svn commit: r307326 - head/sys/boot/efi/loader

[Doug Ambrisko]

On Fri, Oct 14, 2016 at 12:25:54PM -0600, Warner Losh wrote:
|    On Oct 14, 2016 11:55 AM, "Doug Ambrisko" <[1]ambrisko at ambrisko.com>
|    wrote:
|    >
|    > On Fri, Oct 14, 2016 at 10:33:15AM -0700, Ravi Pokala wrote:
|    > | -----Original Message-----
|    > | > From: <[2]owner-src-committers at freebsd.org> on behalf of Doug
|    Ambrisko <[3]ambrisko at ambrisko.com>
|    > | > Date: 2016-10-14, Friday at 10:27
|    > | > To: Warner Losh <[4]imp at bsdimp.com>
|    > | > Cc: Doug Ambrisko <[5]ambrisko at freebsd.org>, src-committers
|    <[6]src-committers at freebsd.org>, "[7]svn-src-all at freebsd.org"
|    <[8]svn-src-all at freebsd.org>, "[9]svn-src-head at freebsd.org"
|    <[10]svn-src-head at freebsd.org>
|    > | > Subject: Re: svn commit: r307326 - head/sys/boot/efi/loader
|    > | >
|    > | > On Fri, Oct 14, 2016 at 11:16:02AM -0600, Warner Losh wrote:
|    > | > | Love the functionality, but don't like using the 'hint'
|    namespace for
|    > | > | this. Can we change it now before too many things depend on it?
|    We had
|    > | > | similar issues in ACPI and moved it to the 'acpi' space. Can we
|    move
|    > | > | this to the 'smbios' space please?
|    > | > |
|    > | > | The reason is that 'hint' is special and sometimes filtered
|    out, so it
|    > | > | is a poor choice to export data from the boot loader to the
|    kernel.
|    > | >
|    > | > The reason I picked hint was it could be put /boot/device.hints
|    > | > to make it work as well and that it was a hint.  Other standards
|    in the
|    > | > future might use other methods.  Looking back over the email I
|    had
|    > | > with John he had suggested hint.smbios.0.anchor to make this look
|    > | > different.  This code had been hanging around for so long I
|    forgot
|    > | > about that and we were using hint.smbios.0.mem in our shipping
|    code base.
|    > | >
|    > | > However, I hope that nothing would use this except for smbios(4)
|    and
|    > | > for people to make smbios(4) useful for this info.
|    > |
|    > | Doug's looking at me when he says that. :-)
|    > |
|    > | We talked about this last night at BAFUG; right now, even if
|    smbios(4)
|    > | is able to find the SMBIOS info -- it currently only looks at the
|    > | aforementioned 0xf0000 - 0xfffff range, so it can't find it on UEFI
|    --
|    > | smbios(4) doesn't actually provide any interface for that
|    information.
|    > | Doug and I have talked about making smbios(4) useful, by parsing
|    the
|    > | data and providing KPIs and APIs, for years now; I think I'll
|    *finally*
|    > | have the time and motivation to do so "soon".
|    >
|    > I've actually talked to a few people.  However, your the first to
|    > step up.  This needs to be designed and will take some time and
|    > review.  I would hope that except for smbios(4), nothing else would
|    > use this kenv but there is nothing to prevent that :-(  I could name
|    > it super_secret_dont_use.
|    >
|    > BTW, to get you started this patch prevents smbios(4) from blowing
|    chunks
|    > when it gets a anchor in high memory and works on legacy machines.
|    >
|    > --- /sys/x86/bios/smbios.c      2013-10-01 14:28:25.000000000 -0700
|    > +++ ./smbios.c  2016-04-11 11:58:03.234969000 -0700
|    > @@ -38,6 +38,7 @@ __FBSDID("$FreeBSD: release/9.2.0/sys/x8
|    >  #include <machine/bus.h>
|    >  #include <machine/resource.h>
|    >  #include <sys/rman.h>
|    > +#include <sys/sysctl.h>
|    >
|    >  #include <vm/vm.h>
|    >  #include <vm/vm_param.h>
|    > @@ -59,7 +60,7 @@ struct smbios_softc {
|    >  };
|    >
|    >  #define        RES2EPS(res)    ((struct smbios_eps
|    *)rman_get_virtual(res))
|    > -#define        ADDR2EPS(addr)  ((struct smbios_eps
|    *)BIOS_PADDRTOVADDR(addr))
|    > +#define        ADDR2EPS(addr)  ((struct smbios_eps
|    *)PHYS_TO_DMAP(addr))
|    >
|    >  static devclass_t      smbios_devclass;
|    >
|    > @@ -71,19 +72,26 @@ static int  smbios_modevent (module_t, in
|    >
|    >  static int     smbios_cksum    (struct smbios_eps *);
|    >
|    > +static unsigned long addr;
|    > +static SYSCTL_NODE(_hw, OID_AUTO, smbios, CTLFLAG_RD, 0,
|    > +    "SMBIOS driver parameters");
|    > +SYSCTL_LONG(_hw_smbios, OID_AUTO, mem, CTLFLAG_RW,
|    > +        &addr, 0, "");
|    > +
|    >  static void
|    >  smbios_identify (driver_t *driver, device_t parent)
|    >  {
|    >         device_t child;
|    > -       u_int32_t addr;
|    >         int length;
|    >         int rid;
|    >
|    >         if (!device_is_alive(parent))
|    >                 return;
|    >
|    > -       addr = bios_sigsearch(SMBIOS_START, SMBIOS_SIG, SMBIOS_LEN,
|    > -                             SMBIOS_STEP, SMBIOS_OFF);
|    > +       if (resource_long_value("smbios", 0, "mem", &addr) != 0 ||
|    > +           addr == 0)
|    > +               addr = bios_sigsearch(SMBIOS_START, SMBIOS_SIG,
|    SMBIOS_LEN,
|    > +                                     SMBIOS_STEP, SMBIOS_OFF);
|    >         if (addr != 0) {
|    >                 rid = 0;
|    >                 length = ADDR2EPS(addr)->length;
|    >
|    > Note I don't plan to commit this since it doesn't really do much and
|    we
|    > need a lot more.
| 
|    I was planning on exporting all smbios stuff via sysctl.
| 
|    You can still put non hints.* in device.hints too.

I wondered about that.  It would look a bit strange.
 
|    I can rework things a bit to be more complete is you like...

That would be great.  A few people have expressed need of this from a
kernel level.  Adrian mentioned it for NUMA, ipmi(4) needs it, it
could be used on Dell machines to fix HW layout issues for NICs.
A couple companies could use it for internal needs.  Anything to
move this forward would help.  I don't mind running dmidecode etc.
but not needing that would be nice.

Thanks,

Doug A.