Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider: - [High] misc: fastrpc: teardown hang when invoke misses fastrpc_notify_users() - [High] misc: fastrpc: fastrpc_rpmsg_remove() fails to drain all context-bank frees --
commit 5c02fd01b12d73075ac0b237229b9d3b84e331dc Author: Jianping Li <[email protected]> misc: fastrpc: drain in-flight invokes before tearing down context banks This commit adds a teardown flag, an in-flight invoke counter, and a wait queue to fastrpc_channel_ctx so that fastrpc_rpmsg_remove() blocks new invocations and waits for active fastrpc_internal_invoke() calls to finish before clearing rpdev and depopulating context bank devices. > diff --git a/drivers/misc/fastrpc.c b/drivers/misc/fastrpc.c > index d201f26d8e463..cffe96a181a2b 100644 > --- a/drivers/misc/fastrpc.c > +++ b/drivers/misc/fastrpc.c [ ... ] > @@ -1365,14 +1385,25 @@ static int fastrpc_internal_invoke(struct > fastrpc_user *fl, u32 kernel, > if (!fl->cctx->rpdev) > return -EPIPE; > > + spin_lock_irqsave(&fl->cctx->lock, flags); > + if (atomic_read(&fl->cctx->teardown)) { > + spin_unlock_irqrestore(&fl->cctx->lock, flags); > + return -EPIPE; > + } > + fastrpc_channel_update_invoke_cnt(fl->cctx, true); > + spin_unlock_irqrestore(&fl->cctx->lock, flags); > + > if (handle == FASTRPC_INIT_HANDLE && !kernel) { > dev_warn_ratelimited(fl->sctx->dev, "user app trying to send a > kernel RPC message (%d)\n", handle); > - return -EPERM; > + err = -EPERM; > + goto out; > } > > ctx = fastrpc_context_alloc(fl, kernel, sc, args); > - if (IS_ERR(ctx)) > - return PTR_ERR(ctx); > + if (IS_ERR(ctx)) { > + err = PTR_ERR(ctx); > + goto out; > + } [Severity: High] Can fastrpc_rpmsg_remove() hang in wait_event() if it runs after fastrpc_internal_invoke() increments cctx->invoke_cnt, but before fastrpc_context_alloc() adds ctx->node to user->pending? In drivers/misc/fastrpc.c, fastrpc_internal_invoke() checks cctx->teardown, increments cctx->invoke_cnt, and drops fl->cctx->lock before calling fastrpc_context_alloc(): drivers/misc/fastrpc.c:fastrpc_context_alloc() { ... ctx = kzalloc_obj(*ctx); ... init_completion(&ctx->work); INIT_WORK(&ctx->put_work, fastrpc_context_put_wq); spin_lock(&user->lock); list_add_tail(&ctx->node, &user->pending); spin_unlock(&user->lock); ... } If fastrpc_rpmsg_remove() runs on another CPU during that window (for example while fastrpc_context_alloc() sleeps in kzalloc_obj()), it sets cctx->teardown to 1, calls fastrpc_notify_users() once across cctx->users, and then blocks in wait_event(cctx->ssr_wait_queue, cctx->invoke_cnt == 0) with cctx->rpdev still valid. Because ctx->node was not on user->pending when fastrpc_notify_users() ran, ctx->work is never completed with -EPIPE. The invoke thread then finishes fastrpc_context_alloc(), runs fastrpc_get_args(), and calls fastrpc_invoke_send() -> rpmsg_send() without re-checking cctx->teardown. When rpmsg_send() succeeds (such as on a remote channel close, driver unbind, module unload, or when a glink rx intent is cached), fastrpc_wait_for_completion() -> fastrpc_wait_for_response() enters wait_for_completion_interruptible(&ctx->work) for user invokes and sleeps waiting for a remote reply that will never arrive, preventing cctx->invoke_cnt from reaching 0 and leaving fastrpc_rpmsg_remove() stuck in wait_event(). For kernel invokes (or signal-interrupted user invokes), fastrpc_wait_for_completion() returns -ETIMEDOUT after 10 seconds (or -ERESTARTSYS), which skips fastrpc_context_put(ctx) at the bail label while the out label decrements cctx->invoke_cnt to 0, allowing fastrpc_rpmsg_remove() to proceed to of_platform_depopulate() with ctx->buf still allocated on the context bank device. > > err = fastrpc_get_args(kernel, ctx); > if (err) > @@ -1428,6 +1459,10 @@ static int fastrpc_internal_invoke(struct fastrpc_user > *fl, u32 kernel, > > if (err) > dev_dbg(fl->sctx->dev, "Error: Invoke Failed %d\n", err); > +out: > + spin_lock_irqsave(&fl->cctx->lock, flags); > + fastrpc_channel_update_invoke_cnt(fl->cctx, false); > + spin_unlock_irqrestore(&fl->cctx->lock, flags); > > return err; > } [ ... ] > @@ -2678,11 +2716,25 @@ static void fastrpc_rpmsg_remove(struct rpmsg_device > *rpdev) > > /* No invocations past this point */ > spin_lock_irqsave(&cctx->lock, flags); > - cctx->rpdev = NULL; > + atomic_set(&cctx->teardown, 1); > list_for_each_entry(user, &cctx->users, user) > fastrpc_notify_users(user); > spin_unlock_irqrestore(&cctx->lock, flags); > > + /* > + * Wait for every invoke that was already past the gate to finish. > + * They have all just been woken with -EPIPE, and no new one can be > + * counted, so this is guaranteed to make progress. > + */ > + spin_lock_irqsave(&cctx->lock, flags); > + while (cctx->invoke_cnt > 0) { > + spin_unlock_irqrestore(&cctx->lock, flags); > + wait_event(cctx->ssr_wait_queue, cctx->invoke_cnt == 0); > + spin_lock_irqsave(&cctx->lock, flags); > + } > + cctx->rpdev = NULL; > + spin_unlock_irqrestore(&cctx->lock, flags); [Severity: High] Does waiting only for cctx->invoke_cnt == 0 in fastrpc_rpmsg_remove() still leave context-bank buffers being freed concurrently with or after of_platform_depopulate(&rpdev->dev), or leaked when woken with -EPIPE? There appear to be three paths in drivers/misc/fastrpc.c where ctx->buf or other fastrpc_buf allocations on fl->sctx->dev outlive cctx->invoke_cnt: 1. In fastrpc_invoke_send(), fastrpc_context_get(ctx) takes a second reference on ctx before calling rpmsg_send(): drivers/misc/fastrpc.c:fastrpc_invoke_send() { ... fastrpc_context_get(ctx); ret = rpmsg_send(cctx->rpdev->ept, (void *)msg, sizeof(*msg)); if (ret) fastrpc_context_put(ctx); ... } If fastrpc_rpmsg_callback() runs just before teardown, it completes ctx->work and calls schedule_work(&ctx->put_work) to drop that second reference via fastrpc_context_put_wq(). The woken invoke thread drops its own reference at the bail label and decrements cctx->invoke_cnt to 0 at the out label, allowing fastrpc_rpmsg_remove() to call of_platform_depopulate(&rpdev->dev) while fastrpc_context_put_wq() is still running fastrpc_context_put() -> fastrpc_context_free() -> fastrpc_buf_free(ctx->buf) -> dma_free_coherent() against the unbound context bank. Conversely, if rpmsg_send() succeeded and fastrpc_notify_users() wakes the waiter with -EPIPE before fastrpc_rpmsg_callback() runs, nothing drops the extra reference taken by fastrpc_invoke_send(). At the bail label, list_del(&ctx->node) unlinks ctx from fl->pending and fastrpc_context_put(ctx) only decrements ctx->refcount from 2 to 1, leaking ctx, ctx->buf, ctx->maps, fl, and cctx. 2. If fastrpc_wait_for_completion() returns -ERESTARTSYS or -ETIMEDOUT, the bail label skips fastrpc_context_put(ctx): drivers/misc/fastrpc.c:fastrpc_internal_invoke() { ... bail: if (err != -ERESTARTSYS && err != -ETIMEDOUT) { /* We are done with this compute context */ spin_lock(&fl->lock); list_del(&ctx->node); spin_unlock(&fl->lock); fastrpc_context_put(ctx); } ... } The out label then decrements cctx->invoke_cnt to 0 while ctx remains on fl->pending to be freed later by fastrpc_device_release() -> fastrpc_user_put() -> fastrpc_user_free() after of_platform_depopulate(). 3. Callers of fastrpc_internal_invoke() allocate or free buffers on fl->sctx->dev outside the cctx->invoke_cnt gate: - fastrpc_init_create_process() and fastrpc_req_mmap() call fastrpc_buf_free() on their error paths after fastrpc_internal_invoke() has already decremented cctx->invoke_cnt to 0 and returned -EPIPE. - fastrpc_device_release() calls fastrpc_release_current_dsp_process(fl) (which exits fastrpc_internal_invoke() and drops cctx->invoke_cnt to 0) and only afterwards calls fastrpc_user_put(fl) -> fastrpc_user_free(), which frees fl->init_mem, fl->pending, fl->maps, and fl->mmaps on fl->sctx->dev without blocking fastrpc_rpmsg_remove(). > + > if (cctx->fdevice) > misc_deregister(&cctx->fdevice->miscdev); > -- Sashiko AI review ยท https://sashiko.dev/#/patchset/[email protected]?part=2
