| Message ID | 1383220962-26836-1-git-send-email-treding@nvidia.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 10/31/2013 06:02 AM, Thierry Reding wrote: > Commit a336ed7 (clk: Implement clk_unregister()) initializes the kref in > clk_set_parent(), which is obviously the wrong place. Further research > shows that the original patches initialized it correctly, so it probably > ended up in clk_set_parent() by mistake during manual application of the > patch. Tested-by: Stephen Warren <swarren@nvidia.com> BTW, it'd be nice to Cc fixes like this to linux-next@vger.kernel.org; I /might/ have avoided doing a bisect if I'd seen this patch first! I see the benefit of that "linux-next plus today's accumulated bug-fixes" tree that I think you proposed:-)
On Thu, Oct 31, 2013 at 10:02 AM, Thierry Reding <thierry.reding@gmail.com> wrote: > Commit a336ed7 (clk: Implement clk_unregister()) initializes the kref in > clk_set_parent(), which is obviously the wrong place. Further research > shows that the original patches initialized it correctly, so it probably > ended up in clk_set_parent() by mistake during manual application of the > patch. > > Signed-off-by: Thierry Reding <treding@nvidia.com> Tested-by: Fabio Estevam <fabio.estevam@freescale.com>
On Thu, Oct 31, 2013 at 01:09:11PM -0600, Stephen Warren wrote: > On 10/31/2013 06:02 AM, Thierry Reding wrote: > > Commit a336ed7 (clk: Implement clk_unregister()) initializes the kref in > > clk_set_parent(), which is obviously the wrong place. Further research > > shows that the original patches initialized it correctly, so it probably > > ended up in clk_set_parent() by mistake during manual application of the > > patch. > > Tested-by: Stephen Warren <swarren@nvidia.com> > > BTW, it'd be nice to Cc fixes like this to linux-next@vger.kernel.org; Yes, perhaps that would've been a good idea. > I /might/ have avoided doing a bisect if I'd seen this patch first! I get that bisect is a really nice tool. But I don't understand why people seem to rely on it to track down *everything* nowadays. In this particular case there was a fairly obvious warning that pretty clearly pointed at something wrong with the reference counting and some simple code inspection revealed the issue at hand. No need to rebuild and boot the kernel dozens of times to find this out. But perhaps other people have much faster machines and bisection is actually faster... > I see the benefit of that "linux-next plus today's accumulated > bug-fixes" tree that I think you proposed:-) Yeah, this is precisely the situation where this would be good to have. Both of these issues together took about 45-60 minutes to track down and fix. I suppose it took Olof and you a similar amount of time. Yet if the fixes were already collected in some standard location it would free you up to do something more productive instead of wasting your time on duplicate work. I'll ask Stephen (Rothwell) if he'd be willing to set up shared access to linux-next so that I can push collected fixes. Alternatively I could do that in a separate repository. Thierry
On 31 October 2013 13:02, Thierry Reding <thierry.reding@gmail.com> wrote: > Commit a336ed7 (clk: Implement clk_unregister()) initializes the kref in > clk_set_parent(), which is obviously the wrong place. Further research > shows that the original patches initialized it correctly, so it probably > ended up in clk_set_parent() by mistake during manual application of the > patch. Tested-by: Jonas Jensen <jonas.jensen@gmail.com> Thanks!
diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c index 6e3f485..b68a999 100644 --- a/drivers/clk/clk.c +++ b/drivers/clk/clk.c @@ -1750,7 +1750,6 @@ int clk_set_parent(struct clk *clk, struct clk *parent) else __clk_recalc_rates(clk, POST_RATE_CHANGE); - kref_init(&clk->ref); out: clk_prepare_unlock(); @@ -1906,6 +1905,7 @@ int __clk_init(struct device *dev, struct clk *clk) if (clk->ops->init) clk->ops->init(clk->hw); + kref_init(&clk->ref); out: clk_prepare_unlock();
Commit a336ed7 (clk: Implement clk_unregister()) initializes the kref in clk_set_parent(), which is obviously the wrong place. Further research shows that the original patches initialized it correctly, so it probably ended up in clk_set_parent() by mistake during manual application of the patch. Signed-off-by: Thierry Reding <treding@nvidia.com> --- drivers/clk/clk.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)