Re: [PATCH v2 1/2] optee: add page list to kernel private shared memory

Hi Sumit,

On Fri, Nov 10, 2023 at 3:36 PM Sumit Garg sumit.garg@linaro.org wrote:

Hi Jens,

On Wed, 8 Nov 2023 at 15:57, Jens Wiklander jens.wiklander@linaro.org wrote:

...

Until now has kernel private shared memory allocated as dynamic shared memory (not from the static shared memory pool) been returned without a list of physical pages on allocations via RPC. To support allocations larger than one page add a list of physical pages for all allocations larger than one page.

Signed-off-by: Jens Wiklander jens.wiklander@linaro.org

drivers/tee/optee/core.c | 40 +++++++++++++++++++++++++-------- drivers/tee/optee/smc_abi.c | 45 +++++++++++++++++-------------------- 2 files changed, 52 insertions(+), 33 deletions(-)

diff --git a/drivers/tee/optee/core.c b/drivers/tee/optee/core.c index 2a258bd3b6b5..a425eca0173d 100644 --- a/drivers/tee/optee/core.c +++ b/drivers/tee/optee/core.c @@ -27,6 +27,7 @@ int optee_pool_op_alloc_helper(struct tee_shm_pool *pool, struct tee_shm *shm, unsigned long start)) { unsigned int order = get_order(size);

•   unsigned int nr_pages = 1 << order;
    struct page *page;
    int rc = 0;
  
  

@@ -39,12 +40,24 @@ int optee_pool_op_alloc_helper(struct tee_shm_pool *pool, struct tee_shm *shm, return -ENOMEM;

    shm->kaddr = page_address(page);

•   shm->paddr = page_to_phys(page);
    shm->size = PAGE_SIZE << order;
  
  

•   if (shm_register) {
  

•           unsigned int nr_pages = 1 << order, i;
  

•   /*
  

•    * If memory is registered immediately use a temporary page list
  

•    * and free it directly after the registration is completed.
  

•    *
  

•    * Else, if we have allocated more than one page store a page
  

•    * list in the struct tee_shm.
  

•    *
  

•    * Else, store the physical address of the single page that was
  

•    * allocated.
  

•    *
  

•    * optee_pool_op_free_helper() is responsible for freeing the
  

•    * resourses allocated in this function.
  

•    */
  

It looks a bit complex to me. Do we really need it to be conditional? I was thinking if we can simplify this API as follows:

int optee_pool_op_alloc_helper(struct tee_shm_pool *pool, struct tee_shm *shm, size_t size, size_t align, int (*shm_register)(struct tee_shm *shm)) { unsigned int order = get_order(size); unsigned int nr_pages = 1 << order, i; struct page **pages; struct page *page; int rc = 0;

    /*
     * Ignore alignment since this is already going to be page aligned
     * and there's no need for any larger alignment.
     */
    page = alloc_pages(GFP_KERNEL | __GFP_ZERO, order);
    if (!page)
            return -ENOMEM;

    shm->kaddr = page_address(page);
    shm->paddr = page_to_phys(page);
    shm->size = PAGE_SIZE << order;

    pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
    if (!pages) {
            rc = -ENOMEM;
            goto err;
    }

    for (i = 0; i < nr_pages; i++)
            pages[i] = page + i;

    shm->pages = pages;
    shm->num_pages = nr_pages;

    if (shm_register) {
            rc = shm_register(shm);
            if (rc)
                    goto err;
    }

    return 0;

err: kfree(pages); free_pages((unsigned long)shm->kaddr, order); return rc; }

Good idea, I'll try it in the v3.

Thanks, Jens

-Sumit

...

•   if (shm_register || nr_pages > 1) {
            struct page **pages;
  

•           unsigned int i;
  
            pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
            if (!pages) {
  

@@ -55,17 +68,24 @@ int optee_pool_op_alloc_helper(struct tee_shm_pool *pool, struct tee_shm *shm, for (i = 0; i < nr_pages; i++) pages[i] = page + i;

•           rc = shm_register(shm->ctx, shm, pages, nr_pages,
  

•                             (unsigned long)shm->kaddr);
  

•           kfree(pages);
  

•           if (rc)
  

•                   goto err;
  

•           if (shm_register) {
  

•                   rc = shm_register(shm->ctx, shm, pages, nr_pages,
  

•                                     (unsigned long)shm->kaddr);
  

•                   kfree(pages);
  

•                   if (rc)
  

•                           goto err;
  

•           } else {
  

•                   shm->pages = pages;
  

•                   shm->num_pages = nr_pages;
  

•           }
  

•   } else {
  

•           shm->paddr = page_to_phys(page);
    }
  
    return 0;
  

err: free_pages((unsigned long)shm->kaddr, order);

•   shm->kaddr = NULL;
    return rc;
  

}

@@ -77,6 +97,8 @@ void optee_pool_op_free_helper(struct tee_shm_pool *pool, struct tee_shm *shm, shm_unregister(shm->ctx, shm); free_pages((unsigned long)shm->kaddr, get_order(shm->size)); shm->kaddr = NULL;

•   kfree(shm->pages);
  

•   shm->pages = NULL;
  

}

static void optee_bus_scan(struct work_struct *work) diff --git a/drivers/tee/optee/smc_abi.c b/drivers/tee/optee/smc_abi.c index d5b28fd35d66..b69410c7cc0a 100644 --- a/drivers/tee/optee/smc_abi.c +++ b/drivers/tee/optee/smc_abi.c @@ -678,10 +678,11 @@ static void handle_rpc_func_cmd_shm_alloc(struct tee_context *ctx, struct optee_msg_arg *arg, struct optee_call_ctx *call_ctx) {

•   phys_addr_t pa;
    struct tee_shm *shm;
    size_t sz;
    size_t n;
  

•   struct page **pages;
  

•   size_t page_count;
  
    arg->ret_origin = TEEC_ORIGIN_COMMS;
  
  

@@ -716,32 +717,23 @@ static void handle_rpc_func_cmd_shm_alloc(struct tee_context *ctx, return; }

•   if (tee_shm_get_pa(shm, 0, &pa)) {
  

•           arg->ret = TEEC_ERROR_BAD_PARAMETERS;
  

•           goto bad;
  

•   }
  

•   sz = tee_shm_get_size(shm);
  

•   if (tee_shm_is_dynamic(shm)) {
  

•           struct page **pages;
  

•   /*
  

•    * If there are pages it's dynamically allocated shared memory (not
  

•    * from the reserved shared memory pool) and needs to be
  

•    * registered.
  

•    */
  

•   pages = tee_shm_get_pages(shm, &page_count);
  

•   if (pages) {
            u64 *pages_list;
  

•           size_t page_num;
  

•           pages = tee_shm_get_pages(shm, &page_num);
  

•           if (!pages || !page_num) {
  

•                   arg->ret = TEEC_ERROR_OUT_OF_MEMORY;
  

•                   goto bad;
  

•           }
  
  

•           pages_list = optee_allocate_pages_list(page_num);
  

•           pages_list = optee_allocate_pages_list(page_count);
            if (!pages_list) {
                    arg->ret = TEEC_ERROR_OUT_OF_MEMORY;
                    goto bad;
            }
  
            call_ctx->pages_list = pages_list;
  

•           call_ctx->num_entries = page_num;
  

•           call_ctx->num_entries = page_count;
  
            arg->params[0].attr = OPTEE_MSG_ATTR_TYPE_TMEM_OUTPUT |
                                  OPTEE_MSG_ATTR_NONCONTIG;
  

@@ -752,17 +744,22 @@ static void handle_rpc_func_cmd_shm_alloc(struct tee_context *ctx, arg->params[0].u.tmem.buf_ptr = virt_to_phys(pages_list) | (tee_shm_get_page_offset(shm) & (OPTEE_MSG_NONCONTIG_PAGE_SIZE - 1));

•           arg->params[0].u.tmem.size = tee_shm_get_size(shm);
  

•           arg->params[0].u.tmem.shm_ref = (unsigned long)shm;
  
  

•           optee_fill_pages_list(pages_list, pages, page_num,
  

•           optee_fill_pages_list(pages_list, pages, page_count,
                                  tee_shm_get_page_offset(shm));
    } else {
  

•           phys_addr_t pa;
  

•           if (tee_shm_get_pa(shm, 0, &pa)) {
  

•                   arg->ret = TEEC_ERROR_BAD_PARAMETERS;
  

•                   goto bad;
  

•           }
  

•           arg->params[0].attr = OPTEE_MSG_ATTR_TYPE_TMEM_OUTPUT;
            arg->params[0].u.tmem.buf_ptr = pa;
  

•           arg->params[0].u.tmem.size = sz;
  

•           arg->params[0].u.tmem.shm_ref = (unsigned long)shm;
    }
  

•   arg->params[0].u.tmem.size = tee_shm_get_size(shm);
  

•   arg->params[0].u.tmem.shm_ref = (unsigned long)shm;
  
    arg->ret = TEEC_SUCCESS;
    return;
  

-- 2.34.1