linux-kernel - Re: [PATCH 1/2] media: verisilicon: fix excessive stack usage

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Wed, 21 Jun 2023 10:44:23 -0400
From:   Nicolas Dufresne <nicolas.dufresne@...labora.com>
To:     Arnd Bergmann <arnd@...nel.org>,
        Ezequiel Garcia <ezequiel@...guardiasur.com.ar>,
        Philipp Zabel <p.zabel@...gutronix.de>,
        Mauro Carvalho Chehab <mchehab@...nel.org>,
        Hans Verkuil <hverkuil-cisco@...all.nl>,
        Benjamin Gaignard <benjamin.gaignard@...labora.com>
Cc:     Arnd Bergmann <arnd@...db.de>, kernel test robot <lkp@...el.com>,
        Jernej Skrabec <jernej.skrabec@...il.com>,
        linux-media@...r.kernel.org, linux-rockchip@...ts.infradead.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 1/2] media: verisilicon: fix excessive stack usage

Hi,

thanks you.

Le vendredi 16 juin 2023 à 16:48 +0200, Arnd Bergmann a écrit :
> From: Arnd Bergmann <arnd@...db.de>
> 
> In some configurations, gcc decides not to inline the register accessor functions,
> which in turn leads to lots of temporary hantro_reg structures on the stack that
> cannot be eliminated because they escape into an uninlined function:
> 
> drivers/media/platform/verisilicon/rockchip_vpu981_hw_av1_dec.c:1022:1: warning: the frame size of 1112 bytes is larger than 1024 bytes [-Wframe-larger-than=]
> 
> Mark all of these as __always_inline so the compiler is able to completely
> eliminate the temporary structures instead, which brings the stack usage
> back down to just the normal local variables.

This is falling into compiler bug territory, though I see no harm in forcing
these to inline, as in the old days these would have been macros anyway.

> 
> Reported-by: kernel test robot <lkp@...el.com>
> Closes: https://lore.kernel.org/oe-kbuild-all/202306151506.goHEegOd-lkp@intel.com/
> Fixes: 727a400686a2c ("media: verisilicon: Add Rockchip AV1 decoder")
> Signed-off-by: Arnd Bergmann <arnd@...db.de>

Thanks again,

Reviewed-by: Nicolas Dufresne <nicolas.dufresne@...labora.com>

> ---
>  drivers/media/platform/verisilicon/hantro.h | 22 ++++++++++-----------
>  1 file changed, 11 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/media/platform/verisilicon/hantro.h b/drivers/media/platform/verisilicon/hantro.h
> index 6523ffb748812..6c5e56ce5b351 100644
> --- a/drivers/media/platform/verisilicon/hantro.h
> +++ b/drivers/media/platform/verisilicon/hantro.h
> @@ -370,26 +370,26 @@ extern int hantro_debug;
>  	pr_err("%s:%d: " fmt, __func__, __LINE__, ##args)
>  
>  /* Structure access helpers. */
> -static inline struct hantro_ctx *fh_to_ctx(struct v4l2_fh *fh)
> +static __always_inline struct hantro_ctx *fh_to_ctx(struct v4l2_fh *fh)
>  {
>  	return container_of(fh, struct hantro_ctx, fh);
>  }
>  
>  /* Register accessors. */
> -static inline void vepu_write_relaxed(struct hantro_dev *vpu,
> +static __always_inline void vepu_write_relaxed(struct hantro_dev *vpu,
>  				      u32 val, u32 reg)
>  {
>  	vpu_debug(6, "0x%04x = 0x%08x\n", reg / 4, val);
>  	writel_relaxed(val, vpu->enc_base + reg);
>  }
>  
> -static inline void vepu_write(struct hantro_dev *vpu, u32 val, u32 reg)
> +static __always_inline void vepu_write(struct hantro_dev *vpu, u32 val, u32 reg)
>  {
>  	vpu_debug(6, "0x%04x = 0x%08x\n", reg / 4, val);
>  	writel(val, vpu->enc_base + reg);
>  }
>  
> -static inline u32 vepu_read(struct hantro_dev *vpu, u32 reg)
> +static __always_inline u32 vepu_read(struct hantro_dev *vpu, u32 reg)
>  {
>  	u32 val = readl(vpu->enc_base + reg);
>  
> @@ -397,27 +397,27 @@ static inline u32 vepu_read(struct hantro_dev *vpu, u32 reg)
>  	return val;
>  }
>  
> -static inline void vdpu_write_relaxed(struct hantro_dev *vpu,
> +static __always_inline void vdpu_write_relaxed(struct hantro_dev *vpu,
>  				      u32 val, u32 reg)
>  {
>  	vpu_debug(6, "0x%04x = 0x%08x\n", reg / 4, val);
>  	writel_relaxed(val, vpu->dec_base + reg);
>  }
>  
> -static inline void vdpu_write(struct hantro_dev *vpu, u32 val, u32 reg)
> +static __always_inline void vdpu_write(struct hantro_dev *vpu, u32 val, u32 reg)
>  {
>  	vpu_debug(6, "0x%04x = 0x%08x\n", reg / 4, val);
>  	writel(val, vpu->dec_base + reg);
>  }
>  
> -static inline void hantro_write_addr(struct hantro_dev *vpu,
> +static __always_inline void hantro_write_addr(struct hantro_dev *vpu,
>  				     unsigned long offset,
>  				     dma_addr_t addr)
>  {
>  	vdpu_write(vpu, addr & 0xffffffff, offset);
>  }
>  
> -static inline u32 vdpu_read(struct hantro_dev *vpu, u32 reg)
> +static __always_inline u32 vdpu_read(struct hantro_dev *vpu, u32 reg)
>  {
>  	u32 val = readl(vpu->dec_base + reg);
>  
> @@ -425,7 +425,7 @@ static inline u32 vdpu_read(struct hantro_dev *vpu, u32 reg)
>  	return val;
>  }
>  
> -static inline u32 vdpu_read_mask(struct hantro_dev *vpu,
> +static __always_inline u32 vdpu_read_mask(struct hantro_dev *vpu,
>  				 const struct hantro_reg *reg,
>  				 u32 val)
>  {
> @@ -437,14 +437,14 @@ static inline u32 vdpu_read_mask(struct hantro_dev *vpu,
>  	return v;
>  }
>  
> -static inline void hantro_reg_write(struct hantro_dev *vpu,
> +static __always_inline void hantro_reg_write(struct hantro_dev *vpu,
>  				    const struct hantro_reg *reg,
>  				    u32 val)
>  {
>  	vdpu_write_relaxed(vpu, vdpu_read_mask(vpu, reg, val), reg->base);
>  }
>  
> -static inline void hantro_reg_write_s(struct hantro_dev *vpu,
> +static __always_inline void hantro_reg_write_s(struct hantro_dev *vpu,
>  				      const struct hantro_reg *reg,
>  				      u32 val)
>  {