* Re: [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV region as boot services data
2018-01-02 15:50 [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV region as boot services data Ard Biesheuvel
@ 2018-01-02 20:42 ` Vladimir Olovyannikov
2018-01-03 5:09 ` Udit Kumar
2018-02-28 15:03 ` Leif Lindholm
2 siblings, 0 replies; 7+ messages in thread
From: Vladimir Olovyannikov @ 2018-01-02 20:42 UTC (permalink / raw)
To: Ard Biesheuvel, edk2-devel; +Cc: leif.lindholm, udit.kumar, meenakshi.aggarwal
> -----Original Message-----
> From: Ard Biesheuvel [mailto:ard.biesheuvel@linaro.org]
> Sent: Tuesday, January 2, 2018 7:51 AM
> To: edk2-devel@lists.01.org
> Cc: leif.lindholm@linaro.org; vladimir.olovyannikov@broadcom.com;
> udit.kumar@nxp.com; meenakshi.aggarwal@nxp.com; Ard Biesheuvel
> Subject: [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV
> region as boot services data
>
> Commit 8ae5fc182941 ("ArmPlatformPkg/MemoryInitPeiLib: don't reserve
> primary FV in memory") deleted the code that removes the memory
> covering the primary firmware volume from the memory map. The
> assumption was that this is no longer necessary now that we no longer
> expose compression and PE/COFF loader library code from the PrePi module
> to DXE core.
>
> However, the FV is still declared, and so code may attempt to access it
> anyway, which may cause unexpected results depending on whether the
> memory has been reused for other purposes in the mean time.
>
> So reinstate the code that splits off the resource descriptor HOB that
> describes the firmware device, but this time, don't mark the memory as
> unusable, but create a memory allocation HOB that marks the region as
boot
> services data.
>
> Contributed-under: TianoCore Contribution Agreement 1.1
> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
> ---
> Vladimir, Udit, Meenakshi: please confirm whether this code works for
you.
Thanks Ard,
Works for me.
Vladimir
>
> ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c | 74
> ++++++++++++++++++++
> 1 file changed, 74 insertions(+)
>
> diff --git a/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
> b/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
> index d03214b5df66..d1b5c0be9497 100644
> --- a/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
> +++ b/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
> @@ -70,7 +70,11 @@ MemoryPeim (
> {
> ARM_MEMORY_REGION_DESCRIPTOR *MemoryTable;
> EFI_RESOURCE_ATTRIBUTE_TYPE ResourceAttributes;
> + UINT64 ResourceLength;
> EFI_PEI_HOB_POINTERS NextHob;
> + EFI_PHYSICAL_ADDRESS FdTop;
> + EFI_PHYSICAL_ADDRESS SystemMemoryTop;
> + EFI_PHYSICAL_ADDRESS ResourceTop;
> BOOLEAN Found;
>
> // Get Virtual Memory Map from the Platform Library @@ -117,6 +121,76
> @@ MemoryPeim (
> );
> }
>
> + //
> + // Reserve the memory space occupied by the firmware volume //
> +
> + SystemMemoryTop = (EFI_PHYSICAL_ADDRESS)PcdGet64
> + (PcdSystemMemoryBase) + (EFI_PHYSICAL_ADDRESS)PcdGet64
> + (PcdSystemMemorySize); FdTop = (EFI_PHYSICAL_ADDRESS)PcdGet64
> + (PcdFdBaseAddress) + (EFI_PHYSICAL_ADDRESS)PcdGet32 (PcdFdSize);
> +
> + // EDK2 does not have the concept of boot firmware copied into DRAM.
> + To avoid the DXE // core to overwrite this area we must create a
> + memory allocation HOB for the region, // but this only works if we
split off
> the underlying resource descriptor as well.
> + if ((PcdGet64 (PcdFdBaseAddress) >= PcdGet64
> (PcdSystemMemoryBase)) && (FdTop <= SystemMemoryTop)) {
> + Found = FALSE;
> +
> + // Search for System Memory Hob that contains the firmware
> + NextHob.Raw = GetHobList ();
> + while ((NextHob.Raw = GetNextHob
> (EFI_HOB_TYPE_RESOURCE_DESCRIPTOR, NextHob.Raw)) != NULL) {
> + if ((NextHob.ResourceDescriptor->ResourceType ==
> EFI_RESOURCE_SYSTEM_MEMORY) &&
> + (PcdGet64 (PcdFdBaseAddress) >= NextHob.ResourceDescriptor-
> >PhysicalStart) &&
> + (FdTop <= NextHob.ResourceDescriptor->PhysicalStart +
> NextHob.ResourceDescriptor->ResourceLength))
> + {
> + ResourceAttributes = NextHob.ResourceDescriptor-
> >ResourceAttribute;
> + ResourceLength = NextHob.ResourceDescriptor->ResourceLength;
> + ResourceTop = NextHob.ResourceDescriptor->PhysicalStart +
> + ResourceLength;
> +
> + if (PcdGet64 (PcdFdBaseAddress) == NextHob.ResourceDescriptor-
> >PhysicalStart) {
> + if (SystemMemoryTop != FdTop) {
> + // Create the System Memory HOB for the firmware with the
non-
> present attribute
> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
> + ResourceAttributes,
> + PcdGet64 (PcdFdBaseAddress),
> + PcdGet32 (PcdFdSize));
> +
> + // Top of the FD is system memory available for UEFI
> + NextHob.ResourceDescriptor->PhysicalStart +=
PcdGet32(PcdFdSize);
> + NextHob.ResourceDescriptor->ResourceLength -=
> PcdGet32(PcdFdSize);
> + }
> + } else {
> + // Create the System Memory HOB for the firmware
> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
> + ResourceAttributes,
> + PcdGet64 (PcdFdBaseAddress),
> + PcdGet32 (PcdFdSize));
> +
> + // Update the HOB
> + NextHob.ResourceDescriptor->ResourceLength = PcdGet64
> + (PcdFdBaseAddress) - NextHob.ResourceDescriptor->PhysicalStart;
> +
> + // If there is some memory available on the top of the FD
then create a
> HOB
> + if (FdTop < NextHob.ResourceDescriptor->PhysicalStart +
> ResourceLength) {
> + // Create the System Memory HOB for the remaining region
(top of
> the FD)
> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
> + ResourceAttributes,
> + FdTop,
> + ResourceTop - FdTop);
> + }
> + }
> +
> + // Mark the memory covering the Firmware Device as boot
services
> data
> + BuildMemoryAllocationHob (PcdGet64 (PcdFdBaseAddress),
> + PcdGet32 (PcdFdSize),
> + EfiBootServicesData);
> +
> + Found = TRUE;
> + break;
> + }
> + NextHob.Raw = GET_NEXT_HOB (NextHob);
> + }
> +
> + ASSERT(Found);
> + }
> +
> // Build Memory Allocation Hob
> InitMmu (MemoryTable);
>
> --
> 2.11.0
^ permalink raw reply [flat|nested] 7+ messages in thread* Re: [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV region as boot services data
2018-01-02 15:50 [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV region as boot services data Ard Biesheuvel
2018-01-02 20:42 ` Vladimir Olovyannikov
@ 2018-01-03 5:09 ` Udit Kumar
2018-01-03 7:44 ` Ard Biesheuvel
2018-02-28 15:03 ` Leif Lindholm
2 siblings, 1 reply; 7+ messages in thread
From: Udit Kumar @ 2018-01-03 5:09 UTC (permalink / raw)
To: Ard Biesheuvel, edk2-devel@lists.01.org
Cc: leif.lindholm@linaro.org, vladimir.olovyannikov@broadcom.com,
Meenakshi Aggarwal
Thanks Ard,
This works for us as well
Few comments inline
Regards
Udit
> -----Original Message-----
> From: Ard Biesheuvel [mailto:ard.biesheuvel@linaro.org]
> Sent: Tuesday, January 02, 2018 9:21 PM
> To: edk2-devel@lists.01.org
> Cc: leif.lindholm@linaro.org; vladimir.olovyannikov@broadcom.com; Udit
> Kumar <udit.kumar@nxp.com>; Meenakshi Aggarwal
> <meenakshi.aggarwal@nxp.com>; Ard Biesheuvel
> <ard.biesheuvel@linaro.org>
> Subject: [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV region
> as boot services data
>
> Commit 8ae5fc182941 ("ArmPlatformPkg/MemoryInitPeiLib: don't reserve
> primary FV in memory") deleted the code that removes the memory covering
> the primary firmware volume from the memory map. The assumption was
> that
> this is no longer necessary now that we no longer expose compression and
> PE/COFF loader library code from the PrePi module to DXE core.
>
> However, the FV is still declared, and so code may attempt to access it
> anyway, which may cause unexpected results depending on whether the
> memory has been reused for other purposes in the mean time.
>
> So reinstate the code that splits off the resource descriptor HOB that
> describes the firmware device, but this time, don't mark the memory as
> unusable, but create a memory allocation HOB that marks the region as
> boot services data.
>
> Contributed-under: TianoCore Contribution Agreement 1.1
> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
> ---
> Vladimir, Udit, Meenakshi: please confirm whether this code works for you.
>
> ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c | 74
> ++++++++++++++++++++
> 1 file changed, 74 insertions(+)
>
> diff --git a/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
> b/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
> index d03214b5df66..d1b5c0be9497 100644
> --- a/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
> +++ b/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
> @@ -70,7 +70,11 @@ MemoryPeim (
> {
> ARM_MEMORY_REGION_DESCRIPTOR *MemoryTable;
> EFI_RESOURCE_ATTRIBUTE_TYPE ResourceAttributes;
> + UINT64 ResourceLength;
> EFI_PEI_HOB_POINTERS NextHob;
> + EFI_PHYSICAL_ADDRESS FdTop;
> + EFI_PHYSICAL_ADDRESS SystemMemoryTop;
> + EFI_PHYSICAL_ADDRESS ResourceTop;
> BOOLEAN Found;
>
> // Get Virtual Memory Map from the Platform Library
> @@ -117,6 +121,76 @@ MemoryPeim (
> );
> }
>
> + //
> + // Reserve the memory space occupied by the firmware volume
> + //
> +
> + SystemMemoryTop = (EFI_PHYSICAL_ADDRESS)PcdGet64
> (PcdSystemMemoryBase) + (EFI_PHYSICAL_ADDRESS)PcdGet64
> (PcdSystemMemorySize);
> + FdTop = (EFI_PHYSICAL_ADDRESS)PcdGet64 (PcdFdBaseAddress) +
> (EFI_PHYSICAL_ADDRESS)PcdGet32 (PcdFdSize);
> +
> + // EDK2 does not have the concept of boot firmware copied into DRAM. To
> avoid the DXE
> + // core to overwrite this area we must create a memory allocation HOB for
> the region,
> + // but this only works if we split off the underlying resource descriptor as
> well.
> + if ((PcdGet64 (PcdFdBaseAddress) >= PcdGet64 (PcdSystemMemoryBase))
> && (FdTop <= SystemMemoryTop)) {
> + Found = FALSE;
> +
> + // Search for System Memory Hob that contains the firmware
> + NextHob.Raw = GetHobList ();
> + while ((NextHob.Raw = GetNextHob
> (EFI_HOB_TYPE_RESOURCE_DESCRIPTOR, NextHob.Raw)) != NULL) {
> + if ((NextHob.ResourceDescriptor->ResourceType ==
> EFI_RESOURCE_SYSTEM_MEMORY) &&
> + (PcdGet64 (PcdFdBaseAddress) >= NextHob.ResourceDescriptor-
> >PhysicalStart) &&
> + (FdTop <= NextHob.ResourceDescriptor->PhysicalStart +
> NextHob.ResourceDescriptor->ResourceLength))
> + {
> + ResourceAttributes = NextHob.ResourceDescriptor->ResourceAttribute;
> + ResourceLength = NextHob.ResourceDescriptor->ResourceLength;
> + ResourceTop = NextHob.ResourceDescriptor->PhysicalStart +
> ResourceLength;
> +
> + if (PcdGet64 (PcdFdBaseAddress) == NextHob.ResourceDescriptor-
> >PhysicalStart) {
> + if (SystemMemoryTop != FdTop) {
> + // Create the System Memory HOB for the firmware with the non-
> present attribute
Please correct comments, now this memory is present
> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
> + ResourceAttributes,
> + PcdGet64 (PcdFdBaseAddress),
> + PcdGet32 (PcdFdSize));
> +
> + // Top of the FD is system memory available for UEFI
> + NextHob.ResourceDescriptor->PhysicalStart += PcdGet32(PcdFdSize);
> + NextHob.ResourceDescriptor->ResourceLength -=
> PcdGet32(PcdFdSize);
> + }
> + } else {
> + // Create the System Memory HOB for the firmware
> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
> + ResourceAttributes,
> + PcdGet64 (PcdFdBaseAddress),
> + PcdGet32 (PcdFdSize));
Hob List is already created for PcdSystemMemoryBase and its size
Within this, we got Fd, then do we want to create another Hob here
> +
> + // Update the HOB
> + NextHob.ResourceDescriptor->ResourceLength = PcdGet64
> (PcdFdBaseAddress) - NextHob.ResourceDescriptor->PhysicalStart;
> +
> + // If there is some memory available on the top of the FD then create
> a HOB
> + if (FdTop < NextHob.ResourceDescriptor->PhysicalStart +
> ResourceLength) {
> + // Create the System Memory HOB for the remaining region (top of
> the FD)
> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
> + ResourceAttributes,
> + FdTop,
> + ResourceTop - FdTop);
> + }
> + }
> +
> + // Mark the memory covering the Firmware Device as boot services
> data
> + BuildMemoryAllocationHob (PcdGet64 (PcdFdBaseAddress),
> + PcdGet32 (PcdFdSize),
> + EfiBootServicesData);
IMO, only this call should be enough to protect FD area.
> +
> + Found = TRUE;
> + break;
> + }
> + NextHob.Raw = GET_NEXT_HOB (NextHob);
> + }
> +
> + ASSERT(Found);
> + }
> +
> // Build Memory Allocation Hob
> InitMmu (MemoryTable);
>
> --
> 2.11.0
^ permalink raw reply [flat|nested] 7+ messages in thread* Re: [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV region as boot services data
2018-01-03 5:09 ` Udit Kumar
@ 2018-01-03 7:44 ` Ard Biesheuvel
2018-02-01 11:41 ` Ard Biesheuvel
0 siblings, 1 reply; 7+ messages in thread
From: Ard Biesheuvel @ 2018-01-03 7:44 UTC (permalink / raw)
To: Udit Kumar
Cc: edk2-devel@lists.01.org, leif.lindholm@linaro.org,
vladimir.olovyannikov@broadcom.com, Meenakshi Aggarwal
On 3 January 2018 at 05:09, Udit Kumar <udit.kumar@nxp.com> wrote:
> Thanks Ard,
> This works for us as well
> Few comments inline
>
>
> Regards
> Udit
>
>> -----Original Message-----
>> From: Ard Biesheuvel [mailto:ard.biesheuvel@linaro.org]
>> Sent: Tuesday, January 02, 2018 9:21 PM
>> To: edk2-devel@lists.01.org
>> Cc: leif.lindholm@linaro.org; vladimir.olovyannikov@broadcom.com; Udit
>> Kumar <udit.kumar@nxp.com>; Meenakshi Aggarwal
>> <meenakshi.aggarwal@nxp.com>; Ard Biesheuvel
>> <ard.biesheuvel@linaro.org>
>> Subject: [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV region
>> as boot services data
>>
>> Commit 8ae5fc182941 ("ArmPlatformPkg/MemoryInitPeiLib: don't reserve
>> primary FV in memory") deleted the code that removes the memory covering
>> the primary firmware volume from the memory map. The assumption was
>> that
>> this is no longer necessary now that we no longer expose compression and
>> PE/COFF loader library code from the PrePi module to DXE core.
>>
>> However, the FV is still declared, and so code may attempt to access it
>> anyway, which may cause unexpected results depending on whether the
>> memory has been reused for other purposes in the mean time.
>>
>> So reinstate the code that splits off the resource descriptor HOB that
>> describes the firmware device, but this time, don't mark the memory as
>> unusable, but create a memory allocation HOB that marks the region as
>> boot services data.
>>
>> Contributed-under: TianoCore Contribution Agreement 1.1
>> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
>> ---
>> Vladimir, Udit, Meenakshi: please confirm whether this code works for you.
>>
>> ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c | 74
>> ++++++++++++++++++++
>> 1 file changed, 74 insertions(+)
>>
>> diff --git a/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
>> b/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
>> index d03214b5df66..d1b5c0be9497 100644
>> --- a/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
>> +++ b/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
>> @@ -70,7 +70,11 @@ MemoryPeim (
>> {
>> ARM_MEMORY_REGION_DESCRIPTOR *MemoryTable;
>> EFI_RESOURCE_ATTRIBUTE_TYPE ResourceAttributes;
>> + UINT64 ResourceLength;
>> EFI_PEI_HOB_POINTERS NextHob;
>> + EFI_PHYSICAL_ADDRESS FdTop;
>> + EFI_PHYSICAL_ADDRESS SystemMemoryTop;
>> + EFI_PHYSICAL_ADDRESS ResourceTop;
>> BOOLEAN Found;
>>
>> // Get Virtual Memory Map from the Platform Library
>> @@ -117,6 +121,76 @@ MemoryPeim (
>> );
>> }
>>
>> + //
>> + // Reserve the memory space occupied by the firmware volume
>> + //
>> +
>> + SystemMemoryTop = (EFI_PHYSICAL_ADDRESS)PcdGet64
>> (PcdSystemMemoryBase) + (EFI_PHYSICAL_ADDRESS)PcdGet64
>> (PcdSystemMemorySize);
>> + FdTop = (EFI_PHYSICAL_ADDRESS)PcdGet64 (PcdFdBaseAddress) +
>> (EFI_PHYSICAL_ADDRESS)PcdGet32 (PcdFdSize);
>> +
>> + // EDK2 does not have the concept of boot firmware copied into DRAM. To
>> avoid the DXE
>> + // core to overwrite this area we must create a memory allocation HOB for
>> the region,
>> + // but this only works if we split off the underlying resource descriptor as
>> well.
>> + if ((PcdGet64 (PcdFdBaseAddress) >= PcdGet64 (PcdSystemMemoryBase))
>> && (FdTop <= SystemMemoryTop)) {
>> + Found = FALSE;
>> +
>> + // Search for System Memory Hob that contains the firmware
>> + NextHob.Raw = GetHobList ();
>> + while ((NextHob.Raw = GetNextHob
>> (EFI_HOB_TYPE_RESOURCE_DESCRIPTOR, NextHob.Raw)) != NULL) {
>> + if ((NextHob.ResourceDescriptor->ResourceType ==
>> EFI_RESOURCE_SYSTEM_MEMORY) &&
>> + (PcdGet64 (PcdFdBaseAddress) >= NextHob.ResourceDescriptor-
>> >PhysicalStart) &&
>> + (FdTop <= NextHob.ResourceDescriptor->PhysicalStart +
>> NextHob.ResourceDescriptor->ResourceLength))
>> + {
>> + ResourceAttributes = NextHob.ResourceDescriptor->ResourceAttribute;
>> + ResourceLength = NextHob.ResourceDescriptor->ResourceLength;
>> + ResourceTop = NextHob.ResourceDescriptor->PhysicalStart +
>> ResourceLength;
>> +
>> + if (PcdGet64 (PcdFdBaseAddress) == NextHob.ResourceDescriptor-
>> >PhysicalStart) {
>> + if (SystemMemoryTop != FdTop) {
>> + // Create the System Memory HOB for the firmware with the non-
>> present attribute
>
> Please correct comments, now this memory is present
>
Yes
>> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
>> + ResourceAttributes,
>> + PcdGet64 (PcdFdBaseAddress),
>> + PcdGet32 (PcdFdSize));
>> +
>> + // Top of the FD is system memory available for UEFI
>> + NextHob.ResourceDescriptor->PhysicalStart += PcdGet32(PcdFdSize);
>> + NextHob.ResourceDescriptor->ResourceLength -=
>> PcdGet32(PcdFdSize);
>> + }
>> + } else {
>> + // Create the System Memory HOB for the firmware
>> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
>> + ResourceAttributes,
>> + PcdGet64 (PcdFdBaseAddress),
>> + PcdGet32 (PcdFdSize));
>
> Hob List is already created for PcdSystemMemoryBase and its size
> Within this, we got Fd, then do we want to create another Hob here
>
The resource descriptor for PcdSystemMemoryBase/Size is updated in the
next line so it no longer covers the FD
>> +
>> + // Update the HOB
>> + NextHob.ResourceDescriptor->ResourceLength = PcdGet64
>> (PcdFdBaseAddress) - NextHob.ResourceDescriptor->PhysicalStart;
>> +
>> + // If there is some memory available on the top of the FD then create
>> a HOB
>> + if (FdTop < NextHob.ResourceDescriptor->PhysicalStart +
>> ResourceLength) {
>> + // Create the System Memory HOB for the remaining region (top of
>> the FD)
>> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
>> + ResourceAttributes,
>> + FdTop,
>> + ResourceTop - FdTop);
>> + }
>> + }
>> +
>> + // Mark the memory covering the Firmware Device as boot services
>> data
>> + BuildMemoryAllocationHob (PcdGet64 (PcdFdBaseAddress),
>> + PcdGet32 (PcdFdSize),
>> + EfiBootServicesData);
>
> IMO, only this call should be enough to protect FD area.
>
I agree, but the reality is that it is not enough. in
CoreInitializeGcdServices, the first system memory resource descriptor
is claimed before the memory allocations are processed, and any memory
allocation HOBs that intersect that region are ignored.
>> +
>> + Found = TRUE;
>> + break;
>> + }
>> + NextHob.Raw = GET_NEXT_HOB (NextHob);
>> + }
>> +
>> + ASSERT(Found);
>> + }
>> +
>> // Build Memory Allocation Hob
>> InitMmu (MemoryTable);
>>
>> --
>> 2.11.0
>
^ permalink raw reply [flat|nested] 7+ messages in thread* Re: [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV region as boot services data
2018-01-03 7:44 ` Ard Biesheuvel
@ 2018-02-01 11:41 ` Ard Biesheuvel
0 siblings, 0 replies; 7+ messages in thread
From: Ard Biesheuvel @ 2018-02-01 11:41 UTC (permalink / raw)
To: Udit Kumar, leif.lindholm@linaro.org
Cc: edk2-devel@lists.01.org, vladimir.olovyannikov@broadcom.com,
Meenakshi Aggarwal
Leif,
On 3 January 2018 at 07:44, Ard Biesheuvel <ard.biesheuvel@linaro.org> wrote:
> On 3 January 2018 at 05:09, Udit Kumar <udit.kumar@nxp.com> wrote:
>> Thanks Ard,
>> This works for us as well
>> Few comments inline
>>
>>
>> Regards
>> Udit
>>
>>> -----Original Message-----
>>> From: Ard Biesheuvel [mailto:ard.biesheuvel@linaro.org]
>>> Sent: Tuesday, January 02, 2018 9:21 PM
>>> To: edk2-devel@lists.01.org
>>> Cc: leif.lindholm@linaro.org; vladimir.olovyannikov@broadcom.com; Udit
>>> Kumar <udit.kumar@nxp.com>; Meenakshi Aggarwal
>>> <meenakshi.aggarwal@nxp.com>; Ard Biesheuvel
>>> <ard.biesheuvel@linaro.org>
>>> Subject: [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV region
>>> as boot services data
>>>
>>> Commit 8ae5fc182941 ("ArmPlatformPkg/MemoryInitPeiLib: don't reserve
>>> primary FV in memory") deleted the code that removes the memory covering
>>> the primary firmware volume from the memory map. The assumption was
>>> that
>>> this is no longer necessary now that we no longer expose compression and
>>> PE/COFF loader library code from the PrePi module to DXE core.
>>>
>>> However, the FV is still declared, and so code may attempt to access it
>>> anyway, which may cause unexpected results depending on whether the
>>> memory has been reused for other purposes in the mean time.
>>>
>>> So reinstate the code that splits off the resource descriptor HOB that
>>> describes the firmware device, but this time, don't mark the memory as
>>> unusable, but create a memory allocation HOB that marks the region as
>>> boot services data.
>>>
>>> Contributed-under: TianoCore Contribution Agreement 1.1
>>> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
>>> ---
>>> Vladimir, Udit, Meenakshi: please confirm whether this code works for you.
>>>
>>> ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c | 74
>>> ++++++++++++++++++++
>>> 1 file changed, 74 insertions(+)
>>>
>>> diff --git a/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
>>> b/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
>>> index d03214b5df66..d1b5c0be9497 100644
>>> --- a/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
>>> +++ b/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
>>> @@ -70,7 +70,11 @@ MemoryPeim (
>>> {
>>> ARM_MEMORY_REGION_DESCRIPTOR *MemoryTable;
>>> EFI_RESOURCE_ATTRIBUTE_TYPE ResourceAttributes;
>>> + UINT64 ResourceLength;
>>> EFI_PEI_HOB_POINTERS NextHob;
>>> + EFI_PHYSICAL_ADDRESS FdTop;
>>> + EFI_PHYSICAL_ADDRESS SystemMemoryTop;
>>> + EFI_PHYSICAL_ADDRESS ResourceTop;
>>> BOOLEAN Found;
>>>
>>> // Get Virtual Memory Map from the Platform Library
>>> @@ -117,6 +121,76 @@ MemoryPeim (
>>> );
>>> }
>>>
>>> + //
>>> + // Reserve the memory space occupied by the firmware volume
>>> + //
>>> +
>>> + SystemMemoryTop = (EFI_PHYSICAL_ADDRESS)PcdGet64
>>> (PcdSystemMemoryBase) + (EFI_PHYSICAL_ADDRESS)PcdGet64
>>> (PcdSystemMemorySize);
>>> + FdTop = (EFI_PHYSICAL_ADDRESS)PcdGet64 (PcdFdBaseAddress) +
>>> (EFI_PHYSICAL_ADDRESS)PcdGet32 (PcdFdSize);
>>> +
>>> + // EDK2 does not have the concept of boot firmware copied into DRAM. To
>>> avoid the DXE
>>> + // core to overwrite this area we must create a memory allocation HOB for
>>> the region,
>>> + // but this only works if we split off the underlying resource descriptor as
>>> well.
>>> + if ((PcdGet64 (PcdFdBaseAddress) >= PcdGet64 (PcdSystemMemoryBase))
>>> && (FdTop <= SystemMemoryTop)) {
>>> + Found = FALSE;
>>> +
>>> + // Search for System Memory Hob that contains the firmware
>>> + NextHob.Raw = GetHobList ();
>>> + while ((NextHob.Raw = GetNextHob
>>> (EFI_HOB_TYPE_RESOURCE_DESCRIPTOR, NextHob.Raw)) != NULL) {
>>> + if ((NextHob.ResourceDescriptor->ResourceType ==
>>> EFI_RESOURCE_SYSTEM_MEMORY) &&
>>> + (PcdGet64 (PcdFdBaseAddress) >= NextHob.ResourceDescriptor-
>>> >PhysicalStart) &&
>>> + (FdTop <= NextHob.ResourceDescriptor->PhysicalStart +
>>> NextHob.ResourceDescriptor->ResourceLength))
>>> + {
>>> + ResourceAttributes = NextHob.ResourceDescriptor->ResourceAttribute;
>>> + ResourceLength = NextHob.ResourceDescriptor->ResourceLength;
>>> + ResourceTop = NextHob.ResourceDescriptor->PhysicalStart +
>>> ResourceLength;
>>> +
>>> + if (PcdGet64 (PcdFdBaseAddress) == NextHob.ResourceDescriptor-
>>> >PhysicalStart) {
>>> + if (SystemMemoryTop != FdTop) {
>>> + // Create the System Memory HOB for the firmware with the non-
>>> present attribute
>>
>> Please correct comments, now this memory is present
>>
>
> Yes
>
>>> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
>>> + ResourceAttributes,
>>> + PcdGet64 (PcdFdBaseAddress),
>>> + PcdGet32 (PcdFdSize));
>>> +
>>> + // Top of the FD is system memory available for UEFI
>>> + NextHob.ResourceDescriptor->PhysicalStart += PcdGet32(PcdFdSize);
>>> + NextHob.ResourceDescriptor->ResourceLength -=
>>> PcdGet32(PcdFdSize);
>>> + }
>>> + } else {
>>> + // Create the System Memory HOB for the firmware
>>> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
>>> + ResourceAttributes,
>>> + PcdGet64 (PcdFdBaseAddress),
>>> + PcdGet32 (PcdFdSize));
>>
>> Hob List is already created for PcdSystemMemoryBase and its size
>> Within this, we got Fd, then do we want to create another Hob here
>>
>
> The resource descriptor for PcdSystemMemoryBase/Size is updated in the
> next line so it no longer covers the FD
>
>>> +
>>> + // Update the HOB
>>> + NextHob.ResourceDescriptor->ResourceLength = PcdGet64
>>> (PcdFdBaseAddress) - NextHob.ResourceDescriptor->PhysicalStart;
>>> +
>>> + // If there is some memory available on the top of the FD then create
>>> a HOB
>>> + if (FdTop < NextHob.ResourceDescriptor->PhysicalStart +
>>> ResourceLength) {
>>> + // Create the System Memory HOB for the remaining region (top of
>>> the FD)
>>> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
>>> + ResourceAttributes,
>>> + FdTop,
>>> + ResourceTop - FdTop);
>>> + }
>>> + }
>>> +
>>> + // Mark the memory covering the Firmware Device as boot services
>>> data
>>> + BuildMemoryAllocationHob (PcdGet64 (PcdFdBaseAddress),
>>> + PcdGet32 (PcdFdSize),
>>> + EfiBootServicesData);
>>
>> IMO, only this call should be enough to protect FD area.
>>
>
> I agree, but the reality is that it is not enough. in
> CoreInitializeGcdServices, the first system memory resource descriptor
> is claimed before the memory allocations are processed, and any memory
> allocation HOBs that intersect that region are ignored.
>
>>> +
>>> + Found = TRUE;
>>> + break;
>>> + }
>>> + NextHob.Raw = GET_NEXT_HOB (NextHob);
>>> + }
>>> +
>>> + ASSERT(Found);
>>> + }
>>> +
>>> // Build Memory Allocation Hob
>>> InitMmu (MemoryTable);
>>>
>>> --
>>> 2.11.0
>>
Any thoughts? I can resend the patch if you like.
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV region as boot services data
2018-01-02 15:50 [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV region as boot services data Ard Biesheuvel
2018-01-02 20:42 ` Vladimir Olovyannikov
2018-01-03 5:09 ` Udit Kumar
@ 2018-02-28 15:03 ` Leif Lindholm
2018-02-28 15:03 ` Ard Biesheuvel
2 siblings, 1 reply; 7+ messages in thread
From: Leif Lindholm @ 2018-02-28 15:03 UTC (permalink / raw)
To: Ard Biesheuvel
Cc: edk2-devel, vladimir.olovyannikov, udit.kumar, meenakshi.aggarwal
On Tue, Jan 02, 2018 at 03:50:34PM +0000, Ard Biesheuvel wrote:
> Commit 8ae5fc182941 ("ArmPlatformPkg/MemoryInitPeiLib: don't reserve
> primary FV in memory") deleted the code that removes the memory covering
> the primary firmware volume from the memory map. The assumption was that
> this is no longer necessary now that we no longer expose compression and
> PE/COFF loader library code from the PrePi module to DXE core.
>
> However, the FV is still declared, and so code may attempt to access it
> anyway, which may cause unexpected results depending on whether the
> memory has been reused for other purposes in the mean time.
>
> So reinstate the code that splits off the resource descriptor HOB that
> describes the firmware device, but this time, don't mark the memory as
> unusable, but create a memory allocation HOB that marks the region as
> boot services data.
>
> Contributed-under: TianoCore Contribution Agreement 1.1
> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
So, this looks like a necessary fix.
However, there are many line-length violations below.
And it's not apparent how fixing that would make the code more
readable.
So rather than asking for a rewrite - could you split the revert
from the modification?
/
Leif
> Vladimir, Udit, Meenakshi: please confirm whether this code works for you.
>
> ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c | 74 ++++++++++++++++++++
> 1 file changed, 74 insertions(+)
>
> diff --git a/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c b/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
> index d03214b5df66..d1b5c0be9497 100644
> --- a/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
> +++ b/ArmPlatformPkg/MemoryInitPei/MemoryInitPeiLib.c
> @@ -70,7 +70,11 @@ MemoryPeim (
> {
> ARM_MEMORY_REGION_DESCRIPTOR *MemoryTable;
> EFI_RESOURCE_ATTRIBUTE_TYPE ResourceAttributes;
> + UINT64 ResourceLength;
> EFI_PEI_HOB_POINTERS NextHob;
> + EFI_PHYSICAL_ADDRESS FdTop;
> + EFI_PHYSICAL_ADDRESS SystemMemoryTop;
> + EFI_PHYSICAL_ADDRESS ResourceTop;
> BOOLEAN Found;
>
> // Get Virtual Memory Map from the Platform Library
> @@ -117,6 +121,76 @@ MemoryPeim (
> );
> }
>
> + //
> + // Reserve the memory space occupied by the firmware volume
> + //
> +
> + SystemMemoryTop = (EFI_PHYSICAL_ADDRESS)PcdGet64 (PcdSystemMemoryBase) + (EFI_PHYSICAL_ADDRESS)PcdGet64 (PcdSystemMemorySize);
> + FdTop = (EFI_PHYSICAL_ADDRESS)PcdGet64 (PcdFdBaseAddress) + (EFI_PHYSICAL_ADDRESS)PcdGet32 (PcdFdSize);
> +
> + // EDK2 does not have the concept of boot firmware copied into DRAM. To avoid the DXE
> + // core to overwrite this area we must create a memory allocation HOB for the region,
> + // but this only works if we split off the underlying resource descriptor as well.
> + if ((PcdGet64 (PcdFdBaseAddress) >= PcdGet64 (PcdSystemMemoryBase)) && (FdTop <= SystemMemoryTop)) {
> + Found = FALSE;
> +
> + // Search for System Memory Hob that contains the firmware
> + NextHob.Raw = GetHobList ();
> + while ((NextHob.Raw = GetNextHob (EFI_HOB_TYPE_RESOURCE_DESCRIPTOR, NextHob.Raw)) != NULL) {
> + if ((NextHob.ResourceDescriptor->ResourceType == EFI_RESOURCE_SYSTEM_MEMORY) &&
> + (PcdGet64 (PcdFdBaseAddress) >= NextHob.ResourceDescriptor->PhysicalStart) &&
> + (FdTop <= NextHob.ResourceDescriptor->PhysicalStart + NextHob.ResourceDescriptor->ResourceLength))
> + {
> + ResourceAttributes = NextHob.ResourceDescriptor->ResourceAttribute;
> + ResourceLength = NextHob.ResourceDescriptor->ResourceLength;
> + ResourceTop = NextHob.ResourceDescriptor->PhysicalStart + ResourceLength;
> +
> + if (PcdGet64 (PcdFdBaseAddress) == NextHob.ResourceDescriptor->PhysicalStart) {
> + if (SystemMemoryTop != FdTop) {
> + // Create the System Memory HOB for the firmware with the non-present attribute
> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
> + ResourceAttributes,
> + PcdGet64 (PcdFdBaseAddress),
> + PcdGet32 (PcdFdSize));
> +
> + // Top of the FD is system memory available for UEFI
> + NextHob.ResourceDescriptor->PhysicalStart += PcdGet32(PcdFdSize);
> + NextHob.ResourceDescriptor->ResourceLength -= PcdGet32(PcdFdSize);
> + }
> + } else {
> + // Create the System Memory HOB for the firmware
> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
> + ResourceAttributes,
> + PcdGet64 (PcdFdBaseAddress),
> + PcdGet32 (PcdFdSize));
> +
> + // Update the HOB
> + NextHob.ResourceDescriptor->ResourceLength = PcdGet64 (PcdFdBaseAddress) - NextHob.ResourceDescriptor->PhysicalStart;
> +
> + // If there is some memory available on the top of the FD then create a HOB
> + if (FdTop < NextHob.ResourceDescriptor->PhysicalStart + ResourceLength) {
> + // Create the System Memory HOB for the remaining region (top of the FD)
> + BuildResourceDescriptorHob (EFI_RESOURCE_SYSTEM_MEMORY,
> + ResourceAttributes,
> + FdTop,
> + ResourceTop - FdTop);
> + }
> + }
> +
> + // Mark the memory covering the Firmware Device as boot services data
> + BuildMemoryAllocationHob (PcdGet64 (PcdFdBaseAddress),
> + PcdGet32 (PcdFdSize),
> + EfiBootServicesData);
> +
> + Found = TRUE;
> + break;
> + }
> + NextHob.Raw = GET_NEXT_HOB (NextHob);
> + }
> +
> + ASSERT(Found);
> + }
> +
> // Build Memory Allocation Hob
> InitMmu (MemoryTable);
>
> --
> 2.11.0
>
^ permalink raw reply [flat|nested] 7+ messages in thread* Re: [PATCH] ArmPlatformPkg/MemoryInitPeiLib: mark primary FV region as boot services data
2018-02-28 15:03 ` Leif Lindholm
@ 2018-02-28 15:03 ` Ard Biesheuvel
0 siblings, 0 replies; 7+ messages in thread
From: Ard Biesheuvel @ 2018-02-28 15:03 UTC (permalink / raw)
To: Leif Lindholm
Cc: edk2-devel@lists.01.org, Vladimir Olovyannikov, Udit Kumar,
Meenakshi Aggarwal
On 28 February 2018 at 15:03, Leif Lindholm <leif.lindholm@linaro.org> wrote:
> On Tue, Jan 02, 2018 at 03:50:34PM +0000, Ard Biesheuvel wrote:
>> Commit 8ae5fc182941 ("ArmPlatformPkg/MemoryInitPeiLib: don't reserve
>> primary FV in memory") deleted the code that removes the memory covering
>> the primary firmware volume from the memory map. The assumption was that
>> this is no longer necessary now that we no longer expose compression and
>> PE/COFF loader library code from the PrePi module to DXE core.
>>
>> However, the FV is still declared, and so code may attempt to access it
>> anyway, which may cause unexpected results depending on whether the
>> memory has been reused for other purposes in the mean time.
>>
>> So reinstate the code that splits off the resource descriptor HOB that
>> describes the firmware device, but this time, don't mark the memory as
>> unusable, but create a memory allocation HOB that marks the region as
>> boot services data.
>>
>> Contributed-under: TianoCore Contribution Agreement 1.1
>> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
>
> So, this looks like a necessary fix.
> However, there are many line-length violations below.
> And it's not apparent how fixing that would make the code more
> readable.
>
> So rather than asking for a rewrite - could you split the revert
> from the modification?
>
Sure.
^ permalink raw reply [flat|nested] 7+ messages in thread