Re: [PATCH v2 07/10] xen/arm: ffa: Transmit RXTX buffers to the SPMC

[Bertrand Marquis <Bertrand.Marquis@xxxxxxx>]

Hi Jens,

> On 23 Oct 2024, at 16:51, Jens Wiklander <jens.wiklander@xxxxxxxxxx> wrote:
> 
> Hi Bertrand,
> 
> On Wed, Oct 16, 2024 at 10:32 AM Bertrand Marquis
> <bertrand.marquis@xxxxxxx> wrote:
>> 
>> When an RXTX buffer is mapped by a VM transmit it to the SPMC when it
>> supports RX_ACQUIRE.
>> As a consequence of that, we must acquire the RX buffer of a VM from the
>> SPMC when we want to use it:
>> - create a generic acquire and release function to get the rx buffer of
>>  a VM which gets it from the SPMC when supported
>> - rename the rx_acquire to hyp_rx_acquire to remove confusion
>> - rework the rx_lock to only lock access to rx_is_free and only allow
>>  usage of the rx buffer to one who managed to acquire it, thus removing
>>  the trylock and returning busy if rx_is_free is false
>> 
>> As part of this change move some structure definition to ffa_private
>> from ffa_shm as those are need for the MAP call with the SPMC.
>> 
>> Signed-off-by: Bertrand Marquis <bertrand.marquis@xxxxxxx>
>> ---
>> Changes in v2:
>> - unmap VM rxtx buffer in SPMC on unmap call or on VM destroy
>> - rework the unmap call to the SPMC to properly pass the VM ID
>> ---
>> xen/arch/arm/tee/ffa.c          |   2 +-
>> xen/arch/arm/tee/ffa_partinfo.c |  29 ++----
>> xen/arch/arm/tee/ffa_private.h  |  22 ++++-
>> xen/arch/arm/tee/ffa_rxtx.c     | 158 ++++++++++++++++++++++++++------
>> xen/arch/arm/tee/ffa_shm.c      |  15 ---
>> 5 files changed, 161 insertions(+), 65 deletions(-)
>> 
>> diff --git a/xen/arch/arm/tee/ffa.c b/xen/arch/arm/tee/ffa.c
>> index a292003ca9fe..40ea5398fa21 100644
>> --- a/xen/arch/arm/tee/ffa.c
>> +++ b/xen/arch/arm/tee/ffa.c
>> @@ -347,7 +347,7 @@ static bool ffa_handle_call(struct cpu_user_regs *regs)
>>         ffa_handle_partition_info_get(regs);
>>         return true;
>>     case FFA_RX_RELEASE:
>> -        e = ffa_handle_rx_release();
>> +        e = ffa_rx_release(d);
>>         break;
>>     case FFA_MSG_SEND_DIRECT_REQ_32:
>>     case FFA_MSG_SEND_DIRECT_REQ_64:
>> diff --git a/xen/arch/arm/tee/ffa_partinfo.c 
>> b/xen/arch/arm/tee/ffa_partinfo.c
>> index 3cf801523296..fde187dba4e5 100644
>> --- a/xen/arch/arm/tee/ffa_partinfo.c
>> +++ b/xen/arch/arm/tee/ffa_partinfo.c
>> @@ -121,11 +121,9 @@ void ffa_handle_partition_info_get(struct cpu_user_regs 
>> *regs)
>>         goto out;
>>     }
>> 
>> -    if ( !spin_trylock(&ctx->rx_lock) )
>> -    {
>> -        ret = FFA_RET_BUSY;
>> +    ret = ffa_rx_acquire(d);
>> +    if ( ret != FFA_RET_OK )
>>         goto out;
>> -    }
>> 
>>     dst_buf = ctx->rx;
>> 
>> @@ -135,22 +133,16 @@ void ffa_handle_partition_info_get(struct 
>> cpu_user_regs *regs)
>>         goto out_rx_release;
>>     }
>> 
>> -    if ( !ctx->page_count || !ctx->rx_is_free )
>> -    {
>> -        ret = FFA_RET_DENIED;
>> -        goto out_rx_release;
>> -    }
>> -
>>     spin_lock(&ffa_rx_buffer_lock);
>> 
>>     ret = ffa_partition_info_get(uuid, 0, &ffa_sp_count, &src_size);
>> 
>>     if ( ret )
>> -        goto out_rx_buf_unlock;
>> +        goto out_rx_hyp_unlock;
>> 
>>     /*
>>      * ffa_partition_info_get() succeeded so we now own the RX buffer we
>> -     * share with the SPMC. We must give it back using ffa_rx_release()
>> +     * share with the SPMC. We must give it back using ffa_hyp_rx_release()
>>      * once we've copied the content.
>>      */
>> 
>> @@ -193,15 +185,13 @@ void ffa_handle_partition_info_get(struct 
>> cpu_user_regs *regs)
>>         }
>>     }
>> 
>> -    ctx->rx_is_free = false;
>> -
>> out_rx_hyp_release:
>> -    ffa_rx_release();
>> -out_rx_buf_unlock:
>> +    ffa_hyp_rx_release();
>> +out_rx_hyp_unlock:
>>     spin_unlock(&ffa_rx_buffer_lock);
>> out_rx_release:
>> -    spin_unlock(&ctx->rx_lock);
>> -
>> +    if ( ret != FFA_RET_OK )
>> +        ffa_rx_release(d);
> 
> Please comment on why ffa_rx_release() must only be called on failure.

It is because the buffer contains data for the caller in case of success so
it must be released by the caller.

Do i get it right that you want me to add a comment saying that in the code
and not only tell you here ?

> 
>> out:
>>     if ( ret )
>>         ffa_set_regs_error(regs, ret);
>> @@ -368,8 +358,7 @@ bool ffa_partinfo_init(void)
>>     ret = init_subscribers(count, fpi_size);
>> 
>> out:
>> -    ffa_rx_release();
>> -
>> +    ffa_hyp_rx_release();
>>     return ret;
>> }
>> 
>> diff --git a/xen/arch/arm/tee/ffa_private.h b/xen/arch/arm/tee/ffa_private.h
>> index afe69b43dbef..9adfe687c3c9 100644
>> --- a/xen/arch/arm/tee/ffa_private.h
>> +++ b/xen/arch/arm/tee/ffa_private.h
>> @@ -265,6 +265,21 @@
>> #define FFA_ABI_BITNUM(id)    ((FFA_ABI_ID(id) - FFA_ABI_MIN) << 1 | \
>>                                FFA_ABI_CONV(id))
>> 
>> +/* Constituent memory region descriptor */
>> +struct ffa_address_range {
>> +    uint64_t address;
>> +    uint32_t page_count;
>> +    uint32_t reserved;
>> +};
>> +
>> +/* Composite memory region descriptor */
>> +struct ffa_mem_region {
>> +    uint32_t total_page_count;
>> +    uint32_t address_range_count;
>> +    uint64_t reserved;
>> +    struct ffa_address_range address_range_array[];
>> +};
>> +
>> struct ffa_ctx_notif {
>>     bool enabled;
>> 
>> @@ -292,7 +307,7 @@ struct ffa_ctx {
>>     struct ffa_ctx_notif notif;
>>     /*
>>      * tx_lock is used to serialize access to tx
>> -     * rx_lock is used to serialize access to rx
>> +     * rx_lock is used to serialize access to rx_is_free
>>      * lock is used for the rest in this struct
>>      */
>>     spinlock_t tx_lock;
>> @@ -331,7 +346,8 @@ void ffa_rxtx_domain_destroy(struct domain *d);
>> uint32_t ffa_handle_rxtx_map(uint32_t fid, register_t tx_addr,
>>                             register_t rx_addr, uint32_t page_count);
>> uint32_t ffa_handle_rxtx_unmap(void);
>> -int32_t ffa_handle_rx_release(void);
>> +int32_t ffa_rx_acquire(struct domain *d);
>> +int32_t ffa_rx_release(struct domain *d);
>> 
>> void ffa_notif_init(void);
>> void ffa_notif_init_interrupt(void);
>> @@ -420,7 +436,7 @@ static inline int32_t ffa_simple_call(uint32_t fid, 
>> register_t a1,
>>     return ffa_get_ret_code(&resp);
>> }
>> 
>> -static inline int32_t ffa_rx_release(void)
>> +static inline int32_t ffa_hyp_rx_release(void)
>> {
>>     return ffa_simple_call(FFA_RX_RELEASE, 0, 0, 0, 0);
>> }
>> diff --git a/xen/arch/arm/tee/ffa_rxtx.c b/xen/arch/arm/tee/ffa_rxtx.c
>> index b6931c855779..a5d43e51f843 100644
>> --- a/xen/arch/arm/tee/ffa_rxtx.c
>> +++ b/xen/arch/arm/tee/ffa_rxtx.c
>> @@ -30,6 +30,17 @@ struct ffa_endpoint_rxtx_descriptor_1_1 {
>>     uint32_t tx_region_offs;
>> };
>> 
>> +static int32_t ffa_rxtx_map(paddr_t tx_addr, paddr_t rx_addr,
>> +                            uint32_t page_count)
>> +{
>> +    return ffa_simple_call(FFA_RXTX_MAP_64, tx_addr, rx_addr, page_count, 
>> 0);
>> +}
>> +
>> +static int32_t ffa_rxtx_unmap(uint16_t id)
>> +{
>> +    return ffa_simple_call(FFA_RXTX_UNMAP, ((uint64_t)id)<<16, 0, 0, 0);
>> +}
>> +
>> uint32_t ffa_handle_rxtx_map(uint32_t fid, register_t tx_addr,
>>                             register_t rx_addr, uint32_t page_count)
>> {
>> @@ -42,6 +53,9 @@ uint32_t ffa_handle_rxtx_map(uint32_t fid, register_t 
>> tx_addr,
>>     void *rx;
>>     void *tx;
>> 
>> +    /* The code is considering that we only get one page for now */
>> +    BUILD_BUG_ON(FFA_MAX_RXTX_PAGE_COUNT != 1);
>> +
>>     if ( !smccc_is_conv_64(fid) )
>>     {
>>         /*
>> @@ -87,6 +101,65 @@ uint32_t ffa_handle_rxtx_map(uint32_t fid, register_t 
>> tx_addr,
>>     if ( !rx )
>>         goto err_unmap_tx;
>> 
>> +    /*
>> +     * Transmit the RX/TX buffer information to the SPM if acquire is 
>> supported
>> +     * as the spec says that if not there is not need to acquire/release/map
>> +     * rxtx buffers from the SPMC
>> +     */
>> +    if ( ffa_fw_supports_fid(FFA_RX_ACQUIRE) )
>> +    {
>> +        struct ffa_endpoint_rxtx_descriptor_1_1 *rxtx_desc;
>> +        struct ffa_mem_region *mem_reg;
>> +
>> +        /* All must fit in our TX buffer */
>> +        BUILD_BUG_ON((sizeof(*rxtx_desc) + sizeof(*mem_reg)*2 +
>> +                      sizeof(struct ffa_address_range)*2) > FFA_PAGE_SIZE);
> 
> Please add space before and after the binary operator "*".
Ack

> The size of the TX buffer is FFA_RXTX_PAGE_COUNT * FFA_PAGE_SIZE.
Very right, I will fix that in v3.

> Nit: the outer parenthesis around the left expression aren't needed.
Ack

> 
>> +
>> +        spin_lock(&ffa_tx_buffer_lock);
>> +        rxtx_desc = ffa_tx;
>> +
>> +        /*
>> +         * We have only one page for each so we pack everything:
>> +         * - rx region descriptor
>> +         * - rx region range
>> +         * - tx region descriptor
>> +         * - tx region range
>> +         */
>> +        rxtx_desc->sender_id = ffa_get_vm_id(d);
>> +        rxtx_desc->reserved = 0;
>> +        rxtx_desc->rx_region_offs = sizeof(*rxtx_desc);
>> +        rxtx_desc->tx_region_offs = sizeof(*rxtx_desc) +
>> +                                    offsetof(struct ffa_mem_region,
>> +                                             address_range_array[1]);
>> +
>> +        /* rx buffer */
>> +        mem_reg = ffa_tx + sizeof(*rxtx_desc);
>> +        mem_reg->total_page_count = 1;
>> +        mem_reg->address_range_count = 1;
>> +        mem_reg->reserved = 0;
>> +
>> +        mem_reg->address_range_array[0].address = page_to_maddr(rx_pg);
>> +        mem_reg->address_range_array[0].page_count = 1;
>> +        mem_reg->address_range_array[0].reserved = 0;
>> +
>> +        /* tx buffer */
>> +        mem_reg = ffa_tx + rxtx_desc->tx_region_offs;
>> +        mem_reg->total_page_count = 1;
>> +        mem_reg->address_range_count = 1;
>> +        mem_reg->reserved = 0;
>> +
>> +        mem_reg->address_range_array[0].address = page_to_maddr(tx_pg);
>> +        mem_reg->address_range_array[0].page_count = 1;
>> +        mem_reg->address_range_array[0].reserved = 0;
>> +
>> +        ret = ffa_rxtx_map(0, 0, 1);
> 
> The last parameter is also MBZ when forwarding on behalf of an endpoint.

Ack, will fix in v3.

Cheers
Bertrand

> 
> Cheers,
> Jens
> 
>> +
>> +        spin_unlock(&ffa_tx_buffer_lock);
>> +
>> +        if ( ret != FFA_RET_OK )
>> +            goto err_unmap_rx;
>> +    }
>> +
>>     ctx->rx = rx;
>>     ctx->tx = tx;
>>     ctx->rx_pg = rx_pg;
>> @@ -95,6 +168,8 @@ uint32_t ffa_handle_rxtx_map(uint32_t fid, register_t 
>> tx_addr,
>>     ctx->rx_is_free = true;
>>     return FFA_RET_OK;
>> 
>> +err_unmap_rx:
>> +    unmap_domain_page_global(rx);
>> err_unmap_tx:
>>     unmap_domain_page_global(tx);
>> err_put_rx_pg:
>> @@ -105,8 +180,22 @@ err_put_tx_pg:
>>     return ret;
>> }
>> 
>> -static void rxtx_unmap(struct ffa_ctx *ctx)
>> +static uint32_t  rxtx_unmap(struct domain *d)
>> {
>> +    struct ffa_ctx *ctx = d->arch.tee;
>> +
>> +    if ( !ctx->page_count )
>> +        return FFA_RET_INVALID_PARAMETERS;
>> +
>> +    if ( ffa_fw_supports_fid(FFA_RX_ACQUIRE) )
>> +    {
>> +        uint32_t ret;
>> +
>> +        ret = ffa_rxtx_unmap(ffa_get_vm_id(d));
>> +        if ( ret != FFA_RET_OK )
>> +            return ret;
>> +    }
>> +
>>     unmap_domain_page_global(ctx->rx);
>>     unmap_domain_page_global(ctx->tx);
>>     put_page(ctx->rx_pg);
>> @@ -117,32 +206,63 @@ static void rxtx_unmap(struct ffa_ctx *ctx)
>>     ctx->tx_pg = NULL;
>>     ctx->page_count = 0;
>>     ctx->rx_is_free = false;
>> +
>> +    return FFA_RET_OK;
>> }
>> 
>> uint32_t ffa_handle_rxtx_unmap(void)
>> {
>> -    struct domain *d = current->domain;
>> +    return rxtx_unmap(current->domain);
>> +}
>> +
>> +int32_t ffa_rx_acquire(struct domain *d)
>> +{
>> +    int32_t ret = FFA_RET_OK;
>>     struct ffa_ctx *ctx = d->arch.tee;
>> 
>> -    if ( !ctx->rx )
>> -        return FFA_RET_INVALID_PARAMETERS;
>> +    spin_lock(&ctx->rx_lock);
>> 
>> -    rxtx_unmap(ctx);
>> +    if ( !ctx->page_count )
>> +    {
>> +        ret = FFA_RET_DENIED;
>> +        goto out;
>> +    }
>> 
>> -    return FFA_RET_OK;
>> +    if ( !ctx->rx_is_free )
>> +    {
>> +        ret = FFA_RET_BUSY;
>> +        goto out;
>> +    }
>> +
>> +    if ( ffa_fw_supports_fid(FFA_RX_ACQUIRE) )
>> +    {
>> +        ret = ffa_simple_call(FFA_RX_ACQUIRE, ffa_get_vm_id(d), 0, 0, 0);
>> +        if ( ret != FFA_RET_OK )
>> +            goto out;
>> +    }
>> +    ctx->rx_is_free = false;
>> +out:
>> +    spin_unlock(&ctx->rx_lock);
>> +
>> +    return ret;
>> }
>> 
>> -int32_t ffa_handle_rx_release(void)
>> +int32_t ffa_rx_release(struct domain *d)
>> {
>>     int32_t ret = FFA_RET_DENIED;
>> -    struct domain *d = current->domain;
>>     struct ffa_ctx *ctx = d->arch.tee;
>> 
>> -    if ( !spin_trylock(&ctx->rx_lock) )
>> -        return FFA_RET_BUSY;
>> +    spin_lock(&ctx->rx_lock);
>> 
>>     if ( !ctx->page_count || ctx->rx_is_free )
>>         goto out;
>> +
>> +    if ( ffa_fw_supports_fid(FFA_RX_ACQUIRE) )
>> +    {
>> +        ret = ffa_simple_call(FFA_RX_RELEASE, ffa_get_vm_id(d), 0, 0, 0);
>> +        if ( ret != FFA_RET_OK )
>> +            goto out;
>> +    }
>>     ret = FFA_RET_OK;
>>     ctx->rx_is_free = true;
>> out:
>> @@ -151,23 +271,9 @@ out:
>>     return ret;
>> }
>> 
>> -static int32_t ffa_rxtx_map(paddr_t tx_addr, paddr_t rx_addr,
>> -                            uint32_t page_count)
>> -{
>> -    return ffa_simple_call(FFA_RXTX_MAP_64, tx_addr, rx_addr, page_count, 
>> 0);
>> -}
>> -
>> -static int32_t ffa_rxtx_unmap(void)
>> -{
>> -    return ffa_simple_call(FFA_RXTX_UNMAP, 0, 0, 0, 0);
>> -}
>> -
>> void ffa_rxtx_domain_destroy(struct domain *d)
>> {
>> -    struct ffa_ctx *ctx = d->arch.tee;
>> -
>> -    if ( ctx->rx )
>> -        rxtx_unmap(ctx);
>> +    rxtx_unmap(d);
>> }
>> 
>> void ffa_rxtx_destroy(void)
>> @@ -186,7 +292,7 @@ void ffa_rxtx_destroy(void)
>>     }
>> 
>>     if ( need_unmap )
>> -        ffa_rxtx_unmap();
>> +        ffa_rxtx_unmap(0);
>> }
>> 
>> bool ffa_rxtx_init(void)
>> diff --git a/xen/arch/arm/tee/ffa_shm.c b/xen/arch/arm/tee/ffa_shm.c
>> index 29675f9ba3f7..d628c1b70609 100644
>> --- a/xen/arch/arm/tee/ffa_shm.c
>> +++ b/xen/arch/arm/tee/ffa_shm.c
>> @@ -16,21 +16,6 @@
>> 
>> #include "ffa_private.h"
>> 
>> -/* Constituent memory region descriptor */
>> -struct ffa_address_range {
>> -    uint64_t address;
>> -    uint32_t page_count;
>> -    uint32_t reserved;
>> -};
>> -
>> -/* Composite memory region descriptor */
>> -struct ffa_mem_region {
>> -    uint32_t total_page_count;
>> -    uint32_t address_range_count;
>> -    uint64_t reserved;
>> -    struct ffa_address_range address_range_array[];
>> -};
>> -
>> /* Memory access permissions descriptor */
>> struct ffa_mem_access_perm {
>>     uint16_t endpoint_id;
>> --
>> 2.47.0