Re: [PATCH 10/10] xen/arm: ffa: advertise FF-A v1.2

[Bertrand Marquis <Bertrand.Marquis@xxxxxxx>]

Hi Jens,

> On 4 Dec 2025, at 08:59, Jens Wiklander <jens.wiklander@xxxxxxxxxx> wrote:
> 
> Hi Bertrand,
> 
> On Thu, Nov 27, 2025 at 4:53 PM Bertrand Marquis
> <bertrand.marquis@xxxxxxx> wrote:
>> 
>> Advertise RUN/YIELD/INTERRUPT via FFA_FEATURES now that the runtime
>> helpers are in place, and expose the RX/TX MAP capacity field only once
>> a guest has negotiated FF-A v1.2.
>> 
>> While there, drop the stale <asm/tee/ffa.h> include.
>> 
>> To comply with the wider v1.2 register ABI, zero registers x8–x17 when
>> responding to a v1.2 VM. The dispatcher also rejects SMCCC64 calls from
>> AArch32 guests.
>> 
>> Finally, bump Xen's FF-A version to 1.2.
>> 
>> Signed-off-by: Bertrand Marquis <bertrand.marquis@xxxxxxx>
>> ---
>> xen/arch/arm/tee/ffa.c         | 25 ++++++++++++++++++++-----
>> xen/arch/arm/tee/ffa_private.h | 28 +++++++++++++++++++---------
>> 2 files changed, 39 insertions(+), 14 deletions(-)
>> 
>> diff --git a/xen/arch/arm/tee/ffa.c b/xen/arch/arm/tee/ffa.c
>> index 8b2f042287fc..c8fe97856d06 100644
>> --- a/xen/arch/arm/tee/ffa.c
>> +++ b/xen/arch/arm/tee/ffa.c
>> @@ -24,9 +24,6 @@
>>  * o FFA_MEM_DONATE_* and FFA_MEM_LEND_* - Used when tranferring ownership
>>  *   or access of a memory region
>>  * o FFA_MSG_SEND2 and FFA_MSG_WAIT - Used for indirect messaging
>> - * o FFA_MSG_YIELD
>> - * o FFA_INTERRUPT - Used to report preemption
>> - * o FFA_RUN
>>  *
>>  * Limitations in the implemented FF-A interfaces:
>>  * o FFA_RXTX_MAP_*:
>> @@ -68,7 +65,6 @@
>> #include <asm/event.h>
>> #include <asm/regs.h>
>> #include <asm/smccc.h>
>> -#include <asm/tee/ffa.h>
>> #include <asm/tee/tee.h>
>> 
>> #include "ffa_private.h"
>> @@ -262,6 +258,9 @@ static void handle_features(struct cpu_user_regs *regs)
>>     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);
>>         break;
>>     case FFA_MSG_SEND_DIRECT_REQ2:
>> @@ -290,7 +289,16 @@ static void handle_features(struct cpu_user_regs *regs)
>>          * differs from FFA_PAGE_SIZE (SZ_4K).
>>          */
>>         BUILD_BUG_ON(PAGE_SIZE != FFA_PAGE_SIZE);
>> -        ffa_set_regs_success(regs, 0, 0);
>> +
>> +        /*
>> +         * From FFA v1.2, we can give the maximum number of pages we support
>> +         * for the RX/TX buffers.
>> +         */
>> +        if ( ctx->guest_vers < FFA_VERSION_1_2 )
>> +            ffa_set_regs_success(regs, 0, 0);
>> +        else
>> +            ffa_set_regs_success(regs, FFA_MAX_RXTX_PAGE_COUNT << 16, 0);
>> +
>>         break;
>>     case FFA_FEATURE_NOTIF_PEND_INTR:
>>         ffa_set_regs_success(regs, GUEST_FFA_NOTIF_PEND_INTR_ID, 0);
>> @@ -323,6 +331,13 @@ static bool ffa_handle_call(struct cpu_user_regs *regs)
>>     if ( !ctx )
>>         return false;
>> 
>> +    if ( !is_64bit_domain(d) && smccc_is_conv_64(fid) )
> 
> You use ctx->is_64bit in ffa_set_regs() below, but here
> is_64bit_domain(d). Which one do we prefer?

It would definitely be more coherent to use ctx->is_64bit.

I will fix that in next version.

Cheers
Bertrand