x86/hyperv : Fixing removal of __iomem address space warning

[PATCH v2] x86/hyperv : Fixing removal of __iomem address space warning

Posted by Abhinav Singh 2 years, 2 months ago

This patch fixes two sparse warnings

1. sparse complaining about the removal of __iomem address
space when casting the return value of this ioremap_cache(...)
from `void __ioremap*` to `void*`.
Fixed this by replacing the ioremap_cache(...)
by memremap(...) and using MEMREMAP_DEC and MEMREMAP_WB flag for making
sure the memory is always decrypted and it will support full write back
cache.

2. sparse complaining `expected void volatile [noderef] __iomem *addr`
when calling iounmap with a non __iomem pointer.
Fixed this by replacing iounmap(...) with memumap(...).

Signed-off-by: Abhinav Singh <singhabhinav9051571833@gmail.com>
---

v1:
https://lore.kernel.org/all/19cec6f0-e176-4bcc-95a0-9d6eb0261ed1@gmail.com/T/

v1 to v2:
1. Fixed the comment which was earlier describing ioremap_cache(...).
2. Replaced iounmap(...) with memremap(...) inside function hv_cpu_die(...).

 arch/x86/hyperv/hv_init.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/arch/x86/hyperv/hv_init.c b/arch/x86/hyperv/hv_init.c
index 21556ad87f4b..2a14928b8a36 100644
--- a/arch/x86/hyperv/hv_init.c
+++ b/arch/x86/hyperv/hv_init.c
@@ -68,9 +68,11 @@ static int hyperv_init_ghcb(void)
 	 */
 	rdmsrl(MSR_AMD64_SEV_ES_GHCB, ghcb_gpa);

-	/* Mask out vTOM bit. ioremap_cache() maps decrypted */
+	/* Mask out vTOM bit.
+	MEMREMAP_WB full write back cache
+	MEMREMAP_DEC maps decrypted memory */
 	ghcb_gpa &= ~ms_hyperv.shared_gpa_boundary;
-	ghcb_va = (void *)ioremap_cache(ghcb_gpa, HV_HYP_PAGE_SIZE);
+	ghcb_va = memremap(ghcb_gpa, HV_HYP_PAGE_SIZE, MEMREMAP_WB | MEMREMAP_DEC);
 	if (!ghcb_va)
 		return -ENOMEM;

@@ -238,7 +240,7 @@ static int hv_cpu_die(unsigned int cpu)
 	if (hv_ghcb_pg) {
 		ghcb_va = (void **)this_cpu_ptr(hv_ghcb_pg);
 		if (*ghcb_va)
-			iounmap(*ghcb_va);
+			memunmap(*ghcb_va);
 		*ghcb_va = NULL;
 	}

--
2.39.2

RE: [PATCH v2] x86/hyperv : Fixing removal of __iomem address space warning

Posted by Michael Kelley 2 years, 1 month ago

From: Abhinav Singh <singhabhinav9051571833@gmail.com> Sent: Monday, October 30, 2023 3:50 PM
> 
> This patch fixes two sparse warnings
> 
> 1. sparse complaining about the removal of __iomem address
> space when casting the return value of this ioremap_cache(...)
> from `void __ioremap*` to `void*`.
> Fixed this by replacing the ioremap_cache(...)
> by memremap(...) and using MEMREMAP_DEC and MEMREMAP_WB flag for
> making
> sure the memory is always decrypted and it will support full write back
> cache.
> 
> 2. sparse complaining `expected void volatile [noderef] __iomem *addr`
> when calling iounmap with a non __iomem pointer.
> Fixed this by replacing iounmap(...) with memumap(...).
> 
> Signed-off-by: Abhinav Singh <singhabhinav9051571833@gmail.com>

Since Nischala has posted her more comprehensive patch,
this patch can be dropped.  But see one comment below for
future reference.

> ---
> 
> v1:
> https://lore.kernel.org/all/19cec6f0-e176-4bcc-95a0-9d6eb0261ed1@gmail.com/T/
> 
> v1 to v2:
> 1. Fixed the comment which was earlier describing ioremap_cache(...).
> 2. Replaced iounmap(...) with memremap(...) inside function hv_cpu_die(...).
> 
>  arch/x86/hyperv/hv_init.c | 8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/hyperv/hv_init.c b/arch/x86/hyperv/hv_init.c
> index 21556ad87f4b..2a14928b8a36 100644
> --- a/arch/x86/hyperv/hv_init.c
> +++ b/arch/x86/hyperv/hv_init.c
> @@ -68,9 +68,11 @@ static int hyperv_init_ghcb(void)
>  	 */
>  	rdmsrl(MSR_AMD64_SEV_ES_GHCB, ghcb_gpa);
> 
> -	/* Mask out vTOM bit. ioremap_cache() maps decrypted */
> +	/* Mask out vTOM bit.
> +	MEMREMAP_WB full write back cache
> +	MEMREMAP_DEC maps decrypted memory */

This isn't the right style for multi-line patches.  Correct would be:

	/*
	 * Mask out vTOM bit.
	 * MEMREMAP_WB full write back cache
	 * MEMREMAP_DEC maps decrypted memory
	 */

Section 8 of coding-style.rst under Documentation/process covers
these details.   Note that the style is slightly different for code
under net and drivers/net.

Michael

>  	ghcb_gpa &= ~ms_hyperv.shared_gpa_boundary;
> -	ghcb_va = (void *)ioremap_cache(ghcb_gpa, HV_HYP_PAGE_SIZE);
> +	ghcb_va = memremap(ghcb_gpa, HV_HYP_PAGE_SIZE, MEMREMAP_WB | MEMREMAP_DEC);
>  	if (!ghcb_va)
>  		return -ENOMEM;
> 
> @@ -238,7 +240,7 @@ static int hv_cpu_die(unsigned int cpu)
>  	if (hv_ghcb_pg) {
>  		ghcb_va = (void **)this_cpu_ptr(hv_ghcb_pg);
>  		if (*ghcb_va)
> -			iounmap(*ghcb_va);
> +			memunmap(*ghcb_va);
>  		*ghcb_va = NULL;
>  	}
> 
> --
> 2.39.2

Re: [PATCH v2] x86/hyperv : Fixing removal of __iomem address space warning

Posted by Abhinav Singh 2 years, 1 month ago

On 11/5/23 20:55, Michael Kelley wrote:
> From: Abhinav Singh <singhabhinav9051571833@gmail.com> Sent: Monday, October 30, 2023 3:50 PM
>>
>> This patch fixes two sparse warnings
>>
>> 1. sparse complaining about the removal of __iomem address
>> space when casting the return value of this ioremap_cache(...)
>> from `void __ioremap*` to `void*`.
>> Fixed this by replacing the ioremap_cache(...)
>> by memremap(...) and using MEMREMAP_DEC and MEMREMAP_WB flag for
>> making
>> sure the memory is always decrypted and it will support full write back
>> cache.
>>
>> 2. sparse complaining `expected void volatile [noderef] __iomem *addr`
>> when calling iounmap with a non __iomem pointer.
>> Fixed this by replacing iounmap(...) with memumap(...).
>>
>> Signed-off-by: Abhinav Singh <singhabhinav9051571833@gmail.com>
> 
> Since Nischala has posted her more comprehensive patch,
> this patch can be dropped.  But see one comment below for
> future reference.
> 
>> ---
>>
>> v1:
>> https://lore.kernel.org/all/19cec6f0-e176-4bcc-95a0-9d6eb0261ed1@gmail.com/T/
>>
>> v1 to v2:
>> 1. Fixed the comment which was earlier describing ioremap_cache(...).
>> 2. Replaced iounmap(...) with memremap(...) inside function hv_cpu_die(...).
>>
>>   arch/x86/hyperv/hv_init.c | 8 +++++---
>>   1 file changed, 5 insertions(+), 3 deletions(-)
>>
>> diff --git a/arch/x86/hyperv/hv_init.c b/arch/x86/hyperv/hv_init.c
>> index 21556ad87f4b..2a14928b8a36 100644
>> --- a/arch/x86/hyperv/hv_init.c
>> +++ b/arch/x86/hyperv/hv_init.c
>> @@ -68,9 +68,11 @@ static int hyperv_init_ghcb(void)
>>   	 */
>>   	rdmsrl(MSR_AMD64_SEV_ES_GHCB, ghcb_gpa);
>>
>> -	/* Mask out vTOM bit. ioremap_cache() maps decrypted */
>> +	/* Mask out vTOM bit.
>> +	MEMREMAP_WB full write back cache
>> +	MEMREMAP_DEC maps decrypted memory */
> 
> This isn't the right style for multi-line patches.  Correct would be:
> 
> 	/*
> 	 * Mask out vTOM bit.
> 	 * MEMREMAP_WB full write back cache
> 	 * MEMREMAP_DEC maps decrypted memory
> 	 */
> 
> Section 8 of coding-style.rst under Documentation/process covers
> these details.   Note that the style is slightly different for code
> under net and drivers/net.
> 
> Michael
> 
>>   	ghcb_gpa &= ~ms_hyperv.shared_gpa_boundary;
>> -	ghcb_va = (void *)ioremap_cache(ghcb_gpa, HV_HYP_PAGE_SIZE);
>> +	ghcb_va = memremap(ghcb_gpa, HV_HYP_PAGE_SIZE, MEMREMAP_WB | MEMREMAP_DEC);
>>   	if (!ghcb_va)
>>   		return -ENOMEM;
>>
>> @@ -238,7 +240,7 @@ static int hv_cpu_die(unsigned int cpu)
>>   	if (hv_ghcb_pg) {
>>   		ghcb_va = (void **)this_cpu_ptr(hv_ghcb_pg);
>>   		if (*ghcb_va)
>> -			iounmap(*ghcb_va);
>> +			memunmap(*ghcb_va);
>>   		*ghcb_va = NULL;
>>   	}
>>
>> --
>> 2.39.2
> 
Thanks a lot for taking a look and the suggestion, will keep this in 
mind in future patches.

Thanks,
Abhinav