during receive the nic_ep may block as long as the guest does not provide
another receive network descriptor. In the meantime, all Genode signals
regarding the network interface, e.g. tx, will be postponed, which may
effect the throughput.
Instead use the nic_ep for rx packets unblocking. Add an notification mechanism
to the e1000 vbox network model, to notify us as soon as the guest added new
receive descriptors in the model.
Issue #5146
This commit replaces the mapping of DMA buffers and gets rid of the
bounce-buffer handling, which was introduced to prevent data
corruption noticed when utilizing USB storage with Windows 10 guests,
with accessing the buffers directly.
Due to the way Windows 10 at times manages its DMA memory (many small
pieces instead of few larger ones) the unbounded registry becomes a
problem when containing stale entries.
Changing the 'qemu-usb' implementation allows for using 'read_dma'
and 'write_dma' directly.
Fixes#5121.
The classes Genode::Mmio, Genode::Register_set, Genode::Attached_mmio, and
Platform::Device::Mmio now receive a template parameter 'size_t SIZE'. In each
type that derives from one of these classes, it is now statically checked that
the range of each Genode::Register::Register- and
Genode::Register_set::Register_array-deriving sub-type is within [0..SIZE).
That said, SIZE is the minimum size of the memory region provided to the above
mentioned Mmio classes in order to avoid page faults or memory corruption when
accessing the registers and register arrays declared inside.
Note, that the range end of a register array is not the end of the last item
but the end of integer access that is used for accessing the last bit in the
last item.
The constructors of Genode::Mmio, Genode::Attached_mmio, and
Platform::Device::Mmio now receive an argument 'Byte_range_ptr range' that is
expected to be the range of the backing memory region. In each type that derives
from on of these classes, it is now dynamically checked that 'range.num_bytes
>= SIZE', thereby implementing the above mention protection against page faults
and memory corruption.
The rest of the commit adapts the code throughout the Genode Labs repositories
regarding the changes. Note that for that code inside Core, the commits mostly
uses a simplified approach by constructing MMIO objects with range
[base..base+SIZE) and not with a mapping- or specification-related range size.
This should be fixed in the future.
Furthermore, there are types that derive from an MMIO class but don't declare
any registers or register arrays (especially with Platform::Device::Mmio). In
this case SIZE is set to 0. This way, the parameters must be actively corrected
by someone who later wants to add registers or register arrays, plus the places
can be easily found by grep'ing for Mmio<0>.
Fix#4081
Hitting a BOSD in a Windows guest in Virtualbox 6 leads to calling
`DBGFR3ReportBugCheck`, which leads to the following error since it is
unimplemented in our port:
```
Error: DBGFR3ReportBugCheck: DBGFR3ReportBugCheck called, not implemented, eip=0x1a5c4e4
```
Including the actual method from Virtualbox 6 enables logging
diagnostics for the Windows Blue Screen of Death (BSOD).
To see Blue Screen logging output, enable "dbgf" and "gim" logging
via the `VBOX_LOG` following environment variable in the vbox6 runtime
config.
When manually triggering a BSOD in the Windows guest, this should
produce the following logging diagnostics:
```
GIM: HyperV: Guest indicates a fatal condition! P0=0xe2 P1=0x0 P2=0x0 P3=0x0 P4=0x0
GIMHv: BugCheck e2 {0, 0, 0, 0}
MANUALLY_INITIATED_CRASH
```
Fixes#5099
Uploading large files via HTTP PUT failed when the file system was
accessed via a file system session because lighttpd opened the
destination file with O_NONBLOCK and the write operation stalled
at some point.
genodelabs/genode#5093
When multiple threads (EMT-0..X + nic_ep) enter the very same
critical section, the use of RTCritSectTryEnter may reflect the contention
case to the Network Model (E1000). Since no one notifies the model, when the
critical section is free again, solely the next packet/event triggered by the
guest will resume the former operation. This may lead to long delays until
packets are sent actually.
Instead of using the RTCritSectTryEnter use RTCritSecEnter to avoid the
situation. All of our network code is non blocking, so the network backend
will only be contented a short time.
Follow up commit to
Issue #5045
- finished tx ack queue should be checked before new allocations
- packets which got not sent must be released in packet stream,
otherwise the network packet stream gets filled up and starves after a while
- rRegister for ack avail packets and process them concurrently to EMT-* threads
by nic_ep thread (thanks @Peter for the findings) + add synchronization.
- add sigh_ready_to_submit to network adapter to improve latency by notifying
the network model explicitly in case we had a full packet stream error case
(_retry resp. VERR_TRY_LATER)
Fixes#5045
Per Affinity::Location a system control cap can be requested. The capability
provides an RPC interface to request and set Cpu_state, as provided by the
former Pd::managing_system(Cpu_state) method. Invocation of those system
control capabilities then *can* (see below) be executed on the desired CPU
as described by Affinity::Location.
The system control cap will be invalid for kernels that don't support
system_control/managing_system functionality at all.
The system control cap will be ever by the same, e.g. ignoring the
Affinity::Location parameter, if the used kernel doesn't support or doesn't
require the feature to execute the system control per CPU.
The commit is a preparation step to add guarded and selective x86 MSR
access per CPU.
Fixes#5009
To prevent exessive allocations the bounce buffers are allocated
once and afterwards re-used. The DMA buffers are provided by an
range allocator whose backing store increases in 4 MiB chunks.
The range allocator might not merge the ranges as those chunks
are not necessarily allocated continously. Allocations larger
than the static chunk size are therefor treated as error.
instead via the hardware registers of the FPU. On Genode all components and
so VMMs are built such, that the compiler may generate optimized code by
using the FPU at any time. We had to make sure to save the
FPU state as early as possible before the VMM component touches the FPU,
to avoid corrupting & losing guest FPU state. This caused headache again
and again. To avoid the uncertainty, we remove this feature and explicitly
transfer the FPU state via the UTCB.
When the wait value is too small the HALT attempt turnes into
busy-polling in the VMM. To prevent that always wait a minimal
amount of time.
Issue #4990.
Some Ubuntu installations (e.g., 20.04) create installation disks with
strange C/H/S MBR partitions like 3988/255/2. Normally, VirtualBox reads
the MBR to guess disk geometry information for the virtual BIOS.
Unfortuantely, the strange values from Ubuntu lead to a heavy
virtual-disk activity on boot. Therefore, this commit forces the use of
calculated values based on the assumption that large disks use LBA
addressing anyway.
Fixes#4978
This patch restricts 'Region_map::attach_executable' to create read-only
mappings, while offering the option to map the full rights using a new
'attach_rwx' method.
The 'attach_rwx' method is now used by the dynamic linker to explicitly
attach the linker area with full rwx rights. With the old page-fault
handling code, the execute flag was evaluated only for leaf dataspaces,
not for managed dataspaces while traversing region-map hierarchies.
With the new page-fault handling code, the execute bit is downgraded
to no-execute when passing a managed dataspace that is not attached
as executable.
Issue #4920
In 'SUPR3InitEx' (SUPLib.cpp) a 'SUPQUERYFUNCS' structure is allocated
with
! (PSUPQUERYFUNCS)RTMemAllocZ(SUP_IOCTL_QUERY_FUNCS_SIZE(CookieReq.u.Out.cFunctions));
where 'CookieReq.u.Out.cFunctions' is 0. To determine the size of the
allocation
! #define SUP_IOCTL_QUERY_FUNCS_SIZE(cFuncs) \
! RT_UOFFSETOF_DYN(SUPQUERYFUNCS, u.Out.aFunctions[(cFuncs)])
is used with cFuncs = 0 (SUPDrvIOC.h) leading to an allocation up to the
arrow below
! typedef struct SUPQUERYFUNCS
! {
! /** The header. */
! SUPREQHDR Hdr;
! union
! {
! struct
! {
! /** Number of functions returned. */
! uint32_t cFunctions;
! /** Array of functions. */
==> end of allocation
! SUPFUNC aFunctions[1];
! } Out;
! } u;
==> sizeof(SUPQUERYFUNCS)
! } SUPQUERYFUNCS, *PSUPQUERYFUNCS;
In sup.cc (Genode) 'ioctl(SUPQUERYFUNCS &request)' will lead to
'with_out_ioctl'
! auto &out = request.u.Out;
where auto is 'SUPQUERYFUNCS' and finally
! out = { };
will zero out 'SUPQUERYFUNCS' up to the second arrow above. Because
'RTMemAllocZ' will call 'calloc' to allocate the memory 'out = { };'
will corrupt the slab block after the allocation. Therefore, it is
reasonable to allocate at least 'sizeof(SUPQUERYFUNCS)'.
Note there might be other 'ioctl' cases like this. A better way might be
to use 'SUPQUERYFUNCS.Hdr.cbOut' to determine the 'out' size.
fixes#4675
