diff mbox

[3/3] omap: iovmm: don't check 'da' to set IOVMF_DA_FIXED/IOVMF_DA_ANON flags

Message ID 1299588365-2749-4-git-send-email-dacohen@gmail.com (mailing list archive)
State Superseded, archived
Headers

Commit Message

David Cohen March 8, 2011, 12:46 p.m. UTC 
  Currently IOVMM driver sets IOVMF_DA_FIXED/IOVMF_DA_ANON flags according
to input 'da' address when mapping memory:
da == 0: IOVMF_DA_ANON
da != 0: IOVMF_DA_FIXED

It prevents IOMMU to map first page with fixed 'da'. To avoid such
issue, IOVMM will not automatically set IOVMF_DA_FIXED. It should now
come from the user. IOVMF_DA_ANON will be automatically set if
IOVMF_DA_FIXED isn't set.

Signed-off-by: David Cohen <dacohen@gmail.com>
---
 arch/arm/plat-omap/iovmm.c |   12 ++++++++----
 1 files changed, 8 insertions(+), 4 deletions(-)

Comments

Guzman Lugo, Fernando March 8, 2011, 5:59 p.m. UTC | #1 
On Tue, Mar 8, 2011 at 6:46 AM, David Cohen <dacohen@gmail.com> wrote:
> Currently IOVMM driver sets IOVMF_DA_FIXED/IOVMF_DA_ANON flags according
> to input 'da' address when mapping memory:
> da == 0: IOVMF_DA_ANON
> da != 0: IOVMF_DA_FIXED
>
> It prevents IOMMU to map first page with fixed 'da'. To avoid such
> issue, IOVMM will not automatically set IOVMF_DA_FIXED. It should now
> come from the user. IOVMF_DA_ANON will be automatically set if
> IOVMF_DA_FIXED isn't set.
>
> Signed-off-by: David Cohen <dacohen@gmail.com>
> ---
>  arch/arm/plat-omap/iovmm.c |   12 ++++++++----
>  1 files changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/arch/arm/plat-omap/iovmm.c b/arch/arm/plat-omap/iovmm.c
> index 11c9b76..dde9cb0 100644
> --- a/arch/arm/plat-omap/iovmm.c
> +++ b/arch/arm/plat-omap/iovmm.c
> @@ -654,7 +654,8 @@ u32 iommu_vmap(struct iommu *obj, u32 da, const struct sg_table *sgt,
>        flags &= IOVMF_HW_MASK;
>        flags |= IOVMF_DISCONT;
>        flags |= IOVMF_MMIO;
> -       flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
> +       if (~flags & IOVMF_DA_FIXED)
> +               flags |= IOVMF_DA_ANON;

could we use only one? both are mutual exclusive, what happen if flag
is IOVMF_DA_FIXED | IOVMF_DA_ANON? so, I suggest to get rid of
IOVMF_DA_ANON.

Regards,
Fernando.

>
>        da = __iommu_vmap(obj, da, sgt, va, bytes, flags);
>        if (IS_ERR_VALUE(da))
> @@ -713,7 +714,8 @@ u32 iommu_vmalloc(struct iommu *obj, u32 da, size_t bytes, u32 flags)
>        flags &= IOVMF_HW_MASK;
>        flags |= IOVMF_DISCONT;
>        flags |= IOVMF_ALLOC;
> -       flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
> +       if (~flags & IOVMF_DA_FIXED)
> +               flags |= IOVMF_DA_ANON;
>
>        sgt = sgtable_alloc(bytes, flags, da, 0);
>        if (IS_ERR(sgt)) {
> @@ -803,7 +805,8 @@ u32 iommu_kmap(struct iommu *obj, u32 da, u32 pa, size_t bytes,
>        flags &= IOVMF_HW_MASK;
>        flags |= IOVMF_LINEAR;
>        flags |= IOVMF_MMIO;
> -       flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
> +       if (~flags & IOVMF_DA_FIXED)
> +               flags |= IOVMF_DA_ANON;
>
>        da = __iommu_kmap(obj, da, pa, va, bytes, flags);
>        if (IS_ERR_VALUE(da))
> @@ -862,7 +865,8 @@ u32 iommu_kmalloc(struct iommu *obj, u32 da, size_t bytes, u32 flags)
>        flags &= IOVMF_HW_MASK;
>        flags |= IOVMF_LINEAR;
>        flags |= IOVMF_ALLOC;
> -       flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
> +       if (~flags & IOVMF_DA_FIXED)
> +               flags |= IOVMF_DA_ANON;
>
>        da = __iommu_kmap(obj, da, pa, va, bytes, flags);
>        if (IS_ERR_VALUE(da))
> --
> 1.7.0.4
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Hiroshi DOYU March 8, 2011, 6:09 p.m. UTC | #2 
From: "ext Guzman Lugo, Fernando" <fernando.lugo@ti.com>
Subject: Re: [PATCH 3/3] omap: iovmm: don't check 'da' to set IOVMF_DA_FIXED/IOVMF_DA_ANON flags
Date: Tue, 8 Mar 2011 11:59:43 -0600

> On Tue, Mar 8, 2011 at 6:46 AM, David Cohen <dacohen@gmail.com> wrote:
>> Currently IOVMM driver sets IOVMF_DA_FIXED/IOVMF_DA_ANON flags according
>> to input 'da' address when mapping memory:
>> da == 0: IOVMF_DA_ANON
>> da != 0: IOVMF_DA_FIXED
>>
>> It prevents IOMMU to map first page with fixed 'da'. To avoid such
>> issue, IOVMM will not automatically set IOVMF_DA_FIXED. It should now
>> come from the user. IOVMF_DA_ANON will be automatically set if
>> IOVMF_DA_FIXED isn't set.
>>
>> Signed-off-by: David Cohen <dacohen@gmail.com>
>> ---
>>  arch/arm/plat-omap/iovmm.c |   12 ++++++++----
>>  1 files changed, 8 insertions(+), 4 deletions(-)
>>
>> diff --git a/arch/arm/plat-omap/iovmm.c b/arch/arm/plat-omap/iovmm.c
>> index 11c9b76..dde9cb0 100644
>> --- a/arch/arm/plat-omap/iovmm.c
>> +++ b/arch/arm/plat-omap/iovmm.c
>> @@ -654,7 +654,8 @@ u32 iommu_vmap(struct iommu *obj, u32 da, const struct sg_table *sgt,
>>        flags &= IOVMF_HW_MASK;
>>        flags |= IOVMF_DISCONT;
>>        flags |= IOVMF_MMIO;
>> -       flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
>> +       if (~flags & IOVMF_DA_FIXED)
>> +               flags |= IOVMF_DA_ANON;
> 
> could we use only one? both are mutual exclusive, what happen if flag
> is IOVMF_DA_FIXED | IOVMF_DA_ANON? so, I suggest to get rid of
> IOVMF_DA_ANON.

Then, what about introducing some MACRO? Better names?

#define set_iovmf_da_anon(flags)
#define set_iovmf_da_fix(flags)
#define set_iovmf_mmio(flags)
......
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Guzman Lugo, Fernando March 8, 2011, 6:53 p.m. UTC | #3 
On Tue, Mar 8, 2011 at 12:09 PM, Hiroshi DOYU <Hiroshi.DOYU@nokia.com> wrote:
> From: "ext Guzman Lugo, Fernando" <fernando.lugo@ti.com>
> Subject: Re: [PATCH 3/3] omap: iovmm: don't check 'da' to set IOVMF_DA_FIXED/IOVMF_DA_ANON flags
> Date: Tue, 8 Mar 2011 11:59:43 -0600
>
>> On Tue, Mar 8, 2011 at 6:46 AM, David Cohen <dacohen@gmail.com> wrote:
>>> Currently IOVMM driver sets IOVMF_DA_FIXED/IOVMF_DA_ANON flags according
>>> to input 'da' address when mapping memory:
>>> da == 0: IOVMF_DA_ANON
>>> da != 0: IOVMF_DA_FIXED
>>>
>>> It prevents IOMMU to map first page with fixed 'da'. To avoid such
>>> issue, IOVMM will not automatically set IOVMF_DA_FIXED. It should now
>>> come from the user. IOVMF_DA_ANON will be automatically set if
>>> IOVMF_DA_FIXED isn't set.
>>>
>>> Signed-off-by: David Cohen <dacohen@gmail.com>
>>> ---
>>>  arch/arm/plat-omap/iovmm.c |   12 ++++++++----
>>>  1 files changed, 8 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/arch/arm/plat-omap/iovmm.c b/arch/arm/plat-omap/iovmm.c
>>> index 11c9b76..dde9cb0 100644
>>> --- a/arch/arm/plat-omap/iovmm.c
>>> +++ b/arch/arm/plat-omap/iovmm.c
>>> @@ -654,7 +654,8 @@ u32 iommu_vmap(struct iommu *obj, u32 da, const struct sg_table *sgt,
>>>        flags &= IOVMF_HW_MASK;
>>>        flags |= IOVMF_DISCONT;
>>>        flags |= IOVMF_MMIO;
>>> -       flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
>>> +       if (~flags & IOVMF_DA_FIXED)
>>> +               flags |= IOVMF_DA_ANON;
>>
>> could we use only one? both are mutual exclusive, what happen if flag
>> is IOVMF_DA_FIXED | IOVMF_DA_ANON? so, I suggest to get rid of
>> IOVMF_DA_ANON.
>
> Then, what about introducing some MACRO? Better names?
>
> #define set_iovmf_da_anon(flags)
> #define set_iovmf_da_fix(flags)
> #define set_iovmf_mmio(flags)

will they be used by the users?

I think people are more used to use

iommu_vmap(obj, da, sgt, IOVMF_MMIO | IOVMF_DA_ANON);

than

set_iovmf_da_anon(flags)
set_iovmf_mmio(flags)
iommu_vmap(obj, da, sgt, flags);

I don't have problem with the change, but I think how it is now is ok,
just that we don't we two bits to handle anon/fixed da, it can be
managed it only 1 bit (one flag), or is there a issue?

Regards,
Fernando.
> ......
>
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
David Cohen March 8, 2011, 6:57 p.m. UTC | #4 
Hi Hiroshi, Fernando,

On Tue, Mar 8, 2011 at 8:53 PM, Guzman Lugo, Fernando
<fernando.lugo@ti.com> wrote:
> On Tue, Mar 8, 2011 at 12:09 PM, Hiroshi DOYU <Hiroshi.DOYU@nokia.com> wrote:
>> From: "ext Guzman Lugo, Fernando" <fernando.lugo@ti.com>
>> Subject: Re: [PATCH 3/3] omap: iovmm: don't check 'da' to set IOVMF_DA_FIXED/IOVMF_DA_ANON flags
>> Date: Tue, 8 Mar 2011 11:59:43 -0600
>>
>>> On Tue, Mar 8, 2011 at 6:46 AM, David Cohen <dacohen@gmail.com> wrote:
>>>> Currently IOVMM driver sets IOVMF_DA_FIXED/IOVMF_DA_ANON flags according
>>>> to input 'da' address when mapping memory:
>>>> da == 0: IOVMF_DA_ANON
>>>> da != 0: IOVMF_DA_FIXED
>>>>
>>>> It prevents IOMMU to map first page with fixed 'da'. To avoid such
>>>> issue, IOVMM will not automatically set IOVMF_DA_FIXED. It should now
>>>> come from the user. IOVMF_DA_ANON will be automatically set if
>>>> IOVMF_DA_FIXED isn't set.
>>>>
>>>> Signed-off-by: David Cohen <dacohen@gmail.com>
>>>> ---
>>>>  arch/arm/plat-omap/iovmm.c |   12 ++++++++----
>>>>  1 files changed, 8 insertions(+), 4 deletions(-)
>>>>
>>>> diff --git a/arch/arm/plat-omap/iovmm.c b/arch/arm/plat-omap/iovmm.c
>>>> index 11c9b76..dde9cb0 100644
>>>> --- a/arch/arm/plat-omap/iovmm.c
>>>> +++ b/arch/arm/plat-omap/iovmm.c
>>>> @@ -654,7 +654,8 @@ u32 iommu_vmap(struct iommu *obj, u32 da, const struct sg_table *sgt,
>>>>        flags &= IOVMF_HW_MASK;
>>>>        flags |= IOVMF_DISCONT;
>>>>        flags |= IOVMF_MMIO;
>>>> -       flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
>>>> +       if (~flags & IOVMF_DA_FIXED)
>>>> +               flags |= IOVMF_DA_ANON;
>>>
>>> could we use only one? both are mutual exclusive, what happen if flag
>>> is IOVMF_DA_FIXED | IOVMF_DA_ANON? so, I suggest to get rid of
>>> IOVMF_DA_ANON.
>>
>> Then, what about introducing some MACRO? Better names?
>>
>> #define set_iovmf_da_anon(flags)
>> #define set_iovmf_da_fix(flags)
>> #define set_iovmf_mmio(flags)
>
> will they be used by the users?
>
> I think people are more used to use
>
> iommu_vmap(obj, da, sgt, IOVMF_MMIO | IOVMF_DA_ANON);

I'd be happier with this approach, instead of the macros. :)
It's intuitive and very common on kernel.

>
> than
>
> set_iovmf_da_anon(flags)
> set_iovmf_mmio(flags)
> iommu_vmap(obj, da, sgt, flags);
>
> I don't have problem with the change, but I think how it is now is ok,
> just that we don't we two bits to handle anon/fixed da, it can be
> managed it only 1 bit (one flag), or is there a issue?

We can exclude IOVMF_DA_ANON and stick with IOVMF_DA_FIXED only.
I can resend my patch if we agree it's OK.

Regards,

David

>
> Regards,
> Fernando.
>> ......
>>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Guzman Lugo, Fernando March 8, 2011, 7:33 p.m. UTC | #5 
On Tue, Mar 8, 2011 at 12:57 PM, David Cohen <dacohen@gmail.com> wrote:
> Hi Hiroshi, Fernando,
>
> On Tue, Mar 8, 2011 at 8:53 PM, Guzman Lugo, Fernando
> <fernando.lugo@ti.com> wrote:
>> On Tue, Mar 8, 2011 at 12:09 PM, Hiroshi DOYU <Hiroshi.DOYU@nokia.com> wrote:
>>> From: "ext Guzman Lugo, Fernando" <fernando.lugo@ti.com>
>>> Subject: Re: [PATCH 3/3] omap: iovmm: don't check 'da' to set IOVMF_DA_FIXED/IOVMF_DA_ANON flags
>>> Date: Tue, 8 Mar 2011 11:59:43 -0600
>>>
>>>> On Tue, Mar 8, 2011 at 6:46 AM, David Cohen <dacohen@gmail.com> wrote:
>>>>> Currently IOVMM driver sets IOVMF_DA_FIXED/IOVMF_DA_ANON flags according
>>>>> to input 'da' address when mapping memory:
>>>>> da == 0: IOVMF_DA_ANON
>>>>> da != 0: IOVMF_DA_FIXED
>>>>>
>>>>> It prevents IOMMU to map first page with fixed 'da'. To avoid such
>>>>> issue, IOVMM will not automatically set IOVMF_DA_FIXED. It should now
>>>>> come from the user. IOVMF_DA_ANON will be automatically set if
>>>>> IOVMF_DA_FIXED isn't set.
>>>>>
>>>>> Signed-off-by: David Cohen <dacohen@gmail.com>
>>>>> ---
>>>>>  arch/arm/plat-omap/iovmm.c |   12 ++++++++----
>>>>>  1 files changed, 8 insertions(+), 4 deletions(-)
>>>>>
>>>>> diff --git a/arch/arm/plat-omap/iovmm.c b/arch/arm/plat-omap/iovmm.c
>>>>> index 11c9b76..dde9cb0 100644
>>>>> --- a/arch/arm/plat-omap/iovmm.c
>>>>> +++ b/arch/arm/plat-omap/iovmm.c
>>>>> @@ -654,7 +654,8 @@ u32 iommu_vmap(struct iommu *obj, u32 da, const struct sg_table *sgt,
>>>>>        flags &= IOVMF_HW_MASK;
>>>>>        flags |= IOVMF_DISCONT;
>>>>>        flags |= IOVMF_MMIO;
>>>>> -       flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
>>>>> +       if (~flags & IOVMF_DA_FIXED)
>>>>> +               flags |= IOVMF_DA_ANON;
>>>>
>>>> could we use only one? both are mutual exclusive, what happen if flag
>>>> is IOVMF_DA_FIXED | IOVMF_DA_ANON? so, I suggest to get rid of
>>>> IOVMF_DA_ANON.
>>>
>>> Then, what about introducing some MACRO? Better names?
>>>
>>> #define set_iovmf_da_anon(flags)
>>> #define set_iovmf_da_fix(flags)
>>> #define set_iovmf_mmio(flags)
>>
>> will they be used by the users?
>>
>> I think people are more used to use
>>
>> iommu_vmap(obj, da, sgt, IOVMF_MMIO | IOVMF_DA_ANON);
>
> I'd be happier with this approach, instead of the macros. :)
> It's intuitive and very common on kernel.
>
>>
>> than
>>
>> set_iovmf_da_anon(flags)
>> set_iovmf_mmio(flags)
>> iommu_vmap(obj, da, sgt, flags);
>>
>> I don't have problem with the change, but I think how it is now is ok,
>> just that we don't we two bits to handle anon/fixed da, it can be
>> managed it only 1 bit (one flag), or is there a issue?
>
> We can exclude IOVMF_DA_ANON and stick with IOVMF_DA_FIXED only.
> I can resend my patch if we agree it's OK.

sounds perfect to me.

Regards,
Fernando.

>
> Regards,
>
> David
>
>>
>> Regards,
>> Fernando.
>>> ......
>>>
>>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
David Cohen March 8, 2011, 7:46 p.m. UTC | #6 
[snip]

>>>>>> -       flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
>>>>>> +       if (~flags & IOVMF_DA_FIXED)
>>>>>> +               flags |= IOVMF_DA_ANON;
>>>>>
>>>>> could we use only one? both are mutual exclusive, what happen if flag
>>>>> is IOVMF_DA_FIXED | IOVMF_DA_ANON? so, I suggest to get rid of
>>>>> IOVMF_DA_ANON.
>>>>
>>>> Then, what about introducing some MACRO? Better names?
>>>>
>>>> #define set_iovmf_da_anon(flags)
>>>> #define set_iovmf_da_fix(flags)
>>>> #define set_iovmf_mmio(flags)
>>>
>>> will they be used by the users?
>>>
>>> I think people are more used to use
>>>
>>> iommu_vmap(obj, da, sgt, IOVMF_MMIO | IOVMF_DA_ANON);
>>
>> I'd be happier with this approach, instead of the macros. :)
>> It's intuitive and very common on kernel.
>>
>>>
>>> than
>>>
>>> set_iovmf_da_anon(flags)
>>> set_iovmf_mmio(flags)
>>> iommu_vmap(obj, da, sgt, flags);
>>>
>>> I don't have problem with the change, but I think how it is now is ok,
>>> just that we don't we two bits to handle anon/fixed da, it can be
>>> managed it only 1 bit (one flag), or is there a issue?
>>
>> We can exclude IOVMF_DA_ANON and stick with IOVMF_DA_FIXED only.
>> I can resend my patch if we agree it's OK.
>
> sounds perfect to me.

Not sure indeed if this change fits to this same patch. Looks like a
4th patch sounds better.

Br,

David Cohen
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
David Cohen March 8, 2011, 7:51 p.m. UTC | #7 
On Tue, Mar 8, 2011 at 9:46 PM, David Cohen <dacohen@gmail.com> wrote:
> [snip]
>
>>>>>>> -       flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
>>>>>>> +       if (~flags & IOVMF_DA_FIXED)
>>>>>>> +               flags |= IOVMF_DA_ANON;
>>>>>>
>>>>>> could we use only one? both are mutual exclusive, what happen if flag
>>>>>> is IOVMF_DA_FIXED | IOVMF_DA_ANON? so, I suggest to get rid of
>>>>>> IOVMF_DA_ANON.
>>>>>
>>>>> Then, what about introducing some MACRO? Better names?
>>>>>
>>>>> #define set_iovmf_da_anon(flags)
>>>>> #define set_iovmf_da_fix(flags)
>>>>> #define set_iovmf_mmio(flags)
>>>>
>>>> will they be used by the users?
>>>>
>>>> I think people are more used to use
>>>>
>>>> iommu_vmap(obj, da, sgt, IOVMF_MMIO | IOVMF_DA_ANON);
>>>
>>> I'd be happier with this approach, instead of the macros. :)
>>> It's intuitive and very common on kernel.
>>>
>>>>
>>>> than
>>>>
>>>> set_iovmf_da_anon(flags)
>>>> set_iovmf_mmio(flags)
>>>> iommu_vmap(obj, da, sgt, flags);
>>>>
>>>> I don't have problem with the change, but I think how it is now is ok,
>>>> just that we don't we two bits to handle anon/fixed da, it can be
>>>> managed it only 1 bit (one flag), or is there a issue?
>>>
>>> We can exclude IOVMF_DA_ANON and stick with IOVMF_DA_FIXED only.
>>> I can resend my patch if we agree it's OK.
>>
>> sounds perfect to me.
>
> Not sure indeed if this change fits to this same patch. Looks like a
> 4th patch sounds better.

Indeed not. :)
A new set is coming soon.

Br,

David

>
> Br,
>
> David Cohen
>
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/arch/arm/plat-omap/iovmm.c b/arch/arm/plat-omap/iovmm.c
index 11c9b76..dde9cb0 100644
--- a/arch/arm/plat-omap/iovmm.c
+++ b/arch/arm/plat-omap/iovmm.c
@@ -654,7 +654,8 @@  u32 iommu_vmap(struct iommu *obj, u32 da, const struct sg_table *sgt,
 	flags &= IOVMF_HW_MASK;
 	flags |= IOVMF_DISCONT;
 	flags |= IOVMF_MMIO;
-	flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
+	if (~flags & IOVMF_DA_FIXED)
+		flags |= IOVMF_DA_ANON;
 
 	da = __iommu_vmap(obj, da, sgt, va, bytes, flags);
 	if (IS_ERR_VALUE(da))
@@ -713,7 +714,8 @@  u32 iommu_vmalloc(struct iommu *obj, u32 da, size_t bytes, u32 flags)
 	flags &= IOVMF_HW_MASK;
 	flags |= IOVMF_DISCONT;
 	flags |= IOVMF_ALLOC;
-	flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
+	if (~flags & IOVMF_DA_FIXED)
+		flags |= IOVMF_DA_ANON;
 
 	sgt = sgtable_alloc(bytes, flags, da, 0);
 	if (IS_ERR(sgt)) {
@@ -803,7 +805,8 @@  u32 iommu_kmap(struct iommu *obj, u32 da, u32 pa, size_t bytes,
 	flags &= IOVMF_HW_MASK;
 	flags |= IOVMF_LINEAR;
 	flags |= IOVMF_MMIO;
-	flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
+	if (~flags & IOVMF_DA_FIXED)
+		flags |= IOVMF_DA_ANON;
 
 	da = __iommu_kmap(obj, da, pa, va, bytes, flags);
 	if (IS_ERR_VALUE(da))
@@ -862,7 +865,8 @@  u32 iommu_kmalloc(struct iommu *obj, u32 da, size_t bytes, u32 flags)
 	flags &= IOVMF_HW_MASK;
 	flags |= IOVMF_LINEAR;
 	flags |= IOVMF_ALLOC;
-	flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON);
+	if (~flags & IOVMF_DA_FIXED)
+		flags |= IOVMF_DA_ANON;
 
 	da = __iommu_kmap(obj, da, pa, va, bytes, flags);
 	if (IS_ERR_VALUE(da))