Skip to contentSkip to conversation

ARM: xip: Use correct symbol for end of ROM marker

PATCH Superseded Chris Brandt 8 participants 53 messages Nov 11, 2015 – Feb 2, 2016

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ARM: xip: Use correct symbol for end of ROM marker

For an XIP build, _edata_loc, not _etext, represents the end of the binary image that will be programmed into ROM and mapped into the MODULES_VADDR area.
With an XIP kernel, nothing is loaded into RAM before boot, meaning you have to take into account the size of the entire binary image that was programmed, including the init data values that will be copied to RAM during kernel boot.

This fixes the bug where you might lose the end of your kernel area after page table setup is complete.

---
 arch/arm/mm/mmu.c              |    4 ++--
 include/asm-generic/sections.h |    2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
arch/arm/mm/mmu.c +2 -2
index 4867f5d..dd5a56b 100644
@@ -1210,7 +1210,7 @@ static inline void prepare_page_table(void)
 
 #ifdef CONFIG_XIP_KERNEL
 	/* The XIP kernel is mapped in the module area -- skip over it */
-	addr = ((unsigned long)_etext + PMD_SIZE - 1) & PMD_MASK;
+	addr = ((unsigned long)_edata_loc + PMD_SIZE - 1) & PMD_MASK;
 #endif
 	for ( ; addr < PAGE_OFFSET; addr += PMD_SIZE)
 		pmd_clear(pmd_off_k(addr));
@@ -1292,7 +1292,7 @@ static void __init devicemaps_init(const struct machine_desc *mdesc)
 #ifdef CONFIG_XIP_KERNEL
 	map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
 	map.virtual = MODULES_VADDR;
-	map.length = ((unsigned long)_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
+	map.length = ((unsigned long)_edata_loc - map.virtual + ~SECTION_MASK) & SECTION_MASK;
 	map.type = MT_ROM;
 	create_mapping(&map);
 #endif
include/asm-generic/sections.h +1 -1
index b58fd66..195554d 100644
@@ -26,7 +26,7 @@
  *	__ctors_start, __ctors_end
  */
 extern char _text[], _stext[], _etext[];
-extern char _data[], _sdata[], _edata[];
+extern char _data[], _sdata[], _edata[], _edata_loc[];
 extern char __bss_start[], __bss_stop[];
 extern char __init_begin[], __init_end[];
 extern char _sinittext[], _einittext[];
Signature · 1 line
1.7.9.5
#1 lore raw
Quoted 46 lines · Chris Brandt

For an XIP build, _edata_loc, not _etext, represents the end of the binary image that will be programmed into ROM and mapped into the MODULES_VADDR area.
With an XIP kernel, nothing is loaded into RAM before boot, meaning you have to take into account the size of the entire binary image that was programmed, including the init data values that will be copied to RAM during kernel boot.

This fixes the bug where you might lose the end of your kernel area after page table setup is complete.

---
 arch/arm/mm/mmu.c              |    4 ++--
 include/asm-generic/sections.h |    2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
arch/arm/mm/mmu.c +2 -2
index 4867f5d..dd5a56b 100644
@@ -1210,7 +1210,7 @@ static inline void prepare_page_table(void)
 
 #ifdef CONFIG_XIP_KERNEL
 	/* The XIP kernel is mapped in the module area -- skip over it */
-	addr = ((unsigned long)_etext + PMD_SIZE - 1) & PMD_MASK;
+	addr = ((unsigned long)_edata_loc + PMD_SIZE - 1) & PMD_MASK;
 #endif
 	for ( ; addr < PAGE_OFFSET; addr += PMD_SIZE)
 		pmd_clear(pmd_off_k(addr));
@@ -1292,7 +1292,7 @@ static void __init devicemaps_init(const struct machine_desc *mdesc)
 #ifdef CONFIG_XIP_KERNEL
 	map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
 	map.virtual = MODULES_VADDR;
-	map.length = ((unsigned long)_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
+	map.length = ((unsigned long)_edata_loc - map.virtual + ~SECTION_MASK) & SECTION_MASK;
 	map.type = MT_ROM;
 	create_mapping(&map);
 #endif
include/asm-generic/sections.h +0 -0
index b58fd66..195554d 100644

include/asm-generic/sections.h line 32

Quoted 8 lines
include/asm-generic/sections.h +1 -1
@@ -26,7 +26,7 @@
  *	__ctors_start, __ctors_end
  */
 extern char _text[], _stext[], _etext[];
-extern char _data[], _sdata[], _edata[];
+extern char _data[], _sdata[], _edata[], _edata_loc[];
 extern char __bss_start[], __bss_stop[];
 extern char __init_begin[], __init_end[];
 extern char _sinittext[], _einittext[];

I think below is required as well.

arch/arm/kernel/module.c +1 -1
@@ -34,7 +36,7 @@
  * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned on/off.
  */
 #undef MODULES_VADDR
-#define MODULES_VADDR	(((unsigned long)_etext + ~PMD_MASK) & PMD_MASK)
+#define MODULES_VADDR	(((unsigned long)_edata_loc + ~PMD_MASK) & PMD_MASK)
 #endif
 

Regards,
Peter Hurley

#2 lore raw

arch/arm/kernel/module.c line 40

Quoted 11 lines

I think below is required as well.

arch/arm/kernel/module.c +1 -1
@@ -34,7 +36,7 @@
  * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned on/off.
  */
 #undef MODULES_VADDR
-#define MODULES_VADDR	(((unsigned long)_etext + ~PMD_MASK) & PMD_MASK)
+#define MODULES_VADDR	(((unsigned long)_edata_loc + ~PMD_MASK) & PMD_MASK)
 #endif

Thank you!
Somehow I missed that file when I was moving from one tree to another when preparing my commits.

I'll update and resubmit the patch.

Chris

#3 lore raw
In reply to Chris Brandt

include/asm-generic/sections.h line 29

Quoted 10 lines · Chris Brandt
include/asm-generic/sections.h +1 -1
index b58fd66..195554d 100644
@@ -26,7 +26,7 @@
  *	__ctors_start, __ctors_end
  */
 extern char _text[], _stext[], _etext[];
-extern char _data[], _sdata[], _edata[];
+extern char _data[], _sdata[], _edata[], _edata_loc[];

It's not good practice to put non-generic symbols in generic header files. _edata_loc is an ARM specific symbol which no other architecture defines. Please create an ARM private asm/sections.h header, which should include the asm-generic version, and also declare the additional symbol.

Thanks.

Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
#4 lore raw

PATCH v2 1/1 ARM: xip: Use correct symbol for end of ROM marker

In reply to Chris Brandt

For an XIP build, _edata_loc, not _etext, represents the end of the binary image that will be programmed into ROM and mapped into the MODULES_VADDR area.
With an XIP kernel, nothing is loaded into RAM before boot, meaning you have to take into account the size of the entire binary image that was programmed, including the init data values that will be copied to RAM during kernel boot.

This fixes the bug where you might lose the end of your kernel area after page table setup is complete.

---
v2
* Added change for MODULES_VADDR
* Moved extern to new file asm/sections.h

---
 arch/arm/include/asm/sections.h |    8 ++++++++
 arch/arm/kernel/module.c        |    2 +-
 arch/arm/mm/mmu.c               |    4 ++--
 3 files changed, 11 insertions(+), 3 deletions(-)
 create mode 100644 arch/arm/include/asm/sections.h
arch/arm/include/asm/sections.h new file +8 -0
new file mode 100644
index 0000000..401eb3c2
@@ -0,0 +1,8 @@
+#ifndef _ASM_ARM_SECTIONS_H
+#define _ASM_ARM_SECTIONS_H
+
+#include <asm-generic/sections.h>
+
+extern char _edata_loc[];
+
+#endif	/* _ASM_ARM_SECTIONS_H */
arch/arm/kernel/module.c +1 -1
index efdddcb..41ae2cc 100644
@@ -34,7 +34,7 @@
  * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned on/off.
  */
 #undef MODULES_VADDR
-#define MODULES_VADDR	(((unsigned long)_etext + ~PMD_MASK) & PMD_MASK)
+#define MODULES_VADDR	(((unsigned long)_edata_loc + ~PMD_MASK) & PMD_MASK)
 #endif
 
 #ifdef CONFIG_MMU
arch/arm/mm/mmu.c +2 -2
index 4867f5d..dd5a56b 100644
@@ -1210,7 +1210,7 @@ static inline void prepare_page_table(void)
 
 #ifdef CONFIG_XIP_KERNEL
 	/* The XIP kernel is mapped in the module area -- skip over it */
-	addr = ((unsigned long)_etext + PMD_SIZE - 1) & PMD_MASK;
+	addr = ((unsigned long)_edata_loc + PMD_SIZE - 1) & PMD_MASK;
 #endif
 	for ( ; addr < PAGE_OFFSET; addr += PMD_SIZE)
 		pmd_clear(pmd_off_k(addr));
@@ -1292,7 +1292,7 @@ static void __init devicemaps_init(const struct machine_desc *mdesc)
 #ifdef CONFIG_XIP_KERNEL
 	map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
 	map.virtual = MODULES_VADDR;
-	map.length = ((unsigned long)_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
+	map.length = ((unsigned long)_edata_loc - map.virtual + ~SECTION_MASK) & SECTION_MASK;
 	map.type = MT_ROM;
 	create_mapping(&map);
 #endif
Signature · 1 line
1.7.9.5
#5 lore raw

PATCH v2 1/1 ARM: xip: Use correct symbol for end of ROM marker

Hi Chris,

Quoted 10 lines · Chris Brandt

For an XIP build, _edata_loc, not _etext, represents the end of the binary image that will be programmed into ROM and mapped into the MODULES_VADDR area.
With an XIP kernel, nothing is loaded into RAM before boot, meaning you have to take into account the size of the entire binary image that was programmed, including the init data values that will be copied to RAM during kernel boot.

This fixes the bug where you might lose the end of your kernel area after page table setup is complete.

Thanks for your patch!

Quoted 11 lines

---
v2
* Added change for MODULES_VADDR
* Moved extern to new file asm/sections.h

---
 arch/arm/include/asm/sections.h |    8 ++++++++
 arch/arm/kernel/module.c        |    2 +-
 arch/arm/mm/mmu.c               |    4 ++--
 3 files changed, 11 insertions(+), 3 deletions(-)
 create mode 100644 arch/arm/include/asm/sections.h

If you create an arm-specific <asm/sections.h>, you also have to remove the line

        generic-y += sections.h

from arch/arm/include/asm/Kbuild.

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

#6 lore raw

PATCH v2 1/1 ARM: xip: Use correct symbol for end of ROM marker

Hi Geert,

Quoted 5 lines

If you create an arm-specific <asm/sections.h>, you also have to remove the line

        generic-y += sections.h

from arch/arm/include/asm/Kbuild.

That, I didn't know about. Thank you.

I'll submit a new patch on Monday.

Chris

#7 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

In reply to Chris Brandt

For an XIP build, _edata_loc, not _etext, represents the end of the binary image that will be programmed into ROM and mapped into the MODULES_VADDR area.
With an XIP kernel, nothing is loaded into RAM before boot, meaning you have to take into account the size of the entire binary image that was programmed, including the init data values that will be copied to RAM during kernel boot.

This fixes the bug where you might lose the end of your kernel area after page table setup is complete.

---
v3:
* Removed sections.h from Kbuild
v2:
* Added change for MODULES_VADDR
* Moved extern to new file asm/sections.h

---
 arch/arm/include/asm/Kbuild     |    1 -
 arch/arm/include/asm/sections.h |    8 ++++++++
 arch/arm/kernel/module.c        |    2 +-
 arch/arm/mm/mmu.c               |    4 ++--
 4 files changed, 11 insertions(+), 4 deletions(-)
 create mode 100644 arch/arm/include/asm/sections.h
arch/arm/include/asm/Kbuild +0 -1
index be648eb..6d3da22 100644
@@ -21,7 +21,6 @@ generic-y += preempt.h
 generic-y += resource.h
 generic-y += rwsem.h
 generic-y += seccomp.h
-generic-y += sections.h
 generic-y += segment.h
 generic-y += sembuf.h
 generic-y += serial.h
arch/arm/include/asm/sections.h new file +8 -0
new file mode 100644
index 0000000..401eb3c2
@@ -0,0 +1,8 @@
+#ifndef _ASM_ARM_SECTIONS_H
+#define _ASM_ARM_SECTIONS_H
+
+#include <asm-generic/sections.h>
+
+extern char _edata_loc[];
+
+#endif	/* _ASM_ARM_SECTIONS_H */
arch/arm/kernel/module.c +1 -1
index efdddcb..41ae2cc 100644
@@ -34,7 +34,7 @@
  * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned on/off.
  */
 #undef MODULES_VADDR
-#define MODULES_VADDR	(((unsigned long)_etext + ~PMD_MASK) & PMD_MASK)
+#define MODULES_VADDR	(((unsigned long)_edata_loc + ~PMD_MASK) & PMD_MASK)
 #endif
 
 #ifdef CONFIG_MMU
arch/arm/mm/mmu.c +2 -2
index 4867f5d..dd5a56b 100644
@@ -1210,7 +1210,7 @@ static inline void prepare_page_table(void)
 
 #ifdef CONFIG_XIP_KERNEL
 	/* The XIP kernel is mapped in the module area -- skip over it */
-	addr = ((unsigned long)_etext + PMD_SIZE - 1) & PMD_MASK;
+	addr = ((unsigned long)_edata_loc + PMD_SIZE - 1) & PMD_MASK;
 #endif
 	for ( ; addr < PAGE_OFFSET; addr += PMD_SIZE)
 		pmd_clear(pmd_off_k(addr));
@@ -1292,7 +1292,7 @@ static void __init devicemaps_init(const struct machine_desc *mdesc)
 #ifdef CONFIG_XIP_KERNEL
 	map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
 	map.virtual = MODULES_VADDR;
-	map.length = ((unsigned long)_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
+	map.length = ((unsigned long)_edata_loc - map.virtual + ~SECTION_MASK) & SECTION_MASK;
 	map.type = MT_ROM;
 	create_mapping(&map);
 #endif
Signature · 1 line
1.7.9.5
#8 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Nico,

As you originally created the XIP stuff, I hope you can remember the details - can you check this patch please?

I'm thinking that we need a new symbol around here:

#ifdef CONFIG_XIP_KERNEL
        __data_loc = ALIGN(4);          /* location in binary */
					<=== here
        . = PAGE_OFFSET + TEXT_OFFSET;

#else

to denote the end of the XIP kernel image which must remain accessible after boot. We don't need the data sections because they will have been copied to RAM, and we probably don't want to keep those exposed (it's potentially useful for attackers.)

Quoted 91 lines · Chris Brandt

For an XIP build, _edata_loc, not _etext, represents the end of the binary image that will be programmed into ROM and mapped into the MODULES_VADDR area.
With an XIP kernel, nothing is loaded into RAM before boot, meaning you have to take into account the size of the entire binary image that was programmed, including the init data values that will be copied to RAM during kernel boot.

This fixes the bug where you might lose the end of your kernel area after page table setup is complete.

---
v3:
* Removed sections.h from Kbuild
v2:
* Added change for MODULES_VADDR
* Moved extern to new file asm/sections.h

---
 arch/arm/include/asm/Kbuild     |    1 -
 arch/arm/include/asm/sections.h |    8 ++++++++
 arch/arm/kernel/module.c        |    2 +-
 arch/arm/mm/mmu.c               |    4 ++--
 4 files changed, 11 insertions(+), 4 deletions(-)
 create mode 100644 arch/arm/include/asm/sections.h
arch/arm/include/asm/Kbuild +0 -1
index be648eb..6d3da22 100644
@@ -21,7 +21,6 @@ generic-y += preempt.h
 generic-y += resource.h
 generic-y += rwsem.h
 generic-y += seccomp.h
-generic-y += sections.h
 generic-y += segment.h
 generic-y += sembuf.h
 generic-y += serial.h
arch/arm/include/asm/sections.h new file +8 -0
new file mode 100644
index 0000000..401eb3c2
@@ -0,0 +1,8 @@
+#ifndef _ASM_ARM_SECTIONS_H
+#define _ASM_ARM_SECTIONS_H
+
+#include <asm-generic/sections.h>
+
+extern char _edata_loc[];
+
+#endif	/* _ASM_ARM_SECTIONS_H */
arch/arm/kernel/module.c +1 -1
index efdddcb..41ae2cc 100644
@@ -34,7 +34,7 @@
  * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned on/off.
  */
 #undef MODULES_VADDR
-#define MODULES_VADDR	(((unsigned long)_etext + ~PMD_MASK) & PMD_MASK)
+#define MODULES_VADDR	(((unsigned long)_edata_loc + ~PMD_MASK) & PMD_MASK)
 #endif
 
 #ifdef CONFIG_MMU
arch/arm/mm/mmu.c +2 -2
index 4867f5d..dd5a56b 100644
@@ -1210,7 +1210,7 @@ static inline void prepare_page_table(void)
 
 #ifdef CONFIG_XIP_KERNEL
 	/* The XIP kernel is mapped in the module area -- skip over it */
-	addr = ((unsigned long)_etext + PMD_SIZE - 1) & PMD_MASK;
+	addr = ((unsigned long)_edata_loc + PMD_SIZE - 1) & PMD_MASK;
 #endif
 	for ( ; addr < PAGE_OFFSET; addr += PMD_SIZE)
 		pmd_clear(pmd_off_k(addr));
@@ -1292,7 +1292,7 @@ static void __init devicemaps_init(const struct machine_desc *mdesc)
 #ifdef CONFIG_XIP_KERNEL
 	map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
 	map.virtual = MODULES_VADDR;
-	map.length = ((unsigned long)_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
+	map.length = ((unsigned long)_edata_loc - map.virtual + ~SECTION_MASK) & SECTION_MASK;
 	map.type = MT_ROM;
 	create_mapping(&map);
 #endif
Signature · 1 line
1.7.9.5
Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
#9 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 3 lines

We don't need the data sections because they will have been copied to RAM, and we probably don't want to keep those exposed (it's potentially useful for attackers.)

The init sections also hang around after boot as well (it's XIP code, so there is nothing to 'free' in terms of executable init code). Any potential security issues there as well? Should the data and init-text sections be put in a separate section that gets blown away after init-data is freed?

Chris

#10 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 9 lines · Chris Brandt
Quoted 3 lines

We don't need the data sections because they will have been copied to RAM, and we probably don't want to keep those exposed (it's potentially useful for attackers.)

The init sections also hang around after boot as well (it's XIP code, so there is nothing to 'free' in terms of executable init code). Any potential security issues there as well? Should the data and init- text sections be put in a separate section that gets blown away after init-data is freed?

That's much harder to do - generally for XIP, people are space limited (which is why they're using XIP rather than putting the kernel in RAM.) They won't take kindly to having the kernel image bloated by 1MB just to pad it out so that the init stuff can be unmapped.

However, from the security point of view, the less that's mapped at known addresses, the better.

Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
#11 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 7 lines

That's much harder to do - generally for XIP, people are space limited (which is why they're using XIP rather than putting the kernel in RAM.) They won't take kindly to having the kernel image bloated by 1MB just to pad it out so that the init stuff can be unmapped.

However, from the security point of view, the less that's mapped at known addresses, the better.

Maybe the only true solution is to have some additional XIP Kconfig option that tells the kernel to not copy any of the init or data sections at all at boot, but instead leave it up to pre-boot to preload the physical locations with the correct values. That would get rid of your 1MB bloated boundary and you could even save some valuable XIP-able ROM/Flash space (you could pre-load from serial flash).

However, while that might provide more flexibility and ROM savings, it means now the pre-boot needs to know what values need to get pre-loaded where...adding more complexity to what right now is a very simple boot (set the PC and run).

- Chris

#12 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Please wrap your messages at or before column 72, thanks.

Quoted 3 lines · Chris Brandt

Maybe the only true solution is to have some additional XIP Kconfig option that tells the kernel to not copy any of the init or data sections at all at boot,

An XIP kernel will copy the .data section from ROM into RAM for you. You don't need additional code external to the kernel to perfor this.

Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
#13 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

In reply to Russell King - ARM Linux
Quoted 11 lines · Russell King - ARM Linux

Nico,

As you originally created the XIP stuff, I hope you can remember the details - can you check this patch please?

I'm thinking that we need a new symbol around here:

#ifdef CONFIG_XIP_KERNEL
        __data_loc = ALIGN(4);          /* location in binary */
					<=== here
        . = PAGE_OFFSET + TEXT_OFFSET;
Quoted 6 lines

#else

to denote the end of the XIP kernel image which must remain accessible after boot. We don't need the data sections because they will have been copied to RAM, and we probably don't want to keep those exposed (it's potentially useful for attackers.)

The _etext symbol is already used for that purpose.

Now we round it up to the next section mapping which might leave quite a lot of data content exposed in ROM. But given it is more or less the same data present in RAM except for those bits that are modified at run time, I don't see what an attacker would gain from the data in ROM that cannot already be obtained from kernel RAM. If the section mapping extends to part of the ROM that is no longer kernel data then maybe this would expose sensitive data. Is that what you're worried about?

Nicolas

#14 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 3 lines

If the section mapping extends to part of the ROM that is no longer kernel data then maybe this would expose sensitive data. Is that what you're worried about?

That is a good point that I did not consider before. Thank you.

Chris

#15 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

In reply to Chris Brandt
Quoted 3 lines · Chris Brandt

For an XIP build, _edata_loc, not _etext, represents the end of the binary image that will be programmed into ROM and mapped into the MODULES_VADDR area.

This statement is wrong. The kernel data is copied to RAM before the MMU is turned on and that's the copy that the kernel must use. There is no reason having anything past _etext accessible from ROM.

Quoted 4 lines

With an XIP kernel, nothing is loaded into RAM before boot, meaning you have to take into account the size of the entire binary image that was programmed, including the init data values that will be copied to RAM during kernel boot.

But this is done way before the code you're modifying is executed.

Quoted 2 lines

This fixes the bug where you might lose the end of your kernel area after page table setup is complete.

Could you elaborate about a possible bug please?

Quoted 87 lines

---
v3:
* Removed sections.h from Kbuild
v2:
* Added change for MODULES_VADDR
* Moved extern to new file asm/sections.h

---
 arch/arm/include/asm/Kbuild     |    1 -
 arch/arm/include/asm/sections.h |    8 ++++++++
 arch/arm/kernel/module.c        |    2 +-
 arch/arm/mm/mmu.c               |    4 ++--
 4 files changed, 11 insertions(+), 4 deletions(-)
 create mode 100644 arch/arm/include/asm/sections.h
arch/arm/include/asm/Kbuild +0 -1
index be648eb..6d3da22 100644
@@ -21,7 +21,6 @@ generic-y += preempt.h
 generic-y += resource.h
 generic-y += rwsem.h
 generic-y += seccomp.h
-generic-y += sections.h
 generic-y += segment.h
 generic-y += sembuf.h
 generic-y += serial.h
arch/arm/include/asm/sections.h new file +8 -0
new file mode 100644
index 0000000..401eb3c2
@@ -0,0 +1,8 @@
+#ifndef _ASM_ARM_SECTIONS_H
+#define _ASM_ARM_SECTIONS_H
+
+#include <asm-generic/sections.h>
+
+extern char _edata_loc[];
+
+#endif	/* _ASM_ARM_SECTIONS_H */
arch/arm/kernel/module.c +1 -1
index efdddcb..41ae2cc 100644
@@ -34,7 +34,7 @@
  * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned on/off.
  */
 #undef MODULES_VADDR
-#define MODULES_VADDR	(((unsigned long)_etext + ~PMD_MASK) & PMD_MASK)
+#define MODULES_VADDR	(((unsigned long)_edata_loc + ~PMD_MASK) & PMD_MASK)
 #endif
 
 #ifdef CONFIG_MMU
arch/arm/mm/mmu.c +2 -2
index 4867f5d..dd5a56b 100644
@@ -1210,7 +1210,7 @@ static inline void prepare_page_table(void)
 
 #ifdef CONFIG_XIP_KERNEL
 	/* The XIP kernel is mapped in the module area -- skip over it */
-	addr = ((unsigned long)_etext + PMD_SIZE - 1) & PMD_MASK;
+	addr = ((unsigned long)_edata_loc + PMD_SIZE - 1) & PMD_MASK;
 #endif
 	for ( ; addr < PAGE_OFFSET; addr += PMD_SIZE)
 		pmd_clear(pmd_off_k(addr));
@@ -1292,7 +1292,7 @@ static void __init devicemaps_init(const struct machine_desc *mdesc)
 #ifdef CONFIG_XIP_KERNEL
 	map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
 	map.virtual = MODULES_VADDR;
-	map.length = ((unsigned long)_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
+	map.length = ((unsigned long)_edata_loc - map.virtual + ~SECTION_MASK) & SECTION_MASK;
 	map.type = MT_ROM;
 	create_mapping(&map);
 #endif
Signature · 8 lines
1.7.9.5



_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#16 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 7 lines
Quoted 3 lines

For an XIP build, _edata_loc, not _etext, represents the end of the binary image that will be programmed into ROM and mapped into the MODULES_VADDR area.

This statement is wrong. The kernel data is copied to RAM before the MMU is turned on and that's the copy that the kernel must use. There is no reason having anything past _etext accessible from ROM.

If I remember correctly, when I first ran into an issue was my initramfs got cut off because it spanned the 1MB boundary after _etext.

Also, your init code resides after _etext, so that get cut off too.

Quoted 6 lines
Quoted 4 lines

With an XIP kernel, nothing is loaded into RAM before boot, meaning you have to take into account the size of the entire binary image that was programmed, including the init data values that will be copied to RAM during kernel boot.

But this is done way before the code you're modifying is executed.

If the MMU is set up to cut everything off after _etext early on, then anything in the init section gets blown away, and there are lots of init functions in the kernel that get run at various times during boot.

Quoted 4 lines
Quoted 2 lines

This fixes the bug where you might lose the end of your kernel area after page table setup is complete.

Could you elaborate about a possible bug please?

During boot, the kernel attempts to execute code that is not mapped anywhere, and hence crashes.

-Chris

#17 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 11 lines · Chris Brandt
Quoted 7 lines
Quoted 3 lines

For an XIP build, _edata_loc, not _etext, represents the end of the binary image that will be programmed into ROM and mapped into the MODULES_VADDR area.

This statement is wrong. The kernel data is copied to RAM before the MMU is turned on and that's the copy that the kernel must use. There is no reason having anything past _etext accessible from ROM.

If I remember correctly, when I first ran into an issue was my initramfs got cut off because it spanned the 1MB boundary after _etext.

If you are convinced that you need kernel XIP, then you normally would want _not_ to use initramfs at all.

Quoted 2 lines

Also, your init code resides after _etext, so that get cut off too.

/me looks at the linker script and cry

Forget it. XIP kernel is simply completely broken at the moment.

There used to be conditional placement of .init.text that was located in the .text section when XIP_KERNEL was selected, or in the .init section otherwise. This particular section was abstracted away into an INIT_TEXT_SECTION macro defined in include/asm-generic/vmlinux.lds.h and the XIP conditional placement got lost. New sections (lots of them) were added as well since the introduction of XIP kernel with no consideration for XIP usage.

The old method with conditional placement in the linker script is broken beyond repair. There is simply way too much stuff these days to do this sanely. I think the only way to fix it properly is to have a separate linker script for XIP usage with proper (and obvious) section placement.

Nicolas

#18 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 2 lines

If you are convinced that you need kernel XIP, then you normally would want _not_ to use initramfs at all.

True, but in trying to test upstream kernels, initramfs was easier than fighting with device drivers (although now I've just switched to using memory mapped mtd-rom).

Quoted 1 line

Forget it. XIP kernel is simply completely broken at the moment.

Yes, there were multiple things wrong that took me a while to find, but once you get them all working, it's pretty cool. Note, the Kconfig issues are going to be a bit harder to get approval.

Quoted 2 lines

New sections (lots of them) were added as well since the introduction of XIP kernel with no consideration for XIP usage.

So I've noticed. As far as I can tell, the XIP kernel stuff broke years ago.

Quoted 2 lines

I think the only way to fix it properly is to have a separate linker script for XIP usage with proper (and obvious) section placement.

I guess then the parts of the kernel that refer to section locations need to be looked at again. As you can see from my other patches, there's a bit of miss-match that I was trying to align.

So at this point...I'm assume all my XIP 'hacks' are being rejected.

As for making a new XIP-only linker script, note that I really only work with the ARMv7. I know the M4 and M5 guys do XIP_KERNEL builds, but I've got no way of testing their XIP_KERNEL builds or knowing what will break.

Chris

#19 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 5 lines · Chris Brandt
Quoted 2 lines

New sections (lots of them) were added as well since the introduction of XIP kernel with no consideration for XIP usage.

So I've noticed. As far as I can tell, the XIP kernel stuff broke years ago.

That, of course, brings the question about keeping XIP around.

Quoted 6 lines
Quoted 2 lines

I think the only way to fix it properly is to have a separate linker script for XIP usage with proper (and obvious) section placement.

I guess then the parts of the kernel that refer to section locations need to be looked at again. As you can see from my other patches, there's a bit of miss-match that I was trying to align.

I don't think that's a big issue. Those section locations are obtained from symbols that the linker script creates explicitly.

Quoted 1 line

So at this point...I'm assume all my XIP 'hacks' are being rejected.

Pretty much. Everything can be fixed at once with a new linker script.

Quoted 4 lines

As for making a new XIP-only linker script, note that I really only work with the ARMv7. I know the M4 and M5 guys do XIP_KERNEL builds, but I've got no way of testing their XIP_KERNEL builds or knowing what will break.

Things can hardly be worse than they are right now.

Nicolas

#20 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 4 lines
Quoted 2 lines

So I've noticed. As far as I can tell, the XIP kernel stuff broke years ago.

That, of course, brings the question about keeping XIP around.

Well, as the email address suggests, I work for Renesas. And since the ARCH_R7S72100 has up to 10MB internal RAM and can XIP directly from QSPI Flash, that's a pretty simple BOM.

I also revisted the AXFS file system (that never made it into the kernel) and like the kernel, required some fixing up. However, I can get a functioning and usefull ARMv7 system up and running in under 3MB of total system RAM.
That's got to be worth something to someone ;)

Quoted 1 line

Everything can be fixed at once with a new linker script.

I do have to agree that a separate linker script would be a cleaner solution for the section arrangements.
Of course, there are still some other XIP patches needed (like you can't use text areas as a temporary stack on boot), but the sections are a pretty logical place to start.

Chris

#21 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 9 lines · Chris Brandt
Quoted 4 lines
Quoted 2 lines

So I've noticed. As far as I can tell, the XIP kernel stuff broke years ago.

That, of course, brings the question about keeping XIP around.

Well, as the email address suggests, I work for Renesas. And since the ARCH_R7S72100 has up to 10MB internal RAM and can XIP directly from QSPI Flash, that's a pretty simple BOM.

Quoted 5 lines

I also revisted the AXFS file system (that never made it into the kernel) and like the kernel, required some fixing up. However, I can get a functioning and usefull ARMv7 system up and running in under 3MB of total system RAM.
That's got to be worth something to someone ;)

Absolutely. I'll be glad to review your patches.

Quoted 7 lines
Quoted 1 line

Everything can be fixed at once with a new linker script.

I do have to agree that a separate linker script would be a cleaner solution for the section arrangements.
Of course, there are still some other XIP patches needed (like you can't use text areas as a temporary stack on boot), but the sections are a pretty logical place to start.

Maybe this can be prevented automatically by the linker with some read-only memory regions or the like.

Nicolas

#22 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 1 line

Absolutely. I'll be glad to review your patches.

What do you think about this submission?
I think it still holds true regardless of the incorrect sectioning.

http://lists.infradead.org/pipermail/linux-arm-kernel/2015-November/384354.html

Quoted 6 lines
Quoted 3 lines

Of course, there are still some other XIP patches needed (like you can't use text areas as a temporary stack on boot), but the sections are a pretty logical place to start.

Maybe this can be prevented automatically by the linker with some read-only memory regions or the like.

I think this one is more of a coding issue. These were the attempts to fix the temporary stack issue:

My first patch:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-July/357106.html

Magnus's try:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-November/383394.html

Chris

#23 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 6 lines · Chris Brandt
Quoted 1 line

Absolutely. I'll be glad to review your patches.

What do you think about this submission?
I think it still holds true regardless of the incorrect sectioning.

http://lists.infradead.org/pipermail/linux-arm-kernel/2015-November/384354.html

This patch is correct.

Quoted 4 lines
Quoted 4 lines
Quoted 3 lines

Of course, there are still some other XIP patches needed (like you can't use text areas as a temporary stack on boot), but the sections are a pretty logical place to start.

Quoted 8 lines
Quoted 2 lines

Maybe this can be prevented automatically by the linker with some read-only memory regions or the like.

I think this one is more of a coding issue. These were the attempts to fix the temporary stack issue:

My first patch:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-July/357106.html

I think this ought to be fixed regardless of the XIP issue.

Quoted 2 lines

Magnus's try:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-November/383394.html

This one is wrong as it clobbers r11.

Nicolas

#24 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

In reply to Chris Brandt
Quoted 8 lines · Chris Brandt

I think this one is more of a coding issue. These were the attempts to fix the temporary stack issue:

My first patch:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-July/357106.html

Magnus's try:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-November/383394.html

Here's my proposal:

arch/arm/mm/proc-v7.S +16 -7
index de2b246fed..2d0ac32320 100644
@@ -274,10 +274,12 @@ __v7_ca15mp_setup:
 __v7_b15mp_setup:
 __v7_ca17mp_setup:
 	mov	r10, #0
-1:	adr	r12, __v7_setup_stack		@ the local stack
-	stmia	r12, {r0-r5, lr}		@ v7_invalidate_l1 touches r0-r6
+1:	adr	r0, __v7_setup_stack_ptr
+	ldr	r12, [r0]
+	add	r12, r12, r0			@ the local stack
+	stmia	r12, {r1-r6, lr}		@ v7_invalidate_l1 touches r0-r6
 	bl      v7_invalidate_l1
-	ldmia	r12, {r0-r5, lr}
+	ldmia	r12, {r1-r6, lr}
 #ifdef CONFIG_SMP
 	ALT_SMP(mrc	p15, 0, r0, c1, c0, 1)
 	ALT_UP(mov	r0, #(1 << 6))		@ fake it for UP
@@ -415,10 +417,12 @@ __v7_pj4b_setup:
 #endif /* CONFIG_CPU_PJ4B */
 
 __v7_setup:
-	adr	r12, __v7_setup_stack		@ the local stack
-	stmia	r12, {r0-r5, lr}		@ v7_invalidate_l1 touches r0-r6
+	adr	r0, __v7_setup_stack_ptr
+	ldr	r12, [r0]
+	add	r12, r12, r0			@ the local stack
+	stmia	r12, {r1-r6, lr}		@ v7_invalidate_l1 touches r0-r6
 	bl      v7_invalidate_l1
-	ldmia	r12, {r0-r5, lr}
+	ldmia	r12, {r1-r6, lr}
 
 __v7_setup_cont:
 	and	r0, r9, #0xff000000		@ ARM?
@@ -480,11 +484,16 @@ __errata_finish:
 	orr	r0, r0, r6			@ set them
  THUMB(	orr	r0, r0, #1 << 30	)	@ Thumb exceptions
 	ret	lr				@ return to head.S:__ret
+
+	.align	2
+__v7_setup_stack_ptr:
+	.word	__v7_setup_stack - .
 ENDPROC(__v7_setup)
 
+	.bss
 	.align	2
 __v7_setup_stack:
-	.space	4 * 7				@ 12 registers
+	.space	4 * 7				@ 7 registers
 
 	__INITDATA
 
arch/arm/mm/proc-v7m.S +2 -0
index 67d9209077..8cd465927a 100644
@@ -123,10 +123,12 @@ __v7m_setup:
 	ret	lr
 ENDPROC(__v7m_setup)
 
+	.pushsection .bss
 	.align 2
 __v7m_setup_stack:
 	.space	4 * 8				@ 8 registers
 __v7m_setup_stack_top:
+	.previous
 
 	define_processor_functions v7m, dabort=nommu_early_abort, pabort=legacy_pabort, nommu=1
 
#25 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Hi Nicolas,

Quoted 33 lines · Nicolas Pitre
Quoted 8 lines · Chris Brandt

I think this one is more of a coding issue. These were the attempts to fix the temporary stack issue:

My first patch:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-July/357106.html

Magnus's try:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-November/383394.html

Here's my proposal:

arch/arm/mm/proc-v7.S +5 -3
index de2b246fed..2d0ac32320 100644
@@ -274,10 +274,12 @@ __v7_ca15mp_setup:
 __v7_b15mp_setup:
 __v7_ca17mp_setup:
        mov     r10, #0
-1:     adr     r12, __v7_setup_stack           @ the local stack
-       stmia   r12, {r0-r5, lr}                @ v7_invalidate_l1 touches r0-r6
+1:     adr     r0, __v7_setup_stack_ptr
+       ldr     r12, [r0]
+       add     r12, r12, r0                    @ the local stack
+       stmia   r12, {r1-r6, lr}                @ v7_invalidate_l1 touches r0-r6
        bl      v7_invalidate_l1
-       ldmia   r12, {r0-r5, lr}
+       ldmia   r12, {r1-r6, lr}
 #ifdef CONFIG_SMP
        ALT_SMP(mrc     p15, 0, r0, c1, c0, 1)
        ALT_UP(mov      r0, #(1 << 6))          @ fake it for UP

[snip]

arch/arm/mm/proc-v7.S __errata_finish: · line 491

Quoted 9 lines
arch/arm/mm/proc-v7.S +4 -0
@@ -480,11 +484,16 @@ __errata_finish:
        orr     r0, r0, r6                      @ set them
  THUMB(        orr     r0, r0, #1 << 30        )       @ Thumb exceptions
        ret     lr                              @ return to head.S:__ret
+
+       .align  2
+__v7_setup_stack_ptr:
+       .word   __v7_setup_stack - .
 ENDPROC(__v7_setup)

Thanks for your take on this. I did a couple of local implementations before submitting, and one of the issues I ran into was the need to get rid of PAGE_OFFSET due to the code running without MMU enabled. I suppose that is taken care of the "__v7_setup_stack - ." calculation above?

Cheers,

/ magnus

#26 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 39 lines · Magnus Damm

Hi Nicolas,

Quoted 33 lines · Nicolas Pitre
Quoted 8 lines · Chris Brandt

I think this one is more of a coding issue. These were the attempts to fix the temporary stack issue:

My first patch:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-July/357106.html

Magnus's try:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-November/383394.html

Here's my proposal:

arch/arm/mm/proc-v7.S +5 -3
index de2b246fed..2d0ac32320 100644
@@ -274,10 +274,12 @@ __v7_ca15mp_setup:
 __v7_b15mp_setup:
 __v7_ca17mp_setup:
        mov     r10, #0
-1:     adr     r12, __v7_setup_stack           @ the local stack
-       stmia   r12, {r0-r5, lr}                @ v7_invalidate_l1 touches r0-r6
+1:     adr     r0, __v7_setup_stack_ptr
+       ldr     r12, [r0]
+       add     r12, r12, r0                    @ the local stack
+       stmia   r12, {r1-r6, lr}                @ v7_invalidate_l1 touches r0-r6
        bl      v7_invalidate_l1
-       ldmia   r12, {r0-r5, lr}
+       ldmia   r12, {r1-r6, lr}
 #ifdef CONFIG_SMP
        ALT_SMP(mrc     p15, 0, r0, c1, c0, 1)
        ALT_UP(mov      r0, #(1 << 6))          @ fake it for UP

[snip]

Quoted 15 lines

arch/arm/mm/proc-v7.S __errata_finish: · line 491

Quoted 9 lines
arch/arm/mm/proc-v7.S +4 -0
@@ -480,11 +484,16 @@ __errata_finish:
        orr     r0, r0, r6                      @ set them
  THUMB(        orr     r0, r0, #1 << 30        )       @ Thumb exceptions
        ret     lr                              @ return to head.S:__ret
+
+       .align  2
+__v7_setup_stack_ptr:
+       .word   __v7_setup_stack - .
 ENDPROC(__v7_setup)

Thanks for your take on this. I did a couple of local implementations before submitting, and one of the issues I ran into was the need to get rid of PAGE_OFFSET due to the code running without MMU enabled. I suppose that is taken care of the "__v7_setup_stack - ." calculation above?

Yes. That provides the offset from __v7_setup_stack_ptr to reach __v7_setup_stack. And __v7_setup_stack_ptr is obtained with adr which is relative to the current pc. So this works whether or not the MMU is enabled.

Nicolas

#27 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 8 lines · Nicolas Pitre
Quoted 5 lines · Magnus Damm

Thanks for your take on this. I did a couple of local implementations before submitting, and one of the issues I ran into was the need to get rid of PAGE_OFFSET due to the code running without MMU enabled. I suppose that is taken care of the "__v7_setup_stack - ." calculation above?

Quoted 4 lines

Yes. That provides the offset from __v7_setup_stack_ptr to reach __v7_setup_stack. And __v7_setup_stack_ptr is obtained with adr which is relative to the current pc. So this works whether or not the MMU is enabled.

Here's the patch with proper changelog, etc. I don't have XIP capable hardware to test it with though.

------ >8
From: Nicolas Pitre <[email protected]>
Subject: [PATCH] ARM: proc-v7*.S: don't locate temporary stack space in .text section

Both proc-v7.S and proc-v7m.S are using a small temporary stack to preserve register content within their respective setup code. This stack is located in the .text section which is normally meant to be read-only. This is even more true in the context of an XIP kernel where .text is accessed from ROM directly.

Move that temporary stack to the .bss section and get its address in a position independent way.

While at it, one comments was updated to reflect reality, and the list of saved registers in the proc-v7.S case is updated to match the comment next to it for coherency.

arch/arm/mm/proc-v7.S +16 -7
index de2b246fed..2d0ac32320 100644
@@ -274,10 +274,12 @@ __v7_ca15mp_setup:
 __v7_b15mp_setup:
 __v7_ca17mp_setup:
 	mov	r10, #0
-1:	adr	r12, __v7_setup_stack		@ the local stack
-	stmia	r12, {r0-r5, lr}		@ v7_invalidate_l1 touches r0-r6
+1:	adr	r0, __v7_setup_stack_ptr
+	ldr	r12, [r0]
+	add	r12, r12, r0			@ the local stack
+	stmia	r12, {r1-r6, lr}		@ v7_invalidate_l1 touches r0-r6
 	bl      v7_invalidate_l1
-	ldmia	r12, {r0-r5, lr}
+	ldmia	r12, {r1-r6, lr}
 #ifdef CONFIG_SMP
 	ALT_SMP(mrc	p15, 0, r0, c1, c0, 1)
 	ALT_UP(mov	r0, #(1 << 6))		@ fake it for UP
@@ -415,10 +417,12 @@ __v7_pj4b_setup:
 #endif /* CONFIG_CPU_PJ4B */
 
 __v7_setup:
-	adr	r12, __v7_setup_stack		@ the local stack
-	stmia	r12, {r0-r5, lr}		@ v7_invalidate_l1 touches r0-r6
+	adr	r0, __v7_setup_stack_ptr
+	ldr	r12, [r0]
+	add	r12, r12, r0			@ the local stack
+	stmia	r12, {r1-r6, lr}		@ v7_invalidate_l1 touches r0-r6
 	bl      v7_invalidate_l1
-	ldmia	r12, {r0-r5, lr}
+	ldmia	r12, {r1-r6, lr}
 
 __v7_setup_cont:
 	and	r0, r9, #0xff000000		@ ARM?
@@ -480,11 +484,16 @@ __errata_finish:
 	orr	r0, r0, r6			@ set them
  THUMB(	orr	r0, r0, #1 << 30	)	@ Thumb exceptions
 	ret	lr				@ return to head.S:__ret
+
+	.align	2
+__v7_setup_stack_ptr:
+	.word	__v7_setup_stack - .
 ENDPROC(__v7_setup)
 
+	.bss
 	.align	2
 __v7_setup_stack:
-	.space	4 * 7				@ 12 registers
+	.space	4 * 7				@ 7 registers
 
 	__INITDATA
 
arch/arm/mm/proc-v7m.S +2 -0
index 67d9209077..8cd465927a 100644
@@ -123,10 +123,12 @@ __v7m_setup:
 	ret	lr
 ENDPROC(__v7m_setup)
 
+	.pushsection .bss
 	.align 2
 __v7m_setup_stack:
 	.space	4 * 8				@ 8 registers
 __v7m_setup_stack_top:
+	.previous
 
 	define_processor_functions v7m, dabort=nommu_early_abort, pabort=legacy_pabort, nommu=1
 
#28 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Hi Nicolas,

Quoted 2 lines

Here's the patch with proper changelog, etc. I don't have XIP capable hardware to test it with though.

I'm testing it now...but it's crashing.

I fired up GDB, so here the reason:

__v7_ca17mp_setup:

	mov	r10, #0

1: adr r0, __v7_setup_stack_ptr
r0=0x18213df4

	ldr	r12, [r0]

r12=0x10174cc

	add	r12, r12, r0			@ the local stack

r12=0x1922b2c0

	stmia	r12, {r1-r6, lr}		@ v7_invalidate_l1 touches r0-r6
	bl      v7_invalidate_l1

0x1922b2c0 is NOT RAM....it's nothing.

As point of reference, here's the memory map of my XIP system:
Physical ROM address: 0x18000000 (I have my XIP kernel starting at 0x18200000) Physical RAM address: 0x20000000
Virtual ROM address: 0xBF0000000
Virtual RAM address: 0xC00000000

Basically, you made the same mistake that Magnus first did: You can't rely on the current PC address to obtain an address in physical RAM because the ROM virt-to-phys relationship is different than the RAM virt-to-phys relationship.

No matter how you look at it, this early in boot, the only 'thing' that really knows where physical RAM is located is PLAT_PHYS_OFFSET.

That's why both Magnus's an my patches (that I both tested) contain PLAT_PHYS_OFFSET.

The only other 'slight' piece of info would be the DTB pointer is still sitting in R2 from boot, and at least for my system, that is in RAM. However, the DTB could be in ROM if the user wanted it to be, and I also think that this code might run twice in a SMP system...so even R2 would not be reliable. So...forget I mentioned it...

I don't see any way around not having a "#ifdef XIP_KERNEL" statement in order to expose PLAT_PHYS_OFFSET so you can properly do your translation math.

Chris

ps, I will add that I appreciate your efforts with this.

#29 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 36 lines · Chris Brandt

Hi Nicolas,

Quoted 2 lines

Here's the patch with proper changelog, etc. I don't have XIP capable hardware to test it with though.

I'm testing it now...but it's crashing.

I fired up GDB, so here the reason:

__v7_ca17mp_setup:

	mov	r10, #0

1: adr r0, __v7_setup_stack_ptr
r0=0x18213df4

	ldr	r12, [r0]

r12=0x10174cc

	add	r12, r12, r0			@ the local stack

r12=0x1922b2c0

	stmia	r12, {r1-r6, lr}		@ v7_invalidate_l1 touches r0-r6
	bl      v7_invalidate_l1

0x1922b2c0 is NOT RAM....it's nothing.

As point of reference, here's the memory map of my XIP system:
Physical ROM address: 0x18000000 (I have my XIP kernel starting at 0x18200000) Physical RAM address: 0x20000000
Virtual ROM address: 0xBF0000000
Virtual RAM address: 0xC00000000

Quoted 4 lines

Basically, you made the same mistake that Magnus first did: You can't rely on the current PC address to obtain an address in physical RAM because the ROM virt-to-phys relationship is different than the RAM virt-to-phys relationship.

Crap... you're right of course. I suspect a couple other places might have problems as they use similar constructs. See kernel/sleep.S for example.

Probably the best way to fix it would be something like:

in asm/memory.h or similar:

#ifdef CONFIG_XIP_KERNEL

#define PHYS_OFFSET_FIXUP \
	( XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR) - PAGE_OFFSET + \
	  PLAT_PHYS_OFFSET - CONFIG_XIP_PHYS_ADDR )

#else
#define PHYS_OFFSET_FIXUP 0
#endif

And then, after my patch is applied, changing:

__v7_setup_stack_ptr:

        .word   __v7_setup_stack - .

into:

__v7_setup_stack_ptr:
.word __v7_setup_stack - . + PHYS_OFFSET_FIXUP

should do the trick. This way it'll work for all those places where the code is getting at the data area when the MMU is off with no XIP conditionals in the code.

I think my patch should be applied as is (minus the mention of XIP) to remove the write access to the .text area for the general case which is a worthy goal in itself. We did a bunch of similar cleanups a while ago.

Then another patch could bring all those places XIP compatible with the simple addition of that PHYS_OFFSET_FIXUP constant.

Nicolas

#30 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 15 lines

Probably the best way to fix it would be something like:

in asm/memory.h or similar:

#ifdef CONFIG_XIP_KERNEL

#define PHYS_OFFSET_FIXUP \
	( XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR) - PAGE_OFFSET + \
	  PLAT_PHYS_OFFSET - CONFIG_XIP_PHYS_ADDR ) #else #define

PHYS_OFFSET_FIXUP 0 #endif

And then, after my patch is applied, changing:

__v7_setup_stack_ptr:

       .word   __v7_setup_stack - .
Quoted 8 lines

into:

__v7_setup_stack_ptr:
.word __v7_setup_stack - . + PHYS_OFFSET_FIXUP

should do the trick. This way it'll work for all those places where the code is getting at the data area when the MMU is off with no XIP conditionals in the code.

Tested....it works!

Here's your new results:

__v7_ca17mp_setup:

	mov	r10, #0

1: adr r0, __v7_setup_stack_ptr
r0=0x18213df4

	ldr	r12, [r0]

r12=0x7e174cc

	add	r12, r12, r0			@ the local stack

r12=0x2002b2c0

	stmia	r12, {r1-r6, lr}		@ v7_invalidate_l1 touches r0-r6
	bl      v7_invalidate_l1

0x2002b2c0 is indeed the correct physical address of __v7_setup_stack_ptr.

So, you still need to use #ifdef XIP_KERNEL and PLAT_PHYS_OFFSET....but...you bury it in another file. I'm good with that!!

Quoted 2 lines

Then another patch could bring all those places XIP compatible with the simple addition of that PHYS_OFFSET_FIXUP constant.

Yes, the less CONFIG_XIP_KERNEL is sprinkled around the arm tree code, the better.

Chris

#31 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 20 lines · Chris Brandt
Quoted 20 lines

Probably the best way to fix it would be something like:

in asm/memory.h or similar:

#ifdef CONFIG_XIP_KERNEL

#define PHYS_OFFSET_FIXUP \
	( XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR) - PAGE_OFFSET + \
	  PLAT_PHYS_OFFSET - CONFIG_XIP_PHYS_ADDR ) #else #define

PHYS_OFFSET_FIXUP 0 #endif

And then, after my patch is applied, changing:

__v7_setup_stack_ptr:

       .word   __v7_setup_stack - .

into:

__v7_setup_stack_ptr:
.word __v7_setup_stack - . + PHYS_OFFSET_FIXUP

Quoted 6 lines
Quoted 3 lines

should do the trick. This way it'll work for all those places where the code is getting at the data area when the MMU is off with no XIP conditionals in the code.

Tested....it works!

Excellent.

Nicolas

#32 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 27 lines · Nicolas Pitre
Quoted 25 lines · Chris Brandt
Quoted 23 lines

Probably the best way to fix it would be something like:

in asm/memory.h or similar:

#ifdef CONFIG_XIP_KERNEL

#define PHYS_OFFSET_FIXUP \
	( XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR) - PAGE_OFFSET + \
	  PLAT_PHYS_OFFSET - CONFIG_XIP_PHYS_ADDR ) #else #define  

PHYS_OFFSET_FIXUP 0 #endif

And then, after my patch is applied, changing:

__v7_setup_stack_ptr:

       .word   __v7_setup_stack - .

into:

__v7_setup_stack_ptr:
.word __v7_setup_stack - . + PHYS_OFFSET_FIXUP

should do the trick. This way it'll work for all those places where the code is getting at the data area when the MMU is off with no XIP conditionals in the code.

Quoted 6 lines
Quoted 1 line

Tested....it works!

Excellent.

Nicolas

Nicolas,

I applied your PHYS_OFFSET_FIXUP macro code (shown above) to the current tree which already has your commit "ARM: 8453/2: proc-v7.S: don't locate temporary stack space in .text section".

After testing, it shows the same results for an XIP_KERNEL build as it did in November:

No PHYS_OFFSET_FIXUP = crash

with PHYS_OFFSET_FIXUP = good

Can I submit a patch with your code for review and inclusion to the kernel?

Chris

#33 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 41 lines · Chris Brandt
Quoted 27 lines · Nicolas Pitre
Quoted 25 lines · Chris Brandt
Quoted 23 lines

Probably the best way to fix it would be something like:

in asm/memory.h or similar:

#ifdef CONFIG_XIP_KERNEL

#define PHYS_OFFSET_FIXUP \
	( XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR) - PAGE_OFFSET + \
	  PLAT_PHYS_OFFSET - CONFIG_XIP_PHYS_ADDR ) #else #define  

PHYS_OFFSET_FIXUP 0 #endif

And then, after my patch is applied, changing:

__v7_setup_stack_ptr:

       .word   __v7_setup_stack - .

into:

__v7_setup_stack_ptr:
.word __v7_setup_stack - . + PHYS_OFFSET_FIXUP

should do the trick. This way it'll work for all those places where the code is getting at the data area when the MMU is off with no XIP conditionals in the code.

Quoted 6 lines
Quoted 1 line

Tested....it works!

Excellent.

Nicolas

Nicolas,

I applied your PHYS_OFFSET_FIXUP macro code (shown above) to the current tree which already has your commit "ARM: 8453/2: proc-v7.S: don't locate temporary stack space in .text section".

Quoted 8 lines

After testing, it shows the same results for an XIP_KERNEL build as it did in November:

No PHYS_OFFSET_FIXUP = crash

with PHYS_OFFSET_FIXUP = good

Can I submit a patch with your code for review and inclusion to the kernel?

Sure.

Nicolas

#34 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

In reply to Nicolas Pitre
Quoted 5 lines

The old method with conditional placement in the linker script is broken beyond repair. There is simply way too much stuff these days to do this sanely. I think the only way to fix it properly is to have a separate linker script for XIP usage with proper (and obvious) section placement.

Step 1: Change the Makefiles to select an alternative linker script.

What do you think about doing the following?

Makefile +4 -0
index fd46821..956ad21 100644
@@ -898,7 +898,11 @@ libs-y		:= $(libs-y1) $(libs-y2)
 # Externally visible symbols (used by link-vmlinux.sh)
 export KBUILD_VMLINUX_INIT := $(head-y) $(init-y)
 export KBUILD_VMLINUX_MAIN := $(core-y) $(libs-y) $(drivers-y) $(net-y)
+ifdef CONFIG_XIP_KERNEL
+export KBUILD_LDS          := arch/$(SRCARCH)/kernel/vmlinux-xip.lds
+else
 export KBUILD_LDS          := arch/$(SRCARCH)/kernel/vmlinux.lds
+endif
 export LDFLAGS_vmlinux
 # used by scripts/pacmage/Makefile
 export KBUILD_ALLDIRS := $(sort $(filter-out arch/%,$(vmlinux-alldirs)) arch Documentation include samples scripts tools virt)
arch/arm/kernel/Makefile +5 -0
index af9e59b..d97fc2e 100644
@@ -3,6 +3,7 @@
 #
 
 CPPFLAGS_vmlinux.lds := -DTEXT_OFFSET=$(TEXT_OFFSET)
+CPPFLAGS_vmlinux-xip.lds := -DTEXT_OFFSET=$(TEXT_OFFSET)
 AFLAGS_head.o        := -DTEXT_OFFSET=$(TEXT_OFFSET)
 
 ifdef CONFIG_FUNCTION_TRACER
@@ -92,4 +93,8 @@ obj-y				+= psci-call.o
 obj-$(CONFIG_SMP)		+= psci_smp.o
 endif
 
+ifdef CONFIG_XIP_KERNEL
+extra-y := $(head-y) vmlinux-xip.lds
+else
 extra-y := $(head-y) vmlinux.lds
+endif

Chris

#35 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 11 lines · Chris Brandt
Quoted 5 lines

The old method with conditional placement in the linker script is broken beyond repair. There is simply way too much stuff these days to do this sanely. I think the only way to fix it properly is to have a separate linker script for XIP usage with proper (and obvious) section placement.

Step 1: Change the Makefiles to select an alternative linker script.

What do you think about doing the following?

That's fine. Obviously this is the first step when developing this, but logically this will have to be the last patch in the series when you're done.

Quoted 56 lines
Makefile +4 -0
index fd46821..956ad21 100644
@@ -898,7 +898,11 @@ libs-y		:= $(libs-y1) $(libs-y2)
 # Externally visible symbols (used by link-vmlinux.sh)
 export KBUILD_VMLINUX_INIT := $(head-y) $(init-y)
 export KBUILD_VMLINUX_MAIN := $(core-y) $(libs-y) $(drivers-y) $(net-y)
+ifdef CONFIG_XIP_KERNEL
+export KBUILD_LDS          := arch/$(SRCARCH)/kernel/vmlinux-xip.lds
+else
 export KBUILD_LDS          := arch/$(SRCARCH)/kernel/vmlinux.lds
+endif
 export LDFLAGS_vmlinux
 # used by scripts/pacmage/Makefile
 export KBUILD_ALLDIRS := $(sort $(filter-out arch/%,$(vmlinux-alldirs)) arch Documentation include samples scripts tools virt)
arch/arm/kernel/Makefile +5 -0
index af9e59b..d97fc2e 100644
@@ -3,6 +3,7 @@
 #
 
 CPPFLAGS_vmlinux.lds := -DTEXT_OFFSET=$(TEXT_OFFSET)
+CPPFLAGS_vmlinux-xip.lds := -DTEXT_OFFSET=$(TEXT_OFFSET)
 AFLAGS_head.o        := -DTEXT_OFFSET=$(TEXT_OFFSET)
 
 ifdef CONFIG_FUNCTION_TRACER
@@ -92,4 +93,8 @@ obj-y				+= psci-call.o
 obj-$(CONFIG_SMP)		+= psci_smp.o
 endif
 
+ifdef CONFIG_XIP_KERNEL
+extra-y := $(head-y) vmlinux-xip.lds
+else
 extra-y := $(head-y) vmlinux.lds
+endif

Chris

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#36 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

So after looking back in history, I think I now understand how/when the XIP stuff broke.

It looks like originally, the init section came first, followed by text and data.

This is why all the XIP code used _etext as its marker for the end of what needed to be mapped.
Like you said, the initialized data values get copied from ROM to RAM early in boot so it's not an issue if the data section (that comes after text) gets blown away when the full MMU setup is done.

However, after the init section was moved from the first section to almost the last, all the _etext references scatter through the kernel were never updated.

So basically, even though I've now got a separate linker script for XIP builds, it really hasn't changed much because the real issue all along was that the XIP_KERNEL references to _etext should have been changed to _data_loc.

With that said, I still think the correct way to fix this issue is to export _data_loc and use that as the end marker. (notice I changed my mind and now say _data_loc instead of _edata_loc because there's no reason to map the full data section anymore).

The other option, if we really wanted to keep the references to _etext as they are, is to simply move the one line " _etext = .;" to the bottom of the file....right before _data_loc is declared. But...that would be a bit misleading if you ask me.

It sounds like there would still be a request to split up the linker scripts in order to remove the 6 XIP_KERNEL references scattered throughout vmlinux.lds.S. However, it would still be nice to keep the two versions looking as close as they can to each other with only the obvious changes (like removing XIP_KERNEL from the non-XIP version, and removing SMP_ON_UP from the XIP version)

Thoughts?

Chris

#37 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 6 lines · Chris Brandt

It sounds like there would still be a request to split up the linker scripts in order to remove the 6 XIP_KERNEL references scattered throughout vmlinux.lds.S. However, it would still be nice to keep the two versions looking as close as they can to each other with only the obvious changes (like removing XIP_KERNEL from the non-XIP version, and removing SMP_ON_UP from the XIP version)

I don't think so. XIP and non-XIP are fundamentally looking different these days. Trying to keep them as similar as possible is rather an impediment to both cases and brings no real benefits.

Not only should .init.smpalt be gone from the XIP version, but it might be safer to put that into a special "should_be_empty" section and bail out with ASSERT() if that section size isn't zero. Same thing for pv_table. This way we'll be sure that nothing in the kernel config expects self-modified code to be present.

With XIP there is still an init section, but it is pointless to store .init.text stuff in there. That should instead stay in ROM and never be discarded (obviously). No need to page align it either at that point which should save some ROM space.

Things like .kprobes.text should be in ROM, but the entire executable in ROM should be surrounded by __kprobes_text_start and __kprobes_text_end to prevent any attempt to put kprobes there. This way, kprobes could still work for modules.

Etc. Etc.

There are simply too many things these days that require special considerations with XIP. And freeing the main script from XIP considerations will ease maintenance as well, more than the added maintenance from the duplicated parts.

Nicolas

maintainer
#38 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 29 lines · On Tuesday 17 November 2015 22:17:41 Nicolas Pitre wrote:
Quoted 6 lines · Chris Brandt

It sounds like there would still be a request to split up the linker scripts in order to remove the 6 XIP_KERNEL references scattered throughout vmlinux.lds.S. However, it would still be nice to keep the two versions looking as close as they can to each other with only the obvious changes (like removing XIP_KERNEL from the non-XIP version, and removing SMP_ON_UP from the XIP version)

I don't think so. XIP and non-XIP are fundamentally looking different these days. Trying to keep them as similar as possible is rather an impediment to both cases and brings no real benefits.

Not only should .init.smpalt be gone from the XIP version, but it might be safer to put that into a special "should_be_empty" section and bail out with ASSERT() if that section size isn't zero. Same thing for pv_table. This way we'll be sure that nothing in the kernel config expects self-modified code to be present.

With XIP there is still an init section, but it is pointless to store .init.text stuff in there. That should instead stay in ROM and never be discarded (obviously). No need to page align it either at that point which should save some ROM space.

Things like .kprobes.text should be in ROM, but the entire executable in ROM should be surrounded by __kprobes_text_start and __kprobes_text_end to prevent any attempt to put kprobes there. This way, kprobes could still work for modules.

Quoted 6 lines

Etc. Etc.

There are simply too many things these days that require special considerations with XIP. And freeing the main script from XIP considerations will ease maintenance as well, more than the added maintenance from the duplicated parts.

[Adding Uwe and Maxime to Cc]

I believe Uwe has been running XIP_KERNEL on efm32 for a while now, but he also mentioned that he needed some extra hacks in the linker script and elsewhere to get it to work.

It would be good to coordinate this, he may have found additional problems, or he can maybe help test the new approach on his hardware.

Arnd

#39 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 5 lines

[Adding Uwe and Maxime to Cc]

I believe Uwe has been running XIP_KERNEL on efm32 for a while now, but he also mentioned that he needed some extra hacks in the linker script and elsewhere to get it to work.

Yes, Maxime got the XIP fix for scripts/link-vmlinux.sh into the kernel ([PATCH v6 01/15] scripts: link-vmlinux: Don't pass page offset to kallsyms if XIP Kernel), which of course made one less file I have to patch.

Quoted 2 lines

It would be good to coordinate this, he may have found additional problems, or he can maybe help test the new approach on his hardware.

The part I'm using (RZ/A1H) is a Cortex A9 and the EFM32 is a Cortex M3, so we'd be able to test both MMU and non-MMU XIP systems.

Chris

#40 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

In reply to Nicolas Pitre
Quoted 3 lines

With XIP there is still an init section, [snip]
No need to page align it either at that point which should save some ROM space.

Good point.

Quoted 3 lines

Things like .kprobes.text should be in ROM, [snip]

Etc. Etc.

Here's the thing...I'm probably out of my depth on most of this. But, I'd be happy to post up a RFC patch and then start collecting the feedback on what can be removed. I can at least test it on a real XIP system.

Chris

#41 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 6 lines · Chris Brandt
Quoted 3 lines

With XIP there is still an init section, [snip]
No need to page align it either at that point which should save some ROM space.

Good point.

Quoted 8 lines
Quoted 3 lines

Things like .kprobes.text should be in ROM, [snip]

Etc. Etc.

Here's the thing...I'm probably out of my depth on most of this. But, I'd be happy to post up a RFC patch and then start collecting the feedback on what can be removed. I can at least test it on a real XIP system.

Sure. You should start by splitting the XIP stuff out to a separate script so functionally it is the same (broken) end result. And from there adding incremental fixes to the XIP script.

Nicolas

#42 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 3 lines

Sure. You should start by splitting the XIP stuff out to a separate script so functionally it is the same (broken) end result. And from there adding incremental fixes to the XIP script.

More or less that's what I'm doing now on my system. You can diff the 2 in order to see what I've removed.

But wait...you said 'the last step' was to add a 2nd linker file and change the Makefiles to use it.

Are you saying I should first submit a patch that simply adds the new file, but not hook it up to any Makefiles yet?
Then, once it has been updated a few times and is in better shape, I submit another patch that hooks it up with the Makefiles so I can then finally remove the XIP_KERNEL parts from the current vmlinux.lds.S?

Chris

#43 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 11 lines · Chris Brandt
Quoted 3 lines

Sure. You should start by splitting the XIP stuff out to a separate script so functionally it is the same (broken) end result. And from there adding incremental fixes to the XIP script.

More or less that's what I'm doing now on my system. You can diff the 2 in order to see what I've removed.

But wait...you said 'the last step' was to add a 2nd linker file and change the Makefiles to use it.

Are you saying I should first submit a patch that simply adds the new file, but not hook it up to any Makefiles yet?
Then, once it has been updated a few times and is in better shape, I submit another patch that hooks it up with the Makefiles so I can then finally remove the XIP_KERNEL parts from the current vmlinux.lds.S?

Well... I'm of the opinion now that the first patch should be the creation of the new script being a copy of the original with the XIP conditionals removed from both, plus the makefile changes. That should be easy to verify that nothing changed result wise, and then additional patches could bring XIP efinements on top without disturbing the non-XIP case.

Nicolas

#44 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 7 lines
Quoted 6 lines

Are you saying I should first submit a patch that simply adds the new file, but not hook it up to any Makefiles yet?
Then, once it has been updated a few times and is in better shape, I submit another patch that hooks it up with the Makefiles so I can then finally remove the XIP_KERNEL parts from the current vmlinux.lds.S?

Quoted 8 lines

Well... I'm of the opinion now that the first patch should be the creation of the new script being a copy of the original with the XIP conditionals removed from both, plus the makefile changes. That should be easy to verify that nothing changed result wise, and then additional patches could bring XIP efinements on top without disturbing the non-XIP case.

Nicolas

OK. I'm good with that. I'll submit a patch and see what happens.

Chris

#45 lore raw

PATCH v3 1/1 ARM: xip: Use correct symbol for end of ROM marker

In reply to Nicolas Pitre
Quoted 4 lines · Nicolas Pitre

The old method with conditional placement in the linker script is broken beyond repair. There is simply way too much stuff these days to do this sanely. I think the only way to fix it properly is to have a separate linker script for XIP usage with proper (and obvious) section placement.

Absolutely the right thing to do.

The XIP stuff has been broken and getting in the way for a long time now because of the preprocessor mess. I've made several changes to the linker script over the years, and each time I've ended up deciding to I'll do the best job I can for XIP, but I'm not going to spend too long making sure that it's correct - especially as there appeared to be no users of it. I wouldn't describe the current file as being "maintainable" in any way - It has devolved into a total mess, and I would like to see the XIP stuff gone from the non-XIP linker script
- and doing that should mean it's less likely for XIP to break.

I'd also like to see a lot more of the preprocessor gunk gone (or at least cleaned up) in that file too...

Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
#46 lore raw

PATCH v4 1/1 ARM: xip: Use correct symbol for end of ROM marker

In reply to Chris Brandt

For an XIP build, __data_loc, not _etext, represents the end of the binary image that needs to stay mapped into the MODULES_VADDR area. Years ago, data came before text in the memory map. However, now that the order is text/init/data, an XIP_KERNEL needs to map up to the __data_loc symbol in order to keep from cutting off parts of the kernel that are needed.
We only map up to the beginning of data because data has already been copied, so there's no reason to keep it around anymore.

This fixes the bug where you might lose the end of your kernel area after page table setup is complete.

---
v4:
* Switched from using _edata_loc __data_loc.
Made commit text more accurate.
v3:
* Removed sections.h from Kbuild
v2:
* Added change for MODULES_VADDR
* Moved extern to new file asm/sections.h

---
 arch/arm/include/asm/Kbuild     | 1 -
 arch/arm/include/asm/sections.h | 8 ++++++++
 arch/arm/kernel/module.c        | 2 +-
 arch/arm/mm/mmu.c               | 8 ++++++--
 4 files changed, 15 insertions(+), 4 deletions(-)
 create mode 100644 arch/arm/include/asm/sections.h
arch/arm/include/asm/Kbuild +0 -1
index 16da638..3f6616b 100644
@@ -23,7 +23,6 @@ generic-y += preempt.h
 generic-y += resource.h
 generic-y += rwsem.h
 generic-y += seccomp.h
-generic-y += sections.h
 generic-y += segment.h
 generic-y += sembuf.h
 generic-y += serial.h
arch/arm/include/asm/sections.h new file +8 -0
new file mode 100644
index 0000000..8d50cce
@@ -0,0 +1,8 @@
+#ifndef _ASM_ARM_SECTIONS_H
+#define _ASM_ARM_SECTIONS_H
+
+#include <asm-generic/sections.h>
+
+extern char __data_loc[];
+
+#endif	/* _ASM_ARM_SECTIONS_H */
arch/arm/kernel/module.c +1 -1
index efdddcb..b48e3bc 100644
@@ -34,7 +34,7 @@
  * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned on/off.
  */
 #undef MODULES_VADDR
-#define MODULES_VADDR	(((unsigned long)_etext + ~PMD_MASK) & PMD_MASK)
+#define MODULES_VADDR	(((unsigned long)__data_loc + ~PMD_MASK) & PMD_MASK)
 #endif
 
 #ifdef CONFIG_MMU
arch/arm/mm/mmu.c +6 -2
index 434d76f..53d7a88 100644
@@ -1253,7 +1253,7 @@ static inline void prepare_page_table(void)
 
 #ifdef CONFIG_XIP_KERNEL
 	/* The XIP kernel is mapped in the module area -- skip over it */
-	addr = ((unsigned long)_etext + PMD_SIZE - 1) & PMD_MASK;
+	addr = ((unsigned long)__data_loc + PMD_SIZE - 1) & PMD_MASK;
 #endif
 	for ( ; addr < PAGE_OFFSET; addr += PMD_SIZE)
 		pmd_clear(pmd_off_k(addr));
@@ -1335,7 +1335,7 @@ static void __init devicemaps_init(const struct machine_desc *mdesc)
 #ifdef CONFIG_XIP_KERNEL
 	map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
 	map.virtual = MODULES_VADDR;
-	map.length = ((unsigned long)_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
+	map.length = ((unsigned long)__data_loc - map.virtual + ~SECTION_MASK) & SECTION_MASK;
 	map.type = MT_ROM;
 	create_mapping(&map);
 #endif
@@ -1426,7 +1426,11 @@ static void __init kmap_init(void)
 static void __init map_lowmem(void)
 {
 	struct memblock_region *reg;
+#ifdef CONFIG_XIP_KERNEL
+	phys_addr_t kernel_x_start = round_down(__pa(_sdata), SECTION_SIZE);
+#else
 	phys_addr_t kernel_x_start = round_down(__pa(_stext), SECTION_SIZE);
+#endif
 	phys_addr_t kernel_x_end = round_up(__pa(__init_end), SECTION_SIZE);
 
 	/* Map all the lowmem memory banks. */
Signature · 1 line
1.9.1
#47 lore raw

PATCH v4 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 9 lines · Chris Brandt

For an XIP build, __data_loc, not _etext, represents the end of the binary image that needs to stay mapped into the MODULES_VADDR area. Years ago, data came before text in the memory map. However, now that the order is text/init/data, an XIP_KERNEL needs to map up to the __data_loc symbol in order to keep from cutting off parts of the kernel that are needed.
We only map up to the beginning of data because data has already been copied, so there's no reason to keep it around anymore.

Quoted 4 lines

This fixes the bug where you might lose the end of your kernel area after page table setup is complete.

The __data_loc symbol doesn't convey much to say with regards to XIP mapping. It would be better to dedicate another symbol with a proper name instead. Something like xiprom_start and xiprom_end perhaps. This way, someone looking at the linker script won't be fooled by a hidden meaning tied to _etext or __data_loc.

Quoted 95 lines

---
v4:
* Switched from using _edata_loc __data_loc.
Made commit text more accurate.
v3:
* Removed sections.h from Kbuild
v2:
* Added change for MODULES_VADDR
* Moved extern to new file asm/sections.h

---
 arch/arm/include/asm/Kbuild     | 1 -
 arch/arm/include/asm/sections.h | 8 ++++++++
 arch/arm/kernel/module.c        | 2 +-
 arch/arm/mm/mmu.c               | 8 ++++++--
 4 files changed, 15 insertions(+), 4 deletions(-)
 create mode 100644 arch/arm/include/asm/sections.h
arch/arm/include/asm/Kbuild +0 -1
index 16da638..3f6616b 100644
@@ -23,7 +23,6 @@ generic-y += preempt.h
 generic-y += resource.h
 generic-y += rwsem.h
 generic-y += seccomp.h
-generic-y += sections.h
 generic-y += segment.h
 generic-y += sembuf.h
 generic-y += serial.h
arch/arm/include/asm/sections.h new file +8 -0
new file mode 100644
index 0000000..8d50cce
@@ -0,0 +1,8 @@
+#ifndef _ASM_ARM_SECTIONS_H
+#define _ASM_ARM_SECTIONS_H
+
+#include <asm-generic/sections.h>
+
+extern char __data_loc[];
+
+#endif	/* _ASM_ARM_SECTIONS_H */
arch/arm/kernel/module.c +1 -1
index efdddcb..b48e3bc 100644
@@ -34,7 +34,7 @@
  * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned on/off.
  */
 #undef MODULES_VADDR
-#define MODULES_VADDR	(((unsigned long)_etext + ~PMD_MASK) & PMD_MASK)
+#define MODULES_VADDR	(((unsigned long)__data_loc + ~PMD_MASK) & PMD_MASK)
 #endif
 
 #ifdef CONFIG_MMU
arch/arm/mm/mmu.c +6 -2
index 434d76f..53d7a88 100644
@@ -1253,7 +1253,7 @@ static inline void prepare_page_table(void)
 
 #ifdef CONFIG_XIP_KERNEL
 	/* The XIP kernel is mapped in the module area -- skip over it */
-	addr = ((unsigned long)_etext + PMD_SIZE - 1) & PMD_MASK;
+	addr = ((unsigned long)__data_loc + PMD_SIZE - 1) & PMD_MASK;
 #endif
 	for ( ; addr < PAGE_OFFSET; addr += PMD_SIZE)
 		pmd_clear(pmd_off_k(addr));
@@ -1335,7 +1335,7 @@ static void __init devicemaps_init(const struct machine_desc *mdesc)
 #ifdef CONFIG_XIP_KERNEL
 	map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
 	map.virtual = MODULES_VADDR;
-	map.length = ((unsigned long)_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
+	map.length = ((unsigned long)__data_loc - map.virtual + ~SECTION_MASK) & SECTION_MASK;
 	map.type = MT_ROM;
 	create_mapping(&map);
 #endif
@@ -1426,7 +1426,11 @@ static void __init kmap_init(void)
 static void __init map_lowmem(void)
 {
 	struct memblock_region *reg;
+#ifdef CONFIG_XIP_KERNEL
+	phys_addr_t kernel_x_start = round_down(__pa(_sdata), SECTION_SIZE);
+#else
 	phys_addr_t kernel_x_start = round_down(__pa(_stext), SECTION_SIZE);
+#endif
 	phys_addr_t kernel_x_end = round_up(__pa(__init_end), SECTION_SIZE);
 
 	/* Map all the lowmem memory banks. */
Signature · 1 line
1.9.1
#48 lore raw

PATCH v4 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 5 lines · Nicolas Pitre

The __data_loc symbol doesn't convey much to say with regards to XIP mapping. It would be better to dedicate another symbol with a proper name instead. Something like xiprom_start and xiprom_end perhaps. This way, someone looking at the linker script won't be fooled by a hidden meaning tied to _etext or __data_loc.

That's fine by me.

However, I just saw the email "[PATCH] ARM: vmlinux.lds: assert that ROM and RAM don't overlap when XIP_KERNEL=y" sent in a little while ago.

So, do we first want to reconsider splitting out the XIP stuff into a separate linker script? Hence my proposal form November "[PATCH v2] ARM: xip: Move XIP linking to a separate file".

Or, should we keep piling more stuff in vmlinux.lds.S?

Chris

#49 lore raw

PATCH v4 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 15 lines · Chris Brandt
Quoted 5 lines · Nicolas Pitre

The __data_loc symbol doesn't convey much to say with regards to XIP mapping. It would be better to dedicate another symbol with a proper name instead. Something like xiprom_start and xiprom_end perhaps. This way, someone looking at the linker script won't be fooled by a hidden meaning tied to _etext or __data_loc.

That's fine by me.

However, I just saw the email "[PATCH] ARM: vmlinux.lds: assert that ROM and RAM don't overlap when XIP_KERNEL=y" sent in a little while ago.

Quoted 5 lines

So, do we first want to reconsider splitting out the XIP stuff into a separate linker script? Hence my proposal form November "[PATCH v2]
ARM: xip: Move XIP linking to a separate file".

Or, should we keep piling more stuff in vmlinux.lds.S?

I still think the split linker file is the way to go.

Nicolas

#50 lore raw

PATCH v4 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 5 lines · Nicolas Pitre

The __data_loc symbol doesn't convey much to say with regards to XIP mapping. It would be better to dedicate another symbol with a proper name instead. Something like xiprom_start and xiprom_end perhaps. This way, someone looking at the linker script won't be fooled by a hidden meaning tied to _etext or __data_loc.

In keeping with the convention that's already in the vmlinux.lds, I'm going to go with "_xiprom" and "_exiprom"

Quoted 3 lines
Quoted 1 line

Or, should we keep piling more stuff in vmlinux.lds.S?

I still think the split linker file is the way to go.

So...since I didn't hear any clapping and cheering, first I'd like to fix what I know is broken and get that in. Then go back to the splitting linker script idea.

Chris

#51 lore raw

PATCH v5 1/1 ARM: xip: Use correct symbol for end of ROM marker

In reply to Chris Brandt

For an XIP build, _etext does not represent the end of the binary image that needs to stay mapped into the MODULES_VADDR area. Years ago, data came before text in the memory map. However, now that the order is text/init/data, an XIP_KERNEL needs to map up to the data location in order to keep from cutting off parts of the kernel that are needed.
We only map up to the beginning of data because data has already been copied, so there's no reason to keep it around anymore.

This fixes the bug where you might lose the end of your kernel area after page table setup is complete.

---
v5:
* Switched from using __data_loc to _exiprom (new symbol) v4:
* Switched from using _edata_loc __data_loc.
* Made commit text more accurate.
v3:
* Removed sections.h from Kbuild
v2:
* Added change for MODULES_VADDR
* Moved extern to new file asm/sections.h

---
 arch/arm/include/asm/Kbuild     | 1 -
 arch/arm/include/asm/sections.h | 8 ++++++++
 arch/arm/kernel/module.c        | 2 +-
 arch/arm/kernel/vmlinux.lds.S   | 2 ++
 arch/arm/mm/mmu.c               | 8 ++++++--
 5 files changed, 17 insertions(+), 4 deletions(-)
 create mode 100644 arch/arm/include/asm/sections.h
arch/arm/include/asm/Kbuild +0 -1
index 16da638..3f6616b 100644
@@ -23,7 +23,6 @@ generic-y += preempt.h
 generic-y += resource.h
 generic-y += rwsem.h
 generic-y += seccomp.h
-generic-y += sections.h
 generic-y += segment.h
 generic-y += sembuf.h
 generic-y += serial.h
arch/arm/include/asm/sections.h new file +8 -0
new file mode 100644
index 0000000..803bbf2
@@ -0,0 +1,8 @@
+#ifndef _ASM_ARM_SECTIONS_H
+#define _ASM_ARM_SECTIONS_H
+
+#include <asm-generic/sections.h>
+
+extern char _exiprom[];
+
+#endif	/* _ASM_ARM_SECTIONS_H */
arch/arm/kernel/module.c +1 -1
index efdddcb..4f14b5c 100644
@@ -34,7 +34,7 @@
  * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned on/off.
  */
 #undef MODULES_VADDR
-#define MODULES_VADDR	(((unsigned long)_etext + ~PMD_MASK) & PMD_MASK)
+#define MODULES_VADDR	(((unsigned long)_exiprom + ~PMD_MASK) & PMD_MASK)
 #endif
 
 #ifdef CONFIG_MMU
arch/arm/kernel/vmlinux.lds.S +2 -0
index 8b60fde..d18e4e5 100644
@@ -86,6 +86,7 @@ SECTIONS
 
 #ifdef CONFIG_XIP_KERNEL
 	. = XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR);
+	_xiprom = .;			/* XIP ROM area to be mapped */
 #else
 	. = PAGE_OFFSET + TEXT_OFFSET;
 #endif
@@ -228,6 +229,7 @@ SECTIONS
 #endif
 
 #ifdef CONFIG_XIP_KERNEL
+	_exiprom = .;			/* End of XIP ROM area */
 	__data_loc = ALIGN(4);		/* location in binary */
 	. = PAGE_OFFSET + TEXT_OFFSET;
 #else
arch/arm/mm/mmu.c +6 -2
index 434d76f..e4b681a 100644
@@ -1253,7 +1253,7 @@ static inline void prepare_page_table(void)
 
 #ifdef CONFIG_XIP_KERNEL
 	/* The XIP kernel is mapped in the module area -- skip over it */
-	addr = ((unsigned long)_etext + PMD_SIZE - 1) & PMD_MASK;
+	addr = ((unsigned long)_exiprom + PMD_SIZE - 1) & PMD_MASK;
 #endif
 	for ( ; addr < PAGE_OFFSET; addr += PMD_SIZE)
 		pmd_clear(pmd_off_k(addr));
@@ -1335,7 +1335,7 @@ static void __init devicemaps_init(const struct machine_desc *mdesc)
 #ifdef CONFIG_XIP_KERNEL
 	map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
 	map.virtual = MODULES_VADDR;
-	map.length = ((unsigned long)_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
+	map.length = ((unsigned long)_exiprom - map.virtual + ~SECTION_MASK) & SECTION_MASK;
 	map.type = MT_ROM;
 	create_mapping(&map);
 #endif
@@ -1426,7 +1426,11 @@ static void __init kmap_init(void)
 static void __init map_lowmem(void)
 {
 	struct memblock_region *reg;
+#ifdef CONFIG_XIP_KERNEL
+	phys_addr_t kernel_x_start = round_down(__pa(_sdata), SECTION_SIZE);
+#else
 	phys_addr_t kernel_x_start = round_down(__pa(_stext), SECTION_SIZE);
+#endif
 	phys_addr_t kernel_x_end = round_up(__pa(__init_end), SECTION_SIZE);
 
 	/* Map all the lowmem memory banks. */
Signature · 1 line
1.9.1
#52 lore raw

PATCH v5 1/1 ARM: xip: Use correct symbol for end of ROM marker

Quoted 9 lines · Chris Brandt

For an XIP build, _etext does not represent the end of the binary image that needs to stay mapped into the MODULES_VADDR area. Years ago, data came before text in the memory map. However, now that the order is text/init/data, an XIP_KERNEL needs to map up to the data location in order to keep from cutting off parts of the kernel that are needed.
We only map up to the beginning of data because data has already been copied, so there's no reason to keep it around anymore.

Quoted 4 lines

This fixes the bug where you might lose the end of your kernel area after page table setup is complete.

Quoted 118 lines

---
v5:
* Switched from using __data_loc to _exiprom (new symbol) v4:
* Switched from using _edata_loc __data_loc.
* Made commit text more accurate.
v3:
* Removed sections.h from Kbuild
v2:
* Added change for MODULES_VADDR
* Moved extern to new file asm/sections.h

---
 arch/arm/include/asm/Kbuild     | 1 -
 arch/arm/include/asm/sections.h | 8 ++++++++
 arch/arm/kernel/module.c        | 2 +-
 arch/arm/kernel/vmlinux.lds.S   | 2 ++
 arch/arm/mm/mmu.c               | 8 ++++++--
 5 files changed, 17 insertions(+), 4 deletions(-)
 create mode 100644 arch/arm/include/asm/sections.h
arch/arm/include/asm/Kbuild +0 -1
index 16da638..3f6616b 100644
@@ -23,7 +23,6 @@ generic-y += preempt.h
 generic-y += resource.h
 generic-y += rwsem.h
 generic-y += seccomp.h
-generic-y += sections.h
 generic-y += segment.h
 generic-y += sembuf.h
 generic-y += serial.h
arch/arm/include/asm/sections.h new file +8 -0
new file mode 100644
index 0000000..803bbf2
@@ -0,0 +1,8 @@
+#ifndef _ASM_ARM_SECTIONS_H
+#define _ASM_ARM_SECTIONS_H
+
+#include <asm-generic/sections.h>
+
+extern char _exiprom[];
+
+#endif	/* _ASM_ARM_SECTIONS_H */
arch/arm/kernel/module.c +1 -1
index efdddcb..4f14b5c 100644
@@ -34,7 +34,7 @@
  * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned on/off.
  */
 #undef MODULES_VADDR
-#define MODULES_VADDR	(((unsigned long)_etext + ~PMD_MASK) & PMD_MASK)
+#define MODULES_VADDR	(((unsigned long)_exiprom + ~PMD_MASK) & PMD_MASK)
 #endif
 
 #ifdef CONFIG_MMU
arch/arm/kernel/vmlinux.lds.S +2 -0
index 8b60fde..d18e4e5 100644
@@ -86,6 +86,7 @@ SECTIONS
 
 #ifdef CONFIG_XIP_KERNEL
 	. = XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR);
+	_xiprom = .;			/* XIP ROM area to be mapped */
 #else
 	. = PAGE_OFFSET + TEXT_OFFSET;
 #endif
@@ -228,6 +229,7 @@ SECTIONS
 #endif
 
 #ifdef CONFIG_XIP_KERNEL
+	_exiprom = .;			/* End of XIP ROM area */
 	__data_loc = ALIGN(4);		/* location in binary */
 	. = PAGE_OFFSET + TEXT_OFFSET;
 #else
arch/arm/mm/mmu.c +6 -2
index 434d76f..e4b681a 100644
@@ -1253,7 +1253,7 @@ static inline void prepare_page_table(void)
 
 #ifdef CONFIG_XIP_KERNEL
 	/* The XIP kernel is mapped in the module area -- skip over it */
-	addr = ((unsigned long)_etext + PMD_SIZE - 1) & PMD_MASK;
+	addr = ((unsigned long)_exiprom + PMD_SIZE - 1) & PMD_MASK;
 #endif
 	for ( ; addr < PAGE_OFFSET; addr += PMD_SIZE)
 		pmd_clear(pmd_off_k(addr));
@@ -1335,7 +1335,7 @@ static void __init devicemaps_init(const struct machine_desc *mdesc)
 #ifdef CONFIG_XIP_KERNEL
 	map.pfn = __phys_to_pfn(CONFIG_XIP_PHYS_ADDR & SECTION_MASK);
 	map.virtual = MODULES_VADDR;
-	map.length = ((unsigned long)_etext - map.virtual + ~SECTION_MASK) & SECTION_MASK;
+	map.length = ((unsigned long)_exiprom - map.virtual + ~SECTION_MASK) & SECTION_MASK;
 	map.type = MT_ROM;
 	create_mapping(&map);
 #endif
@@ -1426,7 +1426,11 @@ static void __init kmap_init(void)
 static void __init map_lowmem(void)
 {
 	struct memblock_region *reg;
+#ifdef CONFIG_XIP_KERNEL
+	phys_addr_t kernel_x_start = round_down(__pa(_sdata), SECTION_SIZE);
+#else
 	phys_addr_t kernel_x_start = round_down(__pa(_stext), SECTION_SIZE);
+#endif
 	phys_addr_t kernel_x_end = round_up(__pa(__init_end), SECTION_SIZE);
 
 	/* Map all the lowmem memory banks. */
Signature · 1 line
1.9.1