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!
1. Assume that errno is non-zero when entering read_char()
and that read(2) returns 0 (indicating end of file).
Then, the code will clear errno before returning.
(Obviously, the statement "errno = 0" is almost always
a bug unless there is save_errno = errno right before it
and the previous value is properly restored later,
in all reachable code paths.)
2. When encountering an invalid byte sequence, the code discards
all following bytes until MB_LEN_MAX overflows; consider, for
example, 0xc2 immediately followed by a few valid ASCII bytes.
Three of those ASCII bytes will be discarded.
3. On a POSIX system, EILSEQ will always be set after reading a
valid (yes, valid, not invalid!) UTF-8 character. The reason
is that mbtowc(3) will first be called with a length limit
(third argument) of 1, which will fail, return -1, and - on
a POSIX system - set errno to EILSEQ.
This third bug is mitigated a bit because i couldn't find any
system that actually conforms to POSIX in this respect: None
of OpenBSD, NetBSD, FreeBSD, Solaris 11, and glibc set errno
when an incomplete character is passed to mbtowc(3), even though
that is required by POSIX.
Anyway, that mbtowc(3) bug will be fixed at least in OpenBSD
after release unlock, so it would be good to fix this bug in
libedit before fixing the bug in mbtowc(3).
How can these three bugs be fixed?
1. As far as i understand it, the intention of the bogus errno = 0
is to undo the effects of failing system calls in el_wset(),
sig_set(), and read__fixio() if the subsequent read(2) indicates
end of file. So, restoring errno has to be moved right after
read__fixio(). Of course, neither 0 nor e is the right value
to restore: 0 is wrong if errno happened to be set on entry, e
would be wrong because if one read(2) fails but a second attempt
succeeds after read__fixio(), errno should not be touched. So,
the errno to be restored in this case has to be saved before
calling read(2) for the first time.
2. Solving the second issue requires distinguishing invalid and
incomplete characters, but that is impossible with the function
mbtowc(3) because it returns -1 in both cases and sets errno
to EILSEQ in both cases (once properly implemented).
It is vital that each input character is processed right away.
It is not acceptable to wait for the next input character before
processing the previous one because this is an interactive
library, not a batch system. Consequently, the only situation
where it is acceptable to wait for the next byte without first
processing the previous one(s) is when the previous one(s) form
an incomplete sequence that can be continued to form a valid
character.
Consequently, short of reimplementing a full UTF-8 state machine
by hand, the only correct way forward is to use mbrtowc(3).
Even then, care is needed to always have the state object
properly initialized before using it, and to not discard a valid
ASCII or UTF-8 lead byte if it happens to follow an invalid
sequence.
3. Fortunately, solution 2. also solves issue 3. as a side effect,
by no longer using mbtowc(3) in the first place.
and return a single error value EEXIST. When making a recursive
call (to load required modules), treat a pre-existing module as
success.
Without this change, when a module was loaded by specific request
(as opposed to being loaded as a requirement of some other module),
we would always load the module from the file-system, and then
after making various sanity/compatability checks we would destroy
the new copy if there was a pre-existing copy.
Fixes PR kern/40764
XXX Note that if the module exists, we bypass all of the various
XXX "compatability" checks, including whether or not the existing
XXX module is of any particular class! (In the previous code, we
XXX checked to see if the newly-loaded copy had the correct class,
XXX but not the pre-existing copy, which could have been loaded
XXX from a different path/filename.)
This file is used in tools, where _BSD_SIZE_T_ is unreliable. In the
kernel, <sys/types.h> will define size_t anyway. Outside the kernel,
<stddef.h> is the standard way to get size_t.
usbd_clear_port_feature(dev, port, UHF_C_PORT_RESET)
The Gadget USB HUB in the IBM x3250 sborrill sent me returns the wrong
status and this would appear to leave the attached cdce(4) in the default
state. The next device probed fails as a result.
skrll