This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Question regarding minimal supported Linux kernel headers (now it is 3.2.0)


On Tue, 12 Nov 2019, Lukasz Majewski wrote:

> The pselect6 conversion has been postponed on purpose as microblaze
> port gained support for pselect6 from Linux kernel 3.15. Between 3.2
> (minimal supported glibc Linux headers) and 3.15 this syscall is
> emulated and hence the generic code is convoluted.
> 
> For the above reason the conversion of pselect6 has been postponed until
> the minimal glibc's Linux headers are bumped to the version after 3.15.
> 
> Dear community - when the minimal glibc version may be increased?
> 
> Are there any guidelines when the minimal version shall be increased
> (like once per X glibc releases) ?

Typically we increase the minimum kernel version used at runtime (which is 
generally the same as the minimum kernel headers version), to the oldest 
currently supported LTS Linux kernel version as listed at 
<https://www.kernel.org/category/releases.html>, when this allows some 
significant cleanup in glibc.

I think the next increase that is likely to allow significant cleanup is 
moving to 4.4 as minimum kernel version (once 3.16 is EOL); 3.16 doesn't 
allow much cleanup, but 4.4 allows removing most of the socketcall support 
from glibc as all architectures have most of the separate socket syscalls 
in 4.4 (a few were only added later for 32-bit SPARC in the compat syscall 
table on 64-bit kernels).

I was *also* hoping that, before the next increase to the minimum kernel 
version, we would have the changes Carlos proposed to remove the error on 
startup for a too-old kernel, so that new glibc running under an old 
kernel would only fail on actually trying to use one of the newer 
syscalls, which might be rare on x86_64, and might not occur in the case 
of container hosting systems that have backported newer syscalls to their 
older kernel.

For pselect6, I suggest a preparatory change increasing the minimum 
version for microblaze only and removing the racy emulation code from 
glibc citing bug 9813, which could then be marked as FIXED with 
appropriate milestone set once the emulation code has been completely 
removed.  (misc/pselect.c would change to an ENOSYS stub complete with 
stub_warning call.)

-- 
Joseph S. Myers
joseph@codesourcery.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]