[PATCH] - UV fix for size of hub mappings

[PATCH] - UV fix for size of hub mappings

[Jack Steiner]

Fix the size of the mappings of UV hub registers. Size must
be a function of the maximum node number within the SSI.


Signed-off-by: Jack Steiner <steiner@sgi.com>

---
 arch/x86/kernel/genx2apic_uv_x.c |   13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)

Index: linux/arch/x86/kernel/genx2apic_uv_x.c
===================================================================
--- linux.orig/arch/x86/kernel/genx2apic_uv_x.c	2008-08-17 01:21:43.000000000 -0500
+++ linux/arch/x86/kernel/genx2apic_uv_x.c	2008-08-17 07:24:38.000000000 -0500
@@ -286,12 +286,13 @@ static __init void map_low_mmrs(void)
 
 enum map_type {map_wb, map_uc};
 
-static __init void map_high(char *id, unsigned long base, int shift, enum map_type map_type)
+static __init void map_high(char *id, unsigned long base, int shift,
+			    int max_pnode, enum map_type map_type)
 {
 	unsigned long bytes, paddr;
 
 	paddr = base << shift;
-	bytes = (1UL << shift);
+	bytes = (1UL << shift) * (max_pnode + 1);
 	printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
 	       					paddr + bytes);
 	if (map_type == map_uc)
@@ -307,7 +308,7 @@ static __init void map_gru_high(int max_
 
 	gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
 	if (gru.s.enable)
-		map_high("GRU", gru.s.base, shift, map_wb);
+		map_high("GRU", gru.s.base, shift, max_pnode, map_wb);
 }
 
 static __init void map_config_high(int max_pnode)
@@ -317,7 +318,7 @@ static __init void map_config_high(int m
 
 	cfg.v = uv_read_local_mmr(UVH_RH_GAM_CFG_OVERLAY_CONFIG_MMR);
 	if (cfg.s.enable)
-		map_high("CONFIG", cfg.s.base, shift, map_uc);
+		map_high("CONFIG", cfg.s.base, shift, max_pnode, map_uc);
 }
 
 static __init void map_mmr_high(int max_pnode)
@@ -327,7 +328,7 @@ static __init void map_mmr_high(int max_
 
 	mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
 	if (mmr.s.enable)
-		map_high("MMR", mmr.s.base, shift, map_uc);
+		map_high("MMR", mmr.s.base, shift, max_pnode, map_uc);
 }
 
 static __init void map_mmioh_high(int max_pnode)
@@ -337,7 +338,7 @@ static __init void map_mmioh_high(int ma
 
 	mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
 	if (mmioh.s.enable)
-		map_high("MMIOH", mmioh.s.base, shift, map_uc);
+		map_high("MMIOH", mmioh.s.base, shift, max_pnode, map_uc);
 }
 
 static __init void uv_rtc_init(void)

Re: [PATCH] - UV fix for size of hub mappings

[Ingo Molnar]

* Jack Steiner <steiner@sgi.com> wrote:

> Fix the size of the mappings of UV hub registers. Size must be a 
> function of the maximum node number within the SSI.
> 
> Signed-off-by: Jack Steiner <steiner@sgi.com>

applied to tip/x86/uv, thanks Jack!

i'm wondering, the code still has a couple of ZZZ fixmes like:

 #ifdef ZZZ              /* Needs x2apic patch */
 static void uv_send_IPI_self(int vector)
 {
         apic_write(APIC_SELF_IPI, vector);
 }
 #endif

x2apic support is available in tip/master, so it would be nice to glue 
UV to generic x2apic support properly and remove duplication and fixmes.
For example, do we really need apic_x2apic_uv_x, or could we use 
apic_x2apic_cluster?

	Ingo

Re: [PATCH] - UV fix for size of hub mappings

[Jack Steiner]

On Sat, Sep 27, 2008 at 07:42:33PM +0200, Ingo Molnar wrote:
> 
> * Jack Steiner <steiner@sgi.com> wrote:
> 
> > Fix the size of the mappings of UV hub registers. Size must be a 
> > function of the maximum node number within the SSI.
> > 
> > Signed-off-by: Jack Steiner <steiner@sgi.com>
> 
> applied to tip/x86/uv, thanks Jack!
> 
> i'm wondering, the code still has a couple of ZZZ fixmes like:
> 
>  #ifdef ZZZ              /* Needs x2apic patch */
>  static void uv_send_IPI_self(int vector)
>  {
>          apic_write(APIC_SELF_IPI, vector);
>  }
>  #endif
> 
> x2apic support is available in tip/master, so it would be nice to glue 
> UV to generic x2apic support properly and remove duplication and fixmes.

Good point. I'll send the fixes to you early next week. Recently,
I've been focused on a distro release which is still waiting for
the x2apic code. However, I don't need to wait to fix this upstream...


> For example, do we really need apic_x2apic_uv_x, or could we use 
> apic_x2apic_cluster?

Unfortunately, we really need a uv genapic. IPIs on large UV systems
can't use an x2apic model. Because of the size of our large systems, APICIDs
in the cpu MSRs are not globally unique. We have a special feature
in the UV chipset that must be used for on large systems for IPIs.
Small & medium sized system can and do use the x2apic genapic.

--- jack

Re: [PATCH] - UV fix for size of hub mappings

[Ingo Molnar]

* Jack Steiner <steiner@sgi.com> wrote:

> > x2apic support is available in tip/master, so it would be nice to 
> > glue UV to generic x2apic support properly and remove duplication 
> > and fixmes.
> 
> Good point. I'll send the fixes to you early next week. Recently, I've 
> been focused on a distro release which is still waiting for the x2apic 
> code. However, I don't need to wait to fix this upstream...

great!

> > For example, do we really need apic_x2apic_uv_x, or could we use 
> > apic_x2apic_cluster?
> 
> Unfortunately, we really need a uv genapic. IPIs on large UV systems 
> can't use an x2apic model. Because of the size of our large systems, 
> APICIDs in the cpu MSRs are not globally unique. We have a special 
> feature in the UV chipset that must be used for on large systems for 
> IPIs. Small & medium sized system can and do use the x2apic genapic.

not a problem really - it's not really complex or intrusive, we just 
want to make sure all unnecessary duplication is eliminated.

	Ingo