Hi Bertrand,
On Tue, Feb 3, 2026 at 6:38 PM Bertrand Marquis
<[email protected]> wrote:
>
> FFA_FEATURES currently accepts non-zero input properties (w2-w7) from
> guests and advertises several ABIs unconditionally, even when firmware
> support is missing or when the ABI is physical-instance-only. This can
> mislead guests about what Xen can actually provide and violates FF-A
> calling conventions. Some SPMCs (Hafnium v2.14 or earlier) also fail to
> report FFA_RX_ACQUIRE despite supporting it.
>
> Update FFA_FEATURES validation to match spec and firmware support:
> - reject non-zero w2-w7 input properties with INVALID_PARAMETERS
> - reject 64-bit calling conventions from 32-bit guests with NOT_SUPPORTED
> - return NOT_SUPPORTED for physical-instance-only ABIs
> (FFA_NOTIFICATION_BITMAP_{CREATE,DESTROY}, FFA_RX_ACQUIRE)
> - advertise FFA_INTERRUPT as supported
> - gate message ABIs on firmware support:
> - FFA_MSG_SEND_DIRECT_REQ_{32,64}
> - FFA_MSG_SEND_DIRECT_REQ2 (also requires FF-A 1.2 negotiation)
> - FFA_MSG_SEND2 (or VM-to-VM enabled)
> - report MEM_SHARE_{32,64} only when FFA_MEM_SHARE_64 is supported
> - stop advertising FFA_MSG_YIELD (not implemented)
>
> Update firmware probing: drop FFA_MEM_SHARE_32 checks (deprecated) and
> add FFA_RX_ACQUIRE to the probed set. If FFA_MSG_SEND2 is reported but
> FFA_RX_ACQUIRE is not, assume RX_ACQUIRE support and warn to work
> around the Hafnium bug.
>
> Functional impact: guests now see ABI support that reflects firmware
> capabilities and Xen implementation status. When SEND2 is present but
> RX_ACQUIRE is not reported, Xen assumes RX_ACQUIRE support.
>
> Signed-off-by: Bertrand Marquis <[email protected]>
> ---
> xen/arch/arm/tee/ffa.c | 62 +++++++++++++++++++++++++++++++++++++-----
> 1 file changed, 55 insertions(+), 7 deletions(-)
>
> diff --git a/xen/arch/arm/tee/ffa.c b/xen/arch/arm/tee/ffa.c
> index 6de2b9f8ac8e..e9e020bb0cb3 100644
> --- a/xen/arch/arm/tee/ffa.c
> +++ b/xen/arch/arm/tee/ffa.c
> @@ -91,10 +91,10 @@ static const struct ffa_fw_abi ffa_fw_abi_needed[] = {
> FW_ABI(FFA_PARTITION_INFO_GET),
> FW_ABI(FFA_NOTIFICATION_INFO_GET_64),
> FW_ABI(FFA_NOTIFICATION_GET),
> + FW_ABI(FFA_RX_ACQUIRE),
> FW_ABI(FFA_RX_RELEASE),
> FW_ABI(FFA_RXTX_MAP_64),
> FW_ABI(FFA_RXTX_UNMAP),
> - FW_ABI(FFA_MEM_SHARE_32),
> FW_ABI(FFA_MEM_SHARE_64),
> FW_ABI(FFA_MEM_RECLAIM),
> FW_ABI(FFA_MSG_SEND_DIRECT_REQ_32),
> @@ -240,19 +240,39 @@ static void handle_features(struct cpu_user_regs *regs)
> struct ffa_ctx *ctx = d->arch.tee;
> unsigned int n;
>
> + /*
> + * Xen does not accept any non-zero FFA_FEATURES input properties from
> + * VMs. The spec only defines w2 input properties for
> FFA_MEM_RETRIEVE_REQ
> + * (NS-bit negotiation for SP/SPMC) and FFA_RXTX_MAP (buffer size and
> + * alignment), so w2 must be MBZ for our callers.
> + */
The spec (version 1.2) lists them as SBZ, except for w2, which is MBZ,
for Feature IDs.
However, if we're to return an error, invalid parameters is a better choice.
Cheers,
Jens
> for ( n = 2; n <= 7; n++ )
> {
> if ( get_user_reg(regs, n) )
> {
> - ffa_set_regs_error(regs, FFA_RET_NOT_SUPPORTED);
> + ffa_set_regs_error(regs, FFA_RET_INVALID_PARAMETERS);
> return;
> }
> }
>
> + if ( !is_64bit_domain(d) && smccc_is_conv_64(a1) )
> + {
> + /* 32bit guests should only use 32bit convention calls */
> + ffa_set_regs_error(regs, FFA_RET_NOT_SUPPORTED);
> + return;
> + }
> +
> switch ( a1 )
> {
> + case FFA_NOTIFICATION_BITMAP_CREATE:
> + case FFA_NOTIFICATION_BITMAP_DESTROY:
> + case FFA_RX_ACQUIRE:
> + /* Physical-instance-only ABIs are not exposed to VMs. */
> + ffa_set_regs_error(regs, FFA_RET_NOT_SUPPORTED);
> + break;
> case FFA_ERROR:
> case FFA_VERSION:
> + case FFA_INTERRUPT:
> case FFA_SUCCESS_32:
> case FFA_SUCCESS_64:
> case FFA_FEATURES:
> @@ -261,16 +281,25 @@ static void handle_features(struct cpu_user_regs *regs)
> case FFA_RXTX_UNMAP:
> case FFA_MEM_RECLAIM:
> case FFA_PARTITION_INFO_GET:
> + ffa_set_regs_success(regs, 0, 0);
> + break;
> case FFA_MSG_SEND_DIRECT_REQ_32:
> case FFA_MSG_SEND_DIRECT_REQ_64:
> - case FFA_MSG_SEND2:
> case FFA_RUN:
> - case FFA_INTERRUPT:
> - case FFA_MSG_YIELD:
> - ffa_set_regs_success(regs, 0, 0);
> + if ( ffa_fw_supports_fid(a1) )
> + ffa_set_regs_success(regs, 0, 0);
> + else
> + ffa_set_regs_error(regs, FFA_RET_NOT_SUPPORTED);
> + break;
> + case FFA_MSG_SEND2:
> + if ( ffa_fw_supports_fid(a1) || IS_ENABLED(CONFIG_FFA_VM_TO_VM) )
> + ffa_set_regs_success(regs, 0, 0);
> + else
> + ffa_set_regs_error(regs, FFA_RET_NOT_SUPPORTED);
> break;
> case FFA_MSG_SEND_DIRECT_REQ2:
> - if ( ACCESS_ONCE(ctx->guest_vers) >= FFA_VERSION_1_2 )
> + if ( ACCESS_ONCE(ctx->guest_vers) >= FFA_VERSION_1_2 &&
> + ffa_fw_supports_fid(FFA_MSG_SEND_DIRECT_REQ2) )
> {
> ffa_set_regs_success(regs, 0, 0);
> }
> @@ -281,6 +310,11 @@ static void handle_features(struct cpu_user_regs *regs)
> break;
> case FFA_MEM_SHARE_64:
> case FFA_MEM_SHARE_32:
> + if ( !ffa_fw_supports_fid(FFA_MEM_SHARE_64) )
> + {
> + ffa_set_regs_error(regs, FFA_RET_NOT_SUPPORTED);
> + break;
> + }
> /*
> * We currently don't support dynamically allocated buffers. Report
> * that with 0 in bit[0] of w2.
> @@ -688,6 +722,20 @@ static bool ffa_probe_fw(void)
> ffa_fw_abi_needed[i].name);
> }
>
> + /*
> + * Hafnium v2.14 or earlier does not report FFA_RX_ACQUIRE in
> + * FFA_FEATURES even though it supports it.
> + */
> + if ( !ffa_fw_supports_fid(FFA_RX_ACQUIRE) &&
> + ffa_fw_supports_fid(FFA_MSG_SEND2) )
> + {
> + printk(XENLOG_WARNING
> + "ARM FF-A Firmware reports FFA_MSG_SEND2 without
> FFA_RX_ACQUIRE\n");
> + printk(XENLOG_WARNING
> + "ffa: assuming RX_ACQUIRE support (workaround)\n");
> + set_bit(FFA_ABI_BITNUM(FFA_RX_ACQUIRE), ffa_fw_abi_supported);
> + }
> +
> if ( !ffa_rxtx_spmc_init() )
> {
> printk(XENLOG_ERR "ffa: Error during RXTX buffer init\n");
> --
> 2.50.1 (Apple Git-155)
>
