With the swift advancement of deep learning technologies, single hyperspectral image (HSI) super-resolution (SR) algorithms have observed considerable progress. However, methods leveraging 2D convolutions often struggle to effectively extract spectral information, while approaches using 2D/3D hybrid convolutions may introduce feature redundancy and model complexity. This can result in an overemphasis on local regions and, subsequently, a loss of high-frequency (HF) information. To address these challenges, a novel methodology based on dual-domain gating attention (DGA), termed DGANet, has been introduced. This innovative approach employs non-local gating convolution (NGC) to filter spatial HF regions within the image while maintaining attention on global context. Guided by spatial-spectral correlation, DGANet reconstructs HSI without compromising detail, while deep feedforward convolution (DFC) preserves HF information even as network architecture deepens. Comprehensive quantitative comparisons and ablation studies were conducted on the proposed network. The experimental findings indicate that DGANet attains superior performance in terms of retaining rich HF information, establishing it as an effective solution in the field.