xen/arm: Using unsigned long for arm64 MPIDR mask
PATCH v2 1/1 xen/arm: Using unsigned long for arm64 MPIDR mask
Currently, Xen is using UINT32 for MPIDR mask to retrieve affinity[0,1,2,3] values for MPIDR_EL1 register. The value of MPIDR_EL1 is 64-bit unsigned long. The 32-bit unsinged integer will do unsigned extend while doing some operations with 64-bit unsigned integer. This can lead to unexpected result in some use cases.
For example, in gicv3_send_sgi_list of GICv3 driver:uint64_t cluster_id = cpu_logical_map(cpu) & ~MPIDR_AFF0_MASK;
When MPIDR_AFF0_MASK is 0xFFU, compiler output:
f7c: 92785c16 and x22, x0, #0xffffff00
When MPIDR_AFF0_MASK is 0xFFUL, compiler output:
f88: 9278dc75 and x21, x3, #0xffffffffffffff00If cpu_logical_map(cpu) = 0x100000000UL and MPIDR_AFF0_MASK is 0xFFU, the cluster_id returns 0. But the expected value should be 0x100000000.
So, in this patch, we force aarch64 to use unsigned long as MPIDR mask to avoid such unexpected results.
---
v1 -> v2:
1. Remove inaccurate descriptions
2. Update example description
--- xen/include/asm-arm/processor.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
xen/include/asm-arm/processor.h +3 -3
@@ -75,11 +75,11 @@/* MPIDR Multiprocessor Affinity Register */ #define _MPIDR_UP (30) -#define MPIDR_UP (_AC(1,U) << _MPIDR_UP) +#define MPIDR_UP (_AC(1,UL) << _MPIDR_UP)#define _MPIDR_SMP (31) -#define MPIDR_SMP (_AC(1,U) << _MPIDR_SMP) +#define MPIDR_SMP (_AC(1,UL) << _MPIDR_SMP)#define MPIDR_AFF0_SHIFT (0) -#define MPIDR_AFF0_MASK (_AC(0xff,U) << MPIDR_AFF0_SHIFT) +#define MPIDR_AFF0_MASK (_AC(0xff,UL) << MPIDR_AFF0_SHIFT)#ifdef CONFIG_ARM_64 #define MPIDR_HWID_MASK _AC(0xff00ffffff,UL)#else
--
2.25.1
Hi Wei,
How about the following title:
"xen/arm: Don't ignore the affinity level 3 in the MPIDR"
Quoted 3 lines · Wei Chen
Currently, Xen is using UINT32 for MPIDR mask to retrieve affinity[0,1,2,3] values for
MPIDR_EL1register. The value ofMPIDR_EL1is 64-bit unsigned long. > The 32-bit unsinged
s/unsinged/unsigned/
Quoted 12 lines
integer will do unsigned extend while doing some operations with 64-bit unsigned integer. This can lead to unexpected result in some use cases.
For example, in
gicv3_send_sgi_listof GICv3 driver:uint64_tcluster_id=cpu_logical_map(cpu)& ~MPIDR_AFF0_MASK;When
MPIDR_AFF0_MASKis0xFFU, compiler output:f7c: 92785c16 and x22, x0, #0xffffff00 When MPIDR_AFF0_MASK is 0xFFUL, compiler output: f88: 9278dc75 and x21, x3, #0xffffffffffffff00
Quoted 6 lines
If
cpu_logical_map(cpu)=0x100000000ULandMPIDR_AFF0_MASKis0xFFU, thecluster_idreturns 0. But the expected value should be0x100000000.So, in this patch, we force aarch64 to use unsigned long as MPIDR mask to avoid such unexpected results.
How about the following commit message:
"Currently, Xen is considering that all the affinity bits are defined below 32-bit. However, Arm64 define a 3rd level affinity in bits 32-39.
The function gicv3_send_sgi_list in the GICv3 driver will compute the cluser using the following code:
uint64_t cluster_id = cpu_logical_map(cpu) & ~MPIDR_AFF0_MASK;
Because MPIDR_AFF0_MASK is defined as a 32-bit value, we will miss out the 3rd level affinity. As a consequence, the IPI would not be sent to the correct vCPU.
This particular error can be solved by switching MPIDR_AFF0_MASK to use unsigned long. However, take the opportunity to switch all the MPIDR_* define to use unsigned long to avoid anymore issue. "
I can update the commit message while committing if you are happy with it.
Cheers,
Quoted 35 lines
Signed-off-by Wei Chen <[email protected]>---
v1 -> v2:
1. Remove inaccurate descriptions
2. Update example description--- xen/include/asm-arm/processor.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)xen/include/asm-arm/processor.h +3
-4@@ -75,11 +75,11 @@/* MPIDR Multiprocessor Affinity Register */ #define _MPIDR_UP (30) -#define MPIDR_UP (_AC(1,U) << _MPIDR_UP) +#define MPIDR_UP (_AC(1,UL) << _MPIDR_UP)#define _MPIDR_SMP (31) -#define MPIDR_SMP (_AC(1,U) << _MPIDR_SMP) +#define MPIDR_SMP (_AC(1,UL) << _MPIDR_SMP)#define MPIDR_AFF0_SHIFT (0) -#define MPIDR_AFF0_MASK (_AC(0xff,U) << MPIDR_AFF0_SHIFT) +#define MPIDR_AFF0_MASK (_AC(0xff,UL) << MPIDR_AFF0_SHIFT)#ifdef CONFIG_ARM_64 #define MPIDR_HWID_MASK _AC(0xff00ffffff,UL)#else --2.25.1
Signature · 1 line
Julien Grall
Hi Julien
Quoted 10 lines
-----Original Message-----
From: Julien Grall <[email protected]>
Sent: 2021年1月8日 19:46
To: Wei Chen <[email protected]>; [email protected]; [email protected]
Cc: Bertrand Marquis <[email protected]>; Penny Zheng <[email protected]>; Jiamei Xie <[email protected]>; nd <[email protected]>
Subject: Re: [PATCH v2] xen/arm: Using unsigned long for arm64 MPIDR mask
Quoted 6 lines
Hi Wei,
How about the following title:
"xen/arm: Don't ignore the affinity level 3 in the MPIDR"
This title makes more sense.
Quoted 7 lines · Wei Chen
Quoted 3 lines
Currently, Xen is using UINT32 for MPIDR mask to retrieve affinity[0,1,2,3] values for
MPIDR_EL1register. The value ofMPIDR_EL1is 64-bit unsigned long. > The 32-bit unsingeds/unsinged/unsigned/
thanks.
Quoted 33 lines
Quoted 12 lines
integer will do unsigned extend while doing some operations with 64-bit unsigned integer. This can lead to unexpected result in some use cases.
For example, in
gicv3_send_sgi_listof GICv3 driver:uint64_tcluster_id=cpu_logical_map(cpu)& ~MPIDR_AFF0_MASK;When
MPIDR_AFF0_MASKis0xFFU, compiler output:f7c: 92785c16 and x22, x0, #0xffffff00 When MPIDR_AFF0_MASK is 0xFFUL, compiler output: f88: 9278dc75 and x21, x3, #0xffffffffffffff00Quoted 6 lines
If
cpu_logical_map(cpu)=0x100000000ULandMPIDR_AFF0_MASKis0xFFU, thecluster_idreturns 0. But the expected value should be0x100000000.So, in this patch, we force aarch64 to use unsigned long as MPIDR mask to avoid such unexpected results.
How about the following commit message:
"Currently, Xen is considering that all the affinity bits are defined below 32-bit. However, Arm64 define a 3rd level affinity in bits 32-39.
The function
gicv3_send_sgi_listin the GICv3 driver will compute the cluser using the following code:
uint64_tcluster_id=cpu_logical_map(cpu)& ~MPIDR_AFF0_MASK;Because
MPIDR_AFF0_MASKis defined as a 32-bit value, we will miss out the 3rd level affinity. As a consequence, the IPI would not be sent to the correct vCPU.
Quoted 7 lines
This particular error can be solved by switching
MPIDR_AFF0_MASKto use unsigned long. However, take the opportunity to switch all the MPIDR_* define to use unsigned long to avoid anymore issue.
"I can update the commit message while committing if you are happy with it.
Yes, that would be good, thank you very much : )
Quoted 41 lines
Cheers,
xen/include/asm-
Quoted 14 lines
Signed-off-by Wei Chen <[email protected]>---
v1 -> v2:
1. Remove inaccurate descriptions
2. Update example description--- xen/include/asm-arm/processor.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)xen/include/asm- renamed from xen/include/asm-arm/processor.h +0
-0
arm/processor.hQuoted 19 lines
index
87c8136022..5c1768cdec100644xen/include/asm-arm/processor.h +3
-4@@ -75,11 +75,11 @@/* MPIDR Multiprocessor Affinity Register */ #define _MPIDR_UP (30) -#define MPIDR_UP (_AC(1,U) << _MPIDR_UP) +#define MPIDR_UP (_AC(1,UL) << _MPIDR_UP)#define _MPIDR_SMP (31) -#define MPIDR_SMP (_AC(1,U) << _MPIDR_SMP) +#define MPIDR_SMP (_AC(1,UL) << _MPIDR_SMP)#define MPIDR_AFF0_SHIFT (0) -#define MPIDR_AFF0_MASK (_AC(0xff,U) << MPIDR_AFF0_SHIFT) +#define MPIDR_AFF0_MASK (_AC(0xff,UL) << MPIDR_AFF0_SHIFT)#ifdef CONFIG_ARM_64 #define MPIDR_HWID_MASK _AC(0xff00ffffff,UL)#else --Quoted 2 lines
2.25.1
--
Julien Grall
Julien Grall
Reviewed-byQuoted 1 line · Wei Chen
Hi Julien
Hi Wei,
Sorry for the late answer. While cleaning my inbox today, I noticied that I didn't reply to this thread :(.
Quoted 38 lines
Quoted 38 lines
Quoted 12 lines
integer will do unsigned extend while doing some operations with 64-bit unsigned integer. This can lead to unexpected result in some use cases.
For example, in
gicv3_send_sgi_listof GICv3 driver:uint64_tcluster_id=cpu_logical_map(cpu)& ~MPIDR_AFF0_MASK;When
MPIDR_AFF0_MASKis0xFFU, compiler output:f7c: 92785c16 and x22, x0, #0xffffff00 When MPIDR_AFF0_MASK is 0xFFUL, compiler output: f88: 9278dc75 and x21, x3, #0xffffffffffffff00Quoted 6 lines
If
cpu_logical_map(cpu)=0x100000000ULandMPIDR_AFF0_MASKis0xFFU, thecluster_idreturns 0. But the expected value should be0x100000000.So, in this patch, we force aarch64 to use unsigned long as MPIDR mask to avoid such unexpected results.
How about the following commit message:
"Currently, Xen is considering that all the affinity bits are defined below 32-bit. However, Arm64 define a 3rd level affinity in bits 32-39.
The function
gicv3_send_sgi_listin the GICv3 driver will compute the cluser using the following code:
uint64_tcluster_id=cpu_logical_map(cpu)& ~MPIDR_AFF0_MASK;Because
MPIDR_AFF0_MASKis defined as a 32-bit value, we will miss out the 3rd level affinity. As a consequence, the IPI would not be sent to the correct vCPU.This particular error can be solved by switching
MPIDR_AFF0_MASKto use unsigned long. However, take the opportunity to switch all the MPIDR_* define to use unsigned long to avoid anymore issue.
"
Quoted 4 lines
Quoted 2 lines
I can update the commit message while committing if you are happy with it.
Yes, that would be good, thank you very much : )
And committed.
Cheers,
Signature · 1 line
Julien Grall
Bertrand Marquis
Reviewed-byHi,
Quoted 47 lines · Julien Grall
Quoted 1 line · Wei Chen
Hi Julien
Hi Wei,
Sorry for the late answer. While cleaning my inbox today, I noticied that I didn't reply to this thread :(.
Quoted 38 lines
Quoted 38 lines
Quoted 12 lines
integer will do unsigned extend while doing some operations with 64-bit unsigned integer. This can lead to unexpected result in some use cases.
For example, in
gicv3_send_sgi_listof GICv3 driver:uint64_tcluster_id=cpu_logical_map(cpu)& ~MPIDR_AFF0_MASK;When
MPIDR_AFF0_MASKis0xFFU, compiler output:f7c: 92785c16 and x22, x0, #0xffffff00 When MPIDR_AFF0_MASK is 0xFFUL, compiler output: f88: 9278dc75 and x21, x3, #0xffffffffffffff00Quoted 6 lines
If
cpu_logical_map(cpu)=0x100000000ULandMPIDR_AFF0_MASKis0xFFU, thecluster_idreturns 0. But the expected value should be0x100000000.So, in this patch, we force aarch64 to use unsigned long as MPIDR mask to avoid such unexpected results.
How about the following commit message:
"Currently, Xen is considering that all the affinity bits are defined below 32-bit. However, Arm64 define a 3rd level affinity in bits 32-39.
The function
gicv3_send_sgi_listin the GICv3 driver will compute the cluser using the following code:
uint64_tcluster_id=cpu_logical_map(cpu)& ~MPIDR_AFF0_MASK;Because
MPIDR_AFF0_MASKis defined as a 32-bit value, we will miss out the 3rd level affinity. As a consequence, the IPI would not be sent to the correct vCPU.This particular error can be solved by switching
MPIDR_AFF0_MASKto use unsigned long. However, take the opportunity to switch all the MPIDR_* define to use unsigned long to avoid anymore issue.
"
Quoted 5 lines
Quoted 3 lines
Quoted 2 lines
I can update the commit message while committing if you are happy with it.
Yes, that would be good, thank you very much : )
Reviewed-by Julien Grall <[email protected]>
Also tested on a platform where Xen was not booting without this patch and i can confirm it fixed the issue :-)
Cheers
Bertrand
Quoted 7 lines
And committed.
Cheers,
Signature · 1 line
Julien Grall