All of lore.kernel.org
 help / color / mirror / Atom feed
From: sashiko-bot@kernel.org
To: "Yanli Yang" <yanli.yang@bedmex.com>
Cc: devicetree@vger.kernel.org, robh@kernel.org, conor+dt@kernel.org
Subject: Re: [PATCH wireless-next v5 4/4] wifi: aic: Integrate the AIC8800 SDIO driver
Date: Mon, 16 Nov 2026 07:48:40 +0000	[thread overview]
Message-ID: <sashiko-outbox-216466@kernel.org> (raw)
In-Reply-To: <179481380242.4.1693438763870948060@bedmex.com>

Thank you for your contribution! Sashiko AI review found 10 potential
issues to consider:
- [High] wifi: aic: stuck drv_conn_state in connect and NULL sta.ap in
  disconnect
- [High] wifi: aic: uninitialized cfm and OOB access in station info and
  survey
- [High] wifi: aic: wrong spinlock unlock, atomic sleep, and leaks in RX
  reorder
- [High] wifi: aic: off-by-one flow control check and stuck cmd_txstate
  in aicwf_sdio_tx_msg()
- [High] wifi: aic: missing credit update and bounds check in
  aicwf_txq_credit_update()
- [High] wifi: aic: unlocked aicwf_txdatacfm() call from
  aicwf_sdio_host_tx_cfm_handler()
- [Medium] wifi: aic: 32-bit BIT() used on 64-bit avail_idx_map in
  aicwf_main.c
- [Medium] wifi: aic: wrong array index for
  ofst_2g4_ofdm_lowrate_chan_10_13
- [Medium] wifi: aic: uninitialized HE radiotap and 40MHz VHT overflow
  in aicwf_rx_add_rtap_hdr()
- [Low] wifi: aic: missing devicetree binding file entry in MAINTAINERS
--

commit e2756605b7e0074b675225d5cc53b5d095cd0e8e
Author: Zhirun Liu <zhirunliu@aicsemi.com>

wifi: aic: Integrate the AIC8800 SDIO driver

This commit connects the AIC wireless vendor directory to the wireless
Kconfig and Makefile and adds a MAINTAINERS entry for the AIC8800 driver.

> diff --git a/MAINTAINERS b/MAINTAINERS
> index 7336600952c8e..4e327fd62b26a 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -746,6 +746,16 @@ L:	linux-scsi@vger.kernel.org
>  S:	Maintained
>  F:	drivers/scsi/aic7xxx/
>  
> +AIC8800 WIRELESS DRIVER
> +M:	Zhirun Liu <zhirunliu@aicsemi.com>
> +M:	Dijia Xu <dijiaxu@aicsemi.com>
> +M:	Chunqiu Liu <chunqiuliu@aicsemi.com>
> +R:	Liheng Wei <liheng.wei@bedmex.com>
> +R:	Yanli Yang <yanli.yang@bedmex.com>
> +L:	linux-wireless@vger.kernel.org
> +S:	Maintained
> +F:	drivers/net/wireless/aic/

[Severity: Low]
Earlier in the series, commit "dt-bindings: net: wireless: Add AIC8800D80
SDIO Wi-Fi" adds
Documentation/devicetree/bindings/net/wireless/aicsemi,aic8800d80-fmac.yaml.
Should that binding file also be listed under F: here?

[ ... ]

> diff --git a/drivers/net/wireless/Makefile b/drivers/net/wireless/Makefile
> index fa3100ff3bf7b..47b999e8014ca 100644
> --- a/drivers/net/wireless/Makefile
> +++ b/drivers/net/wireless/Makefile
> @@ -4,6 +4,7 @@
>  #
>  
>  obj-$(CONFIG_WLAN_VENDOR_ADMTEK) += admtek/
> +obj-$(CONFIG_WLAN_VENDOR_AIC) += aic/
>  obj-$(CONFIG_WLAN_VENDOR_ATH) += ath/
>  obj-$(CONFIG_WLAN_VENDOR_ATMEL) += atmel/
>  obj-$(CONFIG_WLAN_VENDOR_BROADCOM) += broadcom/

[Severity: High]
In the driver enabled here, can aicwf_cfg80211_connect() leave
aicwf_vif->drv_conn_state stuck in AICWF_DRV_STATUS_CONNECTING or
AICWF_DRV_STATUS_ROAMING, and can aicwf_rx_sm_disconnect_ind() dereference a
NULL aicwf_vif->sta.ap?

drivers/net/wireless/aic/aic8800_fdrv/aicwf_main.c:aicwf_cfg80211_connect() {
	if ((int)atomic_read(&aicwf_vif->drv_conn_state) == (int)AICWF_DRV_STATUS_DISCONNECTING ||
	    (int)atomic_read(&aicwf_vif->drv_conn_state) == (int)AICWF_DRV_STATUS_CONNECTING) {
		dev_err(&wiphy->dev, "driver is disconnecting or connecting ,return it\n");
		return 0;
	}

	if (aicwf_vif->sta.is_roam)
		aicwf_set_conn_state(&aicwf_vif->drv_conn_state, (int)AICWF_DRV_STATUS_ROAMING);
	else
		aicwf_set_conn_state(&aicwf_vif->drv_conn_state, (int)AICWF_DRV_STATUS_CONNECTING);
    ...
	error = aicwf_send_sm_connect_req(aicwf_hw, aicwf_vif, sme, &sm_connect_cfm);
	if (error)
		return error;
}

If the SAE check, aicwf_send_sm_connect_req(), or sm_connect_cfm.status fails
synchronously, drv_conn_state is never reset.  Once stuck in
AICWF_DRV_STATUS_CONNECTING, later scans in aicwf_scan() fail with -EBUSY and
later aicwf_cfg80211_connect() calls return 0 without starting a connection.

Conversely, if aicwf_cfg80211_disconnect() is called while drv_conn_state is
AICWF_DRV_STATUS_CONNECTING (before aicwf_rx_sm_connect_ind() assigns
aicwf_vif->sta.ap), aicwf_rx_sm_disconnect_ind() unconditionally passes
aicwf_vif->sta.ap to aicwf_txq_sta_deinit() and clears
aicwf_vif->sta.ap->valid, causing a NULL pointer dereference:

drivers/net/wireless/aic/aic8800_fdrv/aicwf_msg_rx.c:aicwf_rx_sm_disconnect_ind() {
    ...
	aicwf_txq_sta_deinit(aicwf_hw, aicwf_vif->sta.ap);
	aicwf_txq_tdls_vif_deinit(aicwf_vif);
	aicwf_vif->sta.ap->valid = false;
	aicwf_vif->sta.ap = NULL;
    ...
}

[Severity: High]
Can aicwf_fill_station_info() and aicwf_rx_channel_survey_ind() access
uninitialized stack memory and out-of-bounds array elements?

drivers/net/wireless/aic/aic8800_fdrv/aicwf_main.c:aicwf_fill_station_info() {
    ...
	union aicwf_rate_ctrl_info *rate_info;
	struct mm_get_sta_info_cfm cfm;

	aicwf_send_get_sta_info_req(vif->aicwf_hw, sta->sta_idx, &cfm);
	sinfo->tx_failed = cfm.txfailed;
	rate_info = (union aicwf_rate_ctrl_info *)&cfm.rate_info;
    ...
	switch (rate_info->format_mod_tx) {
	case FORMATMOD_NON_HT:
	case FORMATMOD_NON_HT_DUP_OFDM:
		sinfo->txrate.flags = 0;
		sinfo->txrate.legacy = tx_legrates_lut_rate[rate_info->mcs_index_tx];
		break;
    ...
}

If aicwf_send_get_sta_info_req() fails, cfm is read uninitialized from the
stack, and the 7-bit rate_info->mcs_index_tx field indexes the 12-element
tx_legrates_lut_rate[] array without a bounds check.

In addition, aicwf_rx_channel_survey_ind() checks idx >
ARRAY_SIZE(aicwf_hw->survey) instead of >=, allowing an out-of-bounds write
to aicwf_hw->survey[ARRAY_SIZE(aicwf_hw->survey)]:

drivers/net/wireless/aic/aic8800_fdrv/aicwf_msg_rx.c:aicwf_rx_channel_survey_ind() {
    ...
	if (idx > ARRAY_SIZE(aicwf_hw->survey))
		return 0;

	aicwf_survey = &aicwf_hw->survey[idx];
    ...
}

[Severity: Medium]
Does using BIT() instead of BIT_ULL() on aicwf_hw->avail_idx_map break on
32-bit architectures?

drivers/net/wireless/aic/aic8800_fdrv/aicwf_main.c:aicwf_cfg80211_init() {
    ...
	for (i = 0; i < NX_VIRT_DEV_MAX + NX_REMOTE_STA_MAX; i++)
		aicwf_hw->avail_idx_map |= BIT(i);
    ...
}

avail_idx_map is a u64 holding NX_VIRT_DEV_MAX + NX_REMOTE_STA_MAX (36) bits.
On 32-bit architectures, BIT(i) overflows a 32-bit unsigned long for i >= 32,
and aicwf_hw->avail_idx_map &= ~BIT(vif_idx) in aicwf_interface_add() and
aicwf_virtual_interface_add() zero-extends the 32-bit mask and clears the
upper bits of avail_idx_map.

[Severity: Medium]
Does parsing "ofst_2g4_ofdm_lowrate_chan_10_13" in
aicwf_plat_nvram_set_value_group() and aicwf_plat_nvram_set_value_v3() write
to the wrong element of pwrofst2x_tbl_2g4?

drivers/net/wireless/aic/aic8800_fdrv/aicwf_platform.c:aicwf_plat_nvram_set_value_group() {
    ...
	} else if (!strcmp(command, "ofst_2g4_ofdm_lowrate_chan_1_4")) {
		uc->txpwr_ofst2x.pwrofst2x_tbl_2g4[2][0] = aicwf_atoi(value);
	} else if (!strcmp(command, "ofst_2g4_ofdm_lowrate_chan_5_9")) {
		uc->txpwr_ofst2x.pwrofst2x_tbl_2g4[2][1] = aicwf_atoi(value);
	} else if (!strcmp(command, "ofst_2g4_ofdm_lowrate_chan_10_13")) {
		uc->txpwr_ofst2x.pwrofst2x_tbl_2g4[2][0] = aicwf_atoi(value);
    ...
}

Both functions write to pwrofst2x_tbl_2g4[2][0] instead of
pwrofst2x_tbl_2g4[2][2], overwriting the channel 1-4 offset and leaving
channels 10-13 unconfigured.

[Severity: High]
Does the RX reorder code in aicwf_rx.c and aicwf_msg_rx.c unlock the wrong
spinlock, sleep in atomic context, and leak recv_msdu frames?

drivers/net/wireless/aic/aic8800_fdrv/aicwf_rx.c:reord_process_unit() {
    ...
	spin_unlock_bh(&rx_priv->stas_reord_lock);

	if (!preorder_ctrl->enable) {
		spin_lock_bh(&preorder_ctrl->reord_list_lock);
		preorder_ctrl->ind_sn = pframe->seq_num;
		reord_single_frame_ind(rx_priv, pframe);
		preorder_ctrl->ind_sn = (preorder_ctrl->ind_sn + 1) % 4096;
		spin_unlock_bh(&rx_priv->stas_reord_lock);
		return 0;
	}
    ...
}

When !preorder_ctrl->enable, reord_process_unit() locks
preorder_ctrl->reord_list_lock but unlocks rx_priv->stas_reord_lock a second
time, leaving reord_list_lock held and deadlocking the next acquisition.
Earlier in reord_process_unit(), the unsupported iftype and !reord_info error
returns free skb without returning pframe to rx_priv->rxframes_freequeue via
reord_rxframe_free(), exhausting the 250-entry pool.

In reord_flush_tid(), reord_single_frame_ind() is called under
spin_lock_irqsave(&preorder_ctrl->reord_list_lock, flags), but
reord_single_frame_ind() calls spin_unlock_bh(), local_bh_enable(), and
arpoffload_proc() -> handle_dhcp_ack() -> aicwf_send_arpoffload_en_req(),
which allocates with GFP_KERNEL and waits for firmware command completion.

Finally, reord_deinit_sta() calls
cancel_work_sync(&preorder_ctrl->reord_timer_work) while callers such as
aicwf_cfg80211_del_station(), apm_staloss_work_process(), and
aicwf_rx_deinit() hold spin_lock_bh(&rx_priv->stas_reord_lock), whereas
aicwf_rx_sm_disconnect_ind() has stas_reord_lock commented out when walking
and removing from rx_priv->stas_reord_list.

[Severity: Medium]
Can aicwf_rx_add_rtap_hdr() overflow the radiotap buffer on 40 MHz VHT frames
and copy uninitialized stack data on HE SU frames?

drivers/net/wireless/aic/aic8800_fdrv/aicwf_rx.c:aicwf_rx_add_rtap_hdr() {
    ...
		// bandwidth
		if (rxvect->ch_bw == PHY_CHNL_BW_40)
			*pos++ = 1;
		if (rxvect->ch_bw == PHY_CHNL_BW_80)
			*pos++ = 4;
		else if ((rxvect->ch_bw == PHY_CHNL_BW_160) &&
			 phy_info->phy_center2_freq)
			*pos++ = 0; // 80P80
		else if (rxvect->ch_bw == PHY_CHNL_BW_160)
			*pos++ = 11;
		else // 20 MHz
			*pos++ = 0;
    ...
	if (rxvect->format_mod == FORMATMOD_HE_SU) {
		struct ieee80211_radiotap_he he;
    ...
}

Because the second if is missing an else, a PHY_CHNL_BW_40 VHT frame writes
both *pos++ = 1 and the final else's *pos++ = 0, advancing pos by 2 bytes
instead of 1 and writing 1 byte past rtap_len into the frame payload.  In the
FORMATMOD_HE_SU block, struct ieee80211_radiotap_he he is not zero-initialized
on the stack before |= updates to data3, data5, and data6 (and with data4
never touched) before copying he into the skb.

[Severity: High]
In aicwf_sdio_tx_msg(), does the flow-control check reject transfers when
len == buffer_cnt * BUFFER_SIZE and leave cmd_txstate stuck at true?

drivers/net/wireless/aic/aic8800_fdrv/aicwf_sdio.c:aicwf_sdio_tx_msg() {
    ...
	if (sdiodev->chip_hw->use_flowctrl_msg) {
		buffer_cnt = aicwf_sdio_flow_ctrl_msg(sdiodev);
		while ((buffer_cnt <= 0 ||
			(buffer_cnt > 0 && len > (buffer_cnt * BUFFER_SIZE))) &&
			   retry < 10) {
			retry++;
			buffer_cnt = aicwf_sdio_flow_ctrl_msg(sdiodev);
			dev_dbg(sdiodev->dev, "buffer_cnt = %d\n", buffer_cnt);
		}
	}
	down(&sdiodev->tx_priv->cmd_txsema);

	if (sdiodev->chip_hw->use_flowctrl_msg) {
		if (buffer_cnt > 0 && len < (buffer_cnt * BUFFER_SIZE)) {
			err = aicwf_sdio_send_pkt(sdiodev, payload, len);
			if (err)
				dev_err(sdiodev->dev, "aicwf_sdio_send_pkt fail%d\n", err);
		} else {
			dev_err(sdiodev->dev, "tx msg fc retry fail:%d, %d\n", buffer_cnt, len);
			up(&sdiodev->tx_priv->cmd_txsema);
			return -1;
		}
	}
    ...
	sdiodev->tx_priv->cmd_txstate = false;
}

The retry loop stops waiting when len == buffer_cnt * BUFFER_SIZE (for
example, a 1536-byte firmware write command when buffer_cnt is 1), but the
send condition requires len < (buffer_cnt * BUFFER_SIZE).  When they are
equal, the function takes the else branch and returns -1 without clearing
sdiodev->tx_priv->cmd_txstate, causing aicwf_sdio_tx_process() to repeatedly
resend the command buffer.

[Severity: High]
Does aicwf_txq_credit_update() forget to add update to txq->credits, leaving
stopped TX queues permanently stalled?

drivers/net/wireless/aic/aic8800_fdrv/aicwf_tx.c:aicwf_txq_credit_update() {
	struct aicwf_sta *sta = &aicwf_hw->sta_table[sta_idx];
	struct aicwf_txq *txq;

	txq = aicwf_txq_sta_get(sta, tid, aicwf_hw);

	spin_lock_bh(&aicwf_hw->tx_lock);

	if (txq->idx != TXQ_INACTIVE) {
#ifdef AIC8800_CREATE_TRACE_POINTS
		trace_credit_update(txq, update);
#endif
		if (txq->credits <= 0)
			aicwf_txq_stop(txq, AICWF_TXQ_STOP_FULL);
		else
			aicwf_txq_start(txq, AICWF_TXQ_STOP_FULL);
	}

	spin_unlock_bh(&aicwf_hw->tx_lock);
}

txq->credits is never updated with update before checking txq->credits <= 0,
so any queue that reaches 0 credits and stops with AICWF_TXQ_STOP_FULL will
never be restarted by ME_TX_CREDITS_UPDATE_IND.  In addition, neither
aicwf_rx_me_tx_credits_update_ind() nor aicwf_txq_credit_update() bounds-checks
sta_idx or tid before indexing aicwf_hw->sta_table[sta_idx] and calling
aicwf_txq_sta_get().

[Severity: High]
Does calling aicwf_txdatacfm() from aicwf_sdio_host_tx_cfm_handler() without
aicwf_hw->tx_lock race with the TX path?

sdio_busrx_thread                 TX / netdev context
aicwf_process_rxframes()          aicwf_start_xmit() / aicwf_hwq_process_all()
  aicwf_sdio_host_tx_cfm_handler()  spin_lock_bh(&aicwf_hw->tx_lock)
    aicwf_txdatacfm()               aicwf_txq_queue_skb() / aicwf_hwq_process()
      aicwf_txq_confirm_any()         /* mutates txq->sk_list, hwq->list,
      aicwf_tx_retry()                   txq->status, and credits */
      aicwf_txq_add_to_hw_list()    spin_unlock_bh(&aicwf_hw->tx_lock)
        <-- unlocked list and credit mutation

aicwf_txdatacfm() and its helpers (aicwf_txq_confirm_any(), aicwf_tx_retry(),
and aicwf_txq_add_to_hw_list()) require aicwf_hw->tx_lock to serialize access
to txq->sk_list, hwq->list, txq->status, and queue credits, but
aicwf_sdio_host_tx_cfm_handler() invokes aicwf_txdatacfm() without acquiring
aicwf_hw->tx_lock.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/179481380235.4.7388336477392853157@bedmex.com?part=4

  reply	other threads:[~2026-11-16  7:48 UTC|newest]

Thread overview: 12+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-16  7:23 [PATCH wireless-next v5 0/4] wifi: aic: Add AIC8800D80 and D80N SDIO FullMAC driver Yanli Yang
2026-11-16  7:23 ` [PATCH wireless-next v5 2/4] wifi: aic: Add firmware protocol and SDIO chip support Yanli Yang
2026-11-16 11:22   ` Krzysztof Kozlowski
2026-11-16  7:23 ` [PATCH wireless-next v5 3/4] wifi: aic: Add the AIC8800D80 and D80N FullMAC driver Yanli Yang
2026-11-16 11:24   ` Krzysztof Kozlowski
2026-11-16  7:23 ` [PATCH wireless-next v5 1/4] dt-bindings: net: wireless: Add AIC8800D80 SDIO Wi-Fi Yanli Yang
2026-11-16  7:35   ` sashiko-bot
2026-11-16 11:08   ` Krzysztof Kozlowski
2026-11-16 11:12   ` Krzysztof Kozlowski
2026-11-16  7:23 ` [PATCH wireless-next v5 4/4] wifi: aic: Integrate the AIC8800 SDIO driver Yanli Yang
2026-11-16  7:48   ` sashiko-bot [this message]
2026-11-16 11:10   ` Krzysztof Kozlowski

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=sashiko-outbox-216466@kernel.org \
    --to=sashiko-bot@kernel.org \
    --cc=conor+dt@kernel.org \
    --cc=devicetree@vger.kernel.org \
    --cc=robh@kernel.org \
    --cc=sashiko-reviews@lists.linux.dev \
    --cc=yanli.yang@bedmex.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.