This commit contains features and buf fixes:
* Catch errors during resource allocation
* Because Mesa tries to allocate fence (hardware) registers for each
batch buffer execution, do not allocate new fences for buffer objects
that are already fenced
* Add support for global hardware status page. Each context additionally
has a per-process hardware status page, which we used to set the
global hardware status page during Vgpu switch. This was obviously
wrong. There is only one global hardware status page (set once during
initialization) and a distinct per-process page for contexts.
* Write the sequence number of the currently executing batch buffer to
dword 52 of the per-process hardware status page. We use the pipe line
command with QW_WRITE (quad word write), GLOBAL_GTT_IVB disabled
(address space is per-process address space), and STORE_DATA_INDEX
enabled (write goes to offset of hardware status page). This command
used to write to the scratch page. But Linux now uses the first
reserved word of the per-process hardware status page.
* Add Gen9+ WaEnableGapsTsvCreditFix workaround. This sets the "GAPS TSV
Credit fix Enable" bit of the Arbiter control register (GARBCNTLREG)
as described by the documentation this bit should be set by the BIOS
but is not on most Gen9/9.5 platforms. Not setting this bit leads to
random GPU hangs.
* Increase the context size from 20 to 22 pages for Gen9. On Gen8 the
hardware context is 20 pages (1 hardware status page + 19 ring context
register pages). On Gen9 the size of the ring context registers has
increased by two pages to 21 pages or 81.3125 KBytes as the IGD
documentation states.
* The logical ring size in the ring buffer control of the execlist
context has to be programmed with number of pages - 1. So 0 is 1 page.
We programmed the actual number of pages before, leading to ring
buffer execution of NOOPs if page behind our ring buffer was empty or
GPU hangs if there was data on the page.
issue #4260
* Wait for for completion before return from 'execbuffer2'. This makes
buffer execution synchronous.
* Because the Iris driver manages the virtual address space of the GPU
and creates one GEM context for each batch buffer we have to map/unmap
all buffer objects before and after batch buffer execution.
issue #4260
The lx_emul_virt_to_pages implementation initialized the page ref
counter only for the first page, leaving the remaining elements in
uninitialized state. This, in turn, rendered the Linux page_pool (as
used by the emac network driver) ineffective, ultimately leading the a
memory leak. The fix changes the call of 'init_page_count' to take the
loop variable as argument.
Issue #4225
Increased number of trace subjects since the test sporadically fails on
some platforms.
Also added a sanity check to print an error message in case we run into
the same issue again.
Fixesgenodelabs/genode#4261
This patch adds a switch to internal poll function in libssh that
allows to force this function to immediately return without actually
polling for data and in consequence processing this data. This switch
is used to avoid calling callback functions when flushing output
streams which caused locks due to recursive access to internal
ssh_terminal sessions registry.
Issue #4258
This patch allows to replace sftp packet read and write with
completely asynchronous versions needed to properly hook in existing
ssh_terminal implementation.
Issue #4258
Using 'alignas' in declarations might cause GCC to request for an
implementation of 'operator delete(void*, unsigned long, std::align_val_t)'
although it might actually never be called. This commit adds a dummy
implementation to 'cxx/new_delete.cc' that does nothing more than printing an
error to the log that a proper implementation is missing. This approach is
coherent with our treatment of other global delete operators.
Ref #4217
If one has an object X that has a minimum alignment requirement specified
through 'alignas' this requirement is normally inherited by objects that have
object X as member, and by those that have objects as member that have X as
member, and so on... . However, this chain used to get silently interrupted
(dropping the minimum alignment requirement to 8 again) at objects that are
managed with Genode::Reconstructible or Genode::Constructible. In order to fix
this, the commit ensures that Genode::Reconstructible (and therefore also
Genode::Constructible) has at least the minimum alignment requirement (using
'alignas') as the object it manages.
Ref #4217
The NIC router parses, stores and forwards DNS domain names from DHCP replies.
Yet the routers DHCP client used to not request DNS domain-name information on
DHCP requests. This caused DHCP servers to skip this information on their
replies although it was available. This commit fixes the issue by adding the
DNS domain name code to the request parameter list of requests from the routers
DHCP client.
The 'black_hole' component provides dummy implementations of common
session interfaces.
At this time, only the 'Audio_out' session is provided if enabled
in the configuration of the component:
<config>
<audio_out/>
</config>
Issue #3653
* Switch mesa support from DRI to gallium
Supported drivers are
- softpipe (Sebstian Sumpf)
- iris for Intel GPUs (Alexander Boetcher)
- etnaviv for Vivante GPUs (Josef Söntgen)
* Mesa's generated files are placed into 'contrib/mesa-<hash>/generated'
and are cloned per default from a separate Git repo in order to avoid
hash updates upon package build. In case you need to generate files
yourself use
! prepare_port mesa GENERATE_FILES=1
issue #4254
Support for iris and etnvaviv
* entaviv:
- libdrm on FreeBSD is not prepared for !PCI (and libc our is missing
<sys/pciio.h>
- missing <sys/types.h> include in xf86drmMode.c
- etnaviv relies on linux header files - dummy in $(INC_DIR)
- IOCTL FreeBSD ↔ Linux have swapped IO/OUT bit
- O_CLOEXEC differs between FreeBSD ↔ Linux
issue #4254
According to spec the tail pointer points to the next qword instructions
which will be used by the software.
p 1354, Doc Ref # IHD-OS-BDW-Vol 2c-11.15
issue #4254
Superpages (2M, 1G) are not supported by now, but partially copied over code
from base-hw was around. Remove unused register definitions and remove
non-working super page code do avoid confusion.
issue #4254
Size argument of ggtt free range check is ignored, which leads to
overlapping allocations inside, which leads to unavailable IO-MEM exceptions
thrown by core.
issue #4254
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
Driver code such as mfd-core.c may pass 0 as argument n to kcalloc,
which eventually results in an allocation size 0.
res = kcalloc(cell->num_resources, sizeof(*res), GFP_KERNEL);
Since 'res' is checked against NULL for success, kmalloc must not return
a NULL pointer in this case. The patch works around this issue by
forcing an allocation size of 1 byte in this case.
Issue #4253
Clock providers such as drivers/clk/sunxi-ng/ccu-sun8i-r.c don't use
regular init calls but declare their init functions via CLK_OF_DECLARE,
which fill the __clk_of_table. Linux populates the table statically by
using special sections declared in the linker script. In contrast, we
populate the table by expanding the macro to global constructor
functions.
The __clk_of_table is then processed by the call of of_clk_init(NULL).
Issue #4253
* Disable trace source and release ownership on subject destruction.
* Note, since the policy module is also destroyed on descruction of the
session component, the traced component must not access the policy
module when acknowledging the disabled state (else: page fault).
Fixesgenodelabs/genode#4247
If the trace subjects are not properly destructed when the TRACE client
disappears, enabled sources will be owned by a non-existing client.
In other words, when a TRACE client disappears all sources owned by the
client must be disabled.
genodelabs/genode#4247
test-trace always passed, although tracing was never enabled because the
trace subject was not within the first 32 subjects.
* increase number of queried subjects
* output error if trace subject was not found
genodelabs/genode#4247
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
WARNING: BREAKS CONFIG COMPATIBILITY!
This commit changes the configuration interface of the NIC router in a way that
may break systems that use the component without proper adjustment!
How to adjust:
At each occurrence of the 'dns_server_from' attribute in a NIC router
configuration replace the attribute name with 'dns_config_from'. The attribute
value remains unaltered.
DETAILED DESCRIPTION
The new attribute name 'dns_config_from' reflects that also other aspects of
the DNS configuration of the denominated domain are used by the DHCP server
that holds the attribute. This commit is a preparation for forwarding also the
domain name (DHCP option 15) with the mechanism behind the attribute.
Ref #4246
The fact that the IPv4 config was a struct with all data members public was a
mere leftover of an early state of the NIC router. Today, the router
implementation style is to avoid structs and public data members wherever
possible.
This commit slightly changes the behavior of the router regarding log output.
The router used to print malformed IPv4 configurations to the log only if
the 'verbose' config flag was set using this style:
! [my_domain] malformed dynamic IP config: interface 10.0.2.1/24 ...
Now, malformed IPv4 configurations are only printed if the
'verbose_domain_state' config flag is set (like with any IP4v configuration
states) using this style:
! [my_domain] dynamic IP config: malformed (interface 10.0.2.1/24 ...)
Fixes#4242
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
File size must be the same as the number of bytes that can be read from
the file. Otherwise, this will trigger a `Truncated_during_read`
exception.
Fixesgenodelabs/genode#4240
Via a new configuration attribute, the user can decide whether the router
should answer dropped fragmented IPv4 with an ICMP "destination unreachable"
packet and, if so, which value the ICMP code field of this packet should have.
The default is that the router doesn't send such responses (silently dropping
fragmented IPv4). The behavior is tested by the 'nic_router_ipv4_fragm' test.
Fixes#4236
If the new attribute 'dropped_fragm_ipv4' of the <report> tag in the NIC router
config is set "yes", the router will report the number of packets that were
dropped per interface respectively domain because fragmented IPv4 is not
supported. The default is not to report the counter. The behavior is tested by
the 'nic_router_ipv4_fragm' test.
Ref #4236
The NIC router used to ignore the IPv4 header fields "More fragments" and
"Fragment offset" completely. Therefore higher-level protocols of fragmented
IPv4 were interpreted wrong because each fragment was considered a self-
standing packet, expecting, for instance UDP/TCP headers somewhere inside of
the UDP/TCP data field. Normally, such packets were dropped as soon as the
UDP/TCP checksum check failed because of the misinterpretation. However,
it was also possible for fragmented IPv4 to pass the router although normally
only partially.
IPv4 fragmentation support in the router would introduce some potential
security risks and is presumably not an easy endeavor. So, for now, we settled
on not supporting IPv4 fragmentation. With this commit, the router simply drops
all fragmented IPv4. This is reflected to the log for each fragment as "drop
packet (fragmented IPv4 not supported)" when 'verbose_packet_drop="yes"' is
configured.
The new test 'run/nic_router_ipv4_fragm' is an automated test for this
behavior. The test is added to the autopilot list.
Ref #4236
- remove redundant file system factory
- remove dead code block
The code was guarded by preprocessor directives checking whether the
contrib code define "_USE_MKFS" is 1. As "_USE_MKFS" is not set one
for our port of FAT, the code was never executed and can be removed.
- remove uneffective config attributes
Apparently, the former XML attributes to the plugin 'drive' and
'codepage' had no effect. I tested them in a scenario with the VFS
block server on a disk-image boot-module as back end. Regardless of
the 'drive' value, the block session label was always "0". Regardless
of the 'codepage' value, the FAT on the disk image succeeded to mount
when not using '--codepage' for 'mkfs.fat' and failed to mount when
using '--codepage' to specify a supported but foreign codepage for
'mkfs.fat' (e.g. "720").
Ref #4220
There was one global static constructor:
! namespace Fatfs { static Constructible<Platform> _platform; }
This caused applications that used the lib or the <fatfs> VFS plugin to end up
in an uncaught exception due to Genode::Component complaining that method
'construct' returned without executing pending static constructors if they
didn't call Genode::Env::exec_static_constructors().
As the use of Genode::Env::exec_static_constructors() is discouraged in Genode,
this commit rather moves the '_platform' object to the scope of the
initializing function and introduces a global static pointer to the object that
gets set by the initializing function. Although this prevents the exception, it
is, technically speaking even worse than the former solution as the new pointer
isn't checked for validity in contrast to the 'Constructible' object.
However, so far, I don't see a clean solution to this problem without the need
for Genode::Env::exec_static_constructors().
Fixes#4220
* the GPU multiplexer now offers the platform service to the Intel
framebuffer driver (driver_manager)
* ajdusted drivers_managed-pc to hand out resources to the GPU driver
* adjust quotas
issue #4233
The platform services is intented to be used by dde_linux's intel_fb_drv
in order to initlialize displays.
* implement and announce platform session
* limit accessible GTT and aperture of client to 64 MB
* forward display engine IRQs to platform client
* move all PCI resources to 'Igd::Resources' class in order to make them
accessible by the platform service and the GPU driver
* fix fence register allocation for id zero (return true)
issue #4233
For mesa-21 the client takes care and manages
the virtual address space of the vGPU by itself and the intel/gpu driver
can't add silently a guard page anymore. Move the patch to the drm/ioctl
of the former mesa version.
Issue #4148#4233
_unmap_dataspace_ggtt requires the cap of Ggtt::Mapping (ring_map, ctx_map)
in order to find the right metadata and to free up the ggtt entries. Also the
pte range is removed already if the metadata was found.
Issue #4148#4233
BREAKS CONFIG COMPATIBILITY:
This commit changes the configuration interface of the NIC router in a way that
may break systems that use the component without proper adjustment!
HOW TO ADJUST:
At each occurrence of the '<uplink ...>' tag in a NIC router configuration
replace the tag name 'uplink' with 'nic-client'. The rest of the tag stays the
same.
The term "uplink" for network interfaces in the router that have a NIC session
client as back end was introduced in a time when Uplink sessions didn't yet
exist. Now, they do and, although both an uplink and an Uplink session
normally describe a network session between router and network device driver,
they are based on two different service types (NIC and Uplink). This can easily
cause confusion when integrating the router (the <uplink> is not related to
Uplink sessions) or trying to understand its functioning (an 'Uplink' object
has nothing to do with the Uplink service).
Therefore, this commit introduces the more specific term "NIC client" for an
interface that is based on a NIC session requested by the router. This doesn't
imply any semantic changes at the NIC router. However, the commit also brings a
broader update of the router's README and removes the term "downlink" that was
used only in documentation to refer to interfaces backed by a NIC session
provided by the router. The term was only associated with this meaning because
it is the natural counterpart to an uplink. This isn't appropriate anymore as
the terms for interface types have moved to a more technical level.
The commit adjusts all scenarios in the basic Genode repositories properly.
Fixes#4238
An interface that received a signal for a link-state change accessed its
domain reference without assuming that it could not be attached to a domain
at that moment. This caused the NIC router to crash with an uncaught exception
of type 'Net::Pointer<Net::Domain>::Invalid'. The commit adds a catch
directive for this exception resulting in the handler doing nothing if not
attached to any domain.
Fixes#4222
The test script failed during preparation of the on-target execution for
USB Armory with the following error:
! can't read "tz_vmm_block_irq": no such variable
Presumably, the script wasn't run anymore since the introduction of the
'tz_vmm_block_irq' variable for i.MX53 QSB. As we do not have infrastructure
for automated testing of the USB Armory and there seems to be not much
interest in using Genode on this platform, this commit simply removes the
support from the script.
Filtering boards in a run script by specs isn't the right way anymore (the
specs do not exist anymore). Nowadays, we have to use [have_board] instead.
Ref #4229
For unknown reasons, the former 'wget genode.org' call, that was meant to test
network in the Trustzone guest on imx53_qsb_tz, didn't succeed anymore although
the same call succeeded on my Sculpt VM Linux. However, 'ping 1.1.1.1' still
works, so, the script now uses this as test for networking instead.
Fixes#4229
So far, in order to create an ARP reply, the NIC router merely created a copy
of the corresponding ARP request and modified only those values that differ.
This approach has the disadvantage of re-using bad parameters from a broken
request. The specific use-case that made this visible was an early version of
the Pine board network driver that used to forward ARP requests with a greater
size than required. The ARP replies of the router re-used this size and
confused other network nodes with that. In general, the NIC router should
rely on the data of incoming packets the least possible. Therefore, with this
commit, the router creates a new ARP reply from scratch and uses only those
values required from the corresponding ARP request.
Fixes#4235
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
The re-newed approach currently supports ARM 64-bit only.
It depends on the Platform API of the ARM architecture.
It tries to meet the original semantic of the Linux kernel
functions as far as possible. To achieve this, device drivers
using this library should reference the original Linux kernel
headers at foremost. Only the headers in `src/include/lx_emul/shadow`
have to shadow clone the original ones.
Fix#4225
skb_push() already increases the skb->len by ETH_HLEN, hence adding
ETH_HLEN to the packet_size is redundant.
A too large packet size becomes a problem for large MTUs. With a maximum
MTU of 1500, adding ETH_HLEN twice will lead to a packet size of 1528.
Since this is larger than what we expect for good-old Ethernet (max. 1522),
some clients (e.g. the e1000 model in vbox5) may drop these packets.
Fixesgenodelabs/genode#4228
I discovered thinkbroadband.com requires the User-Agent header field and
rejects requests missing it with HTTP response code 403 "access to the
requested resource is forbidden". Now, fetchurl always adds the
User-Agent header fetchurl/LIBCURL_VERSION.
Also the error message now contains the HTTP response code.
The symlink implementation wrongly constructed a 'Sync' object within
the context of a monitor call. The 'Sync' constructor indirectly
depended on libc I/O for obtaining the current time, ultimately
resulting in a nested attempt of a monitor call. This could be
reproduced via the base.run script:
$ cd /home
$ ln -s a b
The 'ln' command resulted in the following log message:
[init -> /bin/bash -> 7] Error: deadlock ahead, mutex=0x10ff8c70, return ip=0x500583a7
The patch fixes the problem by splitting the single monitor call into
two monitor calls and moving the construction of the 'Sync' object
in-between both monitor calls, thereby executing the constructor at the
libc application level.
Fixes#4219
Building the elfloader in kernel-sel4.inc has a problem with Genodes CCACHE
make variable. When issuing ...
! ./tool/depot/create mstein/bin/*/base-sel4-* CCACHE=yes
..., building the elfloader used to consume all memory of the host system and
then run into a segmentation fault:
! make[6]: *** [elfloader/elfloader.o] Segmentation fault (core dumped)
This is because the other build system invokes the CCACHE variable as a command
in front of the compiler command. If CCACHE is set to 'yes', the 'yes' command
is called and produces an endless output into some output file. The problem
can be fixed by locally re-setting the CCACHE variable for the
'make ... elfloader' command to 'ccache' (Genode CCACHE==yes) or '' (Genode
CCACHE!=yes).
Fixes#4212
Adds try-catch-statement with diagnostic errors in Dhcp_server::free_ip in
order to guard against exceptions from the underlying bit allocator. These
exceptions should never happen given that the router is programmed correctly
and always feeds Dhcp_server::free_ip with sane arguments (which it should).
However, should this not be the case, we can assume that the failed IP freeing
indicates that the IP isn't allocated anyway and it's fine to continue using
the router. Furthermore, IP allocations are a mere client service and not
relevant for the integrity or safety of the router.
Ref #4200
When Interface::handle_config_3 (third step of applying a new configuration to
interfaces) tried to detach the interface from the current IP config because
the old and new IP config differed, it did so using the new domain. The former
steps of the reconfiguration already installed the new domain reference at the
interface. Therefore, also the DHCP server of the new domain was used. This,
however caused uncaught exceptions because detaching from an IP config
includes dissolving all DHCP allocations. This dissolving of DHCP allocations
now operated on a DHCP server (the one of the new domain) that wasn't related
to the allocations and, in the worst case, caused an uncaught exception
because the IPs were out of its range.
That said, this commit ensures that detaching an interface from an IP config
is always done on the domain from which the IP config originated. Normally,
this is the domain the interface is attached to. But in the case of
Interface::handle_config_3, it is another - the former domain the interface
was attached to.
The commit also adapts the nic_router_dhcp_* tests in a way that they
reconfigure the router in a way that would trigger the uncaught exception
without the fix.
Fixes#4200
Introduce two new cache maintainance functions:
* cache_clean_invalidate_data
* cache_invalidate_data
used to flush or invalidate data-cache lines.
Both functions are typically empty, accept for the ARM architecture.
The commit provides implementations for the base-hw kernel, and Fiasco.OC.
Fixes#4207
The implementation conflicted with the implicit declaration of bzero:
.../repos/dde_bsd/src/lib/audio/mem.cc: In function ‘void bzero(void*, size_t)’:
.../repos/dde_bsd/src/lib/audio/mem.cc:377:2: warning: ‘nonnull’ argument ‘b’ compared to NULL [-Wnonnull-compare]
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
The fs_query component used to exit with an uncaught exception if a queried
directory didn't exist. Now, fs_query will catch this event and simply skip the
affected query, thereby indicating to the user the inexistence of the
queried directory.
Ref #4032
