Message-ID: <CAFBinCBd=dOZjv4KEDY9ODOONDO=WtaBqAMRi_a+gH3_iZEAHQ@mail.gmail.com>
Date:   Sat, 16 Mar 2019 11:55:00 +0100
From:   Martin Blumenstingl <martin.blumenstingl@...glemail.com>
To:     Liang Yang <liang.yang@...ogic.com>
Cc:     Miquel Raynal <miquel.raynal@...tlin.com>, mark.rutland@....com,
        devicetree@...r.kernel.org, jianxin.pan@...ogic.com,
        linux-kernel@...r.kernel.org, robh+dt@...nel.org,
        linux-mtd@...ts.infradead.org, linux-amlogic@...ts.infradead.org,
        linux-arm-kernel@...ts.infradead.org
Subject: Re: [RFC PATCH nand-next 0/2] meson-nand: support for older SoCs

Hi Liang,

On Tue, Mar 12, 2019 at 10:05 AM Liang Yang <liang.yang@...ogic.com> wrote:
>
> Hi Martin and Miquel,
>
> On 2019/3/7 21:09, Miquel Raynal wrote:
> > Hello,
> >
> > Martin Blumenstingl <martin.blumenstingl@...glemail.com> wrote on Tue,
> > 5 Mar 2019 23:12:51 +0100:
> >
> >> Hi Liang,
> >>
> >> On Mon, Mar 4, 2019 at 5:55 AM Liang Yang <liang.yang@...ogic.com> wrote:
> >>>
> >>> Hello Martin,
> >>>
> >>> On 2019/3/2 2:29, Martin Blumenstingl wrote:
> >>>> Hi Liang,
> >>>>
> >>>> I am trying to add support for older SoCs to the meson-nand driver.
> >>>> Back when the driver was in development I used an early revision (of
> >>>> your driver) and did some modifications to make it work on older SoCs.
> >>>>
> >>>> Now that the driver is upstream I wanted to give it another try and
> >>>> make a real patch out of it. Unfortunately it's not working anymore.
> >>>>
> >>>> As far as I know the NFC IP block revision on GXL is similar (or even
> >>>> the same?) as on all older SoCs. As far as I can tell only the clock
> >>>> setup is different on the older SoCs (which have a dedicated NAND
> >>>> clock):
> >>>> - we don't need the "amlogic,mmc-syscon" property on the older SoCs
> >>>>     because we don't need to setup any muxing (common clock framework
> >>>>     will do everything for us)
> >>>> - "rx" and "tx" clocks don't exist
> >>>> - I could not find any other differences between Meson8, Meson8b,
> >>>>     Meson8m2, GXBB and GXL
> >>>>
> >>> That is right. the serials NFC is almost the same except:
> >>> 1) The clock control and source that M8-serials are not share with EMMC.
> >>> 2) The base register address
> >>> 3) DMA encryption option which we don't care on NFC driver.
> >> great, thank you for confirming this!
> >>
> >>>> In this series I'm sending two patches which add support for the older
> >>>> SoCs.
> >>>>
> >>>> Unfortunately these patches are currently not working for me (hence the
> >>>> "RFC" prefix). I get a (strange) crash which is triggered by the
> >>>> kzalloc() in meson_nfc_read_buf() - see below for more details.
> >>>>
> >>>> Can you please help me on this one? I'd like to know whether:
> >>>> - the meson-nand driver works for you on GXL or AXG on linux-next?
> >>>>     (I was running these patches on top of next-20190301 on my M8S
> >>>>     board which uses a 32-bit Meson8m2 SoC. I don't have any board using
> >>>>     a GXL SoC which also has NAND)
> >>> Yes, it works on AXG platform using a MXIC slc nand flash(MX30LF4G); but
> >>> i an not sure it runs the same flow with yours. because i see the print
> >>> "Counld not find a valid ONFI parameter page, ...." in yours. i will try
> >>> to reproduce it on AXG(i don't have a M8 platform now).
> >> I'm looking forward to hear about the test results on your AXG boards
> >> for reference: my board has a SK Hynix H27UCG8T2B (ID bytes: 0xad 0xde
> >> 0x94 0xeb 0x74 0x44, 20nm MLC)
> >> I have another board (where I haven't tested the NFC driver yet) with
> >> a SK Hynix H27UCG8T2E (ID bytes: 0xad 0xde 0x14 0xa7 0x42 0x4a, 1Ynm
> >> MLC). if it helps with your analysis I can test on that board as well
> >
> > Liang, you just have to fake the output of the ONFI page detection and
> > you will probably run into this error which will then be easy to
> > reproduce.
> >
> i don't reproduce it by using a SK Hynix nand flash H27UCG8T2E on gxl
> platform. it runs well.
> [......]
> [    0.977127] loop: module loaded
> [    0.998625] Could not find a valid ONFI parameter page, trying
> bit-wise majority to recover it
> [    1.001619] ONFI parameter recovery failed, aborting
> [    1.006684] Could not find valid JEDEC parameter page; aborting
> [    1.012391] nand: device found, Manufacturer ID: 0xad, Chip ID: 0xde
> [    1.018660] nand: Hynix NAND 8GiB 3,3V 8-bit
> [    1.022885] nand: 8192 MiB, MLC, erase size: 4096 KiB, page size:
> 16384, OOB size: 1664
> [    1.047033] Bad block table not found for chip 0
> [    1.054950] Bad block table not found for chip 0
> [    1.054970] Scanning device for bad blocks
> [    1.522664] random: fast init done
> [    4.893731] Bad eraseblock 1985 at 0x0001f07fc000
> [    5.020637] Bad block table written to 0x0001ffc00000, version 0x01
> [    5.028258] Bad block table written to 0x0001ff800000, version 0x01
> [    5.029905] 5 fixed-partitions partitions found on MTD device
> d0074800.nfc
> [    5.035714] Creating 5 MTD partitions on "d0074800.nfc":
> [......]
>
> Martin, Now i am not sure whether NFC driver leads to kernel panic when
> calling kmem_cache_alloc_trace.
thank you for confirming that it works for you on GXL

I'm not sure that this is a NFC driver problem.
after enabling CONFIG_SLAB_FREELIST_HARDENED in my kernel config the
crash moves. it's now crashing in slub.c's kfree() at
BUG_ON(!PageCompound(page));

maybe this is related to some difference in 32-bit ARM and arm64
or it could even be some memory management issue
I'm not sure yet so I'll try to dig deeper


Regards
Martin

View attachment "meson-nfc-32bit-crash.txt" of type "text/plain" (6059 bytes)

Powered by blists - more mailing lists