Skip to contentSkip to conversation

genirq: Make nr_irqs runtime expandable

tip:irq/core Yinghai Lu 3 participants 7 messages Feb 19, 2011 – Feb 21, 2011 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore

Yinghai Lu

author
#1 lore raw
In reply to a message not in the archive
Quoted 58 lines · tip-bot for Thomas Gleixner
Commit-ID:  e7bcecb7b1d29b9ad5af939149a945658620ca8f
Gitweb:     http://git.kernel.org/tip/e7bcecb7b1d29b9ad5af939149a945658620ca8f
Author:     Thomas Gleixner <[email protected]>
AuthorDate: Wed, 16 Feb 2011 17:12:57 +0100
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Sat, 19 Feb 2011 12:58:06 +0100

genirq: Make nr_irqs runtime expandable

We face more and more the requirement to expand nr_irqs at runtime. The reason are irq expanders which can not be detected in the early boot stage. So we speculate nr_irqs to have enough room. Further Xen needs extra irq numbers and we really want to avoid adding more "detection" code into the early boot. There is no real good reason why we need to limit nr_irqs at early boot.

Allow the allocation code to expand nr_irqs. We have already 8k extra number space in the allocation bitmap, so lets use it.

---
 kernel/irq/irqdesc.c |   22 +++++++++++++++++++---
 1 files changed, 19 insertions(+), 3 deletions(-)
kernel/irq/irqdesc.c +14 -0
index a250d3a..6f6644f 100644
@@ -206,6 +206,14 @@ struct irq_desc * __ref irq_to_desc_alloc_node(unsigned int irq, int node)
 	return NULL;
 }
 
+static int irq_expand_nr_irqs(unsigned int cnt)
+{
+	if (nr_irqs + cnt > IRQ_BITMAP_BITS)
+		return -ENOMEM;
+	nr_irqs += cnt;
+	return 0;
+}
+
 int __init early_irq_init(void)
 {
 	int i, initcnt, node = first_online_node;
@@ -287,6 +295,12 @@ static inline int alloc_descs(unsigned int start, unsigned int cnt, int node)
 {
 	return start;
 }
+
+static int irq_expand_nr_irqs(unsigned int cnt)
+{
+	return -ENOMEM;
+}
+
 #endif /* !CONFIG_SPARSE_IRQ */
 
 /* Dynamic interrupt handling */
@@ -335,9 +349,11 @@ irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node)
 	if (irq >=0 && start != irq)

kernel/irq/irqdesc.c irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node) · line 356

Quoted 10 lines
kernel/irq/irqdesc.c +5 -3
 		goto err;
 
-	ret = -ENOMEM;
-	if (start >= nr_irqs)
-		goto err;
+	if (start >= nr_irqs) {
+		ret = irq_expand_nr_irqs(cnt);
+		if (ret)
+			goto err;
+	}

may need following patch.

[PATCH] genirq: use IRQ_BITMAP_BITS as search size

instead of nr_irqs.

Otherwise bitmap_find_next_area could exit with larger start and in extreme case we could fail to get wrong irqs return.

For example:
IRQ_BITMAP_BITS=10240
nr_irqs=8192
cnt=2048

and bit 0 to bit 8190 are set already.

before patch start from bit_find_next_area() will be 8191+2048. later irq_expand_nr_irqs will set nr_irqs 10240.
finally irq_alloc_descs will return [8191+2048, 8191+2048+2047] happily..

with this patch, will get correct [8191, 8191+2047]

---
 kernel/irq/irqdesc.c |    5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
kernel/irq/irqdesc.c +3 -2
Index: linux-2.6/kernel/irq/irqdesc.c
===================================================================
@@ -346,12 +346,13 @@ irq_alloc_descs(int irq, unsigned int fr
 
 	mutex_lock(&sparse_irq_lock);
 
-	start = bitmap_find_next_zero_area(allocated_irqs, nr_irqs, from, cnt, 0);
+	start = bitmap_find_next_zero_area(allocated_irqs, IRQ_BITMAP_BITS,
+						 from, cnt, 0);
 	ret = -EEXIST;
 	if (irq >=0 && start != irq)
 		goto err;
 
-	if (start >= nr_irqs) {
+	if (start + cnt > nr_irqs) {
 		ret = irq_expand_nr_irqs(cnt);
 		if (ret)
 			goto err;
#2 lore raw
Quoted 13 lines · Yinghai Lu

On 02/19/2011 04:15 AM, tip-bot for Thomas Gleixner wrote:
[PATCH] genirq: use IRQ_BITMAP_BITS as search size

instead of nr_irqs.

Otherwise bitmap_find_next_area could exit with larger start and in extreme case we could fail to get wrong irqs return.

For example:
IRQ_BITMAP_BITS=10240
nr_irqs=8192
cnt=2048

Quoted 7 lines

and bit 0 to bit 8190 are set already.

before patch start from bit_find_next_area() will be 8191+2048. later irq_expand_nr_irqs will set nr_irqs 10240.
finally irq_alloc_descs will return [8191+2048, 8191+2048+2047] happily..

with this patch, will get correct [8191, 8191+2047]

What a horrible changelog.

Quoted 13 lines
---
 kernel/irq/irqdesc.c |    5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

Index: linux-2.6/kernel/irq/irqdesc.c
===================================================================

kernel/irq/irqdesc.c +0 -0
@@ -346,12 +346,13 @@ irq_alloc_descs(int irq, unsigned int fr
 
 	mutex_lock(&sparse_irq_lock);

kernel/irq/irqdesc.c irq_alloc_descs(int irq, unsigned int fr · line 355

Quoted 10 lines
kernel/irq/irqdesc.c +3 -2
 
-	start = bitmap_find_next_zero_area(allocated_irqs, nr_irqs, from, cnt, 0);
+	start = bitmap_find_next_zero_area(allocated_irqs, IRQ_BITMAP_BITS,
+						 from, cnt, 0);
 	ret = -EEXIST;
 	if (irq >=0 && start != irq)
 		goto err;
 
-	if (start >= nr_irqs) {
+	if (start + cnt > nr_irqs) {

One off. That needs to be >= nr_irqs. Good catch otherwise.

Thanks,

tglx

#3 lore raw
Quoted 5 lines · Thomas Gleixner

kernel/irq/irqdesc.c irq_alloc_descs(int irq, unsigned int fr · line 355

Quoted 2 lines · Yinghai Lu
kernel/irq/irqdesc.c +1 -1
-	if (start >= nr_irqs) {
+	if (start + cnt > nr_irqs) {

One off. That needs to be >= nr_irqs. Good catch otherwise.

Confused myself. Check is correct as is.

#4 lore raw
In reply to Yinghai Lu
Quoted 106 lines · Yinghai Lu
Quoted 58 lines · tip-bot for Thomas Gleixner
Commit-ID:  e7bcecb7b1d29b9ad5af939149a945658620ca8f
Gitweb:     http://git.kernel.org/tip/e7bcecb7b1d29b9ad5af939149a945658620ca8f
Author:     Thomas Gleixner <[email protected]>
AuthorDate: Wed, 16 Feb 2011 17:12:57 +0100
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Sat, 19 Feb 2011 12:58:06 +0100

genirq: Make nr_irqs runtime expandable

We face more and more the requirement to expand nr_irqs at runtime. The reason are irq expanders which can not be detected in the early boot stage. So we speculate nr_irqs to have enough room. Further Xen needs extra irq numbers and we really want to avoid adding more "detection" code into the early boot. There is no real good reason why we need to limit nr_irqs at early boot.

Allow the allocation code to expand nr_irqs. We have already 8k extra number space in the allocation bitmap, so lets use it.

---
 kernel/irq/irqdesc.c |   22 +++++++++++++++++++---
 1 files changed, 19 insertions(+), 3 deletions(-)
kernel/irq/irqdesc.c +14 -0
index a250d3a..6f6644f 100644
@@ -206,6 +206,14 @@ struct irq_desc * __ref irq_to_desc_alloc_node(unsigned int irq, int node)
 	return NULL;
 }
 
+static int irq_expand_nr_irqs(unsigned int cnt)
+{
+	if (nr_irqs + cnt > IRQ_BITMAP_BITS)
+		return -ENOMEM;
+	nr_irqs += cnt;
+	return 0;
+}
+
 int __init early_irq_init(void)
 {
 	int i, initcnt, node = first_online_node;
@@ -287,6 +295,12 @@ static inline int alloc_descs(unsigned int start, unsigned int cnt, int node)
 {
 	return start;
 }
+
+static int irq_expand_nr_irqs(unsigned int cnt)
+{
+	return -ENOMEM;
+}
+
 #endif /* !CONFIG_SPARSE_IRQ */
 
 /* Dynamic interrupt handling */
@@ -335,9 +349,11 @@ irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node)
 	if (irq >=0 && start != irq)

kernel/irq/irqdesc.c irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node) · line 356

Quoted 10 lines
kernel/irq/irqdesc.c +5 -3
 		goto err;
 
-	ret = -ENOMEM;
-	if (start >= nr_irqs)
-		goto err;
+	if (start >= nr_irqs) {
+		ret = irq_expand_nr_irqs(cnt);
+		if (ret)
+			goto err;
+	}

may need following patch.

[PATCH] genirq: use IRQ_BITMAP_BITS as search size

instead of nr_irqs.

Otherwise bitmap_find_next_area could exit with larger start and in extreme case we could fail to get wrong irqs return.

For example:
IRQ_BITMAP_BITS=10240
nr_irqs=8192
cnt=2048

and bit 0 to bit 8190 are set already.

before patch start from bit_find_next_area() will be 8191+2048. later irq_expand_nr_irqs will set nr_irqs 10240.
finally irq_alloc_descs will return [8191+2048, 8191+2048+2047] happily..

with this patch, will get correct [8191, 8191+2047]

---
 kernel/irq/irqdesc.c |    5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

Index: linux-2.6/kernel/irq/irqdesc.c
===================================================================

kernel/irq/irqdesc.c +0 -0
@@ -346,12 +346,13 @@ irq_alloc_descs(int irq, unsigned int fr
 
 	mutex_lock(&sparse_irq_lock);
 

kernel/irq/irqdesc.c irq_alloc_descs(int irq, unsigned int fr · line 356

Quoted 10 lines
kernel/irq/irqdesc.c +3 -2
-	start = bitmap_find_next_zero_area(allocated_irqs, nr_irqs, from, cnt, 0);
+	start = bitmap_find_next_zero_area(allocated_irqs, IRQ_BITMAP_BITS,
+						 from, cnt, 0);
 	ret = -EEXIST;
 	if (irq >=0 && start != irq)
 		goto err;
 
-	if (start >= nr_irqs) {
+	if (start + cnt > nr_irqs) {
 		ret = irq_expand_nr_irqs(cnt);

Just a minor thing, but if there are still unused irqs available at the end of the current range, you'll end up expanding the range more then you need to. So either do

	irq_expand_nr_irqs(nr_irqs - start + cnt);

or change irq_expand_nr_irqs to let it take the new total number of irqs.

Btw., with this patch in place does it make sense to initialize nr_irqs to anything else then initcnt in early_irq_init?

Quoted 7 lines
kernel/irq/irqdesc.c +0 -1
 		if (ret)
 			goto err;
--

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/

author
#5 lore raw
Quoted 1 line · Lars-Peter Clausen

On 02/19/2011 08:07 PM, Yinghai Lu wrote:

...

Quoted 41 lines
Quoted 41 lines

[PATCH] genirq: use IRQ_BITMAP_BITS as search size

instead of nr_irqs.

Otherwise bitmap_find_next_area could exit with larger start and in extreme case we could fail to get wrong irqs return.

For example:
IRQ_BITMAP_BITS=10240
nr_irqs=8192
cnt=2048

and bit 0 to bit 8190 are set already.

before patch start from bit_find_next_area() will be 8191+2048. later irq_expand_nr_irqs will set nr_irqs 10240.
finally irq_alloc_descs will return [8191+2048, 8191+2048+2047] happily..

with this patch, will get correct [8191, 8191+2047]

---
 kernel/irq/irqdesc.c |    5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

Index: linux-2.6/kernel/irq/irqdesc.c
===================================================================

kernel/irq/irqdesc.c +2 -1
@@ -346,12 +346,13 @@ irq_alloc_descs(int irq, unsigned int fr
 
 	mutex_lock(&sparse_irq_lock);
 
-	start = bitmap_find_next_zero_area(allocated_irqs, nr_irqs, from, cnt, 0);
+	start = bitmap_find_next_zero_area(allocated_irqs, IRQ_BITMAP_BITS,
+						 from, cnt, 0);
 	ret = -EEXIST;
 	if (irq >=0 && start != irq)
 		goto err;
 
Quoted 8 lines

kernel/irq/irqdesc.c irq_alloc_descs(int irq, unsigned int fr · line 356

Quoted 3 lines
kernel/irq/irqdesc.c +1 -1
-	if (start >= nr_irqs) {
+	if (start + cnt > nr_irqs) {
 		ret = irq_expand_nr_irqs(cnt);

Just a minor thing, but if there are still unused irqs available at the end of the current range, you'll end up expanding the range more then you need to. So either do

	irq_expand_nr_irqs(nr_irqs - start + cnt);

or change irq_expand_nr_irqs to let it take the new total number of irqs.

yes. please check second way that you suggested.

[PATCH -v2] genirq: use IRQ_BITMAP_BITS as search size

Otherwise bitmap_find_next_area could exit with larger start and in extreme case we could fail to get wrong irqs return.

For example:
IRQ_BITMAP_BITS=10240
nr_irqs=8192
cnt=2048

and bit 0 to bit 8190 are set already.

before patch start from bit_find_next_area() will be 8191+2048. later irq_expand_nr_irqs will set nr_irqs 10240.
finally irq_alloc_descs will return [8191+2048, 8191+2048+2047] happily..

with this patch, will get correct [8191, 8191+2047]

-v2: let irq_expand_nr_irqs take new nr_irqs instead of cnt, so do not
increase nr_irqs too much extra.
Suggested by "Lars-Peter" Clausen <[email protected]>

---
 kernel/irq/irqdesc.c |   15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)
kernel/irq/irqdesc.c +8 -7
Index: linux-2.6/kernel/irq/irqdesc.c
===================================================================
@@ -207,11 +207,11 @@ struct irq_desc * __ref irq_to_desc_allo
 	return NULL;
 }
 
-static int irq_expand_nr_irqs(unsigned int cnt)
+static int irq_expand_nr_irqs(int new_nr_irqs)
 {
-	if (nr_irqs + cnt > IRQ_BITMAP_BITS)
+	if (new_nr_irqs > IRQ_BITMAP_BITS)
 		return -ENOMEM;
-	nr_irqs += cnt;
+	nr_irqs = new_nr_irqs;
 	return 0;
 }
 
@@ -298,7 +298,7 @@ static inline int alloc_descs(unsigned i
 	return start;
 }
 
-static int irq_expand_nr_irqs(unsigned int cnt)
+static int irq_expand_nr_irqs(int new_nr_irqs)
 {
 	return -ENOMEM;
 }
@@ -346,13 +346,14 @@ irq_alloc_descs(int irq, unsigned int fr
 
 	mutex_lock(&sparse_irq_lock);
 
-	start = bitmap_find_next_zero_area(allocated_irqs, nr_irqs, from, cnt, 0);
+	start = bitmap_find_next_zero_area(allocated_irqs, IRQ_BITMAP_BITS,
+						 from, cnt, 0);
 	ret = -EEXIST;
 	if (irq >=0 && start != irq)
 		goto err;
 
-	if (start >= nr_irqs) {
-		ret = irq_expand_nr_irqs(cnt);
+	if (start + cnt > nr_irqs) {
+		ret = irq_expand_nr_irqs(start + cnt);
 		if (ret)
 			goto err;
 	}
Quoted 3 lines

Btw., with this patch in place does it make sense to initialize nr_irqs to anything else then initcnt in early_irq_init?

i have one for x86 and it will set nr_irqs to nr_irqs_gsi

[PATCH] x86, irq: keep nr_irqs as low as possible

For sparseirq, We can expand nr_irqs later if needed, so could keep it from small for beginning.

Assign it to minium value of vectors of all cpus and gsi pins

---
 arch/x86/kernel/apic/io_apic.c |   13 ++++---------
 1 file changed, 4 insertions(+), 9 deletions(-)
arch/x86/kernel/apic/io_apic.c +4 -9
Index: linux-2.6/arch/x86/kernel/apic/io_apic.c
===================================================================
@@ -3642,16 +3642,11 @@ int __init arch_probe_nr_irqs(void)
 {
 	int nr;
 
-	if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
-		nr_irqs = NR_VECTORS * nr_cpu_ids;
+	nr = NR_VECTORS * nr_cpu_ids;
+	if (nr < nr_irqs)
+		nr_irqs = nr;
 
-	nr = nr_irqs_gsi + 8 * nr_cpu_ids;
-#if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
-	/*
-	 * for MSI and HT dyn irq
-	 */
-	nr += nr_irqs_gsi * 16;
-#endif
+	nr = nr_irqs_gsi;
 	if (nr < nr_irqs)
 		nr_irqs = nr;
 
#6 lore raw
In reply to Lars-Peter Clausen
Quoted 10 lines · Lars-Peter Clausen

kernel/irq/irqdesc.c irq_alloc_descs(int irq, unsigned int fr · line 356

Quoted 4 lines · Yinghai Lu

On 02/19/2011 04:15 AM, tip-bot for Thomas Gleixner wrote:

kernel/irq/irqdesc.c +1 -1
-	if (start >= nr_irqs) {
+	if (start + cnt > nr_irqs) {
 		ret = irq_expand_nr_irqs(cnt);

Just a minor thing, but if there are still unused irqs available at the end of the current range, you'll end up expanding the range more then you need to. So either do

	irq_expand_nr_irqs(nr_irqs - start + cnt);

or change irq_expand_nr_irqs to let it take the new total number of irqs.

Fun, I already changed YH's patch in that way :)

Quoted 2 lines

Btw., with this patch in place does it make sense to initialize nr_irqs to anything else then initcnt in early_irq_init?

Probably not.

Thanks,

tglx

#7 lore raw

[tip:irq/core] genirq: Use IRQ_BITMAP_BITS as search size in irq_alloc_descs()

In reply to Yinghai Lu
Commit-ID:  ed4dea6e0e33a3e58d8b77b775a8f0e433e7a005
Gitweb:     http://git.kernel.org/tip/ed4dea6e0e33a3e58d8b77b775a8f0e433e7a005
Author:     Yinghai Lu <[email protected]>
AuthorDate: Sat, 19 Feb 2011 11:07:37 -0800
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Mon, 21 Feb 2011 21:20:00 +0100

genirq: Use IRQ_BITMAP_BITS as search size in irq_alloc_descs()

The runtime expansion of nr_irqs does not take into account that bitmap_find_next_zero_area() returns "start" + size in case the search for an matching zero area fails. That results in a start value which can be completely off and is not covered by the following expand_nr_irqs() and possibly outside of the absolute limit. But we use it without further checking.

Use IRQ_BITMAP_BITS as the limit for the bitmap search and expand nr_irqs when the start bit is beyond nr_irqs. So start is always pointing to the correct area in the bitmap. nr_irqs is just the limit for irq enumerations, not the real limit for the irq space.

[ tglx: Let irq_expand_nr_irqs() take the new upper end so we do not
expand nr_irqs more than necessary. Made changelog readable ]

LKML-Reference: <[email protected]>

---
 kernel/irq/irqdesc.c |   15 ++++++++-------
 1 files changed, 8 insertions(+), 7 deletions(-)
kernel/irq/irqdesc.c +8 -7
index 394ab6a..dbccc79 100644
@@ -207,11 +207,11 @@ struct irq_desc * __ref irq_to_desc_alloc_node(unsigned int irq, int node)
 	return NULL;
 }
 
-static int irq_expand_nr_irqs(unsigned int cnt)
+static int irq_expand_nr_irqs(unsigned int nr)
 {
-	if (nr_irqs + cnt > IRQ_BITMAP_BITS)
+	if (nr > IRQ_BITMAP_BITS)
 		return -ENOMEM;
-	nr_irqs += cnt;
+	nr_irqs = nr;
 	return 0;
 }
 
@@ -298,7 +298,7 @@ static inline int alloc_descs(unsigned int start, unsigned int cnt, int node)
 	return start;
 }
 
-static int irq_expand_nr_irqs(unsigned int cnt)
+static int irq_expand_nr_irqs(unsigned int nr)
 {
 	return -ENOMEM;
 }
@@ -346,13 +346,14 @@ irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node)
 
 	mutex_lock(&sparse_irq_lock);
 
-	start = bitmap_find_next_zero_area(allocated_irqs, nr_irqs, from, cnt, 0);
+	start = bitmap_find_next_zero_area(allocated_irqs, IRQ_BITMAP_BITS,
+					   from, cnt, 0);
 	ret = -EEXIST;
 	if (irq >=0 && start != irq)
 		goto err;
 
-	if (start >= nr_irqs) {
-		ret = irq_expand_nr_irqs(cnt);
+	if (start + cnt > nr_irqs) {
+		ret = irq_expand_nr_irqs(start + cnt);
 		if (ret)
 			goto err;
 	}