From: Neil Armstrong <neil.armstrong@linaro.org>
To: sashiko-reviews@lists.linux.dev
Cc: linux-sound@vger.kernel.org, devicetree@vger.kernel.org,
linux-clk@vger.kernel.org, linux-gpio@vger.kernel.org,
robh@kernel.org, conor+dt@kernel.org
Subject: Re: [PATCH RFC 15/23] ASoC: qcom: lpass: allocate LPM buffer memory to DAIs
Date: Mon, 16 Nov 2026 08:56:29 +0100 [thread overview]
Message-ID: <73036d41-bec4-4d09-891b-8a2432a08f2d@linaro.org> (raw)
In-Reply-To: <sashiko-outbox-213524@kernel.org>
On 11/13/26 20:12, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 5 potential
> issues to consider:
> - [High] ASoC: qcom: lpass: u32 underflow and OOB in
> lpass_platform_allocate_lpm()
> - [Medium] ASoC: qcom: lpass: uninitialized capture buffer on
> full-duplex LPM MI2S PCM
> - [Medium] ASoC: qcom: lpass: unconditional VA LPM reservation blocks VA
> LPM MI2S
> - [Medium] ASoC: qcom: lpass: missing LPM clock parsing in
> asoc_qcom_lpass_cpu_platform_probe()
> - [Low] ASoC: qcom: lpass: wrong LPM offset in
> lpass_platform_prealloc_cdc_dma_buffer()
> --
>
> commit 038ab20def13fb9da7aca847692783be3d4964b4
> Author: Neil Armstrong <neil.armstrong@linaro.org>
>
> ASoC: qcom: lpass: allocate LPM buffer memory to DAIs
>
> This commit dynamically allocates LPASS LPM SRAM buffer offsets for Codec
> DMA and non-core LPAIF MI2S DAIs instead of using fixed static offsets.
>
>> diff --git a/sound/soc/qcom/lpass-cpu.c b/sound/soc/qcom/lpass-cpu.c
>> index 9ebbc422c63c3..4bbcd59ebd602 100644
>> --- a/sound/soc/qcom/lpass-cpu.c
>> +++ b/sound/soc/qcom/lpass-cpu.c
>> @@ -1258,16 +1258,6 @@ int asoc_qcom_lpass_cpu_platform_probe(struct platform_device *pdev)
>> dev_err(dev, "failed to get cdc dma clocks %d\n", ret);
>> return ret;
>> }
>> -
>> - res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lpass-rxtx-cdc-dma-lpm");
>> - if (!res)
>> - return -EINVAL;
>> - drvdata->rxtx_cdc_dma_lpm_buf = res->start;
>> -
>> - res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lpass-va-cdc-dma-lpm");
>> - if (!res)
>> - return -EINVAL;
>> - drvdata->va_cdc_dma_lpm_buf = res->start;
>> }
>>
>> if (variant->mi2s_desc) {
>> @@ -1281,6 +1271,9 @@ int asoc_qcom_lpass_cpu_platform_probe(struct platform_device *pdev)
>> variant->mi2s_desc[i].reg_name);
>> if (ret)
>> return ret;
>> +
>> + if (variant->mi2s_desc[i].lpm_type != LPASS_LPM_NONE)
>> + drvdata->mi2s_lpm_enable = true;
>> }
>> } else {
>> drvdata->num_lpaif = 1;
>> @@ -1290,6 +1283,20 @@ int asoc_qcom_lpass_cpu_platform_probe(struct platform_device *pdev)
>> return ret;
>> }
>>
>> + if (drvdata->codec_dma_enable || drvdata->mi2s_lpm_enable) {
>> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lpass-rxtx-cdc-dma-lpm");
>> + if (!res)
>> + return -EINVAL;
>> + drvdata->rxtx_cdc_dma_lpm_buf = res->start;
>> + drvdata->rxtx_cdc_dma_lpm_size = resource_size(res);
>> +
>> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lpass-va-cdc-dma-lpm");
>> + if (!res)
>> + return -EINVAL;
>> + drvdata->va_cdc_dma_lpm_buf = res->start;
>> + drvdata->va_cdc_dma_lpm_size = resource_size(res);
>> + }
>
> [Severity: Medium]
> Should of_lpass_cdc_dma_clks_parse() also be called when
> drvdata->mi2s_lpm_enable is set?
>
> Because of_lpass_cdc_dma_clks_parse() remains inside the
> if (drvdata->codec_dma_enable) block above, a board that enables LPM-backed
> MI2S DAIs without any Codec DMA DAI leaves drvdata->codec_mem0..3 and
> drvdata->va_mem0 as NULL. Even with commit "ASoC: qcom: lpass: enable the
> right codec_mem clock when usin LPM memory" later in this series,
> lpass_cpu_daiops_startup() will pass a NULL clk pointer to
> clk_prepare_enable(), leaving the LPM SRAM unclocked when accessed by the
> CPU or LPAIF DMA.
Good catch will fix
>
>> diff --git a/sound/soc/qcom/lpass-platform.c b/sound/soc/qcom/lpass-platform.c
>> index a953db262aa00..75d2c872f5123 100644
>> --- a/sound/soc/qcom/lpass-platform.c
>> +++ b/sound/soc/qcom/lpass-platform.c
>
> [ ... ]
>
>> @@ -1315,23 +1348,55 @@ static int lpass_platform_prealloc_cdc_dma_buffer(struct snd_soc_component *comp
>> switch (dai_id) {
>> case LPASS_CDC_DMA_RX0 ... LPASS_CDC_DMA_RX9:
>> buf->bytes = lpass_platform_rxtx_hardware.buffer_bytes_max;
>> - buf->addr = drvdata->rxtx_cdc_dma_lpm_buf;
>> + buf->addr = drvdata->rxtx_cdc_dma_lpm_buf + drvdata->va_lpm_offset;
>> break;
>> case LPASS_CDC_DMA_TX0 ... LPASS_CDC_DMA_TX8:
>> buf->bytes = lpass_platform_rxtx_hardware.buffer_bytes_max;
>> - buf->addr = drvdata->rxtx_cdc_dma_lpm_buf + LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE;
>> + buf->addr = drvdata->rxtx_cdc_dma_lpm_buf +
>> + drvdata->va_lpm_offset + LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE;
>> break;
>
> [Severity: Low]
> Should these two cases use drvdata->rxtx_lpm_offset instead of
> drvdata->va_lpm_offset? lpass_platform_allocate_lpm() populates
> drvdata->rxtx_lpm_offset for RXTX Codec DMA, and commit "ASoC: qcom: lpass:
> enable the right codec_mem clock when usin LPM memory" uses
> drvdata->rxtx_lpm_offset to select the RXTX LPM clock, so using
> va_lpm_offset here will desynchronize the buffer address from the clock if
> the two offsets ever differ.
Good catch will fix
>
>> case LPASS_CDC_DMA_VA_TX0 ... LPASS_CDC_DMA_VA_TX8:
>> buf->bytes = lpass_platform_va_hardware.buffer_bytes_max;
>> - buf->addr = drvdata->va_cdc_dma_lpm_buf;
>> + buf->addr = drvdata->va_cdc_dma_lpm_buf + drvdata->va_lpm_offset;
>> break;
>> case LPASS_CDC_DMA_WSA_RX0 ... LPASS_CDC_DMA_WSA_RX9:
>> buf->bytes = lpass_platform_wsa_hardware.buffer_bytes_max;
>> - buf->addr = drvdata->rxtx_cdc_dma_lpm_buf + LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE * 2;
>> + buf->addr = drvdata->rxtx_cdc_dma_lpm_buf + drvdata->wsa_lpm_offset;
>> break;
>> case LPASS_CDC_DMA_WSA_TX0 ... LPASS_CDC_DMA_WSA_TX8:
>> buf->bytes = lpass_platform_wsa_hardware.buffer_bytes_max;
>> - buf->addr = drvdata->rxtx_cdc_dma_lpm_buf + LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE * 3;
>> + buf->addr = drvdata->rxtx_cdc_dma_lpm_buf +
>> + drvdata->wsa_lpm_offset + LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE;
>> + break;
>> + case MI2S_PRIMARY ... MI2S_QUINARY:
>> + case MI2S_SENARY ... MI2S_SEPTENARY:
>> + if (is_lpm_mi2s_port(drvdata, dai_id)) {
>> + unsigned int port_id = mi2s_port_id(dai_id);
>> +
>> + buf->bytes = lpass_platform_lpm_mi2s_hardware.buffer_bytes_max;
>> + switch (drvdata->mi2s_lpm[port_id]) {
>> + case LPASS_RXTX_CDC_LPM:
>> + if (pcm->streams[SNDRV_PCM_STREAM_PLAYBACK].substream)
>> + buf->addr = drvdata->rxtx_cdc_dma_lpm_buf +
>> + drvdata->mi2s_lpm_offset[port_id];
>> + else
>> + buf->addr = drvdata->rxtx_cdc_dma_lpm_buf +
>> + drvdata->mi2s_lpm_offset[port_id] +
>> + LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE;
>> + break;
>> + case LPASS_VA_CDC_LPM:
>> + if (pcm->streams[SNDRV_PCM_STREAM_PLAYBACK].substream)
>> + buf->addr = drvdata->va_cdc_dma_lpm_buf +
>> + drvdata->mi2s_lpm_offset[port_id];
>> + else
>> + buf->addr = drvdata->va_cdc_dma_lpm_buf +
>> + drvdata->mi2s_lpm_offset[port_id] +
>> + LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE;
>> + break;
>> + default:
>> + break;
>> + }
>> + }
>> break;
>> default:
>> break;
>
> [Severity: Medium]
> Does lpass_platform_prealloc_cdc_dma_buffer() leave the capture
> substream->dma_buffer uninitialized when an LPM MI2S PCM has both playback
> and capture streams?
>
> At the top of lpass_platform_prealloc_cdc_dma_buffer(), only one substream
> is selected:
>
> sound/soc/qcom/lpass-platform.c:lpass_platform_prealloc_cdc_dma_buffer() {
> if (pcm->streams[SNDRV_PCM_STREAM_PLAYBACK].substream)
> substream = pcm->streams[SNDRV_PCM_STREAM_PLAYBACK].substream;
> else
> substream = pcm->streams[SNDRV_PCM_STREAM_CAPTURE].substream;
>
> buf = &substream->dma_buffer;
> ...
> }
>
> Unlike Codec DMA DAIs (which have separate unidirectional RX and TX DAI
> IDs), a single MI2S DAI can have both playback and capture substreams on
> the same struct snd_pcm. When both exist, only the playback dma_buffer is
> initialized and memremapped, while the capture substream->dma_buffer stays
> zeroed. Opening the capture stream in lpass_platform_pcmops_open() then
> sets runtime->dma_area to NULL via snd_pcm_set_runtime_buffer(), causing a
> NULL pointer dereference in lpass_platform_copy().
Good catch, will fix
>
> [ ... ]
>
>> @@ -1464,6 +1527,64 @@ static const struct snd_soc_component_driver lpass_component_driver = {
>>
>> };
>>
>> +static int lpass_platform_allocate_lpm(struct lpass_data *drvdata)
>> +{
>> + u32 rxtx_cdc_dma_lpm_remain = drvdata->rxtx_cdc_dma_lpm_size;
>> + u32 va_cdc_dma_lpm_remain = drvdata->va_cdc_dma_lpm_size;
>> + u32 rxtx_cdc_dma_lpm_offset = 0;
>> + u32 va_cdc_dma_lpm_offset = 0;
>> + int port_id;
>> +
>> + if (drvdata->codec_dma_enable) {
>> + drvdata->rxtx_lpm_offset = rxtx_cdc_dma_lpm_offset;
>> + rxtx_cdc_dma_lpm_offset += LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE * 2;
>> + rxtx_cdc_dma_lpm_remain -= LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE * 2;
>> +
>> + drvdata->va_lpm_offset = va_cdc_dma_lpm_offset;
>> + va_cdc_dma_lpm_offset += LPASS_VA_CDC_DMA_LPM_BUFF_SIZE;
>> + va_cdc_dma_lpm_remain -= LPASS_VA_CDC_DMA_LPM_BUFF_SIZE;
>> +
>> + if (drvdata->wsa_enable) {
>> + drvdata->wsa_lpm_offset = rxtx_cdc_dma_lpm_offset;
>> + rxtx_cdc_dma_lpm_offset += LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE * 2;
>> + rxtx_cdc_dma_lpm_remain -= LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE * 2;
>> + }
>> + }
>
> [Severity: Medium]
> Does reserving LPASS_VA_CDC_DMA_LPM_BUFF_SIZE whenever
> drvdata->codec_dma_enable is set prevent using LPASS_VA_CDC_LPM for MI2S
> when only RXTX or WSA Codec DMA is enabled?
>
> of_lpass_cpu_parse_dai_data() sets drvdata->codec_dma_enable = 1 for any
> RXTX, WSA, or VA Codec DMA DAI. Because the hardware lpass-va-cdc-dma-lpm
> region is 12 KiB (LPASS_VA_CDC_DMA_LPM_BUFF_SIZE), subtracting
> LPASS_VA_CDC_DMA_LPM_BUFF_SIZE here even when no VA Codec DMA DAI is in DT
> reduces va_cdc_dma_lpm_remain to 0, causing the LPASS_VA_CDC_LPM check
> below to always fail with -EINVAL.
Good catch will add the check for VA
>
>> +
>> + if (!drvdata->mi2s_lpm_enable)
>> + return 0;
>> +
>> + for (port_id = 0; port_id < LPASS_MAX_MI2S_PORTS; port_id++) {
>> + if (!test_bit(port_id, &drvdata->lpaif_port_enable))
>> + continue;
>> +
>> + switch (drvdata->mi2s_lpm[port_id]) {
>> + case LPASS_RXTX_CDC_LPM:
>> + if (rxtx_cdc_dma_lpm_remain < LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE)
>> + return -EINVAL;
>> + drvdata->mi2s_lpm_offset[port_id] = rxtx_cdc_dma_lpm_offset;
>> + rxtx_cdc_dma_lpm_offset += LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE * 2;
>> + rxtx_cdc_dma_lpm_remain -= LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE * 2;
>> + break;
>> + case LPASS_VA_CDC_LPM:
>> + /*
>> + * VA Codec uses 12KiB, but MI2S uses 8KiB
>> + * Meaning only VA _or_ a single MI2S can be enabled
>> + */
>> + if (va_cdc_dma_lpm_remain < LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE)
>> + return -EINVAL;
>> + drvdata->mi2s_lpm_offset[port_id] = va_cdc_dma_lpm_offset;
>> + va_cdc_dma_lpm_offset += LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE * 2;
>> + va_cdc_dma_lpm_remain -= LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE * 2;
>> + break;
>
> [Severity: High]
> Can rxtx_cdc_dma_lpm_remain and va_cdc_dma_lpm_remain underflow and assign
> out-of-bounds LPM buffer offsets here?
>
> Both cases check for LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE (8 KiB) remaining,
> but then advance the offset and decrement the unsigned u32 remaining
> counter by LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE * 2 (16 KiB). On a 12 KiB
> lpass-va-cdc-dma-lpm region (0x3000 bytes), enabling one MI2S port on
> LPASS_VA_CDC_LPM passes the < 0x2000 check and subtracts 0x4000, wrapping
> va_cdc_dma_lpm_remain to 0xfffff000:
>
> 1. If a second MI2S port on LPASS_VA_CDC_LPM is enabled (such as both
> MI2S_QUATERNARY and MI2S_QUINARY), 0xfffff000 < 0x2000 is false, so it
> is given an offset 16 KiB past the start of the 12 KiB VA LPM region.
> 2. Even with a single MI2S port on LPASS_VA_CDC_LPM, a capture stream in
> lpass_platform_prealloc_cdc_dma_buffer() is placed at
> mi2s_lpm_offset[port_id] + LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE (8 KiB ..
> 16 KiB), overflowing the 12 KiB VA LPM region by 4 KiB during
> memremap(), mmap(), and DMA.
> 3. The same u32 underflow occurs on LPASS_RXTX_CDC_LPM whenever 8 KiB
> remains before subtracting 16 KiB.
>
Good catch, will check for LPASS_RXTX_CDC_DMA_LPM_BUFF_SIZE*2
next prev parent reply other threads:[~2026-11-16 7:56 UTC|newest]
Thread overview: 54+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-13 18:38 [PATCH RFC 00/23] arm64: qcom: enable SM8650 audio without ADSP firmware Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 01/23] dt-bindings: clock: qcom: add SM8650 LPASS core & audio clock controllers Neil Armstrong
2026-11-13 18:57 ` sashiko-bot
2026-11-16 7:41 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 02/23] clk: " Neil Armstrong
2026-11-13 19:14 ` sashiko-bot
2026-11-16 7:44 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 03/23] dt-bindings: pinctrl: qcom,sm8650-lpass-lpi-pinctrl: document clocks and power domaines when ADSP is not used Neil Armstrong
2026-11-13 19:02 ` sashiko-bot
2026-11-16 7:45 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 04/23] ASoC: dt-bindings: qcom,lpass: add Senary & Septenary MI2S ids Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 05/23] ASoC: dt-bindings: qcom,lpass: add WSA RX and TX channels ids Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 06/23] ASoC: dt-bindings: qcom,lpass-*-macro: document less clock when ADSP is not used Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 07/23] ASoC: dt-bindings: qcom,lpass-cpu: document the SM8650 LPASS Neil Armstrong
2026-11-13 18:57 ` sashiko-bot
2026-11-16 7:45 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 08/23] ASoC: qcom: lpass-cpu: get the optional codec_mem3 clock Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 09/23] ASoC: qcom: lpass: add support for optional WSA interface Neil Armstrong
2026-11-13 19:11 ` sashiko-bot
2026-11-16 7:49 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 10/23] ASoC: qcom: lpass-cdc-dma: add support for WSA codec DMA Neil Armstrong
2026-11-13 19:07 ` sashiko-bot
2026-11-16 7:50 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 11/23] ASoC: qcom: lpass: prepare to handle multiple lpaifs for i2s Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 12/23] ASoC: qcom: lpass: add support for MI2S_SENARY & MI2S_SEPTENARY Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 13/23] ASoC: qcom: lpass: support mapping I2S ports to multiple LPAIFs Neil Armstrong
2026-11-13 19:04 ` sashiko-bot
2026-11-16 7:51 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 14/23] ASoC: qcom: lpass: support LPM memory type for I2S LPAIFs Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 15/23] ASoC: qcom: lpass: allocate LPM buffer memory to DAIs Neil Armstrong
2026-11-13 19:12 ` sashiko-bot
2026-11-16 7:56 ` Neil Armstrong [this message]
2026-11-13 18:38 ` [PATCH RFC 16/23] ASoC: qcom: lpass: enable the right codec_mem clock when usin LPM memory Neil Armstrong
2026-11-13 19:06 ` sashiko-bot
2026-11-16 7:58 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 17/23] ASoC: qcom: add SM8650 LPASS CPU driver Neil Armstrong
2026-11-13 19:22 ` sashiko-bot
2026-11-16 7:59 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 18/23] ASoC: qcom: swd: allow LPASS_CDC_DMA_WSA_RX0 as soundwire dai Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 19/23] ASoC: qcom: sc8280xp: allow using the sc8280xp card with lpass-cpu Neil Armstrong
2026-11-13 19:12 ` sashiko-bot
2026-11-16 8:01 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 20/23] arm64: dts: qcom: sm8650: move EL1 Gunyah nodes to sm8650-el1.dtsi Neil Armstrong
2026-11-13 19:08 ` Val Packett
2026-11-16 7:40 ` Neil Armstrong
2026-11-13 19:11 ` sashiko-bot
2026-11-16 8:02 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 21/23] arm64: dts: qcom: add SM8650 LPASS CPU dtsi Neil Armstrong
2026-11-13 19:18 ` sashiko-bot
2026-11-16 8:04 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 22/23] arm64: dts: qcom: sm8650-valve-deckard: enable audio Neil Armstrong
2026-11-13 19:23 ` sashiko-bot
2026-11-16 8:05 ` Neil Armstrong
2026-11-13 18:38 ` [PATCH RFC 23/23] arm64: defconfig: enable the SM8650 LPASS CPU driver Neil Armstrong
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=73036d41-bec4-4d09-891b-8a2432a08f2d@linaro.org \
--to=neil.armstrong@linaro.org \
--cc=conor+dt@kernel.org \
--cc=devicetree@vger.kernel.org \
--cc=linux-clk@vger.kernel.org \
--cc=linux-gpio@vger.kernel.org \
--cc=linux-sound@vger.kernel.org \
--cc=robh@kernel.org \
--cc=sashiko-reviews@lists.linux.dev \
/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