linux-kernel - Re: [RFC PATCH] memory: renesas-rpc-if: Correct QSPI data transfer in Manual mode

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]

Message-ID: <beb6c373-e9d8-b0f1-78c5-cf21d01a5a9b@canonical.com>
Date:   Tue, 28 Sep 2021 11:46:41 +0200
From:   Krzysztof Kozlowski <krzysztof.kozlowski@...onical.com>
To:     Wolfram Sang <wsa+renesas@...g-engineering.com>,
        linux-kernel@...r.kernel.org
Cc:     linux-renesas-soc@...r.kernel.org,
        Duc Nguyen <duc.nguyen.ub@...esas.com>,
        "Lad, Prabhakar" <prabhakar.csengg@...il.com>,
        Geert Uytterhoeven <geert@...ux-m68k.org>
Subject: Re: [RFC PATCH] memory: renesas-rpc-if: Correct QSPI data transfer in
 Manual mode

On 22/09/2021 11:10, Wolfram Sang wrote:
> This patch fixes 2 problems:
> [1] The output warning logs and data loss when performing
> mount/umount then remount the device with jffs2 format.
> [2] The access width of SMWDR[0:1]/SMRDR[0:1] register is wrong.
> 
> This is the sample warning logs when performing mount/umount then
> remount the device with jffs2 format:
> jffs2: jffs2_scan_inode_node(): CRC failed on node at 0x031c51d4:
> Read 0x00034e00, calculated 0xadb272a7
> 
> The reason for issue [1] is that the writing data seems to
> get messed up.
> Data is only completed when the number of bytes is divisible by 4.
> If you only have 3 bytes of data left to write, 1 garbage byte
> is inserted after the end of the write stream.
> If you only have 2 bytes of data left to write, 2 bytes of '00'
> are added into the write stream.
> If you only have 1 byte of data left to write, 2 bytes of '00'
> are added into the write stream. 1 garbage byte is inserted after
> the end of the write stream.
> 
> To solve problem [1], data must be written continuously in serial
> and the write stream ends when data is out.
> 
> Following HW manual 62.2.15, access to SMWDR0 register should be
> in the same size as the transfer size specified in the SPIDE[3:0]
> bits in the manual mode enable setting register (SMENR).
> Be sure to access from address 0.
> 
> So, in 16-bit transfer (SPIDE[3:0]=b'1100), SMWDR0 should be
> accessed by 16-bit width.
> Similar to SMWDR1, SMDDR0/1 registers.
> In current code, SMWDR0 register is accessed by regmap_write()
> that only set up to do 32-bit width.
> 
> To solve problem [2], data must be written 16-bit or 8-bit when
> transferring 1-byte or 2-byte.
> 
> Signed-off-by: Duc Nguyen <duc.nguyen.ub@...esas.com>
> [wsa: refactored to use regmap only via reg_read/reg_write]
> Signed-off-by: Wolfram Sang <wsa+renesas@...g-engineering.com>
> ---
> 
> Hi,
> 
> I could reproduce the issue by a simple:
> 
>   $ echo "Hello" > /dev/mtd10
> 
> The original BSP patch fixed the issue but mixed regmap-acces with
> ioread/iowrite accesses. So, I refactored it to use custom regmap
> accessors. This keeps the code more readable IMO. With this patch, my
> custom test cases work as well as the JFFS2 remount mentioned in the
> commit message. Tested on a Renesas Condor board (R-Car V3M) and a
> Falcon board (R-Car V3U). I send this as RFC because this is my first
> patch for the RPC code and hope for feedback. The BSP team has been
> contacted as well for comments and testing. Nonetheless, this addresses
> a serious issue which has caused broken boards because of writing to
> unintended locations. So, I'd like to see this discussed and applied
> soon if possible.
> 
> Thanks everyone,
> 
>    Wolfram
> 

Thanks for testing and comments. Any Fixes tag for it?

Best regards,
Krzysztof