Re: Clarify drand48() return values
Theo de Raadt wrote:
> Alexander Nasonov wrote:
>
> > Ingo Schwarze wrote:
> > > Looking at our code in lib/libc/stdlib/drand48.c, i conclude that
> > > drand48(3) does return 0.0 with a probability of 2^-48.
> >
> > I looked at the code too and I have some comments.
> >
> > > More generally, the function returns a uniform distribution of
> > > numbers from the set {2^-48 * n | n integer and 0 <= n < 2^48}.
> >
> > You don't need three ldexp calls to compose 2^-48 * n:
> >
> > uint64_t n = (uint64_t)xseed[2] << 32 | xseed[1] << 16 | xseed[0];
> > return ldexp((double)n, -48);
>
> xseed?
s/xseed/rseed/g
--
Alex
Re: Clarify drand48() return values
Ingo Schwarze wrote:
> Looking at our code in lib/libc/stdlib/drand48.c, i conclude that
> drand48(3) does return 0.0 with a probability of 2^-48.
I looked at the code too and I have some comments.
> More generally, the function returns a uniform distribution of
> numbers from the set {2^-48 * n | n integer and 0 <= n < 2^48}.
You don't need three ldexp calls to compose 2^-48 * n:
uint64_t n = (uint64_t)xseed[2] << 32 | xseed[1] << 16 | xseed[0];
return ldexp((double)n, -48);
> Talking about loading bits into the mantissa and adjusting the
> exponent feels mildly confusing, given that the distribution is
> simply uniform over a fixed finite set. I'm not sending a patch
> because i never looked at how floating point representation works
> internally, so i would likely only make matters worse.
Not sure if you wanted to patch the man page or the code.
If the latter, take a binary representation of 1.0 and replace 52
least significant bits with (pseudo)random bits then subtract 1.0.
double d = 1.0;
uint64_t u = 0;
memcpy(, , sizeof(d));
u |= n << 4; /* scale 48 bits to 52 bits, n is from the previous
snippet */
memcpy(, , sizeof(u));
return d - 1.0;
It will only work with IEEE 754 compliant doubles, though.
--
Alex
Re: Clarify drand48() return values
j...@bitminer.ca wrote: > > Clarify that drand48 returns values not including 1.0. It's not clear from the documentation whether drand48 can generate a denormal number. If it can't, you can exclude 0.0 because it's a denormal ;-) -- Alex
Re: syscall call-from verification
Theo de Raadt wrote: > The following change only permits system calls from address-ranges > in the process which system calls are expected from. Just curious if some approximation of pledge can be reimplemented in userspace with more granular libc.so's text segments? -- Alex
Re: [EXTERNAL] Re: Removing PF
Eichert, Diana wrote:
> I wrote a vax BPF jit as a simple exercize some time ago, so all
> you really need now is to implement vax-to-${ARCH} jit on an MD
> basis. This should be very easy to do as long as BPF does not get
> extended to use floating-point values.
I'm afraid you have to rewrite it to risv-to-${ARCH} and vectorise
along the way.
--
Alex
