This change intends to run the whole network stack in softint context
(or normal LWP), not hardware interrupt context. Note that the work is
still incomplete by this change; to that end, we also have to softint-ify
if_link_state_change (and bpf) which can still run in hardware interrupt.
This change softint-ifies at ifp->if_input that is called from
each device driver (and ieee80211_input) to ensure Layer 2 runs
in softint (e.g., ether_input and bridge_input). To this end,
we provide a framework (called percpuq) that utlizes softint(9)
and percpu ifqueues. With this patch, rxintr of most drivers just
queues received packets and schedules a softint, and the softint
dequeues packets and does rest packet processing.
To minimize changes to each driver, percpuq is allocated in struct
ifnet for now and that is initialized by default (in if_attach).
We probably have to move percpuq to softc of each driver, but it's
future work. At this point, only wm(4) has percpuq in its softc
as a reference implementation.
Additional information including performance numbers can be found
in the thread at tech-kern@ and tech-net@:
http://mail-index.netbsd.org/tech-kern/2016/01/14/msg019997.html
Acknowledgment: riastradh@ greatly helped this work.
Thank you very much!
request again (possibly because of compiler optimisations), by using
copies and barrier.
From XSA155:
The compiler can emit optimizations in the PV backend drivers which
can lead to double fetch vulnerabilities. Specifically the shared
memory between the frontend and backend can be fetched twice (during
which time the frontend can alter the contents) possibly leading to
arbitrary code execution in backend.
This would cause the thread to be almost never scheduled when a userland
process could use all CPU.
Should fix the problem reported by Torbjörn Granlund on port-xen@
from hardware interrupt. Instead, release mutex while calling start routine.
The buffer peek/use/get sequence which can no longer be atomic. So consume
the buffer directly and on error privately save and retry the buffer later.
The dk_drain function is used to flush such a deferred buffer together with
the buffer queue.
Adjust drivers to use dk_drain.
Fix an error path where dk_done was called while the lock was already held.
- d_diskstart callback now processes a single buffer only.
The new wrapper function dk_start processes the queue,
performs other buffer handling and also provides locking
for the the data structures.
- add d_discard callback to handle device specific function inside
the new dk_discard helper function.
Replace splbio with mutex to protect queue and disk structure.
Refactor common code in dk_strategy and dk_discard into dk_translate.
Adjust and simplify ld(4), cgd(4) and xbd(4) drivers accordingly.
ld(4) now becomes MP_SAFE.
Bump kernel version.
Merge common disk driver functionality in ld.c with dksubr.c.
Adjust the two previous users of dk_intf (cgd and xbd) to
the changes.
bump kernel version to 7.99.14
I had removed <sys/rnd.h> from files that didn't mention anything of
the rnd(9) API. But they included other files which assumed
<sys/rnd.h> had already been included.
handlers stop being called when the device has been detached.
Should fix PR port-xen/47720 (which turns out to not be related to raidframe).
While there fix possible races in event_remove_handler() and pirq_establish().
altq Drop Type
disklabel Disk Type
file Descriptor Type
(not to mention constants that contain the string DTYPE).
Let's make them two, by changing the disklabel one to be DisK TYPE since the
other disklabel constants seem to do that. Not many userland programs use
these constants (and the ones that they do are mostly in ifdefs). They will
be fixed shortly.
especially for pmap_kernel(), as soon as it is up.
Instead move all pmap-related cpu_info initialisations, including
initializing ci_kpm_mtx, in cpu_attach_common() from cpu_init()
(ci_pmap and ci_tlbstate as already initialized in cpu_attach_common()).
https://mail-index.netbsd.org/tech-kern/2014/05/21/msg017098.html
remove dk_start() and dk_iodone() from dksubr.c and move the related code
to the underlying driver.
This increase complexity only marginally: the underlying drivers have
to do the while() loop themselves, but this can now be done properly with
bufq_peek()/bufq_get(), removing the buffer from the queue at the right time.
This handle both the recursion and reordering issues (the reordering
issue is described here:
https://mail-index.netbsd.org/tech-kern/2014/05/19/msg017089.html
the recursion isssue is PR #25240).
Difference with the patch posted to tech-kern@: KASSERT() that the
buffer we remove with bufq_get() is the same as the one we bufq_peek()'d
just before.
Hopefully this will allow more disk drivers to use dksubr.c
