Re: [PATCH 1/1] x86/boot: clear some fields explicitly

From: John Hubbard
Date: Thu Jul 25 2019 - 16:33:37 EST


On 7/25/19 12:22 AM, Thomas Gleixner wrote:
> On Wed, 24 Jul 2019, hpa@xxxxxxxxx wrote:
>> On July 24, 2019 4:15:28 PM PDT, john.hubbard@xxxxxxxxx wrote:
>>> From: John Hubbard <jhubbard@xxxxxxxxxx>
>>>
>>> Recent gcc compilers (gcc 9.1) generate warnings about an
>>> out of bounds memset, if you trying memset across several fields
>>> of a struct. This generated a couple of warnings on x86_64 builds.
>>>
>>> Because struct boot_params is __packed__, normal variable
>>> variable assignment will work just as well as a memset here.
>>> Change three u32 fields to be cleared to zero that way, and
>>> just memset the _pad4 field.
>>>
>>> This clears up the build warnings for me.
>>>
>>> Signed-off-by: John Hubbard <jhubbard@xxxxxxxxxx>
>>> ---
>>> arch/x86/include/asm/bootparam_utils.h | 11 +++++------
>>> 1 file changed, 5 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/arch/x86/include/asm/bootparam_utils.h
>>> b/arch/x86/include/asm/bootparam_utils.h
>>> index 101eb944f13c..4df87d4a043b 100644
>>> --- a/arch/x86/include/asm/bootparam_utils.h
>>> +++ b/arch/x86/include/asm/bootparam_utils.h
>>> @@ -37,12 +37,11 @@ static void sanitize_boot_params(struct boot_params
>>> *boot_params)
>>> if (boot_params->sentinel) {
>>> /* fields in boot_params are left uninitialized, clear them */
>>> boot_params->acpi_rsdp_addr = 0;
>>> - memset(&boot_params->ext_ramdisk_image, 0,
>>> - (char *)&boot_params->efi_info -
>>> - (char *)&boot_params->ext_ramdisk_image);
>>> - memset(&boot_params->kbd_status, 0,
>>> - (char *)&boot_params->hdr -
>>> - (char *)&boot_params->kbd_status);
>>> + boot_params->ext_ramdisk_image = 0;
>>> + boot_params->ext_ramdisk_size = 0;
>>> + boot_params->ext_cmd_line_ptr = 0;
>>> +
>>> + memset(&boot_params->_pad4, 0, sizeof(boot_params->_pad4));
>>> memset(&boot_params->_pad7[0], 0,
>>> (char *)&boot_params->edd_mbr_sig_buffer[0] -
>>> (char *)&boot_params->_pad7[0]);
>>
>> The problem with this is that it will break silently when changes are
>> made to this structure.
>
> That's not really the worst problem. Changes to that struct which touch any
> of the to be cleared ranges will break anyway if not handled correctly in
> the sanitizer function.
>
> What's worse is that the patch is broken. It 'clears' the build warnings,
> but not all the fields which have been cleared before:
>
> It removes the clearing of the range between kbd_status and hdr without any
> replacement. It neither clears edid_info.


Yes. Somehow I left that chunk out. Not my finest hour.

>
> The above approach is doomed and if we have to handle this GCC0.1 madness
> then this needs to be done smarter. Something like the completely untested
> thing below.
>
> Thanks,
>
> tglx
>
> 8<--------------
> diff --git a/arch/x86/include/asm/bootparam_utils.h b/arch/x86/include/asm/bootparam_utils.h
> index 101eb944f13c..f5bc4c01b66b 100644
> --- a/arch/x86/include/asm/bootparam_utils.h
> +++ b/arch/x86/include/asm/bootparam_utils.h
> @@ -9,6 +9,18 @@
> * add completing #includes to make it standalone.
> */
>
> +struct boot_params_clear {
> + unsigned int offs;
> + unsigned int len;
> +};
> +
> +#define BOOT_PARAM_CLEAR(start, end) \
> +{ \
> + .offs = offsetof(struct boot_params, start), \
> + .len = offsetof(struct boot_params, end) - \
> + offsetof(struct boot_params, start), \
> +}
> +
> /*
> * Deal with bootloaders which fail to initialize unknown fields in
> * boot_params to zero. The list fields in this list are taken from
> @@ -20,7 +32,19 @@
> */
> static void sanitize_boot_params(struct boot_params *boot_params)
> {
> - /*
> + const struct boot_params_clear toclear[] = {
> + BOOT_PARAM_CLEAR(acpi_rdsp_addr, _pad3),
> + BOOT_PARAM_CLEAR(ext_ramdisk_image, efi_info),
> + BOOT_PARAM_CLEAR(kbd_status, hdr),
> + BOOT_PARAM_CLEAR(_pad7, edd_mbr_sig_buffer),
> + BOOT_PARAM_CLEAR(_pad8, eddbuf),
> + {
> + .offs = offsetof(struct boot_params, _pad9),
> + .len = sizeof(boot_params->_pad9),
> + },
> + }
> +
> + /*
> * IMPORTANT NOTE TO BOOTLOADER AUTHORS: do not simply clear
> * this field. The purpose of this field is to guarantee
> * compliance with the x86 boot spec located in
> @@ -36,20 +60,11 @@ static void sanitize_boot_params(struct boot_params *boot_params)
> */
> if (boot_params->sentinel) {
> /* fields in boot_params are left uninitialized, clear them */
> - boot_params->acpi_rsdp_addr = 0;
> - memset(&boot_params->ext_ramdisk_image, 0,
> - (char *)&boot_params->efi_info -
> - (char *)&boot_params->ext_ramdisk_image);
> - memset(&boot_params->kbd_status, 0,
> - (char *)&boot_params->hdr -
> - (char *)&boot_params->kbd_status);
> - memset(&boot_params->_pad7[0], 0,
> - (char *)&boot_params->edd_mbr_sig_buffer[0] -
> - (char *)&boot_params->_pad7[0]);
> - memset(&boot_params->_pad8[0], 0,
> - (char *)&boot_params->eddbuf[0] -
> - (char *)&boot_params->_pad8[0]);
> - memset(&boot_params->_pad9[0], 0, sizeof(boot_params->_pad9));
> + char *p = (char *) boot_params;
> + int i;
> +
> + for (i = 0; i < ARRAY_SIZE(toclear); i++)
> + memset(p + toclear[i].start, 0, toclear[i].len);
> }
> }


Looks nice.


thanks,
--
John Hubbard
NVIDIA