From: Michael Tretter <m.tretter@pengutronix.de>
To: Sascha Hauer <s.hauer@pengutronix.de>
Cc: BAREBOX <barebox@lists.infradead.org>
Subject: Re: [PATCH 2/2] arm: socfpga: agilex5: add SSBL verification
Date: Wed, 18 Nov 2026 11:38:21 +0100 [thread overview]
Message-ID: <av2AHZSRTa-gfZF3@pengutronix.de> (raw)
In-Reply-To: <15cbf71b-3da6-4993-98f9-0d5357fdbc54@pengutronix.de>
On Mon, 26 Oct 2026 13:32:40 +0000, Sascha Hauer wrote:
> On 2026-10-21 22:33, Michael Tretter wrote:
> > Altera VAB usually uses a certificate that is appended to the SSBL image
> > using the fcs_prepare [0] tool. The certificate contains a sha384
> > checksum and is signed using a key chain that ends with the root key
> > that is written to the hardware.
> >
> > Assume that the certificate is appended to the SSBL image. Read the
> > certificate from the same flash partition as the SSBL, pass it to the
> > SDM firmware for verification and check against the root key, and
> > finally verify the contained checksum in barebox.
> >
> > The SDM firmware returns success for the certificate verification if VAB
> > is disabled. Create a dummy certificate to check if VAB boot is disabled
> > and accept an unsigned SSBL in this case. If there is a valid
> > certificate attached, the checksum has to match, even if VAB is
> > disabled.
> >
> > The implementation is a bit quirky, because
> >
> > 1. it needs to deal with the broken SDM firmware behavior, which expects
> > a 32 bit word with value 0 in front of the certificate,
> >
> > 2. needs to walk and parse the (still untrusted) key chain to find the
> > actual size of the certificate, and
> >
> > 3. I haven't found a better way if VAB is enabled than to check a dummy
> > certificate.
> >
> > The VAB code could be moved to a separate module to make it available to
> > other SoCFPGA platforms or to barebox proper. Since it's only used for
> > SSBL verification for now and is mingled with the SSBL loading, I kept
> > it in the xload code.
> >
> > [0] https://github.com/altera-fpga/libfcs
> >
> > Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
> > ---
> > arch/arm/mach-socfpga/agilex5-xload.c | 157 ++++++++++++++++++++++++++++++++++
> > 1 file changed, 157 insertions(+)
> >
> > diff --git a/arch/arm/mach-socfpga/agilex5-xload.c b/arch/arm/mach-socfpga/agilex5-xload.c
> > index 9e55252c04cc..077f67014945 100644
> > --- a/arch/arm/mach-socfpga/agilex5-xload.c
> > +++ b/arch/arm/mach-socfpga/agilex5-xload.c
> > @@ -5,6 +5,11 @@
> > #include <linux/sizes.h>
> > #include <mach/socfpga/generic.h>
> > #include <pbl/bio.h>
> > +#include <mach/socfpga/mailbox_s10.h>
> > +#include <mach/socfpga/soc64-system-manager.h>
> > +
> > +#include <crypto/pbl-sha.h>
> > +#include <crypto/sha.h>
> >
> > static inline u32 barebox_arm_size(const char *head)
> > {
> > @@ -52,9 +57,128 @@ static ssize_t agilex5_bl33_read(struct pbl_bio *bio, void *bl33,
> > return size;
> > }
> >
> > +struct vab_certificate_data {
> > + u32 magic;
> > + u32 flags;
> > + u32 reserved0;
> > + u32 reserved1;
> > + u8 hash[SHA384_DIGEST_SIZE];
> > + u32 measurement;
> > + u32 reserved2;
> > +};
> > +
> > +struct vab_certificate_int {
> > + u32 magic;
> > + u32 size;
> > + u32 version;
> > + u32 type;
> > + struct vab_certificate_data d;
> > + u32 keychain[0];
> > +};
> > +
> > +/*
> > + * Use a wrapper struct that starts with a zero word to be able to pass the
> > + * certificate to the SDM firmware as it. Let the code that reads the
> > + * certificate deal with it to avoid a copy of the certificate.
> > + */
> > +struct vab_certificate {
> > + u32 zero;
> > + struct vab_certificate_int cert;
> > +};
> > +
> > +static ssize_t vab_get_cert_size(struct vab_certificate_int *cert,
> > + size_t size)
> > +{
> > + struct {
> > + u32 magic;
> > + u32 size;
> > + } *header;
> > + ssize_t offset = 0;
> > +
> > + /*
> > + * Each element in the key chain has a header with some magic and the
> > + * size of the element. The length of the entire key chain is attached
> > + * to the end of the key chain. Walk the entire key chain to find the
> > + * actual size of the key chain.
> > + *
> > + * Note that the header is still untrusted!
> > + */
> > + while (offset < size) {
> > + header = ((void *)cert) + offset;
> > + if (header->magic == offset)
> > + break;
> > + offset += header->size;
> > + if (offset < 0 || offset > size)
> > + return -EINVAL;
> > + }
> > +
> > + return offset;
> > +}
> > +
> > +static ssize_t agilex5_vab_read(struct pbl_bio *bio,
> > + struct vab_certificate *cert,
> > + off_t start, size_t size)
> > +{
> > + ssize_t ret;
> > + struct vab_certificate_int *tmp = &(cert->cert);
> > + size_t tmp_size = size - sizeof(u32);
> > +
> > + cert->zero = 0;
> > + ret = pbl_bio_read(bio, start, tmp, tmp_size);
> > + if (ret < 0)
> > + return -EINVAL;
> > +
> > + if (tmp->magic != 0x25d04e7f) {
> > + pr_err("vab cert: invalid magic 0x%x\n", tmp->magic);
> > + return -EINVAL;
> > + }
> > + if (tmp->size != sizeof(*tmp)) {
> > + pr_err("vab cert: invalid size %u", tmp->size);
> > + return -EINVAL;
> > + }
> > +
> > + ret = vab_get_cert_size(tmp, tmp_size);
> > + if (ret <= 0) {
> > + pr_err("vab cert: invalid certificate chain\n");
> > + return -EINVAL;
> > + }
> > +
> > + pr_info("read vab cert to %p from %lx (%zu bytes)\n",
> > + tmp, start, ret);
> > +
> > + return ret + sizeof(u32);
> > +}
> > +
> > +static ssize_t agilex5_vab_dummy(struct vab_certificate *cert)
> > +{
> > + struct vab_certificate_int *tmp = &(cert->cert);
> > +
> > + tmp->magic = 0x25d04e7f;
> > + tmp->size = sizeof(*tmp);
> > + tmp->version = 1;
> > + tmp->type = 6;
> > +
> > + return tmp->size;
> > +}
> > +
> > +static int agilex5_vab_verify(const void *data, size_t size,
> > + struct vab_certificate *cert)
> > +{
> > + char computed_hash[SHA384_DIGEST_SIZE];
> > + char *hash = cert->cert.d.hash;
> > +
> > + pbl_sha384(data, size, computed_hash);
> > +
> > + return memcmp(hash, computed_hash, SHA384_DIGEST_SIZE);
> > +}
> > +
> > int agilex5_xload_bl33(struct pbl_bio *bio, void *bl33)
> > {
> > ssize_t size;
> > + off_t cert_offset;
> > + ssize_t cert_size;
> > + struct vab_certificate *cert;
> > + int ret;
> >
> > if (!bio) {
> > pr_warn("source device for SSBL is not configured\n");
> > @@ -70,5 +194,38 @@ int agilex5_xload_bl33(struct pbl_bio *bio, void *bl33)
> > if (size < 0)
> > return size;
> >
> > + /*
> > + * The certificate is 8 byte aligned and attached to the SSBL binary.
> > + * The certificate must not be larger than the partition and is
> > + * usually smaller than 4K.
> > + */
> > + size = ALIGN(size, 8);
> > + cert_offset = barebox_part->nor_offset + size;
> > + cert_size = min_t(size_t, SZ_4K, barebox_part->nor_size - size);
> > +
> > + /* Use the SDRAM behind the bl33 for the certificate */
> > + cert = bl33 + size;
> > + cert_size = agilex5_vab_read(bio, cert, cert_offset, cert_size);
> > + if (cert_size < 0) {
> > + /*
> > + * There is no valid certificate. Generate a dummy certificate
> > + * to ask the firmware if authentication bypass is allowed.
> > + */
> > + cert_size = agilex5_vab_dummy(cert);
> > + }
> > +
> > + ret = mbox_vab_src_cert(cert, cert_size);
> > + if (ret)
> > + return ret;
> > +
> > + ret = agilex5_vab_verify(bl33, size, cert);
> > + if (ret) {
> > + if (cert_size != sizeof(*cert)) {
> > + pr_err("SSBL hash mismatch\n");
> > + return ret;
> > + }
>
> This is unnecessary hard to read. I assume cert_size != sizeof(*cert)
> means "This is not the dummy cert generated above", because a real cert
> must always have data besides the header, right?
Correct. A real cert has a certificate chain attached as data.
>
> > + pr_debug("SSBL hash mismatch: Bypass authentication\n");
>
> The dummy cert never passes agilex5_vab_verify(), so it's not necessary
> to check its sha384 sum.
>
> I think you should rather have a dummy_cert boolean and return
> successfully in the dummy case right after calling mbox_vab_src_cert().
> With this you would also not have to return successfully after a failed
> verification, which raises eyebrows.
I think I first wanted to accept images with an invalid signature on
devices that don't have VAB enabled, but warn the developer about it,
before I changed my mind to always reject signed images that have an
invalid signature.
I'll fix this and send a v2.
Michael
prev parent reply other threads:[~2026-11-18 11:55 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-10-21 20:33 [PATCH 0/2] arm: socfpga: agilex5: add VAB verification for SSBL Michael Tretter
2026-10-21 20:33 ` [PATCH 1/2] arm: socfpga: mailbox_s10: add certificate validation Michael Tretter
2026-10-21 20:33 ` [PATCH 2/2] arm: socfpga: agilex5: add SSBL verification Michael Tretter
2026-10-26 13:32 ` Sascha Hauer
2026-11-18 10:38 ` Michael Tretter [this message]
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=av2AHZSRTa-gfZF3@pengutronix.de \
--to=m.tretter@pengutronix.de \
--cc=barebox@lists.infradead.org \
--cc=s.hauer@pengutronix.de \
/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