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: <ZXBGTxS7sUSILtLs@trax>
Date:   Wed, 6 Dec 2023 11:00:47 +0100
From:   "Jorge Ramirez-Ortiz, Foundries" <jorge@...ndries.io>
To:     Avri Altman <Avri.Altman@....com>
Cc:     Adrian Hunter <adrian.hunter@...el.com>,
        Jorge Ramirez-Ortiz <jorge@...ndries.io>,
        "ulf.hansson@...aro.org" <ulf.hansson@...aro.org>,
        "christian.loehle@....com" <christian.loehle@....com>,
        "jinpu.wang@...os.com" <jinpu.wang@...os.com>,
        "axboe@...nel.dk" <axboe@...nel.dk>,
        "beanhuo@...ron.com" <beanhuo@...ron.com>,
        "yibin.ding@...soc.com" <yibin.ding@...soc.com>,
        "victor.shih@...esyslogic.com.tw" <victor.shih@...esyslogic.com.tw>,
        "asuk4.q@...il.com" <asuk4.q@...il.com>,
        "hkallweit1@...il.com" <hkallweit1@...il.com>,
        "yangyingliang@...wei.com" <yangyingliang@...wei.com>,
        "yebin10@...wei.com" <yebin10@...wei.com>,
        "linux-mmc@...r.kernel.org" <linux-mmc@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] mmc: rpmb: do not force a retune before RPMB switch

On 06/12/23 07:02:43, Avri Altman wrote:
> >
> > On 4/12/23 17:01, Jorge Ramirez-Ortiz wrote:
> > > Requesting a retune before switching to the RPMB partition has been
> > > observed to cause CRC errors on the RPMB reads (-EILSEQ).
> >
> > There are still 2 concerns:
> > 1) We don't really know the root cause.  Have you determined if here are
> > CRC errors in the main partition also?

right, and I don't disagree with that.

As a test I created a 4GB file from /dev/random which I then copied
several times (dd if= ....)

root@...cg-dwg-sec:/sys/kernel/debug/mmc0# cat err_stats
# Command Timeout Occurred:      0
# Command CRC Errors Occurred:   0
# Data Timeout Occurred:         0
# Data CRC Errors Occurred:      0
# Auto-Cmd Error Occurred:       0
# ADMA Error Occurred:   0
# Tuning Error Occurred:         0
# CMDQ RED Errors:       0
# CMDQ GCE Errors:       0
# CMDQ ICCE Errors:      0
# Request Timedout:      0
# CMDQ Request Timedout:         0
# ICE Config Errors:     0
# Controller Timedout errors:    0
# Unexpected IRQ errors:         0

However as soon as I access RPMB and fails (it takes just a few tries) I see:

I/TC: RPMB: Using generated key
[   86.902118] sdhci-arasan ff160000.mmc: __mmc_blk_ioctl_cmd: data error -84
E/TC:? 0
E/TC:? 0 TA panicked with code 0xffff0000
E/LD:  Status of TA 22250a54-0bf1-48fe-8002-7b20f1c9c9b1
E/LD:   arch: aarch64
E/LD:  region  0: va 0xc0004000 pa 0x7e200000 size 0x002000 flags rw-s (ldelf)
E/LD:  region  1: va 0xc0006000 pa 0x7e202000 size 0x008000 flags r-xs (ldelf)
E/LD:  region  2: va 0xc000e000 pa 0x7e20a000 size 0x001000 flags rw-s (ldelf)
E/LD:  region  3: va 0xc000f000 pa 0x7e20b000 size 0x004000 flags rw-s (ldelf)
E/LD:  region  4: va 0xc0013000 pa 0x7e20f000 size 0x001000 flags r--s
E/LD:  region  5: va 0xc0014000 pa 0x7e22c000 size 0x005000 flags rw-s (stack)
E/LD:  region  6: va 0xc0019000 pa 0x818ea9ba8 size 0x002000 flags rw-- (param)
E/LD:  region  7: va 0xc001b000 pa 0x818e97ba8 size 0x001000 flags rw-- (param)
E/LD:  region  8: va 0xc004f000 pa 0x00001000 size 0x014000 flags r-xs [0]
E/LD:  region  9: va 0xc0063000 pa 0x00015000 size 0x008000 flags rw-s [0]
E/LD:   [0] 22250a54-0bf1-48fe-8002-7b20f1c9c9b1 @ 0xc004f000
E/LD:  Call stack:
E/LD:   0xc0051a14
E/LD:   0xc004f31c
E/LD:   0xc0052d40
E/LD:   0xc004f624

root@...cg-dwg-sec:/var/rootdirs/home/fio# cat /sys/kernel/debug/mmc0/err_stats
# Command Timeout Occurred:      0
# Command CRC Errors Occurred:   0
# Data Timeout Occurred:         0
# Data CRC Errors Occurred:      1
# Auto-Cmd Error Occurred:       0
# ADMA Error Occurred:   0
# Tuning Error Occurred:         0
# CMDQ RED Errors:       0
# CMDQ GCE Errors:       0
# CMDQ ICCE Errors:      0
# Request Timedout:      0
# CMDQ Request Timedout:         0
# ICE Config Errors:     0
# Controller Timedout errors:    0
# Unexpected IRQ errors:         0

> > 2) Forcing this on everyone
> >
> > The original idea was that because re-tuning cannot be done in RPMB, the
> > need to re-rune in RPMB could be avoided by always re-tuning before
> > switching to RPMB and then switching straight back. IIRC re-tuning should
> > guarantee at least 4MB more I/O without issue.
> Performance is hardly an issue in the context of RPMB access -
> For most cases it’s a single frame.

Yes, the security use case typically stores hashes, variables
(bootcount, upgrade_available, versions, that sort of thing) and
certificates in RPMB.

Since you mentioned, I am seeing that tuning before switching to RPMB
has an impact on performance. As a practical test, just reading a 6 byte
variable incurs in 50ms penalty in kernel space due to the need to
retune 5 times. Not great since the request is coming from a Trusted
Application via OP-TEE through the supplicant meaning this TEE thread
(they are statically allocated CFG_NUM_THREADS) will be reserved for
quite a bit of time.

Roughly:
TA --> OP-TEE (core) --> TEE-supplicant --> Kernel (>50ms) --> OP-TEE --> TA

Adrian, I couldn't find the original performance justification for
enabling this feature globally. At which point do you think it becomes
beneficial to retune before accessing RPMB?

>
> Thanks,
> Avri
>
> >
> > The alternative to dropping re-tuning in this case could be to add a retry loop
> > for MMC_DRV_OP_IOCTL_RPMB if the error is -EILSEQ

For the security use case I mentioned above - even if it didn't end up in
the occasional CRC errors - I honestly see little value: dropping the
feature - or controlling it via CFG_ - seems more logical to me. Would you
agree?

> >
> >
> > >
> > > Since RPMB reads can not be retried, the clients would be directly
> > > affected by the errors.
> > >
> > > This commit disables the request prior to RPMB switching while
> > > allowing the pause interface to still request a retune before the
> > > pause for other use cases.
> > >
> > > This was verified with the sdhci-of-arasan driver (ZynqMP) configured
> > > for HS200 using two separate eMMC cards (DG4064 and 064GB2). In both
> > > cases, the error was easy to reproduce triggering every few tenths of
> > > reads.
> > >
> > > Signed-off-by: Jorge Ramirez-Ortiz <jorge@...ndries.io>
> > > ---
> > >  drivers/mmc/core/block.c | 2 +-
> > >  drivers/mmc/core/host.c  | 7 ++++---
> > >  drivers/mmc/core/host.h  | 2 +-
> > >  3 files changed, 6 insertions(+), 5 deletions(-)
> > >
> > > diff --git a/drivers/mmc/core/block.c b/drivers/mmc/core/block.c index
> > > f9a5cffa64b1..1d69078ad9b2 100644
> > > --- a/drivers/mmc/core/block.c
> > > +++ b/drivers/mmc/core/block.c
> > > @@ -859,7 +859,7 @@ static int mmc_blk_part_switch_pre(struct
> > mmc_card *card,
> > >                       if (ret)
> > >                               return ret;
> > >               }
> > > -             mmc_retune_pause(card->host);
> > > +             mmc_retune_pause(card->host, false);
> > >       }
> > >
> > >       return ret;
> > > diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c index
> > > 096093f7be00..a9b95aaa2235 100644
> > > --- a/drivers/mmc/core/host.c
> > > +++ b/drivers/mmc/core/host.c
> > > @@ -119,13 +119,14 @@ void mmc_retune_enable(struct mmc_host
> > *host)
> > >
> > >  /*
> > >   * Pause re-tuning for a small set of operations.  The pause begins
> > > after the
> > > - * next command and after first doing re-tuning.
> > > + * next command and, if retune is set, after first doing re-tuning.
> > >   */
> > > -void mmc_retune_pause(struct mmc_host *host)
> > > +void mmc_retune_pause(struct mmc_host *host, bool retune)
> > >  {
> > >       if (!host->retune_paused) {
> > >               host->retune_paused = 1;
> > > -             mmc_retune_needed(host);
> > > +             if (retune)
> > > +                     mmc_retune_needed(host);
> >
> > Better to just drop mmc_retune_needed(host);
> >
> > >               mmc_retune_hold(host);
> >
> > There is still a small chance that re-tuning is needed anyway in which case it
> > will still be done.
> >
> > >       }
> > >  }
> > > diff --git a/drivers/mmc/core/host.h b/drivers/mmc/core/host.h index
> > > 48c4952512a5..321776b52270 100644
> > > --- a/drivers/mmc/core/host.h
> > > +++ b/drivers/mmc/core/host.h
> > > @@ -18,7 +18,7 @@ void mmc_retune_disable(struct mmc_host *host);
> > > void mmc_retune_hold(struct mmc_host *host);  void
> > > mmc_retune_release(struct mmc_host *host);  int mmc_retune(struct
> > > mmc_host *host); -void mmc_retune_pause(struct mmc_host *host);
> > > +void mmc_retune_pause(struct mmc_host *host, bool retune);
> > >  void mmc_retune_unpause(struct mmc_host *host);
> > >
> > >  static inline void mmc_retune_clear(struct mmc_host *host)
>

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ