linux-kernel - Re: [PATCH RFC 2/2] memory: add Renesas RPC-IF driver

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] [day] [month] [year] [list]
Date:   Thu, 27 Feb 2020 23:32:07 +0300
From:   Sergei Shtylyov <sergei.shtylyov@...entembedded.com>
To:     "Behme Dirk (CM/ESO2)" <dirk.behme@...bosch.com>
Cc:     devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
        Philipp Zabel <p.zabel@...gutronix.de>,
        Mason Yang <masonccyang@...c.com.tw>,
        linux-spi@...r.kernel.org, Chris Brandt <chris.brandt@...esas.com>,
        linux-renesas-soc@...r.kernel.org,
        Geert Uytterhoeven <geert@...ux-m68k.org>
Subject: Re: [PATCH RFC 2/2] memory: add Renesas RPC-IF driver

Hello!

On 02/26/2020 12:54 PM, Behme Dirk (CM/ESO2) wrote:

>>>>>>>    From d72b805cc461ab1e9747c973e9be84e7abb8f828 Mon Sep 17 00:00:00 2001
>>>>>>> From: Dirk Behme <dirk.behme@...bosch.com>
>>>>>>> Date: Tue, 4 Feb 2020 08:39:31 +0100
>>>>>>> Subject: [PATCH] memory: renesas-rpc-if: Correct the STRTIM and some other
>>>>>>>     clean up
>>>>>>>
>>>>>>> This is required to make the driver work correctly in my M3 environment.
>>>>>>>
>>>>>>> Signed-off-by: Dirk Behme <dirk.behme@...bosch.com>
>>>>>>> ---
>>>>>>>     drivers/memory/renesas-rpc-if.c | 42 ++++++++++++++++++++-------------
>>>>>>>     1 file changed, 25 insertions(+), 17 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/memory/renesas-rpc-if.c b/drivers/memory/renesas-rpc-if.c
>>>>>>> index 04be92b64bfa..f4356b066384 100644
>>>>>>> --- a/drivers/memory/renesas-rpc-if.c
>>>>>>> +++ b/drivers/memory/renesas-rpc-if.c
>>>>>> [...]
>>>>>>> @@ -513,19 +525,15 @@ ssize_t rpcif_dirmap_read(struct rpcif *rpc, u64 offs, size_t len, void *buf)
>>>>>>>         pm_runtime_get_sync(rpc->dev);
>>>>>>>
>>>>>>>         regmap_update_bits(rpc->regmap, RPCIF_CMNCR, RPCIF_CMNCR_MD, 0);
>>>>>>> -    regmap_write(rpc->regmap, RPCIF_DRCR,
>>>>>>> -             RPCIF_DRCR_RBURST(32) | RPCIF_DRCR_RBE);
>>>>>>> -    regmap_write(rpc->regmap, RPCIF_DRCMR, rpc->command);
>>>>>>> -    regmap_write(rpc->regmap, RPCIF_DREAR,
>>>>>>> -             RPCIF_DREAR_EAV(offs >> 25) | RPCIF_DREAR_EAC(1));
>>>>>>> -    regmap_write(rpc->regmap, RPCIF_DROPR, rpc->option);
>>>>>>> -    regmap_write(rpc->regmap, RPCIF_DRENR,
>>>>>>> -             rpc->enable & ~RPCIF_SMENR_SPIDE(0xF));
>>>>>>> -    regmap_write(rpc->regmap, RPCIF_DRDMCR, rpc->dummy);
>>>>>>> -    regmap_write(rpc->regmap, RPCIF_DRDRENR, rpc->ddr);
>>>>>>
>>>>>>       The driver somehow works only with this left in place (with 2 bytes eaten
>>>>>> as before), otherwise all the flash reads all 0xff (via dirmap).
>>>>>
>>>>>
>>>>> Do you boot from hyperflash?
>>>>
>>>>      No, I have arewto say 'cpld write 30 1' in U-Boot before a boot a kernel.
>>
>>     s/arewto/to/. :-)
>>
>>>> Normally, the V3x Starter Kit boards are wired for the QSPI flash chips.
>>>>
>>>>> The system I'm using for testing boots from hyperflash. So most probably all registers
>>>>> I don't touch in the driver are put into a reasonable state by the boot code, already.
>>>>> If you don't boot from hyperflash, that at least would explain our different behavior.
>>>>
>>>>      Yes. Mind dumping the registers and sending to me?
>>>
>>> Using the attached debug patch (0001-memory-renesas-rpc-if-DEBUG-Dump-register-
>>> content.patch) on a r8a7796 system booting from Hyperflash with above register
>> dropping reverted (i.e. including touching these registers) I get
>>>
>>> Before:
>>> RPCIF_DRCR:    0x00000000
>>> RPCIF_DRCMR:   0x00a00000
>>> RPCIF_DREAR:   0x00000000
>>> RPCIF_DROPR:   0x00000000
>>> RPCIF_DRENR:   0xa222d400
>>> RPCIF_DRDMCR:  0x0000000e
>>> RPCIF_DRDRENR: 0x00005101
>>>
>>> After:
>>> RPCIF_DRCR:    0x001f0100
>>> RPCIF_DRCMR:   0x00a00000
>>> RPCIF_DREAR:   0x00010001
>>> RPCIF_DROPR:   0x00000000
>>> RPCIF_DRENR:   0xa202d400
>>> RPCIF_DRDMCR:  0x0000000e
>>> RPCIF_DRDRENR: 0x00005101
>>>
>>> Comparing that, just 3 registers are different between my working version ("Before") and the version which shows the 2-byte offset ("After"): RPCIF_DRCR, RPCIF_DREAR and RPCIF_DRENR. With try & error, at least in my setup, I was able to reduce this to just RPCIF_DRCR. Dropping the burst mode I was able to 'fix' the two byte offset issue.
>>
>>     ACK! Thanks a lot for finding it! :-)
>>     That's what I get on the first dirmap read:
>>
>> Before:
>> RPCIF_CMNCR:   0x81fff301
>> RPCIF_DRCR:    0x00000000
>> RPCIF_DRCMR:   0x00030000
>> RPCIF_DREAR:   0x00000000
>> RPCIF_DROPR:   0x00000000
>> RPCIF_DRENR:   0x00004700
>> RPCIF_DRDMCR:  0x00000000
>> RPCIF_DRDRENR: 0x00000000
>>
>> After:
>> RPCIF_CMNCR:   0x01fff301
>> RPCIF_DRCR:    0x001f0100
>> RPCIF_DRCMR:   0x00800000
>> RPCIF_DREAR:   0x00000001
>> RPCIF_DROPR:   0x00000000
>> RPCIF_DRENR:   0xa202d400
>> RPCIF_DRDMCR:  0x0000000e
>> RPCIF_DRDRENR: 0x00005101
>>
>>> Do you like to give the attached 0001-memory-renesas-rpc-if-Don-t-use-burst-mode-on-read.patch a try in your setup?
>>
>>     Works like charm! :-)
> 
> 
> Good news, thanks! :)
> 
> 
>>     Unfortunately, the SPI dirmap and/or writes are still broken.
> 
> 
> I'm unsure about which function we are talking for "SPI dirmap"?

   The same, rpcif_dirmap_read(). I'm now thinking it works correctly, it's just
the writes that are borked.

> Regarding writes, in rpcif_io_xfer() in RPCIF_DATA_IN we have an additional place where burst mode is enabled:

   This works around the hardware bug where the SPI NOR RDID command's data bytes get
spoiled, leading to unrecognized flash chip. Well, there's a comment before this *if*. 

> if (!(smenr & RPCIF_SMENR_ADE(0xf)) && rpc->dirmap) {
> ...
> regmap_write(rpc->regmap, RPCIF_DRCR,
>     RPCIF_DRCR_RBURST(32) | RPCIF_DRCR_RBE);
> ...
> 
> Maybe it's worth a try to replace this by just 0, as well:
> 
> regmap_write(rpc->regmap, RPCIF_DRCR, 0);

   I've just tried that, here's the result:

spi-nor spi0.0: unrecognized JEDEC id bytes: 01 01 01 01 01 01                  

(and the flash doesn't get registered).

> But of course, this is guessing, as I'm not sure if this
> 
> if (!(smenr & RPCIF_SMENR_ADE(0xf)) && rpc->dirmap) {
> 
> path is taken, at all?

   It's taken but only with SPI NOR.

> Or even better, if you could adapt the "before / after" debug patch to the path which are still not working for you and share it and the results, we could compare it with my setup, too :)

  I'll try doing that tomorrow...

> Dirk

MBR, Sergei