Skip to contentSkip to conversation

irqchip, gicv3: Implement Cavium ThunderX errata

PATCH In review Robert Richter 5 participants 14 messages Jun 30, 2015 – Jul 8, 2015

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover irqchip, gicv3: Implement Cavium ThunderX errata

From: Robert Richter <[email protected]>

This patch series implements workarounds for HW errata in Cavium's ThunderX GICV3.

The first one is a gicv3 code update I sent a while ago and is a prerequisit for patch #4. Patch #2 adds generic code to parse the hw revision provided by an IIDR register value and runs specific code if hw matches. This is used to implement the actual errata fixes in patch #3 (gicv3) and #4 (gicv3-its).

Robert Richter (4):
irqchip, gicv3-its: Read typer register outside the loop irqchip, gicv3: Add HW revision detection and configuration irqchip, gicv3: Implement Cavium ThunderX erratum 23154 irqchip, gicv3-its: Implement Cavium ThunderX errata 22375, 24313

 drivers/irqchip/irq-gic-common.c | 11 +++++++++
 drivers/irqchip/irq-gic-common.h |  9 +++++++
 drivers/irqchip/irq-gic-v3-its.c | 51 ++++++++++++++++++++++++++++++++++++----
 drivers/irqchip/irq-gic-v3.c     | 51 +++++++++++++++++++++++++++++++++++++++-
 4 files changed, 116 insertions(+), 6 deletions(-)
Signature · 1 line
2.1.1
#1 lore raw

PATCH 1/4 irqchip, gicv3-its: Read typer register outside the loop

From: Robert Richter <[email protected]>

No need to read the typer register in the loop. Values do not change.

---
 drivers/irqchip/irq-gic-v3-its.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
drivers/irqchip/irq-gic-v3-its.c +2 -3
index 1b7e155869f6..c5ce21277920 100644
@@ -803,6 +803,8 @@ static int its_alloc_tables(struct its_node *its)
 	int psz = SZ_64K;
 	u64 shr = GITS_BASER_InnerShareable;
 	u64 cache = GITS_BASER_WaWb;
+	u64 typer = readq_relaxed(its->base + GITS_TYPER);
+	u32 ids = GITS_TYPER_DEVBITS(typer);
 
 	for (i = 0; i < GITS_BASER_NR_REGS; i++) {
 		u64 val = readq_relaxed(its->base + GITS_BASER + i * 8);
@@ -825,9 +827,6 @@ static int its_alloc_tables(struct its_node *its)
 		 * For other tables, only allocate a single page.
 		 */
 		if (type == GITS_BASER_TYPE_DEVICE) {
-			u64 typer = readq_relaxed(its->base + GITS_TYPER);
-			u32 ids = GITS_TYPER_DEVBITS(typer);
-
 			/*
 			 * 'order' was initialized earlier to the default page
 			 * granule of the the ITS.  We can't have an allocation
Signature · 1 line
2.1.1
#2 lore raw

Re: [PATCH 1/4] irqchip, gicv3-its: Read typer register outside the loop

Quoted 24 lines · Robert Richter

From: Robert Richter <[email protected]>

No need to read the typer register in the loop. Values do not change.

---
 drivers/irqchip/irq-gic-v3-its.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
drivers/irqchip/irq-gic-v3-its.c +2 -0
index 1b7e155869f6..c5ce21277920 100644
@@ -803,6 +803,8 @@ static int its_alloc_tables(struct its_node *its)
 	int psz = SZ_64K;
 	u64 shr = GITS_BASER_InnerShareable;
 	u64 cache = GITS_BASER_WaWb;
+	u64 typer = readq_relaxed(its->base + GITS_TYPER);
+	u32 ids = GITS_TYPER_DEVBITS(typer);
 
 	for (i = 0; i < GITS_BASER_NR_REGS; i++) {
 		u64 val = readq_relaxed(its->base + GITS_BASER + i * 8);

drivers/irqchip/irq-gic-v3-its.c static int its_alloc_tables(struct its_node *its) · line 832

Quoted 9 lines
drivers/irqchip/irq-gic-v3-its.c +0 -3
@@ -825,9 +827,6 @@ static int its_alloc_tables(struct its_node *its)
 		 * For other tables, only allocate a single page.
 		 */
 		if (type == GITS_BASER_TYPE_DEVICE) {
-			u64 typer = readq_relaxed(its->base + GITS_TYPER);
-			u32 ids = GITS_TYPER_DEVBITS(typer);
-
 			/*
 			 * 'order' was initialized earlier to the default page
 			 * granule of the the ITS.  We can't have an allocation

The TYPER value certainly don't change, but there is also only one device table per ITS instance, so we only read it once. What do we gain here?

Thanks,

	M.
Signature · 1 line
Jazz is not dead. It just smells funny...
#3 lore raw

PATCH 2/4 irqchip, gicv3: Add HW revision detection and configuration

In reply to Robert Richter

From: Robert Richter <[email protected]>

Some GIC revisions require an individual configuration to esp. add workarounds for HW bugs. This patch implements generic code to parse the hw revision provided by an IIDR register value and runs specific code if hw matches. There are functions that read the IIDR registers for GICV3 and ITS (GICD_IIDR/GITS_IIDR) and then go through a list of init functions to be called for specific versions.

The patch is needed to implement workarounds for HW errata in Cavium's ThunderX GICV3.

---
 drivers/irqchip/irq-gic-common.c | 11 +++++++++++
 drivers/irqchip/irq-gic-common.h |  9 +++++++++
 drivers/irqchip/irq-gic-v3-its.c | 15 +++++++++++++++
 drivers/irqchip/irq-gic-v3.c     | 14 ++++++++++++++
 4 files changed, 49 insertions(+)
drivers/irqchip/irq-gic-common.c +11 -0
index ad96ebb0c7ab..27ad9bef760f 100644
@@ -21,6 +21,17 @@
 
 #include "irq-gic-common.h"
 
+void gic_check_capabilities(u32 iidr, const struct gic_capabilities *cap,
+			void *data)
+{
+	for (; cap->desc; cap++) {
+		if ((iidr & cap->mask) != cap->id)
+			continue;
+		cap->init(data);
+		pr_info("%s\n", cap->desc);
+	}
+}
+
 int gic_configure_irq(unsigned int irq, unsigned int type,
 		       void __iomem *base, void (*sync_access)(void))
 {
drivers/irqchip/irq-gic-common.h +9 -0
index 35a9884778bd..90d55b97c0df 100644
@@ -20,10 +20,19 @@
 #include <linux/of.h>
 #include <linux/irqdomain.h>
 
+struct gic_capabilities {
+	const char *desc;
+	void (*init)(void *data);
+	u32 id;
+	u32 mask;
+};
+
 int gic_configure_irq(unsigned int irq, unsigned int type,
                        void __iomem *base, void (*sync_access)(void));
 void gic_dist_config(void __iomem *base, int gic_irqs,
 		     void (*sync_access)(void));
 void gic_cpu_config(void __iomem *base, void (*sync_access)(void));
+void gic_check_capabilities(u32 iidr, const struct gic_capabilities *cap,
+			void *data);
 
 #endif /* _IRQ_GIC_COMMON_H */
drivers/irqchip/irq-gic-v3-its.c +15 -0
index c5ce21277920..de2fc82bba83 100644
@@ -36,6 +36,7 @@
 #include <asm/cputype.h>
 #include <asm/exception.h>
 
+#include "irq-gic-common.h"
 #include "irqchip.h"
 
 #define ITS_FLAGS_CMDQ_NEEDS_FLUSHING		(1 << 0)
@@ -1381,6 +1382,18 @@ static int its_force_quiescent(void __iomem *base)
 	}
 }
 
+static const struct gic_capabilities its_errata[] = {
+	{
+	}
+};
+
+static void its_check_capabilities(struct its_node *its)
+{
+	u32 iidr = readl_relaxed(its->base + GITS_IIDR);
+
+	gic_check_capabilities(iidr, its_errata, its);
+}
+
 static int its_probe(struct device_node *node, struct irq_domain *parent)
 {
 	struct resource res;
@@ -1439,6 +1452,8 @@ static int its_probe(struct device_node *node, struct irq_domain *parent)
 	}
 	its->cmd_write = its->cmd_base;
 
+	its_check_capabilities(its);
+
 	err = its_alloc_tables(its);
 	if (err)
 		goto out_free_cmd;
drivers/irqchip/irq-gic-v3.c +14 -0
index 49875adb6b44..f4bafc69cc18 100644
@@ -765,6 +765,18 @@ static const struct irq_domain_ops gic_irq_domain_ops = {
 	.free = gic_irq_domain_free,
 };
 
+static const struct gic_capabilities gicv3_errata[] = {
+	{
+	}
+};
+
+static void gicv3_check_capabilities(void)
+{
+	u32 iidr = readl_relaxed(gic_data.dist_base + GICD_IIDR);
+
+	gic_check_capabilities(iidr, gicv3_errata, NULL);
+}
+
 static int __init gic_of_init(struct device_node *node, struct device_node *parent)
 {
 	void __iomem *dist_base;
@@ -824,6 +836,8 @@ static int __init gic_of_init(struct device_node *node, struct device_node *pare
 	gic_data.nr_redist_regions = nr_redist_regions;
 	gic_data.redist_stride = redist_stride;
 
+	gicv3_check_capabilities();
+
 	/*
 	 * Find out how many interrupts are supported.
 	 * The GIC only supports up to 1020 interrupt sources (SGI+PPI+SPI)
Signature · 1 line
2.1.1
#4 lore raw

PATCH 3/4 irqchip, gicv3: Implement Cavium ThunderX erratum 23154

In reply to Robert Richter

From: Robert Richter <[email protected]>

This patch implements Cavium ThunderX erratum 23154.

The gicv3 of ThunderX requires a modified version for reading the IAR status to ensure data synchronization. Since this is in the fast-path and called with each interrupt, runtime patching is used using jump label patching for smallest overhead (no-op). This is the same technique as used for tracepoints.

---
 drivers/irqchip/irq-gic-v3.c | 37 ++++++++++++++++++++++++++++++++++++-
 1 file changed, 36 insertions(+), 1 deletion(-)
drivers/irqchip/irq-gic-v3.c +36 -1
index f4bafc69cc18..57bb69686ec6 100644
@@ -107,14 +107,38 @@ static void gic_redist_wait_for_rwp(void)
 }
 
 /* Low level accessors */
-static u64 __maybe_unused gic_read_iar(void)
+static u64 gic_read_iar_common(void)
+{
+	u64 irqstat;
+
+	asm volatile("mrs_s %0, " __stringify(ICC_IAR1_EL1) : "=r" (irqstat));
+	return irqstat;
+}
+
+/* Cavium ThunderX erratum 23154 */
+static u64 gic_read_iar_cavium_thunderx(void)
 {
 	u64 irqstat;
 
+	asm volatile("nop;nop;nop;nop;");
+	asm volatile("nop;nop;nop;nop;");
 	asm volatile("mrs_s %0, " __stringify(ICC_IAR1_EL1) : "=r" (irqstat));
+	asm volatile("nop;nop;nop;nop;");
+	mb();
+
 	return irqstat;
 }
 
+struct static_key is_cavium_thunderx = STATIC_KEY_INIT_FALSE;
+
+static u64 __maybe_unused gic_read_iar(void)
+{
+	if (static_key_false(&is_cavium_thunderx))
+		return gic_read_iar_common();
+	else
+		return gic_read_iar_cavium_thunderx();
+}
+
 static void __maybe_unused gic_write_pmr(u64 val)
 {
 	asm volatile("msr_s " __stringify(ICC_PMR_EL1) ", %0" : : "r" (val));
@@ -765,8 +789,19 @@ static const struct irq_domain_ops gic_irq_domain_ops = {
 	.free = gic_irq_domain_free,
 };
 
+static void gicv3_enable_cavium_thunderx(void *data)
+{
+	static_key_slow_inc(&is_cavium_thunderx);
+}
+
 static const struct gic_capabilities gicv3_errata[] = {
 	{
+		.desc	= "GIC: Cavium erratum 23154",
+		.id	= 0xa100034c,	/* ThunderX pass 1.x */
+		.mask	= 0xffff0fff,
+		.init	= gicv3_enable_cavium_thunderx,
+	},
+	{
 	}
 };
 
Signature · 1 line
2.1.1
#5 lore raw

Re: [PATCH 3/4] irqchip, gicv3: Implement Cavium ThunderX erratum 23154

Hi Robert,

Quoted 70 lines · Robert Richter

From: Robert Richter <[email protected]>

This patch implements Cavium ThunderX erratum 23154.

The gicv3 of ThunderX requires a modified version for reading the IAR status to ensure data synchronization. Since this is in the fast-path and called with each interrupt, runtime patching is used using jump label patching for smallest overhead (no-op). This is the same technique as used for tracepoints.

---
 drivers/irqchip/irq-gic-v3.c | 37 ++++++++++++++++++++++++++++++++++++-
 1 file changed, 36 insertions(+), 1 deletion(-)
drivers/irqchip/irq-gic-v3.c +30 -1
index f4bafc69cc18..57bb69686ec6 100644
@@ -107,14 +107,38 @@ static void gic_redist_wait_for_rwp(void)
 }
 
 /* Low level accessors */
-static u64 __maybe_unused gic_read_iar(void)
+static u64 gic_read_iar_common(void)
+{
+	u64 irqstat;
+
+	asm volatile("mrs_s %0, " __stringify(ICC_IAR1_EL1) : "=r" (irqstat));
+	return irqstat;
+}
+
+/* Cavium ThunderX erratum 23154 */
+static u64 gic_read_iar_cavium_thunderx(void)
 {
 	u64 irqstat;
 
+	asm volatile("nop;nop;nop;nop;");
+	asm volatile("nop;nop;nop;nop;");
 	asm volatile("mrs_s %0, " __stringify(ICC_IAR1_EL1) : "=r" (irqstat));
+	asm volatile("nop;nop;nop;nop;");
+	mb();
+
 	return irqstat;
 }
 
+struct static_key is_cavium_thunderx = STATIC_KEY_INIT_FALSE;
+
+static u64 __maybe_unused gic_read_iar(void)
+{
+	if (static_key_false(&is_cavium_thunderx))
+		return gic_read_iar_common();
+	else
+		return gic_read_iar_cavium_thunderx();
+}
+
 static void __maybe_unused gic_write_pmr(u64 val)
 {
 	asm volatile("msr_s " __stringify(ICC_PMR_EL1) ", %0" : : "r" (val));
@@ -765,8 +789,19 @@ static const struct irq_domain_ops gic_irq_domain_ops = {
 	.free = gic_irq_domain_free,
 };
 
+static void gicv3_enable_cavium_thunderx(void *data)
+{
+	static_key_slow_inc(&is_cavium_thunderx);
+}
+
 static const struct gic_capabilities gicv3_errata[] = {

drivers/irqchip/irq-gic-v3.c static const struct irq_domain_ops gic_irq_domain_ops = { · line 807

Quoted 10 lines
drivers/irqchip/irq-gic-v3.c +6 -0
 	{
+		.desc	= "GIC: Cavium erratum 23154",
+		.id	= 0xa100034c,	/* ThunderX pass 1.x */
+		.mask	= 0xffff0fff,
+		.init	= gicv3_enable_cavium_thunderx,
+	},
+	{
 	}
 };
 

How does this work when running a guest? Does the virtualized access suffer from the same erratum? If that's the case, we need a better workaround...

Thanks,

	M.
Signature · 1 line
Jazz is not dead. It just smells funny...
#6 lore raw

Re: [PATCH 3/4] irqchip, gicv3: Implement Cavium ThunderX erratum 23154

Marc,

Quoted 17 lines · Marc Zyngier

drivers/irqchip/irq-gic-v3.c static const struct irq_domain_ops gic_irq_domain_ops = { · line 806

Quoted 12 lines · Robert Richter

static const struct gic_capabilities gicv3_errata[] = {

drivers/irqchip/irq-gic-v3.c +6 -0
 	{
+		.desc	= "GIC: Cavium erratum 23154",
+		.id	= 0xa100034c,	/* ThunderX pass 1.x */
+		.mask	= 0xffff0fff,
+		.init	= gicv3_enable_cavium_thunderx,
+	},
+	{
 	}
 };

How does this work when running a guest? Does the virtualized access suffer from the same erratum? If that's the case, we need a better workaround...

We need to apply the workaround also for guests. So you are right, evaluating GICD_IIDR does not enable the workaround then as the register is emulated with ARM as implementer.

We considering MIDR_EL1 as a version check for this errata now. This should be the host's cpuid when running as a guest, right?

Thanks,

-Robert

#7 lore raw

Re: [PATCH 3/4] irqchip, gicv3: Implement Cavium ThunderX erratum 23154

Quoted 21 lines · Robert Richter

Marc,

Quoted 17 lines · Marc Zyngier

drivers/irqchip/irq-gic-v3.c static const struct irq_domain_ops gic_irq_domain_ops = { · line 806

Quoted 12 lines · Robert Richter

static const struct gic_capabilities gicv3_errata[] = {

drivers/irqchip/irq-gic-v3.c +6 -0
 	{
+		.desc	= "GIC: Cavium erratum 23154",
+		.id	= 0xa100034c,	/* ThunderX pass 1.x */
+		.mask	= 0xffff0fff,
+		.init	= gicv3_enable_cavium_thunderx,
+	},
+	{
 	}
 };

How does this work when running a guest? Does the virtualized access suffer from the same erratum? If that's the case, we need a better workaround...

Quoted 6 lines

We need to apply the workaround also for guests. So you are right, evaluating GICD_IIDR does not enable the workaround then as the register is emulated with ARM as implementer.

We considering MIDR_EL1 as a version check for this errata now. This should be the host's cpuid when running as a guest, right?

Yes, that should work, as we don't repaint MIDR_EL1 *yet*. But it also means that we're going to have a hard time emulating another CPU (such as A57) on top of ThunderX. Probably not a big deal at the moment...

	M.
Signature · 1 line
Jazz is not dead. It just smells funny...
#8 lore raw

Re: [PATCH 3/4] irqchip, gicv3: Implement Cavium ThunderX erratum 23154

In reply to Robert Richter

drivers/irqchip/irq-gic-v3.c static void gic_redist_wait_for_rwp(void) · line 127

Quoted 9 lines · Robert Richter
drivers/irqchip/irq-gic-v3.c +5 -0
+static u64 gic_read_iar_cavium_thunderx(void)
 {
 	u64 irqstat;
 
+	asm volatile("nop;nop;nop;nop;");
+	asm volatile("nop;nop;nop;nop;");
 	asm volatile("mrs_s %0, " __stringify(ICC_IAR1_EL1) : "=r" (irqstat));
+	asm volatile("nop;nop;nop;nop;");
+	mb();

NAK. Please read the GCC manual for proper use of asm(). Even with "volatile" there, it doesn't stop GCC from inserting instructions between these.

If you need an instruction sequence to be consecutive, then it _must_ be one single asm().

There should also be a comment explaining why that code is necessary here. Please think about the future when you've forgotten why those nops are there. The kernel is a long term project, and it's important that people understand why things are coded the way they are so that the kernel can be properly maintained into the future.

Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.
#9 lore raw

Re: [PATCH 3/4] irqchip, gicv3: Implement Cavium ThunderX erratum 23154

Russell,

thanks for your comments.

Quoted 15 lines · Russell King - ARM Linux

drivers/irqchip/irq-gic-v3.c static void gic_redist_wait_for_rwp(void) · line 127

Quoted 9 lines · Robert Richter
drivers/irqchip/irq-gic-v3.c +5 -0
+static u64 gic_read_iar_cavium_thunderx(void)
 {
 	u64 irqstat;
 
+	asm volatile("nop;nop;nop;nop;");
+	asm volatile("nop;nop;nop;nop;");
 	asm volatile("mrs_s %0, " __stringify(ICC_IAR1_EL1) : "=r" (irqstat));
+	asm volatile("nop;nop;nop;nop;");
+	mb();

NAK. Please read the GCC manual for proper use of asm(). Even with "volatile" there, it doesn't stop GCC from inserting instructions between these.

Quoted 2 lines

If you need an instruction sequence to be consecutive, then it _must_ be one single asm().

I will update this section.

Quoted 5 lines

There should also be a comment explaining why that code is necessary here. Please think about the future when you've forgotten why those nops are there. The kernel is a long term project, and it's important that people understand why things are coded the way they are so that the kernel can be properly maintained into the future.

I will add a comment to the code with a brief description. Would the current text from the patch description including the erratum number be sufficient for that?

Thanks,

-Robert

#10 lore raw

PATCH 4/4 irqchip, gicv3-its: Implement Cavium ThunderX errata 22375, 24313

In reply to Robert Richter

From: Robert Richter <[email protected]>

This implements two gicv3-its errata for ThunderX. Both with small impact affecting only ITS table allocation.

erratum 22375: only alloc 8MB table size
erratum 24313: ignore memory access type

The fixes are in ITS initialization and basically ignore memory access type and table size provided by the TYPER and BASER registers.

---
 drivers/irqchip/irq-gic-v3-its.c | 35 +++++++++++++++++++++++++++++++----
 1 file changed, 31 insertions(+), 4 deletions(-)
drivers/irqchip/irq-gic-v3-its.c +31 -4
index de2fc82bba83..049ef761fbc3 100644
@@ -39,7 +39,8 @@
 #include "irq-gic-common.h"
 #include "irqchip.h"
 
-#define ITS_FLAGS_CMDQ_NEEDS_FLUSHING		(1 << 0)
+#define ITS_FLAGS_CMDQ_NEEDS_FLUSHING		(1ULL << 0)
+#define ITS_FLAGS_CAVIUM_THUNDERX		(1ULL << 1)
 
 #define RDIST_FLAGS_PROPBASE_NEEDS_FLUSHING	(1 << 0)
 
@@ -803,9 +804,22 @@ static int its_alloc_tables(struct its_node *its)
 	int i;
 	int psz = SZ_64K;
 	u64 shr = GITS_BASER_InnerShareable;
-	u64 cache = GITS_BASER_WaWb;
-	u64 typer = readq_relaxed(its->base + GITS_TYPER);
-	u32 ids = GITS_TYPER_DEVBITS(typer);
+	u64 cache;
+	u64 typer;
+	u32 ids;
+
+	if (its->flags & ITS_FLAGS_CAVIUM_THUNDERX) {
+		/*
+		 * erratum 22375: only alloc 8MB table size
+		 * erratum 24313: ignore memory access type
+		 */
+		cache	= 0;
+		ids	= 0x13;			/* 20 bits, 8MB */
+	} else {
+		cache	= GITS_BASER_WaWb;
+		typer	= readq_relaxed(its->base + GITS_TYPER);
+		ids	= GITS_TYPER_DEVBITS(typer);
+	}
 
 	for (i = 0; i < GITS_BASER_NR_REGS; i++) {
 		u64 val = readq_relaxed(its->base + GITS_BASER + i * 8);
@@ -1382,8 +1396,21 @@ static int its_force_quiescent(void __iomem *base)
 	}
 }
 
+static void its_enable_cavium_thunderx(void *data)
+{
+	struct its_node *its = data;
+
+	its->flags |= ITS_FLAGS_CAVIUM_THUNDERX;
+}
+
 static const struct gic_capabilities its_errata[] = {
 	{
+		.desc	= "ITS: Cavium errata 22375, 24313",
+		.id	= 0xa100034c,	/* ThunderX pass 1.x */
+		.mask	= 0xffff0fff,
+		.init	= its_enable_cavium_thunderx,
+	},
+	{
 	}
 };
 
Signature · 1 line
2.1.1
#11 lore raw

PATCH 4/4 irqchip, gicv3-its: Implement Cavium ThunderX errata 22375, 24313

Hello!

Quoted 4 lines

This implements two gicv3-its errata for ThunderX. Both with small impact affecting only ITS table allocation.

erratum 22375: only alloc 8MB table size

I have a simpler solution to this one:
---
For some reason on ThunderX GITS_BASER reports entry size == 8 for ITT, while real size, reported by GITS_TYPER is 4. Consequently, allocated table size appears twice bigger than expected. And, if we set GITS_BASER.SIZE accordingly (the value appears to be 0xFF instead of 0x7F) the ITS will not work.

---
 drivers/irqchip/irq-gic-v3-its.c   | 3 ++-
 include/linux/irqchip/arm-gic-v3.h | 2 ++
 2 files changed, 4 insertions(+), 1 deletion(-)
drivers/irqchip/irq-gic-v3-its.c +2 -1
index ed70163..c420f17 100644
@@ -827,6 +827,7 @@ static int its_alloc_tables(struct its_node *its)
 		if (type == GITS_BASER_TYPE_DEVICE) {
 			u64 typer = readq_relaxed(its->base + GITS_TYPER);
 			u32 ids = GITS_TYPER_DEVBITS(typer);
+			u32 itte_size = GITS_TYPER_ITTE_SIZE(typer);
 
 			/*
 			 * 'order' was initialized earlier to the default page
@@ -834,7 +835,7 @@ static int its_alloc_tables(struct its_node *its)
 			 * smaller than that.  If the requested allocation
 			 * is smaller, round up to the default page granule.
 			 */
-			order = max(get_order((1UL << ids) * entry_size),
+			order = max(get_order((1UL << ids) * itte_size),
 				    order);
 			if (order >= MAX_ORDER) {
 				order = MAX_ORDER - 1;
include/linux/irqchip/arm-gic-v3.h +2 -0
index ffbc034..7f515e6 100644
@@ -184,6 +184,8 @@
 #define GITS_CTLR_ENABLE		(1U << 0)
 #define GITS_CTLR_QUIESCENT		(1U << 31)
 
+#define GITS_TYPER_ITTE_SIZE_SHIFT	4
+#define GITS_TYPER_ITTE_SIZE(r)		((((r) >> GITS_TYPER_ITTE_SIZE_SHIFT) & 0xf) + 1)
 #define GITS_TYPER_DEVBITS_SHIFT	13
 #define GITS_TYPER_DEVBITS(r)		((((r) >> GITS_TYPER_DEVBITS_SHIFT) & 0x1f) + 1)
 #define GITS_TYPER_PTA			(1UL << 19)
Signature · 7 lines
2.4.4


Kind regards,
Pavel Fedin
Expert Engineer
Samsung Electronics Research center Russia
#12 lore raw

PATCH 4/4 irqchip, gicv3-its: Implement Cavium ThunderX errata 22375, 24313

Quoted 34 lines · Pavel Fedin

Hello!

Quoted 4 lines

This implements two gicv3-its errata for ThunderX. Both with small impact affecting only ITS table allocation.

erratum 22375: only alloc 8MB table size

I have a simpler solution to this one:
---
For some reason on ThunderX GITS_BASER reports entry size == 8 for ITT, while real size, reported by GITS_TYPER is 4. Consequently, allocated table size appears twice bigger than expected. And, if we set GITS_BASER.SIZE accordingly (the value appears to be 0xFF instead of 0x7F) the ITS will not work.

---
 drivers/irqchip/irq-gic-v3-its.c   | 3 ++-
 include/linux/irqchip/arm-gic-v3.h | 2 ++
 2 files changed, 4 insertions(+), 1 deletion(-)
drivers/irqchip/irq-gic-v3-its.c +1 -0
index ed70163..c420f17 100644
@@ -827,6 +827,7 @@ static int its_alloc_tables(struct its_node *its)
 		if (type == GITS_BASER_TYPE_DEVICE) {
 			u64 typer = readq_relaxed(its->base + GITS_TYPER);
 			u32 ids = GITS_TYPER_DEVBITS(typer);
+			u32 itte_size = GITS_TYPER_ITTE_SIZE(typer);
 
 			/*
 			 * 'order' was initialized earlier to the default page

drivers/irqchip/irq-gic-v3-its.c static int its_alloc_tables(struct its_node *its) · line 838

Quoted 5 lines
drivers/irqchip/irq-gic-v3-its.c +1 -1
@@ -834,7 +835,7 @@ static int its_alloc_tables(struct its_node *its)
 			 * smaller than that.  If the requested allocation
 			 * is smaller, round up to the default page granule.
 			 */
-			order = max(get_order((1UL << ids) * entry_size),
+			order = max(get_order((1UL << ids) * itte_size),

What makes you think that an entry in the device table is the same size as a translation entry? Just because one contains a pointer to the other doesn't imply they have the same size.

	M.
Signature · 1 line
Jazz is not dead. It just smells funny...
#13 lore raw

PATCH 4/4 irqchip, gicv3-its: Implement Cavium ThunderX errata 22375, 24313

Hello!

Quoted 3 lines

What makes you think that an entry in the device table is the same size as a translation entry? Just because one contains a pointer to the other doesn't imply they have the same size.

Oops... I didn't have GICv3 arch reference manual back then, so i was confused. Ignore this then.
:)

Kind regards,
Pavel Fedin
Expert Engineer
Samsung Electronics Research center Russia