9front - general discussion about 9front
 help / color / mirror / Atom feed
From: Lucas Francesco <lucas.francesco93@gmail.com>
To: 9front@9front.org
Subject: Re: [9front] [PATCH] boot/efi: delay GetMemoryMap until right before ExitBootServices
Date: Sat, 19 Nov 2022 13:56:34 -0300	[thread overview]
Message-ID: <CAF=5iUX_Eav33PH8wnVOqxk+VgDUbUXzcFBz3b2j=bqN2C2F6Q@mail.gmail.com> (raw)
In-Reply-To: <CAF=5iUXyn=7uXKxKPkznTN+9KVGYRneKaonNzYncQVDKXWTK7Q@mail.gmail.com>

Yup, works nicely here, tysm.

On Sat, 19 Nov 2022 at 13:14, Lucas Francesco
<lucas.francesco93@gmail.com> wrote:
>
> Oh, this seems to be the exact same problem I have with my devices
> here. Was having empty nvme queues after identify (and reading the
> spec everything else was mostly fine in our driver besides some quirks
> to workaround shitty controllers) so i banged around to see if
> firmware was messing it up before boot and realized the nvme feature
> was pointing to that, couldn't figure out it was actually something
> that would need to change in the bootloader though! (i banged
> pcireset/nssr/initseg() a lot)
>
> Thanks so much for the find, will test this asap.
>
> On Wed, 16 Nov 2022 at 17:13, Michael Forney <mforney@mforney.org> wrote:
> >
> >
> > The UEFI specification suggests calling GetMemoryMap immediately
> > before ExitBootServices to ensure that the loader has the current
> > memory map. The firmware is able to enforce this since we have to
> > pass a "MapKey" from the last GetMemoryMap to ExitBootServices.
> >
> > The T14 AMD gen 1 firmware is quite strict about this, and even a call
> > to OutputString will invalidate the MapKey. This causes
> > ExitBootServices to fail, and we enter 9front with boot services still
> > running. This causes all sorts of problems including leaving the
> > IOMMU enabled, breaking 9front's PCI drivers.
> >
> > To fix this, move memconf() to unload(), right before
> > ExitBootServices. To retain the ability to override the memory map,
> > check if *e820 is already set and if so, ignore the output of
> > GetMemoryMap except for the MapKey.
> > ---
> > Still some room for improvement around the *conf helper functions, but
> > I think this is fine for now.
> >
> > I also wonder what we should be doing if GetMemoryMap fails. If we
> > don't have any potentially valid MapKey, how could ExitBootServices
> > succeed? Perhaps we could make unload return an error string on
> > failure, and pass it back from bootkern so that we can inform the user
> > that something is wrong.
> >
> > diff a854bb07cd792a52c5e75aabca76c22b7dd18fc6 eb07ce3baa9705a30ecf3cd6b31ff851e24cb3d1
> > --- a/sys/src/boot/efi/efi.c
> > +++ b/sys/src/boot/efi/efi.c
> > @@ -36,12 +36,6 @@
> >         eficall(ST->BootServices->Stall, (UINTN)us);
> >  }
> >
> > -void
> > -unload(void)
> > -{
> > -       eficall(ST->BootServices->ExitBootServices, IH, MK);
> > -}
> > -
> >  static void
> >  memconf(char **cfg)
> >  {
> > @@ -72,6 +66,8 @@
> >         entvers = 1;
> >         if(eficall(ST->BootServices->GetMemoryMap, &mapsize, mapbuf, &MK, &entsize, &entvers))
> >                 return;
> > +       if(cfg == nil)
> > +               return;
> >
> >         s = *cfg;
> >         for(p = mapbuf; mapsize >= entsize; p += entsize, mapsize -= entsize){
> > @@ -93,7 +89,6 @@
> >         *s = '\0';
> >         if(s > *cfg){
> >                 s[-1] = '\n';
> > -               print(*cfg);
> >                 *cfg = s;
> >         }
> >  }
> > @@ -276,9 +271,15 @@
> >  void
> >  eficonfig(char **cfg)
> >  {
> > -       memconf(cfg);
> >         acpiconf(cfg);
> >         screenconf(cfg);
> > +}
> > +
> > +void
> > +unload(char **cfg)
> > +{
> > +       memconf(cfg);
> > +       eficall(ST->BootServices->ExitBootServices, IH, MK);
> >  }
> >
> >  EFI_STATUS
> > --- a/sys/src/boot/efi/fns.h
> > +++ b/sys/src/boot/efi/fns.h
> > @@ -16,7 +16,7 @@
> >  void (*close)(void *f);
> >
> >  int readn(void *f, void *data, int len);
> > -void unload(void);
> > +void unload(char **cfg);
> >
> >  int getc(void);
> >  void putc(int c);
> > --- a/sys/src/boot/efi/sub.c
> > +++ b/sys/src/boot/efi/sub.c
> > @@ -156,6 +156,23 @@
> >
> >  char *confend;
> >
> > +static int
> > +hasconf(char *s)
> > +{
> > +       char *p;
> > +       int n;
> > +
> > +       n = strlen(s);
> > +       for(p = BOOTARGS; n <= confend - p; p++){
> > +               if(memcmp(p, s, n) == 0)
> > +                       return 1;
> > +               p = strchr(p, '\n');
> > +               if(p == nil)
> > +                       break;
> > +       }
> > +       return 0;
> > +}
> > +
> >  static char*
> >  getconf(char *s, char *buf)
> >  {
> > @@ -364,7 +381,7 @@
> >
> >         close(f);
> >         print("boot\n");
> > -       unload();
> > +       unload(hasconf("*e820=") ? nil : &confend);
> >
> >         jump(e);
> >

  reply	other threads:[~2022-11-19 16:58 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-11-16 19:40 Michael Forney
2022-11-19 16:14 ` Lucas Francesco
2022-11-19 16:56   ` Lucas Francesco [this message]
2022-12-02 20:17 ` cinap_lenrek

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to='CAF=5iUX_Eav33PH8wnVOqxk+VgDUbUXzcFBz3b2j=bqN2C2F6Q@mail.gmail.com' \
    --to=lucas.francesco93@gmail.com \
    --cc=9front@9front.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).