(Abridged) We use new multi-wavelength radio observations, made with the VLA and Effelsberg telescopes, to study the magnetic field of the nearby galaxy M51 on scales from $200pc$ to several $kpc$. Interferometric and single dish data are combined to
obtain new maps at wwav{3}{6} in total and polarized emission, and earlier wav{20} data are re-reduced. We compare the spatial distribution of the radio emission with observations of the neutral gas, derive radio spectral index and Faraday depolarization maps, and model the large-scale variation in Faraday rotation in order to deduce the structure of the regular magnetic field. We find that the wav{20} emission from the disc is severely depolarized and that a dominating fraction of the observed polarized emission at wav{6} must be due to anisotropic small-scale magnetic fields. Taking this into account, we derive two components for the regular magnetic field in this galaxy: the disc is dominated by a combination of azimuthal modes, $m=0+2$, but in the halo only an $m=1$ mode is required to fit the observations. We disuss how the observed arm-interarm contrast in radio intensities can be reconciled with evidence for strong gas compression in the spiral shocks. The average arm--interam contrast, representative of the radii $r>2kpc$ where the spiral arms are broader, is not compatible with straightforward compression: lower arm--interarm contrasts than expected may be due to resolution effects and emph{decompression} of the magnetic field as it leaves the arms. We suggest a simple method to estimate the turbulent scale in the magneto-ionic medium from the dependence of the standard deviation of the observed Faraday rotation measure on resolution. We thus obtain an estimate of $50pc$ for the size of the turbulent eddies.
Observations show that magnetic fields in the interstellar medium (ISM) often do not respond to increases in gas density as would be naively expected for a frozen-in field. This may suggest that the magnetic field in the diffuse gas becomes detached
from dense clouds as they form. We have investigated this possibility using theoretical estimates, a simple magneto-hydrodynamic model of a flow without mass conservation and numerical simulations of a thermally unstable flow. Our results show that significant magnetic flux can be shed from dense clouds as they form in the diffuse ISM, leaving behind a magnetically dominated diffuse gas.
