[patch] x86/pat: pass correct address to sanitize_phys

[patch] x86/pat: pass correct address to sanitize_phys

[Jeff Moyer]

memtype_reserve takes an address range of the form [start, end).  It
then passes the start and end addresses to sanitize_phys, which is meant
to operate on the inclusive addresses.  If end falls at the end of the
physical address space, sanitize_phys will return 0.  This can result in
drivers failing to load:

[   10.000087] mpt3sas_cm0: unable to map adapter memory! or resource not found
[   10.000334] mpt3sas_cm0: failure at drivers/scsi/mpt3sas/mpt3sas_scsih.c:10597/_scsih_probe()!

Fix this by passing the inclusive end address to sanitize_phys.

Fixes: 510ee090abc3 ("x86/mm/pat: Prepare {reserve, free}_memtype() for "decoy" addresses")
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
--
It might be worth adding a comment, here.  If there are any suggestions
on what a sane wording would be, I'm all ears.

diff --git a/arch/x86/mm/pat/memtype.c b/arch/x86/mm/pat/memtype.c
index 3112ca7786ed..482557905294 100644
--- a/arch/x86/mm/pat/memtype.c
+++ b/arch/x86/mm/pat/memtype.c
@@ -583,7 +583,7 @@ int memtype_reserve(u64 start, u64 end, enum page_cache_mode req_type,
 	int err = 0;
 
 	start = sanitize_phys(start);
-	end = sanitize_phys(end);
+	end = sanitize_phys(end - 1) + 1;
 	if (start >= end) {
 		WARN(1, "%s failed: [mem %#010Lx-%#010Lx], req %s\n", __func__,
 				start, end - 1, cattr_name(req_type));

Re: [patch] x86/pat: pass correct address to sanitize_phys

[Jeff Moyer]

Ping?  This patch fixes a real problem seen in the field.  Can I get
some review?

Thanks!
Jeff

Jeff Moyer <jmoyer@redhat.com> writes:

> memtype_reserve takes an address range of the form [start, end).  It
> then passes the start and end addresses to sanitize_phys, which is meant
> to operate on the inclusive addresses.  If end falls at the end of the
> physical address space, sanitize_phys will return 0.  This can result in
> drivers failing to load:
>
> [   10.000087] mpt3sas_cm0: unable to map adapter memory! or resource not found
> [   10.000334] mpt3sas_cm0: failure at drivers/scsi/mpt3sas/mpt3sas_scsih.c:10597/_scsih_probe()!
>
> Fix this by passing the inclusive end address to sanitize_phys.
>
> Fixes: 510ee090abc3 ("x86/mm/pat: Prepare {reserve, free}_memtype() for "decoy" addresses")
> Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
> --
> It might be worth adding a comment, here.  If there are any suggestions
> on what a sane wording would be, I'm all ears.
>
> diff --git a/arch/x86/mm/pat/memtype.c b/arch/x86/mm/pat/memtype.c
> index 3112ca7786ed..482557905294 100644
> --- a/arch/x86/mm/pat/memtype.c
> +++ b/arch/x86/mm/pat/memtype.c
> @@ -583,7 +583,7 @@ int memtype_reserve(u64 start, u64 end, enum page_cache_mode req_type,
>  	int err = 0;
>  
>  	start = sanitize_phys(start);
> -	end = sanitize_phys(end);
> +	end = sanitize_phys(end - 1) + 1;
>  	if (start >= end) {
>  		WARN(1, "%s failed: [mem %#010Lx-%#010Lx], req %s\n", __func__,
>  				start, end - 1, cattr_name(req_type));

Re: [patch] x86/pat: pass correct address to sanitize_phys

[David Hildenbrand]

On 21.07.21 21:48, Jeff Moyer wrote:
> memtype_reserve takes an address range of the form [start, end).  It
> then passes the start and end addresses to sanitize_phys, which is meant
> to operate on the inclusive addresses.  If end falls at the end of the
> physical address space, sanitize_phys will return 0.  This can result in
> drivers failing to load:
> 
> [   10.000087] mpt3sas_cm0: unable to map adapter memory! or resource not found
> [   10.000334] mpt3sas_cm0: failure at drivers/scsi/mpt3sas/mpt3sas_scsih.c:10597/_scsih_probe()!
> 
> Fix this by passing the inclusive end address to sanitize_phys.
> 
> Fixes: 510ee090abc3 ("x86/mm/pat: Prepare {reserve, free}_memtype() for "decoy" addresses")
> Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
> --
> It might be worth adding a comment, here.  If there are any suggestions
> on what a sane wording would be, I'm all ears.
> 
> diff --git a/arch/x86/mm/pat/memtype.c b/arch/x86/mm/pat/memtype.c
> index 3112ca7786ed..482557905294 100644
> --- a/arch/x86/mm/pat/memtype.c
> +++ b/arch/x86/mm/pat/memtype.c
> @@ -583,7 +583,7 @@ int memtype_reserve(u64 start, u64 end, enum page_cache_mode req_type,
>   	int err = 0;
>   
>   	start = sanitize_phys(start);
> -	end = sanitize_phys(end);
> +	end = sanitize_phys(end - 1) + 1;
>   	if (start >= end) {
>   		WARN(1, "%s failed: [mem %#010Lx-%#010Lx], req %s\n", __func__,
>   				start, end - 1, cattr_name(req_type));
> 
> 

LGTM

Reviewed-by: David Hildenbrand <david@redhat.com>

-- 
Thanks,

David / dhildenb

Re: [patch] x86/pat: pass correct address to sanitize_phys

[Thomas Gleixner]

Jeff,

On Wed, Jul 21 2021 at 15:48, Jeff Moyer wrote:

Please write function names with brackets, i.e. sanitize_phys().

> memtype_reserve takes an address range of the form [start, end).  It

[start, end]


> then passes the start and end addresses to sanitize_phys, which is meant
> to operate on the inclusive addresses.  If end falls at the end of the
> physical address space, sanitize_phys will return 0.  This can result in
> drivers failing to load:
>
> [   10.000087] mpt3sas_cm0: unable to map adapter memory! or resource not found
> [   10.000334] mpt3sas_cm0: failure at drivers/scsi/mpt3sas/mpt3sas_scsih.c:10597/_scsih_probe()!

Doesn't this trigger the WARN() right below that offending line?

> Fix this by passing the inclusive end address to sanitize_phys.
>
> Fixes: 510ee090abc3 ("x86/mm/pat: Prepare {reserve, free}_memtype() for "decoy" addresses")
> Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
> --
> It might be worth adding a comment, here.  If there are any suggestions
> on what a sane wording would be, I'm all ears.

See below.

> diff --git a/arch/x86/mm/pat/memtype.c b/arch/x86/mm/pat/memtype.c
> index 3112ca7786ed..482557905294 100644
> --- a/arch/x86/mm/pat/memtype.c
> +++ b/arch/x86/mm/pat/memtype.c
> @@ -583,7 +583,7 @@ int memtype_reserve(u64 start, u64 end, enum page_cache_mode req_type,
>  	int err = 0;
>  
>  	start = sanitize_phys(start);
> -	end = sanitize_phys(end);

        /*
         * [start, end] is an exclusive address range, but
         * sanitize_phys() expects an inclusive end address
         */

> +	end = sanitize_phys(end - 1) + 1;
>  	if (start >= end) {
>  		WARN(1, "%s failed: [mem %#010Lx-%#010Lx], req %s\n", __func__,
>  				start, end - 1, cattr_name(req_type));

Thanks,

        tglx

Re: [patch] x86/pat: pass correct address to sanitize_phys

[Jeff Moyer]

Thomas Gleixner <tglx@linutronix.de> writes:

> Jeff,
>
> On Wed, Jul 21 2021 at 15:48, Jeff Moyer wrote:
>
> Please write function names with brackets, i.e. sanitize_phys().

OK, will do.

>> memtype_reserve takes an address range of the form [start, end).  It
>
> [start, end]

Start is inclusive, end is exclusive, so start <= x < end.  I used the
notation found here:

  https://en.wikipedia.org/wiki/Interval_(mathematics)

If that's too confusing, I can stick to inclusive vs exclusive verbiage.

>> then passes the start and end addresses to sanitize_phys, which is meant
>> to operate on the inclusive addresses.  If end falls at the end of the
>> physical address space, sanitize_phys will return 0.  This can result in
>> drivers failing to load:
>>
>> [   10.000087] mpt3sas_cm0: unable to map adapter memory! or resource not found
>> [   10.000334] mpt3sas_cm0: failure at drivers/scsi/mpt3sas/mpt3sas_scsih.c:10597/_scsih_probe()!
>
> Doesn't this trigger the WARN() right below that offending line?

It does.  I'll include the warning message in the v2 posting.

>> Fix this by passing the inclusive end address to sanitize_phys.
>>
>> Fixes: 510ee090abc3 ("x86/mm/pat: Prepare {reserve, free}_memtype() for "decoy" addresses")
>> Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
>> --
>> It might be worth adding a comment, here.  If there are any suggestions
>> on what a sane wording would be, I'm all ears.
>
> See below.
>
>> diff --git a/arch/x86/mm/pat/memtype.c b/arch/x86/mm/pat/memtype.c
>> index 3112ca7786ed..482557905294 100644
>> --- a/arch/x86/mm/pat/memtype.c
>> +++ b/arch/x86/mm/pat/memtype.c
>> @@ -583,7 +583,7 @@ int memtype_reserve(u64 start, u64 end, enum page_cache_mode req_type,
>>  	int err = 0;
>>  
>>  	start = sanitize_phys(start);
>> -	end = sanitize_phys(end);
>
>         /*
>          * [start, end] is an exclusive address range, but
>          * sanitize_phys() expects an inclusive end address
>          */

That works for me (modulo the interval notation), thanks for the
suggestion.

Thanks!
Jeff