| Message ID | 20171129094636.18638-2-daniel.vetter@ffwll.ch (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Quoting Daniel Vetter (2017-11-29 09:46:36) > cross-release ftl > > From Chris: > > "Fwiw, this isn't cross-release but us reloading the module many times, > creating a whole host of new lockclasses. Even more fun is when the > module gets a slightly different address and the new lock address hashes > into an old lock... > > "I did think about a module-hook to revoke the stale lockclasses, but > that still leaves all the hashed chains. > > "This particular nuisance was temporarily pushed back by teaching igt not > to reload i915.ko on a whim." > > Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> > Cc: Marta Lofstedt <marta.lofstedt@intel.com> > References: https://bugs.freedesktop.org/show_bug.cgi?id=103707 > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> Yes, I think we need to carry this for CI. cross-release will add many more chains, and since we already exhausted the previous array we will need more (and just reducing the number of module reloads only gives a temporary respite). Doubling the array doesn't seem like it'll buy us much time though? Could we afford 20 bits? Might we not also need to expand the lockclasses array? That we could just double? In principle acked-by for core-for-CI, I think we need a bit surer ground before saying that in general lockdep needs a larger array. -Chris
Quoting Chris Wilson (2017-11-29 09:57:14) > Quoting Daniel Vetter (2017-11-29 09:46:36) > > cross-release ftl > > > > From Chris: > > > > "Fwiw, this isn't cross-release but us reloading the module many times, > > creating a whole host of new lockclasses. Even more fun is when the > > module gets a slightly different address and the new lock address hashes > > into an old lock... > > > > "I did think about a module-hook to revoke the stale lockclasses, but > > that still leaves all the hashed chains. > > > > "This particular nuisance was temporarily pushed back by teaching igt not > > to reload i915.ko on a whim." > > > > Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> > > Cc: Marta Lofstedt <marta.lofstedt@intel.com> > > References: https://bugs.freedesktop.org/show_bug.cgi?id=103707 > > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> > > In principle acked-by for core-for-CI, I think we need a bit surer > ground before saying that in general lockdep needs a larger array. For upstreaming, I wonder if we could sell them on a kconfig? That way Tomi could adjust it more easily in his kconfig.git than having us provide a patch. -Chris
On Wed, Nov 29, 2017 at 10:01:09AM +0000, Chris Wilson wrote: > Quoting Chris Wilson (2017-11-29 09:57:14) > > Quoting Daniel Vetter (2017-11-29 09:46:36) > > > cross-release ftl > > > > > > From Chris: > > > > > > "Fwiw, this isn't cross-release but us reloading the module many times, > > > creating a whole host of new lockclasses. Even more fun is when the > > > module gets a slightly different address and the new lock address hashes > > > into an old lock... > > > > > > "I did think about a module-hook to revoke the stale lockclasses, but > > > that still leaves all the hashed chains. > > > > > > "This particular nuisance was temporarily pushed back by teaching igt not > > > to reload i915.ko on a whim." > > > > > > Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> > > > Cc: Marta Lofstedt <marta.lofstedt@intel.com> > > > References: https://bugs.freedesktop.org/show_bug.cgi?id=103707 > > > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> > > > > In principle acked-by for core-for-CI, I think we need a bit surer > > ground before saying that in general lockdep needs a larger array. > > For upstreaming, I wonder if we could sell them on a kconfig? That way > Tomi could adjust it more easily in his kconfig.git than having us > provide a patch. I'll submit both (once CI is taken care of) to lockdep folks and hear what they think about this issue. Maybe fixing module reload is already on their plans, maybe not. Same really for the kthread fix, lockdep code (and maintainers) are funky enough that I don't really dwell too much in there ... -Daniel
diff --git a/kernel/locking/lockdep_internals.h b/kernel/locking/lockdep_internals.h index d459d624ba2a..41630a5385c6 100644 --- a/kernel/locking/lockdep_internals.h +++ b/kernel/locking/lockdep_internals.h @@ -69,7 +69,7 @@ enum { #else #define MAX_LOCKDEP_ENTRIES 32768UL -#define MAX_LOCKDEP_CHAINS_BITS 16 +#define MAX_LOCKDEP_CHAINS_BITS 17 /* * Stack-trace: tightly packed array of stack backtrace
cross-release ftl From Chris: "Fwiw, this isn't cross-release but us reloading the module many times, creating a whole host of new lockclasses. Even more fun is when the module gets a slightly different address and the new lock address hashes into an old lock... "I did think about a module-hook to revoke the stale lockclasses, but that still leaves all the hashed chains. "This particular nuisance was temporarily pushed back by teaching igt not to reload i915.ko on a whim." Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Marta Lofstedt <marta.lofstedt@intel.com> References: https://bugs.freedesktop.org/show_bug.cgi?id=103707 Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> --- kernel/locking/lockdep_internals.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)