Re: [SeaBIOS] [PATCH v9 2/6] Implementation of the TCG BIOS extensions

[Stefan Berger]

On 03/20/2015 09:15 PM, Kevin O'Connor wrote:
> On Fri, Mar 20, 2015 at 02:00:37PM -0400, Stefan Berger wrote:
> > From: Stefan Berger
> >
> > This patch implements the main part of the TCG BIOS extensions. It provides
> > the following functionality:
> > initialization of the TCPA ACPI table used for logging of measurements
> > initialization of the TPM by sending a sequence of commands to it
> > proper setup of the TPM before the BIOS hands over control to the bootloader
> > support for S3 resume; BIOS sends TPM_Startup(ST_STATE) to TPM
> > enable configuration of SeaBIOS to be built with TCGBIOS extensions
> >  All TCG BIOS extensions are activated with CONFIG_TCGBIOS.
>
> [...]
> > --- a/src/Kconfig
> > +++ b/src/Kconfig
> > @@ -421,6 +421,13 @@ menu "BIOS interfaces"
> >               modified by programs.  However, some old DOS high memory
> >               managers may require the UMB region to be read-only.
> > config TCGBIOS
> >    select S3_RESUME
> >    bool "TPM support and TCG BIOS extensions"
> >    default y
> >    help
> >        Provide TPM support along with TCG BIOS extensions
>
> I don't think TCGBIOS should select S3_RESUME.  If resume processing
> is an absolute requirement then I think it should use "depends on".

Ok.
> [...]
> > --- a/src/boot.c
> > +++ b/src/boot.c
> > @@ -19,6 +19,7 @@
> >   #include "std/disk.h" // struct mbr_s
> >   #include "string.h" // memset
> >   #include "util.h" // irqtimer_calc
> > +#include "tcgbios.h" // tpm_*
> >
> > /****************************************************************
> > @@ -475,6 +476,7 @@ interactive_bootmenu(void)
> >   printf("Select boot device:\n\n");
> >   wait_threads();
> > tpm_leave_bios();
>
> Is this call really supposed to be in the boot menu?  The call wont
> get invoked if the boot menu isn't selected.

Right. I moved it into post.c maininit() after the wait_threads() call 
there.
> > --- a/src/post.c
> > +++ b/src/post.c
> > @@ -28,6 +28,7 @@
> >   #include "output.h" // dprintf
> >   #include "string.h" // memset
> >   #include "util.h" // kbd_init
> > +#include "tcgbios.h" // tpm_*
> >
> > /****************************************************************
> > @@ -220,6 +221,10 @@ maininit(void)
> >       if (threads_during_optionroms())
> >           device_hardware_setup();
> > // Initialize tpm (after acpi tables were written)
> > tpm_acpi_init();
> > tpm_startup();
>
> I think this should be just one call - such as "tpm_setup()", which in
> turn calls tpm_xyz().
>
> Also, functions with "_init" and "_setup" suffixes have specific
> meaning in the bootstrap ordering sequence.  So tpm shouldn't export
> functions with these suffixes unless they are invoked during the given
> sequence.

I move the tpm_acpi_init() call into tpm_startup().
> [...]
> > +u32
> > +tpm_startup(void)
> > +{
> > u32 rc;
> > u32 returnCode;
> > if (!CONFIG_TCGBIOS)
> >    return 0;
> > if (!has_working_tpm())
> >    return TCG_GENERAL_ERROR;
> > dprintf(DEBUG_tcg, "TCGBIOS: Starting with TPM_Startup(ST_CLEAR)\n");
> > rc = build_and_send_cmd(TPM_ORD_Startup,
> >                        Startup_ST_CLEAR, sizeof(Startup_ST_CLEAR),
> >                        NULL, 10, &returnCode, TPM_DURATION_TYPE_SHORT);
> > dprintf(DEBUG_tcg, "Return code from TPM_Startup = 0x%08x\n",
> >        returnCode);
> >
> > +#if CONFIG_COREBOOT
> > /* with other firmware on the system the TPM may already have been
> > * initialized
> > */
> > if (returnCode == TPM_INVALID_POSTINIT)
> >    returnCode = 0;
> >
> > +#endif
>
> That '#if' should be turned into an 'if (...)'.
>
> [...]

Done
> > +u32
> > +tpm_s3_resume(void)
> > +{
> > u32 rc;
> > u32 returnCode;
> > if (!CONFIG_TCGBIOS)
> >    return 0;
> > if (!has_working_tpm())
> >    return TCG_GENERAL_ERROR;
> > timer_setup();
>
> Why is the timer reinitialized here?  There may be implications in
> touching that hardware during an S3 resume.

I am initializing the time because of the TPM driver's usage of 
msleep(). Your comment reminds me of a comment years ago from Keir 
Fraser from Xen. There, we ended up using I think a DRAM refresh timer 
bit that was emulated by QEMU and flipped every few ns.
> > dprintf(DEBUG_tcg, "TCGBIOS: Resuming with TPM_Startup(ST_STATE)\n");
> > rc = build_and_send_cmd(TPM_ORD_Startup,
> >                        Startup_ST_STATE, sizeof(Startup_ST_STATE),
> >                        NULL, 10, &returnCode, TPM_DURATION_TYPE_SHORT);
> > dprintf(DEBUG_tcg, "TCGBIOS: ReturnCode from TPM_Startup = 0x%08x\n",
> >        returnCode);
> > if (rc || returnCode)
> >    goto err_exit;
> > return 0;
> >
> > +err_exit:
> > dprintf(DEBUG_tcg, "TCGBIOS: TPM malfunctioning (line %d).\n", __LINE__);
> > tpm_state.tpm_working = 0;
> > if (rc)
> >    return rc;
> > return TCG_TCG_COMMAND_ERROR;
> >
> > +}
> > diff --git a/src/tcgbios.h b/src/tcgbios.h
> > new file mode 100644
> > index 0000000..7be89f8
> > --- /dev/null
> > +++ b/src/tcgbios.h
> > @@ -0,0 +1,388 @@
> > +#ifndef TCGBIOS_H
> > +#define TCGBIOS_H
> >
> > +#include "types.h"
> > +#include "bregs.h" /* struct bregs */
>
> Why is "bregs.h" needed in the header - can it be included in the
> individual C files that need it?

This #include needs to only appear in the 3rd patch when the interrupt 
handler prototype appears. I move it there now.

Stefan