Skip to contentSkip to conversation

xen/p2m: fix extra memory regions accounting

PATCH In review Roger Pau Monne 3 participants 34 messages Sep 3, 2015 – Sep 4, 2015

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xen/p2m: fix extra memory regions accounting

On systems with memory maps with ranges that don't end at page boundaries, like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00, but the function used to check if a memory address is inside of a protected range works with pfns, which means that an attempt to map 0xdfdf9c00 will be refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
 arch/x86/xen/setup.c | 6 ++++++
 1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +6 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
 {
 	int i;
 
+	start = PAGE_ALIGN(start);
+	size &= PAGE_MASK;
+
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 		/* Add new region. */
 		if (xen_extra_mem[i].size == 0) {
@@ -92,6 +95,9 @@ static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
 	int i;
 	phys_addr_t start_r, size_r;
 
+	start = PAGE_ALIGN(start);
+	size &= PAGE_MASK;
+
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 		start_r = xen_extra_mem[i].start;
 		size_r = xen_extra_mem[i].size;
Signature · 1 line
1.9.5 (Apple Git-50.3)
#1 lore raw

El 03/09/15 a les 14.05, Roger Pau Monne ha escrit:

Quoted 34 lines

On systems with memory maps with ranges that don't end at page boundaries, like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00, but the function used to check if a memory address is inside of a protected range works with pfns, which means that an attempt to map 0xdfdf9c00 will be refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
 arch/x86/xen/setup.c | 6 ++++++
 1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size) · line 71

Quoted 4 lines
arch/x86/xen/setup.c +1 -0
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
 {
 	int i;
 
+	start = PAGE_ALIGN(start);

Sorry, I didn't remember to update the patch before sending it, but I think this should instead be:

start = round_up(start, PAGE_SIZE);

Here and below.

Quoted 7 lines
arch/x86/xen/setup.c +2 -0
+	size &= PAGE_MASK;
+
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 		/* Add new region. */
 		if (xen_extra_mem[i].size == 0) {

arch/x86/xen/setup.c static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size) · line 103

Quoted 9 lines
arch/x86/xen/setup.c +3 -0
@@ -92,6 +95,9 @@ static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
 	int i;
 	phys_addr_t start_r, size_r;
 
+	start = PAGE_ALIGN(start);
+	size &= PAGE_MASK;
+
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 		start_r = xen_extra_mem[i].start;
 		size_r = xen_extra_mem[i].size;

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

#2 lore raw
In reply to Roger Pau Monne

El 03/09/15 a les 14.05, Roger Pau Monne ha escrit:

Quoted 34 lines

On systems with memory maps with ranges that don't end at page boundaries, like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00, but the function used to check if a memory address is inside of a protected range works with pfns, which means that an attempt to map 0xdfdf9c00 will be refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
 arch/x86/xen/setup.c | 6 ++++++
 1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size) · line 71

Quoted 4 lines
arch/x86/xen/setup.c +1 -0
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
 {
 	int i;
 
+	start = PAGE_ALIGN(start);

Sorry, I didn't remember to update the patch before sending it, but I think this should instead be:

start = round_up(start, PAGE_SIZE);

Here and below.

Quoted 16 lines
arch/x86/xen/setup.c +5 -0
+	size &= PAGE_MASK;
+
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 		/* Add new region. */
 		if (xen_extra_mem[i].size == 0) {
@@ -92,6 +95,9 @@ static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
 	int i;
 	phys_addr_t start_r, size_r;
 
+	start = PAGE_ALIGN(start);
+	size &= PAGE_MASK;
+
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 		start_r = xen_extra_mem[i].start;
 		size_r = xen_extra_mem[i].size;
maintainer
#3 lore raw
Quoted 38 lines · Roger Pau Monné

El 03/09/15 a les 14.05, Roger Pau Monne ha escrit:

Quoted 37 lines

On systems with memory maps with ranges that don't end at page boundaries, like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00, but the function used to check if a memory address is inside of a protected range works with pfns, which means that an attempt to map 0xdfdf9c00 will be refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
  arch/x86/xen/setup.c | 6 ++++++
  1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
  {
  	int i;
 
Quoted 6 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size) · line 71

Quoted 1 line
arch/x86/xen/setup.c +1 -0
+	start = PAGE_ALIGN(start);

Sorry, I didn't remember to update the patch before sending it, but I think this should instead be:

start = round_up(start, PAGE_SIZE);

No, PAGE_ALIGN() already does that.

Juergen

Quoted 12 lines

Here and below.

Quoted 9 lines
arch/x86/xen/setup.c +2 -0
+	size &= PAGE_MASK;
+
  	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
  		/* Add new region. */
  		if (xen_extra_mem[i].size == 0) {
@@ -92,6 +95,9 @@ static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
  	int i;
  	phys_addr_t start_r, size_r;
 
Quoted 14 lines

arch/x86/xen/setup.c static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size) · line 103

Quoted 7 lines
arch/x86/xen/setup.c +3 -0
+	start = PAGE_ALIGN(start);
+	size &= PAGE_MASK;
+
  	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
  		start_r = xen_extra_mem[i].start;
  		size_r = xen_extra_mem[i].size;

--
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/

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

maintainer
#4 lore raw
In reply to Roger Pau Monné
Quoted 38 lines · Roger Pau Monné

El 03/09/15 a les 14.05, Roger Pau Monne ha escrit:

Quoted 37 lines

On systems with memory maps with ranges that don't end at page boundaries, like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00, but the function used to check if a memory address is inside of a protected range works with pfns, which means that an attempt to map 0xdfdf9c00 will be refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
  arch/x86/xen/setup.c | 6 ++++++
  1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
  {
  	int i;
 
Quoted 6 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size) · line 71

Quoted 1 line
arch/x86/xen/setup.c +1 -0
+	start = PAGE_ALIGN(start);

Sorry, I didn't remember to update the patch before sending it, but I think this should instead be:

start = round_up(start, PAGE_SIZE);

No, PAGE_ALIGN() already does that.

Juergen

Quoted 26 lines

Here and below.

Quoted 7 lines
arch/x86/xen/setup.c +2 -0
+	size &= PAGE_MASK;
+
  	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
  		/* Add new region. */
  		if (xen_extra_mem[i].size == 0) {

arch/x86/xen/setup.c static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size) · line 103

Quoted 9 lines
arch/x86/xen/setup.c +3 -0
@@ -92,6 +95,9 @@ static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
  	int i;
  	phys_addr_t start_r, size_r;
 
+	start = PAGE_ALIGN(start);
+	size &= PAGE_MASK;
+
  	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
  		start_r = xen_extra_mem[i].start;
  		size_r = xen_extra_mem[i].size;

--
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/

maintainer
#5 lore raw
In reply to Roger Pau Monne
Quoted 34 lines · Roger Pau Monne

On systems with memory maps with ranges that don't end at page boundaries, like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00, but the function used to check if a memory address is inside of a protected range works with pfns, which means that an attempt to map 0xdfdf9c00 will be refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
  arch/x86/xen/setup.c | 6 ++++++
  1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size) · line 72

Quoted 5 lines
arch/x86/xen/setup.c +2 -0
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
  {
  	int i;
 
+	start = PAGE_ALIGN(start);
+	size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

arch/x86/xen/setup.c static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size) · line 99

Quoted 10 lines
arch/x86/xen/setup.c +3 -0
+
  	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
  		/* Add new region. */
  		if (xen_extra_mem[i].size == 0) {
@@ -92,6 +95,9 @@ static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
  	int i;
  	phys_addr_t start_r, size_r;
 
+	start = PAGE_ALIGN(start);
+	size &= PAGE_MASK;

Not really needed, as xen_del_extra_mem() is called with aligned values only. OTOH it does no harm, as long as you correct size as above.

arch/x86/xen/setup.c static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size) · line 103

Quoted 5 lines
arch/x86/xen/setup.c +1 -0
+
  	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
  		start_r = xen_extra_mem[i].start;
  		size_r = xen_extra_mem[i].size;

Juergen

#6 lore raw

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
  arch/x86/xen/setup.c | 6 ++++++
  1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
  {
      int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

Roger.

#7 lore raw
Quoted 55 lines · Roger Pau Monné

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
  arch/x86/xen/setup.c | 6 ++++++
  1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
  {
      int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

Quoted 7 lines

After the fixup added above this would become:

start = 0x9000
size = 0x1000

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add extra memory regions that cover these same areas.

See the use of PFN_DOWN() and PFN_UP() in xen_set_identify_and_remap().

David

#8 lore raw
Quoted 61 lines · David Vrabel
Quoted 60 lines · Roger Pau Monné

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
  arch/x86/xen/setup.c | 6 ++++++
  1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
  {
      int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 6 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add extra memory regions that cover these same areas.

Ignore this. This was nonsense.

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

David

maintainer
#9 lore raw
Quoted 69 lines · David Vrabel
Quoted 61 lines · David Vrabel
Quoted 60 lines · Roger Pau Monné

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
   arch/x86/xen/setup.c | 6 ++++++
   1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
   {
       int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 6 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add extra memory regions that cover these same areas.

Quoted 6 lines

Ignore this. This was nonsense.

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

Nearly. Roger's patch rounds up start and rounds down the size. It might add non-RAM partial pages to xen_extra_mem.

Juergen

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

maintainer
#10 lore raw
In reply to David Vrabel
Quoted 69 lines · David Vrabel
Quoted 61 lines · David Vrabel
Quoted 60 lines · Roger Pau Monné

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
   arch/x86/xen/setup.c | 6 ++++++
   1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
   {
       int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 6 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add extra memory regions that cover these same areas.

Quoted 6 lines

Ignore this. This was nonsense.

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

Nearly. Roger's patch rounds up start and rounds down the size. It might add non-RAM partial pages to xen_extra_mem.

Juergen

#11 lore raw
Quoted 72 lines · Juergen Gross
Quoted 71 lines · David Vrabel
Quoted 61 lines · David Vrabel
Quoted 60 lines · Roger Pau Monné

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
   arch/x86/xen/setup.c | 6 ++++++
   1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
   {
       int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 6 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add extra memory regions that cover these same areas.

Ignore this. This was nonsense.

Quoted 8 lines
Quoted 5 lines

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

Nearly. Roger's patch rounds up start and rounds down the size. It might add non-RAM partial pages to xen_extra_mem.

Yes. You're right.

David

maintainer
#12 lore raw
Quoted 79 lines · David Vrabel
Quoted 78 lines · Juergen Gross
Quoted 76 lines · David Vrabel
Quoted 61 lines · David Vrabel
Quoted 60 lines · Roger Pau Monné

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
    arch/x86/xen/setup.c | 6 ++++++
    1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
    {
        int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 6 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add extra memory regions that cover these same areas.

Ignore this. This was nonsense.

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

Quoted 4 lines
Quoted 2 lines

Nearly. Roger's patch rounds up start and rounds down the size. It might add non-RAM partial pages to xen_extra_mem.

Yes. You're right.

Hmm, thinking more about it, I'd prefer to change xen_extra_mem to use pfns instead of physical addresses. This would make things much more clear.

Roger, do you want to do the patch or should I?

Juergen

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

maintainer
#13 lore raw
In reply to David Vrabel
Quoted 79 lines · David Vrabel
Quoted 78 lines · Juergen Gross
Quoted 76 lines · David Vrabel
Quoted 61 lines · David Vrabel
Quoted 60 lines · Roger Pau Monné

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
    arch/x86/xen/setup.c | 6 ++++++
    1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
    {
        int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 6 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add extra memory regions that cover these same areas.

Ignore this. This was nonsense.

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

Quoted 4 lines
Quoted 2 lines

Nearly. Roger's patch rounds up start and rounds down the size. It might add non-RAM partial pages to xen_extra_mem.

Yes. You're right.

Hmm, thinking more about it, I'd prefer to change xen_extra_mem to use pfns instead of physical addresses. This would make things much more clear.

Roger, do you want to do the patch or should I?

Juergen

#14 lore raw

El 03/09/15 a les 17.20, Juergen Gross ha escrit:

Quoted 88 lines · David Vrabel
Quoted 87 lines · Juergen Gross
Quoted 85 lines · David Vrabel
Quoted 74 lines · David Vrabel
Quoted 65 lines · Roger Pau Monné
Quoted 64 lines

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 46 lines · Roger Pau Monne
Quoted 45 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00
will be
refused because the check is performed against 0xdfdf9000 instead of
0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the
xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
    arch/x86/xen/setup.c | 6 ++++++
    1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
    {
        int i;
 
Quoted 6 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll
add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 7 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should
add extra memory regions that cover these same areas.

Quoted 7 lines

Ignore this. This was nonsense.

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

Nearly. Roger's patch rounds up start and rounds down the size. It might add non-RAM partial pages to xen_extra_mem.

Quoted 7 lines
Quoted 1 line

Yes. You're right.

Hmm, thinking more about it, I'd prefer to change xen_extra_mem to use pfns instead of physical addresses. This would make things much more clear.

Roger, do you want to do the patch or should I?

I can certainly take care of it if you are busy, otherwise I leave it to you since you have more expertise on it :).

Roger.

maintainer
#15 lore raw
Quoted 91 lines · Roger Pau Monné

El 03/09/15 a les 17.20, Juergen Gross ha escrit:

Quoted 90 lines · David Vrabel
Quoted 88 lines · Juergen Gross
Quoted 77 lines · David Vrabel
Quoted 76 lines · David Vrabel
Quoted 65 lines · Roger Pau Monné
Quoted 64 lines

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 46 lines · Roger Pau Monne
Quoted 45 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00
will be
refused because the check is performed against 0xdfdf9000 instead of
0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the
xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
     arch/x86/xen/setup.c | 6 ++++++
     1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
     {
         int i;
 
Quoted 6 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll
add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 7 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should
add extra memory regions that cover these same areas.

Ignore this. This was nonsense.

Quoted 8 lines
Quoted 5 lines

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

Nearly. Roger's patch rounds up start and rounds down the size. It might add non-RAM partial pages to xen_extra_mem.

Yes. You're right.

Quoted 8 lines
Quoted 5 lines

Hmm, thinking more about it, I'd prefer to change xen_extra_mem to use pfns instead of physical addresses. This would make things much more clear.

Roger, do you want to do the patch or should I?

I can certainly take care of it if you are busy, otherwise I leave it to you since you have more expertise on it :).

Okay, I can do it. Should be ready soon...

Juergen

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

maintainer
#16 lore raw
In reply to Roger Pau Monné

Hi Roger,

Quoted 91 lines · Roger Pau Monné

El 03/09/15 a les 17.20, Juergen Gross ha escrit:

Quoted 90 lines · David Vrabel
Quoted 88 lines · Juergen Gross
Quoted 77 lines · David Vrabel
Quoted 76 lines · David Vrabel
Quoted 65 lines · Roger Pau Monné
Quoted 64 lines

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 46 lines · Roger Pau Monne
Quoted 45 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00
will be
refused because the check is performed against 0xdfdf9000 instead of
0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the
xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
     arch/x86/xen/setup.c | 6 ++++++
     1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
     {
         int i;
 
Quoted 6 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll
add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 7 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should
add extra memory regions that cover these same areas.

Ignore this. This was nonsense.

Quoted 8 lines
Quoted 5 lines

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

Nearly. Roger's patch rounds up start and rounds down the size. It might add non-RAM partial pages to xen_extra_mem.

Yes. You're right.

Quoted 8 lines
Quoted 5 lines

Hmm, thinking more about it, I'd prefer to change xen_extra_mem to use pfns instead of physical addresses. This would make things much more clear.

Roger, do you want to do the patch or should I?

I can certainly take care of it if you are busy, otherwise I leave it to you since you have more expertise on it :).

Could you try the attached patch? It should do the job. It is booting fine on my laptop, but I think you should try it on the machine with the memory ranges not at page boundary.

Juergen

commit 3d0f8aa4d1b4c9c16a81902a197b5d6a77e182a0

Author: Juergen Gross <[email protected]>
Date:   Thu Sep 3 17:44:04 2015 +0200

xen: switch extra memory accounting to use pfns

Instead of using physical addresses for accounting of extra memory areas available for ballooning switch to pfns as this is much less error prone regarding partial pages.

arch/x86/xen/setup.c +40 -36
index 7a5d566..aa58bc4 100644
@@ -90,62 +90,65 @@ static void __init xen_parse_512gb(void)
 	xen_512gb_limit = val;
 }
 
-static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
+static void __init xen_add_extra_mem(unsigned long start_pfn,
+				     unsigned long n_pfns)
 {
 	int i;
 
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 		/* Add new region. */
-		if (xen_extra_mem[i].size == 0) {
-			xen_extra_mem[i].start = start;
-			xen_extra_mem[i].size  = size;
+		if (xen_extra_mem[i].n_pfns == 0) {
+			xen_extra_mem[i].start_pfn = start_pfn;
+			xen_extra_mem[i].n_pfns = n_pfns;
 			break;
 		}
 		/* Append to existing region. */
-		if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
-			xen_extra_mem[i].size += size;
+		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
+		    start_pfn) {
+			xen_extra_mem[i].n_pfns += n_pfns;
 			break;
 		}
 	}
 	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
 		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
 
-	memblock_reserve(start, size);
+	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
 }
 
-static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
+static void __init xen_del_extra_mem(unsigned long start_pfn,
+				     unsigned long n_pfns)
 {
 	int i;
-	phys_addr_t start_r, size_r;
+	unsigned long start_r, size_r;
 
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
-		start_r = xen_extra_mem[i].start;
-		size_r = xen_extra_mem[i].size;
+		start_r = xen_extra_mem[i].start_pfn;
+		size_r = xen_extra_mem[i].n_pfns;
 
 		/* Start of region. */
-		if (start_r == start) {
-			BUG_ON(size > size_r);
-			xen_extra_mem[i].start += size;
-			xen_extra_mem[i].size -= size;
+		if (start_r == start_pfn) {
+			BUG_ON(n_pfns > size_r);
+			xen_extra_mem[i].start_pfn += n_pfns;
+			xen_extra_mem[i].n_pfns -= n_pfns;
 			break;
 		}
 		/* End of region. */
-		if (start_r + size_r == start + size) {
-			BUG_ON(size > size_r);
-			xen_extra_mem[i].size -= size;
+		if (start_r + size_r == start_pfn + n_pfns) {
+			BUG_ON(n_pfns > size_r);
+			xen_extra_mem[i].n_pfns -= n_pfns;
 			break;
 		}
 		/* Mid of region. */
-		if (start > start_r && start < start_r + size_r) {
-			BUG_ON(start + size > start_r + size_r);
-			xen_extra_mem[i].size = start - start_r;
+		if (start_pfn > start_r && start_pfn < start_r + size_r) {
+			BUG_ON(start_pfn + n_pfns > start_r + size_r);
+			xen_extra_mem[i].n_pfns = start_pfn - start_r;
 			/* Calling memblock_reserve() again is okay. */
-			xen_add_extra_mem(start + size, start_r + size_r -
-					  (start + size));
+			xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
+					  (start_pfn + n_pfns));
 			break;
 		}
 	}
-	memblock_free(start, size);
+	memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
 }
 
 /*
@@ -156,11 +159,10 @@ static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
 {
 	int i;
-	phys_addr_t addr = PFN_PHYS(pfn);
 
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
-		if (addr >= xen_extra_mem[i].start &&
-		    addr < xen_extra_mem[i].start + xen_extra_mem[i].size)
+		if (pfn >= xen_extra_mem[i].start_pfn &&
+		    pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
 			return INVALID_P2M_ENTRY;
 	}
 
@@ -176,10 +178,10 @@ void __init xen_inv_extra_mem(void)
 	int i;
 
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
-		if (!xen_extra_mem[i].size)
+		if (!xen_extra_mem[i].n_pfns)
 			continue;
-		pfn_s = PFN_DOWN(xen_extra_mem[i].start);
-		pfn_e = PFN_UP(xen_extra_mem[i].start + xen_extra_mem[i].size);
+		pfn_s = xen_extra_mem[i].start_pfn;
+		pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
 		for (pfn = pfn_s; pfn < pfn_e; pfn++)
 			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 	}
@@ -507,7 +509,7 @@ void __init xen_remap_memory(void)
 		} else if (pfn_s + len == xen_remap_buf.target_pfn) {
 			len += xen_remap_buf.size;
 		} else {
-			xen_del_extra_mem(PFN_PHYS(pfn_s), PFN_PHYS(len));
+			xen_del_extra_mem(pfn_s, len);
 			pfn_s = xen_remap_buf.target_pfn;
 			len = xen_remap_buf.size;
 		}
@@ -517,7 +519,7 @@ void __init xen_remap_memory(void)
 	}
 
 	if (pfn_s != ~0UL && len)
-		xen_del_extra_mem(PFN_PHYS(pfn_s), PFN_PHYS(len));
+		xen_del_extra_mem(pfn_s, len);
 
 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
 
@@ -744,7 +746,7 @@ static void __init xen_reserve_xen_mfnlist(void)
  **/
 char * __init xen_memory_setup(void)
 {
-	unsigned long max_pfn;
+	unsigned long max_pfn, pfn_s, n_pfns;
 	phys_addr_t mem_end, addr, size, chunk_size;
 	u32 type;
 	int rc;
@@ -831,9 +833,11 @@ char * __init xen_memory_setup(void)
 				chunk_size = min(size, mem_end - addr);
 			} else if (extra_pages) {
 				chunk_size = min(size, PFN_PHYS(extra_pages));
-				extra_pages -= PFN_DOWN(chunk_size);
-				xen_add_extra_mem(addr, chunk_size);
-				xen_max_p2m_pfn = PFN_DOWN(addr + chunk_size);
+				pfn_s = PFN_UP(addr);
+				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
+				extra_pages -= n_pfns;
+				xen_add_extra_mem(pfn_s, n_pfns);
+				xen_max_p2m_pfn = pfn_s + n_pfns;
 			} else
 				type = E820_UNUSABLE;
 		}
drivers/xen/balloon.c +3 -3
index bf4a23c..1fa633b 100644
@@ -638,9 +638,9 @@ static int __init balloon_init(void)
 	 * regions (see arch/x86/xen/setup.c).
 	 */
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++)
-		if (xen_extra_mem[i].size)
-			balloon_add_region(PFN_UP(xen_extra_mem[i].start),
-					   PFN_DOWN(xen_extra_mem[i].size));
+		if (xen_extra_mem[i].n_pfns)
+			balloon_add_region(xen_extra_mem[i].start_pfn,
+					   xen_extra_mem[i].n_pfns);
 
 	return 0;
 }
include/xen/page.h +2 -2
index c5ed20b..a5983da 100644
@@ -9,8 +9,8 @@ static inline unsigned long page_to_mfn(struct page *page)
 }
 
 struct xen_memory_region {
-	phys_addr_t start;
-	phys_addr_t size;
+	unsigned long start_pfn;
+	unsigned long n_pfns;
 };
 
 #define XEN_EXTRA_MEM_MAX_REGIONS 128 /* == E820MAX */
#17 lore raw

Hello,

El 04/09/15 a les 7.07, Juergen Gross ha escrit:

Quoted 12 lines

Could you try the attached patch? It should do the job. It is booting fine on my laptop, but I think you should try it on the machine with the memory ranges not at page boundary.

Juergen

extramem.patch

commit 3d0f8aa4d1b4c9c16a81902a197b5d6a77e182a0

Quoted 8 lines
Author: Juergen Gross <[email protected]>
Date:   Thu Sep 3 17:44:04 2015 +0200

xen: switch extra memory accounting to use pfns

Instead of using physical addresses for accounting of extra memory areas available for ballooning switch to pfns as this is much less error prone regarding partial pages.

Thanks for taking care of this! I've tested it on the box that has non-page aligned memory ranges and it works fine, only a couple of comments below.

Quoted 8 lines
arch/x86/xen/setup.c +0 -0
index 7a5d566..aa58bc4 100644

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 94

Quoted 6 lines
arch/x86/xen/setup.c +2 -1
@@ -90,62 +90,65 @@ static void __init xen_parse_512gb(void)
 	xen_512gb_limit = val;
 }
 
-static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
+static void __init xen_add_extra_mem(unsigned long start_pfn,
+				     unsigned long n_pfns)

Not very important, but for type consistency this should probably be xen_pfn_t instead of unsigned long here and below.

Quoted 11 lines
arch/x86/xen/setup.c +3 -3
 {
 	int i;
 
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 		/* Add new region. */
-		if (xen_extra_mem[i].size == 0) {
-			xen_extra_mem[i].start = start;
-			xen_extra_mem[i].size  = size;
+		if (xen_extra_mem[i].n_pfns == 0) {
+			xen_extra_mem[i].start_pfn = start_pfn;
+			xen_extra_mem[i].n_pfns = n_pfns;

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 110

Quoted 10 lines
arch/x86/xen/setup.c +3 -2
 			break;
 		}
 		/* Append to existing region. */
-		if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
-			xen_extra_mem[i].size += size;
+		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
+		    start_pfn) {
+			xen_extra_mem[i].n_pfns += n_pfns;
 			break;
 		}

I also noticed this with the original code, why isn't there a case to prepend to an existing region:

if (start_pfn + n_pfns == xen_extra_mem[i].start_pfn) {
    xen_extra_mem[i].n_pfns += n_pfns;
    xen_extra_mem[i].start_pfn = start_pfn;
}
Quoted 7 lines
arch/x86/xen/setup.c +1 -1
 	}
 	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
 		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
 
-	memblock_reserve(start, size);
+	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
 }

[...]

arch/x86/xen/setup.c char * __init xen_memory_setup(void) · line 837

Quoted 9 lines
arch/x86/xen/setup.c +2 -3
@@ -831,9 +833,11 @@ char * __init xen_memory_setup(void)
 				chunk_size = min(size, mem_end - addr);
 			} else if (extra_pages) {
 				chunk_size = min(size, PFN_PHYS(extra_pages));
-				extra_pages -= PFN_DOWN(chunk_size);
-				xen_add_extra_mem(addr, chunk_size);
-				xen_max_p2m_pfn = PFN_DOWN(addr + chunk_size);
+				pfn_s = PFN_UP(addr);
+				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;

Should xen_add_extra_mem check for empty ranges and bail out early, or should the caller make sure it doesn't try to add empty ranges?

IMHO it's easier and cleaner to add the check to xen_add_extra_mem.

Roger.

maintainer
#18 lore raw
Quoted 39 lines · Roger Pau Monné

Hello,

El 04/09/15 a les 7.07, Juergen Gross ha escrit:

Quoted 12 lines

Could you try the attached patch? It should do the job. It is booting fine on my laptop, but I think you should try it on the machine with the memory ranges not at page boundary.

Juergen

extramem.patch

commit 3d0f8aa4d1b4c9c16a81902a197b5d6a77e182a0

Quoted 8 lines
Author: Juergen Gross <[email protected]>
Date:   Thu Sep 3 17:44:04 2015 +0200

xen: switch extra memory accounting to use pfns

Instead of using physical addresses for accounting of extra memory areas available for ballooning switch to pfns as this is much less error prone regarding partial pages.

Thanks for taking care of this! I've tested it on the box that has non-page aligned memory ranges and it works fine, only a couple of comments below.

Quoted 11 lines
arch/x86/xen/setup.c +0 -0
index 7a5d566..aa58bc4 100644
@@ -90,62 +90,65 @@ static void __init xen_parse_512gb(void)
  	xen_512gb_limit = val;
  }
 
Quoted 6 lines

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 94

Quoted 3 lines
arch/x86/xen/setup.c +2 -1
-static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
+static void __init xen_add_extra_mem(unsigned long start_pfn,
+				     unsigned long n_pfns)

Not very important, but for type consistency this should probably be xen_pfn_t instead of unsigned long here and below.

All of the p2m code is using unsigned long for pfns. I wouldn't mind changing this to use xen_pfn_t instead, but this should be done in a separate patch. I'll put it on my list.

Quoted 26 lines
Quoted 12 lines
arch/x86/xen/setup.c +3 -3
  {
  	int i;
 
  	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
  		/* Add new region. */
-		if (xen_extra_mem[i].size == 0) {
-			xen_extra_mem[i].start = start;
-			xen_extra_mem[i].size  = size;
+		if (xen_extra_mem[i].n_pfns == 0) {
+			xen_extra_mem[i].start_pfn = start_pfn;
+			xen_extra_mem[i].n_pfns = n_pfns;

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 110

Quoted 10 lines
arch/x86/xen/setup.c +3 -2
  			break;
  		}
  		/* Append to existing region. */
-		if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
-			xen_extra_mem[i].size += size;
+		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
+		    start_pfn) {
+			xen_extra_mem[i].n_pfns += n_pfns;
  			break;
  		}

I also noticed this with the original code, why isn't there a case to prepend to an existing region:

Quoted 4 lines
if (start_pfn + n_pfns == xen_extra_mem[i].start_pfn) {
     xen_extra_mem[i].n_pfns += n_pfns;
     xen_extra_mem[i].start_pfn = start_pfn;
}

Processing of memory is done from low to high addresses. This case should never happen. And even if it does, the only downside from not handling this scenario is wasting an additional table entry.

Quoted 24 lines
Quoted 8 lines
arch/x86/xen/setup.c +1 -1
  	}
  	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
  		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
 
-	memblock_reserve(start, size);
+	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
  }

[...]

arch/x86/xen/setup.c char * __init xen_memory_setup(void) · line 837

Quoted 9 lines
arch/x86/xen/setup.c +2 -3
@@ -831,9 +833,11 @@ char * __init xen_memory_setup(void)
  				chunk_size = min(size, mem_end - addr);
  			} else if (extra_pages) {
  				chunk_size = min(size, PFN_PHYS(extra_pages));
-				extra_pages -= PFN_DOWN(chunk_size);
-				xen_add_extra_mem(addr, chunk_size);
-				xen_max_p2m_pfn = PFN_DOWN(addr + chunk_size);
+				pfn_s = PFN_UP(addr);
+				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;

Should xen_add_extra_mem check for empty ranges and bail out early, or should the caller make sure it doesn't try to add empty ranges?

Quoted 1 line

IMHO it's easier and cleaner to add the check to xen_add_extra_mem.

This isn't critical at all. Adding an empty range is a nop, as a table entry is regarded to be not used when n_pfns is 0.

Juergen

#19 lore raw

El 04/09/15 a les 9.47, Juergen Gross ha escrit:

Quoted 78 lines · Roger Pau Monné
Quoted 39 lines

Hello,

El 04/09/15 a les 7.07, Juergen Gross ha escrit:

Quoted 12 lines

Could you try the attached patch? It should do the job. It is booting fine on my laptop, but I think you should try it on the machine with the memory ranges not at page boundary.

Juergen

extramem.patch

commit 3d0f8aa4d1b4c9c16a81902a197b5d6a77e182a0

Quoted 8 lines
Author: Juergen Gross <[email protected]>
Date:   Thu Sep 3 17:44:04 2015 +0200

xen: switch extra memory accounting to use pfns

Instead of using physical addresses for accounting of extra memory areas available for ballooning switch to pfns as this is much less error prone regarding partial pages.

Thanks for taking care of this! I've tested it on the box that has non-page aligned memory ranges and it works fine, only a couple of comments below.

Quoted 11 lines
arch/x86/xen/setup.c +0 -0
index 7a5d566..aa58bc4 100644
@@ -90,62 +90,65 @@ static void __init xen_parse_512gb(void)
      xen_512gb_limit = val;
  }
 
Quoted 7 lines

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 94

Quoted 4 lines
arch/x86/xen/setup.c +2 -1
-static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t
 size)
+static void __init xen_add_extra_mem(unsigned long start_pfn,
+                     unsigned long n_pfns)

Not very important, but for type consistency this should probably be xen_pfn_t instead of unsigned long here and below.

All of the p2m code is using unsigned long for pfns. I wouldn't mind changing this to use xen_pfn_t instead, but this should be done in a separate patch. I'll put it on my list.

Quoted 26 lines
Quoted 12 lines
arch/x86/xen/setup.c +3 -3
  {
      int i;
 
      for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
          /* Add new region. */
-        if (xen_extra_mem[i].size == 0) {
-            xen_extra_mem[i].start = start;
-            xen_extra_mem[i].size  = size;
+        if (xen_extra_mem[i].n_pfns == 0) {
+            xen_extra_mem[i].start_pfn = start_pfn;
+            xen_extra_mem[i].n_pfns = n_pfns;

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 110

Quoted 10 lines
arch/x86/xen/setup.c +3 -2
              break;
          }
          /* Append to existing region. */
-        if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
-            xen_extra_mem[i].size += size;
+        if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
+            start_pfn) {
+            xen_extra_mem[i].n_pfns += n_pfns;
              break;
          }

I also noticed this with the original code, why isn't there a case to prepend to an existing region:

Quoted 8 lines
Quoted 4 lines
if (start_pfn + n_pfns == xen_extra_mem[i].start_pfn) {
     xen_extra_mem[i].n_pfns += n_pfns;
     xen_extra_mem[i].start_pfn = start_pfn;
}

Processing of memory is done from low to high addresses. This case should never happen. And even if it does, the only downside from not handling this scenario is wasting an additional table entry.

Right, this case would only be useful for xen_del_extra_mem, and as you say, the worst that can happen is that we end up using an extra slot.

Quoted 25 lines
Quoted 25 lines
Quoted 9 lines
arch/x86/xen/setup.c +1 -1
      }
      if (i == XEN_EXTRA_MEM_MAX_REGIONS)
          printk(KERN_WARNING "Warning: not enough extra memory
 regions\n");
 
-    memblock_reserve(start, size);
+    memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
  }

[...]

arch/x86/xen/setup.c char * __init xen_memory_setup(void) · line 837

Quoted 9 lines
arch/x86/xen/setup.c +2 -3
@@ -831,9 +833,11 @@ char * __init xen_memory_setup(void)
                  chunk_size = min(size, mem_end - addr);
              } else if (extra_pages) {
                  chunk_size = min(size, PFN_PHYS(extra_pages));
-                extra_pages -= PFN_DOWN(chunk_size);
-                xen_add_extra_mem(addr, chunk_size);
-                xen_max_p2m_pfn = PFN_DOWN(addr + chunk_size);
+                pfn_s = PFN_UP(addr);
+                n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;

Should xen_add_extra_mem check for empty ranges and bail out early, or should the caller make sure it doesn't try to add empty ranges?

Quoted 4 lines
Quoted 1 line

IMHO it's easier and cleaner to add the check to xen_add_extra_mem.

This isn't critical at all. Adding an empty range is a nop, as a table entry is regarded to be not used when n_pfns is 0.

Yes, it's just so we can bail out earlier instead of iterating over the whole xen_extra_mem for empty ranges. IMHO, I would add a check to xen_add_extra_mem so we don't waste cycles.

Roger.

maintainer
#20 lore raw
Quoted 84 lines · Roger Pau Monné

El 04/09/15 a les 9.47, Juergen Gross ha escrit:

Quoted 83 lines · Roger Pau Monné
Quoted 39 lines

Hello,

El 04/09/15 a les 7.07, Juergen Gross ha escrit:

Quoted 12 lines

Could you try the attached patch? It should do the job. It is booting fine on my laptop, but I think you should try it on the machine with the memory ranges not at page boundary.

Juergen

extramem.patch

commit 3d0f8aa4d1b4c9c16a81902a197b5d6a77e182a0

Quoted 8 lines
Author: Juergen Gross <[email protected]>
Date:   Thu Sep 3 17:44:04 2015 +0200

xen: switch extra memory accounting to use pfns

Instead of using physical addresses for accounting of extra memory areas available for ballooning switch to pfns as this is much less error prone regarding partial pages.

Thanks for taking care of this! I've tested it on the box that has non-page aligned memory ranges and it works fine, only a couple of comments below.

Quoted 11 lines
arch/x86/xen/setup.c +0 -0
index 7a5d566..aa58bc4 100644
@@ -90,62 +90,65 @@ static void __init xen_parse_512gb(void)
       xen_512gb_limit = val;
   }
 
Quoted 7 lines

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 94

Quoted 4 lines
arch/x86/xen/setup.c +2 -1
-static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t
 size)
+static void __init xen_add_extra_mem(unsigned long start_pfn,
+                     unsigned long n_pfns)

Not very important, but for type consistency this should probably be xen_pfn_t instead of unsigned long here and below.

All of the p2m code is using unsigned long for pfns. I wouldn't mind changing this to use xen_pfn_t instead, but this should be done in a separate patch. I'll put it on my list.

Quoted 30 lines
Quoted 12 lines
arch/x86/xen/setup.c +3 -3
   {
       int i;
 
       for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
           /* Add new region. */
-        if (xen_extra_mem[i].size == 0) {
-            xen_extra_mem[i].start = start;
-            xen_extra_mem[i].size  = size;
+        if (xen_extra_mem[i].n_pfns == 0) {
+            xen_extra_mem[i].start_pfn = start_pfn;
+            xen_extra_mem[i].n_pfns = n_pfns;

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 110

Quoted 10 lines
arch/x86/xen/setup.c +3 -2
               break;
           }
           /* Append to existing region. */
-        if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
-            xen_extra_mem[i].size += size;
+        if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
+            start_pfn) {
+            xen_extra_mem[i].n_pfns += n_pfns;
               break;
           }

I also noticed this with the original code, why isn't there a case to prepend to an existing region:

if (start_pfn + n_pfns == xen_extra_mem[i].start_pfn) {
      xen_extra_mem[i].n_pfns += n_pfns;
      xen_extra_mem[i].start_pfn = start_pfn;
}
Quoted 6 lines
Quoted 3 lines

Processing of memory is done from low to high addresses. This case should never happen. And even if it does, the only downside from not handling this scenario is wasting an additional table entry.

Right, this case would only be useful for xen_del_extra_mem, and as you say, the worst that can happen is that we end up using an extra slot.

Even this case should be impossible. It would require two adjacent regions to be present in xen_extra_mem before calling xen_del_extra_mem which is impossible as stated above.

Quoted 26 lines
Quoted 26 lines
Quoted 26 lines
Quoted 10 lines
arch/x86/xen/setup.c +1 -1
       }
       if (i == XEN_EXTRA_MEM_MAX_REGIONS)
           printk(KERN_WARNING "Warning: not enough extra memory
 regions\n");
 
-    memblock_reserve(start, size);
+    memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
   }

[...]

arch/x86/xen/setup.c char * __init xen_memory_setup(void) · line 837

Quoted 9 lines
arch/x86/xen/setup.c +2 -3
@@ -831,9 +833,11 @@ char * __init xen_memory_setup(void)
                   chunk_size = min(size, mem_end - addr);
               } else if (extra_pages) {
                   chunk_size = min(size, PFN_PHYS(extra_pages));
-                extra_pages -= PFN_DOWN(chunk_size);
-                xen_add_extra_mem(addr, chunk_size);
-                xen_max_p2m_pfn = PFN_DOWN(addr + chunk_size);
+                pfn_s = PFN_UP(addr);
+                n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;

Should xen_add_extra_mem check for empty ranges and bail out early, or should the caller make sure it doesn't try to add empty ranges?

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

IMHO it's easier and cleaner to add the check to xen_add_extra_mem.

This isn't critical at all. Adding an empty range is a nop, as a table entry is regarded to be not used when n_pfns is 0.

Yes, it's just so we can bail out earlier instead of iterating over the whole xen_extra_mem for empty ranges. IMHO, I would add a check to xen_add_extra_mem so we don't waste cycles.

The question is whether the additional check for each call wouldn't add more cycles than the early bail out for the rare case would win.

A comment seems to be an appropriate measure. :-)

Juergen

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

maintainer
#21 lore raw
In reply to Roger Pau Monné
Quoted 84 lines · Roger Pau Monné

El 04/09/15 a les 9.47, Juergen Gross ha escrit:

Quoted 83 lines · Roger Pau Monné
Quoted 39 lines

Hello,

El 04/09/15 a les 7.07, Juergen Gross ha escrit:

Quoted 12 lines

Could you try the attached patch? It should do the job. It is booting fine on my laptop, but I think you should try it on the machine with the memory ranges not at page boundary.

Juergen

extramem.patch

commit 3d0f8aa4d1b4c9c16a81902a197b5d6a77e182a0

Quoted 8 lines
Author: Juergen Gross <[email protected]>
Date:   Thu Sep 3 17:44:04 2015 +0200

xen: switch extra memory accounting to use pfns

Instead of using physical addresses for accounting of extra memory areas available for ballooning switch to pfns as this is much less error prone regarding partial pages.

Thanks for taking care of this! I've tested it on the box that has non-page aligned memory ranges and it works fine, only a couple of comments below.

Quoted 11 lines
arch/x86/xen/setup.c +0 -0
index 7a5d566..aa58bc4 100644
@@ -90,62 +90,65 @@ static void __init xen_parse_512gb(void)
       xen_512gb_limit = val;
   }
 
Quoted 7 lines

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 94

Quoted 4 lines
arch/x86/xen/setup.c +2 -1
-static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t
 size)
+static void __init xen_add_extra_mem(unsigned long start_pfn,
+                     unsigned long n_pfns)

Not very important, but for type consistency this should probably be xen_pfn_t instead of unsigned long here and below.

All of the p2m code is using unsigned long for pfns. I wouldn't mind changing this to use xen_pfn_t instead, but this should be done in a separate patch. I'll put it on my list.

Quoted 30 lines
Quoted 12 lines
arch/x86/xen/setup.c +3 -3
   {
       int i;
 
       for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
           /* Add new region. */
-        if (xen_extra_mem[i].size == 0) {
-            xen_extra_mem[i].start = start;
-            xen_extra_mem[i].size  = size;
+        if (xen_extra_mem[i].n_pfns == 0) {
+            xen_extra_mem[i].start_pfn = start_pfn;
+            xen_extra_mem[i].n_pfns = n_pfns;

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 110

Quoted 10 lines
arch/x86/xen/setup.c +3 -2
               break;
           }
           /* Append to existing region. */
-        if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
-            xen_extra_mem[i].size += size;
+        if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
+            start_pfn) {
+            xen_extra_mem[i].n_pfns += n_pfns;
               break;
           }

I also noticed this with the original code, why isn't there a case to prepend to an existing region:

if (start_pfn + n_pfns == xen_extra_mem[i].start_pfn) {
      xen_extra_mem[i].n_pfns += n_pfns;
      xen_extra_mem[i].start_pfn = start_pfn;
}
Quoted 6 lines
Quoted 3 lines

Processing of memory is done from low to high addresses. This case should never happen. And even if it does, the only downside from not handling this scenario is wasting an additional table entry.

Right, this case would only be useful for xen_del_extra_mem, and as you say, the worst that can happen is that we end up using an extra slot.

Even this case should be impossible. It would require two adjacent regions to be present in xen_extra_mem before calling xen_del_extra_mem which is impossible as stated above.

Quoted 26 lines
Quoted 26 lines
Quoted 26 lines
Quoted 10 lines
arch/x86/xen/setup.c +1 -1
       }
       if (i == XEN_EXTRA_MEM_MAX_REGIONS)
           printk(KERN_WARNING "Warning: not enough extra memory
 regions\n");
 
-    memblock_reserve(start, size);
+    memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
   }

[...]

arch/x86/xen/setup.c char * __init xen_memory_setup(void) · line 837

Quoted 9 lines
arch/x86/xen/setup.c +2 -3
@@ -831,9 +833,11 @@ char * __init xen_memory_setup(void)
                   chunk_size = min(size, mem_end - addr);
               } else if (extra_pages) {
                   chunk_size = min(size, PFN_PHYS(extra_pages));
-                extra_pages -= PFN_DOWN(chunk_size);
-                xen_add_extra_mem(addr, chunk_size);
-                xen_max_p2m_pfn = PFN_DOWN(addr + chunk_size);
+                pfn_s = PFN_UP(addr);
+                n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;

Should xen_add_extra_mem check for empty ranges and bail out early, or should the caller make sure it doesn't try to add empty ranges?

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

IMHO it's easier and cleaner to add the check to xen_add_extra_mem.

This isn't critical at all. Adding an empty range is a nop, as a table entry is regarded to be not used when n_pfns is 0.

Yes, it's just so we can bail out earlier instead of iterating over the whole xen_extra_mem for empty ranges. IMHO, I would add a check to xen_add_extra_mem so we don't waste cycles.

The question is whether the additional check for each call wouldn't add more cycles than the early bail out for the rare case would win.

A comment seems to be an appropriate measure. :-)

Juergen

#22 lore raw
In reply to Juergen Gross

El 04/09/15 a les 9.47, Juergen Gross ha escrit:

Quoted 78 lines · Roger Pau Monné
Quoted 39 lines

Hello,

El 04/09/15 a les 7.07, Juergen Gross ha escrit:

Quoted 12 lines

Could you try the attached patch? It should do the job. It is booting fine on my laptop, but I think you should try it on the machine with the memory ranges not at page boundary.

Juergen

extramem.patch

commit 3d0f8aa4d1b4c9c16a81902a197b5d6a77e182a0

Quoted 8 lines
Author: Juergen Gross <[email protected]>
Date:   Thu Sep 3 17:44:04 2015 +0200

xen: switch extra memory accounting to use pfns

Instead of using physical addresses for accounting of extra memory areas available for ballooning switch to pfns as this is much less error prone regarding partial pages.

Thanks for taking care of this! I've tested it on the box that has non-page aligned memory ranges and it works fine, only a couple of comments below.

Quoted 11 lines
arch/x86/xen/setup.c +0 -0
index 7a5d566..aa58bc4 100644
@@ -90,62 +90,65 @@ static void __init xen_parse_512gb(void)
      xen_512gb_limit = val;
  }
 
Quoted 7 lines

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 94

Quoted 4 lines
arch/x86/xen/setup.c +2 -1
-static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t
 size)
+static void __init xen_add_extra_mem(unsigned long start_pfn,
+                     unsigned long n_pfns)

Not very important, but for type consistency this should probably be xen_pfn_t instead of unsigned long here and below.

All of the p2m code is using unsigned long for pfns. I wouldn't mind changing this to use xen_pfn_t instead, but this should be done in a separate patch. I'll put it on my list.

Quoted 26 lines
Quoted 12 lines
arch/x86/xen/setup.c +3 -3
  {
      int i;
 
      for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
          /* Add new region. */
-        if (xen_extra_mem[i].size == 0) {
-            xen_extra_mem[i].start = start;
-            xen_extra_mem[i].size  = size;
+        if (xen_extra_mem[i].n_pfns == 0) {
+            xen_extra_mem[i].start_pfn = start_pfn;
+            xen_extra_mem[i].n_pfns = n_pfns;

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 110

Quoted 10 lines
arch/x86/xen/setup.c +3 -2
              break;
          }
          /* Append to existing region. */
-        if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
-            xen_extra_mem[i].size += size;
+        if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
+            start_pfn) {
+            xen_extra_mem[i].n_pfns += n_pfns;
              break;
          }

I also noticed this with the original code, why isn't there a case to prepend to an existing region:

Quoted 8 lines
Quoted 4 lines
if (start_pfn + n_pfns == xen_extra_mem[i].start_pfn) {
     xen_extra_mem[i].n_pfns += n_pfns;
     xen_extra_mem[i].start_pfn = start_pfn;
}

Processing of memory is done from low to high addresses. This case should never happen. And even if it does, the only downside from not handling this scenario is wasting an additional table entry.

Right, this case would only be useful for xen_del_extra_mem, and as you say, the worst that can happen is that we end up using an extra slot.

Quoted 25 lines
Quoted 25 lines
Quoted 9 lines
arch/x86/xen/setup.c +1 -1
      }
      if (i == XEN_EXTRA_MEM_MAX_REGIONS)
          printk(KERN_WARNING "Warning: not enough extra memory
 regions\n");
 
-    memblock_reserve(start, size);
+    memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
  }

[...]

arch/x86/xen/setup.c char * __init xen_memory_setup(void) · line 837

Quoted 9 lines
arch/x86/xen/setup.c +2 -3
@@ -831,9 +833,11 @@ char * __init xen_memory_setup(void)
                  chunk_size = min(size, mem_end - addr);
              } else if (extra_pages) {
                  chunk_size = min(size, PFN_PHYS(extra_pages));
-                extra_pages -= PFN_DOWN(chunk_size);
-                xen_add_extra_mem(addr, chunk_size);
-                xen_max_p2m_pfn = PFN_DOWN(addr + chunk_size);
+                pfn_s = PFN_UP(addr);
+                n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;

Should xen_add_extra_mem check for empty ranges and bail out early, or should the caller make sure it doesn't try to add empty ranges?

Quoted 4 lines
Quoted 1 line

IMHO it's easier and cleaner to add the check to xen_add_extra_mem.

This isn't critical at all. Adding an empty range is a nop, as a table entry is regarded to be not used when n_pfns is 0.

Yes, it's just so we can bail out earlier instead of iterating over the whole xen_extra_mem for empty ranges. IMHO, I would add a check to xen_add_extra_mem so we don't waste cycles.

Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

maintainer
#23 lore raw
In reply to Roger Pau Monné
Quoted 39 lines · Roger Pau Monné

Hello,

El 04/09/15 a les 7.07, Juergen Gross ha escrit:

Quoted 12 lines

Could you try the attached patch? It should do the job. It is booting fine on my laptop, but I think you should try it on the machine with the memory ranges not at page boundary.

Juergen

extramem.patch

commit 3d0f8aa4d1b4c9c16a81902a197b5d6a77e182a0

Quoted 8 lines
Author: Juergen Gross <[email protected]>
Date:   Thu Sep 3 17:44:04 2015 +0200

xen: switch extra memory accounting to use pfns

Instead of using physical addresses for accounting of extra memory areas available for ballooning switch to pfns as this is much less error prone regarding partial pages.

Thanks for taking care of this! I've tested it on the box that has non-page aligned memory ranges and it works fine, only a couple of comments below.

Quoted 11 lines
arch/x86/xen/setup.c +0 -0
index 7a5d566..aa58bc4 100644
@@ -90,62 +90,65 @@ static void __init xen_parse_512gb(void)
  	xen_512gb_limit = val;
  }
 
Quoted 6 lines

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 94

Quoted 3 lines
arch/x86/xen/setup.c +2 -1
-static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
+static void __init xen_add_extra_mem(unsigned long start_pfn,
+				     unsigned long n_pfns)

Not very important, but for type consistency this should probably be xen_pfn_t instead of unsigned long here and below.

All of the p2m code is using unsigned long for pfns. I wouldn't mind changing this to use xen_pfn_t instead, but this should be done in a separate patch. I'll put it on my list.

Quoted 26 lines
Quoted 12 lines
arch/x86/xen/setup.c +3 -3
  {
  	int i;
 
  	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
  		/* Add new region. */
-		if (xen_extra_mem[i].size == 0) {
-			xen_extra_mem[i].start = start;
-			xen_extra_mem[i].size  = size;
+		if (xen_extra_mem[i].n_pfns == 0) {
+			xen_extra_mem[i].start_pfn = start_pfn;
+			xen_extra_mem[i].n_pfns = n_pfns;

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 110

Quoted 10 lines
arch/x86/xen/setup.c +3 -2
  			break;
  		}
  		/* Append to existing region. */
-		if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
-			xen_extra_mem[i].size += size;
+		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
+		    start_pfn) {
+			xen_extra_mem[i].n_pfns += n_pfns;
  			break;
  		}

I also noticed this with the original code, why isn't there a case to prepend to an existing region:

Quoted 4 lines
if (start_pfn + n_pfns == xen_extra_mem[i].start_pfn) {
     xen_extra_mem[i].n_pfns += n_pfns;
     xen_extra_mem[i].start_pfn = start_pfn;
}

Processing of memory is done from low to high addresses. This case should never happen. And even if it does, the only downside from not handling this scenario is wasting an additional table entry.

Quoted 24 lines
Quoted 8 lines
arch/x86/xen/setup.c +1 -1
  	}
  	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
  		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
 
-	memblock_reserve(start, size);
+	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
  }

[...]

arch/x86/xen/setup.c char * __init xen_memory_setup(void) · line 837

Quoted 9 lines
arch/x86/xen/setup.c +2 -3
@@ -831,9 +833,11 @@ char * __init xen_memory_setup(void)
  				chunk_size = min(size, mem_end - addr);
  			} else if (extra_pages) {
  				chunk_size = min(size, PFN_PHYS(extra_pages));
-				extra_pages -= PFN_DOWN(chunk_size);
-				xen_add_extra_mem(addr, chunk_size);
-				xen_max_p2m_pfn = PFN_DOWN(addr + chunk_size);
+				pfn_s = PFN_UP(addr);
+				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;

Should xen_add_extra_mem check for empty ranges and bail out early, or should the caller make sure it doesn't try to add empty ranges?

Quoted 1 line

IMHO it's easier and cleaner to add the check to xen_add_extra_mem.

This isn't critical at all. Adding an empty range is a nop, as a table entry is regarded to be not used when n_pfns is 0.

Juergen

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

#24 lore raw
In reply to Juergen Gross

Hello,

El 04/09/15 a les 7.07, Juergen Gross ha escrit:

Quoted 12 lines

Could you try the attached patch? It should do the job. It is booting fine on my laptop, but I think you should try it on the machine with the memory ranges not at page boundary.

Juergen

extramem.patch

commit 3d0f8aa4d1b4c9c16a81902a197b5d6a77e182a0

Quoted 8 lines
Author: Juergen Gross <[email protected]>
Date:   Thu Sep 3 17:44:04 2015 +0200

xen: switch extra memory accounting to use pfns

Instead of using physical addresses for accounting of extra memory areas available for ballooning switch to pfns as this is much less error prone regarding partial pages.

Thanks for taking care of this! I've tested it on the box that has non-page aligned memory ranges and it works fine, only a couple of comments below.

Quoted 8 lines
arch/x86/xen/setup.c +0 -0
index 7a5d566..aa58bc4 100644

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 94

Quoted 6 lines
arch/x86/xen/setup.c +2 -1
@@ -90,62 +90,65 @@ static void __init xen_parse_512gb(void)
 	xen_512gb_limit = val;
 }
 
-static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
+static void __init xen_add_extra_mem(unsigned long start_pfn,
+				     unsigned long n_pfns)

Not very important, but for type consistency this should probably be xen_pfn_t instead of unsigned long here and below.

Quoted 11 lines
arch/x86/xen/setup.c +3 -3
 {
 	int i;
 
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 		/* Add new region. */
-		if (xen_extra_mem[i].size == 0) {
-			xen_extra_mem[i].start = start;
-			xen_extra_mem[i].size  = size;
+		if (xen_extra_mem[i].n_pfns == 0) {
+			xen_extra_mem[i].start_pfn = start_pfn;
+			xen_extra_mem[i].n_pfns = n_pfns;

arch/x86/xen/setup.c static void __init xen_parse_512gb(void) · line 110

Quoted 10 lines
arch/x86/xen/setup.c +3 -2
 			break;
 		}
 		/* Append to existing region. */
-		if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
-			xen_extra_mem[i].size += size;
+		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
+		    start_pfn) {
+			xen_extra_mem[i].n_pfns += n_pfns;
 			break;
 		}

I also noticed this with the original code, why isn't there a case to prepend to an existing region:

if (start_pfn + n_pfns == xen_extra_mem[i].start_pfn) {
    xen_extra_mem[i].n_pfns += n_pfns;
    xen_extra_mem[i].start_pfn = start_pfn;
}
Quoted 7 lines
arch/x86/xen/setup.c +1 -1
 	}
 	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
 		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
 
-	memblock_reserve(start, size);
+	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
 }

[...]

arch/x86/xen/setup.c char * __init xen_memory_setup(void) · line 837

Quoted 9 lines
arch/x86/xen/setup.c +2 -3
@@ -831,9 +833,11 @@ char * __init xen_memory_setup(void)
 				chunk_size = min(size, mem_end - addr);
 			} else if (extra_pages) {
 				chunk_size = min(size, PFN_PHYS(extra_pages));
-				extra_pages -= PFN_DOWN(chunk_size);
-				xen_add_extra_mem(addr, chunk_size);
-				xen_max_p2m_pfn = PFN_DOWN(addr + chunk_size);
+				pfn_s = PFN_UP(addr);
+				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;

Should xen_add_extra_mem check for empty ranges and bail out early, or should the caller make sure it doesn't try to add empty ranges?

IMHO it's easier and cleaner to add the check to xen_add_extra_mem.

Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

maintainer
#25 lore raw
In reply to Roger Pau Monné

Hi Roger,

Quoted 91 lines · Roger Pau Monné

El 03/09/15 a les 17.20, Juergen Gross ha escrit:

Quoted 90 lines · David Vrabel
Quoted 88 lines · Juergen Gross
Quoted 77 lines · David Vrabel
Quoted 76 lines · David Vrabel
Quoted 65 lines · Roger Pau Monné
Quoted 64 lines

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 46 lines · Roger Pau Monne
Quoted 45 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00
will be
refused because the check is performed against 0xdfdf9000 instead of
0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the
xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
     arch/x86/xen/setup.c | 6 ++++++
     1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
     {
         int i;
 
Quoted 6 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll
add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 7 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should
add extra memory regions that cover these same areas.

Ignore this. This was nonsense.

Quoted 8 lines
Quoted 5 lines

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

Nearly. Roger's patch rounds up start and rounds down the size. It might add non-RAM partial pages to xen_extra_mem.

Yes. You're right.

Quoted 8 lines
Quoted 5 lines

Hmm, thinking more about it, I'd prefer to change xen_extra_mem to use pfns instead of physical addresses. This would make things much more clear.

Roger, do you want to do the patch or should I?

I can certainly take care of it if you are busy, otherwise I leave it to you since you have more expertise on it :).

Could you try the attached patch? It should do the job. It is booting fine on my laptop, but I think you should try it on the machine with the memory ranges not at page boundary.

Juergen

commit 3d0f8aa4d1b4c9c16a81902a197b5d6a77e182a0

Author: Juergen Gross <[email protected]>
Date:   Thu Sep 3 17:44:04 2015 +0200

xen: switch extra memory accounting to use pfns

Instead of using physical addresses for accounting of extra memory areas available for ballooning switch to pfns as this is much less error prone regarding partial pages.

arch/x86/xen/setup.c +40 -36
index 7a5d566..aa58bc4 100644
@@ -90,62 +90,65 @@ static void __init xen_parse_512gb(void)
 	xen_512gb_limit = val;
 }
 
-static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
+static void __init xen_add_extra_mem(unsigned long start_pfn,
+				     unsigned long n_pfns)
 {
 	int i;
 
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 		/* Add new region. */
-		if (xen_extra_mem[i].size == 0) {
-			xen_extra_mem[i].start = start;
-			xen_extra_mem[i].size  = size;
+		if (xen_extra_mem[i].n_pfns == 0) {
+			xen_extra_mem[i].start_pfn = start_pfn;
+			xen_extra_mem[i].n_pfns = n_pfns;
 			break;
 		}
 		/* Append to existing region. */
-		if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
-			xen_extra_mem[i].size += size;
+		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
+		    start_pfn) {
+			xen_extra_mem[i].n_pfns += n_pfns;
 			break;
 		}
 	}
 	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
 		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
 
-	memblock_reserve(start, size);
+	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
 }
 
-static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
+static void __init xen_del_extra_mem(unsigned long start_pfn,
+				     unsigned long n_pfns)
 {
 	int i;
-	phys_addr_t start_r, size_r;
+	unsigned long start_r, size_r;
 
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
-		start_r = xen_extra_mem[i].start;
-		size_r = xen_extra_mem[i].size;
+		start_r = xen_extra_mem[i].start_pfn;
+		size_r = xen_extra_mem[i].n_pfns;
 
 		/* Start of region. */
-		if (start_r == start) {
-			BUG_ON(size > size_r);
-			xen_extra_mem[i].start += size;
-			xen_extra_mem[i].size -= size;
+		if (start_r == start_pfn) {
+			BUG_ON(n_pfns > size_r);
+			xen_extra_mem[i].start_pfn += n_pfns;
+			xen_extra_mem[i].n_pfns -= n_pfns;
 			break;
 		}
 		/* End of region. */
-		if (start_r + size_r == start + size) {
-			BUG_ON(size > size_r);
-			xen_extra_mem[i].size -= size;
+		if (start_r + size_r == start_pfn + n_pfns) {
+			BUG_ON(n_pfns > size_r);
+			xen_extra_mem[i].n_pfns -= n_pfns;
 			break;
 		}
 		/* Mid of region. */
-		if (start > start_r && start < start_r + size_r) {
-			BUG_ON(start + size > start_r + size_r);
-			xen_extra_mem[i].size = start - start_r;
+		if (start_pfn > start_r && start_pfn < start_r + size_r) {
+			BUG_ON(start_pfn + n_pfns > start_r + size_r);
+			xen_extra_mem[i].n_pfns = start_pfn - start_r;
 			/* Calling memblock_reserve() again is okay. */
-			xen_add_extra_mem(start + size, start_r + size_r -
-					  (start + size));
+			xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
+					  (start_pfn + n_pfns));
 			break;
 		}
 	}
-	memblock_free(start, size);
+	memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
 }
 
 /*
@@ -156,11 +159,10 @@ static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
 {
 	int i;
-	phys_addr_t addr = PFN_PHYS(pfn);
 
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
-		if (addr >= xen_extra_mem[i].start &&
-		    addr < xen_extra_mem[i].start + xen_extra_mem[i].size)
+		if (pfn >= xen_extra_mem[i].start_pfn &&
+		    pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
 			return INVALID_P2M_ENTRY;
 	}
 
@@ -176,10 +178,10 @@ void __init xen_inv_extra_mem(void)
 	int i;
 
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
-		if (!xen_extra_mem[i].size)
+		if (!xen_extra_mem[i].n_pfns)
 			continue;
-		pfn_s = PFN_DOWN(xen_extra_mem[i].start);
-		pfn_e = PFN_UP(xen_extra_mem[i].start + xen_extra_mem[i].size);
+		pfn_s = xen_extra_mem[i].start_pfn;
+		pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
 		for (pfn = pfn_s; pfn < pfn_e; pfn++)
 			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 	}
@@ -507,7 +509,7 @@ void __init xen_remap_memory(void)
 		} else if (pfn_s + len == xen_remap_buf.target_pfn) {
 			len += xen_remap_buf.size;
 		} else {
-			xen_del_extra_mem(PFN_PHYS(pfn_s), PFN_PHYS(len));
+			xen_del_extra_mem(pfn_s, len);
 			pfn_s = xen_remap_buf.target_pfn;
 			len = xen_remap_buf.size;
 		}
@@ -517,7 +519,7 @@ void __init xen_remap_memory(void)
 	}
 
 	if (pfn_s != ~0UL && len)
-		xen_del_extra_mem(PFN_PHYS(pfn_s), PFN_PHYS(len));
+		xen_del_extra_mem(pfn_s, len);
 
 	set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
 
@@ -744,7 +746,7 @@ static void __init xen_reserve_xen_mfnlist(void)
  **/
 char * __init xen_memory_setup(void)
 {
-	unsigned long max_pfn;
+	unsigned long max_pfn, pfn_s, n_pfns;
 	phys_addr_t mem_end, addr, size, chunk_size;
 	u32 type;
 	int rc;
@@ -831,9 +833,11 @@ char * __init xen_memory_setup(void)
 				chunk_size = min(size, mem_end - addr);
 			} else if (extra_pages) {
 				chunk_size = min(size, PFN_PHYS(extra_pages));
-				extra_pages -= PFN_DOWN(chunk_size);
-				xen_add_extra_mem(addr, chunk_size);
-				xen_max_p2m_pfn = PFN_DOWN(addr + chunk_size);
+				pfn_s = PFN_UP(addr);
+				n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
+				extra_pages -= n_pfns;
+				xen_add_extra_mem(pfn_s, n_pfns);
+				xen_max_p2m_pfn = pfn_s + n_pfns;
 			} else
 				type = E820_UNUSABLE;
 		}
drivers/xen/balloon.c +3 -3
index bf4a23c..1fa633b 100644
@@ -638,9 +638,9 @@ static int __init balloon_init(void)
 	 * regions (see arch/x86/xen/setup.c).
 	 */
 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++)
-		if (xen_extra_mem[i].size)
-			balloon_add_region(PFN_UP(xen_extra_mem[i].start),
-					   PFN_DOWN(xen_extra_mem[i].size));
+		if (xen_extra_mem[i].n_pfns)
+			balloon_add_region(xen_extra_mem[i].start_pfn,
+					   xen_extra_mem[i].n_pfns);
 
 	return 0;
 }
include/xen/page.h +2 -2
index c5ed20b..a5983da 100644
@@ -9,8 +9,8 @@ static inline unsigned long page_to_mfn(struct page *page)
 }
 
 struct xen_memory_region {
-	phys_addr_t start;
-	phys_addr_t size;
+	unsigned long start_pfn;
+	unsigned long n_pfns;
 };
 
 #define XEN_EXTRA_MEM_MAX_REGIONS 128 /* == E820MAX */

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

#26 lore raw
In reply to Juergen Gross

El 03/09/15 a les 17.20, Juergen Gross ha escrit:

Quoted 88 lines · David Vrabel
Quoted 87 lines · Juergen Gross
Quoted 85 lines · David Vrabel
Quoted 74 lines · David Vrabel
Quoted 65 lines · Roger Pau Monné
Quoted 64 lines

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 46 lines · Roger Pau Monne
Quoted 45 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00
will be
refused because the check is performed against 0xdfdf9000 instead of
0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the
xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
    arch/x86/xen/setup.c | 6 ++++++
    1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
    {
        int i;
 
Quoted 6 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll
add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 7 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should
add extra memory regions that cover these same areas.

Quoted 7 lines

Ignore this. This was nonsense.

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

Nearly. Roger's patch rounds up start and rounds down the size. It might add non-RAM partial pages to xen_extra_mem.

Quoted 7 lines
Quoted 1 line

Yes. You're right.

Hmm, thinking more about it, I'd prefer to change xen_extra_mem to use pfns instead of physical addresses. This would make things much more clear.

Roger, do you want to do the patch or should I?

I can certainly take care of it if you are busy, otherwise I leave it to you since you have more expertise on it :).

Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

#27 lore raw
In reply to Juergen Gross
Quoted 72 lines · Juergen Gross
Quoted 71 lines · David Vrabel
Quoted 61 lines · David Vrabel
Quoted 60 lines · Roger Pau Monné

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
   arch/x86/xen/setup.c | 6 ++++++
   1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
   {
       int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 6 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add extra memory regions that cover these same areas.

Ignore this. This was nonsense.

Quoted 8 lines
Quoted 5 lines

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

Nearly. Roger's patch rounds up start and rounds down the size. It might add non-RAM partial pages to xen_extra_mem.

Yes. You're right.

David

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

#28 lore raw
In reply to David Vrabel
Quoted 61 lines · David Vrabel
Quoted 60 lines · Roger Pau Monné

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
  arch/x86/xen/setup.c | 6 ++++++
  1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
  {
      int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

Quoted 6 lines
Quoted 2 lines

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add extra memory regions that cover these same areas.

Ignore this. This was nonsense.

We expand the reserved (i.e., non-RAM) areas so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add the extra memory such that it does not overlap with with expanded regions. i.e., round up the start and round down the end (like Roger's patch does).

David

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

#29 lore raw
In reply to Roger Pau Monné
Quoted 55 lines · Roger Pau Monné

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
  arch/x86/xen/setup.c | 6 ++++++
  1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
  {
      int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

Quoted 7 lines

After the fixup added above this would become:

start = 0x9000
size = 0x1000

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

We expand the reserved (i.e., non-RAM) areas down so they're fully covered with whole pages when we depopulate and 1:1 map them, we should add extra memory regions that cover these same areas.

See the use of PFN_DOWN() and PFN_UP() in xen_set_identify_and_remap().

David

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

maintainer
#30 lore raw
In reply to Roger Pau Monné
Quoted 55 lines · Roger Pau Monné

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
   arch/x86/xen/setup.c | 6 ++++++
   1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
   {
       int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

Quoted 7 lines

After the fixup added above this would become:

start = 0x9000
size = 0x1000

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

You'd add 9c00-9cff which shouldn't be added (in this example you shouldn't add anything as no complete page is covered by the range).

You need something like:

end = (start + size) & PAGE_MASK;
start = PAGE_ALIGN(start);
size = end - start;
if (!size)

     return;

Juergen

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

maintainer
#31 lore raw
In reply to Roger Pau Monné
Quoted 55 lines · Roger Pau Monné

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
   arch/x86/xen/setup.c | 6 ++++++
   1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
   {
       int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

Quoted 7 lines

After the fixup added above this would become:

start = 0x9000
size = 0x1000

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

You'd add 9c00-9cff which shouldn't be added (in this example you shouldn't add anything as no complete page is covered by the range).

You need something like:

end = (start + size) & PAGE_MASK;
start = PAGE_ALIGN(start);
size = end - start;
if (!size)

     return;

Juergen

#32 lore raw
In reply to Juergen Gross

El 03/09/15 a les 14.25, Juergen Gross ha escrit:

Quoted 43 lines · Roger Pau Monne
Quoted 42 lines

On systems with memory maps with ranges that don't end at page boundaries,
like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00,
but the function used to check if a memory address is inside of a protected
range works with pfns, which means that an attempt to map 0xdfdf9c00 will be
refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
  arch/x86/xen/setup.c | 6 ++++++
  1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t
 start, phys_addr_t size)
  {
      int i;
 
Quoted 5 lines

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t · line 73

Quoted 2 lines
arch/x86/xen/setup.c +2 -0
+    start = PAGE_ALIGN(start);
+    size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

I'm not understanding this, let's put an example:

start = 0x8c00
size = 0x1000

After the fixup added above this would become:

start = 0x9000
size = 0x1000

So if anything, I'm adding one page less (because 0x8000 was partly added, and with the fixup it is not added).

Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

maintainer
#33 lore raw
In reply to Roger Pau Monne
Quoted 34 lines · Roger Pau Monne

On systems with memory maps with ranges that don't end at page boundaries, like:

[...]
(XEN) 0000000000100000 - 00000000dfdf9c00 (usable) (XEN) 00000000dfdf9c00 - 00000000dfe4bc00 (ACPI NVS) [...]

xen_add_extra_mem will create a protected range that ends up at 0xdfdf9c00, but the function used to check if a memory address is inside of a protected range works with pfns, which means that an attempt to map 0xdfdf9c00 will be refused because the check is performed against 0xdfdf9000 instead of 0xdfdf9c00.

In order to fix this, make sure that the ranges that are added to the xen_extra_mem array are aligned to page boundaries.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc David Vrabel <[email protected]>
Cc Juergen Gross <[email protected]>

---
AFAICT this patch needs to be backported to 3.19, 4.0, 4.1 and 4.2.

---
  arch/x86/xen/setup.c | 6 ++++++
  1 file changed, 6 insertions(+)
arch/x86/xen/setup.c +0 -0
index 55f388e..dcf5865 100644

arch/x86/xen/setup.c static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size) · line 72

Quoted 5 lines
arch/x86/xen/setup.c +2 -0
@@ -68,6 +68,9 @@ static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
  {
  	int i;
 
+	start = PAGE_ALIGN(start);
+	size &= PAGE_MASK;

This is not correct. If start wasn't page aligned and size was, you'll add one additional page to xen_extra_mem.

arch/x86/xen/setup.c static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size) · line 99

Quoted 10 lines
arch/x86/xen/setup.c +3 -0
+
  	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
  		/* Add new region. */
  		if (xen_extra_mem[i].size == 0) {
@@ -92,6 +95,9 @@ static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
  	int i;
  	phys_addr_t start_r, size_r;
 
+	start = PAGE_ALIGN(start);
+	size &= PAGE_MASK;

Not really needed, as xen_del_extra_mem() is called with aligned values only. OTOH it does no harm, as long as you correct size as above.

arch/x86/xen/setup.c static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size) · line 103

Quoted 5 lines
arch/x86/xen/setup.c +1 -0
+
  	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
  		start_r = xen_extra_mem[i].start;
  		size_r = xen_extra_mem[i].size;

Juergen

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel