Skip to contentSkip to conversation

x86: Change 4-level page table base address to low memory

PATCH New Bin Meng 5 participants 16 messages Jan 31, 2019 – Feb 12, 2019 x86

lore
Share & tools
lore

Bin Meng

author
#0 lore raw

PATCH 1/3 x86: Change 4-level page table base address to low memory

At present the 4-level page table base address for 64-bit U-Boot proper is assigned an address that conflicts with CONFIG_LOADADDR. Change it to an address within the low memory range instead.

Fixes crashes seen when 'dhcp' on QEMU x86_64 with "-net nic -net user,tftp=.,bootfile=u-boot".

---

 arch/x86/cpu/i386/cpu.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
arch/x86/cpu/i386/cpu.c +2 -4
index 208ef08..af42431 100644
@@ -462,6 +462,7 @@ int cpu_has_64bit(void)
 		has_long_mode();
 }
 
+#define PAGETABLE_BASE		0x80000
 #define PAGETABLE_SIZE		(6 * 4096)
 
 /**
@@ -523,10 +524,7 @@ int cpu_jump_to_64bit_uboot(ulong target)
 	uint32_t *pgtable;
 	func_t func;
 
-	/* TODO(sjg at chromium.org): Find a better place for this */
-	pgtable = (uint32_t *)0x1000000;
-	if (!pgtable)
-		return -ENOMEM;
+	pgtable = (uint32_t *)PAGETABLE_BASE;
 
 	build_pagetable(pgtable);
 
Signature · 1 line
2.7.4
author
#1 lore raw

PATCH 2/3 x86: Don't copy the cpu_call64() function to a hardcoded address

Before jumping to 64-bit U-Boot proper, SPL copies the cpu_call64() function to a hardcoded address 0x3000000. This can have potential conflicts with application usage. Switch the destination address to be allocated from the heap to avoid such risk.

---

 arch/x86/cpu/i386/call64.S |  4 ++++
 arch/x86/cpu/i386/cpu.c    | 11 ++++++++---
 2 files changed, 12 insertions(+), 3 deletions(-)
arch/x86/cpu/i386/call64.S +4 -0
index 8f86728..275063c 100644
@@ -79,6 +79,10 @@ lret_target:
 	mov	%eax, %eax		/* Clear bits 63:32 */
 	jmp	*%eax			/* Jump to the 64-bit target */
 
+.globl call64_stub_size
+call64_stub_size:
+	.long	. - cpu_call64
+
 	.data
 	.align	16
 	.globl	gdt64
arch/x86/cpu/i386/cpu.c +8 -3
index af42431..e4b5514 100644
@@ -523,18 +523,23 @@ int cpu_jump_to_64bit_uboot(ulong target)
 	typedef void (*func_t)(ulong pgtable, ulong setup_base, ulong target);
 	uint32_t *pgtable;
 	func_t func;
+	char *ptr;
 
 	pgtable = (uint32_t *)PAGETABLE_BASE;
 
 	build_pagetable(pgtable);
 
-	/* TODO(sjg at chromium.org): Find a better place for this */
-	char *ptr = (char *)0x3000000;
+	extern long call64_stub_size;
+	ptr = malloc(call64_stub_size);
+	if (!ptr) {
+		printf("Failed to allocate the cpu_call64 stub\n");
+		return -ENOMEM;
+	}
 	char *gdt = (char *)0x3100000;
 
 	extern char gdt64[];
 
-	memcpy(ptr, cpu_call64, 0x1000);
+	memcpy(ptr, cpu_call64, call64_stub_size);
 	memcpy(gdt, gdt64, 0x100);
 
 	/*
Signature · 1 line
2.7.4
#2 lore raw

PATCH 2/3 x86: Don't copy the cpu_call64() function to a hardcoded address

Quoted 6 lines · Bin Meng

Before jumping to 64-bit U-Boot proper, SPL copies the cpu_call64() function to a hardcoded address 0x3000000. This can have potential conflicts with application usage. Switch the destination address to be allocated from the heap to avoid such risk.

Quoted 6 lines

---

 arch/x86/cpu/i386/call64.S |  4 ++++
 arch/x86/cpu/i386/cpu.c    | 11 ++++++++---
 2 files changed, 12 insertions(+), 3 deletions(-)
author
#3 lore raw

PATCH 2/3 x86: Don't copy the cpu_call64() function to a hardcoded address

Quoted 8 lines · Simon Glass
Quoted 6 lines · Bin Meng

Before jumping to 64-bit U-Boot proper, SPL copies the cpu_call64() function to a hardcoded address 0x3000000. This can have potential conflicts with application usage. Switch the destination address to be allocated from the heap to avoid such risk.

Quoted 8 lines
Quoted 6 lines

---

 arch/x86/cpu/i386/call64.S |  4 ++++
 arch/x86/cpu/i386/cpu.c    | 11 ++++++++---
 2 files changed, 12 insertions(+), 3 deletions(-)

applied to u-boot-x86, thanks!

author
#4 lore raw

PATCH 3/3 x86: Use the existing GDT in the ROM for 64-bit U-Boot proper

In reply to Bin Meng

It is unnecessary to use a RAM version GDT for 64-bit U-Boot proper. In fact we can just use the ROM version directly, which not only eliminates the risk of being overwritten by application, but also removes the complexity of patching the cpu_call64().

---

 arch/x86/cpu/i386/cpu.c | 14 --------------
 1 file changed, 14 deletions(-)
arch/x86/cpu/i386/cpu.c +0 -14
index e4b5514..3bde44e 100644
@@ -535,23 +535,9 @@ int cpu_jump_to_64bit_uboot(ulong target)
 		printf("Failed to allocate the cpu_call64 stub\n");
 		return -ENOMEM;
 	}
-	char *gdt = (char *)0x3100000;
-
-	extern char gdt64[];
-
 	memcpy(ptr, cpu_call64, call64_stub_size);
-	memcpy(gdt, gdt64, 0x100);
 
-	/*
-	 * TODO(sjg at chromium.org): This manually inserts the pointers into
-	 * the code. Tidy this up to avoid this.
-	 */
 	func = (func_t)ptr;
-	ulong ofs = (ulong)cpu_call64 - (ulong)ptr;
-	*(ulong *)(ptr + 7) = (ulong)gdt;
-	*(ulong *)(ptr + 0xc) = (ulong)gdt + 2;
-	*(ulong *)(ptr + 0x13) = (ulong)gdt;
-	*(ulong *)(ptr + 0x117 - 0xd4) -= ofs;
 
 	/*
 	 * Copy U-Boot from ROM
Signature · 1 line
2.7.4
#5 lore raw

PATCH 3/3 x86: Use the existing GDT in the ROM for 64-bit U-Boot proper

Hi Bin,

Quoted 6 lines · Bin Meng

It is unnecessary to use a RAM version GDT for 64-bit U-Boot proper. In fact we can just use the ROM version directly, which not only eliminates the risk of being overwritten by application, but also removes the complexity of patching the cpu_call64().

Quoted 6 lines

---

 arch/x86/cpu/i386/cpu.c | 14 --------------
 1 file changed, 14 deletions(-)

I am not sure what issue I had with this, but I'm pleased that this works.

author
#6 lore raw

PATCH 3/3 x86: Use the existing GDT in the ROM for 64-bit U-Boot proper

Quoted 12 lines · Simon Glass

Hi Bin,

Quoted 8 lines · Bin Meng

It is unnecessary to use a RAM version GDT for 64-bit U-Boot proper. In fact we can just use the ROM version directly, which not only eliminates the risk of being overwritten by application, but also removes the complexity of patching the cpu_call64().

Quoted 8 lines
Quoted 4 lines

---

 arch/x86/cpu/i386/cpu.c | 14 --------------
 1 file changed, 14 deletions(-)

I am not sure what issue I had with this, but I'm pleased that this works.

applied to u-boot-x86, thanks!

#7 lore raw

PATCH 1/3 x86: Change 4-level page table base address to low memory

In reply to Bin Meng
Quoted 9 lines · Bin Meng

At present the 4-level page table base address for 64-bit U-Boot proper is assigned an address that conflicts with CONFIG_LOADADDR. Change it to an address within the low memory range instead.

Fixes crashes seen when 'dhcp' on QEMU x86_64 with "-net nic -net user,tftp=.,bootfile=u-boot".

Thanks for addressing this. It fixes the problem reported in

x86: #define CONFIG_LOADADDR 0x1100000
https://lists.denx.de/pipermail/u-boot/2019-January/356108.html

I have set aforementioned patch to superseded.

Tested on qemu-x86_64.

#8 lore raw

PATCH 1/3 x86: Change 4-level page table base address to low memory

In reply to Bin Meng
Quoted 7 lines

Am 31.01.2019 um 17:22 schrieb Bin Meng <[email protected]>:

At present the 4-level page table base address for 64-bit U-Boot proper is assigned an address that conflicts with CONFIG_LOADADDR. Change it to an address within the low memory range instead.

Can't you dynamically allocate the PT too?

Alex

Quoted 38 lines

Fixes crashes seen when 'dhcp' on QEMU x86_64 with "-net nic -net user,tftp=.,bootfile=u-boot".

---

arch/x86/cpu/i386/cpu.c | 6 ++----

1 file changed, 2 insertions(+), 4 deletions(-)
arch/x86/cpu/i386/cpu.c +2 -4
index 208ef08..af42431 100644
@@ -462,6 +462,7 @@ int cpu_has_64bit(void)
       has_long_mode();
 }
 
+#define PAGETABLE_BASE        0x80000
 #define PAGETABLE_SIZE        (6 * 4096)
 
 /**
@@ -523,10 +524,7 @@ int cpu_jump_to_64bit_uboot(ulong target)
   uint32_t *pgtable;
   func_t func;
 
-    /* TODO(sjg at chromium.org): Find a better place for this */
-    pgtable = (uint32_t *)0x1000000;
-    if (!pgtable)
-        return -ENOMEM;
+    pgtable = (uint32_t *)PAGETABLE_BASE;
 
   build_pagetable(pgtable);
 
Signature · 1 line
2.7.4
author
#9 lore raw

PATCH 1/3 x86: Change 4-level page table base address to low memory

Hi Alex,

Quoted 11 lines · Alexander Graf
Quoted 8 lines

Am 31.01.2019 um 17:22 schrieb Bin Meng <[email protected]>:

At present the 4-level page table base address for 64-bit U-Boot proper is assigned an address that conflicts with CONFIG_LOADADDR. Change it to an address within the low memory range instead.

Can't you dynamically allocate the PT too?

The dynamically allocated PT only makes sense when in SPL. It then becomes an arbitrary address again when entering in the 64-bit proper.

Regards,
Bin

#10 lore raw

PATCH 1/3 x86: Change 4-level page table base address to low memory

Quoted 16 lines

Am 01.02.2019 um 00:40 schrieb Bin Meng <[email protected]>:

Hi Alex,

Quoted 10 lines · Alexander Graf
Quoted 8 lines

Am 31.01.2019 um 17:22 schrieb Bin Meng <[email protected]>:

At present the 4-level page table base address for 64-bit U-Boot proper is assigned an address that conflicts with CONFIG_LOADADDR. Change it to an address within the low memory range instead.

Quoted 5 lines
Quoted 2 lines

Can't you dynamically allocate the PT too?

The dynamically allocated PT only makes sense when in SPL. It then becomes an arbitrary address again when entering in the 64-bit proper.

I'm not sure I follow? On aarch64, we allocate every level dynamically. I feel like I'm missing a piece of the puzzle here :)

Alex

Quoted 3 lines

Regards,
Bin

author
#11 lore raw

PATCH 1/3 x86: Change 4-level page table base address to low memory

Hi Alex,

Quoted 17 lines · Alexander Graf
Quoted 17 lines

Am 01.02.2019 um 00:40 schrieb Bin Meng <[email protected]>:

Hi Alex,

Quoted 10 lines · Alexander Graf
Quoted 8 lines

Am 31.01.2019 um 17:22 schrieb Bin Meng <[email protected]>:

At present the 4-level page table base address for 64-bit U-Boot proper is assigned an address that conflicts with CONFIG_LOADADDR. Change it to an address within the low memory range instead.

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

Can't you dynamically allocate the PT too?

The dynamically allocated PT only makes sense when in SPL. It then becomes an arbitrary address again when entering in the 64-bit proper.

I'm not sure I follow? On aarch64, we allocate every level dynamically. I feel like I'm missing a piece of the puzzle here :)

The current x86 implementation is the SPL allocates the page table for the 64-bit U-Boot. We can certainly change the implementation but I would leave that for future changes.

Regards,
Bin

#12 lore raw

PATCH 1/3 x86: Change 4-level page table base address to low memory

Quoted 30 lines

Am 01.02.2019 um 08:29 schrieb Bin Meng <[email protected]>:

Hi Alex,

Quoted 24 lines · Alexander Graf
Quoted 22 lines

Am 01.02.2019 um 00:40 schrieb Bin Meng <[email protected]>:

Hi Alex,

Quoted 12 lines · Alexander Graf
Quoted 8 lines

Am 31.01.2019 um 17:22 schrieb Bin Meng <[email protected]>:

At present the 4-level page table base address for 64-bit U-Boot proper is assigned an address that conflicts with CONFIG_LOADADDR. Change it to an address within the low memory range instead.

Can't you dynamically allocate the PT too?

The dynamically allocated PT only makes sense when in SPL. It then becomes an arbitrary address again when entering in the 64-bit proper.

Quoted 6 lines
Quoted 2 lines

I'm not sure I follow? On aarch64, we allocate every level dynamically. I feel like I'm missing a piece of the puzzle here :)

The current x86 implementation is the SPL allocates the page table for the 64-bit U-Boot. We can certainly change the implementation but I would leave that for future changes.

I see. Works for me :)

Alex

Quoted 3 lines

Regards,
Bin

#13 lore raw

PATCH 1/3 x86: Change 4-level page table base address to low memory

In reply to Bin Meng
Quoted 8 lines · Bin Meng

At present the 4-level page table base address for 64-bit U-Boot proper is assigned an address that conflicts with CONFIG_LOADADDR. Change it to an address within the low memory range instead.

Fixes crashes seen when 'dhcp' on QEMU x86_64 with "-net nic -net user,tftp=.,bootfile=u-boot".

Quoted 6 lines

---

 arch/x86/cpu/i386/cpu.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
#14 lore raw

PATCH 1/3 x86: Change 4-level page table base address to low memory

Quoted 13 lines · Simon Glass
Quoted 12 lines · Bin Meng

At present the 4-level page table base address for 64-bit U-Boot proper is assigned an address that conflicts with CONFIG_LOADADDR. Change it to an address within the low memory range instead.

Fixes crashes seen when 'dhcp' on QEMU x86_64 with "-net nic -net user,tftp=.,bootfile=u-boot".

---

Quoted 9 lines
Quoted 2 lines
 arch/x86/cpu/i386/cpu.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

Hello Bin,

this patch is needed together with

test/py: use default load address for tftp
https://lists.denx.de/pipermail/u-boot/2019-January/356265.html

to get the EFI patch queue merged.

Should Tom pick the patch series?

Best regards

Heinrich

author
#15 lore raw

PATCH 1/3 x86: Change 4-level page table base address to low memory

Hi Heinrich,

Quoted 27 lines · Heinrich Schuchardt
Quoted 13 lines · Simon Glass
Quoted 12 lines · Bin Meng

At present the 4-level page table base address for 64-bit U-Boot proper is assigned an address that conflicts with CONFIG_LOADADDR. Change it to an address within the low memory range instead.

Fixes crashes seen when 'dhcp' on QEMU x86_64 with "-net nic -net user,tftp=.,bootfile=u-boot".

---

Quoted 9 lines
Quoted 2 lines
 arch/x86/cpu/i386/cpu.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

Hello Bin,

Quoted 9 lines

this patch is needed together with

test/py: use default load address for tftp
https://lists.denx.de/pipermail/u-boot/2019-January/356265.html

to get the EFI patch queue merged.

Should Tom pick the patch series?

Sorry I just came back from Chinese new year holiday vacation. I will send PR to Tom ASAP.

applied to u-boot-x86, thanks!

Regards,
Bin