lp:~shakerxy/ubuntu/+source/linux/+git/mainline-crack

Author: Linus Torvalds
Author Date: 2023-03-07 20:16:18 UTC

cpumask: be more careful with 'cpumask_setall()'

Commit 596ff4a09b89 ("cpumask: re-introduce constant-sized cpumask
optimizations") changed cpumask_setall() to use "bitmap_set()" instead
of "bitmap_fill()", because bitmap_fill() would explicitly set all the
bits of a constant sized small bitmap, and that's exactly what we don't
want: we want to only set bits up to 'nr_cpu_ids', which is what
"bitmap_set()" does.

However, Yury correctly points out that while "bitmap_set()" does indeed
only set bits up to the required bitmap size, it doesn't _clear_ bits
above that size, so the upper bits would still not have well-defined
values.

Now, none of this should really matter, since any bits set past
'nr_cpu_ids' should always be ignored in the first place. Yes, the bit
scanning functions might return them as a result, but since users should
always consider the ">= nr_cpu_ids" condition to mean "no more bits",
that shouldn't have any actual effect (see previous commit 8ca09d5fa354
"cpumask: fix incorrect cpumask scanning result checks").

But let's just do it right, the way the code was _intended_ to work. We
have had enough lazy code that works but bites us in the *rse later
(again, see previous commit) that there's no reason to not just do this
properly.

It turns out that "bitmap_fill()" gets this all right for the complex
case, and really only fails for the inlined optimized case that just
fills the whole word. And while we could just fix bitmap_fill() to use
the proper last word mask, there's two issues with that:

 - the cpumask case wants to do the _optimization_ based on "NR_CPUS is
   a small constant", but then wants to do the actual bit _fill_ based
   on "nr_cpu_ids" that isn't necessarily that same constant

 - we have lots of non-cpumask users of bitmap_fill(), and while they
   hopefully don't care, and probably would want the proper semantics
   anyway ("only set bits up to the limit"), I do not want the cpumask
   changes to impact other parts

So this ends up just doing the single-word optimization by hand in the
cpumask code. If our cpumask is fundamentally limited to a single word,
just do the proper "fill in that word" exactly. And if it's the more
complex multi-word case, then the generic bitmap_fill() will DTRT.

This is all an example of how our bitmap function optimizations really
are somewhat broken. They conflate the "this is size of the bitmap"
optimizations with the actual bit(s) we want to set.

In many cases we really want to have the two be separate things:
sometimes we base our optimizations on the size of the whole bitmap ("I
know this whole bitmap fits in a single word, so I'll just use
single-word accesses"), and sometimes we base them on the bit we are
looking at ("this is just acting on bits that are in the first word, so
I'll use single-word accesses").

Notice how the end result of the two optimizations are the same, but the
way we get to them are quite different.

And all our cpumask optimization games are really about that fundamental
distinction, and we'd often really want to pass in both the "this is the
bit I'm working on" (which _can_ be a small constant but might be
variable), and "I know it's in this range even if it's variable" (based
on CONFIG_NR_CPUS).

So this cpumask_setall() implementation just makes that explicit. It
checks the "I statically know the size is small" using the known static
size of the cpumask (which is what that 'small_cpumask_bits' is all
about), but then sets the actual bits using the exact number of cpus we
have (ie 'nr_cpumask_bits')

Of course, in a perfect world, the compiler would have done all the
range analysis (possibly with help from us just telling it that
"this value is always in this range"), and would do all of this for us.
But that is not the world we live in.

While we dream of that perfect world, this does that manual logic to
make it all work out. And this was a very long explanation for a small
code change that shouldn't even matter.

Reported-by: Yury Norov <yury.norov@gmail.com>
Link: https://lore.kernel.org/lkml/ZAV9nGG9e1%2FrV+L%2F@yury-laptop/
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Author: Linus Torvalds
Author Date: 2022-10-29 18:45:07 UTC

mm: delay rmap removal until after TLB flush

When we remove a page table entry, we are very careful to only free the
page after we have flushed the TLB, because other CPUs could still be
using the page through stale TLB entries until after the flush.

However, we have removed the rmap entry for that page early, which means
that functions like folio_mkclean() would end up not serializing with
the page table lock because the page had already been made invisible to
rmap.

And that is a problem, because while the TLB entry exists, we could end
up with the followign situation:

 (a) one CPU could come in and clean it, never seeing our mapping of
     the page

 (b) another CPU could continue to use the stale and dirty TLB entry
     and continue to write to said page

resulting in a page that has been dirtied, but then marked clean again,
all while another CPU might have dirtied it some more.

End result: possibly lost dirty data.

This commit uses the same old TLB gather array that we use to delay the
freeing of the page to also say 'remove from rmap after flush', so that
we can keep the rmap entries alive until all TLB entries have been
flushed.

NOTE! While the "possibly lost dirty data" sounds catastrophic, for this
all to happen you need to have a user thread doing either madvise() with
MADV_DONTNEED or a full re-mmap() of the area concurrently with another
thread continuing to use said mapping.

So arguably this is about user space doing crazy things, but from a VM
consistency standpoint it's better if we track the dirty bit properly
even when user space goes off the rails.

Reported-by: Nadav Amit <nadav.amit@gmail.com>
Link: Link: https://lore.kernel.org/all/B88D3073-440A-41C7-95F4-895D3F657EF2@gmail.com/
Cc: Will Deacon <will@kernel.org>
Cc: Aneesh Kumar <aneesh.kumar@linux.ibm.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Nick Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Author: Linus Torvalds
Author Date: 2021-01-21 18:17:25 UTC

tty: fix up iterate_tty_read() EOVERFLOW handling

When I converted the tty_ldisc_ops 'read()' function to take a kernel
pointer, I was a bit too aggressive about the ldisc returning EOVERFLOW.

Yes, we want to have EOVERFLOW override any partially read data (because
the whole point is that the buffer was too small for the whole packet,
and we don't want to see partial packets), but it shouldn't override a
previous EFAULT.

And in fact, it really is just EOVERFLOW that is special and should
throw away any partially read data, not "any error". Admittedly
EOVERFLOW is currently the only one that can happen for a continuation
read - and if the first read iteration returns an error we won't have this issue.

So this is more of a technicality, but let's just make the intent very
explicit, and re-organize the error handling a bit so that this is all
clearer.

Reported-by: Jiri Slaby <jirislaby@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Author: Greg Kroah-Hartman
Author Date: 2018-09-17 05:24:28 UTC

Merge tag 'linux-kselftest-4.19-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest

Pulled kselftest fixes from Shuah:
 "This Kselftest fixes update for 4.9-rc5 consists of:

  -- fixes to build failures
  -- fixes to add missing config files to increase test coverage
  -- fixes to cgroup test and a new cgroup test for memory.oom.group"

Author: Peter Zijlstra
Author Date: 2018-08-22 15:30:16 UTC

x86/mm: Only use tlb_remove_table() for paravirt

If we don't use paravirt; don't play unnecessary and complicated games
to free page-tables.

Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Rik van Riel <riel@surriel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Author: Linus Torvalds
Author Date: 2018-05-17 22:17:33 UTC

fs/proc: simplify and clarify get_mm_cmdline() function

We have some very odd semantics for reading the command line through
/proc, because we allow people to rewrite their own command line pretty
much at will, and things get positively funky when you extend your
command line past the point that used to be the end of the command line,
and is now in the environment variable area.

But our weird semantics doesn't mean that we should write weird and
complex code to handle them.

So re-write get_mm_cmdline() to be much simpler, and much more explicit
about what it is actually doing and why. And avoid the extra check for
"is there a NUL character at the end of the command line where I expect
one to be", by simply making the NUL character handling be part of the
normal "once you hit the end of the command line, stop at the first NUL
character" logic.

It's quite possible that we should stop the crazy "walk into
environment" entirely, but happily it's not really the usual case.

NOTE! We tried to really simplify and limit our odd cmdline parsing some
time ago, but people complained. See commit c2c0bb44620d ("proc: fix
PAGE_SIZE limit of /proc/$PID/cmdline") for details about why we have
this complexity.

Cc: Tejun Heo <tj@kernel.org>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Jarod Wilson <jarod@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Author: Linus Torvalds
Author Date: 2018-02-27 20:39:30 UTC

Broken, but working, ptregs system call conversion for x86-64

Not-yet-signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Author: Linus Torvalds
Author Date: 2014-06-06 23:35:10 UTC

Merge branch 'akpm' (patches from Andrew Morton) into next

Merge more updates from Andrew Morton:

 - Most of the rest of MM.

   This includes "mark remap_file_pages syscall as deprecated" but the
   actual "replace remap_file_pages syscall with emulation" is held
   back. I guess we'll need to work out when to pull the trigger on
   that one.

 - various minor cleanups to obscure filesystems

 - the drivers/rtc queue

 - hfsplus updates

 - ufs, hpfs, fatfs, affs, reiserfs

 - Documentation/

 - signals

 - procfs

 - cpu hotplug

 - lib/idr.c

 - rapidio

 - sysctl

 - ipc updates

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (171 commits)
  ufs: sb mutex merge + mutex_destroy
  powerpc: update comments for generic idle conversion
  cris: update comments for generic idle conversion
  idle: remove cpu_idle() forward declarations
  nbd: zero from and len fields in NBD_CMD_DISCONNECT.
  mm: convert some level-less printks to pr_*
  MAINTAINERS: adi-buildroot-devel is moderated
  MAINTAINERS: add linux-api for review of API/ABI changes
  mm/kmemleak-test.c: use pr_fmt for logging
  fs/dlm/debug_fs.c: replace seq_printf by seq_puts
  fs/dlm/lockspace.c: convert simple_str to kstr
  fs/dlm/config.c: convert simple_str to kstr
  mm: mark remap_file_pages() syscall as deprecated
  mm: memcontrol: remove unnecessary memcg argument from soft limit functions
  mm: memcontrol: clean up memcg zoneinfo lookup
  mm/memblock.c: call kmemleak directly from memblock_(alloc|free)
  mm/mempool.c: update the kmemleak stack trace for mempool allocations
  lib/radix-tree.c: update the kmemleak stack trace for radix tree allocations
  mm: introduce kmemleak_update_trace()
  mm/kmemleak.c: use %u to print ->checksum
  ...