NASM introduced FXSAVE / FXRSTOR support in commit 900fa5b26b8f ("NASM
0.98p3-hpa", 2002-04-30), which commit stands for the nasm-0.98p3-hpa
release.
NASM introduced FXSAVE64 / FXRSTOR64 support in commit 3a014348ca15
("insns: add FXSAVE64/FXRSTOR64, drop np prefix", 2010-07-07), which was
part of the "nasm-2.09" release.
Edk2 requires nasm-2.10 or later for use with the GCC toolchain family,
and nasm-2.12.01 or later for use with all other toolchain families.
Replace the binary encoding of the FXSAVE(64)/FXRSTOR(64) instructions
with mnemonics.
I verified that the "Ia32/SmiException.obj", "X64/SmiEntry.obj" and
"X64/SmiException.obj" files are rebuilt after this patch, without any
change in content.
This patch removes the last instructions encoded with DBs from
PiSmmCpuDxeSmm.
Cc: Eric Dong <eric.dong@intel.com>
Cc: Michael D Kinney <michael.d.kinney@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
(1) SmmRelocationSemaphoreComplete32() runs in 32-bit mode, so wrap it in
a (BITS 32 ... BITS 64) bracket.
(2) SmmRelocationSemaphoreComplete32() currently compiles to:
> 000002AE C6050000000001 mov byte [dword 0x0],0x1
> 000002B5 FF2500000000 jmp dword [dword 0x0]
where the first instruction is patched with the contents of
"mRebasedFlag" (so that (*mRebasedFlag) is set to 1), and the second
instruction is patched with the address of
"mSmmRelocationOriginalAddress" (so that we jump to
"mSmmRelocationOriginalAddress").
In its current form the first instruction could not be patched with
PatchInstructionX86(), given that the operand to patch is not encoded
in the trailing bytes of the instruction. Therefore, adopt an
EAX-based version, inspired by both the IA32 and X64 variants of
SmmRelocationSemaphoreComplete():
> 000002AE 50 push eax
> 000002AF B800000000 mov eax,0x0
> 000002B4 C60001 mov byte [eax],0x1
> 000002B7 58 pop eax
> 000002B8 FF2500000000 jmp dword [dword 0x0]
Here both instructions can be patched with PatchInstructionX86(), and
the DBs can be replaced with native NASM syntax.
(3) Turn the "mRebasedFlagAddr32" and "mSmmRelocationOriginalAddressPtr32"
variables into markers that suit PatchInstructionX86().
Cc: Eric Dong <eric.dong@intel.com>
Cc: Michael D Kinney <michael.d.kinney@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Rename the variable to "gPatchSmmInitStack" so that its association with
PatchInstructionX86() is clear from the declaration, change its type to
X86_ASSEMBLY_PATCH_LABEL, and patch it with PatchInstructionX86(). This
lets us remove the binary (DB) encoding of some instructions in
"SmmInit.nasm".
The size of the patched source operand is (sizeof (UINTN)).
Cc: Eric Dong <eric.dong@intel.com>
Cc: Michael D Kinney <michael.d.kinney@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
The IA32 version of "SmmInit.nasm" does not need "gSmmJmpAddr" at all (its
PiSmmCpuSmmInitFixupAddress() variant doesn't do anything either). We can
simply use the NASM syntax for the following Mixed-Size Jump:
> jmp PROTECT_MODE_CS : dword @32bit
The generated object code for the instruction is unchanged:
> 00000182 66EA5A0000000800 jmp dword 0x8:0x5a
(The NASM manual explains that putting the DWORD prefix after the colon
":" reflects the intent better, since it is the offset that is a DWORD.
Thus, that's what I used. However, both syntaxes are interchangeable,
hence the ndisasm output.)
The X64 version of "SmmInit.nasm" appears to require "gSmmJmpAddr";
however that's accidental, not inherent:
- Bring LONG_MODE_CODE_SEGMENT from
"UefiCpuPkg/PiSmmCpuDxeSmm/PiSmmCpuDxeSmm.h" to "SmmInit.nasm" as
LONG_MODE_CS, same as PROTECT_MODE_CODE_SEGMENT was brought to the IA32
version as PROTECT_MODE_CS earlier.
- Apply the NASM-native Mixed-Size Jump syntax again, but jump to the
fixed zero offset in LONG_MODE_CS. This will produce no relocation
record at all. Add a label after the instruction.
- Modify PiSmmCpuSmmInitFixupAddress() to patch the jump target backwards
from the label. Because we modify the DWORD offset with a DWORD access,
the segment selector is unharmed in the instruction, and we need not set
it from PiCpuSmmEntry().
According to "objdump --reloc", the X64 version undergoes only the
following relocations, after this patch:
> RELOCATION RECORDS FOR [.text]:
> OFFSET TYPE VALUE
> 0000000000000095 R_X86_64_PC32 SmmInitHandler-0x0000000000000004
> 00000000000000e0 R_X86_64_PC32 mRebasedFlag-0x0000000000000004
> 00000000000000ea R_X86_64_PC32 mSmmRelocationOriginalAddress-0x0000000000000004
Therefore the patch does not regress
<https://bugzilla.tianocore.org/show_bug.cgi?id=849> ("Enable XCODE5 tool
chain for UefiCpuPkg with nasm source code").
Cc: Eric Dong <eric.dong@intel.com>
Cc: Michael D Kinney <michael.d.kinney@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Like "gSmmCr4" in the previous patch, "gSmmCr0" is not only used for
machine code patching, but also as a means to communicate the initial CR0
value from SmmRelocateBases() to InitSmmS3ResumeState(). In other words,
the last four bytes of the "mov eax, Cr0Value" instruction's binary
representation are utilized as normal data too.
In order to get rid of the DB for "mov eax, Cr0Value", we have to split
both roles, patching and data flow. Introduce the "mSmmCr0" global (SMRAM)
variable for the data flow purpose. Rename the "gSmmCr0" variable to
"gPatchSmmCr0" so that its association with PatchInstructionX86() is clear
from the declaration, change its type to X86_ASSEMBLY_PATCH_LABEL, and
patch it with PatchInstructionX86(), to the value now contained in
"mSmmCr0".
This lets us remove the binary (DB) encoding of "mov eax, Cr0Value" in
"SmmInit.nasm".
Cc: Eric Dong <eric.dong@intel.com>
Cc: Michael D Kinney <michael.d.kinney@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Unlike "gSmmCr3" in the previous patch, "gSmmCr4" is not only used for
machine code patching, but also as a means to communicate the initial CR4
value from SmmRelocateBases() to InitSmmS3ResumeState(). In other words,
the last four bytes of the "mov eax, Cr4Value" instruction's binary
representation are utilized as normal data too.
In order to get rid of the DB for "mov eax, Cr4Value", we have to split
both roles, patching and data flow. Introduce the "mSmmCr4" global (SMRAM)
variable for the data flow purpose. Rename the "gSmmCr4" variable to
"gPatchSmmCr4" so that its association with PatchInstructionX86() is clear
from the declaration, change its type to X86_ASSEMBLY_PATCH_LABEL, and
patch it with PatchInstructionX86(), to the value now contained in
"mSmmCr4".
This lets us remove the binary (DB) encoding of "mov eax, Cr4Value" in
"SmmInit.nasm".
Cc: Eric Dong <eric.dong@intel.com>
Cc: Michael D Kinney <michael.d.kinney@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Rename the variable to "gPatchSmmCr3" so that its association with
PatchInstructionX86() is clear from the declaration, change its type to
X86_ASSEMBLY_PATCH_LABEL, and patch it with PatchInstructionX86(). This
lets us remove the binary (DB) encoding of some instructions in
"SmmInit.nasm".
Cc: Eric Dong <eric.dong@intel.com>
Cc: Michael D Kinney <michael.d.kinney@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
(This patch is the 64-bit variant of commit e75ee97224,
"UefiCpuPkg/PiSmmCpuDxeSmm: remove unneeded DBs from IA32 SmmStartup()",
2018-01-31.)
The SmmStartup() function executes in SMM, which is very similar to real
mode. Add "BITS 16" before it and "BITS 64" after it (just before the
@LongMode label).
Remove the manual 0x66 operand-size override prefixes, for selecting
32-bit operands -- the sizes of our operands trigger NASM to insert the
prefixes automatically in almost every spot. The one place where we have
to add it back manually is the LGDT instruction. In the LGDT instruction
we also replace the binary 0x2E prefix with the normal NASM syntax for CS
segment override.
The stores to the Control Registers were always 32-bit wide; the source
code only used RAX as source operand because it generated the expected
object code (with NASM compiling the source as if in BITS 64). With BITS
16 added, we can use the actual register width in the source operands
(EAX).
This patch causes NASM to generate byte-identical object code (determined
by disassembling both the pre-patch and post-patch versions, and comparing
the listings), except:
> @@ -231,7 +231,7 @@
> 000001D2 6689D3 mov ebx,edx
> 000001D5 66B800000000 mov eax,0x0
> 000001DB 0F22D8 mov cr3,eax
> -000001DE 662E670F0155F6 o32 lgdt [cs:ebp-0xa]
> +000001DE 2E66670F0155F6 o32 lgdt [cs:ebp-0xa]
> 000001E5 66B800000000 mov eax,0x0
> 000001EB 80CC02 or ah,0x2
> 000001EE 0F22E0 mov cr4,eax
The only difference is the prefix list order, it changes from:
- 0x66, 0x2E, 0x67
to
- 0x2E, 0x66, 0x67
Cc: Eric Dong <eric.dong@intel.com>
Cc: Michael D Kinney <michael.d.kinney@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
"mXdSupported" is a global BOOLEAN variable, initialized to TRUE. The
CheckFeatureSupported() function is executed on all processors (not
concurrently though), called from SmmInitHandler(). If XD support is found
to be missing on any CPU, then "mXdSupported" is set to FALSE, and further
processors omit the check. Afterwards, "mXdSupported" is read by several
assembly and C code locations.
The tricky part is *where* "mXdSupported" is allocated (defined):
- Before commit 717fb60443 ("UefiCpuPkg/PiSmmCpuDxeSmm: Add paging
protection.", 2016-11-17), it used to be a normal global variable,
defined (allocated) in "SmmProfile.c".
- With said commit, we moved the definition (allocation) of "mXdSupported"
into "SmiEntry.nasm". The variable was defined over the last byte of a
"mov al, 1" instruction, so that setting it to FALSE in
CheckFeatureSupported() would patch the instruction to "mov al, 0". The
subsequent conditional jump would change behavior, plus all further read
references to "mXdSupported" (in C and assembly code) would read back
the source (imm8) operand of the patched MOV instruction as data.
This trick required that the MOV instruction be encoded with DB.
In order to get rid of the DB, we have to split both roles: we need a
label for the code patching, and "mXdSupported" has to be defined
(allocated) independently of the code patching. Of course, their values
must always remain in sync.
(1) Reinstate the "mXdSupported" definition and initialization in
"SmmProfile.c" from before commit 717fb60443. Change the assembly
language definition ("global") to a declaration ("extern").
(2) Define the "gPatchXdSupported" label (type X86_ASSEMBLY_PATCH_LABEL)
in "SmiEntry.nasm", and add the C-language declaration to
"SmmProfileInternal.h". Replace the DB with the MOV mnemonic (keeping
the imm8 source operand with value 1).
(3) In CheckFeatureSupported(), whenever "mXdSupported" is set to FALSE,
patch the assembly code in sync, with PatchInstructionX86().
Cc: Eric Dong <eric.dong@intel.com>
Cc: Michael D Kinney <michael.d.kinney@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Rename the variable to "gPatchSmiCr3" so that its association with
PatchInstructionX86() is clear from the declaration, change its type to
X86_ASSEMBLY_PATCH_LABEL, and patch it with PatchInstructionX86(). This
lets us remove the binary (DB) encoding of some instructions in
"SmiEntry.nasm".
Cc: Eric Dong <eric.dong@intel.com>
Cc: Michael D Kinney <michael.d.kinney@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Rename the variable to "gPatchSmiStack" so that its association with
PatchInstructionX86() is clear from the declaration. Also change its type
to X86_ASSEMBLY_PATCH_LABEL.
Unlike "gSmbase" in the previous patch, "gSmiStack"'s patched value is
also de-referenced by C code (in other words, it is read back after
patching): the InstallSmiHandler() function stores "CpuIndex" to the given
CPU's SMI stack through "gSmiStack". Introduce the local variable
"CpuSmiStack" in InstallSmiHandler() for calculating the stack location
separately, then use this variable for both patching into the assembly
code, and for storing "CpuIndex" through it.
It's assumed that "volatile" stood in the declaration of "gSmiStack"
because we used to read "gSmiStack" back for de-referencing; with that use
gone, we can remove "volatile" too. (Note that the *target* of the pointer
was never volatile-qualified.)
Finally, replace the binary (DB) encoding of "mov esp, imm32" in
"SmiEntry.nasm".
Cc: Eric Dong <eric.dong@intel.com>
Cc: Michael D Kinney <michael.d.kinney@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Rename the variable to "gPatchSmbase" so that its association with
PatchInstructionX86() is clear from the declaration, change its type to
X86_ASSEMBLY_PATCH_LABEL, and patch it with PatchInstructionX86(). This
lets us remove the binary (DB) encoding of some instructions in
"SmiEntry.nasm".
Cc: Eric Dong <eric.dong@intel.com>
Cc: Michael D Kinney <michael.d.kinney@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Within function ApWakeupFunction():
When source level debugger is enabled, AP interrupts will be enabled by
EnableDebugAgent(). Then the AP function will be executed by:
Procedure (Parameter);
After the AP function returns, AP interrupts will be disabled when the
APs are placed in loop mode (both HltLoop and MwaiLoop).
However, at ExitBootServices, ApWakeupFunction() is called with
'Procedure' equals to RelocateApLoop().
(ExitBootServices callback registered within InitMpGlobalData())
RelocateApLoop() never returns, so it has to disable the AP interrupts by
itself. However, we find that interrupts are only disabled for the
HltLoop case, but not for the MwaitLoop case (within file MpFuncs.nasm).
This commit adds the missing disabling of AP interrupts for MwaitLoop.
Also, for X64, this commit will disable the interrupts before switching to
32-bit mode.
Cc: Laszlo Ersek <lersek@redhat.com>
Cc: Jian J Wang <jian.j.wang@intel.com>
Cc: Star Zeng <star.zeng@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Hao Wu <hao.a.wu@intel.com>
Reviewed-by: Ruiyu Ni <ruiyu.ni@intel.com>
Reviewed-by: Jiewen Yao <jiewen.yao@intel.com>
Reviewed-by: Jeff Fan <vanjeff_919@hotmail.com>
FixedPcdGetSize() is used as the macro value, PcdGetSize() is used as global
variable or function. Here usage is to access macro value.
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Liming Gao <liming.gao@intel.com>
Cc: Wang Jian J <jian.j.wang@intel.com>
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Michael Kinney <michael.d.kinney@intel.com>
Reviewed-by: Jian J Wang <jian.j.wang@intel.com>
if PcdDxeNxMemoryProtectionPolicy is enabled for EfiReservedMemoryType
of memory, #PF will be triggered for each APs after ExitBootServices
in SCRT test. The root cause is that AP wakeup code executed at that
time is stored in memory of type EfiReservedMemoryType (referenced by
global mReservedApLoopFunc), which is marked as non-executable.
This patch fixes this issue by setting memory of mReservedApLoopFunc to
be executable immediately after allocation.
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Boolean values do not need to use explicit comparisons
to TRUE or FALSE.
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Dandan Bi <dandan.bi@intel.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Ruiyu Ni <ruiyu.ni@intel.com>
This issue is introduced at following commit, which tried to add stack
switch support on behalf of Stack Guard feature.
0ff5aa9cae
The field KnownGoodStackTop in CPU_EXCEPTION_INIT_DATA is initialized to
the start address of array mNewStack. This is wrong. It must be the end
of mNewStack. This patch fixes this mistake.
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
V2: Just update the commit message to reference the hash value of
new performance infrastructure.
Our new performance infrastructure (edk2 trunk commit hash value:
SHA-1: 73fef64f14 ~
SHA-1: 115eae650b)can support to
dump performance date form ACPI table in OS. So we can remove
the old perf code to write performance data to OS.
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Dandan Bi <dandan.bi@intel.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Star Zeng <star.zeng@intel.com>
V2: Just update the commit message.
Add more perf entry to hook BootScriptDonePpi/EndOfPeiPpi/
EndOfS3Resume.
Add the new perf entry with Identifier
PERF_INMODULE_START_ID/PERF_INMODULE_END_ID which are defined
in new performance infrastructure (edk2 trunk commit hash value:
SHA-1: 73fef64f14 ~
SHA-1: 115eae650b).
PERF_INMODULE_START_ID/PERF_INMODULE_END_ID are general Identifier
which are used within a module.
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Dandan Bi <dandan.bi@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Acked-by: Laszlo Ersek <lersek@redhat.com>
Today's McaInitialize() doesn't check State value before initialize
MCi_CTL and MCi_STATUS.
The patch fixes this issue by only initializing the two kinds of
MSRs when State is enabled.
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Ruiyu Ni <ruiyu.ni@intel.com>
Reviewed-by: Eric Dong <eric.dong@intel.com>
Today's implementation only assumes SandyBridge CPU supports
Extended On-Demand Clock Modulation Duty Cycle.
Actually it is supported when CPUID.06h.EAX[5] == 1.
When platform requests 50% throttling, it causes value 1000b
set to the low-4 bits of IA32_CLOCK_MODULATION.
But the wrong code sets 1000b to bits[1-3] which causes assertion.
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Jeff Fan <vanjeff_919@hotmail.com>
Reviewed-by: Eric Dong <eric.dong@intel.com>
> v2:
> Reduce the number of page to update/restore from 3 to 2 because DF
> has no effect in this issue.
The infinite loop is caused by the memory instruction, such as
"rep mov", operating on memory block crossing boundary of NON-PRESENT
pages. Because the address triggering page fault set in CR2 will be in
the first page, SmmProfilePFHandler() will only change the first page
into PRESENT. The page following will be still in NON-PRESENT status.
Since SmmProfilePFHandler() will setup single-step trap for the
instruction causing #PF, when the handler returns back to the
instruction and re-execute it, both #DB and #PF will be triggered
because the instruction wants to access both first and second page
but only first page is PRESENT.
Normally #DB exception will be handled first and its handler will
change first page back to NON-PRESENT status. Then #PF is handled
and its handler will change first page to PRESENT status again and
setup another single-step for the instruction triggering #PF. Then
the whole system falls into an infinite loop and the memory operation
will never move on.
This patch fix above situation by always changing 2 pages to PRESENT
status instead of just 1 page. Those 2 pages include the page causing
#PF and the page after it.
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Jiewen Yao <jiewen.yao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Jiewen Yao <jiewen.yao@intel.com>
SMM emulation under both KVM and QEMU (TCG) crashes the guest when the
"jz" branch, added in commit d4d87596c1 ("UefiCpuPkg/PiSmmCpuDxeSmm:
Enable NXE if it's supported", 2018-01-18), is taken.
Rework the propagation of CPUID.80000001H:EDX.NX [bit 20] to IA32_EFER.NXE
[bit 11] so that no code is executed conditionally.
Cc: Eric Dong <eric.dong@intel.com>
Cc: Jian J Wang <jian.j.wang@intel.com>
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Ref: http://mid.mail-archive.com/d6fff558-6c4f-9ca6-74a7-e7cd9d007276@redhat.com
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
[lersek@redhat.com: XD -> NX code comment updates from Ray]
Reviewed-by: Ruiyu Ni <ruiyu.ni@intel.com>
[lersek@redhat.com: mark QEMU/TCG as well in the commit message]
The SmmStartup() executes in SMM, which is very similar to real mode. Add
"BITS 16" before it and "BITS 32" after it (just before the @32bit label).
Remove the manual 0x66 operand-size override prefixes, for selecting
32-bit operands -- the sizes of our operands trigger NASM to insert the
prefixes automatically in almost every spot. The one place where we have
to add it back manually is the LGDT instruction. (The 0x67 address-size
override prefix is also auto-generated.)
This patch causes NASM to generate byte-identical object code (determined
by disassembling both the pre-patch and post-patch versions, and comparing
the listings), except:
> @@ -158,7 +158,7 @@
> 00000142 6689D3 mov ebx,edx
> 00000145 66B800000000 mov eax,0x0
> 0000014B 0F22D8 mov cr3,eax
> -0000014E 67662E0F0155F6 o32 lgdt [cs:ebp-0xa]
> +0000014E 2E66670F0155F6 o32 lgdt [cs:ebp-0xa]
> 00000155 66B800000000 mov eax,0x0
> 0000015B 0F22E0 mov cr4,eax
> 0000015E 66B9800000C0 mov ecx,0xc0000080
The only difference is the prefix list order, it changes from:
- 0x67, 0x66, 0x2E
to
- 0x2E, 0x66, 0x67
(0x2E is "CS segment override").
Cc: Eric Dong <eric.dong@intel.com>
Cc: Jian J Wang <jian.j.wang@intel.com>
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=866
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Ruiyu Ni <ruiyu.ni@intel.com>
The gSmmCr3, gSmmCr4, gSmmCr0 and gSmmJmpAddr global variables are used
for patching assembly instructions, thus we can't yet remove the DB
encodings for those instructions. At least we should add the intended
meanings in comments.
This patch only changes comments.
Cc: Eric Dong <eric.dong@intel.com>
Cc: Jian J Wang <jian.j.wang@intel.com>
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Ruiyu Ni <ruiyu.ni@intel.com>
[lersek@redhat.com: adapt commit msg to ongoing PatchAssembly discussion]
The reason doing this is that we found that calling StartupAllAps() to
flush TLB for all APs in CpuDxe driver after changing page attributes
will spend a lot of time to complete. If there are many page attributes
update requests, the whole system performance will be slowed down
explicitly, including any shell command and UI operation.
The solution is removing the flush operation for AP in CpuDxe driver
and let AP flush TLB after woken up.
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Ruiyu Ni <ruiyu.ni@intel.com>
The reason doing this is that we found that calling StartupAllAps() to
flush TLB for all APs in CpuDxe driver after changing page attributes
will spend a lot of time to complete. If there are many page attributes
update requests, the whole system performance will be slowed down
explicitly, including any shell command and UI operation.
The solution is removing the flush operation for AP in CpuDxe driver.
Since TLB is always flushed in HLT loop mode, we just need to enforce
a TLB flush for mwait loop mode.
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Ruiyu Ni <ruiyu.ni@intel.com>
This issue is introduced by a patch at
f32bfe6d06
The above patch miss the case of 64-bit PEI, which will link
X64/MpFuncs.nasm instead of Ia32/MpFuncs.nasm. For X64/MpFuncs.nasm,
ExchangeInfo->ModeHighMemory should be always initialized no matter
if separate wakeup buffer is allocated or not. Ia32/MpFuncs.nasm will
not need ModeHighMemory during AP init. So the changes made in this
patch should not affect the functionality of it.
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Ruiyu Ni <ruiyu.ni@intel.com>
Every processor's StartupApSignal is initialized in
MpInitLibInitialize() before calling CollectProcessorCount().
When SortApicId() is called from CollectProcessorCount(), AP Index
is re-assigned by APIC ID. But SortApicId() forgets to set the
correct StartupApSignal when sorting the AP.
The patch fixes this issue.
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Star Zeng <star.zeng@intel.com>
Signed-off-by: Ruiyu Ni <ruiyu.ni@intel.com>
Reviewed-by: Star Zeng <star.zeng@intel.com>
Cc: Chasel Chiu <chasel.chiu@intel.com>
To fix an issue in which enabling NX feature will mark the AP wakeup
buffer as non-executable and fail the AP init, the buffer was split
into two part: the lower part in memory within 1MB and the higher part
within allocated executable memory (EfiBootServicesCode). But the
address of higher part memory was stored in lower part memory, which
is actually shared with legacy components and will be overwritten by
LegacyBiosDxe driver if CSM is enabled.
This patch fixes this issue by storing the address of higher part
memory in CpuMpData instead of ExchangeInfo.
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
MtrrSetMemoryAttributesInMtrrSettings() is a batch-set API.
When setting multiple ranges of memory attributes, the single-set
API (MtrrSetMemoryAttributeInMtrrSettings and MtrrSetMemoryAttribute)
may fail, but batch-set API may succeed.
Add comments to recommend caller to use batch-set API when setting
multiple ranges.
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Ruiyu Ni <ruiyu.ni@intel.com>
Reviewed-by: Ming Shao <ming.shao@intel.com>
GetWakeupBuffer() tries to find a below-1M free memory, it checks
whether the memory is allocated already in
CheckOverlapWithAllocatedBuffer(). When there is a memory allocation
hob (base = 0xff_00000000, size = 0x10000000),
CheckOverlapWithAllocateBuffer() truncates the base to 0 which causes
it always returns TRUE so GetWakeupBuffer() fails to find a below-1MB
memory.
The patch fixes this issue by using UINT64 type.
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Reviewed-by: Star Zeng <star.zeng@intel.com>
Reviewed-by: Jeff Fan <vanjeff_919@hotmail.com>
If features like memory profile, protection and heap guard are enabled,
a lot of more memory page attributes update actions will happen than
usual. An unnecessary sync of CR0.WP setting among APs will then cause
worse performance in memory allocation action. Removing the calling of
SyncMemoryPageAttributesAp() in function DisableReadOnlyPageWriteProtect
and EnableReadOnlyPageWriteProtect can fix this problem. In DEBUG build
case, the boot performance can be boosted from 11 minute to 6 minute.
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Eric Dong <eric.dong@intel.com>
If PcdDxeNxMemoryProtectionPolicy is set to enable protection for memory
of EfiBootServicesCode, EfiConventionalMemory, the BIOS will hang at a page
fault exception triggered by PiSmmCpuDxeSmm.
The root cause is that PiSmmCpuDxeSmm will access default SMM RAM starting
at 0x30000 which is marked as non-executable, but NX feature was not
enabled during SMM initialization. Accessing memory which has invalid
attributes set will cause page fault exception. This patch fixes it by
checking NX capability in cpuid and enable NXE in EFER MSR if it's
available.
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Eric Dong <eric.dong@intel.com>
If PcdDxeNxMemoryProtectionPolicy is set to enable protection for memory
of EfiBootServicesCode, EfiConventionalMemory and EfiReservedMemoryType,
the BIOS will hang at a page fault exception randomly.
The root cause is that the memory allocation for driver images (actually
a memory type conversion from free memory, type of EfiConventionalMemory,
to code memory, type of EfiBootServicesCode/EfiRuntimeServicesCode)
will get memory with NX set, because the CpuDxe driver will keep the NX
attribute (with free memory) in page directory during page table splitting
and then override the NX attribute of all its entries.
This patch fixes this issue by not inheriting NX attribute when turning
a page entry into a page directory during page granularity split.
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Eric Dong <eric.dong@intel.com>
If PcdDxeNxMemoryProtectionPolicy is set to enable protection for memory
of EfiBootServicesData, EfiConventionalMemory, the BIOS will reset after
timer initialized and started.
The root cause is that the memory used to hold the exception and interrupt
handler is allocated with type of EfiBootServicesData and marked as
non-executable due to NX feature enabled. This patch fixes it by allocating
EfiBootServicesCode type of memory for those handlers instead.
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Eric Dong <eric.dong@intel.com>
If PcdDxeNxMemoryProtectionPolicy is set to enable protection for memory
of EfiBootServicesCode, EfiConventionalMemory, the BIOS will hang at a page
fault exception during MP initialization.
The root cause is that the AP wake up buffer, which is below 1MB and used
to hold both AP init code and data, is type of EfiConventionalMemory (not
really allocated because of potential conflict with legacy code), and is
marked as non-executable. During the transition from real address mode
to long mode, the AP init code has to enable paging which will then cause
itself a page fault exception because it's just running in non-executable
memory.
The solution is splitting AP wake up buffer into two part: lower part is
still below 1MB and shared with legacy system, higher part is really
allocated memory of BootServicesCode type. The init code in the memory
below 1MB will not enable paging but just switch to protected mode and
jump to higher memory, in which the init code will enable paging and
switch to long mode.
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Eric Dong <eric.dong@intel.com>
In 32-bit mode, the BIOS will not create page table for memory beyond
4GB and therefore it cannot handle the attributes change request for
those memory. But current CpuDxe doesn't check this situation and still
try to complete the request, which will cause attributes of incorrect
memory address to be changed due to type cast from 64-bit to 32-bit.
This patch fixes this issue by checking the end address of input
memory block and returning EFI_UNSUPPORTED if it's out of range.
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Cc: Ruiyu Ni <ruiyu.ni@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Jian J Wang <jian.j.wang@intel.com>
Reviewed-by: Ruiyu Ni <ruiyu.ni@intel.com>
Commits a2ea6894e6
* UefiCpuPkg/MpInitLib: Fix a bug that AP enters timer INT handler
masked the interrupts in AP.
But it didn't unmask the interrupt in new BSP when Switch BSP
happens.
The patch fixed this issue.
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Ruiyu Ni <ruiyu.ni@intel.com>
Reviewed-by: Jeff Fan <vanjeff_919@hotmail.com>
Cc: Eric Dong <eric.dong@intel.com>
https://bugzilla.tianocore.org/show_bug.cgi?id=849
In V2, use "mov rax, strict qword 0" to replace the hard code db.
1. Use lea instruction to get the address instead of mov instruction.
2. Use the dummy address as jmp destination, and add the logic to fix up
the address to the absolute address at boot time.
3. On MpFuncs.nasm, use ExchangeInfo to record InitializeFloatingPointUnits.
This way is same to MpInitLib.
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Liming Gao <liming.gao@intel.com>
Cc: Andrew Fish <afish@apple.com>
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Cc: Michael Kinney <michael.d.kinney@intel.com>
Reviewed-by: Jiewen Yao <jiewen.yao@intel.com>
https://bugzilla.tianocore.org/show_bug.cgi?id=849
In V2, use "mov rax, strict qword 0" to replace the hard code db.
1. Use lea instruction to get the address instead of mov instruction.
2. Use the dummy address as jmp destination, and add the logic to fix up
the address to the absolute address at boot time.
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Liming Gao <liming.gao@intel.com>
Cc: Andrew Fish <afish@apple.com>
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Cc: Michael Kinney <michael.d.kinney@intel.com>
Reviewed-by: Jiewen Yao <jiewen.yao@intel.com>
https://bugzilla.tianocore.org/show_bug.cgi?id=849
In V2, use mov rax, strict qword 0 to replace the hard code db.
Use the dummy address as jmp destination, and add the logic to fix up
the address to the absolute address at boot time.
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Liming Gao <liming.gao@intel.com>
Cc: Andrew Fish <afish@apple.com>
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Cc: Michael Kinney <michael.d.kinney@intel.com>
Reviewed-by: Jiewen Yao <jiewen.yao@intel.com>
Enhance MCA feature dependency check base on SDM pseudocode example 15-1.
Cc: Eric Dong <eric.dong@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.1
Signed-off-by: Bell Song <binx.song@intel.com>
Reviewed-by: Eric Dong <eric.dong@intel.com>
Laszlo Ersek
Star Zeng
Hao Wu
Liming Gao
Jian J Wang
Dandan Bi
Ruiyu Ni
Eric Dong
Song, BinX