Override 'try_alloc/free' because ethernet frame headers are 14 bytes
(src/dst mac (12) + ethertype (2)) causing the IP header to be 2 byte
aligned, leading to problems on platforms that require load/store
operations to be naturally aligned when reading, for example, 4 byte IP
addresses. Therefore, we align the allocation to 2 bytes, so the IP
header is aligned to 4.
issue #4312
The key changes in this patch are:
* Buffer allocation is moved into a separate Buffer_pool helper. The
implementation of the buffer allocation strategy does not change.
The helper allocates a single RAM dataspace and splits it in multiple,
equally sized chunks.
* Management of main descriptor ring is enacapsulated in Descriptor_ring
helper class.
* Use separate RAM dataspaces for descriptor rings and buffers.
Previously both of them were packed into a single dataspace. This
might have been more RAM efficient, but IMO it made the code uglier and
harder to understand.
* All of the VirtIO::Queue members are now initialized on the class member
initializer list. This is possible due to previously listed changes.
* Since all VirtIO::Queue members are initalized on member initalizer
list, some additional ones can be marked as const, ex _avail, _used ring
pointers.
* Move descriptor writing code into a common method used by both
write_data and write_data_read_reply members. This avoids some code
duplication between those methods.
* Get rid of request_irq argument that most public VirtIO::Queue methods
accept. None of the existing drivers use it and I doubt this will
change any time soon.
* Use Genode namespace by default in Virtio.
This patch also fixes at least one bug that I discovered while working
on VirtIO block device driver. Namely, when chaining descriptors only the
first descriptor in the chain should be exposed in the available ring.
Issue #4347
The const-variant of the data() method contained an erroneous
calculation of the tail size. This led to the size guard throwing
exceptions when trying to parse TCP packets that only contained the
TCP header.
Fixesgenodelabs/genode#4340
Thanks to Piotr Tworek for the fix and his explanation as follows:
The basic idea is to try to fit payload data into the descriptor used to
send the header. If there is no payload, or the payload fits exactly
into the remaining space in the header decriptor, len should be 0 and
only one descriptor should be used. In such case the "next" and "flags"
members of the descriptor structure should be set to 0.
In case there is some extra payload data to send, but its size is
bigger than the remaining free space in the descriptor used to send the
header, len should contain the remaining size of the payload that
can't be sent via the header descriptor. The code will then chain
additional descriptors to handle this remainder.
With the len variable shadowing, the code will never queue the remaining
data.
Issue #4327
Some more advanced devices like VirtIO GPU do expect they can receive
responses to VirtIO commands they issue via VirtIO queue. Such responses
are not sent via a separate device writeable queue. Instead the driver
is expected to queue some additional descriptors and buffers which the
device can then use to provide the reply.
This patch adds support for such write-data-read-response opeartion to
Genode VirtIO::Queue implementation. The implementation is pretty simple
and does not support any fancy features like receiving the response
asynchronously. Instead the operation will use caller provided callback
to wait for the device to process the command. Once this callback
returns the write-data-read-response VirtIO::Queue function will invoke
another callback passing received response as argument.
Mesa queries information about the underlying device and this header
denotes the layout of the information. It is also used by the driver
itself to populate the 'info_dataspace'.
Issue #4329.
This patch changes the 'Allocator' interface to the use of 'Attempt'
return values instead of using exceptions for propagating errors.
To largely uphold compatibility with components using the original
exception-based interface - in particluar use cases where an 'Allocator'
is passed to the 'new' operator - the traditional 'alloc' is still
supported. But it existes merely as a wrapper around the new
'try_alloc'.
Issue #4324
The new interfaces are meant to gradually replace the existing
Gpio_session interface.
- Each session refers to a single pin.
- The session types distiguish the direction of the signal as input or
output.
- Pin coordinates can be selected via session labels.
- GPIO interrupts are covered by the regular IRQ session interface.
The interfaces are accompanied by framework utilities and interfaces:
- os/pin_driver.h
- pin_control_session/component.h
- pin_state_session/component.h
These headers relieve GPIO drivers from implementing boilerplate code by
providing device-agnostic portions. The A64 pio driver serves as
reference for using those utilities.
https://github.com/nfeske/genode-allwinner/tree/master/src/drivers/pin/a64Fixes#4315
Allocating and mapping buffers not only consumes RAM quota, it consumes
CAP quota as well. Extended the Gpu session to allow for dealing with
that on the client side.
On a side note, the amount of initial CAP quota needed to establish
a connection is increased to cover the current costs of the Intel
GPU multiplexer.
Issue #4284.
This patch introduces a C API to be used by input drivers to generate
Genode events. The initial version is limited to multitouch events only.
Fixes#4273
The former implementation did not internally track ROM changes notified
vs. delivered to the client. We adapt the versioning implementation
implemented in dynamic_rom_session.h and enable explicit notification of
the current version.
The feature is used by the clipboard to notify permitted readers of the
clipboard ROM service on focus change via the newly created private
Rom::Module::_notify_permitted_readers() function.
Fixes#4274
The various mapping methods are modelled after the requirements of
the Intel GPUs or rather the Mesa driver back end.
With upcoming support for other driver back ends, we need to
sequeeze their requirements in as well. For now hijack 'map_buffer'
to provide for specifying the kind of attributes the client needs.
For now all buffers mapped in the GGTT for Intel GPUs are treated
as RW.
Issue #4265.
This call allows for checking if the given execution buffer has been
completed and complements the completion signal. Initially the GPU
multiplexer always sent such a signal when the currently scheduled
execution buffer has been completed. During enablement of the 'iris'
driver it became necessary to properly check of sequence number.
In case of the Intel GPU multiplexer the sequence numbers are
continous, which prompted the greater-than-or-equal check in the
DRM back end. By hidding this implementation detail behind the
interface, GPU drivers are free to deal with sequence numbers any
way they like and allows for polling in the client, where the
completion signal is now more of a progress signal.
Issue #4265.
The current info implementation (as RPC) is limited in a few ways:
* The amount of data that may be transferred is constrained by the
underlying base platform
* Most information never changes during run time but is copied
nonetheless
* The information differs depending on the used GPU device and
in its current implementation only contains Intel GPU specific
details
With this commit the 'info' RPC call is replaced with the
'info_dataspace' call that transfers the capability for the dataspace
containing the information only. This is complemented by a client
local 'attached_info' call that allows for getting typed access to
the information. The layout of the information is moved to its own
and GPU-specific header file, e.g., 'gpu/info_intel.h'
Issue #4265.
Rather than using the dataspace capability directly, let the client
choose its own local identifier that is linked to the underlying
capability.
Fixes#4265.
The contents of those descriptor rings can be modified by the device.
Mark them as volatile so the compiler does not make any assumptions
about them.
Issue #4264
Both, trace_logger and vfs_trace had their own trace_buffer.h. This
commit consolidates the existing implementations and provides the
resulting trace_buffer.h at 'include/trace/'. It thereby becomes part of
the trace api archive.
genodelabs/genode#4244
With this commit, the NIC router DHCP client reads out the first DNS domain
name (DHCP option 15) if any from a DHCP reply that generates an IPv4 config
for a domain and stores the name together with the IPv4 config for that domain.
DNS domain names are reported via the new report tag '<dns-domain>' if the
'config' attribute in the config tag '<report>' is set.
Furthermore, the NIC router DHCP server becomes able to obtain a DNS domain
name from another domain that has a DHCP client dynamically (given the config
attribute 'dns_config_from' is set and no static DNS config is given) or
statically from its configuration (new config tag '<dns-domain>') and propagate
this name with DHCP replies (DHCP option 15).
The 'nic_router_dhcp_*' tests are adapted to test the new feautures.
The commit also gets rid of some mirrored files in
'test/nic_router_dhcp/manager'.
Fixes#4246
The NIC router DHCP server used to add an extra option 6 field to DHCP replies
for each DNS server address. This conflicts with RFC #2132 section 3.8 which
states that the addresses should be listed within one option 6 field without
delimiter. The discrepancy is fixed by this commit.
Ref #4242
The former declaration of the IPv4 packet did not only use the questionable
tool of implementation-defined C++ bitsets but also lacked access to flags
"don't fragment" (DF) and "more fragments" (MF). This commit replaces the
C++ bitsets by using the register framework and introduces accessors for the
missing flags.
Ref #4236
This commit introduces a C-API to the Uplink session, as well as to
serve as a Block service. It can be used by drivers ported from
C-only projects, like the Linux kernel, or BSD kernels for instance.
Fix#4226
Adapts Dir_file_system::open_composite_dirs in a way that it returns "success"
when the leaf node of the path is an empty directory but "lookup failed", as
usual, if one of the other directories on the way to the leaf node is empty.
I couldn't find a technical reason why we used to return "lookup failed" when
only the leaf node was empty.
The commit also adds a test for en empty root directory and empty
sub-directories to the fs_query run script.
Fixes#4198
Introduces the notion of a transaction that consists of one or more
messages. Whereby a message has a read or write direction and consists
of one or more bytes.
Issue #4170Fixes#4169
This patch moves the utility from the app/text_area to os/vfs.h to make
it easier to use by other components. By hosting the 'New_file' as a
friend alongside the 'Directory', we can now pass a 'Directory' as
constructor argument, which is consistent with other utilities such as
'File_content'.
As a further improvement, the new version supports the implicit creation
of the directory hierarchy leading to the new file.
Issue #4032
For fs_file_systems, reads are limited to the size of the packets from the
File_system session. Hence, we cannot read the large files in one go.
This fix is particularly helpful for fonts_fs, as it enables including font
files from a File_system.
genodelabs/genode#4135
With the update to GCC 10, the compiler stopped with an error when compiling
places where a MAC address is copied from outside into a packed object using
the Net::Netaddress::copy method (e.g. in
Net::Arp_packet::dst_mac(Mac_address)):
! error: writing 6 bytes into a region of size 4 [-Werror=stringop-overflow=]
While trying to find a clean solution for this error, I found posts on
gcc.gnu.org and github that stated that the size calculations that cause these
errors are incorrect. Indeed, I could verify that the actual size of the two
regions was static and exactly the same in places were the error occured.
Furthermore, I couldn't find a way of making it more clear to the compiler
that the sizes are the same. By accident, we found that using the address of
the first element of the array that forms the second region instead of the
array address itself, somehow circumvents the error.
Fixes#4109
With the update to GCC 10 the compiler used to warn when using the internet
checksum functions on packet classes (like in
Net::Ipv4_packet::update_checksum):
warning: converting a packed ‘Net::[PACKET_CLASS]’ pointer
(alignment 1) to a ‘const uint16_t’ {aka ‘const short
unsigned int’} pointer (alignment 2) may result in an
unaligned pointer value
Apparently, the 'packed' attribute normally used on packet classes sets the
alignment of the packet class to 1. However, for the purpose of the
internet-checksum functions, we can assume that the packet data has no
alignment. This is expressed by casting the packet-object pointer to a pointer
of the new packed helper struct 'Packed_uint16' that contains only a single
uint16_t member before handing it over to the checksum function (instead of
casting it to a uint16_t pointer).
Ref #4109
Introduce a new _overflowed state variable to indicate whether the
horizontal boundary was reached already and to omit subsequent character
output.
This state is necessary to maintain a valid cursor position at all
times. The _overflowed attribute is reset once the cursor is moved into
a valid position again.
To harmonise the bounds checking for _cursor_pos modifications, the
`constrain()` method was added.
Fixesgenodelabs/genode#4093
Also fixes a bug in `vpa()` and `vpb()` which moved the cursor
horizontally instead of vertically.
* The device XML information dataspace is only provided,
when the client's policy states `info="yes"`
* The device XM information gets changed to include the
physical resource names (I/O memory and IRQ addresses)
instead of virtual ids and page offset
Fix#4077
This is a follow-up fix for "Streamline platform-device API on ARM".
There is an ambiguity of the 'local_addr' method between the inherited
'Attached_dataspace' and the local declaration, which results in the
double application of the sub-page Range::start.
Issue #4075
This API rework eases the access to memory-mapped I/O registers and
interrupts when using the platform driver. It introduces the notions of
- Platform::Device - one device obtained from a platform session
- Platform::Device::Mmio - locally-mapped MMIO registers of a device
- Platform::Device::Irq - interface for receiving device interrupts
The patch touches several drivers. Some drivers would require a
significant structural change to adopt the new API (e.g., net/virtio,
dde_linux drivers, imx gpio). In these cases, the patch adds
compatibility shims meant to be temporary. In other cases (e.g., imx
i2c), the adaptation was simple enough to carry through.
Fixes#4075
This patch extends the 'Platform_session::alloc_dma_buffer' interface
with a 'Cache' argument that corresponds to the argument accepted by
'Ram_allocator::alloc', which is used by the platform driver under the
hood.
Since the x86 platform driver used to be hardwired to allocate DMA
buffers as UNCACHED, I adjusted all drivers by specifying the UNCACHED
argument. Right now, this is needed as a hint for core to steer the
allocation of I/O page tables. Once we eliminate the need for such hints
(by introducing an explicit 'Region_map::attach_dma' operation), we can
revisit the drivers individually because cached DMA buffers should
generally be fine on the x86 architecture.
Issue #2243
This change avoids many repetetive Genode:: prefixes, making the code
easier to read. The patch also includes a few consistency fixes
regarding include guards and file headers. It also renames
Platform_device::String to Platform_device::Device:name.
Issue #2243
This patch adds the designated alternative to Dataspace::phys_addr to
the platform-session interface. Under the hood, the platform driver
still calls Dataspace::phys_addr but it should eventuelly become the
only caller before we can abolish this function.
Issue #2243
This patch changes the 'alloc_aligned' interface as follows:
- The former 'from' and 'to' arguments are replaced by a single
'range' argument.
- The distinction of the use cases of regular allocations vs.
address-constrained allocations is now overed by a dedicated
overload instead of relying on a default argument.
- The 'align' argument has been changed from 'int' to 'unsigned'
to be better compatible with 'addr_t' and 'size_t'.
Fixes#4067
The vfs pipe plugin can now be used as named pipe which anables data
transfer via file handles from one component to another. E.g. if one
would like to send data from component A to stdin of a libc component B,
one can do so by simply writing to that fifo file.
Issue #3583
* Update the 'packet_size' information with the actual length for
each isoc frame to be able to handle short reads at the client side.
* Copy the whole transfer buffer because the host controller stores
the data at the original offsets, i.e., the buffer is not densely
packed.
Fixes#4018.
Adds new Uplink session interface, the corresponding client side (Client,
Connection), and the corresponding API archives. An Uplink session is almost
the same as a NIC session with the difference that the roles of the end points
are swapped. An Uplink client is the one that provides a network interface
(for instance, a NIC driver) whereas an Uplink server is the one that uses
that network interface (for instance, a networking stack).
Therefore, in contrast to the NIC session, MAC address and link state come from
the Uplink client. The link state is reflected through the lifetime of an
Uplink session: The client requests the session only when the link state is
"UP" and closes it whenever the link state becomes "DOWN" again. The MAC
address is transmitted from the Uplink client to the Uplink server as an
argument of the session request.
Ref #3961
This commit restores the diag feature for selecting diagnostic output of
services provided by core. This feature became unavailable with commit
"base: remove dependency from deprecated APIs", which hard-wired the
diag flag for core services to false.
To control this feature, three possible policies can be expressed in a
routing target of init's configuration:
* Forcing silence by specifying 'diag="no"'
* Enabling diagnostics by specifying 'diag="yes"'
* Forwarding the preference of the client by omitting the 'diag'
attribute
Fixes#3962
- Vendor devices add addtional data to the config descriptor, read and
added to the usb session
- allow '0' configuration within the usb session
issue #3822
To simplify writing native VirtIO drivers for Genode add helper classes
representing VirtIO device and queue. The queue implementation should
be platform independant. The device abstraction however is closely tied
to the VirtIO transport being used (PCI/MMIO). Both PCI and MMIO
implementations expose the same public API so the actual driver logic
should be the same regardless of which transport is used.
Its also important to note that the PCI version of Virtio::Device
currently does not support MSI-X interrupts. Unfortunately my kowledge
about PCI bus is very limited and my main area of interest was to get
VirtIO drivers working on virt_qemu ARM/Aarch64 platform. As such all
the VirtIO drivers I plan to submit will work with PCI bus, but might
not use some extended capabilities.
Ref #3825
- make GPIO server more robust on imx by not throwing exceptions for
unknown pins, use '_with_gpio' instead
- use 'Gpio::Pin' data type instead of POD 'unsigned'
issue #3900
The combination of Net::Mac_address and
Genode::ascii_to(Net::Mac_address) required shaky quirks in several
places because GCC is not able to resolve the ascii_to overload if
base/xml_node.h was included to early. The current solution moves the
several ascii_to overloads "closer" to the Net types by putting them
into the Net namespace, where GCC reliably picks them up.
Hence, co-locating the ascii_to() utility with the overload type in the
same scope/namespace is good practice.
This patch removes the now obsolete <nic/xml_node.h> header file.
- base/cancelable_lock.h becomes base/lock.h
- all members become private within base/lock.h
- solely Mutex and Blockade are friends to use base/lock.h
Fixes#3819
Until now, Genode's framebuffer session interface was based on the
RGB565 pixel format. This patch changes the pixel format to 32-bit
XRGB where the X part is ignored. It adapts all graphical applications
and device drivers accordingly.
The patch also adjusts the users of the drivers_interactive packages,
assigning 64 MiB RAM and 1500 caps to the drivers subsystem, which is
sufficient for covering high resolutions at 32 bits per pixel and to
accommodate multi-component USB HID input stacks.
Fixes#3784
- Since Genode::strncpy is not 100% compatible with the POSIX
strncpy function, better use a distinct name.
- Remove bogus return value from the function, easing the potential
enforcement of mandatory return-value checks later.
Fixes#3752
This patch changes the 'Single_file_system' to return NO_PERM only if
the to-be-unlinked file corresponds to the single file. This way, a
<rom> co-mounted with a <ram> file-system does not stand in the way of
unlinking files from the <ram>. The concrete symptom occurred the
following scenario:
<vfs>
<dir name="home">
<ram/>
<rom name="..."/>
</dir>
</vfs>
The following sequence of commands wrongly resulted in "Operation not
permitted":
$ mkdir -p /home/a/b/c
$ rm -f /home/a/b/c/d
In this case, rm should not fail (unlink should return ENOENT)
Fixes#3690
When specifying "/" or "" as rel_path to the 'Directory' constructor,
the constructed directory should refer to the same directory. The
implementation of the join utility did not consider this corner case. It
occurred during the attempt to use fs_query with "/" given as path.
This patch also adds a Directory::Entry::dir accessor that returns true
if the entry is a directory.
Fixes#3630
This patch adds a special variant of a 'Content_producer' called
'Xml_producer', which allows the direct use of an 'Xml_generator'
for generating the ROM-module content.
This patch extracts the child-management functionality from the init
component into a new library called "sandbox". The library API is
located at 'os/include/os/sandbox.h'.
The sandbox API allows for the interaction of the component with the
sandboxed children by providing locally implemented services. This
mechanism is illustrated by the new test at os/src/test/sandbox.
Issue #3601
This patch extends the 'Buffered_xml' utility with a new constructor
that fills the buffer with the output of an 'Xml_generator'. It thereby
presents an easy way to generate XML to be consumed locally.
The patch also add a deprecation mark to the original 'xml' accessor
because copying 'Xml_node' objects (here as return value) is dangerous.
The new 'with_xml_node' method should instead be used to access the XML
content stored in the buffer.
Fixes#3602
`volatile` effectively prevents instruction reordering by the compiler
and fixes an issue with -O3 compiled components.
Note, this commit does not address further arguments regarding memory
barriers and volatile voiced in issue #693.
Issue #693
This is a follow-up patch for issue #1784 that solves two
inconsistencies.
- The Vfs::Timestamp::INVALID matches File_system::Timestamp::INVALID
- The Noux libc plugin tests for Timestamp::INVALID instead of a
positive value.
The patch fixes the mtime info as shown in directory listings in
Sculpt's inspect window.
Add a new plugin for creating pipes between pairs of VFS handles. It is
intended to replace the libc_pipe plugin, one of the last remaining libc
plugins.
In contrast to the libc_pipe plugin, this plugin defers cross-handle
notification until I/O signal handling rather than block and unblock
readers using a semaphore. This is a performance regression in the case
of multiple threads blocking on a pipe, but shall be an intermediate
mechanism pending renovations within the libc VFS and threading layers.
As a side effect, threads blocked on a pipe might not be resumed until
the main thread suspends and dispatches I/O signals.
The "test-libc_pipe" test has been adjusted to use the VFS pipe plugin
and tests both local pipes and pipes hosted remotely in the VFS server.
Merge adaptations (such as EOF handling, adjustment to VFS/libc
interface changes) by Norman Feske.
Fix#2303
This change allows for the conversion of directory entries in place
(i.e., in the VFS server) and anticipates the potential future merge of
both types into one.
Issue #3547
This patch adds support for watch notifications for the
'Readonly_value_file_system', which is often used by VFS plugins to
implement pseudo files. It thereby enables VFS clients to respond to
VFS-plugin events (think of terminal resize) dynamically.
Fixes#3523
When the construction of a member of Packet_stream_*::Rpc_object after
the _cap member threw an exception, the object was not dissolved from
the entrypoint although the Rpc_object vanished at this point. This was
because the call to 'manage()' happened in the initializer list (for the
_cap member instantiation). The destruction of the _cap member then did
not dissolve the object.
This first fix moves the call to 'manage()' into the constructor body
after the instantiation of all other members. A more sophisticated fix
would use some kind of 'Managed_object' life-time guard that manages an
object on construction and dissolves on destruction.
Ref #3525
Nowadays, we use standard command-line tools like vim to edit init
configurations dynamically, which alleviates the need for a custom CLI.
The CLI-monitor component was too limited for use cases like Sculpt
anyway.
The patch also removes the ancient (and untested for long time)
terminal_mux.run script, which used to be the only remaining user of the
CLI monitor.
Issue #3512
This patch extends the 'File_system::Status',
'File_system::Directory_entry', and the related 'Vfs' types with
the following additional information:
- Distinction between continuous and transactional files (Node_type)
(issue #3507)
- Readable, writeable, and executable attributes (Node_rwx),
replacing the former 'mode' bits
(issue #3030)
The types 'Node_rwx', 'Node_type' are defined twice,
once for the VFS (vfs/types.h) and once for the 'File_system'
session (file_system_session/file_system_session.h).
Similarly, there is a direct correspondance between
'Vfs::Directory_service::Dirent' and 'File_system::Directory_entry'.
This duplication of types follows the existing pattern of keeping the
VFS and file-system session independent from each other.
The Press event is actually a Press_char event with a default codepoint.
The default codepoint is now
Codepoint { Codepoint::INVALID } /* value 0xfffe */
in contrast to
Codepoint { Input::Event::INVALID } /* value 0 */
Issue #3483
A client may register a signal handler to be notified whenever the
RTC value was changed, i.e., a mis-configured clock was synchronized,
by calling 'set_sigh()'.
Issue #3450
* Make target binaries independent of board SPECS
* Name binaries of one architecture unambigously
* Extend include path to match board specifics
* Adapt run-scripts to use the right binary
Ref #2190
Ref #3180
To enable the use of uncached DMA buffers as RX and TX communication
buffers in between driver (service) and client, introduce a cache
attribute in the constructor of Nic::Session_component
Ref #3291
This patch equips the 'Block::Connection' with a framework API for the
implementation of robust block-session clients that perform block I/O in
an asynchronous fashion.
An application-defined 'JOB' type, inherited from 'Connection::Job',
encapsulates the application's context information associated with a
block operation.
The lifecycle of the jobs is implemented by the 'Connection' and driven
by the application's invokation of 'Connection::update_jobs'. The
'update_jobs' mechanism takes three hook functions as arguments, which
implement the applications-defined policy for producing and consuming
data, and for the completion of jobs.
Issue #3283
As a preparatory step for introducing the new block-client API, we have
to turn the 'Block::Connection' into a class template. The template
argument will be used to tie an application-defined job type to the
block connection.
Issue #3283
This patch adds support for manually triggering the wakeup of the packet
sink by the source. This way, a packet source becomes able to marshal
batches of submissions or unmarshal batches of acknowledgements before
yielding the control over to the sink.
Issue #3283
This patch removes the blocking Block::Session::sync RPC function and
adds the asynchronous operations SYNC and TRIM to the block session's
packet-stream interface.
Even though the patch adjusts all block components to the interface
change, the components keep the former blocking handling of sync
internally for now because of the design of the 'Block::Driver'
interface. This old interface is not worth changing. We should instead
migrate the block servers step by step to the new
'Block::Request_stream' API.
Fixes#3274
The new request tag allows a block-session client to uniquely correlate
acknowledgements with outstanding requests. Until now, this was possible
for read and write operations by taking the value of the request's
packet-stream offset. However, SYNC and TRIM requests do not carry any
packet-stream payload and thereby lack meaningful offset values. By
introducing the notion of a 'tag', we can support multiple outstanding
requests of any type and don't need to overload the meaning of the
'offset' value.
Issue #3274
This patch splits the 'Request' definition into smaller types that are
suitable for the client-side API too.
The new 'Operation' type comprises the block operation's type (opcode)
and the operation's arguments (block number, block count).
The former 'Request::operation_defined' is now 'Operation::valid'.
The 'Request' aggregates an 'Operation', which changes its object
layout.
Note that this commit relaxes the bit-precise definition of 'Request' to
facilitate the use of 'unsigned long' where appropriate, in particular
for the request tag (which should correspond to an 'Id_space::Id'). The
originally bit-precise definition was pursued to allow the sharing of
the 'Request' type between SPARK and C++ code. However, it turns out
that defining a native type in each language and a (set of) converting
constructors is a more natural approach.
Issue #3283
This patch renames 'wakeup_client' to 'wakeup_client_if_needed' to
clarify that the method triggers signals only when needed, not on every
call.
The name 'wakeup_client' is prone to misguide users to call the function
conditionally as an optimization, thereby complicating the code, but to
no effect.
Fixes#3279
This patch replaces the formerly fixed 2 KiB data alignment within the
packet-stream buffer by a server-defined alignment. This has two
benefits.
First, when using block servers that provide small block sizes like 512
bytes, we avoid fragmenting the packet-stream buffer, which occurs when
aligning 512-byte requests at 2 KiB boundaries. This reduces meta data
costs for the packet-stream allocator and also allows fitting more
requests into the buffer.
Second, block drivers with alignment constraints dictated by the
hardware can now pass those constraints to the client, thereby easing
the use of zero-copy DMA directly into the packet stream.
The alignment is determined by the Block::Session_client at construction
time and applied by the Block::Session_client::alloc_packet method.
Block-session clients should always use this method, not the 'alloc_packet'
method of the packet stream (tx source) directly. The latter merely
applies a default alignment of 2 KiB.
At the server side, the alignment is automatically checked by
block/component.h (old API) and block/request_stream.h (new API).
Issue #3274
This patch modernizes the 'Block::Session::info' interface. Instead of
using out parameters, the 'init' RPC function returns a compound 'Info'
object now. The rather complicated 'Operations' struct is replaced by
a 'writeable' attribute in the 'Info' object.
Fixes#3275
Replace the I/O response handler that is passed to the VFS at
construction with an object that is dynamically attached to handles.
This object shall also accept read-ready notifications, and plugins are
encouraged to keep handles awaiting ready-ready notifications separate
from handles that await I/O progress.
Replace the use of handle lists in plugins with handle queues, this
makes the code easier to understand and the ordering of notifications to
the application more explicit.
These changes replace the use of the Post_signal_hook from all VFS
plugins, applications must assume that read-ready and I/O notifications
occur during I/O signal dispatch and use an Io_progress_handler at its
entrypoints to defer response until after signal dispatching.
Fix#3257
