Skip to contentSkip to conversation

xen/arm: Correctly read the GICv3 Re-Distributor stride

PATCH In review Shameerali Kolothum Thodi 3 participants 6 messages Sep 30, 2015

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xen/arm: Correctly read the GICv3 Re-Distributor stride

The GICv3 driver read a 32 bit value for the re-distributor stride, but the dts binding is a two-cell property.

All instances of rdist stride access and reference has been modified to accommodate 64 bit value. The changes are compiled and verified on HiSilicon Hip05 platform.

---
 xen/arch/arm/gic-v3.c        | 9 +++++----
 xen/arch/arm/vgic-v3.c       | 8 ++++----
 xen/include/asm-arm/domain.h | 2 +-
 xen/include/asm-arm/vgic.h   | 2 +-
 4 files changed, 11 insertions(+), 10 deletions(-)
xen/arch/arm/gic-v3.c +5 -4
index d1db1ce..8003cd6 100644
@@ -46,7 +46,7 @@
 static struct {
     void __iomem *map_dbase;  /* Mapped address of distributor registers */
     struct rdist_region *rdist_regions;
-    uint32_t  rdist_stride;
+    uint64_t rdist_stride;
     unsigned int rdist_count; /* Number of rdist regions count */
     unsigned int nr_priorities;
     spinlock_t lock;
@@ -1074,7 +1074,7 @@ static int gicv3_make_hwdom_dt_node(const struct domain *d,
     if ( res )
         return res;
 
-    res = fdt_property_cell(fdt, "redistributor-stride",
+    res = fdt_property_u64(fdt, "redistributor-stride",
                             d->arch.vgic.rdist_stride);
     if ( res )
         return res;
@@ -1224,7 +1224,8 @@ static int __init gicv3_init(void)
     /* The vGIC code requires the region to be sorted */
     sort(rdist_regs, gicv3.rdist_count, sizeof(*rdist_regs), cmp_rdist, NULL);
 
-    if ( !dt_property_read_u32(node, "redistributor-stride", &gicv3.rdist_stride) )
+    if ( !dt_property_read_u64(node, "redistributor-stride",
+                &gicv3.rdist_stride) )
         gicv3.rdist_stride = 0;
 
     gicv3.rdist_regions= rdist_regs;
@@ -1248,7 +1249,7 @@ static int __init gicv3_init(void)
     printk("GICv3 initialization:\n"
            "      gic_dist_addr=%#"PRIpaddr"\n"
            "      gic_maintenance_irq=%u\n"
-           "      gic_rdist_stride=%#x\n"
+           "      gic_rdist_stride=%#lx\n"
            "      gic_rdist_regions=%d\n",
            dbase, gicv3_info.maintenance_irq,
            gicv3.rdist_stride, gicv3.rdist_count);
xen/arch/arm/vgic-v3.c +4 -4
index f1c482d..91442c7 100644
@@ -56,13 +56,13 @@ static struct {
     /* Re-distributor regions */
     unsigned int nr_rdist_regions;
     const struct rdist_region *regions;
-    uint32_t rdist_stride; /* Re-distributor stride */
+    uint64_t rdist_stride; /* Re-distributor stride */
 } vgic_v3_hw;
 
 void vgic_v3_setup_hw(paddr_t dbase,
                       unsigned int nr_rdist_regions,
                       const struct rdist_region *regions,
-                      uint32_t rdist_stride)
+                      uint64_t rdist_stride)
 {
     vgic_v3_hw.enabled = 1;
     vgic_v3_hw.dbase = dbase;
@@ -598,7 +598,7 @@ static inline struct vcpu *get_vcpu_from_rdist(paddr_t gpa,
                                                uint32_t *offset)
 {
     struct domain *d = v->domain;
-    uint32_t stride = d->arch.vgic.rdist_stride;
+    uint64_t stride = d->arch.vgic.rdist_stride;
     paddr_t base;
     int i, vcpu_id;
     struct vgic_rdist_region *region;
@@ -1084,7 +1084,7 @@ static int vgic_v3_vcpu_init(struct vcpu *v)
 
     /* Convenient alias */
     struct domain *d = v->domain;
-    uint32_t rdist_stride = d->arch.vgic.rdist_stride;
+    uint64_t rdist_stride = d->arch.vgic.rdist_stride;
 
     /* For SGI and PPI the target is always this CPU */
     affinity = (MPIDR_AFFINITY_LEVEL(v->arch.vmpidr, 3) << 32 |
xen/include/asm-arm/domain.h +1 -1
index c3f5a95..6d5acff 100644
@@ -111,7 +111,7 @@ struct arch_domain
             unsigned int first_cpu;         /* First CPU handled */
         } rdist_regions[MAX_RDIST_COUNT];
         int nr_regions;                     /* Number of rdist regions */
-        uint32_t rdist_stride;              /* Re-Distributor stride */
+        uint64_t rdist_stride;              /* Re-Distributor stride */
 #endif
     } vgic;
 
xen/include/asm-arm/vgic.h +1 -1
index 41cadb1..dcaeeab 100644
@@ -243,7 +243,7 @@ struct rdist_region;
 void vgic_v3_setup_hw(paddr_t dbase,
                       unsigned int nr_rdist_regions,
                       const struct rdist_region *regions,
-                      uint32_t rdist_stride);
+                      uint64_t rdist_stride);
 #endif
 
 #endif /* __ASM_ARM_VGIC_H__ */
Signature · 1 line
1.9.1
#1 lore raw
Quoted 2 lines · Shameerali Kolothum Thodi

The GICv3 driver read a 32 bit value for the re-distributor stride, but the dts binding is a two-cell property.

The binding doc I have says:

- redistributor-stride : If using padding pages, specifies the stride of consecutive redistributors. Must be a multiple of 64kB.

i.e. it doesn't say anything about the size. The _example_ is 2 cells, but I don't think that is normative.

Unless you can point to a bindings update which specifies strictly two cells then I think the right answer is the use dt_read_number.

Ian.

#5 lore raw
In reply to Shameerali Kolothum Thodi

Hi Shameerali,

Quoted 1 line · Shameerali Kolothum Thodi

The GICv3 driver read a 32 bit value for the re-distributor stride, but

s/read/reads/

Quoted 1 line

the dts binding is a two-cell property.

I would mention that two-cell = 64 bit to make clear that the issue is we are reading only one-cell (i.e 32 bit).

Quoted 5 lines

All instances of rdist stride access and reference has been modified to accommodate 64 bit value. The changes are compiled and verified on HiSilicon Hip05 platform.

FIY, this patch will clash with the series [1].

Regards,

[1] http://www.gossamer-threads.com/lists/xen/devel/400241

Signature · 1 line
Julien Grall
#2 lore raw
In reply to Ian Campbell
Quoted 6 lines · Ian Campbell
Quoted 2 lines · Shameerali Kolothum Thodi

The GICv3 driver read a 32 bit value for the re-distributor stride, but the dts binding is a two-cell property.

The binding doc I have says:

Quoted 8 lines

- redistributor-stride : If using padding pages, specifies the stride of consecutive redistributors. Must be a multiple of 64kB.

i.e. it doesn't say anything about the size. The _example_ is 2 cells, but I don't think that is normative.

Unless you can point to a bindings update which specifies strictly two cells then I think the right answer is the use dt_read_number.

FWIW Linux is using of_property_read_u64 to get the stride. It has always been the case and I don't understand how we end up to use dt_property_u32 in Xen as we port the driver from Linux...

So before using dt_read_number in Xen, I would rather check if the DT binding is not clear enough or the Linux drivers doesn't respect the binding.

Regards,

Signature · 1 line
Julien Grall
#3 lore raw
In reply to Ian Campbell
Quoted 16 lines

-----Original Message-----
From: Ian Campbell [mailto:[email protected]]
Sent: 30 September 2015 12:04
To: Shameerali Kolothum Thodi; [email protected]
Cc: [email protected]; [email protected]
Subject: Re: [PATCH] xen/arm: Correctly read the GICv3 Re-Distributor stride

Quoted 2 lines · Shameerali Kolothum Thodi

The GICv3 driver read a 32 bit value for the re-distributor stride, but the dts binding is a two-cell property.

The binding doc I have says:

Quoted 8 lines

- redistributor-stride : If using padding pages, specifies the stride of consecutive redistributors. Must be a multiple of 64kB.

i.e. it doesn't say anything about the size. The _example_ is 2 cells, but I don't think that is normative.

Unless you can point to a bindings update which specifies strictly two cells then I think the right answer is the use dt_read_number.

You are right. The binding text is not clear. But Linux gic driver from the beginning is using the 64 bit value.

This is what I could find from the forums(https://patches.linaro.org/24165/)
----------------------------

Review

Quoted 5 lines
(quoted diff) +5 -0
+	gic: interrupt-controller@2c010000 {
+		compatible = "arm,gic-v3";
+		#interrupt-cells = <3>;
+		interrupt-controller;
+		redistributor-stride = 0x40000;		// 256kB stride

I raised the point in patch 1 about the stide value being read as a 64-bit value, which would imply that the example should be the following, I think:
------------------------

Actually linux sets the stride to zero, if it's not 64 bit.

irq-gic-v3.c - gic_of_init()

if (of_property_read_u64(node, "redistributor-stride", &redist_stride))

                redist_stride = 0;
Quoted 1 line

Ian.

#4 lore raw
Quoted 18 lines · Shameerali Kolothum Thodi
Quoted 18 lines

-----Original Message-----
From: Ian Campbell [mailto:[email protected]]
Sent: 30 September 2015 12:04
To: Shameerali Kolothum Thodi; [email protected]
Cc: [email protected]; [email protected]
Subject: Re: [PATCH] xen/arm: Correctly read the GICv3 Re-Distributor stride

Quoted 2 lines · Shameerali Kolothum Thodi

The GICv3 driver read a 32 bit value for the re-distributor stride, but the dts binding is a two-cell property.

The binding doc I have says:

- redistributor-stride : If using padding pages, specifies the stride of consecutive redistributors. Must be a multiple of 64kB.

Quoted 8 lines
Quoted 5 lines

i.e. it doesn't say anything about the size. The _example_ is 2 cells, but I don't think that is normative.

Unless you can point to a bindings update which specifies strictly two cells then I think the right answer is the use dt_read_number.

You are right. The binding text is not clear. But Linux gic driver from the beginning is using the 64 bit value.

The Linux gic driver is also not a normative declaration of the spec.

I think the spec needs to be clarified and then Linux and/or Xen need to be fixed to follow the spec.

Ian.