commit 51d3082fe6e55aecfa17113dbe98077c749f724c enabled NAP unconditinally
on all powermacs. Early G3 cpus can not use it, the result is memory corruption.
Only enable powersave_nap in one place: probe_motherboard()
ppc32 gets nap if the cpu supports it
ppc32 smp gets no nap
ppc64 gets nap unconditionally
---
arch/powerpc/platforms/powermac/feature.c | 8 +++++++-
arch/powerpc/platforms/powermac/setup.c | 4 ----
arch/powerpc/platforms/powermac/smp.c | 4 ----
3 files changed, 7 insertions(+), 9 deletions(-)
Index: linux-2.6.16-rc5-olh/arch/powerpc/platforms/powermac/feature.c
===================================================================
--- linux-2.6.16-rc5-olh.orig/arch/powerpc/platforms/powermac/feature.c
+++ linux-2.6.16-rc5-olh/arch/powerpc/platforms/powermac/feature.c
@@ -2513,8 +2513,11 @@ found:
/* Nap mode not supported if flush-on-lock property is present */
if (get_property(np, "flush-on-lock", NULL))
break;
+
+#ifndef CONFIG_SMP
+ /* 32 bits SMP can't NAP */
powersave_nap = 1;
- printk(KERN_INFO "Processor NAP mode on idle enabled.\n");
+#endif
break;
}
@@ -2526,6 +2529,9 @@ found:
#ifdef CONFIG_POWER4
powersave_nap = 1;
#endif
+ if (powersave_nap)
+ printk(KERN_INFO "Using native/NAP idle loop\n");
+
/* Check for "mobile" machine */
if (model && (strncmp(model, "PowerBook", 9) == 0
|| strncmp(model, "iBook", 5) == 0))
Index: linux-2.6.16-rc5-olh/arch/powerpc/platforms/powermac/setup.c
===================================================================
--- linux-2.6.16-rc5-olh.orig/arch/powerpc/platforms/powermac/setup.c
+++ linux-2.6.16-rc5-olh/arch/powerpc/platforms/powermac/setup.c
@@ -621,10 +621,6 @@ static void __init pmac_init_early(void)
/* Probe motherboard chipset */
pmac_feature_init();
- /* We can NAP */
- powersave_nap = 1;
- printk(KERN_INFO "Using native/NAP idle loop\n");
-
/* Initialize debug stuff */
udbg_scc_init(!!strstr(cmd_line, "sccdbg"));
udbg_adb_init(!!strstr(cmd_line, "btextdbg"));
Index: linux-2.6.16-rc5-olh/arch/powerpc/platforms/powermac/smp.c
===================================================================
--- linux-2.6.16-rc5-olh.orig/arch/powerpc/platforms/powermac/smp.c
+++ linux-2.6.16-rc5-olh/arch/powerpc/platforms/powermac/smp.c
@@ -739,10 +739,6 @@ static void __init smp_core99_setup(int
smp_hw_index[i] = i;
}
#endif
-
- /* 32 bits SMP can't NAP */
- if (!machine_is_compatible("MacRISC4"))
- powersave_nap = 0;
}
static int __init smp_core99_probe(void)
On Sat, 2006-03-11 at 22:58 +0100, Olaf Hering wrote:
> commit 51d3082fe6e55aecfa17113dbe98077c749f724c enabled NAP unconditinally
> on all powermacs. Early G3 cpus can not use it, the result is memory corruption.
>
> Only enable powersave_nap in one place: probe_motherboard()
> ppc32 gets nap if the cpu supports it
> ppc32 smp gets no nap
> ppc64 gets nap unconditionally
CONFIG_SMP is certainly not the way to do it though... I wonder how we
got that wrong, I'll have a look.
Ben.
> ---
> arch/powerpc/platforms/powermac/feature.c | 8 +++++++-
> arch/powerpc/platforms/powermac/setup.c | 4 ----
> arch/powerpc/platforms/powermac/smp.c | 4 ----
> 3 files changed, 7 insertions(+), 9 deletions(-)
>
> Index: linux-2.6.16-rc5-olh/arch/powerpc/platforms/powermac/feature.c
> ===================================================================
> --- linux-2.6.16-rc5-olh.orig/arch/powerpc/platforms/powermac/feature.c
> +++ linux-2.6.16-rc5-olh/arch/powerpc/platforms/powermac/feature.c
> @@ -2513,8 +2513,11 @@ found:
> /* Nap mode not supported if flush-on-lock property is present */
> if (get_property(np, "flush-on-lock", NULL))
> break;
> +
> +#ifndef CONFIG_SMP
> + /* 32 bits SMP can't NAP */
> powersave_nap = 1;
> - printk(KERN_INFO "Processor NAP mode on idle enabled.\n");
> +#endif
> break;
> }
>
> @@ -2526,6 +2529,9 @@ found:
> #ifdef CONFIG_POWER4
> powersave_nap = 1;
> #endif
> + if (powersave_nap)
> + printk(KERN_INFO "Using native/NAP idle loop\n");
> +
> /* Check for "mobile" machine */
> if (model && (strncmp(model, "PowerBook", 9) == 0
> || strncmp(model, "iBook", 5) == 0))
> Index: linux-2.6.16-rc5-olh/arch/powerpc/platforms/powermac/setup.c
> ===================================================================
> --- linux-2.6.16-rc5-olh.orig/arch/powerpc/platforms/powermac/setup.c
> +++ linux-2.6.16-rc5-olh/arch/powerpc/platforms/powermac/setup.c
> @@ -621,10 +621,6 @@ static void __init pmac_init_early(void)
> /* Probe motherboard chipset */
> pmac_feature_init();
>
> - /* We can NAP */
> - powersave_nap = 1;
> - printk(KERN_INFO "Using native/NAP idle loop\n");
> -
> /* Initialize debug stuff */
> udbg_scc_init(!!strstr(cmd_line, "sccdbg"));
> udbg_adb_init(!!strstr(cmd_line, "btextdbg"));
> Index: linux-2.6.16-rc5-olh/arch/powerpc/platforms/powermac/smp.c
> ===================================================================
> --- linux-2.6.16-rc5-olh.orig/arch/powerpc/platforms/powermac/smp.c
> +++ linux-2.6.16-rc5-olh/arch/powerpc/platforms/powermac/smp.c
> @@ -739,10 +739,6 @@ static void __init smp_core99_setup(int
> smp_hw_index[i] = i;
> }
> #endif
> -
> - /* 32 bits SMP can't NAP */
> - if (!machine_is_compatible("MacRISC4"))
> - powersave_nap = 0;
> }
>
> static int __init smp_core99_probe(void)
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
On Sat, 2006-03-11 at 22:58 +0100, Olaf Hering wrote:
> commit 51d3082fe6e55aecfa17113dbe98077c749f724c enabled NAP unconditinally
> on all powermacs. Early G3 cpus can not use it, the result is memory corruption.
Ok, here is what I think is a proper fix: Just remove the code from
setup.c since feature.c will already set powersave_nap when needed.
Can you test asap please ? If it fixes the problem for you, I'll send
the patch to Linus/Andrew with hopes that it will make it in 2.6.16
---
This patch fixes incorrect setting of powersave_nap to 1 on all
PowerMac's potentially causing memory corruption on some models. This
bug was introuced by me during the 32/64 bits merge.
Signed-off-by: Benjamin Herrenschmidt <[email protected]>
Index: linux-work/arch/powerpc/platforms/powermac/setup.c
===================================================================
--- linux-work.orig/arch/powerpc/platforms/powermac/setup.c 2006-02-13 14:10:43.000000000 +1100
+++ linux-work/arch/powerpc/platforms/powermac/setup.c 2006-03-12 10:47:39.000000000 +1100
@@ -621,10 +621,6 @@
/* Probe motherboard chipset */
pmac_feature_init();
- /* We can NAP */
- powersave_nap = 1;
- printk(KERN_INFO "Using native/NAP idle loop\n");
-
/* Initialize debug stuff */
udbg_scc_init(!!strstr(cmd_line, "sccdbg"));
udbg_adb_init(!!strstr(cmd_line, "btextdbg"));
Index: linux-work/arch/powerpc/platforms/powermac/feature.c
===================================================================
--- linux-work.orig/arch/powerpc/platforms/powermac/feature.c 2006-03-01 11:48:27.000000000 +1100
+++ linux-work/arch/powerpc/platforms/powermac/feature.c 2006-03-12 10:51:31.000000000 +1100
@@ -2491,9 +2491,7 @@
pmac_mb.model_id = PMAC_TYPE_COMET;
iounmap(mach_id_ptr);
}
-#endif /* CONFIG_POWER4 */
-#ifdef CONFIG_6xx
/* Set default value of powersave_nap on machines that support it.
* It appears that uninorth rev 3 has a problem with it, we don't
* enable it on those. In theory, the flush-on-lock property is
@@ -2522,10 +2520,11 @@
* NAP mode
*/
powersave_lowspeed = 1;
-#endif /* CONFIG_6xx */
-#ifdef CONFIG_POWER4
+
+#else /* CONFIG_POWER4 */
powersave_nap = 1;
-#endif
+#endif /* CONFIG_POWER4 */
+
/* Check for "mobile" machine */
if (model && (strncmp(model, "PowerBook", 9) == 0
|| strncmp(model, "iBook", 5) == 0))
On Sun, Mar 12, Benjamin Herrenschmidt wrote:
> On Sat, 2006-03-11 at 22:58 +0100, Olaf Hering wrote:
> > commit 51d3082fe6e55aecfa17113dbe98077c749f724c enabled NAP unconditinally
> > on all powermacs. Early G3 cpus can not use it, the result is memory corruption.
>
> Ok, here is what I think is a proper fix: Just remove the code from
> setup.c since feature.c will already set powersave_nap when needed.
>
> Can you test asap please ? If it fixes the problem for you, I'll send
> the patch to Linus/Andrew with hopes that it will make it in 2.6.16
Yes, this helps.