Netdev List
 help / color / mirror / Atom feed
From: Vishnu Santhosh <vishnu.santhosh@oss.qualcomm.com>
To: netdev-bot+sashiko@kernel.org
Cc: stephan@gerhold.net, andrew+netdev@lunn.ch, davem@davemloft.net,
	kuba@kernel.org, pabeni@redhat.com, robh@kernel.org,
	krzk+dt@kernel.org, conor+dt@kernel.org,
	loic.poulain@oss.qualcomm.com, ryazanov.s.a@gmail.com,
	johannes@sipsolutions.net, edumazet@kernel.org,
	linux-arm-msm@vger.kernel.org, netdev@vger.kernel.org,
	devicetree@vger.kernel.org, linux-kernel@vger.kernel.org,
	deepak.singh@oss.qualcomm.com
Subject: Re: [PATCH net-next v6 2/2] net: wwan: qcom_bam_dmux: Assign restricted DMA pool to remote processor
Date: Wed, 18 Nov 2026 11:18:23 +0530	[thread overview]
Message-ID: <12a0edcc-2c99-4f68-9ebd-0a9ec76bece7@oss.qualcomm.com> (raw)
In-Reply-To: <179483727219.3136.17411591489340995392@kernel.org>


On 16-11-2026 07:24 pm, netdev-bot+sashiko@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 4 potential
> issue(s) to consider.
>
> Critical: 0 · High: 0 · Medium: 4 · Low: 0
>
> If you believe none of the issues are legitimate, please reply to each
> issue of Medium severity or higher and explain why the AI is wrong.
> If any of them are legitimate and you plan to respin, please let patchwork
> know by including "pw-bot: cr" as a separate line at the end of your reply
> (one such reply per series is enough).
>
> - [Medium] The patch does not do what the commit message says for BAM
>    descriptor FIFOs.
> - [Medium] The driver can unbind while the restricted DMA pool is still
>    assigned to {HLOS, vmid}, and then it loses the only record of that.
> - [Medium] Requesting the PC IRQ with IRQF_NO_AUTOEN and enabling it only
>    at the end of bam_dmux_probe() means any change in the remote's power-…
> - [Medium] bam_dmux_remove() decides whether to take the restricted DMA
>    pool away from the remote VM using the cached software flag…
>
> --
>
> net: wwan: qcom_bam_dmux: Assign restricted DMA pool to remote processor
>
> When qcom,vmid is present, this patch requires a restricted DMA pool to
> be attached to the device. It then calls qcom_scm_assign_mem() to share
> that pool between HLOS and the modem VMID before the DMA channels are
> requested. The pool is handed back to HLOS on remove or on a probe
> error, unless the remote still holds its power vote, and the PC IRQ is
> now requested with IRQF_NO_AUTOEN and enabled at the end of probe.
>
>> diff --git a/drivers/net/wwan/qcom_bam_dmux.c b/drivers/net/wwan/qcom_bam_dmux.c
>> index 1479f8362e04e..04f3d46d4c3f3 100644
>> --- a/drivers/net/wwan/qcom_bam_dmux.c
>> +++ b/drivers/net/wwan/qcom_bam_dmux.c
> [ ... ]
>
>> @@ -760,6 +766,88 @@ static int __maybe_unused bam_dmux_runtime_resume(struct device *dev)
>>   	return 0;
>>   }
>>   
>> +static int bam_dmux_assign_dma_pool(struct bam_dmux *dmux)
>> +{
>> +	struct device *dev = dmux->dev;
>> +	struct device_node *rmem_np;
>> +	struct reserved_mem *rmem;
>> +	struct qcom_scm_vmperm dst[2];
>> +	u64 src = BIT_ULL(QCOM_SCM_VMID_HLOS);
> [ ... ]
>
>> +	if (!is_swiotlb_for_alloc(dev))
>> +		return dev_err_probe(dev, -EINVAL,
>> +				     "Restricted DMA pool not attached\n");
> [Severity: Medium]
> Does this cover the BAM descriptor FIFOs the way the commit message
> describes? The commit message says:
>
>    "BAM-DMUX DMA mappings and BAM descriptor FIFOs must therefore be
>    constrained to a designated memory region"
>
> and:
>
>    "When the BAM DMA controller references the same pool, its descriptor
>    FIFOs are covered by the assignment as well."
>
> The FIFOs are allocated on the BAM DMA controller's own device, not on
> dmux->dev:
>
> drivers/dma/qcom/bam_dma.c:bam_alloc_chan() {
>      ...
> 	bchan->fifo_virt = dma_alloc_wc(bdev->dev, BAM_DESC_FIFO_SIZE,
> 					&bchan->fifo_phys, GFP_KERNEL);
>      ...
> }
>
> The FIFOs only come from the restricted pool if the BAM DMA node has its
> own memory-region that points at the restricted-dma-pool. The binding
> doesn't allow that. qcom,bam-dma.yaml has no memory-region property and
> sets additionalProperties: false, and this series only updates
> qcom,bam-dmux.yaml.
>
> The check here only looks at is_swiotlb_for_alloc(dmux->dev). Nothing
> checks the device behind the rx/tx channels.
>
> With a DT that follows the bindings, probe would succeed without a
> warning. Meanwhile, bam_dmux_power_on()->dma_request_chan()->
> bam_alloc_chan() would allocate the FIFOs outside the assigned region.
> Wouldn't that leave the XPU violation the patch is meant to prevent?
>
> Is a bam-dma binding update or a check on the channel device needed
> here?

This is covered by the separate BAM DMA binding change:
https://lore.kernel.org/all/20261021-qcom-bam-dma-vmid-ext-v3-1-299355e8dd3b@oss.qualcomm.com/

>
> [ ... ]
>
>> +static void bam_dmux_reclaim_dma_pool(struct bam_dmux *dmux)
>> +{
>> +	struct qcom_scm_vmperm hlos = {
>> +		.vmid = QCOM_SCM_VMID_HLOS,
>> +		.perm = QCOM_SCM_PERM_RW,
>> +	};
>> +	u64 src = dmux->dma_pool_perms;
>> +	int ret;
>> +
>> +	if (!dmux->dma_pool_perms)
>> +		return;
>> +
>> +	ret = qcom_scm_assign_mem(dmux->dma_pool_base, dmux->dma_pool_size, &src,
>> +				  &hlos, 1);
>> +	if (ret) {
>> +		dev_err(dmux->dev, "SCM reclaim restricted DMA pool failed: %d\n", ret);
>> +		return;
>> +	}
> [Severity: Medium]
> What happens to the pool ownership when this call fails? The same
> question applies when bam_dmux_remove() skips the reclaim on purpose
> because pc_state is still set:
>
> 	if (!dmux->pc_state)
> 		bam_dmux_reclaim_dma_pool(dmux);
> 	else if (dmux->dma_pool_perms)
> 		dev_warn(dev, "Leaving restricted DMA pool assigned to remote\n");
>
> In both cases the region stays assigned to HLOS and the modem VMID. The
> only record of that is dmux->dma_pool_perms, which lives in the
> devm_kzalloc'd struct bam_dmux. devres frees it as soon as remove() or
> the failed probe returns.
>
> On the next probe (module reload, sysfs rebind, deferred re-probe),
> bam_dmux_assign_dma_pool() always starts from:
>
> 	u64 src = BIT_ULL(QCOM_SCM_VMID_HLOS);
>
> That no longer matches the real owners (HLOS | vmid). Hyp-assign checks
> the source VM list against the current ownership, which is why users
> such as rmtfs_mem keep their perms around for the reclaim.
>
> Wouldn't qcom_scm_assign_mem() then fail with "SCM assign restricted DMA
> pool failed" on every probe until reboot?
>
> The commit message says the execution environments are tracked
> explicitly for the reclaim. Isn't that record lost in exactly the unbind
> case it is meant to handle?

The tracked permissions are used for the reclaim in remove(). If the reclaim
is skipped or fails, the pool stays assigned, which is fail-safe, since the pool
is static reserved memory that is never returned to the page allocator. This
matches the existing qcom_scm_assign_mem() users, which also only log reclaim
failures.

>
>> +
>> +	dmux->dma_pool_perms = 0;
>> +}
> [ ... ]
>
>> @@ -821,7 +913,7 @@ static int bam_dmux_probe(struct platform_device *pdev)
>>   		goto err_disable_pm;
>>   
>>   	ret = devm_request_threaded_irq(dev, dmux->pc_irq, NULL, bam_dmux_pc_irq,
>> -					IRQF_ONESHOT, NULL, dmux);
>> +					IRQF_ONESHOT | IRQF_NO_AUTOEN, NULL, dmux);
>>   	if (ret)
>>   		goto err_disable_pm;
>>   
>> @@ -838,11 +930,14 @@ static int bam_dmux_probe(struct platform_device *pdev)
>>   			bam_dmux_power_off(dmux);
>>   	}
>>   
>> +	enable_irq(dmux->pc_irq);
> [Severity: Medium]
> Can a remote power-control transition get lost between the line-level
> sample and this enable_irq()? In bam_dmux_probe() the sequence is now:
>
> 	ret = irq_get_irqchip_state(dmux->pc_irq, IRQCHIP_STATE_LINE_LEVEL,
> 				    &dmux->pc_state);
> 	...
> 	if (dmux->pc_state) {
> 		if (bam_dmux_power_on(dmux))
> 	...
> 	enable_irq(dmux->pc_irq);
>
> With IRQF_NO_AUTOEN, __setup_irq() skips irq_startup(), so the smp2p
> and smsm unmask callbacks don't run until enable_irq(). During that
> window, qcom_smp2p_notify_in() still updates last_value but only
> dispatches bits set in irq_enabled:
>
> drivers/soc/qcom/smp2p.c:qcom_smp2p_notify_in() {
>      ...
> 		status = val ^ entry->last_value;
> 		entry->last_value = val;
>      ...
> 		for_each_set_bit(i, entry->irq_enabled, 32) {
>      ...
> }
>
> smsm_intr() also updates last_value, and smsm_unmask_irq() resyncs the
> cached state when the IRQ is enabled:
>
> drivers/soc/qcom/smsm.c:smsm_unmask_irq() {
>      ...
> 	/* Make sure our last cached state is up-to-date */
> 	assign_bit(irq, &entry->last_value,
> 		   readl(entry->remote_state) & BIT(irq));
>      ...
> }
>
> handle_nested_irq() is never called, so IRQS_PENDING is never set and
> enable_irq() has nothing to resend.
>
> bam_dmux_pc_irq() flips the cached value instead of reading the line:
>
> 	bool new_state = !dmux->pc_state;
>
> A single missed edge would therefore leave the driver's view inverted
> from then on. The remote's transition is never acked, later edges are
> handled with the wrong polarity, and the pc_state check in
> bam_dmux_remove() that gates the pool reclaim is wrong too.
>
> When the modem is already up at probe time, the window also covers
> bam_dmux_power_on() (dma_request_chan(), skb allocation, DMA mapping).
> IRQF_NO_AUTOEN is set unconditionally. Wouldn't this regression affect
> every BAM-DMUX platform, not just the ones with qcom,vmid?

Yes, this was a known trade-off when moving to IRQF_NO_AUTOEN. Closing
the window needs a level-based bam_dmux_pc_irq(), which changes the PC
handshake on all BAM-DMUX platforms and can be addressed in a separate
patch if needed.

>
>>   
>>   	return 0;
>>   
>>   err_disable_pm:
>>   	pm_runtime_disable(dev);
>>   	pm_runtime_dont_use_autosuspend(dev);
>> +	bam_dmux_reclaim_dma_pool(dmux);
>>   	return ret;
>>   }
>>   
>> @@ -877,6 +972,12 @@ static void bam_dmux_remove(struct platform_device *pdev)
>>   	disable_irq(dmux->pc_irq);
>>   	bam_dmux_power_off(dmux);
>>   	bam_dmux_free_skbs(dmux->tx_skbs, DMA_TO_DEVICE);
>> +
>> +	/* The remote side may still access the pool if it holds its power vote */
>> +	if (!dmux->pc_state)
>> +		bam_dmux_reclaim_dma_pool(dmux);
>> +	else if (dmux->dma_pool_perms)
>> +		dev_warn(dev, "Leaving restricted DMA pool assigned to remote\n");
> [Severity: Medium]
> Can dmux->pc_state still be trusted at this point? Once
> disable_irq(dmux->pc_irq) has run above, bam_dmux_pc_irq() no longer
> runs. pc_state then shows the last edge that was processed, not the
> remote's current power vote.
>
> The RX side is started by the remote. The comment in probe says "the RX
> path may be active even if we are runtime suspended, since it is
> controlled by the remote side".
>
> Suppose the remote raises its vote after disable_irq(), or pc_state is
> already inverted by the lost edge in bam_dmux_probe(). This code would
> still call bam_dmux_reclaim_dma_pool() and remove the modem VMID while
> the remote holds its vote.
>
> That seems to contradict the commit message:
>
>    "Skip the reclaim and leave the pool assigned if the remote side still
>    holds its power vote, since it may still access the pool."
>
> Should the line level be read again here before deciding whether to
> reclaim, for example with irq_get_irqchip_state(dmux->pc_irq,
> IRQCHIP_STATE_LINE_LEVEL, ...)?

In both cases, the reclaim runs after bam_dmux_power_off(), which has
already reset the BAM pipes and freed their descriptor FIFOs, so no
transfers that access the pool are left. A vote raised after
disable_irq() is also never acked. Reading the line level here can be
done together with the level-based PC handling mentioned above if
needed.

Thanks,
Vishnu

>
>>   }

  reply	other threads:[~2026-11-18  5:48 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-12  9:13 [PATCH net-next v6 0/2] net: wwan: qcom_bam_dmux: Add restricted DMA pool support Vishnu Santhosh
2026-11-12  9:13 ` [PATCH net-next v6 1/2] dt-bindings: net: qcom,bam-dmux: Add memory-region and qcom,vmid properties Vishnu Santhosh
2026-11-16 13:54   ` netdev-bot+sashiko
2026-11-18  5:28     ` Vishnu Santhosh
2026-11-12  9:13 ` [PATCH net-next v6 2/2] net: wwan: qcom_bam_dmux: Assign restricted DMA pool to remote processor Vishnu Santhosh
2026-11-16 13:54   ` netdev-bot+sashiko
2026-11-18  5:48     ` Vishnu Santhosh [this message]
2026-11-16 14:43   ` Loic Poulain

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=12a0edcc-2c99-4f68-9ebd-0a9ec76bece7@oss.qualcomm.com \
    --to=vishnu.santhosh@oss.qualcomm.com \
    --cc=andrew+netdev@lunn.ch \
    --cc=conor+dt@kernel.org \
    --cc=davem@davemloft.net \
    --cc=deepak.singh@oss.qualcomm.com \
    --cc=devicetree@vger.kernel.org \
    --cc=edumazet@kernel.org \
    --cc=johannes@sipsolutions.net \
    --cc=krzk+dt@kernel.org \
    --cc=kuba@kernel.org \
    --cc=linux-arm-msm@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=loic.poulain@oss.qualcomm.com \
    --cc=netdev-bot+sashiko@kernel.org \
    --cc=netdev@vger.kernel.org \
    --cc=pabeni@redhat.com \
    --cc=robh@kernel.org \
    --cc=ryazanov.s.a@gmail.com \
    --cc=stephan@gerhold.net \
    /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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox