 - ocelot_port_attr_ageing_set: from drivers/net/ethernet/mscc/ocelot.c

The issue described in the fixed commit below actually happened for the
felix_set_ageing_time code path only, since ocelot_port_attr_ageing_set
was already dividing by 1000. So to make both paths symmetrical (and to
fix addresses getting aged way too fast on Ocelot), stop dividing by
1000 at caller side altogether.

Fixes: c0d7eccbc761 ("net: mscc: ocelot: ANA_AUTOAGE_AGE_PERIOD holds a value in seconds, not ms")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: fix OCP access on RTL8117
Heiner Kallweit [Thu, 21 May 2020 20:03:08 +0000 (22:03 +0200)]
r8169: fix OCP access on RTL8117

According to r8168 vendor driver DASHv3 chips like RTL8168fp/RTL8117
need a special addressing for OCP access.
Fix is compile-tested only due to missing test hardware.

Fixes: 1287723aa139 ("r8169: add support for RTL8117")
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'mlxsw-Various-fixes'
David S. Miller [Fri, 22 May 2020 23:08:14 +0000 (16:08 -0700)]
Merge branch 'mlxsw-Various-fixes'

Ido Schimmel says:

====================
mlxsw: Various fixes

Patch #1 from Jiri fixes a use-after-free discovered while fuzzing mlxsw
/ devlink with syzkaller.

Patch #2 from Amit works around a limitation in new versions of arping,
which is used in several selftests.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: mlxsw: qos_mc_aware: Specify arping timeout as an integer
Amit Cohen [Thu, 21 May 2020 12:11:45 +0000 (15:11 +0300)]
selftests: mlxsw: qos_mc_aware: Specify arping timeout as an integer

Starting from iputils s20190709 (used in Fedora 31), arping does not
support timeout being specified as a decimal:

$ arping -c 1 -I swp1 -b 192.0.2.66 -q -w 0.1
arping: invalid argument: '0.1'

Previously, such timeouts were rounded to an integer.

Fix this by specifying the timeout as an integer.

Fixes: a5ee171d087e ("selftests: mlxsw: qos_mc_aware: Add a test for UC awareness")
Signed-off-by: Amit Cohen <amitc@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Fix use-after-free of split/unsplit/type_set in case reload fails
Jiri Pirko [Thu, 21 May 2020 12:11:44 +0000 (15:11 +0300)]
mlxsw: spectrum: Fix use-after-free of split/unsplit/type_set in case reload fails

In case of reload fail, the mlxsw_sp->ports contains a pointer to a
freed memory (either by reload_down() or reload_up() error path).
Fix this by initializing the pointer to NULL and checking it before
dereferencing in split/unsplit/type_set callpaths.

Fixes: 24cc68ad6c46 ("mlxsw: core: Add support for reload")
Reported-by: Danielle Ratson <danieller@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ethernet: stmmac: Enable interface clocks on probe for IPQ806x
Jonathan McDowell [Thu, 21 May 2020 11:49:34 +0000 (12:49 +0100)]
net: ethernet: stmmac: Enable interface clocks on probe for IPQ806x

The ipq806x_gmac_probe() function enables the PTP clock but not the
appropriate interface clocks. This means that if the bootloader hasn't
done so attempting to bring up the interface will fail with an error
like:

[   59.028131] ipq806x-gmac-dwmac 37600000.ethernet: Failed to reset the dma
[   59.028196] ipq806x-gmac-dwmac 37600000.ethernet eth1: stmmac_hw_setup: DMA engine initialization failed
[   59.034056] ipq806x-gmac-dwmac 37600000.ethernet eth1: stmmac_open: Hw setup failed

This patch, a slightly cleaned up version of one posted by Sergey
Sergeev in:

https://forum.openwrt.org/t/support-for-mikrotik-rb3011uias-rm/4064/257

correctly enables the clock; we have already configured the source just
before this.

Tested on a MikroTik RB3011.

Signed-off-by: Jonathan McDowell <noodles@earth.li>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'netdevsim-Two-small-fixes'
David S. Miller [Fri, 22 May 2020 23:05:42 +0000 (16:05 -0700)]
Merge branch 'netdevsim-Two-small-fixes'

Ido Schimmel says:

====================
netdevsim: Two small fixes

Fix two bugs observed while analyzing regression failures.

Patch #1 fixes a bug where sometimes the drop counter of a packet trap
policer would not increase.

Patch #2 adds a missing initialization of a variable in a related
selftest.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: netdevsim: Always initialize 'RET' variable
Ido Schimmel [Thu, 21 May 2020 11:46:17 +0000 (14:46 +0300)]
selftests: netdevsim: Always initialize 'RET' variable

The variable is used by log_test() to check if the test case completely
successfully or not. In case it is not initialized at the start of a
test case, it is possible for the test case to fail despite not
encountering any errors.

Example:

```
...
TEST: Trap group statistics                                         [ OK ]
TEST: Trap policer                                                  [FAIL]
Policer drop counter was not incremented
TEST: Trap policer binding                                          [FAIL]
Policer drop counter was not incremented
```

Failure of trap_policer_test() caused trap_policer_bind_test() to fail
as well.

Fix by adding missing initialization of the variable.

Fixes: 5fbff58e27a1 ("selftests: netdevsim: Add test cases for devlink-trap policers")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonetdevsim: Ensure policer drop counter always increases
Ido Schimmel [Thu, 21 May 2020 11:46:16 +0000 (14:46 +0300)]
netdevsim: Ensure policer drop counter always increases

In case the policer drop counter is retrieved when the jiffies value is
a multiple of 64, the counter will not be incremented.

This randomly breaks a selftest [1] the reads the counter twice and
checks that it was incremented:

```
TEST: Trap policer                                                  [FAIL]
Policer drop counter was not incremented
```

Fix by always incrementing the counter by 1.

[1] tools/testing/selftests/drivers/net/netdevsim/devlink_trap.sh

Fixes: ad188458d012 ("netdevsim: Add devlink-trap policer support")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'rxrpc-fixes-20200520' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Fri, 22 May 2020 22:53:30 +0000 (15:53 -0700)]
Merge tag 'rxrpc-fixes-20200520' of git://git./linux/kernel/git/dhowells/linux-fs

David Howells says:

====================
rxrpc: Fix retransmission timeout and ACK discard

Here are a couple of fixes and an extra tracepoint for AF_RXRPC:

 (1) Calculate the RTO pretty much as TCP does, rather than making
     something up, including an initial 4s timeout (which causes return
     probes from the fileserver to fail if a packet goes missing), and add
     backoff.

 (2) Fix the discarding of out-of-order received ACKs.  We mustn't let the
     hard-ACK point regress, nor do we want to do unnecessary
     retransmission because the soft-ACK list regresses.  This is not
     trivial, however, due to some loose wording in various old protocol
     specs, the ACK field that should be used for this sometimes has the
     wrong information in it.

 (3) Add a tracepoint to log a discarded ACK.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/ethernet/freescale: rework quiesce/activate for ucc_geth
Valentin Longchamp [Wed, 20 May 2020 15:53:50 +0000 (17:53 +0200)]
net/ethernet/freescale: rework quiesce/activate for ucc_geth

ugeth_quiesce/activate are used to halt the controller when there is a
link change that requires to reconfigure the mac.

The previous implementation called netif_device_detach(). This however
causes the initial activation of the netdevice to fail precisely because
it's detached. For details, see [1].

A possible workaround was the revert of commit
net: linkwatch: add check for netdevice being present to linkwatch_do_dev
However, the check introduced in the above commit is correct and shall be
kept.

The netif_device_detach() is thus replaced with
netif_tx_stop_all_queues() that prevents any tranmission. This allows to
perform mac config change required by the link change, without detaching
the corresponding netdevice and thus not preventing its initial
activation.

[1] https://lists.openwall.net/netdev/2020/01/08/201

Signed-off-by: Valentin Longchamp <valentin@longchamp.me>
Acked-by: Matteo Ghidoni <matteo.ghidoni@ch.abb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosctp: Start shutdown on association restart if in SHUTDOWN-SENT state and socket...
Jere Leppänen [Wed, 20 May 2020 15:15:31 +0000 (18:15 +0300)]
sctp: Start shutdown on association restart if in SHUTDOWN-SENT state and socket is closed

Commit bdf6fa52f01b ("sctp: handle association restarts when the
socket is closed.") starts shutdown when an association is restarted,
if in SHUTDOWN-PENDING state and the socket is closed. However, the
rationale stated in that commit applies also when in SHUTDOWN-SENT
state - we don't want to move an association to ESTABLISHED state when
the socket has been closed, because that results in an association
that is unreachable from user space.

The problem scenario:

1.  Client crashes and/or restarts.

2.  Server (using one-to-one socket) calls close(). SHUTDOWN is lost.

3.  Client reconnects using the same addresses and ports.

4.  Server's association is restarted. The association and the socket
    move to ESTABLISHED state, even though the server process has
    closed its descriptor.

Also, after step 4 when the server process exits, some resources are
leaked in an attempt to release the underlying inet sock structure in
ESTABLISHED state:

    IPv4: Attempt to release TCP socket in state 1 00000000377288c7

Fix by acting the same way as in SHUTDOWN-PENDING state. That is, if
an association is restarted in SHUTDOWN-SENT state and the socket is
closed, then start shutdown and don't move the association or the
socket to ESTABLISHED state.

Fixes: bdf6fa52f01b ("sctp: handle association restarts when the socket is closed.")
Signed-off-by: Jere Leppänen <jere.leppanen@nokia.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotipc: block BH before using dst_cache
Eric Dumazet [Thu, 21 May 2020 18:29:58 +0000 (11:29 -0700)]
tipc: block BH before using dst_cache

dst_cache_get() documents it must be used with BH disabled.

sysbot reported :

BUG: using smp_processor_id() in preemptible [00000000] code: /21697
caller is dst_cache_get+0x3a/0xb0 net/core/dst_cache.c:68
CPU: 0 PID: 21697 Comm:  Not tainted 5.7.0-rc6-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x188/0x20d lib/dump_stack.c:118
 check_preemption_disabled lib/smp_processor_id.c:47 [inline]
 debug_smp_processor_id.cold+0x88/0x9b lib/smp_processor_id.c:57
 dst_cache_get+0x3a/0xb0 net/core/dst_cache.c:68
 tipc_udp_xmit.isra.0+0xb9/0xad0 net/tipc/udp_media.c:164
 tipc_udp_send_msg+0x3e6/0x490 net/tipc/udp_media.c:244
 tipc_bearer_xmit_skb+0x1de/0x3f0 net/tipc/bearer.c:526
 tipc_enable_bearer+0xb2f/0xd60 net/tipc/bearer.c:331
 __tipc_nl_bearer_enable+0x2bf/0x390 net/tipc/bearer.c:995
 tipc_nl_bearer_enable+0x1e/0x30 net/tipc/bearer.c:1003
 genl_family_rcv_msg_doit net/netlink/genetlink.c:673 [inline]
 genl_family_rcv_msg net/netlink/genetlink.c:718 [inline]
 genl_rcv_msg+0x627/0xdf0 net/netlink/genetlink.c:735
 netlink_rcv_skb+0x15a/0x410 net/netlink/af_netlink.c:2469
 genl_rcv+0x24/0x40 net/netlink/genetlink.c:746
 netlink_unicast_kernel net/netlink/af_netlink.c:1303 [inline]
 netlink_unicast+0x537/0x740 net/netlink/af_netlink.c:1329
 netlink_sendmsg+0x882/0xe10 net/netlink/af_netlink.c:1918
 sock_sendmsg_nosec net/socket.c:652 [inline]
 sock_sendmsg+0xcf/0x120 net/socket.c:672
 ____sys_sendmsg+0x6bf/0x7e0 net/socket.c:2362
 ___sys_sendmsg+0x100/0x170 net/socket.c:2416
 __sys_sendmsg+0xec/0x1b0 net/socket.c:2449
 do_syscall_64+0xf6/0x7d0 arch/x86/entry/common.c:295
 entry_SYSCALL_64_after_hwframe+0x49/0xb3
RIP: 0033:0x45ca29

Fixes: e9c1a793210f ("tipc: add dst_cache support for udp media")
Cc: Xin Long <lucien.xin@gmail.com>
Cc: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mvpp2: fix RX hashing for non-10G ports
Russell King [Wed, 20 May 2020 11:26:35 +0000 (12:26 +0100)]
net: mvpp2: fix RX hashing for non-10G ports

When rxhash is enabled on any ethernet port except the first in each CP
block, traffic flow is prevented.  The analysis is below:

I've been investigating this afternoon, and what I've found, comparing
a kernel without 895586d5dc32 and with 895586d5dc32 applied is:

- The table programmed into the hardware via mvpp22_rss_fill_table()
  appears to be identical with or without the commit.

- When rxhash is enabled on eth2, mvpp2_rss_port_c2_enable() reports
  that c2.attr[0] and c2.attr[2] are written back containing:

   - with 895586d5dc32, failing:    00200000 40000000
   - without 895586d5dc32, working: 04000000 40000000

- When disabling rxhash, c2.attr[0] and c2.attr[2] are written back as:

   04000000 00000000

The second value represents the MVPP22_CLS_C2_ATTR2_RSS_EN bit, the
first value is the queue number, which comprises two fields. The high
5 bits are 24:29 and the low three are 21:23 inclusive. This comes
from:

       c2.attr[0] = MVPP22_CLS_C2_ATTR0_QHIGH(qh) |
                     MVPP22_CLS_C2_ATTR0_QLOW(ql);

So, the working case gives eth2 a queue id of 4.0, or 32 as per
port->first_rxq, and the non-working case a queue id of 0.1, or 1.
The allocation of queue IDs seems to be in mvpp2_port_probe():

        if (priv->hw_version == MVPP21)
                port->first_rxq = port->id * port->nrxqs;
        else
                port->first_rxq = port->id * priv->max_port_rxqs;

Where:

        if (priv->hw_version == MVPP21)
                priv->max_port_rxqs = 8;
        else
                priv->max_port_rxqs = 32;

Making the port 0 (eth0 / eth1) have port->first_rxq = 0, and port 1
(eth2) be 32. It seems the idea is that the first 32 queues belong to
port 0, the second 32 queues belong to port 1, etc.

mvpp2_rss_port_c2_enable() gets the queue number from it's parameter,
'ctx', which comes from mvpp22_rss_ctx(port, 0). This returns
port->rss_ctx[0].

mvpp22_rss_context_create() is responsible for allocating that, which
it does by looking for an unallocated priv->rss_tables[] pointer. This
table is shared amongst all ports on the CP silicon.

When we write the tables in mvpp22_rss_fill_table(), the RSS table
entry is defined by:

                u32 sel = MVPP22_RSS_INDEX_TABLE(rss_ctx) |
                          MVPP22_RSS_INDEX_TABLE_ENTRY(i);

where rss_ctx is the context ID (queue number) and i is the index in
the table.

If we look at what is written:

- The first table to be written has "sel" values of 00000000..0000001f,
  containing values 0..3. This appears to be for eth1. This is table 0,
  RX queue number 0.
- The second table has "sel" values of 00000100..0000011f, and appears
  to be for eth2.  These contain values 0x20..0x23. This is table 1,
  RX queue number 0.
- The third table has "sel" values of 00000200..0000021f, and appears
  to be for eth3.  These contain values 0x40..0x43. This is table 2,
  RX queue number 0.

How do queue numbers translate to the RSS table?  There is another
table - the RXQ2RSS table, indexed by the MVPP22_RSS_INDEX_QUEUE field
of MVPP22_RSS_INDEX and accessed through the MVPP22_RXQ2RSS_TABLE
register. Before 895586d5dc32, it was:

       mvpp2_write(priv, MVPP22_RSS_INDEX,
                   MVPP22_RSS_INDEX_QUEUE(port->first_rxq));
       mvpp2_write(priv, MVPP22_RXQ2RSS_TABLE,
                   MVPP22_RSS_TABLE_POINTER(port->id));

and after:

       mvpp2_write(priv, MVPP22_RSS_INDEX, MVPP22_RSS_INDEX_QUEUE(ctx));
       mvpp2_write(priv, MVPP22_RXQ2RSS_TABLE, MVPP22_RSS_TABLE_POINTER(ctx));

Before the commit, for eth2, that would've contained '32' for the
index and '1' for the table pointer - mapping queue 32 to table 1.
Remember that this is queue-high.queue-low of 4.0.

After the commit, we appear to map queue 1 to table 1. That again
looks fine on the face of it.

Section 9.3.1 of the A8040 manual seems indicate the reason that the
queue number is separated. queue-low seems to always come from the
classifier, whereas queue-high can be from the ingress physical port
number or the classifier depending on the MVPP2_CLS_SWFWD_PCTRL_REG.

We set the port bit in MVPP2_CLS_SWFWD_PCTRL_REG, meaning that queue-high
comes from the MVPP2_CLS_SWFWD_P2HQ_REG() register... and this seems to
be where our bug comes from.

mvpp2_cls_oversize_rxq_set() sets this up as:

        mvpp2_write(port->priv, MVPP2_CLS_SWFWD_P2HQ_REG(port->id),
                    (port->first_rxq >> MVPP2_CLS_OVERSIZE_RXQ_LOW_BITS));

        val = mvpp2_read(port->priv, MVPP2_CLS_SWFWD_PCTRL_REG);
        val |= MVPP2_CLS_SWFWD_PCTRL_MASK(port->id);
        mvpp2_write(port->priv, MVPP2_CLS_SWFWD_PCTRL_REG, val);

Setting the MVPP2_CLS_SWFWD_PCTRL_MASK bit means that the queue-high
for eth2 is _always_ 4, so only queues 32 through 39 inclusive are
available to eth2. Yet, we're trying to tell the classifier to set
queue-high, which will be ignored, to zero. Hence, the queue-high
field (MVPP22_CLS_C2_ATTR0_QHIGH()) from the classifier will be
ignored.

This means we end up directing traffic from eth2 not to queue 1, but
to queue 33, and then we tell it to look up queue 33 in the RSS table.
However, RSS table has not been programmed for queue 33, and so it ends
up (presumably) dropping the packets.

It seems that mvpp22_rss_context_create() doesn't take account of the
fact that the upper 5 bits of the queue ID can't actually be changed
due to the settings in mvpp2_cls_oversize_rxq_set(), _or_ it seems that
mvpp2_cls_oversize_rxq_set() has been missed in this commit. Either
way, these two functions mutually disagree with what queue number
should be used.

Looking deeper into what mvpp2_cls_oversize_rxq_set() and the MTU
validation is doing, it seems that MVPP2_CLS_SWFWD_P2HQ_REG() is used
for over-sized packets attempting to egress through this port. With
the classifier having had RSS enabled and directing eth2 traffic to
queue 1, we may still have packets appearing on queue 32 for this port.

However, the only way we may end up with over-sized packets attempting
to egress through eth2 - is if the A8040 forwards frames between its
ports. From what I can see, we don't support that feature, and the
kernel restricts the egress packet size to the MTU. In any case, if we
were to attempt to transmit an oversized packet, we have no support in
the kernel to deal with that appearing in the port's receive queue.

So, this patch attempts to solve the issue by clearing the
MVPP2_CLS_SWFWD_PCTRL_MASK() bit, allowing MVPP22_CLS_C2_ATTR0_QHIGH()
from the classifier to define the queue-high field of the queue number.

My testing seems to confirm my findings above - clearing this bit
means that if I enable rxhash on eth2, the interface can then pass
traffic, as we are now directing traffic to RX queue 1 rather than
queue 33. Traffic still seems to work with rxhash off as well.

Reported-by: Matteo Croce <mcroce@redhat.com>
Tested-by: Matteo Croce <mcroce@redhat.com>
Fixes: 895586d5dc32 ("net: mvpp2: cls: Use RSS contexts to handle RSS tables")
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>