I can (*) resume a Fedora 19 guest with this.
(*) Almost.
(And now that everyone is listening, let's talk about the problems :))
v1->v2:
- Rebase:
- "MdeModulePkg: SmmLockBox: remove wrong DepEx" has been applied.
- Series is on top of Wei's Xen series now.
- Old patch:
"OvmfPkg: PlatformPei: reserve RAM as SMRAM backing":
- fixup param list in ReserveEmuSmmArea() function definition.
- New patch:
"OvmfPkg: S3 Suspend: save boot script after ACPI context":
- The "big problem" of v1 about EFI_DXE_SMM_READY_TO_LOCK_PROTOCOL not
being installed by anyone has been solved -- save the boot script
inside the OVMF specific EFI_ACPI_S3_SAVE_PROTOCOL implementation.
- Old patch:
"OvmfPkg: S3 Resume: detect S3 Resume in CMOS and set boot mode
accordingly"
- Redraw data flow diagram between suspend and resume according to
new, OVMF-internal installation of
EFI_DXE_SMM_READY_TO_LOCK_PROTOCOL; see above.
- After Paolo had proved (thanks!) that the S3 resume trampoline in
"MdeModulePkg/Universal/Acpi/BootScriptExecutorDxe/X64/S3Asm.S" worked
correctly on KVM too, I started protecting the OS's RAM:
<http://thread.gmane.org/gmane.comp.bios.tianocore.devel/5297/focus=5321>
On 12/04/13 18:17, Jordan Justen wrote:
> I can think of one *major* issue for OVMF S3. We run PEI from RAM.
> This allows allows us to compress the PEI code.
>
> Therefore, on S3 resume, we will overwrite a big chunk of memory when
> we decompress the main FV. This is going to trash a big chunk of RAM
> likely used by the OS. It starts at PcdOvmfMemFvBase and has a size of
> PcdOvmfMemFvSize + 2MB. (The 2MB is actually closer to 1MB, and is
> some extra RAM used during decompression).
>
> Oh, and one other chunk of RAM we use is SEC/PEI Temporary RAM at
> 0x70000-0x80000. (SecMain.c SecCoreStartupWithStack)
>
> I guess the first quick hack would be to reserve these memory ranges
> so the OS will not use them. (EfiACPIMemoryNVS) [...]
The following patches are new and concern themselves with this problem:
OvmfPkg: S3 Resume: reserve decompressed firmware area from the OS
OvmfPkg: S3 Resume: reserve fw decompression scratch space from the OS
OvmfPkg: S3 Resume: reserve initial (pre-migr.) SEC/PEI stack and PEI heap
MdeModulePkg: DxeLoad: UpdateStackHob(): honor original allocation type
OvmfPkg: S3 Resume: ensure survival of migrated PEI stack / heap
OvmfPkg: S3 Resume: flatten DXEFV into MAINFV
UefiCpuPkg: S3Resume2Pei: align return stacks explicitly
UefiCpuPkg: S3Resume2Pei: allow preallocation of PEI_S3_RESUME_STATE
OvmfPkg: S3 Resume: preallocate PEI_S3_RESUME_STATE for S3Resume2Pei
I've written these new patches in the following manner:
- Logged all memory allocation HOBs at the beginning of
S3RestoreConfig2().
- For each memory allocation HOB describing a memory type different
from EfiACPIMemoryNVS and EfiRuntimeServicesData, I tracked down the
origin of the allocation, and adapted it to S3 resume.
- Modified MdePkg/Library/PeiMemoryAllocationLib to emit WARNINGs for
EfiBootServicesData requests. Tracked down all such warnings that were
emitted after entering S3RestoreConfig2() (ie. on the resume path).
Good news: x86_64 Fedora 19 guest can be suspended and "resumed", as in:
- I can log in and issue commands over the serial console.
- I can access the virtio-blk disk.
- The dmesg before and after suspend and resume looks generally
sensible.
- The "shutdown -h now" command tends to work.
Bad news:
- The screen remains dead (independently of cirrus vs. stdvga, and
independently of efifb console vs. X11 GUI).
- virtnet_restore() seems to reliably trigger "list_add corruption" in
the kernel. Maybe a virtio-net resume bug in Linux, maybe we're
corrupting OS memory.
- Resume is generally unstable. Sometimes it looks perfect, sometimes a
bunch of userspace processes start dumping core immediately after
resume (probably some memory overwrites left untracked...)
- When X11 is running (let alone the resume is started from the GUI),
the stability worsens greatly.
I think we could discuss the series at this point. It is available under
<https://github.com/lersek/edk2/commits/s3_rfc_02> too.
Thanks,
Laszlo
Laszlo Ersek (24):
OvmfPkg: PlatformPei: reserve RAM as SMRAM backing store
OvmfPkg: S3 Suspend: introduce EmuSmmDxe for emulating SMRAM
OvmfPkg: S3 Suspend: pull in DXE driver for
EFI_SMM_COMMUNICATION_PROTOCOL
OvmfPkg: S3 Suspend: pull in SmmLockBox driver
OvmfPkg: S3 Suspend: use SMM instances for LockBoxLib library class
OvmfPkg: S3 Suspend: import specialized copy of AcpiS3SaveDxe
OvmfPkg: S3 Suspend: save ACPI context
OvmfPkg: S3 Suspend: enable creation/saving of an S3 Boot Script
OvmfPkg: S3 Suspend: save boot script after ACPI context
OvmfPkg: PlatformPei: reserve storage for disclosing SMST
OvmfPkg: S3 Suspend: introduce DiscloseSmstSmm driver
OvmfPkg: S3 Resume: introduce EmuSmmPei for emulating SMRAM in PEI
OvmfPkg: S3 Resume: pull in BootScriptExecutorDxe
OvmfPkg: S3 Resume: pull in PEIM orchestrating S3 Resume
OvmfPkg: S3 Resume: reserve decompressed firmware area from the OS
OvmfPkg: S3 Resume: reserve fw decompression scratch space from the OS
OvmfPkg: S3 Resume: reserve initial (pre-migr.) SEC/PEI stack and PEI
heap
MdeModulePkg: DxeLoad: UpdateStackHob(): honor original allocation
type
OvmfPkg: S3 Resume: ensure survival of migrated PEI stack / heap
OvmfPkg: S3 Resume: flatten DXEFV into MAINFV
UefiCpuPkg: S3Resume2Pei: align return stacks explicitly
UefiCpuPkg: S3Resume2Pei: allow preallocation of PEI_S3_RESUME_STATE
OvmfPkg: S3 Resume: preallocate PEI_S3_RESUME_STATE for S3Resume2Pei
OvmfPkg: S3 Resume: detect S3 Resume in CMOS and set boot mode
accordingly
OvmfPkg/AcpiS3SaveDxe/AcpiS3SaveDxe.inf | 78 +++
OvmfPkg/DiscloseSmstSmm/DiscloseSmstSmm.inf | 51 ++
OvmfPkg/EmuSmmDxe/EmuSmmDxe.inf | 55 ++
OvmfPkg/EmuSmmPei/EmuSmmPei.inf | 60 ++
OvmfPkg/PlatformPei/PlatformPei.inf | 5 +
OvmfPkg/Sec/SecMain.inf | 4 +
.../Universal/Acpi/S3Resume2Pei/S3Resume2Pei.inf | 1 +
OvmfPkg/AcpiS3SaveDxe/AcpiS3Save.h | 59 ++
MdeModulePkg/Core/DxeIplPeim/DxeLoad.c | 3 +-
OvmfPkg/AcpiS3SaveDxe/AcpiS3Save.c | 605 +++++++++++++++++++++
OvmfPkg/DiscloseSmstSmm/DiscloseSmstSmm.c | 38 ++
OvmfPkg/EmuSmmDxe/EmuSmmDxe.c | 276 ++++++++++
OvmfPkg/EmuSmmPei/EmuSmmPei.c | 285 ++++++++++
OvmfPkg/PlatformPei/Fv.c | 27 +-
OvmfPkg/PlatformPei/Platform.c | 61 ++-
OvmfPkg/Sec/SecMain.c | 35 ++
UefiCpuPkg/Universal/Acpi/S3Resume2Pei/S3Resume.c | 19 +-
OvmfPkg/OvmfPkg.dec | 16 +
OvmfPkg/OvmfPkgX64.dsc | 47 +-
OvmfPkg/OvmfPkgX64.fdf | 65 ++-
UefiCpuPkg/UefiCpuPkg.dec | 7 +
21 files changed, 1755 insertions(+), 42 deletions(-)
create mode 100644 OvmfPkg/AcpiS3SaveDxe/AcpiS3SaveDxe.inf
create mode 100644 OvmfPkg/DiscloseSmstSmm/DiscloseSmstSmm.inf
create mode 100644 OvmfPkg/EmuSmmDxe/EmuSmmDxe.inf
create mode 100644 OvmfPkg/EmuSmmPei/EmuSmmPei.inf
create mode 100644 OvmfPkg/AcpiS3SaveDxe/AcpiS3Save.h
create mode 100644 OvmfPkg/AcpiS3SaveDxe/AcpiS3Save.c
create mode 100644 OvmfPkg/DiscloseSmstSmm/DiscloseSmstSmm.c
create mode 100644 OvmfPkg/EmuSmmDxe/EmuSmmDxe.c
create mode 100644 OvmfPkg/EmuSmmPei/EmuSmmPei.c
--
1.8.3.1
------------------------------------------------------------------------------
Sponsored by Intel(R) XDK
Develop, test and display web and hybrid apps with a single code base.
Download it for free now!
http://pubads.g.doubleclick.net/gampad/clk?id=111408631&iu=/4140/ostg.clktrk
_______________________________________________
edk2-devel mailing list
edk2-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/edk2-devel
