Skip to contentSkip to conversation

ARMv8: PSCI: Fix PSCI_TABLE relocation issue

PATCH RESEND Superseded Lars Povlsen 3 participants 6 messages Apr 4, 2019 – Apr 24, 2019 Arm (32-bit)

lore
Share & tools
lore
#0 lore raw

PATCH RESEND 1/1 ARMv8: PSCI: Fix PSCI_TABLE relocation issue

This fixes relaction isses with the PSCI_TABLE entries in the psci_32_table and psci_64_table.

When using 32-bit adress pointers relocation was not being applied to the tables, causing PSCI handlers to point to the un-relocated code area. By using 64-bit data relocation is properly applied. The handlers are thus in the "secure data" area, which is protected by /memreserve/ in the FDT.

---
 arch/arm/cpu/armv8/psci.S | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)
arch/arm/cpu/armv8/psci.S +6 -7
index 358df8fee9..b4568cf053 100644
@@ -19,8 +19,8 @@
 
 /* PSCI function and ID table definition*/
 #define PSCI_TABLE(__id, __fn) \
-	.word __id; \
-	.word __fn
+	.quad __id; \
+	.quad __fn
 
 .pushsection ._secure.text, "ax"
 
@@ -132,16 +132,15 @@ PSCI_TABLE(0, 0)
 /* Caller must put PSCI function-ID table base in x9 */
 handle_psci:
 	psci_enter
-1:	ldr x10, [x9]			/* Load PSCI function table */
-	ubfx x11, x10, #32, #32
-	ubfx x10, x10, #0, #32
+1:	ldr	x10, [x9]		/* Load PSCI function table */
 	cbz	x10, 3f			/* If reach the end, bail out */
 	cmp	x10, x0
 	b.eq	2f			/* PSCI function found */
-	add x9, x9, #8			/* If not match, try next entry */
+	add x9, x9, #16			/* If not match, try next entry */
 	b	1b
 
-2:	blr	x11			/* Call PSCI function */
+2:	ldr	x11, [x9, #8]		/* Load PSCI function */
+	blr	x11			/* Call PSCI function */
 	psci_return
 
 3:	mov	x0, #ARM_PSCI_RET_NI
Signature · 1 line
2.21.0
#1 lore raw

PATCH RESEND 1/1 ARMv8: PSCI: Fix PSCI_TABLE relocation issue

Quoted 55 lines · Lars Povlsen

This fixes relaction isses with the PSCI_TABLE entries in the psci_32_table and psci_64_table.

When using 32-bit adress pointers relocation was not being applied to the tables, causing PSCI handlers to point to the un-relocated code area. By using 64-bit data relocation is properly applied. The handlers are thus in the "secure data" area, which is protected by /memreserve/ in the FDT.

---
 arch/arm/cpu/armv8/psci.S | 13 ++++++-------

 1 file changed, 6 insertions(+), 7 deletions(-)
arch/arm/cpu/armv8/psci.S +0 -0
index 358df8fee9..b4568cf053 100644
@@ -19,8 +19,8 @@


 /* PSCI function and ID table definition*/
 #define PSCI_TABLE(__id, __fn) \

arch/arm/cpu/armv8/psci.S +2 -2
-	.word __id; \
-	.word __fn
+	.quad __id; \
+	.quad __fn


 .pushsection ._secure.text, "ax"

arch/arm/cpu/armv8/psci.S +0 -0
@@ -132,16 +132,15 @@ PSCI_TABLE(0, 0)
  /* Caller must put PSCI function-ID table base in x9 */

 handle_psci:
  psci_enter
-1: ldr x10, [x9] /* Load PSCI function
table */

arch/arm/cpu/armv8/psci.S +1 -2
-	ubfx x11, x10, #32, #32
-	ubfx x10, x10, #0, #32
+1:	ldr	x10, [x9]		/* Load PSCI function

table */
  cbz x10, 3f /* If reach the
end, bail out */
  cmp x10, x0
  b.eq 2f /* PSCI function found
*/

arch/arm/cpu/armv8/psci.S +0 -1
-	add x9, x9, #8			/* If not match, try

next entry */

arch/arm/cpu/armv8/psci.S +1 -0
+	add x9, x9, #16			/* If not match, try

next entry */
  b 1b

-2: blr x11 /* Call PSCI
function */
+2: ldr x11, [x9, #8] /* Load PSCI
function */

arch/arm/cpu/armv8/psci.S +1 -0
+	blr	x11			/* Call PSCI function
Quoted 4 lines

*/
  psci_return

 3: mov x0, #ARM_PSCI_RET_NI

Hmmm...I presumed you made these changes because your relocation address for "secure" section is beyond 32bits (4GB). How your page table for MMU is being setup ? Why do you need such large address space (beyond 4GB) ?

#2 lore raw

PATCH RESEND 1/1 ARMv8: PSCI: Fix PSCI_TABLE relocation issue

Quoted 66 lines

From: Ang, Chee Hong <[email protected]>
Sent: Monday, April 8, 2019 05:10
To: Lars Povlsen - M31675 <[email protected]>;
trini at konsulko.com; u-boot at lists.denx.de; macro.wave.z at gmail.com; albert.u.boot at aribaud.net; york.sun at nxp.com
Subject: Re: [RESEND PATCH] ARMv8: PSCI: Fix PSCI_TABLE relocation issue

Quoted 58 lines · Lars Povlsen

This fixes relaction isses with the PSCI_TABLE entries in the psci_32_table and psci_64_table.

When using 32-bit adress pointers relocation was not being applied to the tables, causing PSCI handlers to point to the un-relocated code area. By using 64-bit data relocation is properly applied. The handlers are thus in the "secure data" area, which is protected by /memreserve/ in the FDT.

---
 arch/arm/cpu/armv8/psci.S | 13 ++++++-------

 1 file changed, 6 insertions(+), 7 deletions(-)
arch/arm/cpu/armv8/psci.S +2 -2
index 358df8fee9..b4568cf053 100644
@@ -19,8 +19,8 @@
 
  /* PSCI function and ID table definition*/
  #define PSCI_TABLE(__id, __fn) \
-	.word __id; \
-	.word __fn
+	.quad __id; \
+	.quad __fn
 

 .pushsection ._secure.text, "ax"

arch/arm/cpu/armv8/psci.S +0 -0
@@ -132,16 +132,15 @@ PSCI_TABLE(0, 0)
  /* Caller must put PSCI function-ID table base in x9 */

 handle_psci:
  psci_enter
-1: ldr x10, [x9] /* Load PSCI function
table */

arch/arm/cpu/armv8/psci.S +1 -2
-	ubfx x11, x10, #32, #32
-	ubfx x10, x10, #0, #32
+1:	ldr	x10, [x9]		/* Load PSCI function

table */
  cbz x10, 3f /* If reach the
end, bail out */
  cmp x10, x0
  b.eq 2f /* PSCI function found
*/

arch/arm/cpu/armv8/psci.S +0 -1
-	add x9, x9, #8			/* If not match, try

next entry */

arch/arm/cpu/armv8/psci.S +1 -0
+	add x9, x9, #16			/* If not match, try

next entry */
  b 1b

-2: blr x11 /* Call PSCI
function */
+2: ldr x11, [x9, #8] /* Load PSCI
function */

arch/arm/cpu/armv8/psci.S +1 -0
+	blr	x11			/* Call PSCI function

*/
  psci_return

Quoted 6 lines
Quoted 1 line

 3: mov x0, #ARM_PSCI_RET_NI

Hmmm...I presumed you made these changes because your relocation address for "secure" section is beyond 32bits (4GB). How your page table for MMU is being setup ? Why do you need such large address space (beyond 4GB) ?

No, as I tried to describe in the log message, the relocation was simply not being applied. Changing the offsets to 64-bit fixed this. The .text base was 0x80000000 and the relocation was done in a 2Gb window (so somewhere ~ 0xfff.....)

Never the less, I assume a 64-bit target should not be constrained to 32-bit addresses?

When debugging the code, I noticed my debugger was able to match the original symbols even though relocation was done. That’s how I became aware. I could see that the vector table and the first level code *was* relocated, but the individual handlers not.

---Lars

#3 lore raw

PATCH RESEND 1/1 ARMv8: PSCI: Fix PSCI_TABLE relocation issue

Quoted 96 lines · Lars.Povlsen at microchip.com
Quoted 80 lines

From: Ang, Chee Hong <[email protected]>
Sent: Monday, April 8, 2019 05:10
To: Lars Povlsen - M31675 <[email protected]>;
trini at konsulko.com; u-boot at lists.denx.de; macro.wave.z at gmail.com; albert.u.boot at aribaud.net; york.sun at nxp.com
Subject: Re: [RESEND PATCH] ARMv8: PSCI: Fix PSCI_TABLE relocation issue

Quoted 70 lines · Lars Povlsen

This fixes relaction isses with the PSCI_TABLE entries in the psci_32_table and psci_64_table.

When using 32-bit adress pointers relocation was not being applied to
the tables, causing PSCI handlers to point to the un-relocated code
area. By using 64-bit data relocation is properly applied. The handlers are thus in the "secure data" area, which is protected by
/memreserve/ in the FDT.

---
 arch/arm/cpu/armv8/psci.S | 13 ++++++-------

 1 file changed, 6 insertions(+), 7 deletions(-)
arch/arm/cpu/armv8/psci.S +2 -2
index 358df8fee9..b4568cf053 100644
@@ -19,8 +19,8 @@
 
  /* PSCI function and ID table definition*/
  #define PSCI_TABLE(__id, __fn) \
-	.word __id; \
-	.word __fn
+	.quad __id; \
+	.quad __fn
 

 .pushsection ._secure.text, "ax"

arch/arm/cpu/armv8/psci.S +0 -0
@@ -132,16 +132,15 @@ PSCI_TABLE(0, 0)
  /* Caller must put PSCI function-ID table base in x9 */

 handle_psci:
  psci_enter
-1: ldr x10, [x9] /* Load PSCI
function
table */

arch/arm/cpu/armv8/psci.S +1 -2
-	ubfx x11, x10, #32, #32
-	ubfx x10, x10, #0, #32
+1:	ldr	x10, [x9]		/* Load PSCI

function
table */
  cbz x10, 3f /* If reach
the
end, bail out */
  cmp x10, x0
  b.eq 2f /* PSCI function
found
*/

arch/arm/cpu/armv8/psci.S +0 -1
-	add x9, x9, #8			/* If not match,

try
next entry */

arch/arm/cpu/armv8/psci.S +1 -0
+	add x9, x9, #16			/* If not match,

try
next entry */
  b 1b

-2: blr x11 /* Call PSCI
function */
+2: ldr x11, [x9, #8] /* Load PSCI
function */

arch/arm/cpu/armv8/psci.S +1 -0
+	blr	x11			/* Call PSCI

function
*/
  psci_return

Quoted 6 lines
Quoted 1 line

 3: mov x0, #ARM_PSCI_RET_NI

Hmmm...I presumed you made these changes because your relocation address for "secure" section is beyond 32bits (4GB). How your page table for MMU is being setup ? Why do you need such large address space
(beyond 4GB) ?

No, as I tried to describe in the log message, the relocation was simply
not being applied. Changing the offsets to 64-bit fixed this. The .text base
was 0x80000000 and the relocation was done in a 2Gb window (so somewhere ~ 0xfff.....)

Never the less, I assume a 64-bit target should not be constrained to 32-bit addresses?

Quoted 7 lines

When debugging the code, I noticed my debugger was able to match the original symbols
even though relocation was done. That’s how I became aware. I could see that the vector table
and the first level code *was* relocated, but the individual handlers not.
---Lars

CONFIG_ARMV8_SECURE_BASE defines the target relocation address for the PSCI code and the table. I just checked the PSCI handler addresses in the PSCI table entry (_psci_32_table & _psci_64_table) of my uboot, they point to the relocated address without using the 64bits size.

#4 lore raw

PATCH RESEND 1/1 ARMv8: PSCI: Fix PSCI_TABLE relocation issue

Quoted 112 lines

From: Ang, Chee Hong <[email protected]>
Sent: Monday, April 8, 2019 11:00
To: Lars Povlsen - M31675 <[email protected]>;
trini at konsulko.com; u-boot at lists.denx.de; macro.wave.z at gmail.com; albert.u.boot at aribaud.net; york.sun at nxp.com
Subject: Re: [RESEND PATCH] ARMv8: PSCI: Fix PSCI_TABLE relocation issue

Quoted 96 lines · Lars.Povlsen at microchip.com
Quoted 80 lines

From: Ang, Chee Hong <[email protected]>
Sent: Monday, April 8, 2019 05:10
To: Lars Povlsen - M31675 <[email protected]>;
trini at konsulko.com; u-boot at lists.denx.de; macro.wave.z at gmail.com; albert.u.boot at aribaud.net; york.sun at nxp.com
Subject: Re: [RESEND PATCH] ARMv8: PSCI: Fix PSCI_TABLE relocation issue

Quoted 70 lines · Lars Povlsen

This fixes relaction isses with the PSCI_TABLE entries in the psci_32_table and psci_64_table.

When using 32-bit adress pointers relocation was not being applied to
the tables, causing PSCI handlers to point to the un-relocated code
area. By using 64-bit data relocation is properly applied. The handlers are thus in the "secure data" area, which is protected by
/memreserve/ in the FDT.

---
 arch/arm/cpu/armv8/psci.S | 13 ++++++-------

 1 file changed, 6 insertions(+), 7 deletions(-)
arch/arm/cpu/armv8/psci.S +2 -2
index 358df8fee9..b4568cf053 100644
@@ -19,8 +19,8 @@
 
  /* PSCI function and ID table definition*/
  #define PSCI_TABLE(__id, __fn) \
-	.word __id; \
-	.word __fn
+	.quad __id; \
+	.quad __fn
 

 .pushsection ._secure.text, "ax"

arch/arm/cpu/armv8/psci.S +0 -0
@@ -132,16 +132,15 @@ PSCI_TABLE(0, 0)
  /* Caller must put PSCI function-ID table base in x9 */

 handle_psci:
  psci_enter
-1: ldr x10, [x9] /* Load PSCI
function
table */

arch/arm/cpu/armv8/psci.S +1 -2
-	ubfx x11, x10, #32, #32
-	ubfx x10, x10, #0, #32
+1:	ldr	x10, [x9]		/* Load PSCI

function
table */
  cbz x10, 3f /* If reach
the
end, bail out */
  cmp x10, x0
  b.eq 2f /* PSCI function
found
*/

arch/arm/cpu/armv8/psci.S +0 -1
-	add x9, x9, #8			/* If not match,

try
next entry */

arch/arm/cpu/armv8/psci.S +1 -0
+	add x9, x9, #16			/* If not match,

try
next entry */
  b 1b

-2: blr x11 /* Call PSCI
function */
+2: ldr x11, [x9, #8] /* Load PSCI
function */

arch/arm/cpu/armv8/psci.S +1 -0
+	blr	x11			/* Call PSCI

function
*/
  psci_return

Quoted 6 lines
Quoted 1 line

 3: mov x0, #ARM_PSCI_RET_NI

Hmmm...I presumed you made these changes because your relocation address for "secure" section is beyond 32bits (4GB). How your page table for MMU is being setup ? Why do you need such large address space
(beyond 4GB) ?

No, as I tried to describe in the log message, the relocation was simply
not being applied. Changing the offsets to 64-bit fixed this. The .text base
was 0x80000000 and the relocation was done in a 2Gb window (so somewhere ~ 0xfff.....)

Never the less, I assume a 64-bit target should not be constrained to 32-bit addresses?

Quoted 7 lines

When debugging the code, I noticed my debugger was able to match the original symbols
even though relocation was done. That’s how I became aware. I could see that the vector table
and the first level code *was* relocated, but the individual handlers not.
---Lars

Quoted 4 lines

CONFIG_ARMV8_SECURE_BASE defines the target relocation address for the PSCI code and the table. I just checked the PSCI handler addresses in the PSCI table entry (_psci_32_table & _psci_64_table) of my uboot, they point to the relocated address without using the 64bits size.

CONFIG_ARMV8_SECURE_BASE depend on SYS_HAS_ARMV8_SECURE_BASE, which is not set on this platform. Might this be the source of the problem? In this case, I would assume that the "secure data" just stays with the relocated U-boot, and gets carved out with a fdt_add_mem_rsv when booting Linux.

For what it matters, I have changed to using the spintable PSCI variant, so I'm not really depending on this patch. But it looks like there is a problem, at least in the configuration I used.

---Lars

#5 lore raw

PATCH RESEND 1/1 ARMv8: PSCI: Fix PSCI_TABLE relocation issue

In reply to Lars Povlsen
Quoted 10 lines · Lars Povlsen

This fixes relaction isses with the PSCI_TABLE entries in the psci_32_table and psci_64_table.

When using 32-bit adress pointers relocation was not being applied to the tables, causing PSCI handlers to point to the un-relocated code area. By using 64-bit data relocation is properly applied. The handlers are thus in the "secure data" area, which is protected by /memreserve/ in the FDT.

Applied to u-boot/master, thanks!

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20190424/bce10c86/attachment.sig>