Oxygen abundances in a sample of ultra-metal-poor subdwarfs have been deriv
ed from measurements of the oxygen triplet at 7771-5 Angstrom and OH lines
in the near-UV performed in high-resolution and high signal-to-noise ratio
spectra obtained with WHT/UES, Keck I/HIRES, and VLT/UVES. Our Fe abundance
s were derived in LTE and then corrected for non-LTE (NLTE) effects. The ne
w oxygen abundances confirm previous findings for a progressive linear rise
in the oxygen-to-iron ratio with a slope -0.33 +/- 0.02 from solar metalli
city to [Fe/H] similar to -3. A slightly higher slope would be obtained if
the Fe NLTE corrections were not considered. Below [Fe/H] = -2.5 our stars
show [O/Fe] ratios as high as similar to1.17 (G64-12), which can be interpr
eted as evidence for oxygen overproduction in the very early epoch of the f
ormation of the halo, possibly associated with supernova events with very m
assive progenitor stars. We show that the arguments against this linear tre
nd given by Fulbright & Kraft in 1999, based on the LTE Fe analysis of two
metal-poor stars, cannot be sustained when an NLTE analysis is performed. W
e discuss how the Fulbright & Kraft LTE ionization balance of Fe lines unde
restimates the gravity of the very metal-poor star BD +23 degrees 3130 ([Fe
/H] = -2.43) and how this leads to an underestimation of the oxygen abundan
ce derived from the forbidden line. Gravities from Hipparcos appear to be i
n good agreement with those determined in NLTE, giving higher values than p
reviously assumed, which reduces the discrepancies between the oxygen abund
ances determined from OH, triplet, and forbidden lines. Using one-dimension
al models, our analysis of three oxygen indicators available for BD +23 deg
rees 3130 gives an average [O/Fe] ratio of 0.78(-0.18)(+0.15). The high oxy
gen abundances at very low metallicities do not pose a problem to theoretic
al modeling since there is a range of parameters in the calculations of nuc
leosynthesis yields from massive stars at low metallicities that can accomm
odate our results.