Re: Recent bugfixes.

[Anders Magnusson <ragge%tethuvudet.se@localhost>]

Den 2023-03-26 kl. 14:54, skrev Rhialto:
> On Sun 26 Mar 2023 at 14:25:21 +0200, Anders Magnusson wrote:
> > Hi,
> >
> > I have just checked in fixes for both the "S0 space > 1G" bug and the
> > CASMAGIC #define.
> > Have fun!
> It still may make the page tables bigger than needed due to double
> adding of several memory sizes. For example, in pmap.c, calc_kvmsize():
>
>          /*
>           * Compute the number of pages kmem_arena will have.
>           */
>          kmeminit_nkmempages();     // this calculates nkmempages
> ...
>          /* kernel malloc arena */
>          kvmsize += nkmempages * PAGE_SIZE;
> ... and then another time ...
>          /* kernel malloc arena */
>          kvmsize += avail_end;
>
> What should be the size of the kernel malloc arena? Whatever is
> calculated in kmeminit_nkmempages(), or just simply the physical memory size?
> Likely not both added together.
> So on smaller memory, it seems that some of it is wasted unnecessarily.
Please look at the patch:

        kvmsize = calc_kvmsize(usrptsize);
+       /*
+        * Ensure that not more than 1G is allocated, since that is
+        * max size of S0 space.
+        * Also note that for full S0 space the SLR should be 0x200000,
+        * since the comparison in the vax microcode is >= SLR.
+        */
+#define        S0SPACE (1*1024*1024*1024)
+       if (kvmsize > S0SPACE)
+               kvmsize = S0SPACE;
        sysptsize = kvmsize >> VAX_PGSHIFT;
        /*
         * Virtual_* and avail_* is used for mapping of system page table.

After everything is calculated I check if kvmsize gets larger than 1G, 
and in that case truncates it to 1G.
I haven't changed the calculations.

-- Ragge