Skip to content

L
linux
Switch branch/tag
  1. 22 Oct, 2021 3 commits
  3. 21 Oct, 2021 2 commits
    • Kai Vehmanen's avatar
      component: do not leave master devres group open after bind · c87761db
      Kai Vehmanen authored 
      In current code, the devres group for aggregate master is left open
after call to component_master_add_*(). This leads to problems when the
master does further managed allocations on its own. When any
participating driver calls component_del(), this leads to immediate
release of resources.

This came up when investigating a page fault occurring with i915 DRM
driver unbind with 5.15-rc1 kernel. The following sequence occurs:

 i915_pci_remove()
   -> intel_display_driver_unregister()
     -> i915_audio_component_cleanup()
       -> component_del()
         -> component.c:take_down_master()
           -> hdac_component_master_unbind() [via master->ops->unbind()]
           -> devres_release_group(master->parent, NULL)

With older kernels this has not caused issues, but with audio driver
moving to use managed interfaces for more of its allocations, this no
longer works. Devres log shows following to occur:

component_master_add_with_match()
[  126.886032] snd_hda_intel 0000:00:1f.3: DEVRES ADD 00000000323ccdc5 devm_component_match_release (24 bytes)
[  126.886045] snd_hda_intel 0000:00:1f.3: DEVRES ADD 00000000865cdb29 grp< (0 bytes)
[  126.886049] snd_hda_intel 0000:00:1f.3: DEVRES ADD 000000001b480725 grp< (0 bytes)

audio driver completes its PCI probe()
[  126.892238] snd_hda_intel 0000:00:1f.3: DEVRES ADD 000000001b480725 pcim_iomap_release (48 bytes)

component_del() called() at DRM/i915 unbind()
[  137.579422] i915 0000:00:02.0: DEVRES REL 00000000ef44c293 grp< (0 bytes)
[  137.579445] snd_hda_intel 0000:00:1f.3: DEVRES REL 00000000865cdb29 grp< (0 bytes)
[  137.579458] snd_hda_intel 0000:00:1f.3: DEVRES REL 000000001b480725 pcim_iomap_release (48 bytes)

So the "devres_release_group(master->parent, NULL)" ends up freeing the
pcim_iomap allocation. Upon next runtime resume, the audio driver will
cause a page fault as the iomap alloc was released without the driver
knowing about it.

Fix this issue by using the "struct master" pointer as identifier for
the devres group, and by closing the devres group after
the master->ops->bind() call is done. This allows devres allocations
done by the driver acting as master to be isolated from the binding state
of the aggregate driver. This modifies the logic originally introduced in
commit 9e1ccb4a ("drivers/base: fix devres handling for master device")

Fixes: 9e1ccb4a ("drivers/base: fix devres handling for master device")
Cc: stable@vger.kernel.org
Acked-by: default avatarImre Deak <imre.deak@intel.com>
Acked-by: default avatarRussell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: default avatarKai Vehmanen <kai.vehmanen@linux.intel.com>
BugLink: https://gitlab.freedesktop.org/drm/intel/-/issues/4136
Link: https://lore.kernel.org/r/20211013161345.3755341-1-kai.vehmanen@linux.intel.comSigned-off-by: default avatarGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      c87761db
    • Jim Cromie's avatar
      dyndbg: refine verbosity 1-4 summary-detail · 09ee10ff
      Jim Cromie authored 
      adjust current v*pr_info() calls to fit an overview..detail scheme:

1- module level activity: add/remove, etc
2- command ingest, splitting, summary of effects.
   per >control write
3- command parsing: op, flags, search terms
4- per-site change msg
   can yield ~3k x 2 logs per echo "+p;-p" > command.

Summarize these 4 levels in MODULE_PARM_DESC, and update verbose=3 in Doc.

2- is new, to isolate a problem where a stress-test script (which
feeds ~4kb multi-command strings) would produce short writes,
truncating last command and causing parsing errors, which confused
test results.  The script fix was to use syswrite, to deliver full
proper commands.

4- gets per-callsite "changed:" pr-infos, which are very noisy during
stress tests, and formerly obscured v1-3 messages, and overwhelmed the
static-key workload being tested.

The verbose parameter has previously seen adjustment:
commit 481c0e33 ("dyndbg: refine debug verbosity; 1 is basic, 2 more chatty")

The script driving these adjustments is:

 !/usr/bin/perl -w

=for Doc

1st purpose was to benchmark the effect of wildcard queries on query
performance; if wildcards are risk free cheap enough, we can deploy
them in the (floating) format search.  1st finding: wildcards take 2x
as long to process.

2nd purpose was to benchmark real static-key changes VS simple flag
changes.  Found ~100x decrease for the hard work.

The script maximizes workload per >control by packing it a ~4kb
string of "+p; -p;" commands; this uncovered some broken stuff.

The 85th query failed, and appears to be truncated, so is gramatically
incorrect.  Its either an error here, or in the kernel.  Its not
happening atm, retest.

Plot thickens: fail only happens doing +-p, not +-mf, likely load
dependent.  Error remains consistent.  Looks like a short write,
longer on writer than kernel-reader.  Try syswrite on handle to
control this.  That fixed short write.

=cut

use Getopt::Std;

getopts('vN:k:', \my %opts) or die <<EOH;
$0 options:
    -v		verbose
    -k=n	kernel dyndbg verbosity
    -N=n	number of loops.. tbrc
EOH
$opts{N} //= 10; # !undef, 0 tests too long.

my $ctrl = '/proc/dynamic_debug/control';

vx($opts{k}) if defined $opts{k}; # works on -k0

open(my $CTL, '>', $ctrl) or die "cant open $ctrl for writing: $!\n";

sub vx {
    my $arg = shift;
    my $cmd = "echo $arg > /sys/module/dynamic_debug/parameters/verbose";
    system($cmd);
    warn("vx problem: rc:$? err:$! qry: $cmd\n") if ($?);
}

sub qryOK {
    my $qry = shift;

    print "syntax test: <\n$qry>\n" if $opts{v};
    my $bytes = syswrite $CTL, $qry;
    printf "short read: $bytes / %d\n", length $qry if $bytes < length $qry;
    if ($?) {
	warn "rc:$? err:$! qry: $qry\n";
	return 0;
    }
    return 1;
}

sub build_queries {
    my ($cmd, $flags, $ct) = @_;

    # build experiment and reference queries

    my $cycle = " $cmd +$flags # on ; $cmd -$flags # off \n";
    my $ref   = " +$flags ; -$flags \n";

    my $len = length $cycle;
    my $max = int(4096 / $len); # break/fit to buffer size
    $ct |= $max;
    print "qry: ct:$max x << \n$cycle >>\n";

    return unless qryOK($ref);
    return unless qryOK($cycle);

    my $wild = $cycle x $ct;
    my $empty = $ref x $ct;

    printf "len: %d, %d\n", length $wild, length $empty;

    return { trial => $wild,
	     ref => $empty,
	     probe => $cycle,
	     zero => $ref,
	     count => $ct,
	     max => $max
    };
}

my $query_set = build_queries(' file "*" module "*" func "*" ', "mf");

qryOK($query_set->{zero});
qryOK($query_set->{probe});

qryOK($query_set->{ref});
qryOK($query_set->{trial});

use Benchmark;
sub dobatch {
    my ($cmd, $flags, $reps, $ct) = @_;
    $reps ||= $opts{N};

    my $qs = build_queries($cmd, $flags, $ct);

    timethese($reps,
	      {
		  wildcards => sub {
		      syswrite $CTL, $qs->{trial};
		  },
		  no_search => sub {
		      syswrite $CTL, $qs->{ref};
		  }
	      }
	);
}

sub bench_static_key_toggle {
    vx 0;
    dobatch(' file "*" module "*" func "*" ', "mf");
    dobatch(' file "*" module "*" func "*" ', "p");
}

sub bench_verbose_levels {
    for my $i (0..4) {
	vx $i;
	dobatch(' file "*" module "*" func "*" ', "mf");
    }
}

bench_static_key_toggle();

__END__

Heres how the test-script runs:

:: verbose=3 parsing info

[   48.401646] dyndbg: query 95: "file "*" module "*" func "*"  -mf # off " mod:*
[   48.402040] dyndbg: split into words: "file" "*" "module" "*" "func" "*" "-mf"
[   48.402456] dyndbg: op='-'
[   48.402615] dyndbg: flags=0x6
[   48.402779] dyndbg: *flagsp=0x0 *maskp=0xfffffff9
[   48.403033] dyndbg: parsed: func="*" file="*" module="*" format="" lineno=0-0
[   48.403674] dyndbg: applied: func="*" file="*" module="*" format="" lineno=0-0

:: verbose=2 >control summary.
   ~300k site matches/changes per 4kb command

[   48.404063] dyndbg: processed 96 queries, with 296160 matches, 0 errs

:: 2 queries against each other, no-search vs all-wildcard-search

qry: ct:48 x <<
  file "*" module "*" func "*"  +mf # on ;  file "*" module "*" func "*"  -mf # off
 >>
len: 4080, 576
Benchmark: timing 10 iterations of no_search, wildcards...
 no_search:  0 wallclock secs ( 0.00 usr +  0.03 sys =  0.03 CPU) @ 333.33/s (n=10)
            (warning: too few iterations for a reliable count)
 wildcards:  0 wallclock secs ( 0.00 usr +  0.09 sys =  0.09 CPU) @ 111.11/s (n=10)
            (warning: too few iterations for a reliable count)

:: 2 queries, both doing real work / changing stati-key states.

qry: ct:49 x <<
  file "*" module "*" func "*"  +p # on ;  file "*" module "*" func "*"  -p # off
 >>
len: 4067, 490
Benchmark: timing 10 iterations of no_search, wildcards...
 no_search: 20 wallclock secs ( 0.00 usr + 20.36 sys = 20.36 CPU) @  0.49/s (n=10)
 wildcards: 21 wallclock secs ( 0.00 usr + 21.08 sys = 21.08 CPU) @  0.47/s (n=10)
bash-5.1#

Thats 150k static-key-toggles / sec
  ~600x slower than simple flags
  on qemu --smp 3 run
Signed-off-by: default avatarJim Cromie <jim.cromie@gmail.com>
Link: https://lore.kernel.org/r/20211019210746.185307-1-jim.cromie@gmail.comSigned-off-by: default avatarGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      09ee10ff
  5. 20 Oct, 2021 3 commits
  7. 18 Oct, 2021 18 commits
  9. 17 Oct, 2021 3 commits
  11. 16 Oct, 2021 11 commits
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7