The commit adds support to throttle the rate of the RX IRQs to a specified
value. The effect is, that no RX IRQs below the time threshold will fire and
therefore the CPU load gets reduced on the host. Trade-off gaming between
cpu load, throughput, overload.
Modular Sculpt 23.10 on S938 as testcase. In brackets the CPU affinity is
denoted.
ipxe (0,0) -> nic_router (1,0) -> Debian VM vbox6 (3,0) and (3,1)
VM: iperf -C X.X.X.X -t 60 -R
iperf server X.X.X.X is outside Sculpt and sends data due to '-R' to VM
Non representative measure points:
cpu load - ipxe - nic_router - iperf throughput
--------------------------------------------------
w/o patch - ~80% - ~50% - ~706 MBit/s - 0 -> throttling off by default on S938
patch 651 - ~20% - ~35% - ~763 MBit/s - 651 -> 0.166ms throttle RX IRQ
patch 5580 - ~15% - ~25% - ~650 MBit/s - 5580 -> 1.4ms throttle RX IRQ
Issue #5149
A bunch of transmit requests received by the Uplink server (nic_router)
are currently added one by one to the ring buffer and every time the hardware
is notified to process each single request.
Instead, add as many as possible transmit requests in the ring buffer of
the hardware and when done trigger the hardware to process the ring.
Additionally, don't receive an "processed" TX IRQ for each element in the
ring, which causes high CPU load.
With this commit the TX IRQs in the ipxe driver for a
iperf -c X.X.X.X -t 60
from within a VM to the outside iperf server is reduced from about
~2'600'000 IRQs to about ~200'000. The overall CPU load for the driver
(when executed alone on CPU 0) is reduced from ~85 percent load to ~45 percent
load.
Issue #5149
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
The `with_sub_node` method is renamed to `with_optional_sub_node` to
better reflect that the non-existence of a sub node with the desired type is
ignored.
At the same time, the new `with_sub_node` now takes a second functor that is
called when no sub node of the desired type exists.
genodelabs/genode#4600
Under high TX load, 'irq_handler()' is rarely executed but packets
are still received since 'netdev_poll()' is called for every TX packet.
'netdev_poll()' not only handles completed transmissions but also
puts the IO buffers of received packets into the RX queue and refills the RX
ring by allocating more IO buffers. However, the IO buffers can only be freed
after they have been dequeued and passed to the 'rx_callback()'. Since
this was only done by 'irq_handler()', we exhaust the slab allocator
under high TX load.
A fix for this is to process the RX queue not only in the IRQ handler
but also when transmitting packets.
genodelabs/genode#4555
Also, the repository URL was adapted to the permanent redirect to github
to prevent the following warning.
dde_ipxe download http://git.ipxe.org/ipxe.git
dde_ipxe git Cloning into 'src/lib/dde_ipxe'...
dde_ipxe git warning: redirecting to https://github.com/ipxe/ipxe/
dde_ipxe update src/lib/dde_ipxe
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
