Hi,
This series introduces GICv3 ITS to 32-bit world. Since I'm limited
with real world 32-bit platforms which uses ITS it was tested with
help of vITS on 64-bit host running 32-bit guest.
I used Andrea's its/v8 branch at [1] with following option passed to
kvmtool: --aarch32 --irqchip=gicv3-its --force-pci
[1] git://http://www.linux-arm.org/kvmtool.git
Changelog:
RFC -> v1
- rebased on 4.9-rc2, gits_read_typer() has been dropped
- spilt ITS and vITS in separate patch sets
Vladimir Murzin (6):
irqchip/gic-v3-its: Change unsigned types for AArch32 compatibility
irqchip/gic-v3-its: Narrow down Entry Size when used as a divider
irqchip/gicv3-its: Specialise flush_dcache operation
irqchip/gicv3-its: Specialise readq and writeq accesses
ARM: gic-v3-its: Add 32bit support to GICv3 ITS
ARM: virt: Select ARM_GIC_V3_ITS
arch/arm/Kconfig | 1 +
arch/arm/include/asm/arch_gicv3.h | 54 +++++++++++++++++++++----
arch/arm64/include/asm/arch_gicv3.h | 17 ++++++++
drivers/irqchip/irq-gic-v3-its.c | 75 +++++++++++++++++------------------
include/linux/irqchip/arm-gic-v3.h | 4 +-
5 files changed, 104 insertions(+), 47 deletions(-)
--
1.7.9.5
Wire-up flush_dcache, readq- and writeq-like gic-v3-its assessors, so
GICv3 ITS gets all it needs to be built and run.
Signed-off-by: Vladimir Murzin <[email protected]>
---
arch/arm/include/asm/arch_gicv3.h | 54 ++++++++++++++++++++++++++++++++-----
1 file changed, 47 insertions(+), 7 deletions(-)
diff --git a/arch/arm/include/asm/arch_gicv3.h b/arch/arm/include/asm/arch_gicv3.h
index a808829..2747590 100644
--- a/arch/arm/include/asm/arch_gicv3.h
+++ b/arch/arm/include/asm/arch_gicv3.h
@@ -22,6 +22,7 @@
#include <linux/io.h>
#include <asm/barrier.h>
+#include <asm/cacheflush.h>
#include <asm/cp15.h>
#define ICC_EOIR1 __ACCESS_CP15(c12, 0, c12, 1)
@@ -230,19 +231,14 @@ static inline void gic_write_bpr1(u32 val)
* AArch32, since the syndrome register doesn't provide any information for
* them.
* Consequently, the following IO helpers use 32bit accesses.
- *
- * There are only two registers that need 64bit accesses in this driver:
- * - GICD_IROUTERn, contain the affinity values associated to each interrupt.
- * The upper-word (aff3) will always be 0, so there is no need for a lock.
- * - GICR_TYPER is an ID register and doesn't need atomicity.
*/
-static inline void gic_write_irouter(u64 val, volatile void __iomem *addr)
+static inline void __gic_writeq_nonatomic(u64 val, volatile void __iomem *addr)
{
writel_relaxed((u32)val, addr);
writel_relaxed((u32)(val >> 32), addr + 4);
}
-static inline u64 gic_read_typer(const volatile void __iomem *addr)
+static inline u64 __gic_readq_nonatomic(const volatile void __iomem *addr)
{
u64 val;
@@ -251,5 +247,49 @@ static inline u64 gic_read_typer(const volatile void __iomem *addr)
return val;
}
+#define gic_flush_dcache_to_poc(a,l) __cpuc_flush_dcache_area((a), (l))
+
+/*
+ * GICD_IROUTERn, contain the affinity values associated to each interrupt.
+ * The upper-word (aff3) will always be 0, so there is no need for a lock.
+ */
+#define gic_write_irouter(v, c) __gic_writeq_nonatomic(v, c)
+
+/*
+ * GICR_TYPER is an ID register and doesn't need atomicity.
+ */
+#define gic_read_typer(c) __gic_readq_nonatomic(c)
+
+/*
+ * GITS_BASER - hi and lo bits may be accessed independently.
+ */
+#define gits_read_baser(c) __gic_readq_nonatomic(c)
+#define gits_write_baser(v, c) __gic_writeq_nonatomic(v, c)
+
+/*
+ * GICR_PENDBASER and GICR_PROPBASE are changed with LPIs disabled, so they
+ * won't be being used during any updates and can be changed non-atomically
+ */
+#define gicr_read_propbaser(c) __gic_readq_nonatomic(c)
+#define gicr_write_propbaser(v, c) __gic_writeq_nonatomic(v, c)
+#define gicr_read_pendbaser(c) __gic_readq_nonatomic(c)
+#define gicr_write_pendbaser(v, c) __gic_writeq_nonatomic(v, c)
+
+/*
+ * GITS_TYPER is an ID register and doesn't need atomicity.
+ */
+#define gits_read_typer(c) __gic_readq_nonatomic(c)
+
+/*
+ * GITS_CBASER - hi and lo bits may be accessed independently.
+ */
+#define gits_read_cbaser(c) __gic_readq_nonatomic(c)
+#define gits_write_cbaser(v, c) __gic_writeq_nonatomic(v, c)
+
+/*
+ * GITS_CWRITER - hi and lo bits may be accessed independently.
+ */
+#define gits_write_cwriter(v, c) __gic_writeq_nonatomic(v, c)
+
#endif /* !__ASSEMBLY__ */
#endif /* !__ASM_ARCH_GICV3_H */
--
1.7.9.5
This patch allows ARM guests to use GICv3 ITS on an arm64 host
Signed-off-by: Vladimir Murzin <[email protected]>
---
arch/arm/Kconfig | 1 +
1 file changed, 1 insertion(+)
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index b5d529f..caef684 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -703,6 +703,7 @@ config ARCH_VIRT
select ARM_GIC
select ARM_GIC_V2M if PCI
select ARM_GIC_V3
+ select ARM_GIC_V3_ITS if PCI
select ARM_PSCI
select HAVE_ARM_ARCH_TIMER
--
1.7.9.5
Make sure that constants which are supposed to be applied on 64-bit
data is actually unsigned long long, so they won't be truncated when
used in 32-bit mode.
Signed-off-by: Vladimir Murzin <[email protected]>
---
drivers/irqchip/irq-gic-v3-its.c | 28 ++++++++++++++--------------
include/linux/irqchip/arm-gic-v3.h | 4 ++--
2 files changed, 16 insertions(+), 16 deletions(-)
diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index c5dee30..bca125e 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c
@@ -196,7 +196,7 @@ struct its_cmd_block {
static void its_encode_cmd(struct its_cmd_block *cmd, u8 cmd_nr)
{
- cmd->raw_cmd[0] &= ~0xffUL;
+ cmd->raw_cmd[0] &= ~0xffULL;
cmd->raw_cmd[0] |= cmd_nr;
}
@@ -208,43 +208,43 @@ static void its_encode_devid(struct its_cmd_block *cmd, u32 devid)
static void its_encode_event_id(struct its_cmd_block *cmd, u32 id)
{
- cmd->raw_cmd[1] &= ~0xffffffffUL;
+ cmd->raw_cmd[1] &= ~0xffffffffULL;
cmd->raw_cmd[1] |= id;
}
static void its_encode_phys_id(struct its_cmd_block *cmd, u32 phys_id)
{
- cmd->raw_cmd[1] &= 0xffffffffUL;
+ cmd->raw_cmd[1] &= 0xffffffffULL;
cmd->raw_cmd[1] |= ((u64)phys_id) << 32;
}
static void its_encode_size(struct its_cmd_block *cmd, u8 size)
{
- cmd->raw_cmd[1] &= ~0x1fUL;
+ cmd->raw_cmd[1] &= ~0x1fULL;
cmd->raw_cmd[1] |= size & 0x1f;
}
static void its_encode_itt(struct its_cmd_block *cmd, u64 itt_addr)
{
- cmd->raw_cmd[2] &= ~0xffffffffffffUL;
- cmd->raw_cmd[2] |= itt_addr & 0xffffffffff00UL;
+ cmd->raw_cmd[2] &= ~0xffffffffffffULL;
+ cmd->raw_cmd[2] |= itt_addr & 0xffffffffff00ULL;
}
static void its_encode_valid(struct its_cmd_block *cmd, int valid)
{
- cmd->raw_cmd[2] &= ~(1UL << 63);
+ cmd->raw_cmd[2] &= ~(1ULL << 63);
cmd->raw_cmd[2] |= ((u64)!!valid) << 63;
}
static void its_encode_target(struct its_cmd_block *cmd, u64 target_addr)
{
- cmd->raw_cmd[2] &= ~(0xffffffffUL << 16);
- cmd->raw_cmd[2] |= (target_addr & (0xffffffffUL << 16));
+ cmd->raw_cmd[2] &= ~(0xffffffffULL << 16);
+ cmd->raw_cmd[2] |= (target_addr & (0xffffffffULL << 16));
}
static void its_encode_collection(struct its_cmd_block *cmd, u16 col)
{
- cmd->raw_cmd[2] &= ~0xffffUL;
+ cmd->raw_cmd[2] &= ~0xffffULL;
cmd->raw_cmd[2] |= col;
}
@@ -657,8 +657,8 @@ static void its_irq_compose_msi_msg(struct irq_data *d, struct msi_msg *msg)
its = its_dev->its;
addr = its->phys_base + GITS_TRANSLATER;
- msg->address_lo = addr & ((1UL << 32) - 1);
- msg->address_hi = addr >> 32;
+ msg->address_lo = lower_32_bits(addr);
+ msg->address_hi = upper_32_bits(addr);
msg->data = its_get_event_id(d);
iommu_dma_map_msi_msg(d->irq, msg);
@@ -935,9 +935,9 @@ static int its_setup_baser(struct its_node *its, struct its_baser *baser,
}
if (val != tmp) {
- pr_err("ITS@%pa: %s doesn't stick: %lx %lx\n",
+ pr_err("ITS@%pa: %s doesn't stick: %llx %llx\n",
&its->phys_base, its_base_type_string[type],
- (unsigned long) val, (unsigned long) tmp);
+ val, tmp);
free_pages((unsigned long)base, order);
return -ENXIO;
}
diff --git a/include/linux/irqchip/arm-gic-v3.h b/include/linux/irqchip/arm-gic-v3.h
index b7e3431..5118d3a 100644
--- a/include/linux/irqchip/arm-gic-v3.h
+++ b/include/linux/irqchip/arm-gic-v3.h
@@ -239,7 +239,7 @@
#define GITS_TYPER_PTA (1UL << 19)
#define GITS_TYPER_HWCOLLCNT_SHIFT 24
-#define GITS_CBASER_VALID (1UL << 63)
+#define GITS_CBASER_VALID (1ULL << 63)
#define GITS_CBASER_SHAREABILITY_SHIFT (10)
#define GITS_CBASER_INNER_CACHEABILITY_SHIFT (59)
#define GITS_CBASER_OUTER_CACHEABILITY_SHIFT (53)
@@ -265,7 +265,7 @@
#define GITS_BASER_NR_REGS 8
-#define GITS_BASER_VALID (1UL << 63)
+#define GITS_BASER_VALID (1ULL << 63)
#define GITS_BASER_INDIRECT (1ULL << 62)
#define GITS_BASER_INNER_CACHEABILITY_SHIFT (59)
--
1.7.9.5
GITS_BASER<n>'s Entry Size is much smaller than 64-bit, but when it
used as a divider it forces compiler to generate __aeabi_uldivmod if
build in 32-bit mode. So, casting it to int (like it is done in other
places) where used as a divider would give a hint to compiler that
32-bit division can be used.
Signed-off-by: Vladimir Murzin <[email protected]>
---
drivers/irqchip/irq-gic-v3-its.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index bca125e..312dd55 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c
@@ -948,7 +948,7 @@ static int its_setup_baser(struct its_node *its, struct its_baser *baser,
tmp = indirect ? GITS_LVL1_ENTRY_SIZE : esz;
pr_info("ITS@%pa: allocated %d %s @%lx (%s, esz %d, psz %dK, shr %d)\n",
- &its->phys_base, (int)(PAGE_ORDER_TO_SIZE(order) / tmp),
+ &its->phys_base, (int)(PAGE_ORDER_TO_SIZE(order) / (int)tmp),
its_base_type_string[type],
(unsigned long)virt_to_phys(base),
indirect ? "indirect" : "flat", (int)esz,
@@ -983,7 +983,7 @@ static bool its_parse_baser_device(struct its_node *its, struct its_baser *baser
* which is reported by ITS hardware times lvl1 table
* entry size.
*/
- ids -= ilog2(psz / esz);
+ ids -= ilog2(psz / (int)esz);
esz = GITS_LVL1_ENTRY_SIZE;
}
}
@@ -998,7 +998,7 @@ static bool its_parse_baser_device(struct its_node *its, struct its_baser *baser
new_order = max_t(u32, get_order(esz << ids), new_order);
if (new_order >= MAX_ORDER) {
new_order = MAX_ORDER - 1;
- ids = ilog2(PAGE_ORDER_TO_SIZE(new_order) / esz);
+ ids = ilog2(PAGE_ORDER_TO_SIZE(new_order) / (int)esz);
pr_warn("ITS@%pa: Device Table too large, reduce ids %u->%u\n",
&its->phys_base, its->device_ids, ids);
}
--
1.7.9.5
Hi,
On 14/11/16 10:06, Marc Zyngier wrote:
> Hi Vladimir,
>
> On 07/11/16 09:21, Vladimir Murzin wrote:
>> Hi,
>>
>> This series introduces GICv3 ITS to 32-bit world. Since I'm limited
>> with real world 32-bit platforms which uses ITS it was tested with
>> help of vITS on 64-bit host running 32-bit guest.
>>
>> I used Andrea's its/v8 branch at [1] with following option passed to
>> kvmtool: --aarch32 --irqchip=gicv3-its --force-pci
>>
>> [1] git://http://www.linux-arm.org/kvmtool.git
>>
>> Changelog:
>>
>> RFC -> v1
>> - rebased on 4.9-rc2, gits_read_typer() has been dropped
>> - spilt ITS and vITS in separate patch sets
>>
>> Vladimir Murzin (6):
>> irqchip/gic-v3-its: Change unsigned types for AArch32 compatibility
>> irqchip/gic-v3-its: Narrow down Entry Size when used as a divider
>> irqchip/gicv3-its: Specialise flush_dcache operation
>> irqchip/gicv3-its: Specialise readq and writeq accesses
>> ARM: gic-v3-its: Add 32bit support to GICv3 ITS
>> ARM: virt: Select ARM_GIC_V3_ITS
>>
>> arch/arm/Kconfig | 1 +
>> arch/arm/include/asm/arch_gicv3.h | 54 +++++++++++++++++++++----
>> arch/arm64/include/asm/arch_gicv3.h | 17 ++++++++
>> drivers/irqchip/irq-gic-v3-its.c | 75 +++++++++++++++++------------------
>> include/linux/irqchip/arm-gic-v3.h | 4 +-
>> 5 files changed, 104 insertions(+), 47 deletions(-)
>
> I've queued all of this in my irq/gic-4.10 branch.
>
Great!
Thanks
Vladimir
> Thanks,
>
> M.
>
Hi Vladimir,
On 07/11/16 09:21, Vladimir Murzin wrote:
> Hi,
>
> This series introduces GICv3 ITS to 32-bit world. Since I'm limited
> with real world 32-bit platforms which uses ITS it was tested with
> help of vITS on 64-bit host running 32-bit guest.
>
> I used Andrea's its/v8 branch at [1] with following option passed to
> kvmtool: --aarch32 --irqchip=gicv3-its --force-pci
>
> [1] git://http://www.linux-arm.org/kvmtool.git
>
> Changelog:
>
> RFC -> v1
> - rebased on 4.9-rc2, gits_read_typer() has been dropped
> - spilt ITS and vITS in separate patch sets
>
> Vladimir Murzin (6):
> irqchip/gic-v3-its: Change unsigned types for AArch32 compatibility
> irqchip/gic-v3-its: Narrow down Entry Size when used as a divider
> irqchip/gicv3-its: Specialise flush_dcache operation
> irqchip/gicv3-its: Specialise readq and writeq accesses
> ARM: gic-v3-its: Add 32bit support to GICv3 ITS
> ARM: virt: Select ARM_GIC_V3_ITS
>
> arch/arm/Kconfig | 1 +
> arch/arm/include/asm/arch_gicv3.h | 54 +++++++++++++++++++++----
> arch/arm64/include/asm/arch_gicv3.h | 17 ++++++++
> drivers/irqchip/irq-gic-v3-its.c | 75 +++++++++++++++++------------------
> include/linux/irqchip/arm-gic-v3.h | 4 +-
> 5 files changed, 104 insertions(+), 47 deletions(-)
I've queued all of this in my irq/gic-4.10 branch.
Thanks,
M.
--
Jazz is not dead. It just smells funny...
readq and writeq type of assessors are not supported in AArch32, so we
need to specialise them and glue later with series of 32-bit accesses
on AArch32 side.
Signed-off-by: Vladimir Murzin <[email protected]>
---
arch/arm64/include/asm/arch_gicv3.h | 14 ++++++++++++++
drivers/irqchip/irq-gic-v3-its.c | 24 ++++++++++++------------
2 files changed, 26 insertions(+), 12 deletions(-)
diff --git a/arch/arm64/include/asm/arch_gicv3.h b/arch/arm64/include/asm/arch_gicv3.h
index 4f0402a..022523b 100644
--- a/arch/arm64/include/asm/arch_gicv3.h
+++ b/arch/arm64/include/asm/arch_gicv3.h
@@ -190,5 +190,19 @@ static inline void gic_write_bpr1(u32 val)
#define gic_flush_dcache_to_poc(a,l) __flush_dcache_area((a), (l))
+#define gits_read_baser(c) readq_relaxed(c)
+#define gits_write_baser(v, c) writeq_relaxed(v, c)
+
+#define gits_read_cbaser(c) readq_relaxed(c)
+#define gits_write_cbaser(v, c) writeq_relaxed(v, c)
+
+#define gits_write_cwriter(v, c) writeq_relaxed(v, c)
+
+#define gicr_read_propbaser(c) readq_relaxed(c)
+#define gicr_write_propbaser(v, c) writeq_relaxed(v, c)
+
+#define gicr_write_pendbaser(v, c) writeq_relaxed(v, c)
+#define gicr_read_pendbaser(c) readq_relaxed(c)
+
#endif /* __ASSEMBLY__ */
#endif /* __ASM_ARCH_GICV3_H */
diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index b2a6e7b..69b040f 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c
@@ -835,7 +835,7 @@ static u64 its_read_baser(struct its_node *its, struct its_baser *baser)
{
u32 idx = baser - its->tables;
- return readq_relaxed(its->base + GITS_BASER + (idx << 3));
+ return gits_read_baser(its->base + GITS_BASER + (idx << 3));
}
static void its_write_baser(struct its_node *its, struct its_baser *baser,
@@ -843,7 +843,7 @@ static void its_write_baser(struct its_node *its, struct its_baser *baser,
{
u32 idx = baser - its->tables;
- writeq_relaxed(val, its->base + GITS_BASER + (idx << 3));
+ gits_write_baser(val, its->base + GITS_BASER + (idx << 3));
baser->val = its_read_baser(its, baser);
}
@@ -1125,8 +1125,8 @@ static void its_cpu_init_lpis(void)
GICR_PROPBASER_WaWb |
((LPI_NRBITS - 1) & GICR_PROPBASER_IDBITS_MASK));
- writeq_relaxed(val, rbase + GICR_PROPBASER);
- tmp = readq_relaxed(rbase + GICR_PROPBASER);
+ gicr_write_propbaser(val, rbase + GICR_PROPBASER);
+ tmp = gicr_read_propbaser(rbase + GICR_PROPBASER);
if ((tmp ^ val) & GICR_PROPBASER_SHAREABILITY_MASK) {
if (!(tmp & GICR_PROPBASER_SHAREABILITY_MASK)) {
@@ -1138,7 +1138,7 @@ static void its_cpu_init_lpis(void)
val &= ~(GICR_PROPBASER_SHAREABILITY_MASK |
GICR_PROPBASER_CACHEABILITY_MASK);
val |= GICR_PROPBASER_nC;
- writeq_relaxed(val, rbase + GICR_PROPBASER);
+ gicr_write_propbaser(val, rbase + GICR_PROPBASER);
}
pr_info_once("GIC: using cache flushing for LPI property table\n");
gic_rdists->flags |= RDIST_FLAGS_PROPBASE_NEEDS_FLUSHING;
@@ -1149,8 +1149,8 @@ static void its_cpu_init_lpis(void)
GICR_PENDBASER_InnerShareable |
GICR_PENDBASER_WaWb);
- writeq_relaxed(val, rbase + GICR_PENDBASER);
- tmp = readq_relaxed(rbase + GICR_PENDBASER);
+ gicr_write_pendbaser(val, rbase + GICR_PENDBASER);
+ tmp = gicr_read_pendbaser(rbase + GICR_PENDBASER);
if (!(tmp & GICR_PENDBASER_SHAREABILITY_MASK)) {
/*
@@ -1160,7 +1160,7 @@ static void its_cpu_init_lpis(void)
val &= ~(GICR_PENDBASER_SHAREABILITY_MASK |
GICR_PENDBASER_CACHEABILITY_MASK);
val |= GICR_PENDBASER_nC;
- writeq_relaxed(val, rbase + GICR_PENDBASER);
+ gicr_write_pendbaser(val, rbase + GICR_PENDBASER);
}
/* Enable LPIs */
@@ -1716,8 +1716,8 @@ static int __init its_probe_one(struct resource *res,
(ITS_CMD_QUEUE_SZ / SZ_4K - 1) |
GITS_CBASER_VALID);
- writeq_relaxed(baser, its->base + GITS_CBASER);
- tmp = readq_relaxed(its->base + GITS_CBASER);
+ gits_write_cbaser(baser, its->base + GITS_CBASER);
+ tmp = gits_read_cbaser(its->base + GITS_CBASER);
if ((tmp ^ baser) & GITS_CBASER_SHAREABILITY_MASK) {
if (!(tmp & GITS_CBASER_SHAREABILITY_MASK)) {
@@ -1729,13 +1729,13 @@ static int __init its_probe_one(struct resource *res,
baser &= ~(GITS_CBASER_SHAREABILITY_MASK |
GITS_CBASER_CACHEABILITY_MASK);
baser |= GITS_CBASER_nC;
- writeq_relaxed(baser, its->base + GITS_CBASER);
+ gits_write_cbaser(baser, its->base + GITS_CBASER);
}
pr_info("ITS: using cache flushing for cmd queue\n");
its->flags |= ITS_FLAGS_CMDQ_NEEDS_FLUSHING;
}
- writeq_relaxed(0, its->base + GITS_CWRITER);
+ gits_write_cwriter(0, its->base + GITS_CWRITER);
writel_relaxed(GITS_CTLR_ENABLE, its->base + GITS_CTLR);
err = its_init_domain(handle, its);
--
1.7.9.5