It seems that the Meltdown leak only happens if (kernel) data is in the L1 cache.

This seems like an onerous restriction on applicability of the attack : it is like one would have to make a system call to the kernel which would read the secret data just prior to the attack.

This is a far cry from the first comments on the issue, which implied that any memory accessible by the kernel (thus, in Linux and MacOS X at least, all of physical RAM) was readable.

Or is there something not yet public ?