From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from lists1p.gnu.org (lists1p.gnu.org [209.51.188.17]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 4296ECC6B27 for ; Wed, 14 Oct 2026 08:24:38 +0000 (UTC) Received: from localhost ([::1] helo=lists1p.gnu.org) by lists1p.gnu.org with esmtp (Exim 4.90_1) (envelope-from ) id 1xGuHb-0000Ae-C9; Wed, 14 Oct 2026 04:24:20 -0400 Received: from eggs.gnu.org ([2001:470:142:3::10]) by lists1p.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1xGuHV-0000A2-KC for qemu-arm@nongnu.org; Wed, 14 Oct 2026 04:24:14 -0400 Received: from mail-ed1-x52c.google.com ([2a00:1450:4864:20::52c]) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_128_GCM_SHA256:128) (Exim 4.90_1) (envelope-from ) id 1xGuHR-0002EY-2R for qemu-arm@nongnu.org; Wed, 14 Oct 2026 04:24:13 -0400 Received: by mail-ed1-x52c.google.com with SMTP id 4fb4d7f45d1cf-6b1aa27963eso4044688a12.1 for ; Wed, 14 Oct 2026 01:24:08 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20251104; t=1791966247; x=1792571047; darn=nongnu.org; h=content-transfer-encoding:content-type:in-reply-to:content-language :references:cc:to:subject:user-agent:mime-version:date:message-id :from:from:to:cc:subject:date:message-id:reply-to:content-type; bh=E77zUabWfnWHHLgTNO0YwCzccziQx5/PzVOktJEn4XU=; b=aypEhdxYbi/UpDHGwN9ain4lrQh98G/V6ceCUzKd4YbtHqmFDTV2a+WcKA/k6HFtRt BvbwsgeaJVrTz2wkEw5D5yQJqPLXmLwtoYrGcPjgZBL0EKxK9/e1iR3PyNSmulfRQ923 dKoKuaL7/ZqOt4aEdP1sPW3RbNVa4t1+hlS9j6ncckSh7zrusoGI/UEsXnRgt5PSnqjM 94kb/OFYss2FSHKfV//L5nqjusenw3WdhSG6/FRHcIz4KHnKai4+/lk4RabG4LNgiMpB 85+MR0kiKrjpTiNsTOKj/rjZZbVcP4ngLuDchrPEsgSYGItChM5cVAw0c2bJcFalfeIL k4sA== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20260707; t=1791966247; x=1792571047; h=content-transfer-encoding:content-type:in-reply-to:content-language :references:cc:to:subject:user-agent:mime-version:date:message-id :from:x-gm-gg:x-gm-message-state:from:to:cc:subject:date:message-id :reply-to:content-type; bh=E77zUabWfnWHHLgTNO0YwCzccziQx5/PzVOktJEn4XU=; b=UnWFZjY+e9YOwsVRyrz9PMUSQucqw7CmYSRYcaauiRgLD44oL+W9GP+niwOsbfYEOt Mx4geZU1539sA218y80tmqVSylUfCuGXEcOSBXrC9gK2+ynZ+MUjq6DGy/XknsGUtKvn BRTkxG8JTfqLReqQQoykPl1LDBXJyvwB7Q5EeTVxhMk0jr8Kk+qSCEQd5OBuc+b6pY0Q Qhoo0YfRIsLTiflrZNRiL4jO9HxINNWhf7q+IAnpVWmn/T3syGSuXSqbl/aMHF1Aqki6 s4H5zrAgfLVnPg3EfLJArEYykTv0M2zuqET78yotrkIbgeCQbb0x71fT1SBvkaXMDNLO i1zw== X-Gm-Message-State: AFq9FYKJUed79YWiyuLGvzySXsXv9umo3HzpRKV7atkcL/W4RiimS0zm uEH6AAjFxFmJaK6LMCJGy0CyZTbHBuHeR7+p4pzbB80/i2SB3T5EB8Hp X-Gm-Gg: AYBFou3Xc/wK5t2pVUgxDKWhVd4RPX1IljWY17gkvtEIAVU44aVK5BmWiBWeNnygSmU DbTfT0hW7N7YcYyqUDI/abCe/WATn5Sva3b7+Eg8PN8f+3ge16EW/wIzNa2iIvI2CR49b0p4QEZ ckfZvintXc6XRHqjkxbTArvfASwNSJDbhk9eNSBBb3x7xW8XXa9JxLgY0khHmnDhYHiMVwPF2FH Ew5JH3L1vbhMFJqBbnEMj1djUIto6wUy0dJmTzfYjcAGc6jfLV7aYlVlTup6vBAgsTLRBG3AV3Y Non0AaJeonNlZA3ukPOqDyUmYE4oN/bRXfcl48C/YMgvI8NnJiM90CzQ9R9cy9LSlyQQjPmwJPg iPVYkxtlJhzziwIu7cVn8JSxgEkr/1HnWmDMu1cWXOky1Hyy/QdqztPjMMCnPrL3HcChgisntjp LJAuly+oohUwOxayeTannMWbLcW+isOOjD4dJm3VLAZaqvA8Xhgx+LfwK3l1nTzDeyVuD5P98/t CeZx3JyedwLRb4Sep+ZmfcmKFgVmYBrGq40+1y9VA== X-Received: by 2002:a17:906:730f:b0:c2d:d4dd:8b5 with SMTP id a640c23a62f3a-c31a9fc0b88mr1422037866b.19.1791966247252; Wed, 14 Oct 2026 01:24:07 -0700 (PDT) Received: from [192.168.1.130] (93-143-204-231.adsl.net.t-com.hr. [93.143.204.231]) by smtp.gmail.com with ESMTPSA id a640c23a62f3a-c31b8c4b407sm635010866b.0.2026.10.14.01.24.05 (version=TLS1_3 cipher=TLS_AES_128_GCM_SHA256 bits=128/128); Wed, 14 Oct 2026 01:24:06 -0700 (PDT) From: Ruslan Ruslichenko X-Google-Original-From: Ruslan Ruslichenko Message-ID: Date: Wed, 14 Oct 2026 10:24:05 +0200 MIME-Version: 1.0 User-Agent: Mozilla Thunderbird Subject: Re: [PATCH v2 11/20] hw/core: Implement Remote Port Memory Master bus transactions To: =?UTF-8?Q?Philippe_Mathieu-Daud=C3=A9?= , Ruslan Ruslichenko , qemu-devel@nongnu.org Cc: qemu-arm@nongnu.org, artem_mygaiev@epam.com, volodymyr_babchuk@epam.com, peter.maydell@linaro.org, francisco.iglesias@amd.com, edgar.iglesias@gmail.com, berrange@redhat.com, eugene@nuvacore.ai, Ruslan_Ruslichenko@epam.com, takahiro.nakata.wr@renesas.com, "Edgar E . Iglesias" , Paolo Bonzini , =?UTF-8?Q?Alex_Benn=C3=A9e?= References: <20260915160044.345788-1-ruslichenko.r@gmail.com> <20260915160044.345788-12-ruslichenko.r@gmail.com> <8f0cfc4a-060c-41a9-81e4-111e0019392f@oss.qualcomm.com> <981c4de8-2b09-49df-bcfa-c1f4fdbf3cfb@oss.qualcomm.com> Content-Language: en-US In-Reply-To: <981c4de8-2b09-49df-bcfa-c1f4fdbf3cfb@oss.qualcomm.com> Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Received-SPF: pass client-ip=2a00:1450:4864:20::52c; envelope-from=ruslichenko.r@gmail.com; helo=mail-ed1-x52c.google.com X-Spam_score_int: -20 X-Spam_score: -2.1 X-Spam_bar: -- X-Spam_report: (-2.1 / 5.0 requ) BAYES_00=-1.9, DKIM_SIGNED=0.1, DKIM_VALID=-0.1, DKIM_VALID_AU=-0.1, DKIM_VALID_EF=-0.1, FREEMAIL_FROM=0.001, RCVD_IN_DNSWL_NONE=-0.0001, SPF_HELO_NONE=0.001, SPF_PASS=-0.001 autolearn=ham autolearn_force=no X-Spam_action: no action X-BeenThere: qemu-arm@nongnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: qemu-arm-bounces+qemu-arm=archiver.kernel.org@nongnu.org Sender: qemu-arm-bounces+qemu-arm=archiver.kernel.org@nongnu.org Hi Philippe, On 10/13/26 18:10, Philippe Mathieu-Daudé wrote: > On 2026-10-13 18:09, Philippe Mathieu-Daudé wrote: >> Hi, >> >> On 2026-09-15 18:00, Ruslan Ruslichenko wrote: >>> From: Ruslan Ruslichenko >>> >>> Implement the core access handlers for the Remote Port Memory >>> Master device, which enables QEMU to initiate memory-mapped >>> transaction to remote peers. >>> >>> The patch adds support for: >>> - Translation of QEMU memory region read/write into Remote Port >>> read/write packets >>> - Support for propagating of transaction attributes >>> (Secure, MasterID) >>> - Synchronous response waiting with optional timeout (timeout >>>    configuration to be added within later patches) >>> - Add trace events for monitoring TX/RX bus transactions >>> >>> Signed-off-by: Edgar E. Iglesias >>> Signed-off-by: Takahiro Nakata >>> Signed-off-by: Ruslan Ruslichenko >>> --- >>>   hw/core/remote-port-memory-master.c | 191 +++++++++++++++++++++++++++- >>>   hw/core/remote-port.c               |   5 + >>>   hw/core/trace-events                |   4 + >>>   include/hw/core/remote-port.h       |  11 ++ >>>   4 files changed, 207 insertions(+), 4 deletions(-) >>> >>> diff --git a/hw/core/remote-port-memory-master.c b/hw/core/remote- port-memory-master.c >>> index 4b0283f9fe..41ef1df518 100644 >>> --- a/hw/core/remote-port-memory-master.c >>> +++ b/hw/core/remote-port-memory-master.c >>> @@ -29,18 +29,201 @@ >>>   #define RP_MAX_ACCESS_SIZE 4096 >>> +static int rp_mm_get_timeout(RPMemoryTransaction *tr) >>> +{ >>> +    RemotePortMap *map = tr->opaque; >>> +    RemotePortMemoryMaster *s; >>> + >>> +    if (!map || !map->parent || >>> +        !object_dynamic_cast(OBJECT(map->parent), >>> +                             TYPE_REMOTE_PORT_MEMORY_MASTER)) { >>> +        return 0; >>> +    } >>> +    s = REMOTE_PORT_MEMORY_MASTER(map->parent); >>> +    return s->rp_timeout; >>> +} >>> + >>> +MemTxResult rp_mm_access_with_def_attr(RemotePort *rp, uint32_t rp_dev, >>> +                                       struct rp_peer_state *peer, >>> +                                       RPMemoryTransaction *tr, >>> +                                       bool relative, uint64_t offset, >>> +                                       uint32_t def_attr) >>> +{ >>> +    uint64_t addr = tr->addr; >>> +    RemotePortRespSlot *rsp_slot; >>> +    RemotePortDynPkt *rsp; >>> +    struct  { >>> +        struct rp_pkt_busaccess_ext_base pkt; >>> +        uint8_t reserved[RP_MAX_ACCESS_SIZE]; >>> +    } pay; >>> +    uint8_t *data = rp_busaccess_tx_dataptr(peer, &pay.pkt); >>> +    struct rp_encode_busaccess_in in = {0}; >>> +    int i; >>> +    int len; >>> +    int rp_timeout = rp_mm_get_timeout(tr); >>> +    MemTxResult ret; >>> + >>> +    if (tr->rw) { >>> +        /* Data up to 8 bytes is passed as values.  */ >>> +        if (tr->size <= 8) { >>> +            for (i = 0; i < tr->size; i++) { >>> +                data[i] = tr->data.u64 >> (i * 8); >>> +            } >>> +        } else { >>> +            memcpy(data, tr->data.p8, tr->size); >>> +        } >>> +    } >>> + >>> +    addr += relative ? 0 : offset; >>> + >>> +    in.cmd = tr->rw ? RP_CMD_write : RP_CMD_read; >>> +    in.id = rp_new_id(rp); >>> +    in.dev = rp_dev; >>> +    in.clk = rp_normalized_vmclk(rp); >>> +    in.master_id = tr->attr.requester_id; >>> +    in.addr = addr; >>> +    in.attr = def_attr; >>> +    in.attr |= tr->attr.secure ? RP_BUS_ATTR_SECURE : 0; >>> +    in.size = tr->size; >>> +    in.stream_width = tr->size; >>> +    len = rp_encode_busaccess(peer, &pay.pkt, &in); >>> +    len += tr->rw ? tr->size : 0; >>> + >>> +    trace_remote_port_memory_master_tx_busaccess(rp_cmd_to_string(in.cmd), >>> +        in.id, in.flags, in.dev, in.addr, in.size, in.attr); >>> + >>> +    rp_rsp_mutex_lock(rp); >>> +    rp_write(rp, (void *) &pay, len); >>> + >>> +    if (!rp_timeout) { >>> +        rsp_slot = rp_dev_wait_resp(rp, in.dev, in.id); >>> +    } else { >>> +        rsp_slot = rp_dev_timed_wait_resp(rp, in.dev, in.id, rp_timeout); >>> +        if (rsp_slot->valid == false) { >>> +            /* >>> +             * Timeout error >>> +             */ >>> +            rp_rsp_mutex_unlock(rp); >>> +            return MEMTX_ERROR; >>> +        } >>> +    } >>> +    rsp = &rsp_slot->rsp; >>> + >>> +    /* We dont support out of order answers yet.  */ >>> +    assert(rsp->pkt->hdr.id == in.id); >>> + >>> +    switch (rp_get_busaccess_response(rsp->pkt)) { >>> +    case RP_RESP_OK: >>> +        ret = MEMTX_OK; >>> +        break; >>> +    case RP_RESP_ADDR_ERROR: >>> +        ret = MEMTX_DECODE_ERROR; >>> +        break; >>> +    default: >>> +        ret = MEMTX_ERROR; >>> +        break; >>> +    } >>> + >>> +    if (ret == MEMTX_OK && !tr->rw) { >>> +        data = rp_busaccess_rx_dataptr(peer, &rsp->pkt- >busaccess_ext_base); >>> +        /* Data up to 8 bytes is return as values.  */ >>> +        if (tr->size <= 8) { >>> +            for (i = 0; i < tr->size; i++) { >>> +                tr->data.u64 |= ((uint64_t) data[i]) << (i * 8); >>> +            } >>> +        } else { >>> +            memcpy(tr->data.p8, data, tr->size); >> >> I don't understand this direct memory access. >> >> Shouldn't this be handled by the FlatView and end as a subpage_t >> leaf instead? >> Thank you for the catch! In fact this is a leftover of handling MemoryTransaction's: https://lore.kernel.org/qemu-devel/20260205195824.2610192-14-ruslichenko.r@gmail.com/, which were used for RemotePort memory access in v1 patch series. In Xilinx tree, large access sizes are possible thanks to downstream patch: https://github.com/Xilinx/qemu/commit/063b379d47856e1e0f71f93974157b76d51a3106, which is not present in upstream. For Remote Port v2 series, switched to standard (read|write)_with_attrs to simplify things for now, so those need to be cleaned-up. -- BR, Ruslan >> Doing some archeology to understand: >> https://lore.kernel.org/qemu-devel/20170204115048.GM12995@toto/ > > I meant v3: > https://lore.kernel.org/qemu-devel/1493391576-29401-1-git-send-email-fred.konrad@greensocs.com/ > >> >> Amusingly a related backtrace: >> https://lore.kernel.org/qemu-devel/CAKmqyKOqbJ+-t0SeyZDzDM9RLEQX63WM3- wyJTbiMGHRKtHPYQ@mail.gmail.com/ >> >>> +        } >>> +    } >>> + >>> +    trace_remote_port_memory_master_rx_busaccess( >>> +        rp_cmd_to_string(rsp->pkt->hdr.cmd), rsp->pkt->hdr.id, >>> +        rsp->pkt->hdr.flags, rsp->pkt->hdr.dev, rsp->pkt- >busaccess.addr, >>> +        rsp->pkt->busaccess.len, rsp->pkt->busaccess.attributes); >>> + >>> +    if (rp_timeout) { >>> +        for (i = 0; i < ARRAY_SIZE(rp->dev_state[rp_dev].rsp_queue); i++) { >>> +            if (rp->dev_state[rp_dev].rsp_queue[i].used && >>> +                rp->dev_state[rp_dev].rsp_queue[i].valid) { >>> +                rp_resp_slot_done(rp, &rp- >dev_state[rp_dev].rsp_queue[i]); >>> +            } >>> +        } >>> +    } else { >>> +        rp_resp_slot_done(rp, rsp_slot); >>> +    } >>> +    rp_rsp_mutex_unlock(rp); >>> + >>> +    /* >>> +     * For strongly ordered or transactions that don't allow Early Acking, >>> +     * we need to drain the pending RP processing queue here. This is >>> +     * because RP handles responses in parallel with normal requests so >>> +     * they may get reordered. This becomes visible for example with reads >>> +     * to read-to-clear registers that clear interrupts. Even though the >>> +     * lowering of the interrupt-wires arrives to us before the read- resp, >>> +     * we may handle the response before the wire update, resulting in >>> +     * spurious interrupts. >>> +     * >>> +     * This has some room for optimization but for now we use the big hammer >>> +     * and drain the entire qeueue. >>> +     */ >>> +    rp_process(rp); >>> + >>> +    /* Reads are sync-points, roll the sync timer.  */ >>> +    rp_restart_sync_timer(rp); >>> +    return ret; >>> +} >>> + >>> +MemTxResult rp_mm_access(RemotePort *rp, uint32_t rp_dev, >>> +                         struct rp_peer_state *peer, >>> +                         RPMemoryTransaction *tr, >>> +                         bool relative, uint64_t offset) >>> +{ >>> +    return rp_mm_access_with_def_attr(rp, rp_dev, peer, tr, relative, offset, >>> +                                      0); >>> +} >>> + >>>   static MemTxResult rp_mm_read(void *opaque, hwaddr addr, uint64_t *data, >>>                                    unsigned size, MemTxAttrs attrs) >>>   { >>> -    /* TBD */ >>> -    return MEMTX_OK; >>> +    MemTxResult ret; >>> +    RemotePortMap *map = opaque; >>> +    RemotePortMemoryMaster *s = map->parent; >>> + >>> +    RPMemoryTransaction tr = {}; >>> + >>> +    tr.opaque = map; >>> +    tr.rw = false; >>> +    tr.addr = addr; >>> +    tr.size = size; >>> +    tr.attr = attrs; >>> + >>> +    ret = rp_mm_access(s->rp, s->rp_dev, s->peer, &tr, s->relative, >>> +                map->offset); >>> + >>> +    if (ret == MEMTX_OK) { >>> +        *data = tr.data.u64; >>> +    } >>> + >>> +    return ret; >>>   } >>>   static MemTxResult rp_mm_write(void *opaque, hwaddr addr, uint64_t data, >>>                                     unsigned size, MemTxAttrs attrs) >>>   { >>> -    /* TBD */ >>> -    return MEMTX_OK; >>> +    RemotePortMap *map = opaque; >>> +    RemotePortMemoryMaster *s = map->parent; >>> + >>> +    RPMemoryTransaction tr = {}; >>> + >>> +    tr.opaque = map; >>> +    tr.rw = true; >>> +    tr.addr = addr; >>> +    tr.size = size; >>> +    tr.attr = attrs; >>> +    tr.data.u64 = data; >>> + >>> +    return rp_mm_access(s->rp, s->rp_dev, s->peer, &tr, s->relative, >>> +                    map->offset); >>>   } >>>   static const MemoryRegionOps rp_ops_template = { >>> diff --git a/hw/core/remote-port.c b/hw/core/remote-port.c >>> index b4c9fa1528..6a6effdd19 100644 >>> --- a/hw/core/remote-port.c >>> +++ b/hw/core/remote-port.c >>> @@ -39,6 +39,11 @@ static void rp_event_read_and_process(RemotePort *s); >>>   static void sync_timer_hit(void *opaque); >>>   static void syncresp_timer_hit(void *opaque); >>> +uint32_t rp_new_id(RemotePort *s) >>> +{ >>> +    return qatomic_fetch_inc(&s->current_id); >>> +} >