* [PATCH] mips: fix execution hazard during watchpoint register probe
@ 2013-06-15 16:34 ` Paul Burton
0 siblings, 0 replies; 3+ messages in thread
From: Paul Burton @ 2013-06-15 16:34 UTC (permalink / raw)
To: linux-mips; +Cc: Paul Burton
Writing a value to a WatchLo* register creates an execution hazard, so
if its value is then read before that hazard is cleared then said value
may be invalid. The mips_probe_watch_registers function must therefore
clear the execution hazard between setting the match bits in a WatchLo*
register & reading the register back in order to check which are set.
This fixes intermittent incorrect watchpoint register probing on some
MIPS cores such as interAptiv & proAptiv.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Reviewed-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Steven J. Hill <Steven.Hill@imgtec.com>
---
arch/mips/kernel/watch.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/arch/mips/kernel/watch.c b/arch/mips/kernel/watch.c
index 7726f61..cbdc4de 100644
--- a/arch/mips/kernel/watch.c
+++ b/arch/mips/kernel/watch.c
@@ -111,6 +111,7 @@ __cpuinit void mips_probe_watch_registers(struct cpuinfo_mips *c)
* disable the register.
*/
write_c0_watchlo0(7);
+ back_to_back_c0_hazard();
t = read_c0_watchlo0();
write_c0_watchlo0(0);
c->watch_reg_masks[0] = t & 7;
@@ -121,12 +122,14 @@ __cpuinit void mips_probe_watch_registers(struct cpuinfo_mips *c)
c->watch_reg_use_cnt = 1;
t = read_c0_watchhi0();
write_c0_watchhi0(t | 0xff8);
+ back_to_back_c0_hazard();
t = read_c0_watchhi0();
c->watch_reg_masks[0] |= (t & 0xff8);
if ((t & 0x80000000) == 0)
return;
write_c0_watchlo1(7);
+ back_to_back_c0_hazard();
t = read_c0_watchlo1();
write_c0_watchlo1(0);
c->watch_reg_masks[1] = t & 7;
@@ -135,12 +138,14 @@ __cpuinit void mips_probe_watch_registers(struct cpuinfo_mips *c)
c->watch_reg_use_cnt = 2;
t = read_c0_watchhi1();
write_c0_watchhi1(t | 0xff8);
+ back_to_back_c0_hazard();
t = read_c0_watchhi1();
c->watch_reg_masks[1] |= (t & 0xff8);
if ((t & 0x80000000) == 0)
return;
write_c0_watchlo2(7);
+ back_to_back_c0_hazard();
t = read_c0_watchlo2();
write_c0_watchlo2(0);
c->watch_reg_masks[2] = t & 7;
@@ -149,12 +154,14 @@ __cpuinit void mips_probe_watch_registers(struct cpuinfo_mips *c)
c->watch_reg_use_cnt = 3;
t = read_c0_watchhi2();
write_c0_watchhi2(t | 0xff8);
+ back_to_back_c0_hazard();
t = read_c0_watchhi2();
c->watch_reg_masks[2] |= (t & 0xff8);
if ((t & 0x80000000) == 0)
return;
write_c0_watchlo3(7);
+ back_to_back_c0_hazard();
t = read_c0_watchlo3();
write_c0_watchlo3(0);
c->watch_reg_masks[3] = t & 7;
@@ -163,6 +170,7 @@ __cpuinit void mips_probe_watch_registers(struct cpuinfo_mips *c)
c->watch_reg_use_cnt = 4;
t = read_c0_watchhi3();
write_c0_watchhi3(t | 0xff8);
+ back_to_back_c0_hazard();
t = read_c0_watchhi3();
c->watch_reg_masks[3] |= (t & 0xff8);
if ((t & 0x80000000) == 0)
--
1.8.3.1
^ permalink raw reply related [flat|nested] 3+ messages in thread
* [PATCH] mips: fix execution hazard during watchpoint register probe
@ 2013-06-15 16:34 ` Paul Burton
0 siblings, 0 replies; 3+ messages in thread
From: Paul Burton @ 2013-06-15 16:34 UTC (permalink / raw)
To: linux-mips; +Cc: Paul Burton
Writing a value to a WatchLo* register creates an execution hazard, so
if its value is then read before that hazard is cleared then said value
may be invalid. The mips_probe_watch_registers function must therefore
clear the execution hazard between setting the match bits in a WatchLo*
register & reading the register back in order to check which are set.
This fixes intermittent incorrect watchpoint register probing on some
MIPS cores such as interAptiv & proAptiv.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Reviewed-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Steven J. Hill <Steven.Hill@imgtec.com>
---
arch/mips/kernel/watch.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/arch/mips/kernel/watch.c b/arch/mips/kernel/watch.c
index 7726f61..cbdc4de 100644
--- a/arch/mips/kernel/watch.c
+++ b/arch/mips/kernel/watch.c
@@ -111,6 +111,7 @@ __cpuinit void mips_probe_watch_registers(struct cpuinfo_mips *c)
* disable the register.
*/
write_c0_watchlo0(7);
+ back_to_back_c0_hazard();
t = read_c0_watchlo0();
write_c0_watchlo0(0);
c->watch_reg_masks[0] = t & 7;
@@ -121,12 +122,14 @@ __cpuinit void mips_probe_watch_registers(struct cpuinfo_mips *c)
c->watch_reg_use_cnt = 1;
t = read_c0_watchhi0();
write_c0_watchhi0(t | 0xff8);
+ back_to_back_c0_hazard();
t = read_c0_watchhi0();
c->watch_reg_masks[0] |= (t & 0xff8);
if ((t & 0x80000000) == 0)
return;
write_c0_watchlo1(7);
+ back_to_back_c0_hazard();
t = read_c0_watchlo1();
write_c0_watchlo1(0);
c->watch_reg_masks[1] = t & 7;
@@ -135,12 +138,14 @@ __cpuinit void mips_probe_watch_registers(struct cpuinfo_mips *c)
c->watch_reg_use_cnt = 2;
t = read_c0_watchhi1();
write_c0_watchhi1(t | 0xff8);
+ back_to_back_c0_hazard();
t = read_c0_watchhi1();
c->watch_reg_masks[1] |= (t & 0xff8);
if ((t & 0x80000000) == 0)
return;
write_c0_watchlo2(7);
+ back_to_back_c0_hazard();
t = read_c0_watchlo2();
write_c0_watchlo2(0);
c->watch_reg_masks[2] = t & 7;
@@ -149,12 +154,14 @@ __cpuinit void mips_probe_watch_registers(struct cpuinfo_mips *c)
c->watch_reg_use_cnt = 3;
t = read_c0_watchhi2();
write_c0_watchhi2(t | 0xff8);
+ back_to_back_c0_hazard();
t = read_c0_watchhi2();
c->watch_reg_masks[2] |= (t & 0xff8);
if ((t & 0x80000000) == 0)
return;
write_c0_watchlo3(7);
+ back_to_back_c0_hazard();
t = read_c0_watchlo3();
write_c0_watchlo3(0);
c->watch_reg_masks[3] = t & 7;
@@ -163,6 +170,7 @@ __cpuinit void mips_probe_watch_registers(struct cpuinfo_mips *c)
c->watch_reg_use_cnt = 4;
t = read_c0_watchhi3();
write_c0_watchhi3(t | 0xff8);
+ back_to_back_c0_hazard();
t = read_c0_watchhi3();
c->watch_reg_masks[3] |= (t & 0xff8);
if ((t & 0x80000000) == 0)
--
1.8.3.1
^ permalink raw reply related [flat|nested] 3+ messages in thread
* Re: [PATCH] mips: fix execution hazard during watchpoint register probe
2013-06-15 16:34 ` Paul Burton
(?)
@ 2013-06-18 14:42 ` Ralf Baechle
-1 siblings, 0 replies; 3+ messages in thread
From: Ralf Baechle @ 2013-06-18 14:42 UTC (permalink / raw)
To: Paul Burton; +Cc: linux-mips, Chris Dearman
On Sat, Jun 15, 2013 at 05:34:40PM +0100, Paul Burton wrote:
> Writing a value to a WatchLo* register creates an execution hazard, so
> if its value is then read before that hazard is cleared then said value
> may be invalid. The mips_probe_watch_registers function must therefore
> clear the execution hazard between setting the match bits in a WatchLo*
> register & reading the register back in order to check which are set.
>
> This fixes intermittent incorrect watchpoint register probing on some
> MIPS cores such as interAptiv & proAptiv.
Obviously correct. Did somebody once have a tool to test for this kind
of back-to-back cp0 issues?
Ralf
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2013-06-18 14:42 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2013-06-15 16:34 [PATCH] mips: fix execution hazard during watchpoint register probe Paul Burton
2013-06-15 16:34 ` Paul Burton
2013-06-18 14:42 ` Ralf Baechle
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.