On Wed, 9 Oct 2019 at 15:18, Hans de Goede <hdegoede@xxxxxxxxxx> wrote:
>
> Hi,
>
> On 09-10-2019 15:07, Ard Biesheuvel wrote:
> > On Fri, 4 Oct 2019 at 16:51, Hans de Goede <hdegoede@xxxxxxxxxx> wrote:
> >>
> >> Sometimes it is useful to be able to dump the efi boot-services code and
> >> data. This commit adds these as debugfs-blobs to /sys/kernel/debug/efi,
> >> but only if efi=debug is passed on the kernel-commandline as this requires
> >> not freeing those memory-regions, which costs 20+ MB of RAM.
> >>
> >> Reviewed-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> >> Acked-by: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx>
> >> Signed-off-by: Hans de Goede <hdegoede@xxxxxxxxxx>
> >> ---
> >> Changes in v5:
> >> -Rename the EFI_BOOT_SERVICES flag to EFI_PRESERVE_BS_REGIONS
> >>
> >> Changes in v4:
> >> -Add new EFI_BOOT_SERVICES flag and use it to determine if the boot-services
> >> memory segments are available (and thus if it makes sense to register the
> >> debugfs bits for them)
> >>
> >> Changes in v2:
> >> -Do not call pr_err on debugfs call failures
> >> ---
> >> arch/x86/platform/efi/efi.c | 1 +
> >> arch/x86/platform/efi/quirks.c | 4 +++
> >> drivers/firmware/efi/efi.c | 53 ++++++++++++++++++++++++++++++++++
> >> include/linux/efi.h | 1 +
> >> 4 files changed, 59 insertions(+)
> >>
> >> diff --git a/arch/x86/platform/efi/efi.c b/arch/x86/platform/efi/efi.c
> >> index c202e1b07e29..847730f7e74b 100644
> >> --- a/arch/x86/platform/efi/efi.c
> >> +++ b/arch/x86/platform/efi/efi.c
> >> @@ -232,6 +232,7 @@ int __init efi_memblock_x86_reserve_range(void)
> >> efi.memmap.desc_version);
> >>
> >> memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
> >> + set_bit(EFI_PRESERVE_BS_REGIONS, &efi.flags);
> >
> > Should we add a Kconfig symbol to opt into this behavior [set by the
> > driver in question], instead of always preserving all boot services
> > regions on all x86 systems?
>
> This bit does not control anything, it merely signals that the arch early
> boot EFI code keeps the boot-services code around, which is something
> which the x86 code already does. Where as e.g. on arm / aarch64 this is
> freed early on, this ties in with the other bits:
>
> >
> >>
> >> return 0;
> >> }
> >> diff --git a/arch/x86/platform/efi/quirks.c b/arch/x86/platform/efi/quirks.c
> >> index 3b9fd679cea9..fab12ebf0ada 100644
> >> --- a/arch/x86/platform/efi/quirks.c
> >> +++ b/arch/x86/platform/efi/quirks.c
> >> @@ -411,6 +411,10 @@ void __init efi_free_boot_services(void)
> >> int num_entries = 0;
> >> void *new, *new_md;
> >>
> >> + /* Keep all regions for /sys/kernel/debug/efi */
> >> + if (efi_enabled(EFI_DBG))
> >> + return;
> >> +
>
> This is the point where normally on x86 we do actually free the boot-services
> code which is a lot later then on other arches. And this new code actually
> does change things to keep the boot-services code *forever* but only
> if EFI debugging is enabled on the kernel commandline.
>
I get this part. But at some point, your driver is going to expect
this memory to be preserved even if EFI_DBG is not set, right? My
question was whether we should only opt into that if such a driver is
enabled in the first place.
> This ties in with the next bit:
>
> >> for_each_efi_memory_desc(md) {
> >> unsigned long long start = md->phys_addr;
> >> unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
> >> diff --git a/drivers/firmware/efi/efi.c b/drivers/firmware/efi/efi.c
> >> index 8d3e778e988b..abba49c4c46d 100644
> >> --- a/drivers/firmware/efi/efi.c
> >> +++ b/drivers/firmware/efi/efi.c
>
> <snip>
>
> >> @@ -370,6 +420,9 @@ static int __init efisubsys_init(void)
> >> goto err_remove_group;
> >> }
> >>
> >> + if (efi_enabled(EFI_DBG) && efi_enabled(EFI_PRESERVE_BS_REGIONS))
> >> + efi_debugfs_init();
> >> +
> >> return 0;
> >>
> >> err_remove_group:
>
> Here we register the debugfs dir + files, but only when the
> boot services code has been kept around, so only if the
> EFI_PRESERVE_BS_REGIONS arch feature flag has been set and
> EFI debugging has been requested on the kernel commandline.
>
> IOW this patch already offers to configurability you ask for, but instead
> of through a Kconfig option (which IMHO would be cumbersome) the decision
> is made runtime based on the presence of efi=debug on the kernel commandline.
>
> Regards,
>
> Hans
>
>
>
>
> >> diff --git a/include/linux/efi.h b/include/linux/efi.h
> >> index bd3837022307..2a30a1bd8bdf 100644
> >> --- a/include/linux/efi.h
> >> +++ b/include/linux/efi.h
> >> @@ -1202,6 +1202,7 @@ extern int __init efi_setup_pcdp_console(char *);
> >> #define EFI_DBG 8 /* Print additional debug info at runtime */
> >> #define EFI_NX_PE_DATA 9 /* Can runtime data regions be mapped non-executable? */
> >> #define EFI_MEM_ATTR 10 /* Did firmware publish an EFI_MEMORY_ATTRIBUTES table? */
> >> +#define EFI_PRESERVE_BS_REGIONS 11 /* Are EFI boot-services memory segments available? */
> >>
> >> #ifdef CONFIG_EFI
> >> /*
> >> --
> >> 2.23.0
> >>
