clocksource: sh_cmt: fix clocksource width for 32-bit machines
diff mbox series

Message ID 749e6480-aed6-041c-0209-8dc74c07f344@cogentembedded.com
State Accepted
Delegated to: Geert Uytterhoeven
Headers show
Series
  • clocksource: sh_cmt: fix clocksource width for 32-bit machines
Related show

Commit Message

Sergei Shtylyov Sept. 10, 2018, 8:22 p.m. UTC
The driver seems to abuse *unsigned long* not only for the (32-bit)
register values but also for the 'sh_cmt_channel::total_cycles' which
needs to always be 64-bit -- as a result, the clocksource's mask is
needlessly clamped down to 32-bits on the 32-bit machines...

Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>

---
This patch is against the 'tip.git' repo's 'timers/core' branch plus the fixup
for the 64-bit machines reposted last Saturday...

 drivers/clocksource/sh_cmt.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

Comments

Sergei Shtylyov Sept. 10, 2018, 8:34 p.m. UTC | #1
On 09/10/2018 11:22 PM, Sergei Shtylyov wrote:

> The driver seems to abuse *unsigned long* not only for the (32-bit)
> register values but also for the 'sh_cmt_channel::total_cycles' which
> needs to always be 64-bit -- as a result, the clocksource's mask is
> needlessly clamped down to 32-bits on the 32-bit machines...

   I thought this bug was present from the start but no... 

> Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
> Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>

Fixes: 19bdc9d061bc ("clocksource: sh_cmt clocksource support")

[...]

MBR, Sergei
Simon Horman Sept. 12, 2018, 9:26 a.m. UTC | #2
On Mon, Sep 10, 2018 at 11:34:49PM +0300, Sergei Shtylyov wrote:
> On 09/10/2018 11:22 PM, Sergei Shtylyov wrote:
> 
> > The driver seems to abuse *unsigned long* not only for the (32-bit)
> > register values but also for the 'sh_cmt_channel::total_cycles' which
> > needs to always be 64-bit -- as a result, the clocksource's mask is
> > needlessly clamped down to 32-bits on the 32-bit machines...
> 
>    I thought this bug was present from the start but no... 
> 
> > Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
> > Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
> 
> Fixes: 19bdc9d061bc ("clocksource: sh_cmt clocksource support")

Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Daniel Lezcano Sept. 13, 2018, 1:25 p.m. UTC | #3
On 10/09/2018 22:22, Sergei Shtylyov wrote:
> The driver seems to abuse *unsigned long* not only for the (32-bit)
> register values but also for the 'sh_cmt_channel::total_cycles' which
> needs to always be 64-bit -- as a result, the clocksource's mask is
> needlessly clamped down to 32-bits on the 32-bit machines...
> 
> Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
> Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
> 
> ---
> This patch is against the 'tip.git' repo's 'timers/core' branch plus the fixup
> for the 64-bit machines reposted last Saturday...

Why not against timers/urgent ?
Sergei Shtylyov Sept. 13, 2018, 4:26 p.m. UTC | #4
Hello!

On 09/13/2018 04:25 PM, Daniel Lezcano wrote:

>> The driver seems to abuse *unsigned long* not only for the (32-bit)
>> register values but also for the 'sh_cmt_channel::total_cycles' which
>> needs to always be 64-bit -- as a result, the clocksource's mask is
>> needlessly clamped down to 32-bits on the 32-bit machines...
>>
>> Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
>> Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
>>
>> ---
>> This patch is against the 'tip.git' repo's 'timers/core' branch plus the fixup
>> for the 64-bit machines reposted last Saturday...
> 
> Why not against timers/urgent ?

   Mhm... nothing blows up because of this apparently (the bug is quite old).

MBR, Sergei
Geert Uytterhoeven Sept. 14, 2018, 11:55 a.m. UTC | #5
On Mon, Sep 10, 2018 at 10:23 PM Sergei Shtylyov
<sergei.shtylyov@cogentembedded.com> wrote:
> The driver seems to abuse *unsigned long* not only for the (32-bit)
> register values but also for the 'sh_cmt_channel::total_cycles' which
> needs to always be 64-bit -- as a result, the clocksource's mask is
> needlessly clamped down to 32-bits on the 32-bit machines...
>
> Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
> Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>

Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>

Gr{oetje,eeting}s,

                        Geert
Daniel Lezcano Sept. 14, 2018, 2:33 p.m. UTC | #6
On 10/09/2018 22:22, Sergei Shtylyov wrote:
> The driver seems to abuse *unsigned long* not only for the (32-bit)
> register values but also for the 'sh_cmt_channel::total_cycles' which
> needs to always be 64-bit -- as a result, the clocksource's mask is
> needlessly clamped down to 32-bits on the 32-bit machines...
> 
> Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
> Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
> 
> ---

Applied for 4.20, thanks.

Patch
diff mbox series

Index: tip/drivers/clocksource/sh_cmt.c
===================================================================
--- tip.orig/drivers/clocksource/sh_cmt.c
+++ tip/drivers/clocksource/sh_cmt.c
@@ -108,7 +108,7 @@  struct sh_cmt_channel {
 	raw_spinlock_t lock;
 	struct clock_event_device ced;
 	struct clocksource cs;
-	unsigned long total_cycles;
+	u64 total_cycles;
 	bool cs_enabled;
 };
 
@@ -613,8 +613,8 @@  static u64 sh_cmt_clocksource_read(struc
 {
 	struct sh_cmt_channel *ch = cs_to_sh_cmt(cs);
 	unsigned long flags;
-	unsigned long value;
 	u32 has_wrapped;
+	u64 value;
 	u32 raw;
 
 	raw_spin_lock_irqsave(&ch->lock, flags);
@@ -688,7 +688,7 @@  static int sh_cmt_register_clocksource(s
 	cs->disable = sh_cmt_clocksource_disable;
 	cs->suspend = sh_cmt_clocksource_suspend;
 	cs->resume = sh_cmt_clocksource_resume;
-	cs->mask = CLOCKSOURCE_MASK(sizeof(unsigned long) * 8);
+	cs->mask = CLOCKSOURCE_MASK(sizeof(u64) * 8);
 	cs->flags = CLOCK_SOURCE_IS_CONTINUOUS;
 
 	dev_info(&ch->cmt->pdev->dev, "ch%u: used as clock source\n",