The Boot#### variables that have become unreferenced in the new BootOrder
variable won't ever be automatically reused for booting. They are
"unreachable" resources that take up room in the variable store. Make an
effort to remove them.
This should plug the leak which, given sufficient reboots, exhausts the
variable store with stale Boot#### variables and renders the VM
unbootable.
Reported-by: Michael Chang <mchang@suse.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@15327 6f19259b-4bc3-4df7-8a09-765794883524
When PI can distinguish the "full config" boot mode from "assume no
changes", then the following BDS logic is correct:
if BootMode == BOOT_WITH_FULL_CONFIGURATION:
//
// connect all devices
// create & append each default boot option that's missing
//
BdsLibConnectAll
BdsLibEnumerateAllBootOption
else if BootMode == BOOT_ASSUMING_NO_CONFIGURATION_CHANGES:
//
// just stick with current BootOrder and the Boot#### variables
// referenced by it
//
In theory, the first branch is intended to run infrequently, and the
"assume no changes" branch should run most of the time.
However, some platforms can't tell these two boot modes apart. The
following substitute had been introduced:
//
// Technically, always assume "full config", but the BootMode HOB is
// actually meaningless wrt. to "full config" or "assume no changes".
//
ASSERT (BootMode == BOOT_WITH_FULL_CONFIGURATION);
//
// Key off the existence of BootOrder. Try to prepare an in-memory list
// of boot options, based on BootOrder and the referenced Boot####
// variables.
//
Status = BdsLibBuildOptionFromVar()
//
// If that succeeded, we'll treat it as "assume no changes". If it
// failed (*only* if it failed), we'll build default boot options,
// calling it "full config":
//
if EFI_ERROR(Status):
BdsLibConnectAll()
BdsLibEnumerateAllBootOption(BootOptionList)
What we have now in OVMF is a mixture of the hack, and the behavior that's
theoretically correct for "full config":
- We assert "full config" -- this is OK.
- We call "connect all" and "enumerate all" deliberately -- this is OK
too. It matches "full config" which we assert.
- However, we also have the hack in place, which had been meant as an
alternative.
In order to clean this up, we either need to restore the hack to its
original form (ie. comment out the unconditional calls again), or we ought
to remove the hack altogether.
The unconditional "connect all" + "enumerate all" calls are the correct
approach for OVMF, because we want, in fact, to start with "full config".
The QEMU boot order specification and the set of emulated devices might
change "out of band", which excludes "assume no changes".
In other words, removing the hack corresponds to the "real production"
case that the comment hints at.
Because SetBootOrderFromQemu() may change the BootOrder NvVar, we must
preserve the BdsLibBuildOptionFromVar() function call, in order to
refresh the in-memory list with the new boot priorities.
(The last step of BdsLibEnumerateAllBootOption() is such a call too.)
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@15326 6f19259b-4bc3-4df7-8a09-765794883524
The S3 suspend/resume infrastructure depends on the LockBox library class.
The edk2 tree currently contains Null and SMM instances. The Null instance
is useless, and the SMM instance would require SMM emulation by including
the SMM core and adding several new drivers, which is deemed too complex.
Hence add a simple LockBoxLib instance for OVMF.
jordan.l.justen@intel.com:
* use PCDs instead of EmuNvramLib
- clear memory in PlatformPei on non S3 boots
* allocate NVS memory and store a pointer to that memory
- reduces memory use at fixed locations
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@15301 6f19259b-4bc3-4df7-8a09-765794883524
Such a packaged query function will come in handy in the following
patches.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
[jordan.l.justen@intel.com: check for enabled rather than disabled]
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@15292 6f19259b-4bc3-4df7-8a09-765794883524
The BOOLEAN IsFinal variable initialization isn't properly seen by
MSVC. To make it compile OVMF the variable needs to be initialized.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Mauro Faccenda <faccenda@gmail.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@15214 6f19259b-4bc3-4df7-8a09-765794883524
When this option is passed to qemu, it appends the word HALT to the
"bootorder" fw_cfg file, as last entry. For example,
/pci@i0cf8/ethernet@3/ethernet-phy@0
/pci@i0cf8/scsi@4/disk@0,0
HALT
The option's purpose is to prevent SeaBIOS from booting from devices that
have not been specified explicitly (with bootindex=N device properties nor
-boot options). When SeaBIOS sees HALT, it doesn't proceed to boot from
default locations (after boot fails from all of the listed locations).
The HALT string currently causes OVMF to reject the entire "bootorder"
fw_cfg contents, with "parse error". This is not good, because since a
recent libvirt commit, libvirt unconditionally passes "-boot strict=on" to
qemu. Consequently, the boot order logic in QemuBootOrder.c has stopped
working for libvirt users.
OVMF's SetBootOrderFromQemu() function actually implements the idea behind
"-boot strict=on": it drops all boot options not in the fw_cfg list. (*)
Therefore, let's recognize HALT, and just do what we've been doing all
along.
(*) Except the UEFI shell, according to the survival policy in
BootOrderComplete(), but the memory mapped UEFI shell is not expressible
via fw_cfg anyway, and its preservation has been requested on edk2-devel.
Hence it's a good boot option to keep in any case.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@15197 6f19259b-4bc3-4df7-8a09-765794883524
The QemuFwCfgSecLib library instance
- is stateless,
- has no library constructor,
- is available to SEC client code,
- must be queried with QemuFwCfgIsAvailable() before use,
- is restricted to SEC in order to limit the explicit querying
requirement. (There is no current user.)
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@15046 6f19259b-4bc3-4df7-8a09-765794883524
The current implementation of QemuFwCfgLib is:
- stateful
- implicitly initialized in the library constructor.
OVMF's SEC runs from read-only memory/flash. When the library is linked
into a SEC binary (which currently never happens), the
"mQemuFwCfgSupported" global variable becomes read-only, making the
library non-functional.
Extract the stateful, implicitly initialized library implementation into a
separate file, making room for a stateless, explicitly queried
implementation that's usable in SEC. Restrict the stateful implementation
to the current, non-SEC clients.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@15045 6f19259b-4bc3-4df7-8a09-765794883524
This internal function allows separation of library-internal and
for-clients external availability of fw_cfg.
The interface contract of QemuFwCfgIsAvailable() is changed so that now it
may modify fw_cfg state. All current users are compliant with the new
contract.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@15044 6f19259b-4bc3-4df7-8a09-765794883524
These functions did not provide much more than the new protocol functions
VIRTIO_DEVICE_PROTOCOL.ReadDevice() / VIRTIO_DEVICE_PROTOCOL.WriteDevice().
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Olivier Martin <olivier.martin@arm.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14968 6f19259b-4bc3-4df7-8a09-765794883524
This change replaces the accesses to the PCI bus from the Block, Scsi and Net drivers by
the use of the new VIRTIO_DEVICE_PROTOCOL protocol that abstracts the transport layer.
It means these drivers can be used on PCI and MMIO transport layer.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Olivier Martin <olivier.martin@arm.com>
v5:
- VirtioFlush(): update comment block in VirtioLib.[hc]; error code is
propagated from VirtIo->SetQueueNotify().
- VirtioBlkInit(): jump to Failed label if SetPageSize() fails
- VirtioBlkInit(): fixup comment, and add error handling, near
SetQueueNum() call
- VirtioBlkDriverBindingStart(): remove redundant (always false) check for
a subsystem device ID different from VIRTIO_SUBSYSTEM_BLOCK_DEVICE;
VirtioBlkDriverBindingSupported() handles it already
- VirtioNetGetFeatures(): update stale comment block
- VirtioNetGetFeatures(): retrieve MAC address byte for byte (open-coded
loop)
- VirtioNetDriverBindingStart(): remove redundant (always false) check for
a subsystem device ID different from VIRTIO_SUBSYSTEM_NETWORK_CARD;
VirtioNetDriverBindingSupported() handles it already
- VirtioNetInitRing(): call SetQueueNum() and SetQueueAlign() for proper
MMIO operation
- VirtioNetInitialize(): fix destination error label for when
SetPageSize() fails
- VirtioScsi.c: fix comment block of VIRTIO_CFG_WRITE()/VIRTIO_CFG_READ()
- VirtioScsiInit(): fix destination error label for when SetPageSize()
fails
- VirtioScsiInit(): call SetQueueNum() and SetQueueAlign() for proper MMIO
operation
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14966 6f19259b-4bc3-4df7-8a09-765794883524
Why is the virtio-mmio implementation of the protocol a library,
instead of a driver binary?
The UEFI driver model would encourage to create a virtio-mmio driver
instead of a library. But the reasons why I created a library are:
- A virtio-mmio driver would imply an additional protocol that would
probably have a single attribute field:
typedef struct {
PHYSICAL_ADDRESS BaseAddress;
} VIRTIO_MMIO_DEVICE_PROTOCOL;
- There is no (easy) way to scan the available VirtIo devices on a
platform. So, the UEFI firmware for this platform would need a driver
to produce instances for every virtio devices it wants to expose in
UEFI. A single call to a helper library (ie: VirtioMmioDeviceLib)
make the porting easier.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Olivier Martin <olivier.martin@arm.com>
v5:
- typo fix in VirtioMmioInstallDevice() comment block
- plug MmioDevice leak in VirtioMmioUninstallDevice()
- return EFI_INVALID_PARAMETER in VirtioMmioGetQueueAddress() if
QueueAddress is NULL
- VirtioMmioSetQueueSize(): fix return value (it's a status code)
- VirtioMmioSetPageSize(): check against EFI_PAGE_SIZE with "if" plus
EFI_UNSUPPORTED, rather than ASSERT()
- VirtioMmioDeviceWrite(), VirtioMmioDeviceRead(): remove redundant
(FieldSize > 8) checks
- VirtioMmioDeviceLib.inf: drop UefiDriverEntryPoint library dependency
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14965 6f19259b-4bc3-4df7-8a09-765794883524
QemuFlashFvbServicesRuntimeDxe provides actual persistent storage for
non-volatile variables. When it is active, any on-disk NvVars file counts
as a stale source of variables -- hence don't load these files in BDS.
This also allows Secure Boot settings (eg. enrolled keys) to survive cold
VM reboots.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14844 6f19259b-4bc3-4df7-8a09-765794883524
LoadLinux() is looking at the wrong field for the kernel's EFI handover
protocol flags. It's not currently possible for JumpToUefiKernel() to
ever be called (even accidentally) because BIT2 and BIT3 of
Bp->hdr.load_flags are never set in modern kernels, which means that
control is always transferred to the kernel via the legacy entry point.
Look at the correct field so that the EFI handover protocol is used
whenever it's available.
Contributed-under: TianoCore Contribution Agreement 1.0
Cc: David Woodhouse <David.Woodhouse@intel.com>
Cc: Jordan Justen <jordan.l.justen@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14721 6f19259b-4bc3-4df7-8a09-765794883524
Some of the active boot options that have not been selected over fw_cfg
should be preserved at the end of the boot order. For now we're adding
back everything that starts with neither PciRoot() nor HD(). This includes
the UEFI shell, memory-mapped from the firmware image.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Tested-by: Michael Chang <mchang@suse.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14668 6f19259b-4bc3-4df7-8a09-765794883524
This will allow us to identify those UEFI boot options (while keeping
their relative order) that have *not* been selected by fw_cfg.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Tested-by: Michael Chang <mchang@suse.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14667 6f19259b-4bc3-4df7-8a09-765794883524
In preparation for the next patch, collect active UEFI boot options in
advance into a new array. Rebase the current inner loop (the matching
loop) to this array.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Tested-by: Michael Chang <mchang@suse.com>
[jordan.l.justen@intel.com: initialize *ActiveOption for GCC IA32 warning]
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14666 6f19259b-4bc3-4df7-8a09-765794883524
The prefix matching logic in Match()
[OvmfPkg/Library/PlatformBdsLib/QemuBootOrder.c] expects UEFI boot options
to specify full (absolute) device paths. However, partial (relative)
device paths starting with a HD() node are valid for booting. By not
recognizing them, QemuBootOrder.c misses (and deletes) valid boot options
that would otherwise match the user's preference.
Just like BdsLibBootViaBootOption() expands such paths with the
BdsExpandPartitionPartialDevicePathToFull() function for booting, do the
same in QemuBootOrder.c for prefix matching.
This moves the very first call to
BdsExpandPartitionPartialDevicePathToFull() to an earlier point. The
following call tree explains it:
BdsEntry() [IntelFrameworkModulePkg/Universal/BdsDxe/BdsEntry.c]
PlatformBdsPolicyBehavior() [OvmfPkg/Library/PlatformBdsLib/BdsPlatform.c]
SetBootOrderFromQemu() [OvmfPkg/Library/PlatformBdsLib/QemuBootOrder.c]
Match() [OvmfPkg/Library/PlatformBdsLib/QemuBootOrder.c]
BdsExpandPartitionPartialDevicePathToFull() [IntelFrameworkModulePkg/Library/GenericBdsLib/BdsBoot.c]
BdsBootDeviceSelect() [IntelFrameworkModulePkg/Universal/BdsDxe/BdsEntry.c]
BdsLibBootViaBootOption() [IntelFrameworkModulePkg/Library/GenericBdsLib/BdsBoot.c]
BdsExpandPartitionPartialDevicePathToFull() [IntelFrameworkModulePkg/Library/GenericBdsLib/BdsBoot.c]
This should be fine, for two reasons:
- the new, earlier call is still under BdsEntry(),
- BdsExpandPartitionPartialDevicePathToFull() expects to be called
repeatedly, even with the same set of HD() device paths. This function
implements its own caching for device paths, likely for performance
reasons.
That fits this patch well because whatever device paths we expand under
PlatformBdsPolicyBehavior() can be quickly looked up in
BdsBootDeviceSelect(), so no work (ie.
BdsLibConnectAllDriversToAllControllers()) should be wasted.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Tested-by: Michael Chang <mchang@suse.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14665 6f19259b-4bc3-4df7-8a09-765794883524
The volatile 'NvVars' variable indicates that the variables do
not need to be loaded from the file again. After we write the
variables out to the file, there is clearly no need to load
them back from the file.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Tested-by: Michael Chang <mchang@suse.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14613 6f19259b-4bc3-4df7-8a09-765794883524
Enforce in-order execution of these steps even on not sequentially
consistent architectures, as discussed in [1]. These changes should be
unnecessary on x86 (the only architecture OVMF currently supports), but
they align the OVMF virtio code with the virtio specification and could be
necessary for future OVMF ports.
[1] http://lists.linuxfoundation.org/pipermail/virtualization/2013-June/024547.html
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14601 6f19259b-4bc3-4df7-8a09-765794883524
In Linux, efi_memblock_x86_reserve_range() and efi_reserve_boot_services()
expect that whoever allocates the EFI memmap allocates it in Loader Data
type memory. Linux's own exit_boot()-->low_alloc() complies, but
SetupLinuxMemmap() in LoadLinuxLib doesn't.
The memory type discrepancy leads to efi_memblock_x86_reserve_range() and
efi_reserve_boot_services() both trying to reserve the range backing the
memmap, resulting in memmap entry truncation in
efi_reserve_boot_services().
This fix also makes this allocation consistent with all other persistent
allocations in "OvmfPkg/Library/LoadLinuxLib/Linux.c".
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reported-and-tested-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14555 6f19259b-4bc3-4df7-8a09-765794883524
When the PM base address was moved from 0x400 to 0xb000, this
code was missed. This prevented shutdown's via the UEFI system
call from working. (For example, at the EFI shell prompt: reset -s)
We now use gUefiOvmfPkgTokenSpaceGuid.PcdAcpiPmBaseAddress
which is currently set at 0xb000.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
git-svn-id: https://svn.code.sf.net/p/edk2/code/trunk/edk2@14492 6f19259b-4bc3-4df7-8a09-765794883524
OvmfPkg's file-based NvVar storage is read back as follows at boot (all
paths under OvmfPkg/Library/):
PlatformBdsPolicyBehavior() [PlatformBdsLib/BdsPlatform.c]
PlatformBdsRestoreNvVarsFromHardDisk()
VisitAllInstancesOfProtocol
for each simple file system:
VisitingFileSystemInstance()
ConnectNvVarsToFileSystem() [NvVarsFileLib/NvVarsFileLib.c]
LoadNvVarsFromFs() [NvVarsFileLib/FsAccess.c]
ReadNvVarsFile()
+-------------> SerializeVariablesSetSerializedVariables() [SerializeVariablesLib/SerializeVariablesLib.c]
| SerializeVariablesIterateInstanceVariables()
| +-------------> IterateVariablesInBuffer()
| | for each loaded / deserialized variable:
| +-|-----------------> IterateVariablesCallbackSetSystemVariable()
| | | gRT->SetVariable()
| | |
| | IterateVariablesInBuffer() stops processing variables as soon as the
| | first error is encountered from the callback function.
| |
| | In this case the callback function is
| IterateVariablesCallbackSetSystemVariable(), selected by
SerializeVariablesSetSerializedVariables().
The result is that no NvVar is restored from the file after the first
gRT->SetVariable() failure.
On my system such a failure
- never happens in an OVMF build with secure boot disabled,
- happens *immediately* with SECURE_BOOT_ENABLE, because the first
variable to restore is "AuthVarKeyDatabase".
"AuthVarKeyDatabase" has the EFI_VARIABLE_AUTHENTICATED_WRITE_ACCESS
attribute set. Since the loop tries to restore it before any keys (PK, KEK
etc) are enrolled, gRT->SetVariable() rejects it with
EFI_SECURITY_VIOLATION. Consequently the NvVar restore loop terminates
immediately, and we never reach non-authenticated variables such as
Boot#### and BootOrder.
Until work on KVM-compatible flash emulation converges between qemu and
OvmfPkg, improve the SECURE_BOOT_ENABLE boot experience by masking
EFI_SECURITY_VIOLATION in the callback:
- authenticated variables continue to be rejected same as before, but
- at least we allow the loop to progress and restore non-authenticated
variables, for example boot options.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@14390 6f19259b-4bc3-4df7-8a09-765794883524
The descriptor table (also known as "queue") consists of descriptors. (The
corresponding type in the code is VRING_DESC.)
An individual descriptor describes a contiguous buffer, to be transferred
uni-directionally between host and guest.
Several descriptors in the descriptor table can be linked into a
descriptor chain, specifying a bi-directional scatter-gather transfer
between host and guest. Such a descriptor chain is also known as "virtio
request".
(The descriptor table can host sereval descriptor chains (in-flight virtio
requests) in parallel, but the OVMF driver supports at most one chain, at
any point in time.)
The first descriptor in any descriptor chain is called "head descriptor".
In order to submit a number of parallel requests (= a set of independent
descriptor chains) from the guest to the host, the guest must put *only*
the head descriptor of each separate chain onto the Available Ring.
VirtioLib currently places the head of its one descriptor chain onto the
Available Ring repeatedly, once for each single (head *or* dependent)
descriptor in said descriptor chain. If the descriptor chain comprises N
descriptors, this error amounts to submitting the same entire chain N
times in parallel.
Available Ring Descriptor table
Ptr to head ----> Desc#0 (head of chain)
Ptr to head --/ Desc#1 (next in same chain)
... / ...
Ptr to head / Desc#(N-1) (last in same chain)
Anatomy of a single virtio-blk READ request (a descriptor chain with three
descriptors):
virtio-blk request header, prepared by guest:
VirtioAppendDesc PhysAddr=3FBC6050 Size=16 Flags=1 Head=1232 Next=1232
payload to be filled in by host:
VirtioAppendDesc PhysAddr=3B934C00 Size=32768 Flags=3 Head=1232 Next=1233
host status, to be filled in by host:
VirtioAppendDesc PhysAddr=3FBC604F Size=1 Flags=2 Head=1232 Next=1234
Processing on the host side -- the descriptor chain is processed three
times in parallel (its head is available to virtqueue_pop() thrice); the
same chain is submitted/collected separately to/from AIO three times:
virtio_queue_notify vdev VDEV vq VQ#0
virtqueue_pop vq VQ#0 elem EL#0 in_num 2 out_num 1
bdrv_aio_readv bs BDRV sector_num 585792 nb_sectors 64 opaque REQ#0
virtqueue_pop vq VQ#0 elem EL#1 in_num 2 out_num 1
bdrv_aio_readv bs BDRV sector_num 585792 nb_sectors 64 opaque REQ#1
virtqueue_pop vq VQ#0 elem EL#2 in_num 2 out_num 1
bdrv_aio_readv bs BDRV sector_num 585792 nb_sectors 64 opaque REQ#2
virtio_blk_rw_complete req REQ#0 ret 0
virtio_blk_req_complete req REQ#0 status 0
virtio_blk_rw_complete req REQ#1 ret 0
virtio_blk_req_complete req REQ#1 status 0
virtio_blk_rw_complete req REQ#2 ret 0
virtio_blk_req_complete req REQ#2 status 0
On my Thinkpad T510 laptop with RHEL-6 as host, this probably leads to
simultaneous DMA transfers targeting the same RAM area. Even though the
source of each transfer is identical, the data is corrupted in the
destination buffer -- the CRC32 calculated over the buffer varies, even
though the origin of the transfers is the same, never rewritten LBA.
SynchronousRequest Lba=585792 BufSiz=32768 ReqIsWrite=0 Crc32=BF68A44D
The problem is invisible on my HP Z400 workstation.
Fix the request submission by:
- building the only one descriptor chain supported by VirtioLib always at
the beginning of the descriptor table,
- ensuring the head descriptor of this chain is put on the Available Ring
only once,
- requesting the virtio spec's language to be cleaned up
<http://lists.linuxfoundation.org/pipermail/virtualization/2013-April/024032.html>.
Available Ring Descriptor table
Ptr to head ----> Desc#0 (head of chain)
Desc#1 (next in same chain)
...
Desc#(N-1) (last in same chain)
VirtioAppendDesc PhysAddr=3FBC6040 Size=16 Flags=1 Head=0 Next=0
VirtioAppendDesc PhysAddr=3B934C00 Size=32768 Flags=3 Head=0 Next=1
VirtioAppendDesc PhysAddr=3FBC603F Size=1 Flags=2 Head=0 Next=2
virtio_queue_notify vdev VDEV vq VQ#0
virtqueue_pop vq VQ#0 elem EL#0 in_num 2 out_num 1
bdrv_aio_readv bs BDRV sector_num 585792 nb_sectors 64 opaque REQ#0
virtio_blk_rw_complete req REQ#0 ret 0
virtio_blk_req_complete req REQ#0 status 0
SynchronousRequest Lba=585792 BufSiz=32768 ReqIsWrite=0 Crc32=1EEB2B07
(The Crc32 was double-checked with edk2's and Linux's guest IDE driver.)
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@14356 6f19259b-4bc3-4df7-8a09-765794883524
Also, add a small delay after the 0xCF9 hard reset request -- on qemu/kvm the
port access is translated to the qemu-internal system reset request by the CPU
thread, and it might progress some more before the IO thread acts upon the
system reset request.
MicroSecondDelay() is implemented by OvmfPkg's own AcpiTimerLib.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@14158 6f19259b-4bc3-4df7-8a09-765794883524
The reset requested via the keyboard controller (port 0x64) is actually a
soft reset, but qemu has supported it since forever (plus qemu has not
distinguished between hard reset and soft reset, although this is changing
now). Therefore leave the current IoWrite() in place for compatibility.
On qemu versions with commit 1ec4ba74 ("PIIX3: reset the VM when the Reset
Control Register's RCPU bit gets set"), use the PIIX3 RCR as first choice.
In the future qemu will act differently on soft vs. hard reset requests,
and we should honor that in ResetCold().
Writing to ioport 0xCF9 on qemu builds prior to commit 1ec4ba74 should
have no effect. Access to the PCI host config register went through
several implementations in qemu. Commit 9f6f0423 ("pci_host: rewrite
using rwhandler") seems safe, both before and after.
Commit d0ed8076 ("pci_host: convert conf index and data ports to memory
API") inadvertently dropped the alignment/size check, causing a boot
regression on NetBSD. It was fixed about six months later in commit
cdde6ffc, which is current. Translating that to qemu releases, the bug
was visible from v1.0 to v1.1.0.
On physical hardware cycling between reset methods is sometimes necessary
<http://mjg59.dreamwidth.org/3561.html>. On qemu the port access should
trap immediately.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@14157 6f19259b-4bc3-4df7-8a09-765794883524
Usage of the EFI entry point was made feasible in the kernel
x64 boot protocol 2.12 where a 32-bit & 64-bit entry point
became well defined.
http://git.kernel.org/linus/09c205af
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@14132 6f19259b-4bc3-4df7-8a09-765794883524
The Xcode assembler is much pickier than GCC. Also the 64-bit
linker is not a fan of relocations so it is better to us IP
relative code, but at least it removes a relocation entry.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Andrew Fish <afish@apple.com>
[jordan.l.justen@intel.com: use .byte for retfq rather than lret]
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@14055 6f19259b-4bc3-4df7-8a09-765794883524
We're supposed to zero everything in the kernel bootparams that we don't
explicitly initialise, other than the setup_header from 0x1f1 onwards
for a precisely defined length, which is copied from the bzImage.
We're *not* supposed to just pass the garbage that we happened to find
in the bzImage file surrounding the setup_header.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@14052 6f19259b-4bc3-4df7-8a09-765794883524
Boot protocol 2.05 just means that the relocatable_kernel field is present
in the header. We should actually check that it's *set*.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@14051 6f19259b-4bc3-4df7-8a09-765794883524
We currently just jump to offset 0x200 in the kernel image, in 64-bit
mode. This is completely broken. If it's a 32-bit kernel, we'll be
jumping into the compressed data payload.
If it's a 64-bit kernel, it'll work... but the 0x200 offset is
explicitly marked as 'may change in the future', has already changed
from 0x100 to 0x200 in the past with no fanfare, and bootloaders are
instructed that they should look at the ELF header to find the offset.
So although it does actually work today, it's still broken in the
"someone needs to whipped for doing it this way" sense of the word.
In fact, the same bug exists in other bootloaders so the 0x200 offset
probably *is* now set in stone. But still it's only valid to use it if
we *know* it's a 64-bit kernel. And we don't. There *is* no ELF header
that we can look at when we're booting a bzImage, and we can't rely on
it having a PE/COFF header either.
The 32-bit entry point is always guaranteed to work, and we need to
support it anyway. So let's just *always* use it, in 32-bit mode, and
then we don't have to make up some horrible heuristics for detecting
32-bit vs. 64-bit kernels.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@14045 6f19259b-4bc3-4df7-8a09-765794883524
If QEMU's -kernel parameter was used, then download the
kernel from the FwCfg interface, and launch it. (See -kernel,
-initrd, -append) The application uses the LoadLinuxLib to boot
the kernel image.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@13923 6f19259b-4bc3-4df7-8a09-765794883524
This code is based on efilinux's bzimage support.
git://git.kernel.org/pub/scm/boot/efilinux/efilinux.git
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Matt Fleming <matt.fleming@intel.com>
git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@13922 6f19259b-4bc3-4df7-8a09-765794883524
AppendDesc() should have a prefix implying its containing library,
VirtioLib. Update its sole client VirtioBlkDxe.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@13843 6f19259b-4bc3-4df7-8a09-765794883524
Laszlo Ersek
Mauro Faccenda
Olivier Martin
Matt Fleming
Jordan Justen
Ruiyu Ni
jljusten
andrewfish