Re: [PATCH v3] memblock: Add flags and nid info in memblock debugfs

[Yuwei Guan <ssawgyw@xxxxxxxxx>]

Anshuman Khandual <anshuman.khandual@xxxxxxx> 于2023年5月18日周四 18:12写道：
>
>
>
> On 5/18/23 14:44, Yuwei Guan wrote:
> > Currently, the memblock debugfs can display the count of memblock_type and
> > the base and end of the reg. However, when memblock_mark_*() or
> > memblock_set_node() is executed on some range, the information in the
> > existing debugfs cannot make it clear why the address is not consecutive.
> >
> > For example,
> > cat /sys/kernel/debug/memblock/memory
> >    0: 0x0000000080000000..0x00000000901fffff
> >    1: 0x0000000090200000..0x00000000905fffff
> >    2: 0x0000000090600000..0x0000000092ffffff
> >    3: 0x0000000093000000..0x00000000973fffff
> >    4: 0x0000000097400000..0x00000000b71fffff
> >    5: 0x00000000c0000000..0x00000000dfffffff
> >    6: 0x00000000e2500000..0x00000000f87fffff
> >    7: 0x00000000f8800000..0x00000000fa7fffff
> >    8: 0x00000000fa800000..0x00000000fd3effff
> >    9: 0x00000000fd3f0000..0x00000000fd3fefff
> >   10: 0x00000000fd3ff000..0x00000000fd7fffff
> >   11: 0x00000000fd800000..0x00000000fd901fff
> >   12: 0x00000000fd902000..0x00000000fd909fff
> >   13: 0x00000000fd90a000..0x00000000fd90bfff
> >   14: 0x00000000fd90c000..0x00000000ffffffff
> >   15: 0x0000000880000000..0x0000000affffffff
> >
> > So we can add flags and nid to this debugfs.
> >
> > For example,
> > cat /sys/kernel/debug/memblock/memory
> >    0: 0x0000000080000000..0x00000000901fffff     NONE    0
> >    1: 0x0000000090200000..0x00000000905fffff    NOMAP    0
> >    2: 0x0000000090600000..0x0000000092ffffff     NONE    0
> >    3: 0x0000000093000000..0x00000000973fffff    NOMAP    0
> >    4: 0x0000000097400000..0x00000000b71fffff     NONE    0
> >    5: 0x00000000c0000000..0x00000000dfffffff     NONE    0
> >    6: 0x00000000e2500000..0x00000000f87fffff     NONE    0
> >    7: 0x00000000f8800000..0x00000000fa7fffff    NOMAP    0
> >    8: 0x00000000fa800000..0x00000000fd3effff     NONE    0
> >    9: 0x00000000fd3f0000..0x00000000fd3fefff    NOMAP    0
> >   10: 0x00000000fd3ff000..0x00000000fd7fffff     NONE    0
> >   11: 0x00000000fd800000..0x00000000fd901fff    NOMAP    0
> >   12: 0x00000000fd902000..0x00000000fd909fff     NONE    0
> >   13: 0x00000000fd90a000..0x00000000fd90bfff    NOMAP    0
> >   14: 0x00000000fd90c000..0x00000000ffffffff     NONE    0
> >   15: 0x0000000880000000..0x0000000affffffff     NONE    0
>
> Although, Mike had suggested to keep these flags print last, above
> format looks good as well.
>
> >
> > Signed-off-by: Yuwei Guan <ssawgyw@xxxxxxxxx>
> > ---
> > v3:
> > - show string value for each memblock flag
> > ---
> >  mm/memblock.c | 12 +++++++++++-
> >  1 file changed, 11 insertions(+), 1 deletion(-)
> >
> > diff --git a/mm/memblock.c b/mm/memblock.c
> > index 511d4783dcf1..5fba53f98b2d 100644
> > --- a/mm/memblock.c
> > +++ b/mm/memblock.c
> > @@ -2143,13 +2143,23 @@ static int memblock_debug_show(struct seq_file *m, void *private)
> >       struct memblock_region *reg;
> >       int i;
> >       phys_addr_t end;
> > +     static const char flagname[BITS_PER_LONG][8] = {
> > +             [0 ... (BITS_PER_LONG-1)] = "?",
>
> Minor nit -
>
> Although checkpatch does not complain, should there be spaces between
> the operator and operands e.g (BITS_PER_LONG - 1).
>
> > +
> > +             [ilog2(MEMBLOCK_HOTPLUG)] = "HOTPLUG",
> > +             [ilog2(MEMBLOCK_MIRROR)] = "MIRROR",
> > +             [ilog2(MEMBLOCK_NOMAP)] = "NOMAP",
> > +             [ilog2(MEMBLOCK_DRIVER_MANAGED)] = "DRV_MNG",
> > +     };
>
> Also, BITS_PER_LONG sized array is really required here ? as there are
> just four available memblock flags.
>
Hi Anshuman,
The main reason to use BITS_PER_LONG is to reserve.
If the flagname buffer is (ilog2(MEMBLOCK_DRIVER_MANAGED) + 1),
memblock_flags adds a new attribute and does not add its name in debugfs,
it will cause an overflow.
But BITS_PER_LONG is too wasteful, so I implement a new solution.
Please help to check it.

        struct memblock_type *type = m->private;
        struct memblock_region *reg;
-       int i;
+       int i, j;
        phys_addr_t end;
+       static const char *flagname[] = {
+               [ilog2(MEMBLOCK_HOTPLUG)] = "HOTPLUG",
+               [ilog2(MEMBLOCK_MIRROR)] = "MIRROR",
+               [ilog2(MEMBLOCK_NOMAP)] = "NOMAP",
+               [ilog2(MEMBLOCK_DRIVER_MANAGED)] = "DRV_MNG",
+       };

        for (i = 0; i < type->cnt; i++) {
                reg = &type->regions[i];
                end = reg->base + reg->size - 1;

                seq_printf(m, "%4d: ", i);
-               seq_printf(m, "%pa..%pa\n", &reg->base, &end);
+               seq_printf(m, "%pa..%pa ", &reg->base, &end);
+               seq_printf(m, "%4d ", memblock_get_region_node(reg));
+               if (reg->flags) {
+                       for (j = 0; j < ARRAY_SIZE(flagname); j++) {
+                               if (reg->flags & (1U << j)) {
+                                       seq_printf(m, "%s\n", flagname[j]);
+                                       break;
+                               }
+                       }
+                       if (j == ARRAY_SIZE(flagname))
+                               seq_printf(m, "%s\n", "UNKNOWN");
+               } else {
+                       seq_printf(m, "%s\n", "NONE");
+               }
        }
> >
> >       for (i = 0; i < type->cnt; i++) {
> >               reg = &type->regions[i];
> >               end = reg->base + reg->size - 1;
> >
> >               seq_printf(m, "%4d: ", i);
> > -             seq_printf(m, "%pa..%pa\n", &reg->base, &end);
> > +             seq_printf(m, "%pa..%pa ", &reg->base, &end);
> > +             seq_printf(m, "%8s ", reg->flags ? flagname[ilog2(reg->flags)] : "NONE");
> > +             seq_printf(m, "%4d\n", memblock_get_region_node(reg));
> >       }
> >       return 0;
> >  }