dri-devel Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: sashiko-bot@kernel.org
To: "Mukesh Ojha" <mukesh.ojha@oss.qualcomm.com>
Cc: dri-devel@lists.freedesktop.org, media-ci@linuxtv.org
Subject: Re: [PATCH 3/8] media: iris: migrate firmware loading to qcom_mdt_pas_load() and qcom_pas_prepare_and_auth_reset()
Date: Sat, 26 Sep 2026 15:59:35 +0000	[thread overview]
Message-ID: <20260926155935.D7FDD1F000FF@smtp.kernel.org> (raw)
In-Reply-To: <20260925155737.4161358-4-mukesh.ojha@oss.qualcomm.com>

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] firmware: qcom: pas: missing qcom_pas_metadata_release() leaks TZ/DMA memory
- [Medium] devres: memory leak due to repeated devm_qcom_pas_context_alloc() on fw reload
--

commit bb3acb8beaa539b1cfa0975914c1f8fd75747e36
Author: Mukesh Ojha <mukesh.ojha@oss.qualcomm.com>

media: iris: migrate firmware loading to qcom_mdt_pas_load() and qcom_pas_prepare_and_auth_reset()

iris_load_fw_to_memory() calls qcom_mdt_load() which bundles metadata
init, memory setup, and segment loading with no PAS context, and
iris_fw_load() calls qcom_pas_auth_and_reset() which skips the shmbridge
prepare step required before TrustZone authentication.

> diff --git a/drivers/media/platform/qcom/iris/iris_firmware.c b/drivers/media/platform/qcom/iris/iris_firmware.c
> index 21cdf1c398357..e73a5dd98b88c 100644
> --- a/drivers/media/platform/qcom/iris/iris_firmware.c
> +++ b/drivers/media/platform/qcom/iris/iris_firmware.c
[ ... ]
> @@ -134,21 +131,20 @@ static int iris_load_fw_to_memory(struct iris_core *core)
>  	core->iris_firmware_data = core->iris_firmware_desc->firmware_data;
>  
>  	fw_size = qcom_mdt_get_size(firmware);
> -	if (fw_size < 0 || res_size < (size_t)fw_size) {
> +	if (fw_size < 0 || resource_size(&res) < (size_t)fw_size) {
>  		ret = -EINVAL;
>  		goto err_release_fw;
>  	}
>  
> -	mem_virt = memremap(mem_phys, res_size, MEMREMAP_WC);
> -	if (!mem_virt) {
> -		ret = -ENOMEM;
> +	core->pas_ctx = devm_qcom_pas_context_alloc(dev, IRIS_PAS_ID, mem_phys,
> +						    resource_size(&res));

[Severity: Medium]
Will this leak device-managed memory when firmware is reloaded?

If the hardware system error recovery path is triggered (for example,
iris_sys_error_handler() calls iris_core_deinit() followed by
iris_core_init() and iris_fw_load()), devm_qcom_pas_context_alloc() is called
again unconditionally. This overwrites core->pas_ctx without freeing the old
devres allocation.

> +	if (IS_ERR(core->pas_ctx)) {
> +		ret = PTR_ERR(core->pas_ctx);
> +		core->pas_ctx = NULL;
>  		goto err_release_fw;
>  	}
>  
> -	ret = qcom_mdt_load(dev, firmware, fw_name,
> -			    IRIS_PAS_ID, mem_virt, mem_phys, res_size, NULL);
> -
> -	memunmap(mem_virt);
> +	ret = qcom_mdt_pas_load(core->pas_ctx, firmware, fw_name, NULL);

[Severity: High]
Does this sequence leak DMA or TrustZone memory buffers allocated for firmware
metadata?

When qcom_mdt_pas_load() is called, the underlying PAS subsystem allocates a
DMA buffer for the metadata and stores it in the PAS context. The PAS API
contract dictates that callers must subsequently call
qcom_pas_metadata_release(core->pas_ctx) to free this memory.

Looking at iris_fw_load(), this step appears to be completely omitted on both
success and error paths before the function returns. Because the DMA
allocation is tied to the SCM device (not devres-managed for the iris
driver), repeated firmware loads due to probe/unbind cycles or hardware
errors could permanently exhaust restricted DMA or secure memory pools.

>  err_release_fw:
>  	release_firmware(firmware);
>  
> @@ -166,7 +162,7 @@ int iris_fw_load(struct iris_core *core)
>  		return ret;
>  	}
>  
> -	ret = qcom_pas_auth_and_reset(IRIS_PAS_ID);
> +	ret = qcom_pas_prepare_and_auth_reset(core->pas_ctx);
>  	if (ret)  {
>  		dev_err(core->dev, "auth and reset failed: %d\n", ret);
>  		return ret;

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260925155737.4161358-1-mukesh.ojha@oss.qualcomm.com?part=3

  reply	other threads:[~2026-09-26 15:59 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-09-25 15:57 [PATCH 0/8] Migrate PAS clients to qcom_mdt_pas_load() and qcom_pas_prepare_and_auth_reset() Mukesh Ojha
2026-09-25 15:57 ` [PATCH 1/8] remoteproc: qcom_wcnss: migrate " Mukesh Ojha
2026-09-26 15:59   ` sashiko-bot
2026-09-25 15:57 ` [PATCH 2/8] drm/msm/adreno: migrate zap shader loading " Mukesh Ojha
2026-09-26 15:59   ` sashiko-bot
2026-09-25 15:57 ` [PATCH 3/8] media: iris: migrate firmware " Mukesh Ojha
2026-09-26 15:59   ` sashiko-bot [this message]
2026-09-25 15:57 ` [PATCH 4/8] media: venus: " Mukesh Ojha
2026-09-26 15:59   ` sashiko-bot
2026-09-25 15:57 ` [PATCH 5/8] net: ipa: " Mukesh Ojha
2026-09-26 15:59   ` sashiko-bot
2026-09-25 15:57 ` [PATCH 6/8] wifi: ath12k: " Mukesh Ojha
2026-09-26 15:59   ` sashiko-bot
2026-09-25 15:57 ` [PATCH 7/8] soc: qcom: mdt_loader: remove qcom_mdt_load() Mukesh Ojha
2026-09-26 15:59   ` sashiko-bot
2026-09-25 15:57 ` [PATCH 8/8] firmware: qcom: pas: remove qcom_pas_auth_and_reset() Mukesh Ojha

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=20260926155935.D7FDD1F000FF@smtp.kernel.org \
    --to=sashiko-bot@kernel.org \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=media-ci@linuxtv.org \
    --cc=mukesh.ojha@oss.qualcomm.com \
    --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